Resguardo inicial

linux-7.2.3/arch/powerpc/mm/book3s64/Makefile

@@ -0,0 +1,33 @@
+# SPDX-License-Identifier: GPL-2.0
+
+obj-y				+= mmu_context.o pgtable.o trace.o
+ifdef CONFIG_PPC_64S_HASH_MMU
+CFLAGS_REMOVE_slb.o = $(CC_FLAGS_FTRACE)
+obj-y				+= hash_pgtable.o hash_utils.o hash_tlb.o slb.o slice.o
+obj-$(CONFIG_PPC_HASH_MMU_NATIVE)	+= hash_native.o
+obj-$(CONFIG_PPC_4K_PAGES)	+= hash_4k.o
+obj-$(CONFIG_PPC_64K_PAGES)	+= hash_64k.o
+obj-$(CONFIG_TRANSPARENT_HUGEPAGE) += hash_hugepage.o
+obj-$(CONFIG_PPC_SUBPAGE_PROT)	+= subpage_prot.o
+endif
+
+obj-$(CONFIG_HUGETLB_PAGE)	+= hugetlbpage.o
+
+obj-$(CONFIG_PPC_RADIX_MMU)	+= radix_pgtable.o radix_tlb.o
+ifdef CONFIG_HUGETLB_PAGE
+obj-$(CONFIG_PPC_RADIX_MMU)	+= radix_hugetlbpage.o
+endif
+obj-$(CONFIG_SPAPR_TCE_IOMMU)	+= iommu_api.o
+obj-$(CONFIG_PPC_PKEY)	+= pkeys.o
+
+# Instrumenting the SLB fault path can lead to duplicate SLB entries
+KCOV_INSTRUMENT_slb.o := n
+
+# Parts of these can run in real mode and therefore are
+# not safe with the current outline KASAN implementation
+KASAN_SANITIZE_mmu_context.o := n
+KASAN_SANITIZE_pgtable.o := n
+KASAN_SANITIZE_radix_pgtable.o := n
+KASAN_SANITIZE_radix_tlb.o := n
+KASAN_SANITIZE_slb.o := n
+KASAN_SANITIZE_pkeys.o := n

linux-7.2.3/arch/powerpc/mm/book3s64/hash_4k.c

@@ -0,0 +1,129 @@
+/*
+ * Copyright IBM Corporation, 2015
+ * Author Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of version 2 of the GNU Lesser General Public License
+ * as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it would be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ */
+
+#include <linux/mm.h>
+#include <asm/machdep.h>
+#include <asm/mmu.h>
+
+#include "internal.h"
+
+int __hash_page_4K(unsigned long ea, unsigned long access, unsigned long vsid,
+		   pte_t *ptep, unsigned long trap, unsigned long flags,
+		   int ssize, int subpg_prot)
+{
+	real_pte_t rpte;
+	unsigned long hpte_group;
+	unsigned long rflags, pa;
+	unsigned long old_pte, new_pte;
+	unsigned long vpn, hash, slot;
+	unsigned long shift = mmu_psize_defs[MMU_PAGE_4K].shift;
+
+	/*
+	 * atomically mark the linux large page PTE busy and dirty
+	 */
+	do {
+		pte_t pte = READ_ONCE(*ptep);
+
+		old_pte = pte_val(pte);
+		/* If PTE busy, retry the access */
+		if (unlikely(old_pte & H_PAGE_BUSY))
+			return 0;
+		/* If PTE permissions don't match, take page fault */
+		if (unlikely(!check_pte_access(access, old_pte)))
+			return 1;
+		/*
+		 * Try to lock the PTE, add ACCESSED and DIRTY if it was
+		 * a write access. Since this is 4K insert of 64K page size
+		 * also add H_PAGE_COMBO
+		 */
+		new_pte = old_pte | H_PAGE_BUSY | _PAGE_ACCESSED;
+		if (access & _PAGE_WRITE)
+			new_pte |= _PAGE_DIRTY;
+	} while (!pte_xchg(ptep, __pte(old_pte), __pte(new_pte)));
+
+	/*
+	 * PP bits. _PAGE_USER is already PP bit 0x2, so we only
+	 * need to add in 0x1 if it's a read-only user page
+	 */
+	rflags = htab_convert_pte_flags(new_pte, flags);
+	rpte = __real_pte(__pte(old_pte), ptep, PTRS_PER_PTE);
+
+	if (cpu_has_feature(CPU_FTR_NOEXECUTE) &&
+	    !cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
+		rflags = hash_page_do_lazy_icache(rflags, __pte(old_pte), trap);
+
+	vpn  = hpt_vpn(ea, vsid, ssize);
+	if (unlikely(old_pte & H_PAGE_HASHPTE)) {
+		/*
+		 * There MIGHT be an HPTE for this pte
+		 */
+		unsigned long gslot = pte_get_hash_gslot(vpn, shift, ssize,
+							 rpte, 0);
+
+		if (mmu_hash_ops.hpte_updatepp(gslot, rflags, vpn, MMU_PAGE_4K,
+					       MMU_PAGE_4K, ssize, flags) == -1)
+			old_pte &= ~_PAGE_HPTEFLAGS;
+	}
+
+	if (likely(!(old_pte & H_PAGE_HASHPTE))) {
+
+		pa = pte_pfn(__pte(old_pte)) << PAGE_SHIFT;
+		hash = hpt_hash(vpn, shift, ssize);
+
+repeat:
+		hpte_group = (hash & htab_hash_mask) * HPTES_PER_GROUP;
+
+		/* Insert into the hash table, primary slot */
+		slot = mmu_hash_ops.hpte_insert(hpte_group, vpn, pa, rflags, 0,
+						MMU_PAGE_4K, MMU_PAGE_4K, ssize);
+		/*
+		 * Primary is full, try the secondary
+		 */
+		if (unlikely(slot == -1)) {
+			hpte_group = (~hash & htab_hash_mask) * HPTES_PER_GROUP;
+			slot = mmu_hash_ops.hpte_insert(hpte_group, vpn, pa,
+							rflags,
+							HPTE_V_SECONDARY,
+							MMU_PAGE_4K,
+							MMU_PAGE_4K, ssize);
+			if (slot == -1) {
+				if (mftb() & 0x1)
+					hpte_group = (hash & htab_hash_mask) *
+							HPTES_PER_GROUP;
+				mmu_hash_ops.hpte_remove(hpte_group);
+				/*
+				 * FIXME!! Should be try the group from which we removed ?
+				 */
+				goto repeat;
+			}
+		}
+		/*
+		 * Hypervisor failure. Restore old pte and return -1
+		 * similar to __hash_page_*
+		 */
+		if (unlikely(slot == -2)) {
+			*ptep = __pte(old_pte);
+			hash_failure_debug(ea, access, vsid, trap, ssize,
+					   MMU_PAGE_4K, MMU_PAGE_4K, old_pte);
+			return -1;
+		}
+		new_pte = (new_pte & ~_PAGE_HPTEFLAGS) | H_PAGE_HASHPTE;
+		new_pte |= pte_set_hidx(ptep, rpte, 0, slot, PTRS_PER_PTE);
+
+		if (stress_hpt())
+			hpt_do_stress(ea, hpte_group);
+	}
+	*ptep = __pte(new_pte & ~H_PAGE_BUSY);
+	return 0;
+}

linux-7.2.3/arch/powerpc/mm/book3s64/hash_64k.c

@@ -0,0 +1,343 @@
+/*
+ * Copyright IBM Corporation, 2015
+ * Author Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of version 2 of the GNU Lesser General Public License
+ * as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it would be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ */
+
+#include <linux/mm.h>
+#include <asm/machdep.h>
+#include <asm/mmu.h>
+
+#include "internal.h"
+
+/*
+ * Return true, if the entry has a slot value which
+ * the software considers as invalid.
+ */
+static inline bool hpte_soft_invalid(unsigned long hidx)
+{
+	return ((hidx & 0xfUL) == 0xfUL);
+}
+
+/*
+ * index from 0 - 15
+ */
+bool __rpte_sub_valid(real_pte_t rpte, unsigned long index)
+{
+	return !(hpte_soft_invalid(__rpte_to_hidx(rpte, index)));
+}
+
+int __hash_page_4K(unsigned long ea, unsigned long access, unsigned long vsid,
+		   pte_t *ptep, unsigned long trap, unsigned long flags,
+		   int ssize, int subpg_prot)
+{
+	real_pte_t rpte;
+	unsigned long hpte_group;
+	unsigned int subpg_index;
+	unsigned long rflags, pa;
+	unsigned long old_pte, new_pte, subpg_pte;
+	unsigned long vpn, hash, slot, gslot;
+	unsigned long shift = mmu_psize_defs[MMU_PAGE_4K].shift;
+
+	/*
+	 * atomically mark the linux large page PTE busy and dirty
+	 */
+	do {
+		pte_t pte = READ_ONCE(*ptep);
+
+		old_pte = pte_val(pte);
+		/* If PTE busy, retry the access */
+		if (unlikely(old_pte & H_PAGE_BUSY))
+			return 0;
+		/* If PTE permissions don't match, take page fault */
+		if (unlikely(!check_pte_access(access, old_pte)))
+			return 1;
+		/*
+		 * Try to lock the PTE, add ACCESSED and DIRTY if it was
+		 * a write access. Since this is 4K insert of 64K page size
+		 * also add H_PAGE_COMBO
+		 */
+		new_pte = old_pte | H_PAGE_BUSY | _PAGE_ACCESSED | H_PAGE_COMBO;
+		if (access & _PAGE_WRITE)
+			new_pte |= _PAGE_DIRTY;
+	} while (!pte_xchg(ptep, __pte(old_pte), __pte(new_pte)));
+
+	/*
+	 * Handle the subpage protection bits
+	 */
+	subpg_pte = new_pte & ~subpg_prot;
+	rflags = htab_convert_pte_flags(subpg_pte, flags);
+
+	if (cpu_has_feature(CPU_FTR_NOEXECUTE) &&
+	    !cpu_has_feature(CPU_FTR_COHERENT_ICACHE)) {
+
+		/*
+		 * No CPU has hugepages but lacks no execute, so we
+		 * don't need to worry about that case
+		 */
+		rflags = hash_page_do_lazy_icache(rflags, __pte(old_pte), trap);
+	}
+
+	subpg_index = (ea & (PAGE_SIZE - 1)) >> shift;
+	vpn  = hpt_vpn(ea, vsid, ssize);
+	rpte = __real_pte(__pte(old_pte), ptep, PTRS_PER_PTE);
+	/*
+	 *None of the sub 4k page is hashed
+	 */
+	if (!(old_pte & H_PAGE_HASHPTE))
+		goto htab_insert_hpte;
+	/*
+	 * Check if the pte was already inserted into the hash table
+	 * as a 64k HW page, and invalidate the 64k HPTE if so.
+	 */
+	if (!(old_pte & H_PAGE_COMBO)) {
+		flush_hash_page(vpn, rpte, MMU_PAGE_64K, ssize, flags);
+		/*
+		 * clear the old slot details from the old and new pte.
+		 * On hash insert failure we use old pte value and we don't
+		 * want slot information there if we have a insert failure.
+		 */
+		old_pte &= ~H_PAGE_HASHPTE;
+		new_pte &= ~H_PAGE_HASHPTE;
+		goto htab_insert_hpte;
+	}
+	/*
+	 * Check for sub page valid and update
+	 */
+	if (__rpte_sub_valid(rpte, subpg_index)) {
+		int ret;
+
+		gslot = pte_get_hash_gslot(vpn, shift, ssize, rpte,
+					   subpg_index);
+		ret = mmu_hash_ops.hpte_updatepp(gslot, rflags, vpn,
+						 MMU_PAGE_4K, MMU_PAGE_4K,
+						 ssize, flags);
+
+		/*
+		 * If we failed because typically the HPTE wasn't really here
+		 * we try an insertion.
+		 */
+		if (ret == -1)
+			goto htab_insert_hpte;
+
+		*ptep = __pte(new_pte & ~H_PAGE_BUSY);
+		return 0;
+	}
+
+htab_insert_hpte:
+
+	/*
+	 * Initialize all hidx entries to invalid value, the first time
+	 * the PTE is about to allocate a 4K HPTE.
+	 */
+	if (!(old_pte & H_PAGE_COMBO))
+		rpte.hidx = INVALID_RPTE_HIDX;
+
+	/*
+	 * handle H_PAGE_4K_PFN case
+	 */
+	if (old_pte & H_PAGE_4K_PFN) {
+		/*
+		 * All the sub 4k page have the same
+		 * physical address.
+		 */
+		pa = pte_pfn(__pte(old_pte)) << HW_PAGE_SHIFT;
+	} else {
+		pa = pte_pfn(__pte(old_pte)) << PAGE_SHIFT;
+		pa += (subpg_index << shift);
+	}
+	hash = hpt_hash(vpn, shift, ssize);
+repeat:
+	hpte_group = (hash & htab_hash_mask) * HPTES_PER_GROUP;
+
+	/* Insert into the hash table, primary slot */
+	slot = mmu_hash_ops.hpte_insert(hpte_group, vpn, pa, rflags, 0,
+					MMU_PAGE_4K, MMU_PAGE_4K, ssize);
+	/*
+	 * Primary is full, try the secondary
+	 */
+	if (unlikely(slot == -1)) {
+		bool soft_invalid;
+
+		hpte_group = (~hash & htab_hash_mask) * HPTES_PER_GROUP;
+		slot = mmu_hash_ops.hpte_insert(hpte_group, vpn, pa,
+						rflags, HPTE_V_SECONDARY,
+						MMU_PAGE_4K, MMU_PAGE_4K,
+						ssize);
+
+		soft_invalid = hpte_soft_invalid(slot);
+		if (unlikely(soft_invalid)) {
+			/*
+			 * We got a valid slot from a hardware point of view.
+			 * but we cannot use it, because we use this special
+			 * value; as defined by hpte_soft_invalid(), to track
+			 * invalid slots. We cannot use it. So invalidate it.
+			 */
+			gslot = slot & _PTEIDX_GROUP_IX;
+			mmu_hash_ops.hpte_invalidate(hpte_group + gslot, vpn,
+						     MMU_PAGE_4K, MMU_PAGE_4K,
+						     ssize, 0);
+		}
+
+		if (unlikely(slot == -1 || soft_invalid)) {
+			/*
+			 * For soft invalid slot, let's ensure that we release a
+			 * slot from the primary, with the hope that we will
+			 * acquire that slot next time we try. This will ensure
+			 * that we do not get the same soft-invalid slot.
+			 */
+			if (soft_invalid || (mftb() & 0x1))
+				hpte_group = (hash & htab_hash_mask) * HPTES_PER_GROUP;
+
+			mmu_hash_ops.hpte_remove(hpte_group);
+			/*
+			 * FIXME!! Should be try the group from which we removed ?
+			 */
+			goto repeat;
+		}
+	}
+	/*
+	 * Hypervisor failure. Restore old pte and return -1
+	 * similar to __hash_page_*
+	 */
+	if (unlikely(slot == -2)) {
+		*ptep = __pte(old_pte);
+		hash_failure_debug(ea, access, vsid, trap, ssize,
+				   MMU_PAGE_4K, MMU_PAGE_4K, old_pte);
+		return -1;
+	}
+
+	new_pte |= pte_set_hidx(ptep, rpte, subpg_index, slot, PTRS_PER_PTE);
+	new_pte |= H_PAGE_HASHPTE;
+
+	if (stress_hpt())
+		hpt_do_stress(ea, hpte_group);
+
+	*ptep = __pte(new_pte & ~H_PAGE_BUSY);
+	return 0;
+}
+
+int __hash_page_64K(unsigned long ea, unsigned long access,
+		    unsigned long vsid, pte_t *ptep, unsigned long trap,
+		    unsigned long flags, int ssize)
+{
+	real_pte_t rpte;
+	unsigned long hpte_group;
+	unsigned long rflags, pa;
+	unsigned long old_pte, new_pte;
+	unsigned long vpn, hash, slot;
+	unsigned long shift = mmu_psize_defs[MMU_PAGE_64K].shift;
+
+	/*
+	 * atomically mark the linux large page PTE busy and dirty
+	 */
+	do {
+		pte_t pte = READ_ONCE(*ptep);
+
+		old_pte = pte_val(pte);
+		/* If PTE busy, retry the access */
+		if (unlikely(old_pte & H_PAGE_BUSY))
+			return 0;
+		/* If PTE permissions don't match, take page fault */
+		if (unlikely(!check_pte_access(access, old_pte)))
+			return 1;
+		/*
+		 * Check if PTE has the cache-inhibit bit set
+		 * If so, bail out and refault as a 4k page
+		 */
+		if (!mmu_has_feature(MMU_FTR_CI_LARGE_PAGE) &&
+		    unlikely(pte_ci(pte)))
+			return 0;
+		/*
+		 * Try to lock the PTE, add ACCESSED and DIRTY if it was
+		 * a write access.
+		 */
+		new_pte = old_pte | H_PAGE_BUSY | _PAGE_ACCESSED;
+		if (access & _PAGE_WRITE)
+			new_pte |= _PAGE_DIRTY;
+	} while (!pte_xchg(ptep, __pte(old_pte), __pte(new_pte)));
+
+	rflags = htab_convert_pte_flags(new_pte, flags);
+	rpte = __real_pte(__pte(old_pte), ptep, PTRS_PER_PTE);
+
+	if (cpu_has_feature(CPU_FTR_NOEXECUTE) &&
+	    !cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
+		rflags = hash_page_do_lazy_icache(rflags, __pte(old_pte), trap);
+
+	vpn  = hpt_vpn(ea, vsid, ssize);
+	if (unlikely(old_pte & H_PAGE_HASHPTE)) {
+		unsigned long gslot;
+
+		/*
+		 * There MIGHT be an HPTE for this pte
+		 */
+		gslot = pte_get_hash_gslot(vpn, shift, ssize, rpte, 0);
+		if (mmu_hash_ops.hpte_updatepp(gslot, rflags, vpn, MMU_PAGE_64K,
+					       MMU_PAGE_64K, ssize,
+					       flags) == -1)
+			old_pte &= ~_PAGE_HPTEFLAGS;
+	}
+
+	if (likely(!(old_pte & H_PAGE_HASHPTE))) {
+
+		pa = pte_pfn(__pte(old_pte)) << PAGE_SHIFT;
+		hash = hpt_hash(vpn, shift, ssize);
+
+repeat:
+		hpte_group = (hash & htab_hash_mask) * HPTES_PER_GROUP;
+
+		/* Insert into the hash table, primary slot */
+		slot = mmu_hash_ops.hpte_insert(hpte_group, vpn, pa, rflags, 0,
+						MMU_PAGE_64K, MMU_PAGE_64K,
+						ssize);
+		/*
+		 * Primary is full, try the secondary
+		 */
+		if (unlikely(slot == -1)) {
+			hpte_group = (~hash & htab_hash_mask) * HPTES_PER_GROUP;
+			slot = mmu_hash_ops.hpte_insert(hpte_group, vpn, pa,
+							rflags,
+							HPTE_V_SECONDARY,
+							MMU_PAGE_64K,
+							MMU_PAGE_64K, ssize);
+			if (slot == -1) {
+				if (mftb() & 0x1)
+					hpte_group = (hash & htab_hash_mask) *
+							HPTES_PER_GROUP;
+				mmu_hash_ops.hpte_remove(hpte_group);
+				/*
+				 * FIXME!! Should be try the group from which we removed ?
+				 */
+				goto repeat;
+			}
+		}
+		/*
+		 * Hypervisor failure. Restore old pte and return -1
+		 * similar to __hash_page_*
+		 */
+		if (unlikely(slot == -2)) {
+			*ptep = __pte(old_pte);
+			hash_failure_debug(ea, access, vsid, trap, ssize,
+					   MMU_PAGE_64K, MMU_PAGE_64K, old_pte);
+			return -1;
+		}
+
+		new_pte = (new_pte & ~_PAGE_HPTEFLAGS) | H_PAGE_HASHPTE;
+		new_pte |= pte_set_hidx(ptep, rpte, 0, slot, PTRS_PER_PTE);
+
+		if (stress_hpt())
+			hpt_do_stress(ea, hpte_group);
+	}
+
+	*ptep = __pte(new_pte & ~H_PAGE_BUSY);
+
+	return 0;
+}

linux-7.2.3/arch/powerpc/mm/book3s64/hash_hugepage.c

@@ -0,0 +1,188 @@
+/*
+ * Copyright IBM Corporation, 2013
+ * Author Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of version 2.1 of the GNU Lesser General Public License
+ * as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it would be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ */
+
+/*
+ * PPC64 THP Support for hash based MMUs
+ */
+#include <linux/mm.h>
+#include <asm/machdep.h>
+
+int __hash_page_thp(unsigned long ea, unsigned long access, unsigned long vsid,
+		    pmd_t *pmdp, unsigned long trap, unsigned long flags,
+		    int ssize, unsigned int psize)
+{
+	unsigned int index, valid;
+	unsigned char *hpte_slot_array;
+	unsigned long rflags, pa, hidx;
+	unsigned long old_pmd, new_pmd;
+	int ret, lpsize = MMU_PAGE_16M;
+	unsigned long vpn, hash, shift, slot;
+
+	/*
+	 * atomically mark the linux large page PMD busy and dirty
+	 */
+	do {
+		pmd_t pmd = READ_ONCE(*pmdp);
+
+		old_pmd = pmd_val(pmd);
+		/* If PMD busy, retry the access */
+		if (unlikely(old_pmd & H_PAGE_BUSY))
+			return 0;
+		/* If PMD permissions don't match, take page fault */
+		if (unlikely(!check_pte_access(access, old_pmd)))
+			return 1;
+		/*
+		 * Try to lock the PTE, add ACCESSED and DIRTY if it was
+		 * a write access
+		 */
+		new_pmd = old_pmd | H_PAGE_BUSY | _PAGE_ACCESSED;
+		if (access & _PAGE_WRITE)
+			new_pmd |= _PAGE_DIRTY;
+	} while (!pmd_xchg(pmdp, __pmd(old_pmd), __pmd(new_pmd)));
+
+	/*
+	 * Make sure this is thp or devmap entry
+	 */
+	if (!(old_pmd & H_PAGE_THP_HUGE))
+		return 0;
+
+	rflags = htab_convert_pte_flags(new_pmd, flags);
+
+	/*
+	 * THPs are only supported on platforms that can do mixed page size
+	 * segments (MPSS) and all such platforms have coherent icache. Hence we
+	 * don't need to do lazy icache flush (hash_page_do_lazy_icache()) on
+	 * noexecute fault.
+	 */
+
+	/*
+	 * Find the slot index details for this ea, using base page size.
+	 */
+	shift = mmu_psize_defs[psize].shift;
+	index = (ea & ~HPAGE_PMD_MASK) >> shift;
+	BUG_ON(index >= PTE_FRAG_SIZE);
+
+	vpn = hpt_vpn(ea, vsid, ssize);
+	hpte_slot_array = get_hpte_slot_array(pmdp);
+	if (psize == MMU_PAGE_4K) {
+		/*
+		 * invalidate the old hpte entry if we have that mapped via 64K
+		 * base page size. This is because demote_segment won't flush
+		 * hash page table entries.
+		 */
+		if ((old_pmd & H_PAGE_HASHPTE) && !(old_pmd & H_PAGE_COMBO)) {
+			flush_hash_hugepage(vsid, ea, pmdp, MMU_PAGE_64K,
+					    ssize, flags);
+			/*
+			 * With THP, we also clear the slot information with
+			 * respect to all the 64K hash pte mapping the 16MB
+			 * page. They are all invalid now. This make sure we
+			 * don't find the slot valid when we fault with 4k
+			 * base page size.
+			 *
+			 */
+			memset(hpte_slot_array, 0, PTE_FRAG_SIZE);
+		}
+	}
+
+	valid = hpte_valid(hpte_slot_array, index);
+	if (valid) {
+		/* update the hpte bits */
+		hash = hpt_hash(vpn, shift, ssize);
+		hidx =  hpte_hash_index(hpte_slot_array, index);
+		if (hidx & _PTEIDX_SECONDARY)
+			hash = ~hash;
+		slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
+		slot += hidx & _PTEIDX_GROUP_IX;
+
+		ret = mmu_hash_ops.hpte_updatepp(slot, rflags, vpn,
+						 psize, lpsize, ssize, flags);
+		/*
+		 * We failed to update, try to insert a new entry.
+		 */
+		if (ret == -1) {
+			/*
+			 * large pte is marked busy, so we can be sure
+			 * nobody is looking at hpte_slot_array. hence we can
+			 * safely update this here.
+			 */
+			valid = 0;
+			hpte_slot_array[index] = 0;
+		}
+	}
+
+	if (!valid) {
+		unsigned long hpte_group;
+
+		hash = hpt_hash(vpn, shift, ssize);
+		/* insert new entry */
+		pa = pmd_pfn(__pmd(old_pmd)) << PAGE_SHIFT;
+		new_pmd |= H_PAGE_HASHPTE;
+
+repeat:
+		hpte_group = (hash & htab_hash_mask) * HPTES_PER_GROUP;
+
+		/* Insert into the hash table, primary slot */
+		slot = mmu_hash_ops.hpte_insert(hpte_group, vpn, pa, rflags, 0,
+						psize, lpsize, ssize);
+		/*
+		 * Primary is full, try the secondary
+		 */
+		if (unlikely(slot == -1)) {
+			hpte_group = (~hash & htab_hash_mask) * HPTES_PER_GROUP;
+			slot = mmu_hash_ops.hpte_insert(hpte_group, vpn, pa,
+							rflags,
+							HPTE_V_SECONDARY,
+							psize, lpsize, ssize);
+			if (slot == -1) {
+				if (mftb() & 0x1)
+					hpte_group = (hash & htab_hash_mask) *
+							HPTES_PER_GROUP;
+
+				mmu_hash_ops.hpte_remove(hpte_group);
+				goto repeat;
+			}
+		}
+		/*
+		 * Hypervisor failure. Restore old pmd and return -1
+		 * similar to __hash_page_*
+		 */
+		if (unlikely(slot == -2)) {
+			*pmdp = __pmd(old_pmd);
+			hash_failure_debug(ea, access, vsid, trap, ssize,
+					   psize, lpsize, old_pmd);
+			return -1;
+		}
+		/*
+		 * large pte is marked busy, so we can be sure
+		 * nobody is looking at hpte_slot_array. hence we can
+		 * safely update this here.
+		 */
+		mark_hpte_slot_valid(hpte_slot_array, index, slot);
+	}
+	/*
+	 * Mark the pte with H_PAGE_COMBO, if we are trying to hash it with
+	 * base page size 4k.
+	 */
+	if (psize == MMU_PAGE_4K)
+		new_pmd |= H_PAGE_COMBO;
+	/*
+	 * The hpte valid is stored in the pgtable whose address is in the
+	 * second half of the PMD. Order this against clearing of the busy bit in
+	 * huge pmd.
+	 */
+	smp_wmb();
+	*pmdp = __pmd(new_pmd & ~H_PAGE_BUSY);
+	return 0;
+}

