arm64: mm: move fixmap code to its own file
authorMark Rutland <mark.rutland@arm.com>
Thu, 6 Apr 2023 15:27:58 +0000 (16:27 +0100)
committerWill Deacon <will@kernel.org>
Tue, 11 Apr 2023 17:55:28 +0000 (18:55 +0100)
Over time, arm64's mm/mmu.c has become increasingly large and painful to
navigate. Move the fixmap code to its own file where it can be understood in
isolation.

There should be no functional change as a result of this patch.

Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Ard Biesheuvel <ardb@kernel.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Will Deacon <will@kernel.org>
Reviewed-by: Ryan Roberts <ryan.roberts@arm.com>
Link: https://lore.kernel.org/r/20230406152759.4164229-3-mark.rutland@arm.com
Signed-off-by: Will Deacon <will@kernel.org>
arch/arm64/include/asm/fixmap.h
arch/arm64/include/asm/mmu.h
arch/arm64/mm/Makefile
arch/arm64/mm/fixmap.c [new file with mode: 0644]
arch/arm64/mm/mmu.c

index 508238f..42e21c6 100644 (file)
@@ -103,6 +103,7 @@ enum fixed_addresses {
 #define FIXMAP_PAGE_IO     __pgprot(PROT_DEVICE_nGnRE)
 
 void __init early_fixmap_init(void);
+void __init fixmap_copy(pgd_t *pgdir);
 
