* [PATCH v4 1/4] x86/mm: Update mapped addresses in phys_{pmd,pud}_init()
2025-06-13 20:09 [PATCH v4 0/4] x86/mm: Improve alloc handling of phys_*_init() Em Sharnoff
@ 2025-06-13 20:10 ` Em Sharnoff
2025-06-13 20:11 ` [PATCH v4 2/4] x86/mm: Allow error returns from phys_*_init() Em Sharnoff
` (2 subsequent siblings)
3 siblings, 0 replies; 5+ messages in thread
From: Em Sharnoff @ 2025-06-13 20:10 UTC (permalink / raw)
To: linux-kernel, x86, linux-mm
Cc: Ingo Molnar, H. Peter Anvin, Dave Hansen, Andy Lutomirski,
Peter Zijlstra, Thomas Gleixner, Borislav Petkov,
Edgecombe, Rick P, Oleg Vasilev, Arthur Petukhovsky, Stefan Radig,
Misha Sakhnov
Currently kernel_physical_mapping_init() and its dependents return the
last physical address mapped ('paddr_last'). This makes it harder to
cleanly handle allocation errors in those functions.
'paddr_last' is used to update 'pfn_mapped'/'max_pfn_mapped', so:
1. Introduce add_paddr_range_mapped() to do the update, translating from
physical addresses to pfns
2. Call add_paddr_range_mapped() in phys_pud_init() where 'paddr_last'
would otherwise be updated due to 1Gi pages.
- Note: this includes places where we set 'paddr_last = paddr_next',
as was added in 20167d3421a0 ("x86-64: Fix accounting in
kernel_physical_mapping_init()")
add_paddr_range_mapped() is probably too expensive to be called every
time a page is updated, so instead, phys_pte_init() continues to return
'paddr_last', and phys_pmd_init() calls add_paddr_range_mapped() only at
the end of the loop (should mean it's called every 1Gi).
Signed-off-by: Em Sharnoff <sharnoff@neon.tech>
---
Changelog:
- v4: Add this patch
---
arch/x86/include/asm/pgtable.h | 3 +-
arch/x86/mm/init.c | 23 +++++----
arch/x86/mm/init_32.c | 6 ++-
arch/x86/mm/init_64.c | 88 +++++++++++++++++-----------------
arch/x86/mm/mm_internal.h | 13 +++--
5 files changed, 69 insertions(+), 64 deletions(-)
diff --git a/arch/x86/include/asm/pgtable.h b/arch/x86/include/asm/pgtable.h
index 7bd6bd6df4a1..138d55f48a4f 100644
--- a/arch/x86/include/asm/pgtable.h
+++ b/arch/x86/include/asm/pgtable.h
@@ -1244,8 +1244,7 @@ extern int direct_gbpages;
void init_mem_mapping(void);
void early_alloc_pgt_buf(void);
void __init poking_init(void);
-unsigned long init_memory_mapping(unsigned long start,
- unsigned long end, pgprot_t prot);
+void init_memory_mapping(unsigned long start, unsigned long end, pgprot_t prot);
#ifdef CONFIG_X86_64
extern pgd_t trampoline_pgd_entry;
diff --git a/arch/x86/mm/init.c b/arch/x86/mm/init.c
index bfa444a7dbb0..1461873b44f1 100644
--- a/arch/x86/mm/init.c
+++ b/arch/x86/mm/init.c
@@ -529,16 +529,24 @@ bool pfn_range_is_mapped(unsigned long start_pfn, unsigned long end_pfn)
return false;
}
+/*
+ * Update max_pfn_mapped and range_pfn_mapped with the range of physical
+ * addresses mapped. The range may overlap with previous calls to this function.
+ */
+void add_paddr_range_mapped(unsigned long start_paddr, unsigned long end_paddr)
+{
+ add_pfn_range_mapped(start_paddr >> PAGE_SHIFT, end_paddr >> PAGE_SHIFT);
+}
+
/*
* Setup the direct mapping of the physical memory at PAGE_OFFSET.
* This runs before bootmem is initialized and gets pages directly from
* the physical memory. To access them they are temporarily mapped.