linux-7.2.3/arch/powerpc/mm/book3s64/hash_native.c

@@ -0,0 +1,875 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * native hashtable management.
+ *
+ * SMP scalability work:
+ *    Copyright (C) 2001 Anton Blanchard <anton@au.ibm.com>, IBM
+ */
+
+#undef DEBUG_LOW
+
+#include <linux/spinlock.h>
+#include <linux/bitops.h>
+#include <linux/of.h>
+#include <linux/processor.h>
+#include <linux/threads.h>
+#include <linux/smp.h>
+#include <linux/pgtable.h>
+
+#include <asm/machdep.h>
+#include <asm/mmu.h>
+#include <asm/mmu_context.h>
+#include <asm/trace.h>
+#include <asm/tlb.h>
+#include <asm/cputable.h>
+#include <asm/udbg.h>
+#include <asm/kexec.h>
+#include <asm/ppc-opcode.h>
+#include <asm/feature-fixups.h>
+
+#ifdef DEBUG_LOW
+#define DBG_LOW(fmt...) udbg_printf(fmt)
+#else
+#define DBG_LOW(fmt...)
+#endif
+
+#ifdef __BIG_ENDIAN__
+#define HPTE_LOCK_BIT 3
+#else
+#define HPTE_LOCK_BIT (56+3)
+#endif
+
+static DEFINE_RAW_SPINLOCK(native_tlbie_lock);
+
+#ifdef CONFIG_LOCKDEP
+static struct lockdep_map hpte_lock_map =
+	STATIC_LOCKDEP_MAP_INIT("hpte_lock", &hpte_lock_map);
+
+static void acquire_hpte_lock(void)
+{
+	lock_map_acquire(&hpte_lock_map);
+}
+
+static void release_hpte_lock(void)
+{
+	lock_map_release(&hpte_lock_map);
+}
+#else
+static void acquire_hpte_lock(void)
+{
+}
+
+static void release_hpte_lock(void)
+{
+}
+#endif
+
+static inline unsigned long  ___tlbie(unsigned long vpn, int psize,
+						int apsize, int ssize)
+{
+	unsigned long va;
+	unsigned int penc;
+	unsigned long sllp;
+
+	/*
+	 * We need 14 to 65 bits of va for a tlibe of 4K page
+	 * With vpn we ignore the lower VPN_SHIFT bits already.
+	 * And top two bits are already ignored because we can
+	 * only accomodate 76 bits in a 64 bit vpn with a VPN_SHIFT
+	 * of 12.
+	 */
+	va = vpn << VPN_SHIFT;
+	/*
+	 * clear top 16 bits of 64bit va, non SLS segment
+	 * Older versions of the architecture (2.02 and earler) require the
+	 * masking of the top 16 bits.
+	 */
+	if (mmu_has_feature(MMU_FTR_TLBIE_CROP_VA))
+		va &= ~(0xffffULL << 48);
+
+	switch (psize) {
+	case MMU_PAGE_4K:
+		/* clear out bits after (52) [0....52.....63] */
+		va &= ~((1ul << (64 - 52)) - 1);
+		va |= ssize << 8;
+		sllp = get_sllp_encoding(apsize);
+		va |= sllp << 5;
+		asm volatile(ASM_FTR_IFCLR("tlbie %0,0", PPC_TLBIE(%1,%0), %2)
+			     : : "r" (va), "r"(0), "i" (CPU_FTR_ARCH_206)
+			     : "memory");
+		break;
+	default:
+		/* We need 14 to 14 + i bits of va */
+		penc = mmu_psize_defs[psize].penc[apsize];
+		va &= ~((1ul << mmu_psize_defs[apsize].shift) - 1);
+		va |= penc << 12;
+		va |= ssize << 8;
+		/*
+		 * AVAL bits:
+		 * We don't need all the bits, but rest of the bits
+		 * must be ignored by the processor.
+		 * vpn cover upto 65 bits of va. (0...65) and we need
+		 * 58..64 bits of va.
+		 */
+		va |= (vpn & 0xfe); /* AVAL */
+		va |= 1; /* L */
+		asm volatile(ASM_FTR_IFCLR("tlbie %0,1", PPC_TLBIE(%1,%0), %2)
+			     : : "r" (va), "r"(0), "i" (CPU_FTR_ARCH_206)
+			     : "memory");
+		break;
+	}
+	return va;
+}
+
+static inline void fixup_tlbie_vpn(unsigned long vpn, int psize,
+				   int apsize, int ssize)
+{
+	if (cpu_has_feature(CPU_FTR_P9_TLBIE_ERAT_BUG)) {
+		/* Radix flush for a hash guest */
+
+		unsigned long rb,rs,prs,r,ric;
+
+		rb = PPC_BIT(52); /* IS = 2 */
+		rs = 0;  /* lpid = 0 */
+		prs = 0; /* partition scoped */
+		r = 1;   /* radix format */
+		ric = 0; /* RIC_FLSUH_TLB */
+
+		/*
+		 * Need the extra ptesync to make sure we don't
+		 * re-order the tlbie
+		 */
+		asm volatile("ptesync": : :"memory");
+		asm volatile(PPC_TLBIE_5(%0, %4, %3, %2, %1)
+			     : : "r"(rb), "i"(r), "i"(prs),
+			       "i"(ric), "r"(rs) : "memory");
+	}
+
+
+	if (cpu_has_feature(CPU_FTR_P9_TLBIE_STQ_BUG)) {
+		/* Need the extra ptesync to ensure we don't reorder tlbie*/
+		asm volatile("ptesync": : :"memory");
+		___tlbie(vpn, psize, apsize, ssize);
+	}
+}
+
+static inline void __tlbie(unsigned long vpn, int psize, int apsize, int ssize)
+{
+	unsigned long rb;
+
+	rb = ___tlbie(vpn, psize, apsize, ssize);
+	trace_tlbie(0, 0, rb, 0, 0, 0, 0);
+}
+
+static inline void __tlbiel(unsigned long vpn, int psize, int apsize, int ssize)
+{
+	unsigned long va;
+	unsigned int penc;
+	unsigned long sllp;
+
+	/* VPN_SHIFT can be atmost 12 */
+	va = vpn << VPN_SHIFT;
+	/*
+	 * clear top 16 bits of 64 bit va, non SLS segment
+	 * Older versions of the architecture (2.02 and earler) require the
+	 * masking of the top 16 bits.
+	 */
+	if (mmu_has_feature(MMU_FTR_TLBIE_CROP_VA))
+		va &= ~(0xffffULL << 48);
+
+	switch (psize) {
+	case MMU_PAGE_4K:
+		/* clear out bits after(52) [0....52.....63] */
+		va &= ~((1ul << (64 - 52)) - 1);
+		va |= ssize << 8;
+		sllp = get_sllp_encoding(apsize);
+		va |= sllp << 5;
+		asm volatile(ASM_FTR_IFSET("tlbiel %0", PPC_TLBIEL_v205(%0, 0), %1)
+			     : : "r" (va), "i" (CPU_FTR_ARCH_206)
+			     : "memory");
+		break;
+	default:
+		/* We need 14 to 14 + i bits of va */
+		penc = mmu_psize_defs[psize].penc[apsize];
+		va &= ~((1ul << mmu_psize_defs[apsize].shift) - 1);
+		va |= penc << 12;
+		va |= ssize << 8;
+		/*
+		 * AVAL bits:
+		 * We don't need all the bits, but rest of the bits
+		 * must be ignored by the processor.
+		 * vpn cover upto 65 bits of va. (0...65) and we need
+		 * 58..64 bits of va.
+		 */
+		va |= (vpn & 0xfe);
+		va |= 1; /* L */
+		asm volatile(ASM_FTR_IFSET("tlbiel %0", PPC_TLBIEL_v205(%0, 1), %1)
+			     : : "r" (va), "i" (CPU_FTR_ARCH_206)
+			     : "memory");
+		break;
+	}
+	trace_tlbie(0, 1, va, 0, 0, 0, 0);
+
+}
+
+static inline void tlbie(unsigned long vpn, int psize, int apsize,
+			 int ssize, int local)
+{
+	unsigned int use_local = local && mmu_has_feature(MMU_FTR_TLBIEL);
+	int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
+
+	if (use_local)
+		use_local = mmu_psize_defs[psize].tlbiel;
+	if (lock_tlbie && !use_local)
+		raw_spin_lock(&native_tlbie_lock);
+	asm volatile("ptesync": : :"memory");
+	if (use_local) {
+		__tlbiel(vpn, psize, apsize, ssize);
+		ppc_after_tlbiel_barrier();
+	} else {
+		__tlbie(vpn, psize, apsize, ssize);
+		fixup_tlbie_vpn(vpn, psize, apsize, ssize);
+		asm volatile("eieio; tlbsync; ptesync": : :"memory");
+	}
+	if (lock_tlbie && !use_local)
+		raw_spin_unlock(&native_tlbie_lock);
+}
+
+static inline void native_lock_hpte(struct hash_pte *hptep)
+{
+	unsigned long *word = (unsigned long *)&hptep->v;
+
+	acquire_hpte_lock();
+	while (1) {
+		if (!test_and_set_bit_lock(HPTE_LOCK_BIT, word))
+			break;
+		spin_begin();
+		while(test_bit(HPTE_LOCK_BIT, word))
+			spin_cpu_relax();
+		spin_end();
+	}
+}
+
+static inline void native_unlock_hpte(struct hash_pte *hptep)
+{
+	unsigned long *word = (unsigned long *)&hptep->v;
+
+	release_hpte_lock();
+	clear_bit_unlock(HPTE_LOCK_BIT, word);
+}
+
+static long native_hpte_insert(unsigned long hpte_group, unsigned long vpn,
+			unsigned long pa, unsigned long rflags,
+			unsigned long vflags, int psize, int apsize, int ssize)
+{
+	struct hash_pte *hptep = htab_address + hpte_group;
+	unsigned long hpte_v, hpte_r;
+	unsigned long flags;
+	int i;
+
+	local_irq_save(flags);
+
+	if (!(vflags & HPTE_V_BOLTED)) {
+		DBG_LOW("    insert(group=%lx, vpn=%016lx, pa=%016lx,"
+			" rflags=%lx, vflags=%lx, psize=%d)\n",
+			hpte_group, vpn, pa, rflags, vflags, psize);
+	}
+
+	for (i = 0; i < HPTES_PER_GROUP; i++) {
+		if (! (be64_to_cpu(hptep->v) & HPTE_V_VALID)) {
+			/* retry with lock held */
+			native_lock_hpte(hptep);
+			if (! (be64_to_cpu(hptep->v) & HPTE_V_VALID))
+				break;
+			native_unlock_hpte(hptep);
+		}
+
+		hptep++;
+	}
+
+	if (i == HPTES_PER_GROUP) {
+		local_irq_restore(flags);
+		return -1;
+	}
+
+	hpte_v = hpte_encode_v(vpn, psize, apsize, ssize) | vflags | HPTE_V_VALID;
+	hpte_r = hpte_encode_r(pa, psize, apsize) | rflags;
+
+	if (!(vflags & HPTE_V_BOLTED)) {
+		DBG_LOW(" i=%x hpte_v=%016lx, hpte_r=%016lx\n",
+			i, hpte_v, hpte_r);
+	}
+
+	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
+		hpte_r = hpte_old_to_new_r(hpte_v, hpte_r);
+		hpte_v = hpte_old_to_new_v(hpte_v);
+	}
+
+	hptep->r = cpu_to_be64(hpte_r);
+	/* Guarantee the second dword is visible before the valid bit */
+	eieio();
+	/*
+	 * Now set the first dword including the valid bit
+	 * NOTE: this also unlocks the hpte
+	 */
+	release_hpte_lock();
+	hptep->v = cpu_to_be64(hpte_v);
+
+	__asm__ __volatile__ ("ptesync" : : : "memory");
+
+	local_irq_restore(flags);
+
+	return i | (!!(vflags & HPTE_V_SECONDARY) << 3);
+}
+
+static long native_hpte_remove(unsigned long hpte_group)
+{
+	unsigned long hpte_v, flags;
+	struct hash_pte *hptep;
+	int i;
+	int slot_offset;
+
+	local_irq_save(flags);
+
+	DBG_LOW("    remove(group=%lx)\n", hpte_group);
+
+	/* pick a random entry to start at */
+	slot_offset = mftb() & 0x7;
+
+	for (i = 0; i < HPTES_PER_GROUP; i++) {
+		hptep = htab_address + hpte_group + slot_offset;
+		hpte_v = be64_to_cpu(hptep->v);
+
+		if ((hpte_v & HPTE_V_VALID) && !(hpte_v & HPTE_V_BOLTED)) {
+			/* retry with lock held */
+			native_lock_hpte(hptep);
+			hpte_v = be64_to_cpu(hptep->v);
+			if ((hpte_v & HPTE_V_VALID)
+			    && !(hpte_v & HPTE_V_BOLTED))
+				break;
+			native_unlock_hpte(hptep);
+		}
+
+		slot_offset++;
+		slot_offset &= 0x7;
+	}
+
+	if (i == HPTES_PER_GROUP) {
+		i = -1;
+		goto out;
+	}
+
+	/* Invalidate the hpte. NOTE: this also unlocks it */
+	release_hpte_lock();
+	hptep->v = 0;
+out:
+	local_irq_restore(flags);
+	return i;
+}
+
+static long native_hpte_updatepp(unsigned long slot, unsigned long newpp,
+				 unsigned long vpn, int bpsize,
+				 int apsize, int ssize, unsigned long flags)
+{
+	struct hash_pte *hptep = htab_address + slot;
+	unsigned long hpte_v, want_v;
+	int ret = 0, local = 0;
+	unsigned long irqflags;
+
+	local_irq_save(irqflags);
+
+	want_v = hpte_encode_avpn(vpn, bpsize, ssize);
+
+	DBG_LOW("    update(vpn=%016lx, avpnv=%016lx, group=%lx, newpp=%lx)",
+		vpn, want_v & HPTE_V_AVPN, slot, newpp);
+
+	hpte_v = hpte_get_old_v(hptep);
+	/*
+	 * We need to invalidate the TLB always because hpte_remove doesn't do
+	 * a tlb invalidate. If a hash bucket gets full, we "evict" a more/less
+	 * random entry from it. When we do that we don't invalidate the TLB
+	 * (hpte_remove) because we assume the old translation is still
+	 * technically "valid".
+	 */
+	if (!HPTE_V_COMPARE(hpte_v, want_v) || !(hpte_v & HPTE_V_VALID)) {
+		DBG_LOW(" -> miss\n");
+		ret = -1;
+	} else {
+		native_lock_hpte(hptep);
+		/* recheck with locks held */
+		hpte_v = hpte_get_old_v(hptep);
+		if (unlikely(!HPTE_V_COMPARE(hpte_v, want_v) ||
+			     !(hpte_v & HPTE_V_VALID))) {
+			ret = -1;
+		} else {
+			DBG_LOW(" -> hit\n");
+			/* Update the HPTE */
+			hptep->r = cpu_to_be64((be64_to_cpu(hptep->r) &
+						~(HPTE_R_PPP | HPTE_R_N)) |
+					       (newpp & (HPTE_R_PPP | HPTE_R_N |
+							 HPTE_R_C)));
+		}
+		native_unlock_hpte(hptep);
+	}
+
+	if (flags & HPTE_LOCAL_UPDATE)
+		local = 1;
+	/*
+	 * Ensure it is out of the tlb too if it is not a nohpte fault
+	 */
+	if (!(flags & HPTE_NOHPTE_UPDATE))
+		tlbie(vpn, bpsize, apsize, ssize, local);
+
+	local_irq_restore(irqflags);
+
+	return ret;
+}
+
+static long __native_hpte_find(unsigned long want_v, unsigned long slot)
+{
+	struct hash_pte *hptep;
+	unsigned long hpte_v;
+	unsigned long i;
+
+	for (i = 0; i < HPTES_PER_GROUP; i++) {
+
+		hptep = htab_address + slot;
+		hpte_v = hpte_get_old_v(hptep);
+		if (HPTE_V_COMPARE(hpte_v, want_v) && (hpte_v & HPTE_V_VALID))
+			/* HPTE matches */
+			return slot;
+		++slot;
+	}
+
+	return -1;
+}
+
+static long native_hpte_find(unsigned long vpn, int psize, int ssize)
+{
+	unsigned long hpte_group;
+	unsigned long want_v;
+	unsigned long hash;
+	long slot;
+
+	hash = hpt_hash(vpn, mmu_psize_defs[psize].shift, ssize);
+	want_v = hpte_encode_avpn(vpn, psize, ssize);
+
+	/*
+	 * We try to keep bolted entries always in primary hash
+	 * But in some case we can find them in secondary too.
+	 */
+	hpte_group = (hash & htab_hash_mask) * HPTES_PER_GROUP;
+	slot = __native_hpte_find(want_v, hpte_group);
+	if (slot < 0) {
+		/* Try in secondary */
+		hpte_group = (~hash & htab_hash_mask) * HPTES_PER_GROUP;
+		slot = __native_hpte_find(want_v, hpte_group);
+		if (slot < 0)
+			return -1;
+	}
+
+	return slot;
+}
+
+/*
+ * Update the page protection bits. Intended to be used to create
+ * guard pages for kernel data structures on pages which are bolted
+ * in the HPT. Assumes pages being operated on will not be stolen.
+ *
+ * No need to lock here because we should be the only user.
+ */
+static void native_hpte_updateboltedpp(unsigned long newpp, unsigned long ea,
+				       int psize, int ssize)
+{
+	unsigned long vpn;
+	unsigned long vsid;
+	long slot;
+	struct hash_pte *hptep;
+	unsigned long flags;
+
+	local_irq_save(flags);
+
+	vsid = get_kernel_vsid(ea, ssize);
+	vpn = hpt_vpn(ea, vsid, ssize);
+
+	slot = native_hpte_find(vpn, psize, ssize);
+	if (slot == -1)
+		panic("could not find page to bolt\n");
+	hptep = htab_address + slot;
+
+	/* Update the HPTE */
+	hptep->r = cpu_to_be64((be64_to_cpu(hptep->r) &
+				~(HPTE_R_PPP | HPTE_R_N)) |
+			       (newpp & (HPTE_R_PPP | HPTE_R_N)));
+	/*
+	 * Ensure it is out of the tlb too. Bolted entries base and
+	 * actual page size will be same.
+	 */
+	tlbie(vpn, psize, psize, ssize, 0);
+
+	local_irq_restore(flags);
+}
+
+/*
+ * Remove a bolted kernel entry. Memory hotplug uses this.
+ *
+ * No need to lock here because we should be the only user.
+ */
+static int native_hpte_removebolted(unsigned long ea, int psize, int ssize)
+{
+	unsigned long vpn;
+	unsigned long vsid;
+	long slot;
+	struct hash_pte *hptep;
+	unsigned long flags;
+
+	local_irq_save(flags);
+
+	vsid = get_kernel_vsid(ea, ssize);
+	vpn = hpt_vpn(ea, vsid, ssize);
+
+	slot = native_hpte_find(vpn, psize, ssize);
+	if (slot == -1)
+		return -ENOENT;
+
+	hptep = htab_address + slot;
+
+	VM_WARN_ON(!(be64_to_cpu(hptep->v) & HPTE_V_BOLTED));
+
+	/* Invalidate the hpte */
+	hptep->v = 0;
+
+	/* Invalidate the TLB */
+	tlbie(vpn, psize, psize, ssize, 0);
+
+	local_irq_restore(flags);
+
+	return 0;
+}
+
+
+static void native_hpte_invalidate(unsigned long slot, unsigned long vpn,
+				   int bpsize, int apsize, int ssize, int local)
+{
+	struct hash_pte *hptep = htab_address + slot;
+	unsigned long hpte_v;
+	unsigned long want_v;
+	unsigned long flags;
+
+	local_irq_save(flags);
+
+	DBG_LOW("    invalidate(vpn=%016lx, hash: %lx)\n", vpn, slot);
+
+	want_v = hpte_encode_avpn(vpn, bpsize, ssize);
+	hpte_v = hpte_get_old_v(hptep);
+
+	if (HPTE_V_COMPARE(hpte_v, want_v) && (hpte_v & HPTE_V_VALID)) {
+		native_lock_hpte(hptep);
+		/* recheck with locks held */
+		hpte_v = hpte_get_old_v(hptep);
+
+		if (HPTE_V_COMPARE(hpte_v, want_v) && (hpte_v & HPTE_V_VALID)) {
+			/* Invalidate the hpte. NOTE: this also unlocks it */
+			release_hpte_lock();
+			hptep->v = 0;
+		} else
+			native_unlock_hpte(hptep);
+	}
+	/*
+	 * We need to invalidate the TLB always because hpte_remove doesn't do
+	 * a tlb invalidate. If a hash bucket gets full, we "evict" a more/less
+	 * random entry from it. When we do that we don't invalidate the TLB
+	 * (hpte_remove) because we assume the old translation is still
+	 * technically "valid".
+	 */
+	tlbie(vpn, bpsize, apsize, ssize, local);
+
+	local_irq_restore(flags);
+}
+
+#ifdef CONFIG_TRANSPARENT_HUGEPAGE
+static void native_hugepage_invalidate(unsigned long vsid,
+				       unsigned long addr,
+				       unsigned char *hpte_slot_array,
+				       int psize, int ssize, int local)
+{
+	int i;
+	struct hash_pte *hptep;
+	int actual_psize = MMU_PAGE_16M;
+	unsigned int max_hpte_count, valid;
+	unsigned long flags, s_addr = addr;
+	unsigned long hpte_v, want_v, shift;
+	unsigned long hidx, vpn = 0, hash, slot;
+
+	shift = mmu_psize_defs[psize].shift;
+	max_hpte_count = 1U << (PMD_SHIFT - shift);
+
+	local_irq_save(flags);
+	for (i = 0; i < max_hpte_count; i++) {
+		valid = hpte_valid(hpte_slot_array, i);
+		if (!valid)
+			continue;
+		hidx =  hpte_hash_index(hpte_slot_array, i);
+
+		/* get the vpn */
+		addr = s_addr + (i * (1ul << shift));
+		vpn = hpt_vpn(addr, vsid, ssize);
+		hash = hpt_hash(vpn, shift, ssize);
+		if (hidx & _PTEIDX_SECONDARY)
+			hash = ~hash;
+
+		slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
+		slot += hidx & _PTEIDX_GROUP_IX;
+
+		hptep = htab_address + slot;
+		want_v = hpte_encode_avpn(vpn, psize, ssize);
+		hpte_v = hpte_get_old_v(hptep);
+
+		/* Even if we miss, we need to invalidate the TLB */
+		if (HPTE_V_COMPARE(hpte_v, want_v) && (hpte_v & HPTE_V_VALID)) {
+			/* recheck with locks held */
+			native_lock_hpte(hptep);
+			hpte_v = hpte_get_old_v(hptep);
+
+			if (HPTE_V_COMPARE(hpte_v, want_v) && (hpte_v & HPTE_V_VALID)) {
+				/* Invalidate the hpte. NOTE: this also unlocks it */
+				release_hpte_lock();
+				hptep->v = 0;
+			} else
+				native_unlock_hpte(hptep);
+		}
+		/*
+		 * We need to do tlb invalidate for all the address, tlbie
+		 * instruction compares entry_VA in tlb with the VA specified
+		 * here
+		 */
+		tlbie(vpn, psize, actual_psize, ssize, local);
+	}
+	local_irq_restore(flags);
+}
+#else
+static void native_hugepage_invalidate(unsigned long vsid,
+				       unsigned long addr,
+				       unsigned char *hpte_slot_array,
+				       int psize, int ssize, int local)
+{
+	WARN(1, "%s called without THP support\n", __func__);
+}
+#endif
+
+static void hpte_decode(struct hash_pte *hpte, unsigned long slot,
+			int *psize, int *apsize, int *ssize, unsigned long *vpn)
+{
+	unsigned long avpn, pteg, vpi;
+	unsigned long hpte_v = be64_to_cpu(hpte->v);
+	unsigned long hpte_r = be64_to_cpu(hpte->r);
+	unsigned long vsid, seg_off;
+	int size, a_size, shift;
+	/* Look at the 8 bit LP value */
+	unsigned int lp = (hpte_r >> LP_SHIFT) & ((1 << LP_BITS) - 1);
+
+	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
+		hpte_v = hpte_new_to_old_v(hpte_v, hpte_r);
+		hpte_r = hpte_new_to_old_r(hpte_r);
+	}
+	if (!(hpte_v & HPTE_V_LARGE)) {
+		size   = MMU_PAGE_4K;
+		a_size = MMU_PAGE_4K;
+	} else {
+		size = hpte_page_sizes[lp] & 0xf;
+		a_size = hpte_page_sizes[lp] >> 4;
+	}
+	/* This works for all page sizes, and for 256M and 1T segments */
+	*ssize = hpte_v >> HPTE_V_SSIZE_SHIFT;
+	shift = mmu_psize_defs[size].shift;
+
+	avpn = (HPTE_V_AVPN_VAL(hpte_v) & ~mmu_psize_defs[size].avpnm);
+	pteg = slot / HPTES_PER_GROUP;
+	if (hpte_v & HPTE_V_SECONDARY)
+		pteg = ~pteg;
+
+	switch (*ssize) {
+	case MMU_SEGSIZE_256M:
+		/* We only have 28 - 23 bits of seg_off in avpn */
+		seg_off = (avpn & 0x1f) << 23;
+		vsid    =  avpn >> 5;
+		/* We can find more bits from the pteg value */
+		if (shift < 23) {
+			vpi = (vsid ^ pteg) & htab_hash_mask;
+			seg_off |= vpi << shift;
+		}
+		*vpn = vsid << (SID_SHIFT - VPN_SHIFT) | seg_off >> VPN_SHIFT;
+		break;
+	case MMU_SEGSIZE_1T:
+		/* We only have 40 - 23 bits of seg_off in avpn */
+		seg_off = (avpn & 0x1ffff) << 23;
+		vsid    = avpn >> 17;
+		if (shift < 23) {
+			vpi = (vsid ^ (vsid << 25) ^ pteg) & htab_hash_mask;
+			seg_off |= vpi << shift;
+		}
+		*vpn = vsid << (SID_SHIFT_1T - VPN_SHIFT) | seg_off >> VPN_SHIFT;
+		break;
+	default:
+		*vpn = size = 0;
+	}
+	*psize  = size;
+	*apsize = a_size;
+}
+
+/*
+ * clear all mappings on kexec.  All cpus are in real mode (or they will
+ * be when they isi), and we are the only one left.  We rely on our kernel
+ * mapping being 0xC0's and the hardware ignoring those two real bits.
+ *
+ * This must be called with interrupts disabled.
+ *
+ * Taking the native_tlbie_lock is unsafe here due to the possibility of
+ * lockdep being on. On pre POWER5 hardware, not taking the lock could
+ * cause deadlock. POWER5 and newer not taking the lock is fine. This only
+ * gets called during boot before secondary CPUs have come up and during
+ * crashdump and all bets are off anyway.
+ *
+ * TODO: add batching support when enabled.  remember, no dynamic memory here,
+ * although there is the control page available...
+ */
+static notrace void native_hpte_clear(void)
+{
+	unsigned long vpn = 0;
+	unsigned long slot, slots;
+	struct hash_pte *hptep = htab_address;
+	unsigned long hpte_v;
+	unsigned long pteg_count;
+	int psize, apsize, ssize;
+
+	pteg_count = htab_hash_mask + 1;
+
+	slots = pteg_count * HPTES_PER_GROUP;
+
+	for (slot = 0; slot < slots; slot++, hptep++) {
+		/*
+		 * we could lock the pte here, but we are the only cpu
+		 * running,  right?  and for crash dump, we probably
+		 * don't want to wait for a maybe bad cpu.
+		 */
+		hpte_v = be64_to_cpu(hptep->v);
+
+		/*
+		 * Call __tlbie() here rather than tlbie() since we can't take the
+		 * native_tlbie_lock.
+		 */
+		if (hpte_v & HPTE_V_VALID) {
+			hpte_decode(hptep, slot, &psize, &apsize, &ssize, &vpn);
+			hptep->v = 0;
+			___tlbie(vpn, psize, apsize, ssize);
+		}
+	}
+
+	asm volatile("eieio; tlbsync; ptesync":::"memory");
+}
+
+/*
+ * Batched hash table flush, we batch the tlbie's to avoid taking/releasing
+ * the lock all the time
+ */
+static void native_flush_hash_range(unsigned long number, int local)
+{
+	unsigned long vpn = 0;
+	unsigned long hash, index, hidx, shift, slot;
+	struct hash_pte *hptep;
+	unsigned long hpte_v;
+	unsigned long want_v;
+	unsigned long flags;
+	real_pte_t pte;
+	struct ppc64_tlb_batch *batch = this_cpu_ptr(&ppc64_tlb_batch);
+	unsigned long psize = batch->psize;
+	int ssize = batch->ssize;
+	int i;
+
+	local_irq_save(flags);
+
+	for (i = 0; i < number; i++) {
+		vpn = batch->vpn[i];
+		pte = batch->pte[i];
+
+		pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) {
+			hash = hpt_hash(vpn, shift, ssize);
+			hidx = __rpte_to_hidx(pte, index);
+			if (hidx & _PTEIDX_SECONDARY)
+				hash = ~hash;
+			slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
+			slot += hidx & _PTEIDX_GROUP_IX;
+			hptep = htab_address + slot;
+			want_v = hpte_encode_avpn(vpn, psize, ssize);
+			hpte_v = hpte_get_old_v(hptep);
+
+			if (!HPTE_V_COMPARE(hpte_v, want_v) || !(hpte_v & HPTE_V_VALID))
+				continue;
+			/* lock and try again */
+			native_lock_hpte(hptep);
+			hpte_v = hpte_get_old_v(hptep);
+
+			if (!HPTE_V_COMPARE(hpte_v, want_v) || !(hpte_v & HPTE_V_VALID))
+				native_unlock_hpte(hptep);
+			else {
+				release_hpte_lock();
+				hptep->v = 0;
+			}
+
+		} pte_iterate_hashed_end();
+	}
+
+	if (mmu_has_feature(MMU_FTR_TLBIEL) &&
+	    mmu_psize_defs[psize].tlbiel && local) {
+		asm volatile("ptesync":::"memory");
+		for (i = 0; i < number; i++) {
+			vpn = batch->vpn[i];
+			pte = batch->pte[i];
+
+			pte_iterate_hashed_subpages(pte, psize,
+						    vpn, index, shift) {
+				__tlbiel(vpn, psize, psize, ssize);
+			} pte_iterate_hashed_end();
+		}
+		ppc_after_tlbiel_barrier();
+	} else {
+		int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
+
+		if (lock_tlbie)
+			raw_spin_lock(&native_tlbie_lock);
+
+		asm volatile("ptesync":::"memory");
+		for (i = 0; i < number; i++) {
+			vpn = batch->vpn[i];
+			pte = batch->pte[i];
+
+			pte_iterate_hashed_subpages(pte, psize,
+						    vpn, index, shift) {
+				__tlbie(vpn, psize, psize, ssize);
+			} pte_iterate_hashed_end();
+		}
+		/*
+		 * Just do one more with the last used values.
+		 */
+		fixup_tlbie_vpn(vpn, psize, psize, ssize);
+		asm volatile("eieio; tlbsync; ptesync":::"memory");
+
+		if (lock_tlbie)
+			raw_spin_unlock(&native_tlbie_lock);
+	}
+
+	local_irq_restore(flags);
+}
+
+void __init hpte_init_native(void)
+{
+	mmu_hash_ops.hpte_invalidate	= native_hpte_invalidate;
+	mmu_hash_ops.hpte_updatepp	= native_hpte_updatepp;
+	mmu_hash_ops.hpte_updateboltedpp = native_hpte_updateboltedpp;
+	mmu_hash_ops.hpte_removebolted = native_hpte_removebolted;
+	mmu_hash_ops.hpte_insert	= native_hpte_insert;
+	mmu_hash_ops.hpte_remove	= native_hpte_remove;
+	mmu_hash_ops.hpte_clear_all	= native_hpte_clear;
+	mmu_hash_ops.flush_hash_range = native_flush_hash_range;
+	mmu_hash_ops.hugepage_invalidate   = native_hugepage_invalidate;
+}