 #define __early_set_fixmap __set_fixmap
 
index 48f8466..4384eaa 100644 (file)
@@ -65,6 +65,8 @@ extern void paging_init(void);
 extern void bootmem_init(void);
 extern void __iomem *early_io_map(phys_addr_t phys, unsigned long virt);
 extern void init_mem_pgprot(void);
+extern void create_mapping_noalloc(phys_addr_t phys, unsigned long virt,
+                                  phys_addr_t size, pgprot_t prot);
 extern void create_pgd_mapping(struct mm_struct *mm, phys_addr_t phys,
                               unsigned long virt, phys_addr_t size,
                               pgprot_t prot, bool page_mappings_only);
index ff1e800..dbd1bc9 100644 (file)
@@ -2,7 +2,7 @@
 obj-y                          := dma-mapping.o extable.o fault.o init.o \
                                   cache.o copypage.o flush.o \
                                   ioremap.o mmap.o pgd.o mmu.o \
-                                  context.o proc.o pageattr.o
+                                  context.o proc.o pageattr.o fixmap.o
 obj-$(CONFIG_HUGETLB_PAGE)     += hugetlbpage.o
 obj-$(CONFIG_PTDUMP_CORE)      += ptdump.o
 obj-$(CONFIG_PTDUMP_DEBUGFS)   += ptdump_debugfs.o
diff --git a/arch/arm64/mm/fixmap.c b/arch/arm64/mm/fixmap.c
new file mode 100644 (file)
index 0000000..7fbdd6f
--- /dev/null
@@ -0,0 +1,215 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Fixmap manipulation code
+ */
+
+#include <linux/bug.h>
+#include <linux/init.h>
+#include <linux/kernel.h>
+#include <linux/libfdt.h>
+#include <linux/memory.h>
+#include <linux/mm.h>
+#include <linux/sizes.h>
+
+#include <asm/fixmap.h>
+#include <asm/kernel-pgtable.h>
+#include <asm/pgalloc.h>
+#include <asm/tlbflush.h>
+
+static pte_t bm_pte[PTRS_PER_PTE] __page_aligned_bss;
+static pmd_t bm_pmd[PTRS_PER_PMD] __page_aligned_bss __maybe_unused;
+static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused;
+
+static inline pud_t *fixmap_pud(unsigned long addr)
+{
+       pgd_t *pgdp = pgd_offset_k(addr);
+       p4d_t *p4dp = p4d_offset(pgdp, addr);
+       p4d_t p4d = READ_ONCE(*p4dp);
+
+       BUG_ON(p4d_none(p4d) || p4d_bad(p4d));
+
+       return pud_offset_kimg(p4dp, addr);
+}
+
+static inline pmd_t *fixmap_pmd(unsigned long addr)
+{
+       pud_t *pudp = fixmap_pud(addr);
+       pud_t pud = READ_ONCE(*pudp);
+
+       BUG_ON(pud_none(pud) || pud_bad(pud));
+
+       return pmd_offset_kimg(pudp, addr);
+}
+
+static inline pte_t *fixmap_pte(unsigned long addr)
+{
+       return &bm_pte[pte_index(addr)];
+}
+
+/*
+ * The p*d_populate functions call virt_to_phys implicitly so they can't be used
+ * directly on kernel symbols (bm_p*d). This function is called too early to use
+ * lm_alias so __p*d_populate functions must be used to populate with the
+ * physical address from __pa_symbol.
+ */
+void __init early_fixmap_init(void)
+{
+       pgd_t *pgdp;
+       p4d_t *p4dp, p4d;
+       pud_t *pudp;
+       pmd_t *pmdp;
+       unsigned long addr = FIXADDR_TOT_START;
+
+       pgdp = pgd_offset_k(addr);
+       p4dp = p4d_offset(pgdp, addr);
+       p4d = READ_ONCE(*p4dp);
+       if (CONFIG_PGTABLE_LEVELS > 3 &&
+           !(p4d_none(p4d) || p4d_page_paddr(p4d) == __pa_symbol(bm_pud))) {
+               /*
+                * We only end up here if the kernel mapping and the fixmap
+                * share the top level pgd entry, which should only happen on
+                * 16k/4 levels configurations.
+                */
+               BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
+               pudp = pud_offset_kimg(p4dp, addr);
+       } else {
+               if (p4d_none(p4d))
+                       __p4d_populate(p4dp, __pa_symbol(bm_pud), P4D_TYPE_TABLE);
+               pudp = fixmap_pud(addr);
+       }
+       if (pud_none(READ_ONCE(*pudp)))
+               __pud_populate(pudp, __pa_symbol(bm_pmd), PUD_TYPE_TABLE);
+       pmdp = fixmap_pmd(addr);
+       __pmd_populate(pmdp, __pa_symbol(bm_pte), PMD_TYPE_TABLE);
+
+       /*
+        * The boot-ioremap range spans multiple pmds, for which
+        * we are not prepared:
+        */
+       BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
+                    != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
+
+       if ((pmdp != fixmap_pmd(__fix_to_virt(FIX_BTMAP_BEGIN)))
+            || pmdp != fixmap_pmd(__fix_to_virt(FIX_BTMAP_END))) {
+               WARN_ON(1);
+               pr_warn("pmdp %p != %p, %p\n",
+                       pmdp, fixmap_pmd(__fix_to_virt(FIX_BTMAP_BEGIN)),
+                       fixmap_pmd(__fix_to_virt(FIX_BTMAP_END)));
+               pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
+                       __fix_to_virt(FIX_BTMAP_BEGIN));
+               pr_warn("fix_to_virt(FIX_BTMAP_END):   %08lx\n",
