From: Dave Hansen <dave.hansen@xxxxxxxxxxxxxxx>
This bug seems familiar. I'm not sure if I hit it a while ago
and ignored it or if it is something I've seen show up in a
couple of different forms.
I booted a kernel in a KVM instance which has a bunch of
debugging turned on. Meminfo shows:
DirectMap4k: 2058232 kB
DirectMap2M: 18446744073709541376 kB
Which is a _bit_ bogus. :) In this case, I think DEBUG_PAGEALLOC
is what actually triggers this:
void free_init_pages(char *what, unsigned long begin, unsigned long end)
{
...
#ifdef CONFIG_DEBUG_PAGEALLOC
printk(KERN_INFO "debug: unmapping init [mem %#010lx-%#010lx]\n",
begin, end - 1);
set_memory_np(begin, (end - begin) >> PAGE_SHIFT);
#else
...
Here, we are freeing memory in this mapping (from
Documentation/x86/x86_64/mm.txt):
ffffffff80000000 - ffffffffa0000000 (=512 MB) kernel text mapping, from phys 0
Which we map with 2M pages. But, this is not a part of the
actual kernel linear map. The change_page_attr() code calls in
to split_page_count() since it is splitting a 2M page. We do
not have any 2M pages for the linear map (because of
DEBUG_PAGEALLOC), and the count underflows.
This patch adds a check (and an argument) to split_page_count()
to make sure that we can tell whether or not the address we are
splitting is part of the linear map.
It also changes the types of the direct_pages_count[] variables
and the accessor functions. The callers are already passing
signed variables in:
update_page_count(PG_LEVEL_1G, -pages);
to a function with unsigned arguments:
void update_page_count(int level, unsigned long pages)
so we might as well make them signed. We make the
direct_pages_count[] variables signed so that we can do easy
checks when they underflow.
---
linux.git-davehans/arch/x86/include/asm/pgtable_types.h | 4 +-
linux.git-davehans/arch/x86/mm/init_32.c | 2 -
linux.git-davehans/arch/x86/mm/pageattr.c | 32 ++++++++++++----
3 files changed, 27 insertions(+), 11 deletions(-)
diff -puN arch/x86/include/asm/pgtable_types.h~mm-meminfo-DirectMap2M-underflow arch/x86/include/asm/pgtable_types.h
--- linux.git/arch/x86/include/asm/pgtable_types.h~mm-meminfo-DirectMap2M-underflow 2013-06-28 10:55:42.528074131 -0700
+++ linux.git-davehans/arch/x86/include/asm/pgtable_types.h 2013-06-28 10:55:42.535074443 -0700
@@ -339,9 +339,9 @@ enum pg_level {
};
#ifdef CONFIG_PROC_FS
-extern void update_page_count(int level, unsigned long pages);
+extern void update_page_count(int level, int pages);
#else
-static inline void update_page_count(int level, unsigned long pages) { }
+static inline void update_page_count(int level, int pages) { }
#endif
/*
diff -puN arch/x86/mm/init_32.c~mm-meminfo-DirectMap2M-underflow arch/x86/mm/init_32.c
--- linux.git/arch/x86/mm/init_32.c~mm-meminfo-DirectMap2M-underflow 2013-06-28 10:55:42.530074220 -0700
+++ linux.git-davehans/arch/x86/mm/init_32.c 2013-06-28 10:55:42.536074487 -0700
@@ -261,7 +261,7 @@ kernel_physical_mapping_init(unsigned lo
pgd_t *pgd;
pmd_t *pmd;
pte_t *pte;
- unsigned pages_2m, pages_4k;
+ int pages_2m, pages_4k;
int mapping_iter;
start_pfn = start >> PAGE_SHIFT;
diff -puN arch/x86/mm/pageattr.c~mm-meminfo-DirectMap2M-underflow arch/x86/mm/pageattr.c
--- linux.git/arch/x86/mm/pageattr.c~mm-meminfo-DirectMap2M-underflow 2013-06-28 10:55:42.532074309 -0700
+++ linux.git-davehans/arch/x86/mm/pageattr.c 2013-06-28 10:55:42.537074532 -0700
@@ -53,31 +53,47 @@ static DEFINE_SPINLOCK(cpa_lock);
#define CPA_PAGES_ARRAY 4
#ifdef CONFIG_PROC_FS
-static unsigned long direct_pages_count[PG_LEVEL_NUM];
+static long direct_pages_count[PG_LEVEL_NUM];
-void update_page_count(int level, unsigned long pages)
+static void check_direct_pages_count(int level)
+{
+ WARN_ONCE(direct_pages_count[level] < 0,
+ "page table count underflow level: %d", level);
+}
+
+void update_page_count(int level, int pages)
{
/* Protect against CPA */
spin_lock(&pgd_lock);
direct_pages_count[level] += pages;
+ check_direct_pages_count(level);
spin_unlock(&pgd_lock);
}
-static void split_page_count(int level)
+static void split_page_count(unsigned long address, int level)
{
+ /*
+ * We only keep these pagetable counts for memory in
+ * the linear map. The things we do not care about
+ * and do not track are all at or above VMALLOC_START.
+ */
+ if (address >= VMALLOC_START)
+ return;
+
direct_pages_count[level]--;
+ check_direct_pages_count(level);
direct_pages_count[level - 1] += PTRS_PER_PTE;
}
void arch_report_meminfo(struct seq_file *m)
{
- seq_printf(m, "DirectMap4k: %8lu kB\n",
+ seq_printf(m, "DirectMap4k: %8ld kB\n",
direct_pages_count[PG_LEVEL_4K] << 2);
#if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
- seq_printf(m, "DirectMap2M: %8lu kB\n",
+ seq_printf(m, "DirectMap2M: %8ld kB\n",
direct_pages_count[PG_LEVEL_2M] << 11);
#else
- seq_printf(m, "DirectMap4M: %8lu kB\n",
+ seq_printf(m, "DirectMap4M: %8ld kB\n",
direct_pages_count[PG_LEVEL_2M] << 12);
#endif
#ifdef CONFIG_X86_64
@@ -87,7 +103,7 @@ void arch_report_meminfo(struct seq_file
#endif
}
#else
-static inline void split_page_count(int level) { }
+static inline void split_page_count(unsigned long address, int level) { }
#endif
#ifdef CONFIG_X86_64
@@ -607,7 +623,7 @@ __split_large_page(pte_t *kpte, unsigned
if (pfn_range_is_mapped(PFN_DOWN(__pa(address)),
PFN_DOWN(__pa(address)) + 1))
- split_page_count(level);
+ split_page_count(address, level);
/*
* Install the new, split up pagetable.
_
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>