linux-7.2.3/arch/powerpc/mm/book3s64/hash_pgtable.c

@@ -0,0 +1,588 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Copyright 2005, Paul Mackerras, IBM Corporation.
+ * Copyright 2009, Benjamin Herrenschmidt, IBM Corporation.
+ * Copyright 2015-2016, Aneesh Kumar K.V, IBM Corporation.
+ */
+
+#include <linux/sched.h>
+#include <linux/mm_types.h>
+#include <linux/mm.h>
+#include <linux/page_table_check.h>
+#include <linux/stop_machine.h>
+
+#include <asm/sections.h>
+#include <asm/mmu.h>
+#include <asm/tlb.h>
+#include <asm/firmware.h>
+
+#include <mm/mmu_decl.h>
+
+#include <trace/events/thp.h>
+
+#if H_PGTABLE_RANGE > (USER_VSID_RANGE * (TASK_SIZE_USER64 / TASK_CONTEXT_SIZE))
+#warning Limited user VSID range means pagetable space is wasted
+#endif
+
+#ifdef CONFIG_SPARSEMEM_VMEMMAP
+/*
+ * vmemmap is the starting address of the virtual address space where
+ * struct pages are allocated for all possible PFNs present on the system
+ * including holes and bad memory (hence sparse). These virtual struct
+ * pages are stored in sequence in this virtual address space irrespective
+ * of the fact whether the corresponding PFN is valid or not. This achieves
+ * constant relationship between address of struct page and its PFN.
+ *
+ * During boot or memory hotplug operation when a new memory section is
+ * added, physical memory allocation (including hash table bolting) will
+ * be performed for the set of struct pages which are part of the memory
+ * section. This saves memory by not allocating struct pages for PFNs
+ * which are not valid.
+ *
+ *		----------------------------------------------
+ *		| PHYSICAL ALLOCATION OF VIRTUAL STRUCT PAGES|
+ *		----------------------------------------------
+ *
+ *	   f000000000000000                  c000000000000000
+ * vmemmap +--------------+                  +--------------+
+ *  +      |  page struct | +--------------> |  page struct |
+ *  |      +--------------+                  +--------------+
+ *  |      |  page struct | +--------------> |  page struct |
+ *  |      +--------------+ |                +--------------+
+ *  |      |  page struct | +       +------> |  page struct |
+ *  |      +--------------+         |        +--------------+
+ *  |      |  page struct |         |   +--> |  page struct |
+ *  |      +--------------+         |   |    +--------------+
+ *  |      |  page struct |         |   |
+ *  |      +--------------+         |   |
+ *  |      |  page struct |         |   |
+ *  |      +--------------+         |   |
+ *  |      |  page struct |         |   |
+ *  |      +--------------+         |   |
+ *  |      |  page struct |         |   |
+ *  |      +--------------+         |   |
+ *  |      |  page struct | +-------+   |
+ *  |      +--------------+             |
+ *  |      |  page struct | +-----------+
+ *  |      +--------------+
+ *  |      |  page struct | No mapping
+ *  |      +--------------+
+ *  |      |  page struct | No mapping
+ *  v      +--------------+
+ *
+ *		-----------------------------------------
+ *		| RELATION BETWEEN STRUCT PAGES AND PFNS|
+ *		-----------------------------------------
+ *
+ * vmemmap +--------------+                 +---------------+
+ *  +      |  page struct | +-------------> |      PFN      |
+ *  |      +--------------+                 +---------------+
+ *  |      |  page struct | +-------------> |      PFN      |
+ *  |      +--------------+                 +---------------+
+ *  |      |  page struct | +-------------> |      PFN      |
+ *  |      +--------------+                 +---------------+
+ *  |      |  page struct | +-------------> |      PFN      |
+ *  |      +--------------+                 +---------------+
+ *  |      |              |
+ *  |      +--------------+
+ *  |      |              |
+ *  |      +--------------+
+ *  |      |              |
+ *  |      +--------------+                 +---------------+
+ *  |      |  page struct | +-------------> |      PFN      |
+ *  |      +--------------+                 +---------------+
+ *  |      |              |
+ *  |      +--------------+
+ *  |      |              |
+ *  |      +--------------+                 +---------------+
+ *  |      |  page struct | +-------------> |      PFN      |
+ *  |      +--------------+                 +---------------+
+ *  |      |  page struct | +-------------> |      PFN      |
+ *  v      +--------------+                 +---------------+
+ */
+/*
+ * On hash-based CPUs, the vmemmap is bolted in the hash table.
+ *
+ */
+int __meminit hash__vmemmap_create_mapping(unsigned long start,
+				       unsigned long page_size,
+				       unsigned long phys)
+{
+	int rc;
+
+	if ((start + page_size) >= H_VMEMMAP_END) {
+		pr_warn("Outside the supported range\n");
+		return -1;
+	}
+
+	rc = htab_bolt_mapping(start, start + page_size, phys,
+			       pgprot_val(PAGE_KERNEL),
+			       mmu_vmemmap_psize, mmu_kernel_ssize);
+	if (rc < 0) {
+		int rc2 = htab_remove_mapping(start, start + page_size,
+					      mmu_vmemmap_psize,
+					      mmu_kernel_ssize);
+		BUG_ON(rc2 && (rc2 != -ENOENT));
+	}
+	return rc;
+}
+
+#ifdef CONFIG_MEMORY_HOTPLUG
+void hash__vmemmap_remove_mapping(unsigned long start,
+			      unsigned long page_size)
+{
+	int rc = htab_remove_mapping(start, start + page_size,
+				     mmu_vmemmap_psize,
+				     mmu_kernel_ssize);
+	BUG_ON((rc < 0) && (rc != -ENOENT));
+	WARN_ON(rc == -ENOENT);
+}
+#endif
+#endif /* CONFIG_SPARSEMEM_VMEMMAP */
+
+/*
+ * map_kernel_page currently only called by __ioremap
+ * map_kernel_page adds an entry to the ioremap page table
+ * and adds an entry to the HPT, possibly bolting it
+ */
+int hash__map_kernel_page(unsigned long ea, unsigned long pa, pgprot_t prot)
+{
+	pgd_t *pgdp;
+	p4d_t *p4dp;
+	pud_t *pudp;
+	pmd_t *pmdp;
+	pte_t *ptep;
+
+	BUILD_BUG_ON(TASK_SIZE_USER64 > H_PGTABLE_RANGE);
+	if (slab_is_available()) {
+		pgdp = pgd_offset_k(ea);
+		p4dp = p4d_offset(pgdp, ea);
+		pudp = pud_alloc(&init_mm, p4dp, ea);
+		if (!pudp)
+			return -ENOMEM;
+		pmdp = pmd_alloc(&init_mm, pudp, ea);
+		if (!pmdp)
+			return -ENOMEM;
+		ptep = pte_alloc_kernel(pmdp, ea);
+		if (!ptep)
+			return -ENOMEM;
+		set_pte_at(&init_mm, ea, ptep, pfn_pte(pa >> PAGE_SHIFT, prot));
+	} else {
+		/*
+		 * If the mm subsystem is not fully up, we cannot create a
+		 * linux page table entry for this mapping.  Simply bolt an
+		 * entry in the hardware page table.
+		 *
+		 */
+		if (htab_bolt_mapping(ea, ea + PAGE_SIZE, pa, pgprot_val(prot),
+				      mmu_io_psize, mmu_kernel_ssize)) {
+			printk(KERN_ERR "Failed to do bolted mapping IO "
+			       "memory at %016lx !\n", pa);
+			return -ENOMEM;
+		}
+	}
+
+	smp_wmb();
+	return 0;
+}
+
+#ifdef CONFIG_TRANSPARENT_HUGEPAGE
+
+unsigned long hash__pmd_hugepage_update(struct mm_struct *mm, unsigned long addr,
+				    pmd_t *pmdp, unsigned long clr,
+				    unsigned long set)
+{
+	__be64 old_be, tmp;
+	unsigned long old;
+
+#ifdef CONFIG_DEBUG_VM
+	WARN_ON(!hash__pmd_trans_huge(*pmdp));
+	assert_spin_locked(pmd_lockptr(mm, pmdp));
+#endif
+
+	__asm__ __volatile__(
+	"1:	ldarx	%0,0,%3\n\
+		and.	%1,%0,%6\n\
+		bne-	1b \n\
+		andc	%1,%0,%4 \n\
+		or	%1,%1,%7\n\
+		stdcx.	%1,0,%3 \n\
+		bne-	1b"
+	: "=&r" (old_be), "=&r" (tmp), "=m" (*pmdp)
+	: "r" (pmdp), "r" (cpu_to_be64(clr)), "m" (*pmdp),
+	  "r" (cpu_to_be64(H_PAGE_BUSY)), "r" (cpu_to_be64(set))
+	: "cc" );
+
+	old = be64_to_cpu(old_be);
+
+	trace_hugepage_update_pmd(addr, old, clr, set);
+	if (old & H_PAGE_HASHPTE)
+		hpte_do_hugepage_flush(mm, addr, pmdp, old);
+	return old;
+}
+
+static void do_nothing(void *arg)
+{
+
+}
+
+/*
+ * Serialize against __find_linux_pte() which does lock-less
+ * lookup in page tables with local interrupts disabled. For huge pages
+ * it casts pmd_t to pte_t. Since format of pte_t is different from
+ * pmd_t we want to prevent transit from pmd pointing to page table
+ * to pmd pointing to huge page (and back) while interrupts are disabled.
+ * We clear pmd to possibly replace it with page table pointer in
+ * different code paths. So make sure we wait for the parallel
+ * __find_linux_pte() to finish.
+ */
+static void serialize_against_pte_lookup(struct mm_struct *mm)
+{
+	smp_mb();
+	smp_call_function_many(mm_cpumask(mm), do_nothing, mm, 1);
+}
+
+pmd_t hash__pmdp_collapse_flush(struct vm_area_struct *vma, unsigned long address,
+			    pmd_t *pmdp)
+{
+	pmd_t pmd;
+
+	VM_BUG_ON(address & ~HPAGE_PMD_MASK);
+	VM_BUG_ON(pmd_trans_huge(*pmdp));
+
+	pmd = *pmdp;
+	pmd_clear(pmdp);
+
+	page_table_check_pmd_clear(vma->vm_mm, address, pmd);
+
+	/*
+	 * Wait for all pending hash_page to finish. This is needed
+	 * in case of subpage collapse. When we collapse normal pages
+	 * to hugepage, we first clear the pmd, then invalidate all
+	 * the PTE entries. The assumption here is that any low level
+	 * page fault will see a none pmd and take the slow path that
+	 * will wait on mmap_lock. But we could very well be in a
+	 * hash_page with local ptep pointer value. Such a hash page
+	 * can result in adding new HPTE entries for normal subpages.
+	 * That means we could be modifying the page content as we
+	 * copy them to a huge page. So wait for parallel hash_page
+	 * to finish before invalidating HPTE entries. We can do this
+	 * by sending an IPI to all the cpus and executing a dummy
+	 * function there.
+	 */
+	serialize_against_pte_lookup(vma->vm_mm);
+	/*
+	 * Now invalidate the hpte entries in the range
+	 * covered by pmd. This make sure we take a
+	 * fault and will find the pmd as none, which will
+	 * result in a major fault which takes mmap_lock and
+	 * hence wait for collapse to complete. Without this
+	 * the __collapse_huge_page_copy can result in copying
+	 * the old content.
+	 */
+	flush_hash_table_pmd_range(vma->vm_mm, &pmd, address);
+	return pmd;
+}
+
+/*
+ * We want to put the pgtable in pmd and use pgtable for tracking
+ * the base page size hptes
+ */
+void hash__pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
+				  pgtable_t pgtable)
+{
+	pgtable_t *pgtable_slot;
+
+	assert_spin_locked(pmd_lockptr(mm, pmdp));
+	/*
+	 * we store the pgtable in the second half of PMD
+	 */
+	pgtable_slot = (pgtable_t *)pmdp + PTRS_PER_PMD;
+	*pgtable_slot = pgtable;
+	/*
+	 * expose the deposited pgtable to other cpus.
+	 * before we set the hugepage PTE at pmd level
+	 * hash fault code looks at the deposted pgtable
+	 * to store hash index values.
+	 */
+	smp_wmb();
+}
+
+pgtable_t hash__pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp)
+{
+	pgtable_t pgtable;
+	pgtable_t *pgtable_slot;
+
+	assert_spin_locked(pmd_lockptr(mm, pmdp));
+
+	pgtable_slot = (pgtable_t *)pmdp + PTRS_PER_PMD;
+	pgtable = *pgtable_slot;
+	/*
+	 * Once we withdraw, mark the entry NULL.
+	 */
+	*pgtable_slot = NULL;
+	/*
+	 * We store HPTE information in the deposited PTE fragment.
+	 * zero out the content on withdraw.
+	 */
+	memset(pgtable, 0, PTE_FRAG_SIZE);
+	return pgtable;
+}
+
+/*
+ * A linux hugepage PMD was changed and the corresponding hash table entries
+ * neesd to be flushed.
+ */
+void hpte_do_hugepage_flush(struct mm_struct *mm, unsigned long addr,
+			    pmd_t *pmdp, unsigned long old_pmd)
+{
+	int ssize;
+	unsigned int psize;
+	unsigned long vsid;
+	unsigned long flags = 0;
+
+	/* get the base page size,vsid and segment size */
+#ifdef CONFIG_DEBUG_VM
+	psize = get_slice_psize(mm, addr);
+	BUG_ON(psize == MMU_PAGE_16M);
+#endif
+	if (old_pmd & H_PAGE_COMBO)
+		psize = MMU_PAGE_4K;
+	else
+		psize = MMU_PAGE_64K;
+
+	if (!is_kernel_addr(addr)) {
+		ssize = user_segment_size(addr);
+		vsid = get_user_vsid(&mm->context, addr, ssize);
+		WARN_ON(vsid == 0);
+	} else {
+		vsid = get_kernel_vsid(addr, mmu_kernel_ssize);
+		ssize = mmu_kernel_ssize;
+	}
+
+	if (mm_is_thread_local(mm))
+		flags |= HPTE_LOCAL_UPDATE;
+
+	return flush_hash_hugepage(vsid, addr, pmdp, psize, ssize, flags);
+}
+
+pmd_t hash__pmdp_huge_get_and_clear(struct mm_struct *mm,
+				unsigned long addr, pmd_t *pmdp)
+{
+	pmd_t old_pmd;
+	pgtable_t pgtable;
+	unsigned long old;
+	pgtable_t *pgtable_slot;
+
+	old = pmd_hugepage_update(mm, addr, pmdp, ~0UL, 0);
+	old_pmd = __pmd(old);
+	/*
+	 * We have pmd == none and we are holding page_table_lock.
+	 * So we can safely go and clear the pgtable hash
+	 * index info.
+	 */
+	pgtable_slot = (pgtable_t *)pmdp + PTRS_PER_PMD;
+	pgtable = *pgtable_slot;
+	/*
+	 * Let's zero out old valid and hash index details
+	 * hash fault look at them.
+	 */
+	memset(pgtable, 0, PTE_FRAG_SIZE);
+	return old_pmd;
+}
+
+int hash__has_transparent_hugepage(void)
+{
+
+	if (!mmu_has_feature(MMU_FTR_16M_PAGE))
+		return 0;
+	/*
+	 * We support THP only if PMD_SIZE is 16MB.
+	 */
+	if (mmu_psize_defs[MMU_PAGE_16M].shift != PMD_SHIFT)
+		return 0;
+	/*
+	 * We need to make sure that we support 16MB hugepage in a segment
+	 * with base page size 64K or 4K. We only enable THP with a PAGE_SIZE
+	 * of 64K.
+	 */
+	/*
+	 * If we have 64K HPTE, we will be using that by default
+	 */
+	if (mmu_psize_defs[MMU_PAGE_64K].shift &&
+	    (mmu_psize_defs[MMU_PAGE_64K].penc[MMU_PAGE_16M] == -1))
+		return 0;
+	/*
+	 * Ok we only have 4K HPTE
+	 */
+	if (mmu_psize_defs[MMU_PAGE_4K].penc[MMU_PAGE_16M] == -1)
+		return 0;
+
+	return 1;
+}
+EXPORT_SYMBOL_GPL(hash__has_transparent_hugepage);
+
+#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
+
+#ifdef CONFIG_STRICT_KERNEL_RWX
+
+struct change_memory_parms {
+	unsigned long start, end, newpp;
+	unsigned int step, nr_cpus;
+	atomic_t master_cpu;
+	atomic_t cpu_counter;
+};
+
+// We'd rather this was on the stack but it has to be in the RMO
+static struct change_memory_parms chmem_parms;
+
+// And therefore we need a lock to protect it from concurrent use
+static DEFINE_MUTEX(chmem_lock);
+
+static void change_memory_range(unsigned long start, unsigned long end,
+				unsigned int step, unsigned long newpp)
+{
+	unsigned long idx;
+
+	pr_debug("Changing page protection on range 0x%lx-0x%lx, to 0x%lx, step 0x%x\n",
+		 start, end, newpp, step);
+
+	for (idx = start; idx < end; idx += step)
+		/* Not sure if we can do much with the return value */
+		mmu_hash_ops.hpte_updateboltedpp(newpp, idx, mmu_linear_psize,
+							mmu_kernel_ssize);
+}
+
+static int notrace chmem_secondary_loop(struct change_memory_parms *parms)
+{
+	unsigned long msr, tmp, flags;
+	int *p;
+
+	p = &parms->cpu_counter.counter;
+
+	local_irq_save(flags);
+	hard_irq_disable();
+
+	asm volatile (
+	// Switch to real mode and leave interrupts off
+	"mfmsr	%[msr]			;"
+	"li	%[tmp], %[MSR_IR_DR]	;"
+	"andc	%[tmp], %[msr], %[tmp]	;"
+	"mtmsrd %[tmp]			;"
+
+	// Tell the master we are in real mode
+	"1:				"
+	"lwarx	%[tmp], 0, %[p]		;"
+	"addic	%[tmp], %[tmp], -1	;"
+	"stwcx.	%[tmp], 0, %[p]		;"
+	"bne-	1b			;"
+
+	// Spin until the counter goes to zero
+	"2:				;"
+	"lwz	%[tmp], 0(%[p])		;"
+	"cmpwi	%[tmp], 0		;"
+	"bne-	2b			;"
+
+	// Switch back to virtual mode
+	"mtmsrd %[msr]			;"
+
+	: // outputs
+	  [msr] "=&r" (msr), [tmp] "=&b" (tmp), "+m" (*p)
+	: // inputs
+	  [p] "b" (p), [MSR_IR_DR] "i" (MSR_IR | MSR_DR)
+	: // clobbers
+	  "cc", "xer"
+	);
+
+	local_irq_restore(flags);
+
+	return 0;
+}
+
+static int change_memory_range_fn(void *data)
+{
+	struct change_memory_parms *parms = data;
+
+	// First CPU goes through, all others wait.
+	if (atomic_xchg(&parms->master_cpu, 1) == 1)
+		return chmem_secondary_loop(parms);
+
+	// Wait for all but one CPU (this one) to call-in
+	while (atomic_read(&parms->cpu_counter) > 1)
+		barrier();
+
+	change_memory_range(parms->start, parms->end, parms->step, parms->newpp);
+
+	mb();
+
+	// Signal the other CPUs that we're done
+	atomic_dec(&parms->cpu_counter);
+
+	return 0;
+}
+
+static bool hash__change_memory_range(unsigned long start, unsigned long end,
+				      unsigned long newpp)
+{
+	unsigned int step, shift;
+
+	shift = mmu_psize_defs[mmu_linear_psize].shift;
+	step = 1 << shift;
+
+	start = ALIGN_DOWN(start, step);
+	end = ALIGN(end, step); // aligns up
+
+	if (start >= end)
+		return false;
+
+	if (firmware_has_feature(FW_FEATURE_LPAR)) {
+		mutex_lock(&chmem_lock);
+
+		chmem_parms.start = start;
+		chmem_parms.end = end;
+		chmem_parms.step = step;
+		chmem_parms.newpp = newpp;
+		atomic_set(&chmem_parms.master_cpu, 0);
+
+		cpus_read_lock();
+
+		atomic_set(&chmem_parms.cpu_counter, num_online_cpus());
+
+		// Ensure state is consistent before we call the other CPUs
+		mb();
+
+		stop_machine_cpuslocked(change_memory_range_fn, &chmem_parms,
+					cpu_online_mask);
+
+		cpus_read_unlock();
+		mutex_unlock(&chmem_lock);
+	} else
+		change_memory_range(start, end, step, newpp);
+
+	return true;
+}
+
+void hash__mark_rodata_ro(void)
+{
+	unsigned long start, end, pp;
+
+	start = (unsigned long)_stext;
+	end = (unsigned long)__end_rodata;
+
+	pp = htab_convert_pte_flags(pgprot_val(PAGE_KERNEL_ROX), HPTE_USE_KERNEL_KEY);
+
+	WARN_ON(!hash__change_memory_range(start, end, pp));
+}
+
+void hash__mark_initmem_nx(void)
+{
+	unsigned long start, end, pp;
+
+	start = (unsigned long)__init_begin;
+	end = (unsigned long)__init_end;
+
+	pp = htab_convert_pte_flags(pgprot_val(PAGE_KERNEL), HPTE_USE_KERNEL_KEY);
+
+	WARN_ON(!hash__change_memory_range(start, end, pp));
+}
+#endif

linux-7.2.3/arch/powerpc/mm/book3s64/hash_tlb.c

@@ -0,0 +1,256 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * This file contains the routines for flushing entries from the
+ * TLB and MMU hash table.
+ *
+ *  Derived from arch/ppc64/mm/init.c:
+ *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
+ *
+ *  Modifications by Paul Mackerras (PowerMac) (paulus@cs.anu.edu.au)
+ *  and Cort Dougan (PReP) (cort@cs.nmt.edu)
+ *    Copyright (C) 1996 Paul Mackerras
+ *
+ *  Derived from "arch/i386/mm/init.c"
+ *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
+ *
+ *  Dave Engebretsen <engebret@us.ibm.com>
+ *      Rework for PPC64 port.
+ */
+
+#include <linux/kernel.h>
+#include <linux/mm.h>
+#include <linux/percpu.h>
+#include <linux/hardirq.h>
+#include <asm/tlbflush.h>
+#include <asm/tlb.h>
+#include <asm/bug.h>
+#include <asm/pte-walk.h>
+#include <kunit/visibility.h>
+
+#include <trace/events/thp.h>
+
+DEFINE_PER_CPU(struct ppc64_tlb_batch, ppc64_tlb_batch);
+EXPORT_SYMBOL_IF_KUNIT(ppc64_tlb_batch);
+
+/*
+ * A linux PTE was changed and the corresponding hash table entry
+ * neesd to be flushed. This function will either perform the flush
+ * immediately or will batch it up if the current CPU has an active
+ * batch on it.
+ */
+void hpte_need_flush(struct mm_struct *mm, unsigned long addr,
+		     pte_t *ptep, unsigned long pte, int huge)
+{
+	unsigned long vpn;
+	struct ppc64_tlb_batch *batch = &get_cpu_var(ppc64_tlb_batch);
+	unsigned long vsid;
+	unsigned int psize;
+	int ssize;
+	real_pte_t rpte;
+	int i, offset;
+
+	i = batch->index;
+
+	/*
+	 * Get page size (maybe move back to caller).
+	 *
+	 * NOTE: when using special 64K mappings in 4K environment like
+	 * for SPEs, we obtain the page size from the slice, which thus
+	 * must still exist (and thus the VMA not reused) at the time
+	 * of this call
+	 */
+	if (huge) {
+#ifdef CONFIG_HUGETLB_PAGE
+		psize = get_slice_psize(mm, addr);
+		/* Mask the address for the correct page size */
+		addr &= ~((1UL << mmu_psize_defs[psize].shift) - 1);
+		if (unlikely(psize == MMU_PAGE_16G))
+			offset = PTRS_PER_PUD;
+		else
+			offset = PTRS_PER_PMD;
+#else
+		BUG();
+		psize = pte_pagesize_index(mm, addr, pte); /* shutup gcc */
+#endif
+	} else {
+		psize = pte_pagesize_index(mm, addr, pte);
+		/*
+		 * Mask the address for the standard page size.  If we
+		 * have a 64k page kernel, but the hardware does not
+		 * support 64k pages, this might be different from the
+		 * hardware page size encoded in the slice table.
+		 */
+		addr &= PAGE_MASK;
+		offset = PTRS_PER_PTE;
+	}
+
+
+	/* Build full vaddr */
+	if (!is_kernel_addr(addr)) {
+		ssize = user_segment_size(addr);
+		vsid = get_user_vsid(&mm->context, addr, ssize);
+	} else {
+		vsid = get_kernel_vsid(addr, mmu_kernel_ssize);
+		ssize = mmu_kernel_ssize;
+	}
+	WARN_ON(vsid == 0);
+	vpn = hpt_vpn(addr, vsid, ssize);
+	rpte = __real_pte(__pte(pte), ptep, offset);
+
+	/*
+	 * Check if we have an active batch on this CPU. If not, just
+	 * flush now and return.
+	 */
+	if (!is_lazy_mmu_mode_active()) {
+		flush_hash_page(vpn, rpte, psize, ssize, mm_is_thread_local(mm));
+		put_cpu_var(ppc64_tlb_batch);
+		return;
+	}
+
+	/*
+	 * This can happen when we are in the middle of a TLB batch and
+	 * we encounter memory pressure (eg copy_page_range when it tries
+	 * to allocate a new pte). If we have to reclaim memory and end
+	 * up scanning and resetting referenced bits then our batch context
+	 * will change mid stream.
+	 *
+	 * We also need to ensure only one page size is present in a given
+	 * batch
+	 */
+	if (i != 0 && (mm != batch->mm || batch->psize != psize ||
+		       batch->ssize != ssize)) {
+		__flush_tlb_pending(batch);
+		i = 0;
+	}
+	if (i == 0) {
+		batch->mm = mm;
+		batch->psize = psize;
+		batch->ssize = ssize;
+	}
+	batch->pte[i] = rpte;
+	batch->vpn[i] = vpn;
+	batch->index = ++i;
+	if (i >= PPC64_TLB_BATCH_NR)
+		__flush_tlb_pending(batch);
+	put_cpu_var(ppc64_tlb_batch);
+}
+
+/*
+ * This function is called when terminating an mmu batch or when a batch
+ * is full. It will perform the flush of all the entries currently stored
+ * in a batch.
+ *
+ * Must be called from within some kind of spinlock/non-preempt region...
+ */
+void __flush_tlb_pending(struct ppc64_tlb_batch *batch)
+{
+	int i, local;
+
+	i = batch->index;
+	local = mm_is_thread_local(batch->mm);
+	if (i == 1)
+		flush_hash_page(batch->vpn[0], batch->pte[0],
+				batch->psize, batch->ssize, local);
+	else
+		flush_hash_range(i, local);
+	batch->index = 0;
+}
+EXPORT_SYMBOL_IF_KUNIT(__flush_tlb_pending);
+
+void hash__tlb_flush(struct mmu_gather *tlb)
+{
+	struct ppc64_tlb_batch *tlbbatch = &get_cpu_var(ppc64_tlb_batch);
+
+	/*
+	 * If there's a TLB batch pending, then we must flush it because the
+	 * pages are going to be freed and we really don't want to have a CPU
+	 * access a freed page because it has a stale TLB
+	 */
+	if (tlbbatch->index)
+		__flush_tlb_pending(tlbbatch);
+
+	put_cpu_var(ppc64_tlb_batch);
+}
+
+/**
+ * __flush_hash_table_range - Flush all HPTEs for a given address range
+ *                            from the hash table (and the TLB). But keeps
+ *                            the linux PTEs intact.
+ *
+ * @start	: starting address
+ * @end         : ending address (not included in the flush)
+ *
+ * This function is mostly to be used by some IO hotplug code in order
+ * to remove all hash entries from a given address range used to map IO
+ * space on a removed PCI-PCI bidge without tearing down the full mapping
+ * since 64K pages may overlap with other bridges when using 64K pages
+ * with 4K HW pages on IO space.
+ *
+ * Because of that usage pattern, it is implemented for small size rather
+ * than speed.
+ */
+void __flush_hash_table_range(unsigned long start, unsigned long end)
+{
+	int hugepage_shift;
+	unsigned long flags;
+
+	start = ALIGN_DOWN(start, PAGE_SIZE);
+	end = ALIGN(end, PAGE_SIZE);
+
+
+	/*
+	 * Note: Normally, we should only ever use a batch within a
+	 * PTE locked section. This violates the rule, but will work
+	 * since we don't actually modify the PTEs, we just flush the
+	 * hash while leaving the PTEs intact (including their reference
+	 * to being hashed). This is not the most performance oriented
+	 * way to do things but is fine for our needs here.
+	 */
+	local_irq_save(flags);
+	lazy_mmu_mode_enable();
+	for (; start < end; start += PAGE_SIZE) {
+		pte_t *ptep = find_init_mm_pte(start, &hugepage_shift);
+		unsigned long pte;
+
+		if (ptep == NULL)
+			continue;
+		pte = pte_val(*ptep);
+		if (!(pte & H_PAGE_HASHPTE))
+			continue;
+		hpte_need_flush(&init_mm, start, ptep, pte, hugepage_shift);
+	}
+	lazy_mmu_mode_disable();
+	local_irq_restore(flags);
+}
+
+void flush_hash_table_pmd_range(struct mm_struct *mm, pmd_t *pmd, unsigned long addr)
+{
+	pte_t *pte;
+	pte_t *start_pte;
+	unsigned long flags;
+
+	addr = ALIGN_DOWN(addr, PMD_SIZE);
+	/*
+	 * Note: Normally, we should only ever use a batch within a
+	 * PTE locked section. This violates the rule, but will work
+	 * since we don't actually modify the PTEs, we just flush the
+	 * hash while leaving the PTEs intact (including their reference
+	 * to being hashed). This is not the most performance oriented
+	 * way to do things but is fine for our needs here.
+	 */
+	local_irq_save(flags);
+	lazy_mmu_mode_enable();
+	start_pte = pte_offset_map(pmd, addr);
+	if (!start_pte)
+		goto out;
+	for (pte = start_pte; pte < start_pte + PTRS_PER_PTE; pte++) {
+		unsigned long pteval = pte_val(*pte);
+		if (pteval & H_PAGE_HASHPTE)
+			hpte_need_flush(mm, addr, pte, pteval, 0);
+		addr += PAGE_SIZE;
+	}
+	pte_unmap(start_pte);
+out:
+	lazy_mmu_mode_disable();
+	local_irq_restore(flags);
+}