*/
-unsigned long __ref init_memory_mapping(unsigned long start,
- unsigned long end, pgprot_t prot)
+void __ref init_memory_mapping(unsigned long start,
+ unsigned long end, pgprot_t prot)
{
struct map_range mr[NR_RANGE_MR];
- unsigned long ret = 0;
int nr_range, i;
pr_debug("init_memory_mapping: [mem %#010lx-%#010lx]\n",
@@ -548,13 +556,10 @@ unsigned long __ref init_memory_mapping(unsigned long start,
nr_range = split_mem_range(mr, 0, start, end);
for (i = 0; i < nr_range; i++)
- ret = kernel_physical_mapping_init(mr[i].start, mr[i].end,
- mr[i].page_size_mask,
- prot);
+ kernel_physical_mapping_init(mr[i].start, mr[i].end,
+ mr[i].page_size_mask, prot);
- add_pfn_range_mapped(start >> PAGE_SHIFT, ret >> PAGE_SHIFT);
-
- return ret >> PAGE_SHIFT;
+ return;
}
/*
diff --git a/arch/x86/mm/init_32.c b/arch/x86/mm/init_32.c
index ad662cc4605c..4427ac433041 100644
--- a/arch/x86/mm/init_32.c
+++ b/arch/x86/mm/init_32.c
@@ -245,7 +245,7 @@ static inline int is_x86_32_kernel_text(unsigned long addr)
* of max_low_pfn pages, by creating page tables starting from address
* PAGE_OFFSET:
*/
-unsigned long __init
+void __init
kernel_physical_mapping_init(unsigned long start,
unsigned long end,
unsigned long page_size_mask,
@@ -382,7 +382,9 @@ kernel_physical_mapping_init(unsigned long start,
mapping_iter = 2;
goto repeat;
}
- return last_map_addr;
+
+ add_paddr_range_mapped(start, last_map_addr);
+ return;
}
#ifdef CONFIG_HIGHMEM
diff --git a/arch/x86/mm/init_64.c b/arch/x86/mm/init_64.c
index 7c4f6f591f2b..e729108bee30 100644
--- a/arch/x86/mm/init_64.c
+++ b/arch/x86/mm/init_64.c
@@ -502,13 +502,13 @@ phys_pte_init(pte_t *pte_page, unsigned long paddr, unsigned long paddr_end,
/*
* Create PMD level page table mapping for physical addresses. The virtual
* and physical address have to be aligned at this level.
- * It returns the last physical address mapped.
*/
-static unsigned long __meminit
+static void __meminit
phys_pmd_init(pmd_t *pmd_page, unsigned long paddr, unsigned long paddr_end,
unsigned long page_size_mask, pgprot_t prot, bool init)
{
unsigned long pages = 0, paddr_next;
+ unsigned long paddr_first = paddr;
unsigned long paddr_last = paddr_end;
int i = pmd_index(paddr);
@@ -579,21 +579,25 @@ phys_pmd_init(pmd_t *pmd_page, unsigned long paddr, unsigned long paddr_end,
spin_unlock(&init_mm.page_table_lock);
}
update_page_count(PG_LEVEL_2M, pages);
- return paddr_last;
+ /*
+ * In case of recovery from previous state, add_paddr_range_mapped() may
+ * be called with an overlapping range from previous operations.
+ * It is idempotent, so this is ok.
+ */
+ add_paddr_range_mapped(paddr_first, paddr_last);
+ return;
}
/*
* Create PUD level page table mapping for physical addresses. The virtual
* and physical address do not have to be aligned at this level. KASLR can
* randomize virtual addresses up to this level.
- * It returns the last physical address mapped.
*/
-static unsigned long __meminit
+static void __meminit
phys_pud_init(pud_t *pud_page, unsigned long paddr, unsigned long paddr_end,
unsigned long page_size_mask, pgprot_t _prot, bool init)
{
unsigned long pages = 0, paddr_next;
- unsigned long paddr_last = paddr_end;
unsigned long vaddr = (unsigned long)__va(paddr);
int i = pud_index(vaddr);
@@ -619,10 +623,8 @@ phys_pud_init(pud_t *pud_page, unsigned long paddr, unsigned long paddr_end,
if (!pud_none(*pud)) {
if (!pud_leaf(*pud)) {
pmd = pmd_offset(pud, 0);
- paddr_last = phys_pmd_init(pmd, paddr,
- paddr_end,
- page_size_mask,
- prot, init);
+ phys_pmd_init(pmd, paddr, paddr_end,
+ page_size_mask, prot, init);
continue;
}
/*
@@ -640,7 +642,7 @@ phys_pud_init(pud_t *pud_page, unsigned long paddr, unsigned long paddr_end,
if (page_size_mask & (1 << PG_LEVEL_1G)) {
if (!after_bootmem)
pages++;
- paddr_last = paddr_next;
+ add_paddr_range_mapped(paddr, paddr_next);
continue;
}
prot = pte_pgprot(pte_clrhuge(*(pte_t *)pud));
@@ -653,13 +655,13 @@ phys_pud_init(pud_t *pud_page, unsigned long paddr, unsigned long paddr_end,
pfn_pud(paddr >> PAGE_SHIFT, prot_sethuge(prot)),
init);
spin_unlock(&init_mm.page_table_lock);