+                       __fix_to_virt(FIX_BTMAP_END));
+
+               pr_warn("FIX_BTMAP_END:       %d\n", FIX_BTMAP_END);
+               pr_warn("FIX_BTMAP_BEGIN:     %d\n", FIX_BTMAP_BEGIN);
+       }
+}
+
+/*
+ * Unusually, this is also called in IRQ context (ghes_iounmap_irq) so if we
+ * ever need to use IPIs for TLB broadcasting, then we're in trouble here.
+ */
+void __set_fixmap(enum fixed_addresses idx,
+                              phys_addr_t phys, pgprot_t flags)
+{
+       unsigned long addr = __fix_to_virt(idx);
+       pte_t *ptep;
+
+       BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
+
+       ptep = fixmap_pte(addr);
+
+       if (pgprot_val(flags)) {
+               set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, flags));
+       } else {
+               pte_clear(&init_mm, addr, ptep);
+               flush_tlb_kernel_range(addr, addr+PAGE_SIZE);
+       }
+}
+
+void *__init fixmap_remap_fdt(phys_addr_t dt_phys, int *size, pgprot_t prot)
+{
+       const u64 dt_virt_base = __fix_to_virt(FIX_FDT);
+       int offset;
+       void *dt_virt;
+
+       /*
+        * Check whether the physical FDT address is set and meets the minimum
+        * alignment requirement. Since we are relying on MIN_FDT_ALIGN to be
+        * at least 8 bytes so that we can always access the magic and size
+        * fields of the FDT header after mapping the first chunk, double check
+        * here if that is indeed the case.
+        */
+       BUILD_BUG_ON(MIN_FDT_ALIGN < 8);
+       if (!dt_phys || dt_phys % MIN_FDT_ALIGN)
+               return NULL;
+
+       /*
+        * Make sure that the FDT region can be mapped without the need to
+        * allocate additional translation table pages, so that it is safe
+        * to call create_mapping_noalloc() this early.
+        *
+        * On 64k pages, the FDT will be mapped using PTEs, so we need to
+        * be in the same PMD as the rest of the fixmap.
+        * On 4k pages, we'll use section mappings for the FDT so we only
+        * have to be in the same PUD.
+        */
+       BUILD_BUG_ON(dt_virt_base % SZ_2M);
+
+       BUILD_BUG_ON(__fix_to_virt(FIX_FDT_END) >> SWAPPER_TABLE_SHIFT !=
+                    __fix_to_virt(FIX_BTMAP_BEGIN) >> SWAPPER_TABLE_SHIFT);
+
+       offset = dt_phys % SWAPPER_BLOCK_SIZE;
+       dt_virt = (void *)dt_virt_base + offset;
+
+       /* map the first chunk so we can read the size from the header */
+       create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE),
+                       dt_virt_base, SWAPPER_BLOCK_SIZE, prot);
+
+       if (fdt_magic(dt_virt) != FDT_MAGIC)
+               return NULL;
+
+       *size = fdt_totalsize(dt_virt);
+       if (*size > MAX_FDT_SIZE)
+               return NULL;
+
+       if (offset + *size > SWAPPER_BLOCK_SIZE)
+               create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE), dt_virt_base,
+                              round_up(offset + *size, SWAPPER_BLOCK_SIZE), prot);
+
+       return dt_virt;
+}
+
+/*
+ * Copy the fixmap region into a new pgdir.
+ */
+void __init fixmap_copy(pgd_t *pgdir)
+{
+       if (!READ_ONCE(pgd_val(*pgd_offset_pgd(pgdir, FIXADDR_TOT_START)))) {
+               /*
+                * The fixmap falls in a separate pgd to the kernel, and doesn't
+                * live in the carveout for the swapper_pg_dir. We can simply
+                * re-use the existing dir for the fixmap.
+                */
+               set_pgd(pgd_offset_pgd(pgdir, FIXADDR_TOT_START),
+                       READ_ONCE(*pgd_offset_k(FIXADDR_TOT_START)));
+       } else if (CONFIG_PGTABLE_LEVELS > 3) {
+               pgd_t *bm_pgdp;
+               p4d_t *bm_p4dp;
+               pud_t *bm_pudp;
+               /*
+                * The fixmap shares its top level pgd entry with the kernel
+                * mapping. This can really only occur when we are running
+                * with 16k/4 levels, so we can simply reuse the pud level
+                * entry instead.
+                */
+               BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
+               bm_pgdp = pgd_offset_pgd(pgdir, FIXADDR_TOT_START);
+               bm_p4dp = p4d_offset(bm_pgdp, FIXADDR_TOT_START);
+               bm_pudp = pud_set_fixmap_offset(bm_p4dp, FIXADDR_TOT_START);
+               pud_populate(&init_mm, bm_pudp, lm_alias(bm_pmd));
+               pud_clear_fixmap();
+       } else {
+               BUG();
+       }
+}
index 4495425..ef59c18 100644 (file)
@@ -73,10 +73,6 @@ long __section(".mmuoff.data.write") __early_cpu_boot_status;
 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] __page_aligned_bss;
 EXPORT_SYMBOL(empty_zero_page);
 