linux-7.2.3/arch/powerpc/mm/book3s64/hugetlbpage.c

@@ -0,0 +1,177 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * PPC64 Huge TLB Page Support for hash based MMUs (POWER4 and later)
+ *
+ * Copyright (C) 2003 David Gibson, IBM Corporation.
+ *
+ * Based on the IA-32 version:
+ * Copyright (C) 2002, Rohit Seth <rohit.seth@intel.com>
+ */
+
+#include <linux/mm.h>
+#include <linux/hugetlb.h>
+#include <asm/cacheflush.h>
+#include <asm/machdep.h>
+
+unsigned int hpage_shift;
+EXPORT_SYMBOL(hpage_shift);
+
+#ifdef CONFIG_PPC_64S_HASH_MMU
+int __hash_page_huge(unsigned long ea, unsigned long access, unsigned long vsid,
+		     pte_t *ptep, unsigned long trap, unsigned long flags,
+		     int ssize, unsigned int shift, unsigned int mmu_psize)
+{
+	real_pte_t rpte;
+	unsigned long vpn;
+	unsigned long old_pte, new_pte;
+	unsigned long rflags, pa;
+	long slot, offset;
+
+	BUG_ON(shift != mmu_psize_defs[mmu_psize].shift);
+
+	/* Search the Linux page table for a match with va */
+	vpn = hpt_vpn(ea, vsid, ssize);
+
+	/*
+	 * At this point, we have a pte (old_pte) which can be used to build
+	 * or update an HPTE. There are 2 cases:
+	 *
+	 * 1. There is a valid (present) pte with no associated HPTE (this is
+	 *	the most common case)
+	 * 2. There is a valid (present) pte with an associated HPTE. The
+	 *	current values of the pp bits in the HPTE prevent access
+	 *	because we are doing software DIRTY bit management and the
+	 *	page is currently not DIRTY.
+	 */
+
+
+	do {
+		old_pte = pte_val(*ptep);
+		/* If PTE busy, retry the access */
+		if (unlikely(old_pte & H_PAGE_BUSY))
+			return 0;
+		/* If PTE permissions don't match, take page fault */
+		if (unlikely(!check_pte_access(access, old_pte)))
+			return 1;
+		/*
+		 * If hash-4k, hugepages use seeral contiguous PxD entries
+		 * so bail out and let mm make the page young or dirty
+		 */
+		if (IS_ENABLED(CONFIG_PPC_4K_PAGES)) {
+			if (!(old_pte & _PAGE_ACCESSED))
+				return 1;
+			if ((access & _PAGE_WRITE) && !(old_pte & _PAGE_DIRTY))
+				return 1;
+		}
+
+		/*
+		 * Try to lock the PTE, add ACCESSED and DIRTY if it was
+		 * a write access
+		 */
+		new_pte = old_pte | H_PAGE_BUSY | _PAGE_ACCESSED;
+		if (access & _PAGE_WRITE)
+			new_pte |= _PAGE_DIRTY;
+	} while(!pte_xchg(ptep, __pte(old_pte), __pte(new_pte)));
+
+	/* Make sure this is a hugetlb entry */
+	if (old_pte & H_PAGE_THP_HUGE)
+		return 0;
+
+	rflags = htab_convert_pte_flags(new_pte, flags);
+	if (unlikely(mmu_psize == MMU_PAGE_16G))
+		offset = PTRS_PER_PUD;
+	else
+		offset = PTRS_PER_PMD;
+	rpte = __real_pte(__pte(old_pte), ptep, offset);
+
+	if (!cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
+		/*
+		 * No CPU has hugepages but lacks no execute, so we
+		 * don't need to worry about that case
+		 */
+		rflags = hash_page_do_lazy_icache(rflags, __pte(old_pte), trap);
+
+	/* Check if pte already has an hpte (case 2) */
+	if (unlikely(old_pte & H_PAGE_HASHPTE)) {
+		/* There MIGHT be an HPTE for this pte */
+		unsigned long gslot;
+
+		gslot = pte_get_hash_gslot(vpn, shift, ssize, rpte, 0);
+		if (mmu_hash_ops.hpte_updatepp(gslot, rflags, vpn, mmu_psize,
+					       mmu_psize, ssize, flags) == -1)
+			old_pte &= ~_PAGE_HPTEFLAGS;
+	}
+
+	if (likely(!(old_pte & H_PAGE_HASHPTE))) {
+		unsigned long hash = hpt_hash(vpn, shift, ssize);
+
+		pa = pte_pfn(__pte(old_pte)) << PAGE_SHIFT;
+
+		/* clear HPTE slot informations in new PTE */
+		new_pte = (new_pte & ~_PAGE_HPTEFLAGS) | H_PAGE_HASHPTE;
+
+		slot = hpte_insert_repeating(hash, vpn, pa, rflags, 0,
+					     mmu_psize, ssize);
+
+		/*
+		 * Hypervisor failure. Restore old pte and return -1
+		 * similar to __hash_page_*
+		 */
+		if (unlikely(slot == -2)) {
+			*ptep = __pte(old_pte);
+			hash_failure_debug(ea, access, vsid, trap, ssize,
+					   mmu_psize, mmu_psize, old_pte);
+			return -1;
+		}
+
+		new_pte |= pte_set_hidx(ptep, rpte, 0, slot, offset);
+	}
+
+	/*
+	 * No need to use ldarx/stdcx here
+	 */
+	*ptep = __pte(new_pte & ~H_PAGE_BUSY);
+	return 0;
+}
+#endif
+
+pte_t huge_ptep_modify_prot_start(struct vm_area_struct *vma,
+				  unsigned long addr, pte_t *ptep)
+{
+	unsigned long pte_val;
+	/*
+	 * Clear the _PAGE_PRESENT so that no hardware parallel update is
+	 * possible. Also keep the pte_present true so that we don't take
+	 * wrong fault.
+	 */
+	pte_val = pte_update(vma->vm_mm, addr, ptep,
+			     _PAGE_PRESENT, _PAGE_INVALID, 1);
+
+	return __pte(pte_val);
+}
+
+void huge_ptep_modify_prot_commit(struct vm_area_struct *vma, unsigned long addr,
+				  pte_t *ptep, pte_t old_pte, pte_t pte)
+{
+	unsigned long psize;
+
+	if (radix_enabled())
+		return radix__huge_ptep_modify_prot_commit(vma, addr, ptep,
+							   old_pte, pte);
+
+	psize = huge_page_size(hstate_vma(vma));
+	set_huge_pte_at(vma->vm_mm, addr, ptep, pte, psize);
+}
+
+void __init hugetlbpage_init_defaultsize(void)
+{
+	/* Set default large page size. Currently, we pick 16M or 1M
+	 * depending on what is available
+	 */
+	if (mmu_psize_defs[MMU_PAGE_16M].shift)
+		hpage_shift = mmu_psize_defs[MMU_PAGE_16M].shift;
+	else if (mmu_psize_defs[MMU_PAGE_1M].shift)
+		hpage_shift = mmu_psize_defs[MMU_PAGE_1M].shift;
+	else if (mmu_psize_defs[MMU_PAGE_2M].shift)
+		hpage_shift = mmu_psize_defs[MMU_PAGE_2M].shift;
+}

linux-7.2.3/arch/powerpc/mm/book3s64/internal.h

@@ -0,0 +1,34 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+#ifndef ARCH_POWERPC_MM_BOOK3S64_INTERNAL_H
+#define ARCH_POWERPC_MM_BOOK3S64_INTERNAL_H
+
+#include <linux/jump_label.h>
+
+extern bool stress_slb_enabled;
+
+DECLARE_STATIC_KEY_FALSE(stress_slb_key);
+
+static inline bool stress_slb(void)
+{
+	return static_branch_unlikely(&stress_slb_key);
+}
+
+extern bool stress_hpt_enabled;
+
+DECLARE_STATIC_KEY_FALSE(stress_hpt_key);
+
+static inline bool stress_hpt(void)
+{
+	return static_branch_unlikely(&stress_hpt_key);
+}
+
+extern bool no_slb_preload;
+DECLARE_STATIC_KEY_FALSE(no_slb_preload_key);
+static inline bool slb_preload_disabled(void)
+{
+	return static_branch_unlikely(&no_slb_preload_key);
+}
+
+void hpt_do_stress(unsigned long ea, unsigned long hpte_group);
+
+#endif /* ARCH_POWERPC_MM_BOOK3S64_INTERNAL_H */

linux-7.2.3/arch/powerpc/mm/book3s64/iommu_api.c

@@ -0,0 +1,402 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ *  IOMMU helpers in MMU context.
+ *
+ *  Copyright (C) 2015 IBM Corp. <aik@ozlabs.ru>
+ */
+
+#include <linux/sched/signal.h>
+#include <linux/slab.h>
+#include <linux/rculist.h>
+#include <linux/vmalloc.h>
+#include <linux/mutex.h>
+#include <linux/migrate.h>
+#include <linux/hugetlb.h>
+#include <linux/swap.h>
+#include <linux/sizes.h>
+#include <linux/mm.h>
+#include <asm/mmu_context.h>
+#include <asm/pte-walk.h>
+#include <linux/mm_inline.h>
+
+static DEFINE_MUTEX(mem_list_mutex);
+
+#define MM_IOMMU_TABLE_GROUP_PAGE_DIRTY	0x1
+#define MM_IOMMU_TABLE_GROUP_PAGE_MASK	~(SZ_4K - 1)
+
+struct mm_iommu_table_group_mem_t {
+	struct list_head next;
+	struct rcu_head rcu;
+	unsigned long used;
+	atomic64_t mapped;
+	unsigned int pageshift;
+	u64 ua;			/* userspace address */
+	u64 entries;		/* number of entries in hpas/hpages[] */
+	/*
+	 * in mm_iommu_get we temporarily use this to store
+	 * struct page address.
+	 *
+	 * We need to convert ua to hpa in real mode. Make it
+	 * simpler by storing physical address.
+	 */
+	union {
+		struct page **hpages;	/* vmalloc'ed */
+		phys_addr_t *hpas;
+	};
+#define MM_IOMMU_TABLE_INVALID_HPA	((uint64_t)-1)
+	u64 dev_hpa;		/* Device memory base address */
+};
+
+bool mm_iommu_preregistered(struct mm_struct *mm)
+{
+	return !list_empty(&mm->context.iommu_group_mem_list);
+}
+EXPORT_SYMBOL_GPL(mm_iommu_preregistered);
+
+static long mm_iommu_do_alloc(struct mm_struct *mm, unsigned long ua,
+			      unsigned long entries, unsigned long dev_hpa,
+			      struct mm_iommu_table_group_mem_t **pmem)
+{
+	struct mm_iommu_table_group_mem_t *mem, *mem2;
+	long i, ret, locked_entries = 0, pinned = 0;
+	unsigned int pageshift;
+	unsigned long entry, chunk;
+
+	if (dev_hpa == MM_IOMMU_TABLE_INVALID_HPA) {
+		ret = account_locked_vm(mm, entries, true);
+		if (ret)
+			return ret;
+
+		locked_entries = entries;
+	}
+
+	mem = kzalloc_obj(*mem);
+	if (!mem) {
+		ret = -ENOMEM;
+		goto unlock_exit;
+	}
+
+	if (dev_hpa != MM_IOMMU_TABLE_INVALID_HPA) {
+		mem->pageshift = __ffs(dev_hpa | (entries << PAGE_SHIFT));
+		mem->dev_hpa = dev_hpa;
+		goto good_exit;
+	}
+	mem->dev_hpa = MM_IOMMU_TABLE_INVALID_HPA;
+
+	/*
+	 * For a starting point for a maximum page size calculation
+	 * we use @ua and @entries natural alignment to allow IOMMU pages
+	 * smaller than huge pages but still bigger than PAGE_SIZE.
+	 */
+	mem->pageshift = __ffs(ua | (entries << PAGE_SHIFT));
+	mem->hpas = vzalloc(array_size(entries, sizeof(mem->hpas[0])));
+	if (!mem->hpas) {
+		kfree(mem);
+		ret = -ENOMEM;
+		goto unlock_exit;
+	}
+
+	mmap_read_lock(mm);
+	chunk = (1UL << (PAGE_SHIFT + MAX_PAGE_ORDER)) /
+			sizeof(struct vm_area_struct *);
+	chunk = min(chunk, entries);
+	for (entry = 0; entry < entries; entry += chunk) {
+		unsigned long n = min(entries - entry, chunk);
+
+		ret = pin_user_pages(ua + (entry << PAGE_SHIFT), n,
+				FOLL_WRITE | FOLL_LONGTERM,
+				mem->hpages + entry);
+		if (ret == n) {
+			pinned += n;
+			continue;
+		}
+		if (ret > 0)
+			pinned += ret;
+		break;
+	}
+	mmap_read_unlock(mm);
+	if (pinned != entries) {
+		if (!ret)
+			ret = -EFAULT;
+		goto free_exit;
+	}
+
+good_exit:
+	atomic64_set(&mem->mapped, 1);
+	mem->used = 1;
+	mem->ua = ua;
+	mem->entries = entries;
+
+	mutex_lock(&mem_list_mutex);
+
+	list_for_each_entry_rcu(mem2, &mm->context.iommu_group_mem_list, next,
+				lockdep_is_held(&mem_list_mutex)) {
+		/* Overlap? */
+		if ((mem2->ua < (ua + (entries << PAGE_SHIFT))) &&
+				(ua < (mem2->ua +
+				       (mem2->entries << PAGE_SHIFT)))) {
+			ret = -EINVAL;
+			mutex_unlock(&mem_list_mutex);
+			goto free_exit;
+		}
+	}
+
+	if (mem->dev_hpa == MM_IOMMU_TABLE_INVALID_HPA) {
+		/*
+		 * Allow to use larger than 64k IOMMU pages. Only do that
+		 * if we are backed by hugetlb. Skip device memory as it is not
+		 * backed with page structs.
+		 */
+		pageshift = PAGE_SHIFT;
+		for (i = 0; i < entries; ++i) {
+			struct page *page = mem->hpages[i];
+
+			if ((mem->pageshift > PAGE_SHIFT) && PageHuge(page))
+				pageshift = page_shift(compound_head(page));
+			mem->pageshift = min(mem->pageshift, pageshift);
+			/*
+			 * We don't need struct page reference any more, switch
+			 * to physical address.
+			 */
+			mem->hpas[i] = page_to_pfn(page) << PAGE_SHIFT;
+		}
+	}
+
+	list_add_rcu(&mem->next, &mm->context.iommu_group_mem_list);
+
+	mutex_unlock(&mem_list_mutex);
+
+	*pmem = mem;
+
+	return 0;
+
+free_exit:
+	/* free the references taken */
+	unpin_user_pages(mem->hpages, pinned);
+
+	vfree(mem->hpas);
+	kfree(mem);
+
+unlock_exit:
+	account_locked_vm(mm, locked_entries, false);
+
+	return ret;
+}
+
+long mm_iommu_new(struct mm_struct *mm, unsigned long ua, unsigned long entries,
+		struct mm_iommu_table_group_mem_t **pmem)
+{
+	return mm_iommu_do_alloc(mm, ua, entries, MM_IOMMU_TABLE_INVALID_HPA,
+			pmem);
+}
+EXPORT_SYMBOL_GPL(mm_iommu_new);
+
+long mm_iommu_newdev(struct mm_struct *mm, unsigned long ua,
+		unsigned long entries, unsigned long dev_hpa,
+		struct mm_iommu_table_group_mem_t **pmem)
+{
+	return mm_iommu_do_alloc(mm, ua, entries, dev_hpa, pmem);
+}
+EXPORT_SYMBOL_GPL(mm_iommu_newdev);
+
+static void mm_iommu_unpin(struct mm_iommu_table_group_mem_t *mem)
+{
+	long i;
+	struct page *page = NULL;
+
+	if (!mem->hpas)
+		return;
+
+	for (i = 0; i < mem->entries; ++i) {
+		if (!mem->hpas[i])
+			continue;
+
+		page = pfn_to_page(mem->hpas[i] >> PAGE_SHIFT);
+		if (!page)
+			continue;
+
+		if (mem->hpas[i] & MM_IOMMU_TABLE_GROUP_PAGE_DIRTY)
+			SetPageDirty(page);
+
+		unpin_user_page(page);
+
+		mem->hpas[i] = 0;
+	}
+}
+
+static void mm_iommu_do_free(struct mm_iommu_table_group_mem_t *mem)
+{
+
+	mm_iommu_unpin(mem);
+	vfree(mem->hpas);
+	kfree(mem);
+}
+
+static void mm_iommu_free(struct rcu_head *head)
+{
+	struct mm_iommu_table_group_mem_t *mem = container_of(head,
+			struct mm_iommu_table_group_mem_t, rcu);
+
+	mm_iommu_do_free(mem);
+}
+
+static void mm_iommu_release(struct mm_iommu_table_group_mem_t *mem)
+{
+	list_del_rcu(&mem->next);
+	call_rcu(&mem->rcu, mm_iommu_free);
+}
+
+long mm_iommu_put(struct mm_struct *mm, struct mm_iommu_table_group_mem_t *mem)
+{
+	long ret = 0;
+	unsigned long unlock_entries = 0;
+
+	mutex_lock(&mem_list_mutex);
+
+	if (mem->used == 0) {
+		ret = -ENOENT;
+		goto unlock_exit;
+	}
+
+	--mem->used;
+	/* There are still users, exit */
+	if (mem->used)
+		goto unlock_exit;
+
+	/* Are there still mappings? */
+	if (atomic64_cmpxchg(&mem->mapped, 1, 0) != 1) {
+		++mem->used;
+		ret = -EBUSY;
+		goto unlock_exit;
+	}
+
+	if (mem->dev_hpa == MM_IOMMU_TABLE_INVALID_HPA)
+		unlock_entries = mem->entries;
+
+	/* @mapped became 0 so now mappings are disabled, release the region */
+	mm_iommu_release(mem);
+
+unlock_exit:
+	mutex_unlock(&mem_list_mutex);
+
+	account_locked_vm(mm, unlock_entries, false);
+
+	return ret;
+}
+EXPORT_SYMBOL_GPL(mm_iommu_put);
+
+struct mm_iommu_table_group_mem_t *mm_iommu_lookup(struct mm_struct *mm,
+		unsigned long ua, unsigned long size)
+{
+	struct mm_iommu_table_group_mem_t *mem, *ret = NULL;
+
+	rcu_read_lock();
+	list_for_each_entry_rcu(mem, &mm->context.iommu_group_mem_list, next) {
+		if ((mem->ua <= ua) &&
+				(ua + size <= mem->ua +
+				 (mem->entries << PAGE_SHIFT))) {
+			ret = mem;
+			break;
+		}
+	}
+	rcu_read_unlock();
+
+	return ret;
+}
+EXPORT_SYMBOL_GPL(mm_iommu_lookup);
+
+struct mm_iommu_table_group_mem_t *mm_iommu_get(struct mm_struct *mm,
+		unsigned long ua, unsigned long entries)
+{
+	struct mm_iommu_table_group_mem_t *mem, *ret = NULL;
+
+	mutex_lock(&mem_list_mutex);
+
+	list_for_each_entry_rcu(mem, &mm->context.iommu_group_mem_list, next,
+				lockdep_is_held(&mem_list_mutex)) {
+		if ((mem->ua == ua) && (mem->entries == entries)) {
+			ret = mem;
+			++mem->used;
+			break;
+		}
+	}
+
+	mutex_unlock(&mem_list_mutex);
+
+	return ret;
+}
+EXPORT_SYMBOL_GPL(mm_iommu_get);
+
+long mm_iommu_ua_to_hpa(struct mm_iommu_table_group_mem_t *mem,
+		unsigned long ua, unsigned int pageshift, unsigned long *hpa)
+{
+	const long entry = (ua - mem->ua) >> PAGE_SHIFT;
+	u64 *va;
+
+	if (entry >= mem->entries)
+		return -EFAULT;
+
+	if (pageshift > mem->pageshift)
+		return -EFAULT;
+
+	if (!mem->hpas) {
+		*hpa = mem->dev_hpa + (ua - mem->ua);
+		return 0;
+	}
+
+	va = &mem->hpas[entry];
+	*hpa = (*va & MM_IOMMU_TABLE_GROUP_PAGE_MASK) | (ua & ~PAGE_MASK);
+
+	return 0;
+}
+EXPORT_SYMBOL_GPL(mm_iommu_ua_to_hpa);
+
+bool mm_iommu_is_devmem(struct mm_struct *mm, unsigned long hpa,
+		unsigned int pageshift, unsigned long *size)
+{
+	struct mm_iommu_table_group_mem_t *mem;
+	unsigned long end;
+
+	rcu_read_lock();
+	list_for_each_entry_rcu(mem, &mm->context.iommu_group_mem_list, next) {
+		if (mem->dev_hpa == MM_IOMMU_TABLE_INVALID_HPA)
+			continue;
+
+		end = mem->dev_hpa + (mem->entries << PAGE_SHIFT);
+		if ((mem->dev_hpa <= hpa) && (hpa < end)) {
+			/*
+			 * Since the IOMMU page size might be bigger than
+			 * PAGE_SIZE, the amount of preregistered memory
+			 * starting from @hpa might be smaller than 1<<pageshift
+			 * and the caller needs to distinguish this situation.
+			 */
+			*size = min(1UL << pageshift, end - hpa);
+			return true;
+		}
+	}
+	rcu_read_unlock();
+
+	return false;
+}
+EXPORT_SYMBOL_GPL(mm_iommu_is_devmem);
+
+long mm_iommu_mapped_inc(struct mm_iommu_table_group_mem_t *mem)
+{
+	if (atomic64_inc_not_zero(&mem->mapped))
+		return 0;
+
+	/* Last mm_iommu_put() has been called, no more mappings allowed() */
+	return -ENXIO;
+}
+EXPORT_SYMBOL_GPL(mm_iommu_mapped_inc);
+
+void mm_iommu_mapped_dec(struct mm_iommu_table_group_mem_t *mem)
+{
+	atomic64_add_unless(&mem->mapped, -1, 1);
+}
+EXPORT_SYMBOL_GPL(mm_iommu_mapped_dec);
+
+void mm_iommu_init(struct mm_struct *mm)
+{
+	INIT_LIST_HEAD_RCU(&mm->context.iommu_group_mem_list);
+}

linux-7.2.3/arch/powerpc/mm/book3s64/mmu_context.c

@@ -0,0 +1,345 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ *  MMU context allocation for 64-bit kernels.
+ *
+ *  Copyright (C) 2004 Anton Blanchard, IBM Corp. <anton@samba.org>
+ */
+
+#include <linux/sched.h>
+#include <linux/kernel.h>
+#include <linux/errno.h>
+#include <linux/string.h>
+#include <linux/types.h>
+#include <linux/mm.h>
+#include <linux/pkeys.h>
+#include <linux/spinlock.h>
+#include <linux/idr.h>
+#include <linux/export.h>
+#include <linux/gfp.h>
+#include <linux/slab.h>
+#include <linux/cpu.h>
+
+#include <asm/mmu_context.h>
+#include <asm/pgalloc.h>
+
+#include "internal.h"
+
+static DEFINE_IDA(mmu_context_ida);
+
+static int alloc_context_id(int min_id, int max_id)
+{
+	return ida_alloc_range(&mmu_context_ida, min_id, max_id, GFP_KERNEL);
+}
+
+#ifdef CONFIG_PPC_64S_HASH_MMU
+void __init hash__reserve_context_id(int id)
+{
+	int result = ida_alloc_range(&mmu_context_ida, id, id, GFP_KERNEL);
+
+	WARN(result != id, "mmu: Failed to reserve context id %d (rc %d)\n", id, result);
+}
+
+int hash__alloc_context_id(void)
+{
+	unsigned long max;
+
+	if (mmu_has_feature(MMU_FTR_68_BIT_VA))
+		max = MAX_USER_CONTEXT;
+	else
+		max = MAX_USER_CONTEXT_65BIT_VA;
+
+	return alloc_context_id(MIN_USER_CONTEXT, max);
+}
+EXPORT_SYMBOL_GPL(hash__alloc_context_id);
+#endif
+
+#ifdef CONFIG_PPC_64S_HASH_MMU
+static int realloc_context_ids(mm_context_t *ctx)
+{
+	int i, id;
+
+	/*
+	 * id 0 (aka. ctx->id) is special, we always allocate a new one, even if
+	 * there wasn't one allocated previously (which happens in the exec
+	 * case where ctx is newly allocated).
+	 *
+	 * We have to be a bit careful here. We must keep the existing ids in
+	 * the array, so that we can test if they're non-zero to decide if we
+	 * need to allocate a new one. However in case of error we must free the
+	 * ids we've allocated but *not* any of the existing ones (or risk a
+	 * UAF). That's why we decrement i at the start of the error handling
+	 * loop, to skip the id that we just tested but couldn't reallocate.
+	 */
+	for (i = 0; i < ARRAY_SIZE(ctx->extended_id); i++) {
+		if (i == 0 || ctx->extended_id[i]) {
+			id = hash__alloc_context_id();
+			if (id < 0)
+				goto error;
+
+			ctx->extended_id[i] = id;
+		}
+	}
+
+	/* The caller expects us to return id */
+	return ctx->id;
+
+error:
+	for (i--; i >= 0; i--) {
+		if (ctx->extended_id[i])
+			ida_free(&mmu_context_ida, ctx->extended_id[i]);
+	}
+
+	return id;
+}
+
+static int hash__init_new_context(struct mm_struct *mm)
+{
+	int index;
+
+	mm->context.hash_context = kmalloc_obj(struct hash_mm_context);
+	if (!mm->context.hash_context)
+		return -ENOMEM;
+
+	/*
+	 * The old code would re-promote on fork, we don't do that when using
+	 * slices as it could cause problem promoting slices that have been
+	 * forced down to 4K.
+	 *
+	 * For book3s we have MMU_NO_CONTEXT set to be ~0. Hence check
+	 * explicitly against context.id == 0. This ensures that we properly
+	 * initialize context slice details for newly allocated mm's (which will
+	 * have id == 0) and don't alter context slice inherited via fork (which
+	 * will have id != 0).
+	 *
+	 * We should not be calling init_new_context() on init_mm. Hence a
+	 * check against 0 is OK.
+	 */
+	if (mm->context.id == 0) {
+		memset(mm->context.hash_context, 0, sizeof(struct hash_mm_context));
+		slice_init_new_context_exec(mm);
+	} else {
+		/* This is fork. Copy hash_context details from current->mm */
+		memcpy(mm->context.hash_context, current->mm->context.hash_context, sizeof(struct hash_mm_context));
+#ifdef CONFIG_PPC_SUBPAGE_PROT
+		/* inherit subpage prot details if we have one. */
+		if (current->mm->context.hash_context->spt) {
+			mm->context.hash_context->spt = kmalloc_obj(struct subpage_prot_table);
+			if (!mm->context.hash_context->spt) {
+				kfree(mm->context.hash_context);
+				return -ENOMEM;
+			}
+		}
+#endif
+	}
+
+	index = realloc_context_ids(&mm->context);
+	if (index < 0) {
+#ifdef CONFIG_PPC_SUBPAGE_PROT
+		kfree(mm->context.hash_context->spt);
+#endif
+		kfree(mm->context.hash_context);
+		return index;
+	}
+
+	pkey_mm_init(mm);
+	return index;
+}
+
+void hash__setup_new_exec(void)
+{
+	slice_setup_new_exec();
+}
+#else
+static inline int hash__init_new_context(struct mm_struct *mm)
+{
+	BUILD_BUG();
+	return 0;
+}
+#endif
+
+static int radix__init_new_context(struct mm_struct *mm)
+{
+	unsigned long rts_field;
+	int index, max_id;
+
+	max_id = (1 << mmu_pid_bits) - 1;
+	index = alloc_context_id(mmu_base_pid, max_id);
+	if (index < 0)
+		return index;
+
+	/*
+	 * set the process table entry,
+	 */
+	rts_field = radix__get_tree_size();
+	process_tb[index].prtb0 = cpu_to_be64(rts_field | __pa(mm->pgd) | RADIX_PGD_INDEX_SIZE);
+
+	/*
+	 * Order the above store with subsequent update of the PID
+	 * register (at which point HW can start loading/caching
+	 * the entry) and the corresponding load by the MMU from
+	 * the L2 cache.
+	 */
+	asm volatile("ptesync;isync" : : : "memory");
+
+#ifdef CONFIG_PPC_64S_HASH_MMU
+	mm->context.hash_context = NULL;
+#endif
+
+	return index;
+}
+
+int init_new_context(struct task_struct *tsk, struct mm_struct *mm)
+{
+	int index;
+
+	if (radix_enabled())
+		index = radix__init_new_context(mm);
+	else
+		index = hash__init_new_context(mm);
+
+	if (index < 0)
+		return index;
+
+	mm->context.id = index;
+
+	mm->context.pte_frag = NULL;
+	mm->context.pmd_frag = NULL;
+#ifdef CONFIG_SPAPR_TCE_IOMMU
+	mm_iommu_init(mm);
+#endif
+	atomic_set(&mm->context.active_cpus, 0);
+	atomic_set(&mm->context.copros, 0);
+
+	return 0;
+}
+
+void __destroy_context(int context_id)
+{
+	ida_free(&mmu_context_ida, context_id);
+}
+EXPORT_SYMBOL_GPL(__destroy_context);
+
+static void destroy_contexts(mm_context_t *ctx)
+{
+	if (radix_enabled()) {
+		ida_free(&mmu_context_ida, ctx->id);
+	} else {
+#ifdef CONFIG_PPC_64S_HASH_MMU
+		int index, context_id;
+
+		for (index = 0; index < ARRAY_SIZE(ctx->extended_id); index++) {
+			context_id = ctx->extended_id[index];
+			if (context_id)
+				ida_free(&mmu_context_ida, context_id);
+		}
+		kfree(ctx->hash_context);
+#else
+		BUILD_BUG(); // radix_enabled() should be constant true
+#endif
+	}
+}
+
+static void pmd_frag_destroy(void *pmd_frag)
+{
+	int count;
+	struct ptdesc *ptdesc;
+
+	ptdesc = virt_to_ptdesc(pmd_frag);
+	/* drop all the pending references */
+	count = ((unsigned long)pmd_frag & ~PAGE_MASK) >> PMD_FRAG_SIZE_SHIFT;
+	/* We allow PTE_FRAG_NR fragments from a PTE page */
+	if (atomic_sub_and_test(PMD_FRAG_NR - count, &ptdesc->pt_frag_refcount)) {
+		pagetable_dtor(ptdesc);
+		pagetable_free(ptdesc);
+	}
+}
+
+static void destroy_pagetable_cache(struct mm_struct *mm)
+{
+	void *frag;
+
+	frag = mm->context.pte_frag;
+	if (frag)
+		pte_frag_destroy(frag);
+
+	frag = mm->context.pmd_frag;
+	if (frag)
+		pmd_frag_destroy(frag);
+	return;
+}
+
+void destroy_context(struct mm_struct *mm)
+{
+#ifdef CONFIG_SPAPR_TCE_IOMMU
+	WARN_ON_ONCE(!list_empty(&mm->context.iommu_group_mem_list));
+#endif
+	/*
+	 * For tasks which were successfully initialized we end up calling
+	 * arch_exit_mmap() which clears the process table entry. And
+	 * arch_exit_mmap() is called before the required fullmm TLB flush
+	 * which does a RIC=2 flush. Hence for an initialized task, we do clear
+	 * any cached process table entries.
+	 *
+	 * The condition below handles the error case during task init. We have
+	 * set the process table entry early and if we fail a task
+	 * initialization, we need to ensure the process table entry is zeroed.
+	 * We need not worry about process table entry caches because the task
+	 * never ran with the PID value.
+	 */
+	if (radix_enabled())
+		process_tb[mm->context.id].prtb0 = 0;
+	else
+		subpage_prot_free(mm);
+	destroy_contexts(&mm->context);
+	mm->context.id = MMU_NO_CONTEXT;
+}
+
+void arch_exit_mmap(struct mm_struct *mm)
+{
+	destroy_pagetable_cache(mm);
+
+	if (radix_enabled()) {
+		/*
+		 * Radix doesn't have a valid bit in the process table
+		 * entries. However we know that at least P9 implementation
+		 * will avoid caching an entry with an invalid RTS field,
+		 * and 0 is invalid. So this will do.
+		 *
+		 * This runs before the "fullmm" tlb flush in exit_mmap,
+		 * which does a RIC=2 tlbie to clear the process table
+		 * entry. See the "fullmm" comments in tlb-radix.c.
+		 *
+		 * No barrier required here after the store because
+		 * this process will do the invalidate, which starts with
+		 * ptesync.
+		 */
+		process_tb[mm->context.id].prtb0 = 0;
+	}
+}
+
+#ifdef CONFIG_PPC_RADIX_MMU
+void radix__switch_mmu_context(struct mm_struct *prev, struct mm_struct *next)
+{
+	mtspr(SPRN_PID, next->context.id);
+	isync();
+}
+#endif
+
+/**
+ * cleanup_cpu_mmu_context - Clean up MMU details for this CPU (newly offlined)
+ *
+ * This clears the CPU from mm_cpumask for all processes, and then flushes the
+ * local TLB to ensure TLB coherency in case the CPU is onlined again.
+ *
+ * KVM guest translations are not necessarily flushed here. If KVM started
+ * using mm_cpumask or the Linux APIs which do, this would have to be resolved.
+ */
+#ifdef CONFIG_HOTPLUG_CPU
+void cleanup_cpu_mmu_context(void)
+{
+	int cpu = smp_processor_id();
+
+	clear_tasks_mm_cpumask(cpu);
+	tlbiel_all();
+}
+#endif