- paddr_last = paddr_next;
+ add_paddr_range_mapped(paddr, paddr_next);
continue;
}
pmd = alloc_low_page();
- paddr_last = phys_pmd_init(pmd, paddr, paddr_end,
- page_size_mask, prot, init);
+ phys_pmd_init(pmd, paddr, paddr_end,
+ page_size_mask, prot, init);
spin_lock(&init_mm.page_table_lock);
pud_populate_init(&init_mm, pud, pmd, init);
@@ -668,22 +670,23 @@ phys_pud_init(pud_t *pud_page, unsigned long paddr, unsigned long paddr_end,
update_page_count(PG_LEVEL_1G, pages);
- return paddr_last;
+ return;
}
-static unsigned long __meminit
+static void __meminit
phys_p4d_init(p4d_t *p4d_page, unsigned long paddr, unsigned long paddr_end,
unsigned long page_size_mask, pgprot_t prot, bool init)
{
- unsigned long vaddr, vaddr_end, vaddr_next, paddr_next, paddr_last;
+ unsigned long vaddr, vaddr_end, vaddr_next, paddr_next;
- paddr_last = paddr_end;
vaddr = (unsigned long)__va(paddr);
vaddr_end = (unsigned long)__va(paddr_end);
- if (!pgtable_l5_enabled())
- return phys_pud_init((pud_t *) p4d_page, paddr, paddr_end,
- page_size_mask, prot, init);
+ if (!pgtable_l5_enabled()) {
+ phys_pud_init((pud_t *) p4d_page, paddr, paddr_end,
+ page_size_mask, prot, init);
+ return;
+ }
for (; vaddr < vaddr_end; vaddr = vaddr_next) {
p4d_t *p4d = p4d_page + p4d_index(vaddr);
@@ -705,33 +708,32 @@ phys_p4d_init(p4d_t *p4d_page, unsigned long paddr, unsigned long paddr_end,
if (!p4d_none(*p4d)) {
pud = pud_offset(p4d, 0);
- paddr_last = phys_pud_init(pud, paddr, __pa(vaddr_end),
- page_size_mask, prot, init);
+ phys_pud_init(pud, paddr, __pa(vaddr_end),
+ page_size_mask, prot, init);
continue;
}
pud = alloc_low_page();
- paddr_last = phys_pud_init(pud, paddr, __pa(vaddr_end),
- page_size_mask, prot, init);
+ phys_pud_init(pud, paddr, __pa(vaddr_end),
+ page_size_mask, prot, init);
spin_lock(&init_mm.page_table_lock);
p4d_populate_init(&init_mm, p4d, pud, init);
spin_unlock(&init_mm.page_table_lock);
}
- return paddr_last;
+ return;
}
-static unsigned long __meminit
+static void __meminit
__kernel_physical_mapping_init(unsigned long paddr_start,
unsigned long paddr_end,
unsigned long page_size_mask,
pgprot_t prot, bool init)
{
bool pgd_changed = false;
- unsigned long vaddr, vaddr_start, vaddr_end, vaddr_next, paddr_last;
+ unsigned long vaddr, vaddr_start, vaddr_end, vaddr_next;
- paddr_last = paddr_end;
vaddr = (unsigned long)__va(paddr_start);
vaddr_end = (unsigned long)__va(paddr_end);
vaddr_start = vaddr;
@@ -744,16 +746,14 @@ __kernel_physical_mapping_init(unsigned long paddr_start,
if (pgd_val(*pgd)) {
p4d = (p4d_t *)pgd_page_vaddr(*pgd);
- paddr_last = phys_p4d_init(p4d, __pa(vaddr),
- __pa(vaddr_end),
- page_size_mask,
- prot, init);
+ phys_p4d_init(p4d, __pa(vaddr), __pa(vaddr_end),
+ page_size_mask, prot, init);
continue;
}
p4d = alloc_low_page();
- paddr_last = phys_p4d_init(p4d, __pa(vaddr), __pa(vaddr_end),
- page_size_mask, prot, init);
+ phys_p4d_init(p4d, __pa(vaddr), __pa(vaddr_end),
+ page_size_mask, prot, init);
spin_lock(&init_mm.page_table_lock);
if (pgtable_l5_enabled())
@@ -769,7 +769,7 @@ __kernel_physical_mapping_init(unsigned long paddr_start,
if (pgd_changed)
sync_global_pgds(vaddr_start, vaddr_end - 1);
- return paddr_last;
+ return;
}
@@ -777,15 +777,15 @@ __kernel_physical_mapping_init(unsigned long paddr_start,
* Create page table mapping for the physical memory for specific physical
* addresses. Note that it can only be used to populate non-present entries.
* The virtual and physical addresses have to be aligned on PMD level
- * down. It returns the last physical address mapped.
+ * down.
*/
-unsigned long __meminit
+void __meminit
kernel_physical_mapping_init(unsigned long paddr_start,
unsigned long paddr_end,
unsigned long page_size_mask, pgprot_t prot)
{
- return __kernel_physical_mapping_init(paddr_start, paddr_end,
- page_size_mask, prot, true);
+ __kernel_physical_mapping_init(paddr_start, paddr_end,
+ page_size_mask, prot, true);
}
/*
@@ -794,14 +794,14 @@ kernel_physical_mapping_init(unsigned long paddr_start,
* when updating the mapping. The caller is responsible to flush the TLBs after
* the function returns.
