[PATCH 2/3] powerpc/64s: POWER10 nest MMU can upgrade PTE access authority without TLB flush

[Nicholas Piggin]

The nest MMU in POWER9 does not re-fetch the PTE in response to
permission mismatch, contrary to the architecture[*] and unlike the core
MMU. This requires a TLB flush before upgrading permissions of valid
PTEs, for any address space with a coprocessor attached.

Per (non-public) Nest MMU Workbook, POWER10 nest MMU conforms to the
architecture in this regard, so skip the workaround.

[*] See: Power ISA Version 3.1B, 6.10.1.2 Modifying a Translation Table
    Entry, Setting a Reference or Change Bit or Upgrading Access
    Authority (PTE Subject to Atomic Hardware Updates):

      "If the only change being made to a valid PTE that is subject to
       atomic hardware updates is to set the Reference or Change bit to
       1 or to upgrade access authority, a simpler sequence suffices
       because the translation hardware will refetch the PTE if an
       access is attempted for which the only problems were reference
       and/or change bits needing to be set or insufficient access
       authority."

Signed-off-by: Nicholas Piggin <npiggin at gmail.com>
---
 arch/powerpc/mm/book3s64/radix_hugetlbpage.c | 10 +++---
 arch/powerpc/mm/book3s64/radix_pgtable.c     | 35 ++++++++++++--------
 2 files changed, 28 insertions(+), 17 deletions(-)

diff --git a/arch/powerpc/mm/book3s64/radix_hugetlbpage.c b/arch/powerpc/mm/book3s64/radix_hugetlbpage.c
index 23d3e08911d3..78618c9b618b 100644
--- a/arch/powerpc/mm/book3s64/radix_hugetlbpage.c
+++ b/arch/powerpc/mm/book3s64/radix_hugetlbpage.c
@@ -103,11 +103,13 @@ void radix__huge_ptep_modify_prot_commit(struct vm_area_struct *vma,
 	struct mm_struct *mm = vma->vm_mm;
 
 	/*
-	 * To avoid NMMU hang while relaxing access we need to flush the tlb before
-	 * we set the new value.
+	 * 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 (is_pte_rw_upgrade(pte_val(old_pte), pte_val(pte)) &&
-	    (atomic_read(&mm->context.copros) > 0))
+	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);
diff --git a/arch/powerpc/mm/book3s64/radix_pgtable.c b/arch/powerpc/mm/book3s64/radix_pgtable.c
index def04631a74d..195719a6c41c 100644
--- a/arch/powerpc/mm/book3s64/radix_pgtable.c
+++ b/arch/powerpc/mm/book3s64/radix_pgtable.c
@@ -1018,16 +1018,21 @@ void radix__ptep_set_access_flags(struct vm_area_struct *vma, pte_t *ptep,
 
 	unsigned long change = pte_val(entry) ^ pte_val(*ptep);
 	/*
-	 * To avoid NMMU hang while relaxing access, we need mark
-	 * the pte invalid in between.
+	 * On POWER9, the NMMU is not able to relax PTE access permissions
+	 * for a translation with a TLB. The PTE must be invalidated, TLB
+	 * flushed before the new PTE is installed.
+	 *
+	 * This only needs to be done for radix, because hash translation does
+	 * flush when updating the linux pte (and we don't support NMMU
+	 * accelerators on HPT on POWER9 anyway XXX: do we?).
+	 *
+	 * POWER10 (and P9P) NMMU does behave as per ISA.
 	 */
-	if ((change & _PAGE_RW) && atomic_read(&mm->context.copros) > 0) {
+	if (!cpu_has_feature(CPU_FTR_ARCH_31) && (change & _PAGE_RW) &&
+	    atomic_read(&mm->context.copros) > 0) {
 		unsigned long old_pte, new_pte;
 
 		old_pte = __radix_pte_update(ptep, _PAGE_PRESENT, _PAGE_INVALID);
-		/*
-		 * new value of pte
-		 */
 		new_pte = old_pte | set;
 		radix__flush_tlb_page_psize(mm, address, psize);
 		__radix_pte_update(ptep, _PAGE_INVALID, new_pte);
@@ -1035,9 +1040,12 @@ void radix__ptep_set_access_flags(struct vm_area_struct *vma, pte_t *ptep,
 		__radix_pte_update(ptep, 0, set);
 		/*
 		 * Book3S does not require a TLB flush when relaxing access
-		 * restrictions when the address space is not attached to a
-		 * NMMU, because the core MMU will reload the pte after taking
-		 * an access fault, which is defined by the architecture.
+		 * restrictions when the address space (modulo the POWER9 nest
+		 * MMU issue above) because the MMU will reload the PTE after
+		 * taking an access fault, as defined by the architecture. See
+		 * "Setting a Reference or Change Bit or Upgrading Access
+		 *  Authority (PTE Subject to Atomic Hardware Updates)" in
+		 *  Power ISA Version 3.1B.
 		 */
 	}
 	/* See ptesync comment in radix__set_pte_at */
@@ -1050,11 +1058,12 @@ void radix__ptep_modify_prot_commit(struct vm_area_struct *vma,
 	struct mm_struct *mm = vma->vm_mm;
 
 	/*
-	 * To avoid NMMU hang while relaxing access we need to flush the tlb before
-	 * we set the new value. We need to do this only for radix, because hash
-	 * translation does flush when updating the linux pte.
+	 * 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 (is_pte_rw_upgrade(pte_val(old_pte), pte_val(pte)) &&
+	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_tlb_page(vma, addr);
 
-- 
2.35.1