linux-7.2.3/arch/powerpc/mm/book3s64/pgtable.c

@@ -0,0 +1,650 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Copyright 2015-2016, Aneesh Kumar K.V, IBM Corporation.
+ */
+
+#include <linux/sched.h>
+#include <linux/mm_types.h>
+#include <linux/memblock.h>
+#include <linux/memremap.h>
+#include <linux/pkeys.h>
+#include <linux/debugfs.h>
+#include <linux/proc_fs.h>
+#include <linux/page_table_check.h>
+
+#include <asm/pgalloc.h>
+#include <asm/tlb.h>
+#include <asm/trace.h>
+#include <asm/powernv.h>
+#include <asm/firmware.h>
+#include <asm/ultravisor.h>
+#include <asm/kexec.h>
+
+#include <mm/mmu_decl.h>
+#include <trace/events/thp.h>
+
+struct mmu_psize_def mmu_psize_defs[MMU_PAGE_COUNT];
+EXPORT_SYMBOL_GPL(mmu_psize_defs);
+
+#ifdef CONFIG_SPARSEMEM_VMEMMAP
+int mmu_vmemmap_psize = MMU_PAGE_4K;
+#endif
+
+unsigned long __pmd_frag_nr;
+EXPORT_SYMBOL(__pmd_frag_nr);
+unsigned long __pmd_frag_size_shift;
+EXPORT_SYMBOL(__pmd_frag_size_shift);
+
+#ifdef CONFIG_KFENCE
+extern bool kfence_early_init;
+static int __init parse_kfence_early_init(char *arg)
+{
+	int val;
+
+	if (get_option(&arg, &val))
+		kfence_early_init = !!val;
+	return 0;
+}
+early_param("kfence.sample_interval", parse_kfence_early_init);
+#endif
+
+#ifdef CONFIG_TRANSPARENT_HUGEPAGE
+/*
+ * This is called when relaxing access to a hugepage. It's also called in the page
+ * fault path when we don't hit any of the major fault cases, ie, a minor
+ * update of _PAGE_ACCESSED, _PAGE_DIRTY, etc... The generic code will have
+ * handled those two for us, we additionally deal with missing execute
+ * permission here on some processors
+ */
+int pmdp_set_access_flags(struct vm_area_struct *vma, unsigned long address,
+			  pmd_t *pmdp, pmd_t entry, int dirty)
+{
+	int changed;
+#ifdef CONFIG_DEBUG_VM
+	WARN_ON(!pmd_trans_huge(*pmdp));
+	assert_spin_locked(pmd_lockptr(vma->vm_mm, pmdp));
+#endif
+	changed = !pmd_same(*(pmdp), entry);
+	if (changed) {
+		/*
+		 * We can use MMU_PAGE_2M here, because only radix
+		 * path look at the psize.
+		 */
+		__ptep_set_access_flags(vma, pmdp_ptep(pmdp),
+					pmd_pte(entry), address, MMU_PAGE_2M);
+	}
+	return changed;
+}
+
+int pudp_set_access_flags(struct vm_area_struct *vma, unsigned long address,
+			  pud_t *pudp, pud_t entry, int dirty)
+{
+	int changed;
+#ifdef CONFIG_DEBUG_VM
+	assert_spin_locked(pud_lockptr(vma->vm_mm, pudp));
+#endif
+	changed = !pud_same(*(pudp), entry);
+	if (changed) {
+		/*
+		 * We can use MMU_PAGE_1G here, because only radix
+		 * path look at the psize.
+		 */
+		__ptep_set_access_flags(vma, pudp_ptep(pudp),
+					pud_pte(entry), address, MMU_PAGE_1G);
+	}
+	return changed;
+}
+
+
+bool pmdp_test_and_clear_young(struct vm_area_struct *vma,
+		unsigned long address, pmd_t *pmdp)
+{
+	return __pmdp_test_and_clear_young(vma->vm_mm, address, pmdp);
+}
+
+bool pudp_test_and_clear_young(struct vm_area_struct *vma,
+		unsigned long address, pud_t *pudp)
+{
+	return __pudp_test_and_clear_young(vma->vm_mm, address, pudp);
+}
+
+/*
+ * set a new huge pmd. We should not be called for updating
+ * an existing pmd entry. That should go via pmd_hugepage_update.
+ */
+void set_pmd_at(struct mm_struct *mm, unsigned long addr,
+		pmd_t *pmdp, pmd_t pmd)
+{
+#ifdef CONFIG_DEBUG_VM
+	/*
+	 * Make sure hardware valid bit is not set. We don't do
+	 * tlb flush for this update.
+	 */
+
+	WARN_ON(pte_hw_valid(pmd_pte(*pmdp)) && !pte_protnone(pmd_pte(*pmdp)));
+	assert_spin_locked(pmd_lockptr(mm, pmdp));
+	WARN_ON(!(pmd_leaf(pmd)));
+#endif
+	trace_hugepage_set_pmd(addr, pmd_val(pmd));
+	page_table_check_pmd_set(mm, addr, pmdp, pmd);
+	return set_pte_at_unchecked(mm, addr, pmdp_ptep(pmdp), pmd_pte(pmd));
+}
+
+void set_pud_at(struct mm_struct *mm, unsigned long addr,
+		pud_t *pudp, pud_t pud)
+{
+#ifdef CONFIG_DEBUG_VM
+	/*
+	 * Make sure hardware valid bit is not set. We don't do
+	 * tlb flush for this update.
+	 */
+
+	WARN_ON(pte_hw_valid(pud_pte(*pudp)));
+	assert_spin_locked(pud_lockptr(mm, pudp));
+	WARN_ON(!(pud_leaf(pud)));
+#endif
+	trace_hugepage_set_pud(addr, pud_val(pud));
+	page_table_check_pud_set(mm, addr, pudp, pud);
+	return set_pte_at_unchecked(mm, addr, pudp_ptep(pudp), pud_pte(pud));
+}
+
+/*
+ * We use this to invalidate a pmdp entry before switching from a
+ * hugepte to regular pmd entry.
+ */
+pmd_t pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
+		     pmd_t *pmdp)
+{
+	pmd_t old_pmd;
+
+	VM_WARN_ON_ONCE(!pmd_present(*pmdp));
+	old_pmd = __pmd(pmd_hugepage_update(vma->vm_mm, address, pmdp, _PAGE_PRESENT, _PAGE_INVALID));
+	flush_pmd_tlb_range(vma, address, address + HPAGE_PMD_SIZE);
+	page_table_check_pmd_clear(vma->vm_mm, address, old_pmd);
+
+	return old_pmd;
+}
+
+pud_t pudp_invalidate(struct vm_area_struct *vma, unsigned long address,
+		      pud_t *pudp)
+{
+	pud_t old_pud;
+
+	VM_WARN_ON_ONCE(!pud_present(*pudp));
+	old_pud = __pud(pud_hugepage_update(vma->vm_mm, address, pudp, _PAGE_PRESENT, _PAGE_INVALID));
+	flush_pud_tlb_range(vma, address, address + HPAGE_PUD_SIZE);
+	page_table_check_pud_clear(vma->vm_mm, address, old_pud);
+
+	return old_pud;
+}
+
+pmd_t pmdp_huge_get_and_clear_full(struct vm_area_struct *vma,
+				   unsigned long addr, pmd_t *pmdp, int full)
+{
+	pmd_t pmd;
+	bool was_present = pmd_present(*pmdp);
+
+	VM_BUG_ON(addr & ~HPAGE_PMD_MASK);
+	VM_BUG_ON(was_present && !pmd_trans_huge(*pmdp));
+	/*
+	 * Check pmdp_huge_get_and_clear() for non-present pmd case.
+	 */
+	pmd = pmdp_huge_get_and_clear(vma->vm_mm, addr, pmdp);
+	/*
+	 * if it not a fullmm flush, then we can possibly end up converting
+	 * this PMD pte entry to a regular level 0 PTE by a parallel page fault.
+	 * Make sure we flush the tlb in this case. TLB flush not needed for
+	 * non-present case.
+	 */
+	if (was_present && !full)
+		flush_pmd_tlb_range(vma, addr, addr + HPAGE_PMD_SIZE);
+	return pmd;
+}
+
+pud_t pudp_huge_get_and_clear_full(struct vm_area_struct *vma,
+				   unsigned long addr, pud_t *pudp, int full)
+{
+	pud_t pud;
+
+	VM_BUG_ON(addr & ~HPAGE_PMD_MASK);
+	VM_BUG_ON(!pud_present(*pudp));
+	pud = pudp_huge_get_and_clear(vma->vm_mm, addr, pudp);
+	/*
+	 * if it not a fullmm flush, then we can possibly end up converting
+	 * this PMD pte entry to a regular level 0 PTE by a parallel page fault.
+	 * Make sure we flush the tlb in this case.
+	 */
+	if (!full)
+		flush_pud_tlb_range(vma, addr, addr + HPAGE_PUD_SIZE);
+	return pud;
+}
+
+static pmd_t pmd_set_protbits(pmd_t pmd, pgprot_t pgprot)
+{
+	return __pmd(pmd_val(pmd) | pgprot_val(pgprot));
+}
+
+static pud_t pud_set_protbits(pud_t pud, pgprot_t pgprot)
+{
+	return __pud(pud_val(pud) | pgprot_val(pgprot));
+}
+
+/*
+ * At some point we should be able to get rid of
+ * pmd_mkhuge() and mk_huge_pmd() when we update all the
+ * other archs to mark the pmd huge in pfn_pmd()
+ */
+pmd_t pfn_pmd(unsigned long pfn, pgprot_t pgprot)
+{
+	unsigned long pmdv;
+
+	pmdv = (pfn << PAGE_SHIFT) & PTE_RPN_MASK;
+
+	return __pmd_mkhuge(pmd_set_protbits(__pmd(pmdv), pgprot));
+}
+
+pud_t pfn_pud(unsigned long pfn, pgprot_t pgprot)
+{
+	unsigned long pudv;
+
+	pudv = (pfn << PAGE_SHIFT) & PTE_RPN_MASK;
+
+	return __pud_mkhuge(pud_set_protbits(__pud(pudv), pgprot));
+}
+
+pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
+{
+	unsigned long pmdv;
+
+	pmdv = pmd_val(pmd);
+	pmdv &= _HPAGE_CHG_MASK;
+	return pmd_set_protbits(__pmd(pmdv), newprot);
+}
+
+pud_t pud_modify(pud_t pud, pgprot_t newprot)
+{
+	unsigned long pudv;
+
+	pudv = pud_val(pud);
+	pudv &= _HPAGE_CHG_MASK;
+	return pud_set_protbits(__pud(pudv), newprot);
+}
+#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
+
+/* For use by kexec, called with MMU off */
+notrace void mmu_cleanup_all(void)
+{
+	if (radix_enabled())
+		radix__mmu_cleanup_all();
+	else if (mmu_hash_ops.hpte_clear_all)
+		mmu_hash_ops.hpte_clear_all();
+
+	reset_sprs();
+}
+
+#ifdef CONFIG_MEMORY_HOTPLUG
+int __meminit create_section_mapping(unsigned long start, unsigned long end,
+				     int nid, pgprot_t prot)
+{
+	if (radix_enabled())
+		return radix__create_section_mapping(start, end, nid, prot);
+
+	return hash__create_section_mapping(start, end, nid, prot);
+}
+
+int __meminit remove_section_mapping(unsigned long start, unsigned long end)
+{
+	if (radix_enabled())
+		return radix__remove_section_mapping(start, end);
+
+	return hash__remove_section_mapping(start, end);
+}
+#endif /* CONFIG_MEMORY_HOTPLUG */
+
+void __init mmu_partition_table_init(void)
+{
+	unsigned long patb_size = 1UL << PATB_SIZE_SHIFT;
+	unsigned long ptcr;
+
+	/* Initialize the Partition Table with no entries */
+	partition_tb = memblock_alloc_or_panic(patb_size, patb_size);
+	ptcr = __pa(partition_tb) | (PATB_SIZE_SHIFT - 12);
+	set_ptcr_when_no_uv(ptcr);
+	powernv_set_nmmu_ptcr(ptcr);
+}
+
+static void flush_partition(unsigned int lpid, bool radix)
+{
+	if (radix) {
+		radix__flush_all_lpid(lpid);
+		radix__flush_all_lpid_guest(lpid);
+	} else {
+		asm volatile("ptesync" : : : "memory");
+		asm volatile(PPC_TLBIE_5(%0,%1,2,0,0) : :
+			     "r" (TLBIEL_INVAL_SET_LPID), "r" (lpid));
+		/* do we need fixup here ?*/
+		asm volatile("eieio; tlbsync; ptesync" : : : "memory");
+		trace_tlbie(lpid, 0, TLBIEL_INVAL_SET_LPID, lpid, 2, 0, 0);
+	}
+}
+
+void mmu_partition_table_set_entry(unsigned int lpid, unsigned long dw0,
+				  unsigned long dw1, bool flush)
+{
+	unsigned long old = be64_to_cpu(partition_tb[lpid].patb0);
+
+	/*
+	 * When ultravisor is enabled, the partition table is stored in secure
+	 * memory and can only be accessed doing an ultravisor call. However, we
+	 * maintain a copy of the partition table in normal memory to allow Nest
+	 * MMU translations to occur (for normal VMs).
+	 *
+	 * Therefore, here we always update partition_tb, regardless of whether
+	 * we are running under an ultravisor or not.
+	 */
+	partition_tb[lpid].patb0 = cpu_to_be64(dw0);
+	partition_tb[lpid].patb1 = cpu_to_be64(dw1);
+
+	/*
+	 * If ultravisor is enabled, we do an ultravisor call to register the
+	 * partition table entry (PATE), which also do a global flush of TLBs
+	 * and partition table caches for the lpid. Otherwise, just do the
+	 * flush. The type of flush (hash or radix) depends on what the previous
+	 * use of the partition ID was, not the new use.
+	 */
+	if (firmware_has_feature(FW_FEATURE_ULTRAVISOR)) {
+		uv_register_pate(lpid, dw0, dw1);
+		pr_info("PATE registered by ultravisor: dw0 = 0x%lx, dw1 = 0x%lx\n",
+			dw0, dw1);
+	} else if (flush) {
+		/*
+		 * Boot does not need to flush, because MMU is off and each
+		 * CPU does a tlbiel_all() before switching them on, which
+		 * flushes everything.
+		 */
+		flush_partition(lpid, (old & PATB_HR));
+	}
+}
+EXPORT_SYMBOL_GPL(mmu_partition_table_set_entry);
+
+static pmd_t *get_pmd_from_cache(struct mm_struct *mm)
+{
+	void *pmd_frag, *ret;
+
+	if (PMD_FRAG_NR == 1)
+		return NULL;
+
+	spin_lock(&mm->page_table_lock);
+	ret = mm->context.pmd_frag;
+	if (ret) {
+		pmd_frag = ret + PMD_FRAG_SIZE;
+		/*
+		 * If we have taken up all the fragments mark PTE page NULL
+		 */
+		if (((unsigned long)pmd_frag & ~PAGE_MASK) == 0)
+			pmd_frag = NULL;
+		mm->context.pmd_frag = pmd_frag;
+	}
+	spin_unlock(&mm->page_table_lock);
+	return (pmd_t *)ret;
+}
+
+static pmd_t *__alloc_for_pmdcache(struct mm_struct *mm)
+{
+	void *ret = NULL;
+	struct ptdesc *ptdesc;
+	gfp_t gfp = GFP_KERNEL_ACCOUNT | __GFP_ZERO;
+
+	if (mm == &init_mm)
+		gfp &= ~__GFP_ACCOUNT;
+	ptdesc = pagetable_alloc(gfp, 0);
+	if (!ptdesc)
+		return NULL;
+	if (!pagetable_pmd_ctor(mm, ptdesc)) {
+		pagetable_free(ptdesc);
+		return NULL;
+	}
+
+	atomic_set(&ptdesc->pt_frag_refcount, 1);
+
+	ret = ptdesc_address(ptdesc);
+	/*
+	 * if we support only one fragment just return the
+	 * allocated page.
+	 */
+	if (PMD_FRAG_NR == 1)
+		return ret;
+
+	spin_lock(&mm->page_table_lock);
+	/*
+	 * If we find ptdesc_page set, we return
+	 * the allocated page with single fragment
+	 * count.
+	 */
+	if (likely(!mm->context.pmd_frag)) {
+		atomic_set(&ptdesc->pt_frag_refcount, PMD_FRAG_NR);
+		mm->context.pmd_frag = ret + PMD_FRAG_SIZE;
+	}
+	spin_unlock(&mm->page_table_lock);
+
+	return (pmd_t *)ret;
+}
+
+pmd_t *pmd_fragment_alloc(struct mm_struct *mm, unsigned long vmaddr)
+{
+	pmd_t *pmd;
+
+	pmd = get_pmd_from_cache(mm);
+	if (pmd)
+		return pmd;
+
+	return __alloc_for_pmdcache(mm);
+}
+
+void pmd_fragment_free(unsigned long *pmd)
+{
+	struct ptdesc *ptdesc = virt_to_ptdesc(pmd);
+
+	if (pagetable_is_reserved(ptdesc))
+		return free_reserved_ptdesc(ptdesc);
+
+	BUG_ON(atomic_read(&ptdesc->pt_frag_refcount) <= 0);
+	if (atomic_dec_and_test(&ptdesc->pt_frag_refcount)) {
+		pagetable_dtor(ptdesc);
+		pagetable_free(ptdesc);
+	}
+}
+
+static inline void pgtable_free(void *table, int index)
+{
+	switch (index) {
+	case PTE_INDEX:
+		pte_fragment_free(table, 0);
+		break;
+	case PMD_INDEX:
+		pmd_fragment_free(table);
+		break;
+	case PUD_INDEX:
+		__pud_free(table);
+		break;
+		/* We don't free pgd table via RCU callback */
+	default:
+		BUG();
+	}
+}
+
+void pgtable_free_tlb(struct mmu_gather *tlb, void *table, int index)
+{
+	unsigned long pgf = (unsigned long)table;
+
+	BUG_ON(index > MAX_PGTABLE_INDEX_SIZE);
+	pgf |= index;
+	tlb_remove_table(tlb, (void *)pgf);
+}
+
+void __tlb_remove_table(void *_table)
+{
+	void *table = (void *)((unsigned long)_table & ~MAX_PGTABLE_INDEX_SIZE);
+	unsigned int index = (unsigned long)_table & MAX_PGTABLE_INDEX_SIZE;
+
+	return pgtable_free(table, index);
+}
+
+#ifdef CONFIG_PROC_FS
+atomic_long_t direct_pages_count[MMU_PAGE_COUNT];
+
+void arch_report_meminfo(struct seq_file *m)
+{
+	seq_printf(m, "DirectMap4k:    %8lu kB\n",
+		   atomic_long_read(&direct_pages_count[MMU_PAGE_4K]) << 2);
+	seq_printf(m, "DirectMap64k:   %8lu kB\n",
+		   atomic_long_read(&direct_pages_count[MMU_PAGE_64K]) << 6);
+	if (radix_enabled()) {
+		seq_printf(m, "DirectMap2M:    %8lu kB\n",
+			   atomic_long_read(&direct_pages_count[MMU_PAGE_2M]) << 11);
+		seq_printf(m, "DirectMap1G:    %8lu kB\n",
+			   atomic_long_read(&direct_pages_count[MMU_PAGE_1G]) << 20);
+	} else {
+		seq_printf(m, "DirectMap16M:   %8lu kB\n",
+			   atomic_long_read(&direct_pages_count[MMU_PAGE_16M]) << 14);
+		seq_printf(m, "DirectMap16G:   %8lu kB\n",
+			   atomic_long_read(&direct_pages_count[MMU_PAGE_16G]) << 24);
+	}
+}
+#endif /* CONFIG_PROC_FS */
+
+pte_t ptep_modify_prot_start(struct vm_area_struct *vma, unsigned long addr,
+			     pte_t *ptep)
+{
+	unsigned long pte_val;
+
+	/*
+	 * Clear the _PAGE_PRESENT so that no hardware parallel update is
+	 * possible. Also keep the pte_present true so that we don't take
+	 * wrong fault.
+	 */
+	pte_val = pte_update(vma->vm_mm, addr, ptep, _PAGE_PRESENT, _PAGE_INVALID, 0);
+
+	return __pte(pte_val);
+
+}
+
+void ptep_modify_prot_commit(struct vm_area_struct *vma, unsigned long addr,
+			     pte_t *ptep, pte_t old_pte, pte_t pte)
+{
+	if (radix_enabled())
+		return radix__ptep_modify_prot_commit(vma, addr,
+						      ptep, old_pte, pte);
+	set_pte_at_unchecked(vma->vm_mm, addr, ptep, pte);
+}
+
+#ifdef CONFIG_TRANSPARENT_HUGEPAGE
+/*
+ * For hash translation mode, we use the deposited table to store hash slot
+ * information and they are stored at PTRS_PER_PMD offset from related pmd
+ * location. Hence a pmd move requires deposit and withdraw.
+ *
+ * For radix translation with split pmd ptl, we store the deposited table in the
+ * pmd page. Hence if we have different pmd page we need to withdraw during pmd
+ * move.
+ *
+ * With hash we use deposited table always irrespective of anon or not.
+ * With radix we use deposited table only for anonymous mapping.
+ */
+int pmd_move_must_withdraw(struct spinlock *new_pmd_ptl,
+			   struct spinlock *old_pmd_ptl,
+			   struct vm_area_struct *vma)
+{
+	if (radix_enabled())
+		return (new_pmd_ptl != old_pmd_ptl) && vma_is_anonymous(vma);
+
+	return true;
+}
+#endif
+
+/*
+ * Does the CPU support tlbie?
+ */
+bool tlbie_capable __read_mostly = IS_ENABLED(CONFIG_PPC_RADIX_BROADCAST_TLBIE);
+EXPORT_SYMBOL(tlbie_capable);
+
+/*
+ * Should tlbie be used for management of CPU TLBs, for kernel and process
+ * address spaces? tlbie may still be used for nMMU accelerators, and for KVM
+ * guest address spaces.
+ */
+bool tlbie_enabled __read_mostly = IS_ENABLED(CONFIG_PPC_RADIX_BROADCAST_TLBIE);
+
+static int __init setup_disable_tlbie(char *str)
+{
+	if (!radix_enabled()) {
+		pr_err("disable_tlbie: Unable to disable TLBIE with Hash MMU.\n");
+		return 1;
+	}
+
+	tlbie_capable = false;
+	tlbie_enabled = false;
+
+        return 1;
+}
+__setup("disable_tlbie", setup_disable_tlbie);
+
+static int __init pgtable_debugfs_setup(void)
+{
+	if (!tlbie_capable)
+		return 0;
+
+	/*
+	 * There is no locking vs tlb flushing when changing this value.
+	 * The tlb flushers will see one value or another, and use either
+	 * tlbie or tlbiel with IPIs. In both cases the TLBs will be
+	 * invalidated as expected.
+	 */
+	debugfs_create_bool("tlbie_enabled", 0600,
+			arch_debugfs_dir,
+			&tlbie_enabled);
+
+	return 0;
+}
+arch_initcall(pgtable_debugfs_setup);
+
+#if defined(CONFIG_ZONE_DEVICE) && defined(CONFIG_ARCH_HAS_MEMREMAP_COMPAT_ALIGN)
+/*
+ * Override the generic version in mm/memremap.c.
+ *
+ * With hash translation, the direct-map range is mapped with just one
+ * page size selected by htab_init_page_sizes(). Consult
+ * mmu_psize_defs[] to determine the minimum page size alignment.
+*/
+unsigned long memremap_compat_align(void)
+{
+	if (!radix_enabled()) {
+		unsigned int shift = mmu_psize_defs[mmu_linear_psize].shift;
+		return max(SUBSECTION_SIZE, 1UL << shift);
+	}
+
+	return SUBSECTION_SIZE;
+}
+EXPORT_SYMBOL_GPL(memremap_compat_align);
+#endif
+
+pgprot_t vm_get_page_prot(vm_flags_t vm_flags)
+{
+	unsigned long prot;
+
+	/* Radix supports execute-only, but protection_map maps X -> RX */
+	if (!radix_enabled() && ((vm_flags & VM_ACCESS_FLAGS) == VM_EXEC))
+		vm_flags |= VM_READ;
+
+	prot = pgprot_val(protection_map[vm_flags & (VM_ACCESS_FLAGS | VM_SHARED)]);
+
+	if (vm_flags & VM_SAO)
+		prot |= _PAGE_SAO;
+
+#ifdef CONFIG_PPC_MEM_KEYS
+	prot |= vmflag_to_pte_pkey_bits(vm_flags);
+#endif
+
+	return __pgprot(prot);
+}
+EXPORT_SYMBOL(vm_get_page_prot);