*/
-unsigned long __meminit
+void __meminit
kernel_physical_mapping_change(unsigned long paddr_start,
unsigned long paddr_end,
unsigned long page_size_mask)
{
- return __kernel_physical_mapping_init(paddr_start, paddr_end,
- page_size_mask, PAGE_KERNEL,
- false);
+ __kernel_physical_mapping_init(paddr_start, paddr_end,
+ page_size_mask, PAGE_KERNEL,
+ false);
}
#ifndef CONFIG_NUMA
diff --git a/arch/x86/mm/mm_internal.h b/arch/x86/mm/mm_internal.h
index 3f37b5c80bb3..6fea5f7edd48 100644
--- a/arch/x86/mm/mm_internal.h
+++ b/arch/x86/mm/mm_internal.h
@@ -10,13 +10,12 @@ static inline void *alloc_low_page(void)
void early_ioremap_page_table_range_init(void);
-unsigned long kernel_physical_mapping_init(unsigned long start,
- unsigned long end,
- unsigned long page_size_mask,
- pgprot_t prot);
-unsigned long kernel_physical_mapping_change(unsigned long start,
- unsigned long end,
- unsigned long page_size_mask);
+void add_paddr_range_mapped(unsigned long start_paddr, unsigned long end_paddr);
+
+void kernel_physical_mapping_init(unsigned long start, unsigned long end,
+ unsigned long page_size_mask, pgprot_t prot);
+void kernel_physical_mapping_change(unsigned long start, unsigned long end,
+ unsigned long page_size_mask);
void zone_sizes_init(void);
extern int after_bootmem;
--
2.39.5
^ permalink raw reply related [flat|nested] 5+ messages in thread* [PATCH v4 2/4] x86/mm: Allow error returns from phys_*_init()
2025-06-13 20:09 [PATCH v4 0/4] x86/mm: Improve alloc handling of phys_*_init() Em Sharnoff
2025-06-13 20:10 ` [PATCH v4 1/4] x86/mm: Update mapped addresses in phys_{pmd,pud}_init() Em Sharnoff
@ 2025-06-13 20:11 ` Em Sharnoff
2025-06-13 20:11 ` [PATCH v4 3/4] x86/mm: Handle alloc failure in phys_*_init() Em Sharnoff
2025-06-13 20:12 ` [PATCH v4 4/4] x86/mm: Use GFP_KERNEL for alloc_low_pages() after boot Em Sharnoff
3 siblings, 0 replies; 5+ messages in thread
From: Em Sharnoff @ 2025-06-13 20:11 UTC (permalink / raw)
To: linux-kernel, x86, linux-mm
Cc: Ingo Molnar, H. Peter Anvin, Dave Hansen, Andy Lutomirski,
Peter Zijlstra, Thomas Gleixner, Borislav Petkov,
Edgecombe, Rick P, Oleg Vasilev, Arthur Petukhovsky, Stefan Radig,
Misha Sakhnov
Preparation for returning errors when alloc_low_page() fails.
phys_pte_init() is excluded because it can't fail, and it's useful for
it to return 'paddr_last' instead.
This patch depends on the previous patch ("x86/mm: Update mapped
addresses in phys_{pmd,pud}_init()").
Signed-off-by: Em Sharnoff <sharnoff@neon.tech>
---
Changleog:
- v2: Switch from special-casing zero value to using ERR_PTR()
- v3: Fix -Wint-conversion errors
- v4: Switch return type to int, split alloc handling into separate patch.
---
arch/x86/include/asm/pgtable.h | 2 +-
arch/x86/mm/init.c | 14 +++--
arch/x86/mm/init_32.c | 4 +-
arch/x86/mm/init_64.c | 100 ++++++++++++++++++++++-----------
arch/x86/mm/mem_encrypt_amd.c | 8 ++-
arch/x86/mm/mm_internal.h | 8 +--
6 files changed, 87 insertions(+), 49 deletions(-)
diff --git a/arch/x86/include/asm/pgtable.h b/arch/x86/include/asm/pgtable.h
index 138d55f48a4f..b09194c42688 100644
--- a/arch/x86/include/asm/pgtable.h
+++ b/arch/x86/include/asm/pgtable.h
@@ -1244,7 +1244,7 @@ extern int direct_gbpages;
void init_mem_mapping(void);
void early_alloc_pgt_buf(void);
void __init poking_init(void);
-void init_memory_mapping(unsigned long start, unsigned long end, pgprot_t prot);
+int init_memory_mapping(unsigned long start, unsigned long end, pgprot_t prot);
#ifdef CONFIG_X86_64
extern pgd_t trampoline_pgd_entry;
diff --git a/arch/x86/mm/init.c b/arch/x86/mm/init.c
index 1461873b44f1..1dc8fbd8cb63 100644
--- a/arch/x86/mm/init.c
+++ b/arch/x86/mm/init.c
@@ -543,11 +543,12 @@ void add_paddr_range_mapped(unsigned long start_paddr, unsigned long end_paddr)
* This runs before bootmem is initialized and gets pages directly from
* the physical memory. To access them they are temporarily mapped.
