diff mbox series

[RFC,v2,08/21] riscv: mm: Reimplement page table entry atomic get function

Message ID 20241205103729.14798-9-luxu.kernel@bytedance.com (mailing list archive)
State New
Headers show
Series riscv: Introduce 64K base page | expand

Checks

Context Check Description
conchuod/vmtest-for-next-PR fail PR summary
conchuod/patch-8-test-1 success .github/scripts/patches/tests/build_rv32_defconfig.sh took 207.91s
conchuod/patch-8-test-2 success .github/scripts/patches/tests/build_rv64_clang_allmodconfig.sh took 2568.41s
conchuod/patch-8-test-3 success .github/scripts/patches/tests/build_rv64_gcc_allmodconfig.sh took 2954.33s
conchuod/patch-8-test-4 success .github/scripts/patches/tests/build_rv64_nommu_k210_defconfig.sh took 79.41s
conchuod/patch-8-test-5 success .github/scripts/patches/tests/build_rv64_nommu_virt_defconfig.sh took 80.93s
conchuod/patch-8-test-6 warning .github/scripts/patches/tests/checkpatch.sh took 0.85s
conchuod/patch-8-test-7 success .github/scripts/patches/tests/dtb_warn_rv64.sh took 48.35s
conchuod/patch-8-test-8 success .github/scripts/patches/tests/header_inline.sh took 0.01s
conchuod/patch-8-test-9 success .github/scripts/patches/tests/kdoc.sh took 0.63s
conchuod/patch-8-test-10 success .github/scripts/patches/tests/module_param.sh took 0.02s
conchuod/patch-8-test-11 success .github/scripts/patches/tests/verify_fixes.sh took 0.00s
conchuod/patch-8-test-12 success .github/scripts/patches/tests/verify_signedoff.sh took 0.04s
conchuod/vmtest-fixes-PR fail merge-conflict

Commit Message

Xu Lu Dec. 5, 2024, 10:37 a.m. UTC
This commit implements lockless functions to atomically fetch pte's
value. For each pte structure, we atomically fetch the first mapping
entry, and then fetch the following entries and compare them with the
first mappin entry plus certain step path in a loop. If we find any
difference in their pfns or prots, then the pte structure has been
modified and need to be reloaded.

Signed-off-by: Xu Lu <luxu.kernel@bytedance.com>
---
 arch/riscv/include/asm/pgtable.h | 156 +++++++++++++++++++++++++++++++
 include/linux/pgtable.h          |  21 +++++
 2 files changed, 177 insertions(+)
diff mbox series

Patch

diff --git a/arch/riscv/include/asm/pgtable.h b/arch/riscv/include/asm/pgtable.h
index ba4a083b7210..fe42afb4441e 100644
--- a/arch/riscv/include/asm/pgtable.h
+++ b/arch/riscv/include/asm/pgtable.h
@@ -220,6 +220,18 @@  static inline unsigned long satp_pfn(unsigned long satp)
 	return hwpfn_to_pfn(hwpfn);
 }
 
+static inline unsigned long __pte_pgprot(unsigned long pteval)
+{
+	unsigned long prot_mask = GENMASK(_PAGE_HWPFN_SHIFT - 1, 0);
+
+	return pteval & prot_mask;
+}
+
+static inline pgprot_t pte_pgprot(pte_t pte)
+{
+	return __pgprot(__pte_pgprot(pte_val(pte)));
+}
+
 static inline int __pgd_leaf(unsigned long pgdval)
 {
 	return __pgd_present(pgdval) && (pgdval & _PAGE_LEAF);
@@ -734,6 +746,150 @@  static inline pgd_t pgdp_get(pgd_t *pgdp)
 }
 #define pgdp_get	pgdp_get
 