linux-7.2.3/arch/powerpc/mm/book3s64/pkeys.c

@@ -0,0 +1,470 @@
+// SPDX-License-Identifier: GPL-2.0+
+/*
+ * PowerPC Memory Protection Keys management
+ *
+ * Copyright 2017, Ram Pai, IBM Corporation.
+ */
+
+#include <asm/mman.h>
+#include <asm/mmu_context.h>
+#include <asm/mmu.h>
+#include <asm/setup.h>
+#include <asm/smp.h>
+#include <asm/firmware.h>
+
+#include <linux/pkeys.h>
+#include <linux/of_fdt.h>
+
+
+int  num_pkey;		/* Max number of pkeys supported */
+/*
+ *  Keys marked in the reservation list cannot be allocated by  userspace
+ */
+u32 reserved_allocation_mask __ro_after_init;
+
+/* Bits set for the initially allocated keys */
+static u32 initial_allocation_mask __ro_after_init;
+
+/*
+ * Even if we allocate keys with sys_pkey_alloc(), we need to make sure
+ * other thread still find the access denied using the same keys.
+ */
+u64 default_amr __ro_after_init  = ~0x0UL;
+u64 default_iamr __ro_after_init = 0x5555555555555555UL;
+u64 default_uamor __ro_after_init;
+EXPORT_SYMBOL(default_amr);
+/*
+ * Key used to implement PROT_EXEC mmap. Denies READ/WRITE
+ * We pick key 2 because 0 is special key and 1 is reserved as per ISA.
+ */
+static int execute_only_key = 2;
+static bool pkey_execute_disable_supported;
+
+
+#define AMR_BITS_PER_PKEY 2
+#define AMR_RD_BIT 0x1UL
+#define AMR_WR_BIT 0x2UL
+#define IAMR_EX_BIT 0x1UL
+#define PKEY_REG_BITS (sizeof(u64) * 8)
+#define pkeyshift(pkey) (PKEY_REG_BITS - ((pkey+1) * AMR_BITS_PER_PKEY))
+
+static int __init dt_scan_storage_keys(unsigned long node,
+				       const char *uname, int depth,
+				       void *data)
+{
+	const char *type = of_get_flat_dt_prop(node, "device_type", NULL);
+	const __be32 *prop;
+	int *pkeys_total = (int *) data;
+
+	/* We are scanning "cpu" nodes only */
+	if (type == NULL || strcmp(type, "cpu") != 0)
+		return 0;
+
+	prop = of_get_flat_dt_prop(node, "ibm,processor-storage-keys", NULL);
+	if (!prop)
+		return 0;
+	*pkeys_total = be32_to_cpu(prop[0]);
+	return 1;
+}
+
+static int __init scan_pkey_feature(void)
+{
+	int ret;
+	int pkeys_total = 0;
+
+	/*
+	 * Pkey is not supported with Radix translation.
+	 */
+	if (early_radix_enabled())
+		return 0;
+
+	ret = of_scan_flat_dt(dt_scan_storage_keys, &pkeys_total);
+	if (ret == 0) {
+		/*
+		 * Let's assume 32 pkeys on P8/P9 bare metal, if its not defined by device
+		 * tree. We make this exception since some version of skiboot forgot to
+		 * expose this property on power8/9.
+		 */
+		if (!firmware_has_feature(FW_FEATURE_LPAR)) {
+			unsigned long pvr = mfspr(SPRN_PVR);
+
+			if (PVR_VER(pvr) == PVR_POWER8 || PVR_VER(pvr) == PVR_POWER8E ||
+			    PVR_VER(pvr) == PVR_POWER8NVL || PVR_VER(pvr) == PVR_POWER9 ||
+			    PVR_VER(pvr) == PVR_HX_C2000)
+				pkeys_total = 32;
+		}
+	}
+
+#ifdef CONFIG_PPC_MEM_KEYS
+	/*
+	 * Adjust the upper limit, based on the number of bits supported by
+	 * arch-neutral code.
+	 */
+	pkeys_total = min_t(int, pkeys_total,
+			    ((ARCH_VM_PKEY_FLAGS >> VM_PKEY_SHIFT) + 1));
+#endif
+	return pkeys_total;
+}
+
+void __init pkey_early_init_devtree(void)
+{
+	int pkeys_total, i;
+
+#ifdef CONFIG_PPC_MEM_KEYS
+	/*
+	 * We define PKEY_DISABLE_EXECUTE in addition to the arch-neutral
+	 * generic defines for PKEY_DISABLE_ACCESS and PKEY_DISABLE_WRITE.
+	 * Ensure that the bits a distinct.
+	 */
+	BUILD_BUG_ON(PKEY_DISABLE_EXECUTE &
+		     (PKEY_DISABLE_ACCESS | PKEY_DISABLE_WRITE));
+
+	/*
+	 * pkey_to_vmflag_bits() assumes that the pkey bits are contiguous
+	 * in the vmaflag. Make sure that is really the case.
+	 */
+	BUILD_BUG_ON(__builtin_clzl(ARCH_VM_PKEY_FLAGS >> VM_PKEY_SHIFT) +
+		     __builtin_popcountl(ARCH_VM_PKEY_FLAGS >> VM_PKEY_SHIFT)
+				!= (sizeof(u64) * BITS_PER_BYTE));
+#endif
+	/*
+	 * Only P7 and above supports SPRN_AMR update with MSR[PR] = 1
+	 */
+	if (!early_cpu_has_feature(CPU_FTR_ARCH_206))
+		return;
+
+	/* scan the device tree for pkey feature */
+	pkeys_total = scan_pkey_feature();
+	if (!pkeys_total)
+		goto out;
+
+	/* Allow all keys to be modified by default */
+	default_uamor = ~0x0UL;
+
+	cur_cpu_spec->mmu_features |= MMU_FTR_PKEY;
+
+	/*
+	 * The device tree cannot be relied to indicate support for
+	 * execute_disable support. Instead we use a PVR check.
+	 */
+	if (pvr_version_is(PVR_POWER7) || pvr_version_is(PVR_POWER7p))
+		pkey_execute_disable_supported = false;
+	else
+		pkey_execute_disable_supported = true;
+
+#ifdef CONFIG_PPC_4K_PAGES
+	/*
+	 * The OS can manage only 8 pkeys due to its inability to represent them
+	 * in the Linux 4K PTE. Mark all other keys reserved.
+	 */
+	num_pkey = min(8, pkeys_total);
+#else
+	num_pkey = pkeys_total;
+#endif
+
+	if (unlikely(num_pkey <= execute_only_key) || !pkey_execute_disable_supported) {
+		/*
+		 * Insufficient number of keys to support
+		 * execute only key. Mark it unavailable.
+		 */
+		execute_only_key = -1;
+	} else {
+		/*
+		 * Mark the execute_only_pkey as not available for
+		 * user allocation via pkey_alloc.
+		 */
+		reserved_allocation_mask |= (0x1 << execute_only_key);
+
+		/*
+		 * Deny READ/WRITE for execute_only_key.
+		 * Allow execute in IAMR.
+		 */
+		default_amr  |= (0x3ul << pkeyshift(execute_only_key));
+		default_iamr &= ~(0x1ul << pkeyshift(execute_only_key));
+
+		/*
+		 * Clear the uamor bits for this key.
+		 */
+		default_uamor &= ~(0x3ul << pkeyshift(execute_only_key));
+	}
+
+	if (unlikely(num_pkey <= 3)) {
+		/*
+		 * Insufficient number of keys to support
+		 * KUAP/KUEP feature.
+		 */
+		disable_kuep = true;
+		disable_kuap = true;
+		WARN(1, "Disabling kernel user protection due to low (%d) max supported keys\n", num_pkey);
+	} else {
+		/*  handle key which is used by kernel for KAUP */
+		reserved_allocation_mask |= (0x1 << 3);
+		/*
+		 * Mark access for kup_key in default amr so that
+		 * we continue to operate with that AMR in
+		 * copy_to/from_user().
+		 */
+		default_amr   &= ~(0x3ul << pkeyshift(3));
+		default_iamr  &= ~(0x1ul << pkeyshift(3));
+		default_uamor &= ~(0x3ul << pkeyshift(3));
+	}
+
+	/*
+	 * Allow access for only key 0. And prevent any other modification.
+	 */
+	default_amr   &= ~(0x3ul << pkeyshift(0));
+	default_iamr  &= ~(0x1ul << pkeyshift(0));
+	default_uamor &= ~(0x3ul << pkeyshift(0));
+	/*
+	 * key 0 is special in that we want to consider it an allocated
+	 * key which is preallocated. We don't allow changing AMR bits
+	 * w.r.t key 0. But one can pkey_free(key0)
+	 */
+	initial_allocation_mask |= (0x1 << 0);
+
+	/*
+	 * key 1 is recommended not to be used. PowerISA(3.0) page 1015,
+	 * programming note.
+	 */
+	reserved_allocation_mask |= (0x1 << 1);
+	default_uamor &= ~(0x3ul << pkeyshift(1));
+
+	/*
+	 * Prevent the usage of OS reserved keys. Update UAMOR
+	 * for those keys. Also mark the rest of the bits in the
+	 * 32 bit mask as reserved.
+	 */
+	for (i = num_pkey; i < 32 ; i++) {
+		reserved_allocation_mask |= (0x1 << i);
+		default_uamor &= ~(0x3ul << pkeyshift(i));
+	}
+	/*
+	 * Prevent the allocation of reserved keys too.
+	 */
+	initial_allocation_mask |= reserved_allocation_mask;
+
+	pr_info("Enabling pkeys with max key count %d\n", num_pkey);
+out:
+	/*
+	 * Setup uamor on boot cpu
+	 */
+	mtspr(SPRN_UAMOR, default_uamor);
+
+	return;
+}
+
+#ifdef CONFIG_PPC_KUEP
+void setup_kuep(bool disabled)
+{
+	if (disabled)
+		return;
+	/*
+	 * On hash if PKEY feature is not enabled, disable KUAP too.
+	 */
+	if (!early_radix_enabled() && !early_mmu_has_feature(MMU_FTR_PKEY))
+		return;
+
+	if (smp_processor_id() == boot_cpuid) {
+		pr_info("Activating Kernel Userspace Execution Prevention\n");
+		cur_cpu_spec->mmu_features |= MMU_FTR_BOOK3S_KUEP;
+	}
+
+	/*
+	 * Radix always uses key0 of the IAMR to determine if an access is
+	 * allowed. We set bit 0 (IBM bit 1) of key0, to prevent instruction
+	 * fetch.
+	 */
+	mtspr(SPRN_IAMR, AMR_KUEP_BLOCKED);
+	isync();
+}
+#endif
+
+#ifdef CONFIG_PPC_KUAP
+void setup_kuap(bool disabled)
+{
+	if (disabled)
+		return;
+	/*
+	 * On hash if PKEY feature is not enabled, disable KUAP too.
+	 */
+	if (!early_radix_enabled() && !early_mmu_has_feature(MMU_FTR_PKEY))
+		return;
+
+	if (smp_processor_id() == boot_cpuid) {
+		pr_info("Activating Kernel Userspace Access Prevention\n");
+		cur_cpu_spec->mmu_features |= MMU_FTR_KUAP;
+	}
+
+	/*
+	 * Set the default kernel AMR values on all cpus.
+	 */
+	mtspr(SPRN_AMR, AMR_KUAP_BLOCKED);
+	isync();
+}
+#endif
+
+#ifdef CONFIG_PPC_MEM_KEYS
+void pkey_mm_init(struct mm_struct *mm)
+{
+	if (!mmu_has_feature(MMU_FTR_PKEY))
+		return;
+	mm_pkey_allocation_map(mm) = initial_allocation_mask;
+	mm->context.execute_only_pkey = execute_only_key;
+}
+
+static inline void init_amr(int pkey, u8 init_bits)
+{
+	u64 new_amr_bits = (((u64)init_bits & 0x3UL) << pkeyshift(pkey));
+	u64 old_amr = current_thread_amr() & ~((u64)(0x3ul) << pkeyshift(pkey));
+
+	current->thread.regs->amr = old_amr | new_amr_bits;
+}
+
+static inline void init_iamr(int pkey, u8 init_bits)
+{
+	u64 new_iamr_bits = (((u64)init_bits & 0x1UL) << pkeyshift(pkey));
+	u64 old_iamr = current_thread_iamr() & ~((u64)(0x1ul) << pkeyshift(pkey));
+
+	if (!likely(pkey_execute_disable_supported))
+		return;
+
+	current->thread.regs->iamr = old_iamr | new_iamr_bits;
+}
+
+/*
+ * Set the access rights in AMR IAMR and UAMOR registers for @pkey to that
+ * specified in @init_val.
+ */
+int __arch_set_user_pkey_access(int pkey, unsigned long init_val)
+{
+	u64 new_amr_bits = 0x0ul;
+	u64 new_iamr_bits = 0x0ul;
+	u64 pkey_bits, uamor_pkey_bits;
+
+	/*
+	 * Check whether the key is disabled by UAMOR.
+	 */
+	pkey_bits = 0x3ul << pkeyshift(pkey);
+	uamor_pkey_bits = (default_uamor & pkey_bits);
+
+	/*
+	 * Both the bits in UAMOR corresponding to the key should be set
+	 */
+	if (uamor_pkey_bits != pkey_bits)
+		return -EINVAL;
+
+	if (init_val & PKEY_DISABLE_EXECUTE) {
+		if (!pkey_execute_disable_supported)
+			return -EINVAL;
+		new_iamr_bits |= IAMR_EX_BIT;
+	}
+	init_iamr(pkey, new_iamr_bits);
+
+	/* Set the bits we need in AMR: */
+	if (init_val & PKEY_DISABLE_ACCESS)
+		new_amr_bits |= AMR_RD_BIT | AMR_WR_BIT;
+	else if (init_val & PKEY_DISABLE_WRITE)
+		new_amr_bits |= AMR_WR_BIT;
+
+	init_amr(pkey, new_amr_bits);
+	return 0;
+}
+
+int execute_only_pkey(struct mm_struct *mm)
+{
+	return mm->context.execute_only_pkey;
+}
+
+static inline bool vma_is_pkey_exec_only(struct vm_area_struct *vma)
+{
+	/* Do this check first since the vm_flags should be hot */
+	if ((vma->vm_flags & VM_ACCESS_FLAGS) != VM_EXEC)
+		return false;
+
+	return (vma_pkey(vma) == vma->vm_mm->context.execute_only_pkey);
+}
+
+/*
+ * This should only be called for *plain* mprotect calls.
+ */
+int __arch_override_mprotect_pkey(struct vm_area_struct *vma, int prot,
+				  int pkey)
+{
+	/*
+	 * If the currently associated pkey is execute-only, but the requested
+	 * protection is not execute-only, move it back to the default pkey.
+	 */
+	if (vma_is_pkey_exec_only(vma) && (prot != PROT_EXEC))
+		return 0;
+
+	/*
+	 * The requested protection is execute-only. Hence let's use an
+	 * execute-only pkey.
+	 */
+	if (prot == PROT_EXEC) {
+		pkey = execute_only_pkey(vma->vm_mm);
+		if (pkey > 0)
+			return pkey;
+	}
+
+	/* Nothing to override. */
+	return vma_pkey(vma);
+}
+
+static bool pkey_access_permitted(int pkey, bool write, bool execute)
+{
+	int pkey_shift;
+	u64 amr;
+
+	pkey_shift = pkeyshift(pkey);
+	if (execute)
+		return !(current_thread_iamr() & (IAMR_EX_BIT << pkey_shift));
+
+	amr = current_thread_amr();
+	if (write)
+		return !(amr & (AMR_WR_BIT << pkey_shift));
+
+	return !(amr & (AMR_RD_BIT << pkey_shift));
+}
+
+bool arch_pte_access_permitted(u64 pte, bool write, bool execute)
+{
+	if (!mmu_has_feature(MMU_FTR_PKEY))
+		return true;
+
+	return pkey_access_permitted(pte_to_pkey_bits(pte), write, execute);
+}
+
+/*
+ * We only want to enforce protection keys on the current thread because we
+ * effectively have no access to AMR/IAMR for other threads or any way to tell
+ * which AMR/IAMR in a threaded process we could use.
+ *
+ * So do not enforce things if the VMA is not from the current mm, or if we are
+ * in a kernel thread.
+ */
+bool arch_vma_access_permitted(struct vm_area_struct *vma, bool write,
+			       bool execute, bool foreign)
+{
+	if (!mmu_has_feature(MMU_FTR_PKEY))
+		return true;
+	/*
+	 * Do not enforce our key-permissions on a foreign vma.
+	 */
+	if (foreign || vma_is_foreign(vma))
+		return true;
+
+	return pkey_access_permitted(vma_pkey(vma), write, execute);
+}
+
+void arch_dup_pkeys(struct mm_struct *oldmm, struct mm_struct *mm)
+{
+	if (!mmu_has_feature(MMU_FTR_PKEY))
+		return;
+
+	/* Duplicate the oldmm pkey state in mm: */
+	mm_pkey_allocation_map(mm) = mm_pkey_allocation_map(oldmm);
+	mm->context.execute_only_pkey = oldmm->context.execute_only_pkey;
+}
+
+#endif /* CONFIG_PPC_MEM_KEYS */

linux-7.2.3/arch/powerpc/mm/book3s64/radix_hugetlbpage.c

@@ -0,0 +1,63 @@
+// SPDX-License-Identifier: GPL-2.0
+#include <linux/mm.h>
+#include <linux/hugetlb.h>
+#include <linux/security.h>
+#include <asm/cacheflush.h>
+#include <asm/machdep.h>
+#include <asm/mman.h>
+#include <asm/tlb.h>
+
+void radix__flush_hugetlb_page(struct vm_area_struct *vma, unsigned long vmaddr)
+{
+	int psize;
+	struct hstate *hstate = hstate_file(vma->vm_file);
+
+	psize = hstate_get_psize(hstate);
+	radix__flush_tlb_page_psize(vma->vm_mm, vmaddr, psize);
+}
+
+void radix__local_flush_hugetlb_page(struct vm_area_struct *vma, unsigned long vmaddr)
+{
+	int psize;
+	struct hstate *hstate = hstate_file(vma->vm_file);
+
+	psize = hstate_get_psize(hstate);
+	radix__local_flush_tlb_page_psize(vma->vm_mm, vmaddr, psize);
+}
+
+void radix__flush_hugetlb_tlb_range(struct vm_area_struct *vma, unsigned long start,
+				   unsigned long end)
+{
+	int psize;
+	struct hstate *hstate = hstate_file(vma->vm_file);
+
+	psize = hstate_get_psize(hstate);
+	/*
+	 * Flush PWC even if we get PUD_SIZE hugetlb invalidate to keep this simpler.
+	 */
+	if (end - start >= PUD_SIZE)
+		radix__flush_tlb_pwc_range_psize(vma->vm_mm, start, end, psize);
+	else
+		radix__flush_tlb_range_psize(vma->vm_mm, start, end, psize);
+	mmu_notifier_arch_invalidate_secondary_tlbs(vma->vm_mm, start, end);
+}
+
+void radix__huge_ptep_modify_prot_commit(struct vm_area_struct *vma,
+					 unsigned long addr, pte_t *ptep,
+					 pte_t old_pte, pte_t pte)
+{
+	struct mm_struct *mm = vma->vm_mm;
+	unsigned long psize = huge_page_size(hstate_vma(vma));
+
+	/*
+	 * POWER9 NMMU must flush the TLB after clearing the PTE before
+	 * installing a PTE with more relaxed access permissions, see
+	 * radix__ptep_set_access_flags.
+	 */
+	if (!cpu_has_feature(CPU_FTR_ARCH_31) &&
+	    is_pte_rw_upgrade(pte_val(old_pte), pte_val(pte)) &&
+	    atomic_read(&mm->context.copros) > 0)
+		radix__flush_hugetlb_page(vma, addr);
+
+	set_huge_pte_at(vma->vm_mm, addr, ptep, pte, psize);
+}