*/
-void __ref init_memory_mapping(unsigned long start,
+int __ref init_memory_mapping(unsigned long start,
unsigned long end, pgprot_t prot)
{
struct map_range mr[NR_RANGE_MR];
int nr_range, i;
+ int ret;
pr_debug("init_memory_mapping: [mem %#010lx-%#010lx]\n",
start, end - 1);
@@ -555,11 +556,14 @@ void __ref init_memory_mapping(unsigned long start,
memset(mr, 0, sizeof(mr));
nr_range = split_mem_range(mr, 0, start, end);
- for (i = 0; i < nr_range; i++)
- kernel_physical_mapping_init(mr[i].start, mr[i].end,
- mr[i].page_size_mask, prot);
+ for (i = 0; i < nr_range; i++) {
+ ret = kernel_physical_mapping_init(mr[i].start, mr[i].end,
+ mr[i].page_size_mask, prot);
+ if (ret)
+ return ret;
+ }
- return;
+ return 0;
}
/*
diff --git a/arch/x86/mm/init_32.c b/arch/x86/mm/init_32.c
index 4427ac433041..57bd154c206d 100644
--- a/arch/x86/mm/init_32.c
+++ b/arch/x86/mm/init_32.c
@@ -245,7 +245,7 @@ static inline int is_x86_32_kernel_text(unsigned long addr)
* of max_low_pfn pages, by creating page tables starting from address
* PAGE_OFFSET:
*/
-void __init
+int __init
kernel_physical_mapping_init(unsigned long start,
unsigned long end,
unsigned long page_size_mask,
@@ -384,7 +384,7 @@ kernel_physical_mapping_init(unsigned long start,
}
add_paddr_range_mapped(start, last_map_addr);
- return;
+ return 0;
}
#ifdef CONFIG_HIGHMEM
diff --git a/arch/x86/mm/init_64.c b/arch/x86/mm/init_64.c
index e729108bee30..b18ab2dcc799 100644
--- a/arch/x86/mm/init_64.c
+++ b/arch/x86/mm/init_64.c
@@ -503,7 +503,7 @@ phys_pte_init(pte_t *pte_page, unsigned long paddr, unsigned long paddr_end,
* Create PMD level page table mapping for physical addresses. The virtual
* and physical address have to be aligned at this level.
*/
-static void __meminit
+static int __meminit
phys_pmd_init(pmd_t *pmd_page, unsigned long paddr, unsigned long paddr_end,
unsigned long page_size_mask, pgprot_t prot, bool init)
{
@@ -585,7 +585,7 @@ phys_pmd_init(pmd_t *pmd_page, unsigned long paddr, unsigned long paddr_end,
* It is idempotent, so this is ok.
*/
add_paddr_range_mapped(paddr_first, paddr_last);
- return;
+ return 0;
}
/*
@@ -593,12 +593,14 @@ phys_pmd_init(pmd_t *pmd_page, unsigned long paddr, unsigned long paddr_end,
* and physical address do not have to be aligned at this level. KASLR can
* randomize virtual addresses up to this level.
*/
-static void __meminit
+static int __meminit
phys_pud_init(pud_t *pud_page, unsigned long paddr, unsigned long paddr_end,
unsigned long page_size_mask, pgprot_t _prot, bool init)
{
unsigned long pages = 0, paddr_next;
unsigned long vaddr = (unsigned long)__va(paddr);
+ int ret;
+
int i = pud_index(vaddr);
for (; i < PTRS_PER_PUD; i++, paddr = paddr_next) {
@@ -623,8 +625,10 @@ phys_pud_init(pud_t *pud_page, unsigned long paddr, unsigned long paddr_end,
if (!pud_none(*pud)) {
if (!pud_leaf(*pud)) {
pmd = pmd_offset(pud, 0);
- phys_pmd_init(pmd, paddr, paddr_end,
- page_size_mask, prot, init);
+ ret = phys_pmd_init(pmd, paddr, paddr_end,
+ page_size_mask, prot, init);
+ if (ret)
+ return ret;
continue;
}
/*
@@ -660,33 +664,39 @@ phys_pud_init(pud_t *pud_page, unsigned long paddr, unsigned long paddr_end,
}
pmd = alloc_low_page();
- phys_pmd_init(pmd, paddr, paddr_end,
- page_size_mask, prot, init);
+ ret = phys_pmd_init(pmd, paddr, paddr_end,
+ page_size_mask, prot, init);
spin_lock(&init_mm.page_table_lock);
pud_populate_init(&init_mm, pud, pmd, init);
spin_unlock(&init_mm.page_table_lock);
+
+ /*
+ * Bail only after updating pud to keep progress from pmd across
+ * retries.