-static pte_t bm_pte[PTRS_PER_PTE] __page_aligned_bss;
-static pmd_t bm_pmd[PTRS_PER_PMD] __page_aligned_bss __maybe_unused;
-static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused;
-
 static DEFINE_SPINLOCK(swapper_pgdir_lock);
 static DEFINE_MUTEX(fixmap_lock);
 
@@ -452,8 +448,8 @@ static phys_addr_t pgd_pgtable_alloc(int shift)
  * without allocating new levels of table. Note that this permits the
  * creation of new section or page entries.
  */
-static void __init create_mapping_noalloc(phys_addr_t phys, unsigned long virt,
-                                 phys_addr_t size, pgprot_t prot)
+void __init create_mapping_noalloc(phys_addr_t phys, unsigned long virt,
+                                  phys_addr_t size, pgprot_t prot)
 {
        if ((virt >= PAGE_END) && (virt < VMALLOC_START)) {
                pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
@@ -795,34 +791,7 @@ static void __init map_kernel(pgd_t *pgdp)
                           &vmlinux_initdata, 0, VM_NO_GUARD);
        map_kernel_segment(pgdp, _data, _end, PAGE_KERNEL, &vmlinux_data, 0, 0);
 
-       if (!READ_ONCE(pgd_val(*pgd_offset_pgd(pgdp, FIXADDR_TOT_START)))) {
-               /*
-                * The fixmap falls in a separate pgd to the kernel, and doesn't
-                * live in the carveout for the swapper_pg_dir. We can simply
-                * re-use the existing dir for the fixmap.
-                */
-               set_pgd(pgd_offset_pgd(pgdp, FIXADDR_TOT_START),
-                       READ_ONCE(*pgd_offset_k(FIXADDR_TOT_START)));
-       } else if (CONFIG_PGTABLE_LEVELS > 3) {
-               pgd_t *bm_pgdp;
-               p4d_t *bm_p4dp;
-               pud_t *bm_pudp;
-               /*
-                * The fixmap shares its top level pgd entry with the kernel
-                * mapping. This can really only occur when we are running
-                * with 16k/4 levels, so we can simply reuse the pud level
-                * entry instead.
-                */
-               BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
-               bm_pgdp = pgd_offset_pgd(pgdp, FIXADDR_TOT_START);
-               bm_p4dp = p4d_offset(bm_pgdp, FIXADDR_TOT_START);
-               bm_pudp = pud_set_fixmap_offset(bm_p4dp, FIXADDR_TOT_START);
-               pud_populate(&init_mm, bm_pudp, lm_alias(bm_pmd));
-               pud_clear_fixmap();
-       } else {
-               BUG();
-       }
-
+       fixmap_copy(pgdp);
        kasan_copy_shadow(pgdp);
 }
 
@@ -1237,166 +1206,6 @@ void vmemmap_free(unsigned long start, unsigned long end,
 }
 #endif /* CONFIG_MEMORY_HOTPLUG */
 