linux-7.2.3/arch/powerpc/mm/book3s64/slb.c

@@ -0,0 +1,795 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * PowerPC64 SLB support.
+ *
+ * Copyright (C) 2004 David Gibson <dwg@au.ibm.com>, IBM
+ * Based on earlier code written by:
+ * Dave Engebretsen and Mike Corrigan {engebret|mikejc}@us.ibm.com
+ *    Copyright (c) 2001 Dave Engebretsen
+ * Copyright (C) 2002 Anton Blanchard <anton@au.ibm.com>, IBM
+ */
+
+#include <asm/interrupt.h>
+#include <asm/mmu.h>
+#include <asm/mmu_context.h>
+#include <asm/paca.h>
+#include <asm/lppaca.h>
+#include <asm/ppc-opcode.h>
+#include <asm/cputable.h>
+#include <asm/cacheflush.h>
+#include <asm/smp.h>
+#include <linux/compiler.h>
+#include <linux/context_tracking.h>
+#include <linux/mm_types.h>
+#include <linux/pgtable.h>
+
+#include <asm/udbg.h>
+#include <asm/text-patching.h>
+
+#include "internal.h"
+
+
+static long slb_allocate_user(struct mm_struct *mm, unsigned long ea);
+
+bool stress_slb_enabled __initdata;
+
+static int __init parse_stress_slb(char *p)
+{
+	stress_slb_enabled = true;
+	return 0;
+}
+early_param("stress_slb", parse_stress_slb);
+
+__ro_after_init DEFINE_STATIC_KEY_FALSE(stress_slb_key);
+
+bool no_slb_preload __initdata;
+static int __init parse_no_slb_preload(char *p)
+{
+	no_slb_preload = true;
+	return 0;
+}
+early_param("no_slb_preload", parse_no_slb_preload);
+__ro_after_init DEFINE_STATIC_KEY_FALSE(no_slb_preload_key);
+
+static void assert_slb_presence(bool present, unsigned long ea)
+{
+#ifdef CONFIG_DEBUG_VM
+	unsigned long tmp;
+
+	WARN_ON_ONCE(mfmsr() & MSR_EE);
+
+	if (!cpu_has_feature(CPU_FTR_ARCH_206))
+		return;
+
+	/*
+	 * slbfee. requires bit 24 (PPC bit 39) be clear in RB. Hardware
+	 * ignores all other bits from 0-27, so just clear them all.
+	 */
+	ea &= ~((1UL << SID_SHIFT) - 1);
+	asm volatile(__PPC_SLBFEE_DOT(%0, %1) : "=r"(tmp) : "r"(ea) : "cr0");
+
+	WARN_ON(present == (tmp == 0));
+#endif
+}
+
+static inline void slb_shadow_update(unsigned long ea, int ssize,
+				     unsigned long flags,
+				     enum slb_index index)
+{
+	struct slb_shadow *p = get_slb_shadow();
+
+	/*
+	 * Clear the ESID first so the entry is not valid while we are
+	 * updating it.  No write barriers are needed here, provided
+	 * we only update the current CPU's SLB shadow buffer.
+	 */
+	WRITE_ONCE(p->save_area[index].esid, 0);
+	WRITE_ONCE(p->save_area[index].vsid, cpu_to_be64(mk_vsid_data(ea, ssize, flags)));
+	WRITE_ONCE(p->save_area[index].esid, cpu_to_be64(mk_esid_data(ea, ssize, index)));
+}
+
+static inline void slb_shadow_clear(enum slb_index index)
+{
+	WRITE_ONCE(get_slb_shadow()->save_area[index].esid, cpu_to_be64(index));
+}
+
+static inline void create_shadowed_slbe(unsigned long ea, int ssize,
+					unsigned long flags,
+					enum slb_index index)
+{
+	/*
+	 * Updating the shadow buffer before writing the SLB ensures
+	 * we don't get a stale entry here if we get preempted by PHYP
+	 * between these two statements.
+	 */
+	slb_shadow_update(ea, ssize, flags, index);
+
+	assert_slb_presence(false, ea);
+	asm volatile("slbmte  %0,%1" :
+		     : "r" (mk_vsid_data(ea, ssize, flags)),
+		       "r" (mk_esid_data(ea, ssize, index))
+		     : "memory" );
+}
+
+/*
+ * Insert bolted entries into SLB (which may not be empty, so don't clear
+ * slb_cache_ptr).
+ */
+void __slb_restore_bolted_realmode(void)
+{
+	struct slb_shadow *p = get_slb_shadow();
+	enum slb_index index;
+
+	 /* No isync needed because realmode. */
+	for (index = 0; index < SLB_NUM_BOLTED; index++) {
+		asm volatile("slbmte  %0,%1" :
+		     : "r" (be64_to_cpu(p->save_area[index].vsid)),
+		       "r" (be64_to_cpu(p->save_area[index].esid)));
+	}
+
+	assert_slb_presence(true, local_paca->kstack);
+}
+
+/*
+ * Insert the bolted entries into an empty SLB.
+ */
+void slb_restore_bolted_realmode(void)
+{
+	__slb_restore_bolted_realmode();
+	get_paca()->slb_cache_ptr = 0;
+
+	get_paca()->slb_kern_bitmap = (1U << SLB_NUM_BOLTED) - 1;
+	get_paca()->slb_used_bitmap = get_paca()->slb_kern_bitmap;
+}
+
+/*
+ * This flushes all SLB entries including 0, so it must be realmode.
+ */
+void slb_flush_all_realmode(void)
+{
+	asm volatile("slbmte %0,%0; slbia" : : "r" (0));
+}
+
+static __always_inline void __slb_flush_and_restore_bolted(bool preserve_kernel_lookaside)
+{
+	struct slb_shadow *p = get_slb_shadow();
+	unsigned long ksp_esid_data, ksp_vsid_data;
+	u32 ih;
+
+	/*
+	 * SLBIA IH=1 on ISA v2.05 and newer processors may preserve lookaside
+	 * information created with Class=0 entries, which we use for kernel
+	 * SLB entries (the SLB entries themselves are still invalidated).
+	 *
+	 * Older processors will ignore this optimisation. Over-invalidation
+	 * is fine because we never rely on lookaside information existing.
+	 */
+	if (preserve_kernel_lookaside)
+		ih = 1;
+	else
+		ih = 0;
+
+	ksp_esid_data = be64_to_cpu(p->save_area[KSTACK_INDEX].esid);
+	ksp_vsid_data = be64_to_cpu(p->save_area[KSTACK_INDEX].vsid);
+
+	asm volatile(PPC_SLBIA(%0)"	\n"
+		     "slbmte	%1, %2	\n"
+		     :: "i" (ih),
+			"r" (ksp_vsid_data),
+			"r" (ksp_esid_data)
+		     : "memory");
+}
+
+/*
+ * This flushes non-bolted entries, it can be run in virtual mode. Must
+ * be called with interrupts disabled.
+ */
+void slb_flush_and_restore_bolted(void)
+{
+	BUILD_BUG_ON(SLB_NUM_BOLTED != 2);
+
+	WARN_ON(!irqs_disabled());
+
+	/*
+	 * We can't take a PMU exception in the following code, so hard
+	 * disable interrupts.
+	 */
+	hard_irq_disable();
+
+	isync();
+	__slb_flush_and_restore_bolted(false);
+	isync();
+
+	assert_slb_presence(true, get_paca()->kstack);
+
+	get_paca()->slb_cache_ptr = 0;
+
+	get_paca()->slb_kern_bitmap = (1U << SLB_NUM_BOLTED) - 1;
+	get_paca()->slb_used_bitmap = get_paca()->slb_kern_bitmap;
+}
+
+void slb_save_contents(struct slb_entry *slb_ptr)
+{
+	int i;
+	unsigned long e, v;
+
+	/* Save slb_cache_ptr value. */
+	get_paca()->slb_save_cache_ptr = get_paca()->slb_cache_ptr;
+
+	if (!slb_ptr)
+		return;
+
+	for (i = 0; i < mmu_slb_size; i++) {
+		asm volatile("slbmfee  %0,%1" : "=r" (e) : "r" (i));
+		asm volatile("slbmfev  %0,%1" : "=r" (v) : "r" (i));
+		slb_ptr->esid = e;
+		slb_ptr->vsid = v;
+		slb_ptr++;
+	}
+}
+
+void slb_dump_contents(struct slb_entry *slb_ptr)
+{
+	int i, n;
+	unsigned long e, v;
+	unsigned long llp;
+
+	if (!slb_ptr)
+		return;
+
+	pr_err("SLB contents of cpu 0x%x\n", smp_processor_id());
+
+	for (i = 0; i < mmu_slb_size; i++) {
+		e = slb_ptr->esid;
+		v = slb_ptr->vsid;
+		slb_ptr++;
+
+		if (!e && !v)
+			continue;
+
+		pr_err("%02d %016lx %016lx %s\n", i, e, v,
+				(e & SLB_ESID_V) ? "VALID" : "NOT VALID");
+
+		if (!(e & SLB_ESID_V))
+			continue;
+
+		llp = v & SLB_VSID_LLP;
+		if (v & SLB_VSID_B_1T) {
+			pr_err("     1T ESID=%9lx VSID=%13lx LLP:%3lx\n",
+			       GET_ESID_1T(e),
+			       (v & ~SLB_VSID_B) >> SLB_VSID_SHIFT_1T, llp);
+		} else {
+			pr_err("   256M ESID=%9lx VSID=%13lx LLP:%3lx\n",
+			       GET_ESID(e),
+			       (v & ~SLB_VSID_B) >> SLB_VSID_SHIFT, llp);
+		}
+	}
+
+	if (!early_cpu_has_feature(CPU_FTR_ARCH_300)) {
+		/* RR is not so useful as it's often not used for allocation */
+		pr_err("SLB RR allocator index %d\n", get_paca()->stab_rr);
+
+		/* Dump slb cache entires as well. */
+		pr_err("SLB cache ptr value = %d\n", get_paca()->slb_save_cache_ptr);
+		pr_err("Valid SLB cache entries:\n");
+		n = min_t(int, get_paca()->slb_save_cache_ptr, SLB_CACHE_ENTRIES);
+		for (i = 0; i < n; i++)
+			pr_err("%02d EA[0-35]=%9x\n", i, get_paca()->slb_cache[i]);
+		pr_err("Rest of SLB cache entries:\n");
+		for (i = n; i < SLB_CACHE_ENTRIES; i++)
+			pr_err("%02d EA[0-35]=%9x\n", i, get_paca()->slb_cache[i]);
+	}
+}
+
+void slb_vmalloc_update(void)
+{
+	/*
+	 * vmalloc is not bolted, so just have to flush non-bolted.
+	 */
+	slb_flush_and_restore_bolted();
+}
+
+static bool preload_hit(struct thread_info *ti, unsigned long esid)
+{
+	unsigned char i;
+
+	for (i = 0; i < ti->slb_preload_nr; i++) {
+		unsigned char idx;
+
+		idx = (ti->slb_preload_tail + i) % SLB_PRELOAD_NR;
+		if (esid == ti->slb_preload_esid[idx])
+			return true;
+	}
+	return false;
+}
+
+static void preload_add(struct thread_info *ti, unsigned long ea)
+{
+	unsigned char idx;
+	unsigned long esid;
+
+	if (slb_preload_disabled())
+		return;
+
+	if (mmu_has_feature(MMU_FTR_1T_SEGMENT)) {
+		/* EAs are stored >> 28 so 256MB segments don't need clearing */
+		if (ea & ESID_MASK_1T)
+			ea &= ESID_MASK_1T;
+	}
+
+	esid = ea >> SID_SHIFT;
+
+	if (preload_hit(ti, esid))
+		return;
+
+	idx = (ti->slb_preload_tail + ti->slb_preload_nr) % SLB_PRELOAD_NR;
+	ti->slb_preload_esid[idx] = esid;
+	if (ti->slb_preload_nr == SLB_PRELOAD_NR)
+		ti->slb_preload_tail = (ti->slb_preload_tail + 1) % SLB_PRELOAD_NR;
+	else
+		ti->slb_preload_nr++;
+}
+
+static void preload_age(struct thread_info *ti)
+{
+	if (!ti->slb_preload_nr)
+		return;
+	ti->slb_preload_nr--;
+	ti->slb_preload_tail = (ti->slb_preload_tail + 1) % SLB_PRELOAD_NR;
+}
+
+static void slb_cache_slbie_kernel(unsigned int index)
+{
+	unsigned long slbie_data = get_paca()->slb_cache[index];
+	unsigned long ksp = get_paca()->kstack;
+
+	slbie_data <<= SID_SHIFT;
+	slbie_data |= 0xc000000000000000ULL;
+	if ((ksp & slb_esid_mask(mmu_kernel_ssize)) == slbie_data)
+		return;
+	slbie_data |= mmu_kernel_ssize << SLBIE_SSIZE_SHIFT;
+
+	asm volatile("slbie %0" : : "r" (slbie_data));
+}
+
+static void slb_cache_slbie_user(unsigned int index)
+{
+	unsigned long slbie_data = get_paca()->slb_cache[index];
+
+	slbie_data <<= SID_SHIFT;
+	slbie_data |= user_segment_size(slbie_data) << SLBIE_SSIZE_SHIFT;
+	slbie_data |= SLBIE_C; /* user slbs have C=1 */
+
+	asm volatile("slbie %0" : : "r" (slbie_data));
+}
+
+/* Flush all user entries from the segment table of the current processor. */
+void switch_slb(struct task_struct *tsk, struct mm_struct *mm)
+{
+	struct thread_info *ti = task_thread_info(tsk);
+	unsigned char i;
+
+	/*
+	 * We need interrupts hard-disabled here, not just soft-disabled,
+	 * so that a PMU interrupt can't occur, which might try to access
+	 * user memory (to get a stack trace) and possible cause an SLB miss
+	 * which would update the slb_cache/slb_cache_ptr fields in the PACA.
+	 */
+	hard_irq_disable();
+	isync();
+	if (stress_slb()) {
+		__slb_flush_and_restore_bolted(false);
+		isync();
+		get_paca()->slb_cache_ptr = 0;
+		get_paca()->slb_kern_bitmap = (1U << SLB_NUM_BOLTED) - 1;
+
+	} else if (cpu_has_feature(CPU_FTR_ARCH_300)) {
+		/*
+		 * SLBIA IH=3 invalidates all Class=1 SLBEs and their
+		 * associated lookaside structures, which matches what
+		 * switch_slb wants. So ARCH_300 does not use the slb
+		 * cache.
+		 */
+		asm volatile(PPC_SLBIA(3));
+
+	} else {
+		unsigned long offset = get_paca()->slb_cache_ptr;
+
+		if (!mmu_has_feature(MMU_FTR_NO_SLBIE_B) &&
+		    offset <= SLB_CACHE_ENTRIES) {
+			/*
+			 * Could assert_slb_presence(true) here, but
+			 * hypervisor or machine check could have come
+			 * in and removed the entry at this point.
+			 */
+
+			for (i = 0; i < offset; i++)
+				slb_cache_slbie_user(i);
+
+			/* Workaround POWER5 < DD2.1 issue */
+			if (!cpu_has_feature(CPU_FTR_ARCH_207S) && offset == 1)
+				slb_cache_slbie_user(0);
+
+		} else {
+			/* Flush but retain kernel lookaside information */
+			__slb_flush_and_restore_bolted(true);
+			isync();
+
+			get_paca()->slb_kern_bitmap = (1U << SLB_NUM_BOLTED) - 1;
+		}
+
+		get_paca()->slb_cache_ptr = 0;
+	}
+	get_paca()->slb_used_bitmap = get_paca()->slb_kern_bitmap;
+
+	copy_mm_to_paca(mm);
+
+	if (slb_preload_disabled())
+		return;
+
+	/*
+	 * We gradually age out SLBs after a number of context switches to
+	 * reduce reload overhead of unused entries (like we do with FP/VEC
+	 * reload). Each time we wrap 256 switches, take an entry out of the
+	 * SLB preload cache.
+	 */
+	tsk->thread.load_slb++;
+	if (!tsk->thread.load_slb) {
+		unsigned long pc = KSTK_EIP(tsk);
+
+		preload_age(ti);
+		preload_add(ti, pc);
+	}
+
+	for (i = 0; i < ti->slb_preload_nr; i++) {
+		unsigned char idx;
+		unsigned long ea;
+
+		idx = (ti->slb_preload_tail + i) % SLB_PRELOAD_NR;
+		ea = (unsigned long)ti->slb_preload_esid[idx] << SID_SHIFT;
+
+		slb_allocate_user(mm, ea);
+	}
+
+	/*
+	 * Synchronize slbmte preloads with possible subsequent user memory
+	 * address accesses by the kernel (user mode won't happen until
+	 * rfid, which is safe).
+	 */
+	isync();
+}
+
+void slb_set_size(u16 size)
+{
+	mmu_slb_size = size;
+}
+
+void slb_initialize(void)
+{
+	unsigned long linear_llp, vmalloc_llp, io_llp;
+	unsigned long lflags;
+	static int slb_encoding_inited;
+#ifdef CONFIG_SPARSEMEM_VMEMMAP
+	unsigned long vmemmap_llp;
+#endif
+
+	/* Prepare our SLB miss handler based on our page size */
+	linear_llp = mmu_psize_defs[mmu_linear_psize].sllp;
+	io_llp = mmu_psize_defs[mmu_io_psize].sllp;
+	vmalloc_llp = mmu_psize_defs[mmu_vmalloc_psize].sllp;
+	get_paca()->vmalloc_sllp = SLB_VSID_KERNEL | vmalloc_llp;
+#ifdef CONFIG_SPARSEMEM_VMEMMAP
+	vmemmap_llp = mmu_psize_defs[mmu_vmemmap_psize].sllp;
+#endif
+	if (!slb_encoding_inited) {
+		slb_encoding_inited = 1;
+		pr_devel("SLB: linear  LLP = %04lx\n", linear_llp);
+		pr_devel("SLB: io      LLP = %04lx\n", io_llp);
+#ifdef CONFIG_SPARSEMEM_VMEMMAP
+		pr_devel("SLB: vmemmap LLP = %04lx\n", vmemmap_llp);
+#endif
+	}
+
+	get_paca()->stab_rr = SLB_NUM_BOLTED - 1;
+	get_paca()->slb_kern_bitmap = (1U << SLB_NUM_BOLTED) - 1;
+	get_paca()->slb_used_bitmap = get_paca()->slb_kern_bitmap;
+
+	lflags = SLB_VSID_KERNEL | linear_llp;
+
+	/* Invalidate the entire SLB (even entry 0) & all the ERATS */
+	asm volatile("isync":::"memory");
+	asm volatile("slbmte  %0,%0"::"r" (0) : "memory");
+	asm volatile("isync; slbia; isync":::"memory");
+	create_shadowed_slbe(PAGE_OFFSET, mmu_kernel_ssize, lflags, LINEAR_INDEX);
+
+	/*
+	 * For the boot cpu, we're running on the stack in init_thread_union,
+	 * which is in the first segment of the linear mapping, and also
+	 * get_paca()->kstack hasn't been initialized yet.
+	 * For secondary cpus, we need to bolt the kernel stack entry now.
+	 */
+	slb_shadow_clear(KSTACK_INDEX);
+	if (raw_smp_processor_id() != boot_cpuid &&
+	    (get_paca()->kstack & slb_esid_mask(mmu_kernel_ssize)) > PAGE_OFFSET)
+		create_shadowed_slbe(get_paca()->kstack,
+				     mmu_kernel_ssize, lflags, KSTACK_INDEX);
+
+	asm volatile("isync":::"memory");
+}
+
+static void slb_cache_update(unsigned long esid_data)
+{
+	int slb_cache_index;
+
+	if (cpu_has_feature(CPU_FTR_ARCH_300))
+		return; /* ISAv3.0B and later does not use slb_cache */
+
+	if (stress_slb())
+		return;
+
+	/*
+	 * Now update slb cache entries
+	 */
+	slb_cache_index = local_paca->slb_cache_ptr;
+	if (slb_cache_index < SLB_CACHE_ENTRIES) {
+		/*
+		 * We have space in slb cache for optimized switch_slb().
+		 * Top 36 bits from esid_data as per ISA
+		 */
+		local_paca->slb_cache[slb_cache_index++] = esid_data >> SID_SHIFT;
+		local_paca->slb_cache_ptr++;
+	} else {
+		/*
+		 * Our cache is full and the current cache content strictly
+		 * doesn't indicate the active SLB contents. Bump the ptr
+		 * so that switch_slb() will ignore the cache.
+		 */
+		local_paca->slb_cache_ptr = SLB_CACHE_ENTRIES + 1;
+	}
+}
+
+static enum slb_index alloc_slb_index(bool kernel)
+{
+	enum slb_index index;
+
+	/*
+	 * The allocation bitmaps can become out of synch with the SLB
+	 * when the _switch code does slbie when bolting a new stack
+	 * segment and it must not be anywhere else in the SLB. This leaves
+	 * a kernel allocated entry that is unused in the SLB. With very
+	 * large systems or small segment sizes, the bitmaps could slowly
+	 * fill with these entries. They will eventually be cleared out
+	 * by the round robin allocator in that case, so it's probably not
+	 * worth accounting for.
+	 */
+
+	/*
+	 * SLBs beyond 32 entries are allocated with stab_rr only
+	 * POWER7/8/9 have 32 SLB entries, this could be expanded if a
+	 * future CPU has more.
+	 */
+	if (local_paca->slb_used_bitmap != U32_MAX) {
+		index = ffz(local_paca->slb_used_bitmap);
+		local_paca->slb_used_bitmap |= 1U << index;
+		if (kernel)
+			local_paca->slb_kern_bitmap |= 1U << index;
+	} else {
+		/* round-robin replacement of slb starting at SLB_NUM_BOLTED. */
+		index = local_paca->stab_rr;
+		if (index < (mmu_slb_size - 1))
+			index++;
+		else
+			index = SLB_NUM_BOLTED;
+		local_paca->stab_rr = index;
+		if (index < 32) {
+			if (kernel)
+				local_paca->slb_kern_bitmap |= 1U << index;
+			else
+				local_paca->slb_kern_bitmap &= ~(1U << index);
+		}
+	}
+	BUG_ON(index < SLB_NUM_BOLTED);
+
+	return index;
+}
+
+static long slb_insert_entry(unsigned long ea, unsigned long context,
+				unsigned long flags, int ssize, bool kernel)
+{
+	unsigned long vsid;
+	unsigned long vsid_data, esid_data;
+	enum slb_index index;
+
+	vsid = get_vsid(context, ea, ssize);
+	if (!vsid)
+		return -EFAULT;
+
+	/*
+	 * There must not be a kernel SLB fault in alloc_slb_index or before
+	 * slbmte here or the allocation bitmaps could get out of whack with
+	 * the SLB.
+	 *
+	 * User SLB faults or preloads take this path which might get inlined
+	 * into the caller, so add compiler barriers here to ensure unsafe
+	 * memory accesses do not come between.
+	 */
+	barrier();
+
+	index = alloc_slb_index(kernel);
+
+	vsid_data = __mk_vsid_data(vsid, ssize, flags);
+	esid_data = mk_esid_data(ea, ssize, index);
+
+	/*
+	 * No need for an isync before or after this slbmte. The exception
+	 * we enter with and the rfid we exit with are context synchronizing.
+	 * User preloads should add isync afterwards in case the kernel
+	 * accesses user memory before it returns to userspace with rfid.
+	 */
+	assert_slb_presence(false, ea);
+	if (stress_slb()) {
+		int slb_cache_index = local_paca->slb_cache_ptr;
+
+		/*
+		 * stress_slb() does not use slb cache, repurpose as a
+		 * cache of inserted (non-bolted) kernel SLB entries. All
+		 * non-bolted kernel entries are flushed on any user fault,
+		 * or if there are already 3 non-boled kernel entries.
+		 */
+		BUILD_BUG_ON(SLB_CACHE_ENTRIES < 3);
+		if (!kernel || slb_cache_index == 3) {
+			int i;
+
+			for (i = 0; i < slb_cache_index; i++)
+				slb_cache_slbie_kernel(i);
+			slb_cache_index = 0;
+		}
+
+		if (kernel)
+			local_paca->slb_cache[slb_cache_index++] = esid_data >> SID_SHIFT;
+		local_paca->slb_cache_ptr = slb_cache_index;
+	}
+	asm volatile("slbmte %0, %1" : : "r" (vsid_data), "r" (esid_data));
+
+	barrier();
+
+	if (!kernel)
+		slb_cache_update(esid_data);
+
+	return 0;
+}
+
+static long slb_allocate_kernel(unsigned long ea, unsigned long id)
+{
+	unsigned long context;
+	unsigned long flags;
+	int ssize;
+
+	if (id == LINEAR_MAP_REGION_ID) {
+
+		/* We only support upto H_MAX_PHYSMEM_BITS */
+		if ((ea & EA_MASK) > (1UL << H_MAX_PHYSMEM_BITS))
+			return -EFAULT;
+
+		flags = SLB_VSID_KERNEL | mmu_psize_defs[mmu_linear_psize].sllp;
+
+#ifdef CONFIG_SPARSEMEM_VMEMMAP
+	} else if (id == VMEMMAP_REGION_ID) {
+
+		if (ea >= H_VMEMMAP_END)
+			return -EFAULT;
+
+		flags = SLB_VSID_KERNEL | mmu_psize_defs[mmu_vmemmap_psize].sllp;
+#endif
+	} else if (id == VMALLOC_REGION_ID) {
+
+		if (ea >= H_VMALLOC_END)
+			return -EFAULT;
+
+		flags = local_paca->vmalloc_sllp;
+
+	} else if (id == IO_REGION_ID) {
+
+		if (ea >= H_KERN_IO_END)
+			return -EFAULT;
+
+		flags = SLB_VSID_KERNEL | mmu_psize_defs[mmu_io_psize].sllp;
+
+	} else {
+		return -EFAULT;
+	}
+
+	ssize = MMU_SEGSIZE_1T;
+	if (!mmu_has_feature(MMU_FTR_1T_SEGMENT))
+		ssize = MMU_SEGSIZE_256M;
+
+	context = get_kernel_context(ea);
+
+	return slb_insert_entry(ea, context, flags, ssize, true);
+}
+
+static long slb_allocate_user(struct mm_struct *mm, unsigned long ea)
+{
+	unsigned long context;
+	unsigned long flags;
+	int bpsize;
+	int ssize;
+
+	/*
+	 * consider this as bad access if we take a SLB miss
+	 * on an address above addr limit.
+	 */
+	if (ea >= mm_ctx_slb_addr_limit(&mm->context))
+		return -EFAULT;
+
+	context = get_user_context(&mm->context, ea);
+	if (!context)
+		return -EFAULT;
+
+	if (unlikely(ea >= H_PGTABLE_RANGE)) {
+		WARN_ON(1);
+		return -EFAULT;
+	}
+
+	ssize = user_segment_size(ea);
+
+	bpsize = get_slice_psize(mm, ea);
+	flags = SLB_VSID_USER | mmu_psize_defs[bpsize].sllp;
+
+	return slb_insert_entry(ea, context, flags, ssize, false);
+}
+
+DEFINE_INTERRUPT_HANDLER_RAW(do_slb_fault)
+{
+	unsigned long ea = regs->dar;
+	unsigned long id = get_region_id(ea);
+
+	/* IRQs are not reconciled here, so can't check irqs_disabled */
+	VM_WARN_ON(mfmsr() & MSR_EE);
+
+	if (regs_is_unrecoverable(regs))
+		return -EINVAL;
+
+	/*
+	 * SLB kernel faults must be very careful not to touch anything that is
+	 * not bolted. E.g., PACA and global variables are okay, mm->context
+	 * stuff is not. SLB user faults may access all of memory (and induce
+	 * one recursive SLB kernel fault), so the kernel fault must not
+	 * trample on the user fault state at those points.
+	 */
+
+	/*
+	 * This is a raw interrupt handler, for performance, so that
+	 * fast_interrupt_return can be used. The handler must not touch local
+	 * irq state, or schedule. We could test for usermode and upgrade to a
+	 * normal process context (synchronous) interrupt for those, which
+	 * would make them first-class kernel code and able to be traced and
+	 * instrumented, although performance would suffer a bit, it would
+	 * probably be a good tradeoff.
+	 */
+	if (id >= LINEAR_MAP_REGION_ID) {
+		long err;
+#ifdef CONFIG_DEBUG_VM
+		/* Catch recursive kernel SLB faults. */
+		BUG_ON(local_paca->in_kernel_slb_handler);
+		local_paca->in_kernel_slb_handler = 1;
+#endif
+		err = slb_allocate_kernel(ea, id);
+#ifdef CONFIG_DEBUG_VM
+		local_paca->in_kernel_slb_handler = 0;
+#endif
+		return err;
+	} else {
+		struct mm_struct *mm = current->mm;
+		long err;
+
+		if (unlikely(!mm))
+			return -EFAULT;
+
+		err = slb_allocate_user(mm, ea);
+		if (!err)
+			preload_add(current_thread_info(), ea);
+
+		return err;
+	}
+}