+ */
+ if (ret)
+ return ret;
}
update_page_count(PG_LEVEL_1G, pages);
- return;
+ return 0;
}
-static void __meminit
+static int __meminit
phys_p4d_init(p4d_t *p4d_page, unsigned long paddr, unsigned long paddr_end,
unsigned long page_size_mask, pgprot_t prot, bool init)
{
unsigned long vaddr, vaddr_end, vaddr_next, paddr_next;
+ int ret;
vaddr = (unsigned long)__va(paddr);
vaddr_end = (unsigned long)__va(paddr_end);
- if (!pgtable_l5_enabled()) {
- phys_pud_init((pud_t *) p4d_page, paddr, paddr_end,
- page_size_mask, prot, init);
- return;
- }
+ if (!pgtable_l5_enabled())
+ return phys_pud_init((pud_t *) p4d_page, paddr, paddr_end,
+ page_size_mask, prot, init);
for (; vaddr < vaddr_end; vaddr = vaddr_next) {
p4d_t *p4d = p4d_page + p4d_index(vaddr);
@@ -708,24 +718,33 @@ phys_p4d_init(p4d_t *p4d_page, unsigned long paddr, unsigned long paddr_end,
if (!p4d_none(*p4d)) {
pud = pud_offset(p4d, 0);
- phys_pud_init(pud, paddr, __pa(vaddr_end),
- page_size_mask, prot, init);
+ ret = phys_pud_init(pud, paddr, __pa(vaddr_end),
+ page_size_mask, prot, init);
+ if (ret)
+ return ret;
continue;
}
pud = alloc_low_page();
- phys_pud_init(pud, paddr, __pa(vaddr_end),
- page_size_mask, prot, init);
+ ret = phys_pud_init(pud, paddr, __pa(vaddr_end),
+ page_size_mask, prot, init);
spin_lock(&init_mm.page_table_lock);
p4d_populate_init(&init_mm, p4d, pud, init);
spin_unlock(&init_mm.page_table_lock);
+
+ /*
+ * Bail only after updating p4d to keep progress from pud across
+ * retries.
+ */
+ if (ret)
+ return ret;
}
- return;
+ return 0;
}
-static void __meminit
+static int __meminit
__kernel_physical_mapping_init(unsigned long paddr_start,
unsigned long paddr_end,
unsigned long page_size_mask,
@@ -733,6 +752,7 @@ __kernel_physical_mapping_init(unsigned long paddr_start,
{
bool pgd_changed = false;
unsigned long vaddr, vaddr_start, vaddr_end, vaddr_next;
+ int ret;
vaddr = (unsigned long)__va(paddr_start);
vaddr_end = (unsigned long)__va(paddr_end);
@@ -746,14 +766,16 @@ __kernel_physical_mapping_init(unsigned long paddr_start,
if (pgd_val(*pgd)) {
p4d = (p4d_t *)pgd_page_vaddr(*pgd);
- phys_p4d_init(p4d, __pa(vaddr), __pa(vaddr_end),
- page_size_mask, prot, init);
+ ret = phys_p4d_init(p4d, __pa(vaddr), __pa(vaddr_end),
+ page_size_mask, prot, init);
+ if (ret)
+ return ret;
continue;
}
p4d = alloc_low_page();
- phys_p4d_init(p4d, __pa(vaddr), __pa(vaddr_end),
- page_size_mask, prot, init);
+ ret = phys_p4d_init(p4d, __pa(vaddr), __pa(vaddr_end),
+ page_size_mask, prot, init);
spin_lock(&init_mm.page_table_lock);
if (pgtable_l5_enabled())
@@ -761,15 +783,22 @@ __kernel_physical_mapping_init(unsigned long paddr_start,
else
p4d_populate_init(&init_mm, p4d_offset(pgd, vaddr),
(pud_t *) p4d, init);
-
spin_unlock(&init_mm.page_table_lock);
+
+ /*
+ * Bail only after updating pgd/p4d to keep progress from p4d
+ * across retries.
+ */
+ if (ret)
+ return ret;
+
pgd_changed = true;
}
if (pgd_changed)
sync_global_pgds(vaddr_start, vaddr_end - 1);
- return;
+ return 0;
}
@@ -779,13 +808,13 @@ __kernel_physical_mapping_init(unsigned long paddr_start,
* The virtual and physical addresses have to be aligned on PMD level
* down.