+#ifdef CONFIG_RISCV_USE_SW_PAGE
+static inline pte_t ptep_get_lockless(pte_t *ptep)
+{
+	unsigned long pteval;
+	pte_t pte;
+	int i;
+
+retry:
+	pteval = READ_ONCE(ptep->ptes[0]);
+	pte = *ptep;
+	for (i = 0; i < HW_PAGES_PER_PAGE; i++) {
+		if (__page_val_to_pfn(pteval) !=
+		    __page_val_to_pfn(pte.ptes[i]))
+			goto retry;
+		if ((__pte_pgprot(pteval) | _PAGE_DIRTY | _PAGE_ACCESSED) !=
+		    (__pte_pgprot(pte.ptes[i]) | _PAGE_DIRTY | _PAGE_ACCESSED))
+			goto retry;
+
+		if (__pte_present(pteval) && !__pte_napot(pteval))
+			pteval += 1 << _PAGE_HWPFN_SHIFT;
+	}
+
+	return pte;
+}
+#define ptep_get_lockless	ptep_get_lockless
+
+static inline pmd_t pmdp_get_lockless(pmd_t *pmdp)
+{
+	unsigned long pmdval;
+	pmd_t pmd;
+	int i;
+
+retry:
+	pmdval = READ_ONCE(pmdp->pmds[0]);
+	pmd = *pmdp;
+	for (i = 0; i < HW_PAGES_PER_PAGE; i++) {
+		if (__page_val_to_pfn(pmdval) !=
+		    __page_val_to_pfn(pmd.pmds[i]))
+			goto retry;
+		if ((__pte_pgprot(pmdval) | _PAGE_DIRTY | _PAGE_ACCESSED) !=
+		    (__pte_pgprot(pmd.pmds[i]) | _PAGE_DIRTY | _PAGE_ACCESSED))
+			goto retry;
+
+		if (__pmd_leaf(pmdval))
+			pmdval += (1 << (PMD_SHIFT - PAGE_SHIFT)) <<
+					_PAGE_HWPFN_SHIFT;
+		else if (__pmd_present(pmdval))
+			pmdval += 1 << _PAGE_HWPFN_SHIFT;
+	}
+
+	return pmd;
+}
+#define pmdp_get_lockless	pmdp_get_lockless
+
+static inline void pmdp_get_lockless_sync(void)
+{
+}
+
+static inline pud_t pudp_get_lockless(pud_t *pudp)
+{
+	unsigned long pudval;
+	pud_t pud;
+	int i;
+
+retry:
+	pudval = READ_ONCE(pudp->puds[0]);
+	pud = *pudp;
+	for (i = 0; i < HW_PAGES_PER_PAGE; i++) {
+		if (__page_val_to_pfn(pudval) !=
+		    __page_val_to_pfn(pud.puds[i]))
+			goto retry;
+		if ((__pte_pgprot(pudval) | _PAGE_DIRTY | _PAGE_ACCESSED) !=
+		    (__pte_pgprot(pud.puds[i]) | _PAGE_DIRTY | _PAGE_ACCESSED))
+			goto retry;
+
+		if (__pud_leaf(pudval))
+			pudval += (1 << (PUD_SHIFT - PAGE_SHIFT)) <<
+					_PAGE_HWPFN_SHIFT;
+		else if (__pud_present(pudval))
+			pudval += 1 << _PAGE_HWPFN_SHIFT;
+	}
+
+	return pud;
+}
+#define pudp_get_lockless	pudp_get_lockless
+
+static inline p4d_t p4dp_get_lockless(p4d_t *p4dp)
+{
+	unsigned long p4dval;
+	p4d_t p4d;
+	int i;
+
+retry:
+	p4dval = READ_ONCE(p4dp->p4ds[0]);
+	p4d = *p4dp;
+	for (i = 0; i < HW_PAGES_PER_PAGE; i++) {
+		if (__page_val_to_pfn(p4dval) !=
+		    __page_val_to_pfn(p4d.p4ds[i]))
+			goto retry;
+		if ((__pte_pgprot(p4dval) | _PAGE_DIRTY | _PAGE_ACCESSED) !=
+		    (__pte_pgprot(p4d.p4ds[i]) | _PAGE_DIRTY | _PAGE_ACCESSED))
+			goto retry;
+
+		if (__p4d_leaf(p4dval))
+			p4dval += (1 << (P4D_SHIFT - PAGE_SHIFT)) <<
+					_PAGE_HWPFN_SHIFT;
+		else if (__p4d_present(p4dval))
+			p4dval += 1 << _PAGE_HWPFN_SHIFT;
+	}
+
+	return p4d;
+}
+#define p4dp_get_lockless	p4dp_get_lockless
+
+static inline pgd_t pgdp_get_lockless(pgd_t *pgdp)
+{
+	unsigned long pgdval;
+	pgd_t pgd;
+	int i;
+
+retry:
+	pgdval = READ_ONCE(pgdp->pgds[0]);
+	pgd = *pgdp;
+	for (i = 0; i < HW_PAGES_PER_PAGE; i++) {
+		if (__page_val_to_pfn(pgdval) !=
+		    __page_val_to_pfn(pgd.pgds[i]))
+			goto retry;
+		if ((__pte_pgprot(pgdval) | _PAGE_DIRTY | _PAGE_ACCESSED) !=
+		    (__pte_pgprot(pgd.pgds[i]) | _PAGE_DIRTY | _PAGE_ACCESSED))
+			goto retry;
+
+		if (__pgd_leaf(pgdval))
+			pgdval += (1 << (PGDIR_SHIFT - PAGE_SHIFT)) <<
+					_PAGE_HWPFN_SHIFT;
+		else if (__pgd_present(pgdval))
+			pgdval += 1 << _PAGE_HWPFN_SHIFT;
+	}
+
+	return pgd;
+}
+#define pgdp_get_lockless	pgdp_get_lockless
+
+#endif /* CONFIG_RISCV_USE_SW_PAGE */
+
 void flush_icache_pte(struct mm_struct *mm, pte_t pte);
 
 static inline void __set_pte_at(struct mm_struct *mm, pte_t *ptep, pte_t pteval)
diff --git a/include/linux/pgtable.h b/include/linux/pgtable.h
index e8b2ac6bd2ae..b629c48b980b 100644
--- a/include/linux/pgtable.h
+++ b/include/linux/pgtable.h
@@ -598,6 +598,27 @@  static inline void pmdp_get_lockless_sync(void)
 }
 #endif
 
+#ifndef pudp_get_lockless
+static inline pud_t pudp_get_lockless(pud_t *pudp)
+{
+	return pudp_get(pudp);
+}
+#endif
+
+#ifndef p4dp_get_lockless
+static inline p4d_t p4dp_get_lockless(p4d_t *p4dp)
+{
+	return p4dp_get(p4dp);
+}
+#endif
+
+#ifndef pgdp_get_lockless
+static inline pgd_t pgdp_get_lockless(pgd_t *pgdp)
+{
+	return pgdp_get(pgdp);
+}
+#endif
+
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,