linux-7.2.3/arch/powerpc/mm/book3s64/slice.c

@@ -0,0 +1,819 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * address space "slices" (meta-segments) support
+ *
+ * Copyright (C) 2007 Benjamin Herrenschmidt, IBM Corporation.
+ *
+ * Based on hugetlb implementation
+ *
+ * Copyright (C) 2003 David Gibson, IBM Corporation.
+ */
+
+#undef DEBUG
+
+#include <linux/kernel.h>
+#include <linux/mm.h>
+#include <linux/pagemap.h>
+#include <linux/err.h>
+#include <linux/spinlock.h>
+#include <linux/export.h>
+#include <linux/hugetlb.h>
+#include <linux/sched/mm.h>
+#include <linux/security.h>
+#include <asm/mman.h>
+#include <asm/mmu.h>
+#include <asm/spu.h>
+#include <asm/hugetlb.h>
+#include <asm/mmu_context.h>
+
+static DEFINE_SPINLOCK(slice_convert_lock);
+
+#ifdef DEBUG
+int _slice_debug = 1;
+
+static void slice_print_mask(const char *label, const struct slice_mask *mask)
+{
+	if (!_slice_debug)
+		return;
+	pr_devel("%s low_slice: %*pbl\n", label,
+			(int)SLICE_NUM_LOW, &mask->low_slices);
+	pr_devel("%s high_slice: %*pbl\n", label,
+			(int)SLICE_NUM_HIGH, mask->high_slices);
+}
+
+#define slice_dbg(fmt...) do { if (_slice_debug) pr_devel(fmt); } while (0)
+
+#else
+
+static void slice_print_mask(const char *label, const struct slice_mask *mask) {}
+#define slice_dbg(fmt...)
+
+#endif
+
+static inline notrace bool slice_addr_is_low(unsigned long addr)
+{
+	u64 tmp = (u64)addr;
+
+	return tmp < SLICE_LOW_TOP;
+}
+
+static void slice_range_to_mask(unsigned long start, unsigned long len,
+				struct slice_mask *ret)
+{
+	unsigned long end = start + len - 1;
+
+	ret->low_slices = 0;
+	if (SLICE_NUM_HIGH)
+		bitmap_zero(ret->high_slices, SLICE_NUM_HIGH);
+
+	if (slice_addr_is_low(start)) {
+		unsigned long mend = min(end,
+					 (unsigned long)(SLICE_LOW_TOP - 1));
+
+		ret->low_slices = (1u << (GET_LOW_SLICE_INDEX(mend) + 1))
+			- (1u << GET_LOW_SLICE_INDEX(start));
+	}
+
+	if (SLICE_NUM_HIGH && !slice_addr_is_low(end)) {
+		unsigned long start_index = GET_HIGH_SLICE_INDEX(start);
+		unsigned long align_end = ALIGN(end, (1UL << SLICE_HIGH_SHIFT));
+		unsigned long count = GET_HIGH_SLICE_INDEX(align_end) - start_index;
+
+		bitmap_set(ret->high_slices, start_index, count);
+	}
+}
+
+static int slice_area_is_free(struct mm_struct *mm, unsigned long addr,
+			      unsigned long len)
+{
+	struct vm_area_struct *vma;
+
+	if ((mm_ctx_slb_addr_limit(&mm->context) - len) < addr)
+		return 0;
+	vma = find_vma(mm, addr);
+	return (!vma || (addr + len) <= vm_start_gap(vma));
+}
+
+static int slice_low_has_vma(struct mm_struct *mm, unsigned long slice)
+{
+	return !slice_area_is_free(mm, slice << SLICE_LOW_SHIFT,
+				   1ul << SLICE_LOW_SHIFT);
+}
+
+static int slice_high_has_vma(struct mm_struct *mm, unsigned long slice)
+{
+	unsigned long start = slice << SLICE_HIGH_SHIFT;
+	unsigned long end = start + (1ul << SLICE_HIGH_SHIFT);
+
+	/* Hack, so that each addresses is controlled by exactly one
+	 * of the high or low area bitmaps, the first high area starts
+	 * at 4GB, not 0 */
+	if (start == 0)
+		start = (unsigned long)SLICE_LOW_TOP;
+
+	return !slice_area_is_free(mm, start, end - start);
+}
+
+static void slice_mask_for_free(struct mm_struct *mm, struct slice_mask *ret,
+				unsigned long high_limit)
+{
+	unsigned long i;
+
+	ret->low_slices = 0;
+	if (SLICE_NUM_HIGH)
+		bitmap_zero(ret->high_slices, SLICE_NUM_HIGH);
+
+	for (i = 0; i < SLICE_NUM_LOW; i++)
+		if (!slice_low_has_vma(mm, i))
+			ret->low_slices |= 1u << i;
+
+	if (slice_addr_is_low(high_limit - 1))
+		return;
+
+	for (i = 0; i < GET_HIGH_SLICE_INDEX(high_limit); i++)
+		if (!slice_high_has_vma(mm, i))
+			__set_bit(i, ret->high_slices);
+}
+
+static bool slice_check_range_fits(struct mm_struct *mm,
+			   const struct slice_mask *available,
+			   unsigned long start, unsigned long len)
+{
+	unsigned long end = start + len - 1;
+	u64 low_slices = 0;
+
+	if (slice_addr_is_low(start)) {
+		unsigned long mend = min(end,
+					 (unsigned long)(SLICE_LOW_TOP - 1));
+
+		low_slices = (1u << (GET_LOW_SLICE_INDEX(mend) + 1))
+				- (1u << GET_LOW_SLICE_INDEX(start));
+	}
+	if ((low_slices & available->low_slices) != low_slices)
+		return false;
+
+	if (SLICE_NUM_HIGH && !slice_addr_is_low(end)) {
+		unsigned long start_index = GET_HIGH_SLICE_INDEX(start);
+		unsigned long align_end = ALIGN(end, (1UL << SLICE_HIGH_SHIFT));
+		unsigned long count = GET_HIGH_SLICE_INDEX(align_end) - start_index;
+		unsigned long i;
+
+		for (i = start_index; i < start_index + count; i++) {
+			if (!test_bit(i, available->high_slices))
+				return false;
+		}
+	}
+
+	return true;
+}
+
+static void slice_flush_segments(void *parm)
+{
+#ifdef CONFIG_PPC64
+	struct mm_struct *mm = parm;
+	unsigned long flags;
+
+	if (mm != current->active_mm)
+		return;
+
+	copy_mm_to_paca(current->active_mm);
+
+	local_irq_save(flags);
+	slb_flush_and_restore_bolted();
+	local_irq_restore(flags);
+#endif
+}
+
+static void slice_convert(struct mm_struct *mm,
+				const struct slice_mask *mask, int psize)
+{
+	int index, mask_index;
+	/* Write the new slice psize bits */
+	unsigned char *hpsizes, *lpsizes;
+	struct slice_mask *psize_mask, *old_mask;
+	unsigned long i, flags;
+	int old_psize;
+
+	slice_dbg("slice_convert(mm=%p, psize=%d)\n", mm, psize);
+	slice_print_mask(" mask", mask);
+
+	psize_mask = slice_mask_for_size(&mm->context, psize);
+
+	/* We need to use a spinlock here to protect against
+	 * concurrent 64k -> 4k demotion ...
+	 */
+	spin_lock_irqsave(&slice_convert_lock, flags);
+
+	lpsizes = mm_ctx_low_slices(&mm->context);
+	for (i = 0; i < SLICE_NUM_LOW; i++) {
+		if (!(mask->low_slices & (1u << i)))
+			continue;
+
+		mask_index = i & 0x1;
+		index = i >> 1;
+
+		/* Update the slice_mask */
+		old_psize = (lpsizes[index] >> (mask_index * 4)) & 0xf;
+		old_mask = slice_mask_for_size(&mm->context, old_psize);
+		old_mask->low_slices &= ~(1u << i);
+		psize_mask->low_slices |= 1u << i;
+
+		/* Update the sizes array */
+		lpsizes[index] = (lpsizes[index] & ~(0xf << (mask_index * 4))) |
+				(((unsigned long)psize) << (mask_index * 4));
+	}
+
+	hpsizes = mm_ctx_high_slices(&mm->context);
+	for (i = 0; i < GET_HIGH_SLICE_INDEX(mm_ctx_slb_addr_limit(&mm->context)); i++) {
+		if (!test_bit(i, mask->high_slices))
+			continue;
+
+		mask_index = i & 0x1;
+		index = i >> 1;
+
+		/* Update the slice_mask */
+		old_psize = (hpsizes[index] >> (mask_index * 4)) & 0xf;
+		old_mask = slice_mask_for_size(&mm->context, old_psize);
+		__clear_bit(i, old_mask->high_slices);
+		__set_bit(i, psize_mask->high_slices);
+
+		/* Update the sizes array */
+		hpsizes[index] = (hpsizes[index] & ~(0xf << (mask_index * 4))) |
+				(((unsigned long)psize) << (mask_index * 4));
+	}
+
+	slice_dbg(" lsps=%lx, hsps=%lx\n",
+		  (unsigned long)mm_ctx_low_slices(&mm->context),
+		  (unsigned long)mm_ctx_high_slices(&mm->context));
+
+	spin_unlock_irqrestore(&slice_convert_lock, flags);
+
+#ifdef CONFIG_SPU_BASE
+	spu_flush_all_slbs(mm);
+#endif
+}
+
+/*
+ * Compute which slice addr is part of;
+ * set *boundary_addr to the start or end boundary of that slice
+ * (depending on 'end' parameter);
+ * return boolean indicating if the slice is marked as available in the
+ * 'available' slice_mark.
+ */
+static bool slice_scan_available(unsigned long addr,
+				 const struct slice_mask *available,
+				 int end, unsigned long *boundary_addr)
+{
+	unsigned long slice;
+	if (slice_addr_is_low(addr)) {
+		slice = GET_LOW_SLICE_INDEX(addr);
+		*boundary_addr = (slice + end) << SLICE_LOW_SHIFT;
+		return !!(available->low_slices & (1u << slice));
+	} else {
+		slice = GET_HIGH_SLICE_INDEX(addr);
+		*boundary_addr = (slice + end) ?
+			((slice + end) << SLICE_HIGH_SHIFT) : SLICE_LOW_TOP;
+		return !!test_bit(slice, available->high_slices);
+	}
+}
+
+static unsigned long slice_find_area_bottomup(struct mm_struct *mm,
+					      unsigned long addr, unsigned long len,
+					      const struct slice_mask *available,
+					      int psize, unsigned long high_limit)
+{
+	int pshift = max_t(int, mmu_psize_defs[psize].shift, PAGE_SHIFT);
+	unsigned long found, next_end;
+	struct vm_unmapped_area_info info = {
+		.length = len,
+		.align_mask = PAGE_MASK & ((1ul << pshift) - 1),
+	};
+	/*
+	 * Check till the allow max value for this mmap request
+	 */
+	while (addr < high_limit) {
+		info.low_limit = addr;
+		if (!slice_scan_available(addr, available, 1, &addr))
+			continue;
+
+ next_slice:
+		/*
+		 * At this point [info.low_limit; addr) covers
+		 * available slices only and ends at a slice boundary.
+		 * Check if we need to reduce the range, or if we can
+		 * extend it to cover the next available slice.
+		 */
+		if (addr >= high_limit)
+			addr = high_limit;
+		else if (slice_scan_available(addr, available, 1, &next_end)) {
+			addr = next_end;
+			goto next_slice;
+		}
+		info.high_limit = addr;
+
+		found = vm_unmapped_area(&info);
+		if (!(found & ~PAGE_MASK))
+			return found;
+	}
+
+	return -ENOMEM;
+}
+
+static unsigned long slice_find_area_topdown(struct mm_struct *mm,
+					     unsigned long addr, unsigned long len,
+					     const struct slice_mask *available,
+					     int psize, unsigned long high_limit)
+{
+	int pshift = max_t(int, mmu_psize_defs[psize].shift, PAGE_SHIFT);
+	unsigned long found, prev;
+	struct vm_unmapped_area_info info = {
+		.flags = VM_UNMAPPED_AREA_TOPDOWN,
+		.length = len,
+		.align_mask = PAGE_MASK & ((1ul << pshift) - 1),
+	};
+	unsigned long min_addr = max(PAGE_SIZE, mmap_min_addr);
+
+	/*
+	 * If we are trying to allocate above DEFAULT_MAP_WINDOW
+	 * Add the different to the mmap_base.
+	 * Only for that request for which high_limit is above
+	 * DEFAULT_MAP_WINDOW we should apply this.
+	 */
+	if (high_limit > DEFAULT_MAP_WINDOW)
+		addr += mm_ctx_slb_addr_limit(&mm->context) - DEFAULT_MAP_WINDOW;
+
+	while (addr > min_addr) {
+		info.high_limit = addr;
+		if (!slice_scan_available(addr - 1, available, 0, &addr))
+			continue;
+
+ prev_slice:
+		/*
+		 * At this point [addr; info.high_limit) covers
+		 * available slices only and starts at a slice boundary.
+		 * Check if we need to reduce the range, or if we can
+		 * extend it to cover the previous available slice.
+		 */
+		if (addr < min_addr)
+			addr = min_addr;
+		else if (slice_scan_available(addr - 1, available, 0, &prev)) {
+			addr = prev;
+			goto prev_slice;
+		}
+		info.low_limit = addr;
+
+		found = vm_unmapped_area(&info);
+		if (!(found & ~PAGE_MASK))
+			return found;
+	}
+
+	/*
+	 * A failed mmap() very likely causes application failure,
+	 * so fall back to the bottom-up function here. This scenario
+	 * can happen with large stack limits and large mmap()
+	 * allocations.
+	 */
+	return slice_find_area_bottomup(mm, TASK_UNMAPPED_BASE, len, available, psize, high_limit);
+}
+
+
+static unsigned long slice_find_area(struct mm_struct *mm, unsigned long len,
+				     const struct slice_mask *mask, int psize,
+				     int topdown, unsigned long high_limit)
+{
+	if (topdown)
+		return slice_find_area_topdown(mm, mm->mmap_base, len, mask, psize, high_limit);
+	else
+		return slice_find_area_bottomup(mm, mm->mmap_base, len, mask, psize, high_limit);
+}
+
+static inline void slice_copy_mask(struct slice_mask *dst,
+					const struct slice_mask *src)
+{
+	dst->low_slices = src->low_slices;
+	if (!SLICE_NUM_HIGH)
+		return;
+	bitmap_copy(dst->high_slices, src->high_slices, SLICE_NUM_HIGH);
+}
+
+static inline void slice_or_mask(struct slice_mask *dst,
+					const struct slice_mask *src1,
+					const struct slice_mask *src2)
+{
+	dst->low_slices = src1->low_slices | src2->low_slices;
+	if (!SLICE_NUM_HIGH)
+		return;
+	bitmap_or(dst->high_slices, src1->high_slices, src2->high_slices, SLICE_NUM_HIGH);
+}
+
+static inline void slice_andnot_mask(struct slice_mask *dst,
+					const struct slice_mask *src1,
+					const struct slice_mask *src2)
+{
+	dst->low_slices = src1->low_slices & ~src2->low_slices;
+	if (!SLICE_NUM_HIGH)
+		return;
+	bitmap_andnot(dst->high_slices, src1->high_slices, src2->high_slices, SLICE_NUM_HIGH);
+}
+
+#ifdef CONFIG_PPC_64K_PAGES
+#define MMU_PAGE_BASE	MMU_PAGE_64K
+#else
+#define MMU_PAGE_BASE	MMU_PAGE_4K
+#endif
+
+unsigned long slice_get_unmapped_area(unsigned long addr, unsigned long len,
+				      unsigned long flags, unsigned int psize,
+				      int topdown)
+{
+	struct slice_mask good_mask;
+	struct slice_mask potential_mask;
+	const struct slice_mask *maskp;
+	const struct slice_mask *compat_maskp = NULL;
+	int fixed = (flags & MAP_FIXED);
+	int pshift = max_t(int, mmu_psize_defs[psize].shift, PAGE_SHIFT);
+	unsigned long page_size = 1UL << pshift;
+	struct mm_struct *mm = current->mm;
+	unsigned long newaddr;
+	unsigned long high_limit;
+
+	high_limit = DEFAULT_MAP_WINDOW;
+	if (addr >= high_limit || (fixed && (addr + len > high_limit)))
+		high_limit = TASK_SIZE;
+
+	if (len > high_limit)
+		return -ENOMEM;
+	if (len & (page_size - 1))
+		return -EINVAL;
+	if (fixed) {
+		if (addr & (page_size - 1))
+			return -EINVAL;
+		if (addr > high_limit - len)
+			return -ENOMEM;
+	}
+
+	if (high_limit > mm_ctx_slb_addr_limit(&mm->context)) {
+		/*
+		 * Increasing the slb_addr_limit does not require
+		 * slice mask cache to be recalculated because it should
+		 * be already initialised beyond the old address limit.
+		 */
+		mm_ctx_set_slb_addr_limit(&mm->context, high_limit);
+
+		on_each_cpu(slice_flush_segments, mm, 1);
+	}
+
+	/* Sanity checks */
+	BUG_ON(mm->task_size == 0);
+	BUG_ON(mm_ctx_slb_addr_limit(&mm->context) == 0);
+	VM_BUG_ON(radix_enabled());
+
+	slice_dbg("slice_get_unmapped_area(mm=%p, psize=%d...\n", mm, psize);
+	slice_dbg(" addr=%lx, len=%lx, flags=%lx, topdown=%d\n",
+		  addr, len, flags, topdown);
+
+	/* If hint, make sure it matches our alignment restrictions */
+	if (!fixed && addr) {
+		addr = ALIGN(addr, page_size);
+		slice_dbg(" aligned addr=%lx\n", addr);
+		/* Ignore hint if it's too large or overlaps a VMA */
+		if (addr > high_limit - len || addr < mmap_min_addr ||
+		    !slice_area_is_free(mm, addr, len))
+			addr = 0;
+	}
+
+	/* First make up a "good" mask of slices that have the right size
+	 * already
+	 */
+	maskp = slice_mask_for_size(&mm->context, psize);
+
+	/*
+	 * Here "good" means slices that are already the right page size,
+	 * "compat" means slices that have a compatible page size (i.e.
+	 * 4k in a 64k pagesize kernel), and "free" means slices without
+	 * any VMAs.
+	 *
+	 * If MAP_FIXED:
+	 *	check if fits in good | compat => OK
+	 *	check if fits in good | compat | free => convert free
+	 *	else bad
+	 * If have hint:
+	 *	check if hint fits in good => OK
+	 *	check if hint fits in good | free => convert free
+	 * Otherwise:
+	 *	search in good, found => OK
+	 *	search in good | free, found => convert free
+	 *	search in good | compat | free, found => convert free.
+	 */
+
+	/*
+	 * If we support combo pages, we can allow 64k pages in 4k slices
+	 * The mask copies could be avoided in most cases here if we had
+	 * a pointer to good mask for the next code to use.
+	 */
+	if (IS_ENABLED(CONFIG_PPC_64K_PAGES) && psize == MMU_PAGE_64K) {
+		compat_maskp = slice_mask_for_size(&mm->context, MMU_PAGE_4K);
+		if (fixed)
+			slice_or_mask(&good_mask, maskp, compat_maskp);
+		else
+			slice_copy_mask(&good_mask, maskp);
+	} else {
+		slice_copy_mask(&good_mask, maskp);
+	}
+
+	slice_print_mask(" good_mask", &good_mask);
+	if (compat_maskp)
+		slice_print_mask(" compat_mask", compat_maskp);
+
+	/* First check hint if it's valid or if we have MAP_FIXED */
+	if (addr != 0 || fixed) {
+		/* Check if we fit in the good mask. If we do, we just return,
+		 * nothing else to do
+		 */
+		if (slice_check_range_fits(mm, &good_mask, addr, len)) {
+			slice_dbg(" fits good !\n");
+			newaddr = addr;
+			goto return_addr;
+		}
+	} else {
+		/* Now let's see if we can find something in the existing
+		 * slices for that size
+		 */
+		newaddr = slice_find_area(mm, len, &good_mask,
+					  psize, topdown, high_limit);
+		if (newaddr != -ENOMEM) {
+			/* Found within the good mask, we don't have to setup,
+			 * we thus return directly
+			 */
+			slice_dbg(" found area at 0x%lx\n", newaddr);
+			goto return_addr;
+		}
+	}
+	/*
+	 * We don't fit in the good mask, check what other slices are
+	 * empty and thus can be converted
+	 */
+	slice_mask_for_free(mm, &potential_mask, high_limit);
+	slice_or_mask(&potential_mask, &potential_mask, &good_mask);
+	slice_print_mask(" potential", &potential_mask);
+
+	if (addr != 0 || fixed) {
+		if (slice_check_range_fits(mm, &potential_mask, addr, len)) {
+			slice_dbg(" fits potential !\n");
+			newaddr = addr;
+			goto convert;
+		}
+	}
+
+	/* If we have MAP_FIXED and failed the above steps, then error out */
+	if (fixed)
+		return -EBUSY;
+
+	slice_dbg(" search...\n");
+
+	/* If we had a hint that didn't work out, see if we can fit
+	 * anywhere in the good area.
+	 */
+	if (addr) {
+		newaddr = slice_find_area(mm, len, &good_mask,
+					  psize, topdown, high_limit);
+		if (newaddr != -ENOMEM) {
+			slice_dbg(" found area at 0x%lx\n", newaddr);
+			goto return_addr;
+		}
+	}
+
+	/* Now let's see if we can find something in the existing slices
+	 * for that size plus free slices
+	 */
+	newaddr = slice_find_area(mm, len, &potential_mask,
+				  psize, topdown, high_limit);
+
+	if (IS_ENABLED(CONFIG_PPC_64K_PAGES) && newaddr == -ENOMEM &&
+	    psize == MMU_PAGE_64K) {
+		/* retry the search with 4k-page slices included */
+		slice_or_mask(&potential_mask, &potential_mask, compat_maskp);
+		newaddr = slice_find_area(mm, len, &potential_mask,
+					  psize, topdown, high_limit);
+	}
+
+	if (newaddr == -ENOMEM)
+		return -ENOMEM;
+
+	slice_range_to_mask(newaddr, len, &potential_mask);
+	slice_dbg(" found potential area at 0x%lx\n", newaddr);
+	slice_print_mask(" mask", &potential_mask);
+
+ convert:
+	/*
+	 * Try to allocate the context before we do slice convert
+	 * so that we handle the context allocation failure gracefully.
+	 */
+	if (need_extra_context(mm, newaddr)) {
+		if (alloc_extended_context(mm, newaddr) < 0)
+			return -ENOMEM;
+	}
+
+	slice_andnot_mask(&potential_mask, &potential_mask, &good_mask);
+	if (compat_maskp && !fixed)
+		slice_andnot_mask(&potential_mask, &potential_mask, compat_maskp);
+	if (potential_mask.low_slices ||
+		(SLICE_NUM_HIGH &&
+		 !bitmap_empty(potential_mask.high_slices, SLICE_NUM_HIGH))) {
+		slice_convert(mm, &potential_mask, psize);
+		if (psize > MMU_PAGE_BASE)
+			on_each_cpu(slice_flush_segments, mm, 1);
+	}
+	return newaddr;
+
+return_addr:
+	if (need_extra_context(mm, newaddr)) {
+		if (alloc_extended_context(mm, newaddr) < 0)
+			return -ENOMEM;
+	}
+	return newaddr;
+}
+EXPORT_SYMBOL_GPL(slice_get_unmapped_area);
+
+#ifdef CONFIG_HUGETLB_PAGE
+static int file_to_psize(struct file *file)
+{
+	struct hstate *hstate = hstate_file(file);
+
+	return shift_to_mmu_psize(huge_page_shift(hstate));
+}
+#else
+static int file_to_psize(struct file *file)
+{
+	return 0;
+}
+#endif
+
+unsigned long arch_get_unmapped_area(struct file *filp,
+				     unsigned long addr,
+				     unsigned long len,
+				     unsigned long pgoff,
+				     unsigned long flags,
+				     vm_flags_t vm_flags)
+{
+	unsigned int psize;
+
+	if (radix_enabled())
+		return generic_get_unmapped_area(filp, addr, len, pgoff, flags, vm_flags);
+
+	if (filp && is_file_hugepages(filp))
+		psize = file_to_psize(filp);
+	else
+		psize = mm_ctx_user_psize(&current->mm->context);
+
+	return slice_get_unmapped_area(addr, len, flags, psize, 0);
+}
+
+unsigned long arch_get_unmapped_area_topdown(struct file *filp,
+					     const unsigned long addr0,
+					     const unsigned long len,
+					     const unsigned long pgoff,
+					     const unsigned long flags,
+					     vm_flags_t vm_flags)
+{
+	unsigned int psize;
+
+	if (radix_enabled())
+		return generic_get_unmapped_area_topdown(filp, addr0, len, pgoff, flags, vm_flags);
+
+	if (filp && is_file_hugepages(filp))
+		psize = file_to_psize(filp);
+	else
+		psize = mm_ctx_user_psize(&current->mm->context);
+
+	return slice_get_unmapped_area(addr0, len, flags, psize, 1);
+}
+
+unsigned int notrace get_slice_psize(struct mm_struct *mm, unsigned long addr)
+{
+	unsigned char *psizes;
+	int index, mask_index;
+
+	VM_BUG_ON(radix_enabled());
+
+	if (slice_addr_is_low(addr)) {
+		psizes = mm_ctx_low_slices(&mm->context);
+		index = GET_LOW_SLICE_INDEX(addr);
+	} else {
+		psizes = mm_ctx_high_slices(&mm->context);
+		index = GET_HIGH_SLICE_INDEX(addr);
+	}
+	mask_index = index & 0x1;
+	return (psizes[index >> 1] >> (mask_index * 4)) & 0xf;
+}
+EXPORT_SYMBOL_GPL(get_slice_psize);
+
+void slice_init_new_context_exec(struct mm_struct *mm)
+{
+	unsigned char *hpsizes, *lpsizes;
+	struct slice_mask *mask;
+	unsigned int psize = mmu_virtual_psize;
+
+	slice_dbg("slice_init_new_context_exec(mm=%p)\n", mm);
+
+	/*
+	 * In the case of exec, use the default limit. In the
+	 * case of fork it is just inherited from the mm being
+	 * duplicated.
+	 */
+	mm_ctx_set_slb_addr_limit(&mm->context, SLB_ADDR_LIMIT_DEFAULT);
+	mm_ctx_set_user_psize(&mm->context, psize);
+
+	/*
+	 * Set all slice psizes to the default.
+	 */
+	lpsizes = mm_ctx_low_slices(&mm->context);
+	memset(lpsizes, (psize << 4) | psize, SLICE_NUM_LOW >> 1);
+
+	hpsizes = mm_ctx_high_slices(&mm->context);
+	memset(hpsizes, (psize << 4) | psize, SLICE_NUM_HIGH >> 1);
+
+	/*
+	 * Slice mask cache starts zeroed, fill the default size cache.
+	 */
+	mask = slice_mask_for_size(&mm->context, psize);
+	mask->low_slices = ~0UL;
+	if (SLICE_NUM_HIGH)
+		bitmap_fill(mask->high_slices, SLICE_NUM_HIGH);
+}
+
+void slice_setup_new_exec(void)
+{
+	struct mm_struct *mm = current->mm;
+
+	slice_dbg("slice_setup_new_exec(mm=%p)\n", mm);
+
+	if (!is_32bit_task())
+		return;
+
+	mm_ctx_set_slb_addr_limit(&mm->context, DEFAULT_MAP_WINDOW);
+}
+
+void slice_set_range_psize(struct mm_struct *mm, unsigned long start,
+			   unsigned long len, unsigned int psize)
+{
+	struct slice_mask mask;
+
+	VM_BUG_ON(radix_enabled());
+
+	slice_range_to_mask(start, len, &mask);
+	slice_convert(mm, &mask, psize);
+}
+
+#ifdef CONFIG_HUGETLB_PAGE
+/*
+ * is_hugepage_only_range() is used by generic code to verify whether
+ * a normal mmap mapping (non hugetlbfs) is valid on a given area.
+ *
+ * until the generic code provides a more generic hook and/or starts
+ * calling arch get_unmapped_area for MAP_FIXED (which our implementation
+ * here knows how to deal with), we hijack it to keep standard mappings
+ * away from us.
+ *
+ * because of that generic code limitation, MAP_FIXED mapping cannot
+ * "convert" back a slice with no VMAs to the standard page size, only
+ * get_unmapped_area() can. It would be possible to fix it here but I
+ * prefer working on fixing the generic code instead.
+ *
+ * WARNING: This will not work if hugetlbfs isn't enabled since the
+ * generic code will redefine that function as 0 in that. This is ok
+ * for now as we only use slices with hugetlbfs enabled. This should
+ * be fixed as the generic code gets fixed.
+ */
+int slice_is_hugepage_only_range(struct mm_struct *mm, unsigned long addr,
+			   unsigned long len)
+{
+	const struct slice_mask *maskp;
+	unsigned int psize = mm_ctx_user_psize(&mm->context);
+
+	VM_BUG_ON(radix_enabled());
+
+	maskp = slice_mask_for_size(&mm->context, psize);
+
+	/* We need to account for 4k slices too */
+	if (IS_ENABLED(CONFIG_PPC_64K_PAGES) && psize == MMU_PAGE_64K) {
+		const struct slice_mask *compat_maskp;
+		struct slice_mask available;
+
+		compat_maskp = slice_mask_for_size(&mm->context, MMU_PAGE_4K);
+		slice_or_mask(&available, maskp, compat_maskp);
+		return !slice_check_range_fits(mm, &available, addr, len);
+	}
+
+	return !slice_check_range_fits(mm, maskp, addr, len);
+}
+
+unsigned long vma_mmu_pagesize(struct vm_area_struct *vma)
+{
+	/* With radix we don't use slice, so derive it from vma*/
+	if (radix_enabled())
+		return vma_kernel_pagesize(vma);
+
+	return 1UL << mmu_psize_to_shift(get_slice_psize(vma->vm_mm, vma->vm_start));
+}
+#endif

linux-7.2.3/arch/powerpc/mm/book3s64/subpage_prot.c

@@ -0,0 +1,281 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Copyright 2007-2008 Paul Mackerras, IBM Corp.
+ */
+
+#include <linux/errno.h>
+#include <linux/kernel.h>
+#include <linux/gfp.h>
+#include <linux/types.h>
+#include <linux/pagewalk.h>
+#include <linux/hugetlb.h>
+#include <linux/syscalls.h>
+
+#include <linux/pgtable.h>
+#include <linux/uaccess.h>
+
+/*
+ * Free all pages allocated for subpage protection maps and pointers.
+ * Also makes sure that the subpage_prot_table structure is
+ * reinitialized for the next user.
+ */
+void subpage_prot_free(struct mm_struct *mm)
+{
+	struct subpage_prot_table *spt = mm_ctx_subpage_prot(&mm->context);
+	unsigned long i, j, addr;
+	u32 **p;
+
+	if (!spt)
+		return;
+
+	for (i = 0; i < 4; ++i) {
+		if (spt->low_prot[i]) {
+			free_page((unsigned long)spt->low_prot[i]);
+			spt->low_prot[i] = NULL;
+		}
+	}
+	addr = 0;
+	for (i = 0; i < (TASK_SIZE_USER64 >> 43); ++i) {
+		p = spt->protptrs[i];
+		if (!p)
+			continue;
+		spt->protptrs[i] = NULL;
+		for (j = 0; j < SBP_L2_COUNT && addr < spt->maxaddr;
+		     ++j, addr += PAGE_SIZE)
+			if (p[j])
+				free_page((unsigned long)p[j]);
+		free_page((unsigned long)p);
+	}
+	spt->maxaddr = 0;
+	kfree(spt);
+}
+
+static void hpte_flush_range(struct mm_struct *mm, unsigned long addr,
+			     int npages)
+{
+	pgd_t *pgd;
+	p4d_t *p4d;
+	pud_t *pud;
+	pmd_t *pmd;
+	pte_t *pte;
+	spinlock_t *ptl;
+
+	pgd = pgd_offset(mm, addr);
+	p4d = p4d_offset(pgd, addr);
+	if (p4d_none(*p4d))
+		return;
+	pud = pud_offset(p4d, addr);
+	if (pud_none(*pud))
+		return;
+	pmd = pmd_offset(pud, addr);
+	if (pmd_none(*pmd))
+		return;
+	pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
+	if (!pte)
+		return;
+	lazy_mmu_mode_enable();
+	for (; npages > 0; --npages) {
+		pte_update(mm, addr, pte, 0, 0, 0);
+		addr += PAGE_SIZE;
+		++pte;
+	}
+	lazy_mmu_mode_disable();
+	pte_unmap_unlock(pte - 1, ptl);
+}
+
+/*
+ * Clear the subpage protection map for an address range, allowing
+ * all accesses that are allowed by the pte permissions.
+ */
+static void subpage_prot_clear(unsigned long addr, unsigned long len)
+{
+	struct mm_struct *mm = current->mm;
+	struct subpage_prot_table *spt;
+	u32 **spm, *spp;
+	unsigned long i;
+	size_t nw;
+	unsigned long next, limit;
+
+	mmap_write_lock(mm);
+
+	spt = mm_ctx_subpage_prot(&mm->context);
+	if (!spt)
+		goto err_out;
+
+	limit = addr + len;
+	if (limit > spt->maxaddr)
+		limit = spt->maxaddr;
+	for (; addr < limit; addr = next) {
+		next = pmd_addr_end(addr, limit);
+		if (addr < 0x100000000UL) {
+			spm = spt->low_prot;
+		} else {
+			spm = spt->protptrs[addr >> SBP_L3_SHIFT];
+			if (!spm)
+				continue;
+		}
+		spp = spm[(addr >> SBP_L2_SHIFT) & (SBP_L2_COUNT - 1)];
+		if (!spp)
+			continue;
+		spp += (addr >> PAGE_SHIFT) & (SBP_L1_COUNT - 1);
+
+		i = (addr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1);
+		nw = PTRS_PER_PTE - i;
+		if (addr + (nw << PAGE_SHIFT) > next)
+			nw = (next - addr) >> PAGE_SHIFT;
+
+		memset(spp, 0, nw * sizeof(u32));
+
+		/* now flush any existing HPTEs for the range */
+		hpte_flush_range(mm, addr, nw);
+	}
+
+err_out:
+	mmap_write_unlock(mm);
+}
+
+#ifdef CONFIG_TRANSPARENT_HUGEPAGE
+static int subpage_walk_pmd_entry(pmd_t *pmd, unsigned long addr,
+				  unsigned long end, struct mm_walk *walk)
+{
+	struct vm_area_struct *vma = walk->vma;
+	split_huge_pmd(vma, pmd, addr);
+	return 0;
+}
+
+static const struct mm_walk_ops subpage_walk_ops = {
+	.pmd_entry	= subpage_walk_pmd_entry,
+	.walk_lock	= PGWALK_WRLOCK_VERIFY,
+};
+
+static void subpage_mark_vma_nohuge(struct mm_struct *mm, unsigned long addr,
+				    unsigned long len)
+{
+	struct vm_area_struct *vma;
+	VMA_ITERATOR(vmi, mm, addr);
+
+	/*
+	 * We don't try too hard, we just mark all the vma in that range
+	 * VM_NOHUGEPAGE and split them.
+	 */
+	for_each_vma_range(vmi, vma, addr + len) {
+		vm_flags_set(vma, VM_NOHUGEPAGE);
+		walk_page_vma(vma, &subpage_walk_ops, NULL);
+	}
+}
+#else
+static void subpage_mark_vma_nohuge(struct mm_struct *mm, unsigned long addr,
+				    unsigned long len)
+{
+	return;
+}
+#endif
+
+/*
+ * Copy in a subpage protection map for an address range.
+ * The map has 2 bits per 4k subpage, so 32 bits per 64k page.
+ * Each 2-bit field is 0 to allow any access, 1 to prevent writes,
+ * 2 or 3 to prevent all accesses.
+ * Note that the normal page protections also apply; the subpage
+ * protection mechanism is an additional constraint, so putting 0
+ * in a 2-bit field won't allow writes to a page that is otherwise
+ * write-protected.
+ */
+SYSCALL_DEFINE3(subpage_prot, unsigned long, addr,
+		unsigned long, len, u32 __user *, map)
+{
+	struct mm_struct *mm = current->mm;
+	struct subpage_prot_table *spt;
+	u32 **spm, *spp;
+	unsigned long i;
+	size_t nw;
+	unsigned long next, limit;
+	int err;
+
+	if (radix_enabled())
+		return -ENOENT;
+
+	/* Check parameters */
+	if ((addr & ~PAGE_MASK) || (len & ~PAGE_MASK) ||
+	    addr >= mm->task_size || len >= mm->task_size ||
+	    addr + len > mm->task_size)
+		return -EINVAL;
+
+	if (is_hugepage_only_range(mm, addr, len))
+		return -EINVAL;
+
+	if (!map) {
+		/* Clear out the protection map for the address range */
+		subpage_prot_clear(addr, len);
+		return 0;
+	}
+
+	if (!access_ok(map, (len >> PAGE_SHIFT) * sizeof(u32)))
+		return -EFAULT;
+
+	mmap_write_lock(mm);
+
+	spt = mm_ctx_subpage_prot(&mm->context);
+	if (!spt) {
+		/*
+		 * Allocate subpage prot table if not already done.
+		 * Do this with mmap_lock held
+		 */
+		spt = kzalloc_obj(struct subpage_prot_table);
+		if (!spt) {
+			err = -ENOMEM;
+			goto out;
+		}
+		mm->context.hash_context->spt = spt;
+	}
+
+	subpage_mark_vma_nohuge(mm, addr, len);
+	for (limit = addr + len; addr < limit; addr = next) {
+		next = pmd_addr_end(addr, limit);
+		err = -ENOMEM;
+		if (addr < 0x100000000UL) {
+			spm = spt->low_prot;
+		} else {
+			spm = spt->protptrs[addr >> SBP_L3_SHIFT];
+			if (!spm) {
+				spm = (u32 **)get_zeroed_page(GFP_KERNEL);
+				if (!spm)
+					goto out;
+				spt->protptrs[addr >> SBP_L3_SHIFT] = spm;
+			}
+		}
+		spm += (addr >> SBP_L2_SHIFT) & (SBP_L2_COUNT - 1);
+		spp = *spm;
+		if (!spp) {
+			spp = (u32 *)get_zeroed_page(GFP_KERNEL);
+			if (!spp)
+				goto out;
+			*spm = spp;
+		}
+		spp += (addr >> PAGE_SHIFT) & (SBP_L1_COUNT - 1);
+
+		local_irq_disable();
+		demote_segment_4k(mm, addr);
+		local_irq_enable();
+
+		i = (addr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1);
+		nw = PTRS_PER_PTE - i;
+		if (addr + (nw << PAGE_SHIFT) > next)
+			nw = (next - addr) >> PAGE_SHIFT;
+
+		mmap_write_unlock(mm);
+		if (__copy_from_user(spp, map, nw * sizeof(u32)))
+			return -EFAULT;
+		map += nw;
+		mmap_write_lock(mm);
+
+		/* now flush any existing HPTEs for the range */
+		hpte_flush_range(mm, addr, nw);
+	}
+	if (limit > spt->maxaddr)
+		spt->maxaddr = limit;
+	err = 0;
+ out:
+	mmap_write_unlock(mm);
+	return err;
+}

linux-7.2.3/arch/powerpc/mm/book3s64/trace.c

@@ -0,0 +1,7 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * This file is for defining trace points and trace related helpers.
+ */
+#ifdef CONFIG_TRANSPARENT_HUGEPAGE
+#include <trace/events/thp.h>
+#endif