*/
-void __meminit
+int __meminit
kernel_physical_mapping_init(unsigned long paddr_start,
unsigned long paddr_end,
unsigned long page_size_mask, pgprot_t prot)
{
- __kernel_physical_mapping_init(paddr_start, paddr_end,
- page_size_mask, prot, true);
+ return __kernel_physical_mapping_init(paddr_start, paddr_end,
+ page_size_mask, prot, true);
}
/*
@@ -794,14 +823,14 @@ kernel_physical_mapping_init(unsigned long paddr_start,
* when updating the mapping. The caller is responsible to flush the TLBs after
* the function returns.
*/
-void __meminit
+int __meminit
kernel_physical_mapping_change(unsigned long paddr_start,
unsigned long paddr_end,
unsigned long page_size_mask)
{
- __kernel_physical_mapping_init(paddr_start, paddr_end,
- page_size_mask, PAGE_KERNEL,
- false);
+ return __kernel_physical_mapping_init(paddr_start, paddr_end,
+ page_size_mask, PAGE_KERNEL,
+ false);
}
#ifndef CONFIG_NUMA
@@ -980,8 +1009,11 @@ int arch_add_memory(int nid, u64 start, u64 size,
{
unsigned long start_pfn = start >> PAGE_SHIFT;
unsigned long nr_pages = size >> PAGE_SHIFT;
+ int ret;
- init_memory_mapping(start, start + size, params->pgprot);
+ ret = init_memory_mapping(start, start + size, params->pgprot);
+ if (ret)
+ return ret;
return add_pages(nid, start_pfn, nr_pages, params);
}
diff --git a/arch/x86/mm/mem_encrypt_amd.c b/arch/x86/mm/mem_encrypt_amd.c
index 7490ff6d83b1..04e0b92eb9ad 100644
--- a/arch/x86/mm/mem_encrypt_amd.c
+++ b/arch/x86/mm/mem_encrypt_amd.c
@@ -444,9 +444,11 @@ static int __init early_set_memory_enc_dec(unsigned long vaddr,
* kernel_physical_mapping_change() does not flush the TLBs, so
* a TLB flush is required after we exit from the for loop.
*/
- kernel_physical_mapping_change(__pa(vaddr & pmask),
- __pa((vaddr_end & pmask) + psize),
- split_page_size_mask);
+ ret = kernel_physical_mapping_change(__pa(vaddr & pmask),
+ __pa((vaddr_end & pmask) + psize),
+ split_page_size_mask);
+ if (ret)
+ return ret;
}
ret = 0;
diff --git a/arch/x86/mm/mm_internal.h b/arch/x86/mm/mm_internal.h
index 6fea5f7edd48..dacf3c924fbd 100644
--- a/arch/x86/mm/mm_internal.h
+++ b/arch/x86/mm/mm_internal.h
@@ -12,10 +12,10 @@ void early_ioremap_page_table_range_init(void);
void add_paddr_range_mapped(unsigned long start_paddr, unsigned long end_paddr);
-void kernel_physical_mapping_init(unsigned long start, unsigned long end,
- unsigned long page_size_mask, pgprot_t prot);
-void kernel_physical_mapping_change(unsigned long start, unsigned long end,
- unsigned long page_size_mask);
+int kernel_physical_mapping_init(unsigned long start, unsigned long end,
+ unsigned long page_size_mask, pgprot_t prot);
+int kernel_physical_mapping_change(unsigned long start, unsigned long end,
+ unsigned long page_size_mask);
void zone_sizes_init(void);
extern int after_bootmem;
--
2.39.5
^ permalink raw reply related [flat|nested] 5+ messages in thread* [PATCH v4 3/4] x86/mm: Handle alloc failure in phys_*_init()
2025-06-13 20:09 [PATCH v4 0/4] x86/mm: Improve alloc handling of phys_*_init() Em Sharnoff
2025-06-13 20:10 ` [PATCH v4 1/4] x86/mm: Update mapped addresses in phys_{pmd,pud}_init() Em Sharnoff
2025-06-13 20:11 ` [PATCH v4 2/4] x86/mm: Allow error returns from phys_*_init() Em Sharnoff
@ 2025-06-13 20:11 ` Em Sharnoff
2025-06-13 20:12 ` [PATCH v4 4/4] x86/mm: Use GFP_KERNEL for alloc_low_pages() after boot Em Sharnoff
3 siblings, 0 replies; 5+ messages in thread
From: Em Sharnoff @ 2025-06-13 20:11 UTC (permalink / raw)
To: linux-kernel, x86, linux-mm
Cc: Ingo Molnar, H. Peter Anvin, Dave Hansen, Andy Lutomirski,
Peter Zijlstra, Thomas Gleixner, Borislav Petkov,
Edgecombe, Rick P, Oleg Vasilev, Arthur Petukhovsky, Stefan Radig,
Misha Sakhnov
During memory hotplug, allocation failures in phys_*_init() aren't
handled, which results in a null pointer dereference if they occur.