-static inline pud_t *fixmap_pud(unsigned long addr)
-{
-       pgd_t *pgdp = pgd_offset_k(addr);
-       p4d_t *p4dp = p4d_offset(pgdp, addr);
-       p4d_t p4d = READ_ONCE(*p4dp);
-
-       BUG_ON(p4d_none(p4d) || p4d_bad(p4d));
-
-       return pud_offset_kimg(p4dp, addr);
-}
-
-static inline pmd_t *fixmap_pmd(unsigned long addr)
-{
-       pud_t *pudp = fixmap_pud(addr);
-       pud_t pud = READ_ONCE(*pudp);
-
-       BUG_ON(pud_none(pud) || pud_bad(pud));
-
-       return pmd_offset_kimg(pudp, addr);
-}
-
-static inline pte_t *fixmap_pte(unsigned long addr)
-{
-       return &bm_pte[pte_index(addr)];
-}
-
-/*
- * The p*d_populate functions call virt_to_phys implicitly so they can't be used
- * directly on kernel symbols (bm_p*d). This function is called too early to use
- * lm_alias so __p*d_populate functions must be used to populate with the
- * physical address from __pa_symbol.
- */
-void __init early_fixmap_init(void)
-{
-       pgd_t *pgdp;
-       p4d_t *p4dp, p4d;
-       pud_t *pudp;
-       pmd_t *pmdp;
-       unsigned long addr = FIXADDR_TOT_START;
-
-       pgdp = pgd_offset_k(addr);
-       p4dp = p4d_offset(pgdp, addr);
-       p4d = READ_ONCE(*p4dp);
-       if (CONFIG_PGTABLE_LEVELS > 3 &&
-           !(p4d_none(p4d) || p4d_page_paddr(p4d) == __pa_symbol(bm_pud))) {
-               /*
-                * We only end up here if the kernel mapping and the fixmap
-                * share the top level pgd entry, which should only happen on
-                * 16k/4 levels configurations.
-                */
-               BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
-               pudp = pud_offset_kimg(p4dp, addr);
-       } else {
-               if (p4d_none(p4d))
-                       __p4d_populate(p4dp, __pa_symbol(bm_pud), P4D_TYPE_TABLE);
-               pudp = fixmap_pud(addr);
-       }
-       if (pud_none(READ_ONCE(*pudp)))
-               __pud_populate(pudp, __pa_symbol(bm_pmd), PUD_TYPE_TABLE);
-       pmdp = fixmap_pmd(addr);
-       __pmd_populate(pmdp, __pa_symbol(bm_pte), PMD_TYPE_TABLE);
-
-       /*
-        * The boot-ioremap range spans multiple pmds, for which
-        * we are not prepared:
-        */
-       BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
-                    != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
-
-       if ((pmdp != fixmap_pmd(__fix_to_virt(FIX_BTMAP_BEGIN)))
-            || pmdp != fixmap_pmd(__fix_to_virt(FIX_BTMAP_END))) {
-               WARN_ON(1);
-               pr_warn("pmdp %p != %p, %p\n",
-                       pmdp, fixmap_pmd(__fix_to_virt(FIX_BTMAP_BEGIN)),
-                       fixmap_pmd(__fix_to_virt(FIX_BTMAP_END)));
-               pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
-                       __fix_to_virt(FIX_BTMAP_BEGIN));
-               pr_warn("fix_to_virt(FIX_BTMAP_END):   %08lx\n",
-                       __fix_to_virt(FIX_BTMAP_END));
-
-               pr_warn("FIX_BTMAP_END:       %d\n", FIX_BTMAP_END);
-               pr_warn("FIX_BTMAP_BEGIN:     %d\n", FIX_BTMAP_BEGIN);
-       }
-}
-
-/*
- * Unusually, this is also called in IRQ context (ghes_iounmap_irq) so if we
- * ever need to use IPIs for TLB broadcasting, then we're in trouble here.
- */
-void __set_fixmap(enum fixed_addresses idx,
-                              phys_addr_t phys, pgprot_t flags)
-{
-       unsigned long addr = __fix_to_virt(idx);
-       pte_t *ptep;
-
-       BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
-
-       ptep = fixmap_pte(addr);
-
-       if (pgprot_val(flags)) {
-               set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, flags));
-       } else {
-               pte_clear(&init_mm, addr, ptep);
-               flush_tlb_kernel_range(addr, addr+PAGE_SIZE);
-       }
-}
-
-void *__init fixmap_remap_fdt(phys_addr_t dt_phys, int *size, pgprot_t prot)
-{
-       const u64 dt_virt_base = __fix_to_virt(FIX_FDT);
-       int offset;
-       void *dt_virt;
-
-       /*
-        * Check whether the physical FDT address is set and meets the minimum
-        * alignment requirement. Since we are relying on MIN_FDT_ALIGN to be
-        * at least 8 bytes so that we can always access the magic and size
-        * fields of the FDT header after mapping the first chunk, double check
-        * here if that is indeed the case.
-        */
-       BUILD_BUG_ON(MIN_FDT_ALIGN < 8);
-       if (!dt_phys || dt_phys % MIN_FDT_ALIGN)
-               return NULL;
-
-       /*
-        * Make sure that the FDT region can be mapped without the need to
-        * allocate additional translation table pages, so that it is safe
-        * to call create_mapping_noalloc() this early.
-        *
-        * On 64k pages, the FDT will be mapped using PTEs, so we need to
-        * be in the same PMD as the rest of the fixmap.
-        * On 4k pages, we'll use section mappings for the FDT so we only
-        * have to be in the same PUD.
-        */
-       BUILD_BUG_ON(dt_virt_base % SZ_2M);
-
-       BUILD_BUG_ON(__fix_to_virt(FIX_FDT_END) >> SWAPPER_TABLE_SHIFT !=
-                    __fix_to_virt(FIX_BTMAP_BEGIN) >> SWAPPER_TABLE_SHIFT);
-
-       offset = dt_phys % SWAPPER_BLOCK_SIZE;
-       dt_virt = (void *)dt_virt_base + offset;
-
-       /* map the first chunk so we can read the size from the header */
-       create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE),
-                       dt_virt_base, SWAPPER_BLOCK_SIZE, prot);
-
-       if (fdt_magic(dt_virt) != FDT_MAGIC)
-               return NULL;
-
-       *size = fdt_totalsize(dt_virt);
-       if (*size > MAX_FDT_SIZE)
-               return NULL;
-
-       if (offset + *size > SWAPPER_BLOCK_SIZE)
-               create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE), dt_virt_base,
-                              round_up(offset + *size, SWAPPER_BLOCK_SIZE), prot);
-
-       return dt_virt;
-}
-
 int pud_set_huge(pud_t *pudp, phys_addr_t phys, pgprot_t prot)
 {
        pud_t new_pud = pfn_pud(__phys_to_pfn(phys), mk_pud_sect_prot(prot));