This patch depends on the previous patch ("x86/mm: Allow error returns
from phys_*_init()").
Signed-off-by: Em Sharnoff <sharnoff@neon.tech>
---
Changelog:
- v4: Split this patch out from the error handling changes
---
arch/x86/mm/init_64.c | 8 ++++++++
1 file changed, 8 insertions(+)
diff --git a/arch/x86/mm/init_64.c b/arch/x86/mm/init_64.c
index b18ab2dcc799..a585daa76d69 100644
--- a/arch/x86/mm/init_64.c
+++ b/arch/x86/mm/init_64.c
@@ -572,6 +572,8 @@ phys_pmd_init(pmd_t *pmd_page, unsigned long paddr, unsigned long paddr_end,
}
pte = alloc_low_page();
+ if (!pte)
+ return -ENOMEM;
paddr_last = phys_pte_init(pte, paddr, paddr_end, new_prot, init);
spin_lock(&init_mm.page_table_lock);
@@ -664,6 +666,8 @@ phys_pud_init(pud_t *pud_page, unsigned long paddr, unsigned long paddr_end,
}
pmd = alloc_low_page();
+ if (!pmd)
+ return -ENOMEM;
ret = phys_pmd_init(pmd, paddr, paddr_end,
page_size_mask, prot, init);
@@ -726,6 +730,8 @@ phys_p4d_init(p4d_t *p4d_page, unsigned long paddr, unsigned long paddr_end,
}
pud = alloc_low_page();
+ if (!pud)
+ return -ENOMEM;
ret = phys_pud_init(pud, paddr, __pa(vaddr_end),
page_size_mask, prot, init);
@@ -774,6 +780,8 @@ __kernel_physical_mapping_init(unsigned long paddr_start,
}
p4d = alloc_low_page();
+ if (!p4d)
+ return -ENOMEM;
ret = phys_p4d_init(p4d, __pa(vaddr), __pa(vaddr_end),
page_size_mask, prot, init);
--
2.39.5
^ permalink raw reply related [flat|nested] 5+ messages in thread* [PATCH v4 4/4] x86/mm: Use GFP_KERNEL for alloc_low_pages() after boot
2025-06-13 20:09 [PATCH v4 0/4] x86/mm: Improve alloc handling of phys_*_init() Em Sharnoff
` (2 preceding siblings ...)
2025-06-13 20:11 ` [PATCH v4 3/4] x86/mm: Handle alloc failure in phys_*_init() Em Sharnoff
@ 2025-06-13 20:12 ` Em Sharnoff
3 siblings, 0 replies; 5+ messages in thread
From: Em Sharnoff @ 2025-06-13 20:12 UTC (permalink / raw)
To: linux-kernel, x86, linux-mm
Cc: Ingo Molnar, H. Peter Anvin, Dave Hansen, Andy Lutomirski,
Peter Zijlstra, Thomas Gleixner, Borislav Petkov,
Edgecombe, Rick P, Oleg Vasilev, Arthur Petukhovsky, Stefan Radig,
Misha Sakhnov
Currently it's GFP_ATOMIC. GFP_KERNEL seems more correct.
From Ingo M. [1]
> There's no real reason why it should be GFP_ATOMIC AFAICS, other than
> some historic inertia that nobody bothered to fix.
and previously Mike R. [2]
> The few callers that effectively use page allocator for the direct map
> updates are gart_iommu_init() and memory hotplug. Neither of them
> happen in an atomic context so there is no reason to use GFP_ATOMIC
> for these allocations.
>
> Replace GFP_ATOMIC with GFP_KERNEL to avoid using atomic reserves for
> allocations that do not require that.
[1]: https://lore.kernel.org/all/aEE6_S2a-1tk1dtI@gmail.com/
[2]: https://lore.kernel.org/all/20211111110241.25968-5-rppt@kernel.org/
Signed-off-by: Em Sharnoff <sharnoff@neon.tech>
---
Changelog:
- v2: Add this patch
- v3: No changes
- v4: No changes
---
arch/x86/mm/init.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/arch/x86/mm/init.c b/arch/x86/mm/init.c
index 1dc8fbd8cb63..80eddcef992b 100644
--- a/arch/x86/mm/init.c
+++ b/arch/x86/mm/init.c
@@ -131,7 +131,7 @@ __ref void *alloc_low_pages(unsigned int num)
unsigned int order;
order = get_order((unsigned long)num << PAGE_SHIFT);
- return (void *)__get_free_pages(GFP_ATOMIC | __GFP_ZERO, order);
+ return (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, order);
}
if ((pgt_buf_end + num) > pgt_buf_top || !can_use_brk_pgt) {
--
2.39.5
^ permalink raw reply related [flat|nested] 5+ messages in thread