Skip to contentSkip to conversation

generic debug pagealloc

PATCH In review Akinobu Mita 6 participants 9 messages Mar 3, 2009 – Mar 6, 2009

Subsystems (8)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 generic debug pagealloc

CONFIG_DEBUG_PAGEALLOC is now supported by x86, powerpc, sparc (64bit), and s390. This patch implements it for the rest of the architectures by filling the pages with poison byte patterns after free_pages() and verifying the poison patterns before alloc_pages().

This generic one cannot detect invalid read accesses and it can only detect invalid write accesses after a long delay. But it is an feasible way for nommu architectures.

---
 arch/avr32/mm/fault.c      |   18 ----------
 arch/powerpc/Kconfig       |    3 +
 arch/powerpc/Kconfig.debug |    1 
 arch/s390/Kconfig          |    3 +
 arch/s390/Kconfig.debug    |    1 
 arch/sparc/Kconfig         |    3 +
 arch/sparc/Kconfig.debug   |    3 +
 arch/x86/Kconfig           |    3 +
 arch/x86/Kconfig.debug     |    1 
 include/linux/mm_types.h   |    4 ++
 include/linux/poison.h     |    3 +
 lib/Kconfig.debug          |    1 
 mm/Kconfig.debug           |    8 ++++
 mm/Makefile                |    1 
 mm/debug-pagealloc.c       |   81 +++++++++++++++++++++++++++++++++++++++++++++
 15 files changed, 115 insertions(+), 19 deletions(-)
include/linux/poison.h +3 -0
Index: 2.6-poison/include/linux/poison.h
===================================================================
@@ -17,6 +17,9 @@
  */
 #define TIMER_ENTRY_STATIC	((void *) 0x74737461)
 
+/********** mm/debug-pagealloc.c **********/
+#define PAGE_POISON 0xaa
+
 /********** mm/slab.c **********/
 /*
  * Magic nums for obj red zoning.
mm/debug-pagealloc.c new file +81 -0

81 lines not on this page · raw message

mm/Kconfig.debug new file +8 -0
Index: 2.6-poison/mm/Kconfig.debug
===================================================================
@@ -0,0 +1,8 @@
+config PAGE_POISONING
+	bool "Debug page memory allocations"
+	depends on !ARCH_SUPPORTS_DEBUG_PAGEALLOC
+	select DEBUG_PAGEALLOC
+	help
+	   Fill the pages with poison patterns after free_pages() and verify
+	   the patterns before alloc_pages(). This results in a large slowdown,
+	   but helps to find certain types of memory corruptions.
mm/Makefile +1 -0
Index: 2.6-poison/mm/Makefile
===================================================================
@@ -33,3 +33,4 @@ obj-$(CONFIG_MIGRATION) += migrate.o
 obj-$(CONFIG_SMP) += allocpercpu.o
 obj-$(CONFIG_QUICKLIST) += quicklist.o
 obj-$(CONFIG_CGROUP_MEM_RES_CTLR) += memcontrol.o page_cgroup.o
+obj-$(CONFIG_PAGE_POISONING) += debug-pagealloc.o
lib/Kconfig.debug +1 -0
Index: 2.6-poison/lib/Kconfig.debug
===================================================================
@@ -796,6 +796,7 @@ config SYSCTL_SYSCALL_CHECK
 	  to properly maintain and use. This enables checks that help
 	  you to keep things correct.
 
+source mm/Kconfig.debug
 source kernel/trace/Kconfig
 
 config PROVIDE_OHCI1394_DMA_INIT
arch/powerpc/Kconfig.debug +1 -0
Index: 2.6-poison/arch/powerpc/Kconfig.debug
===================================================================
@@ -30,6 +30,7 @@ config DEBUG_STACK_USAGE
 config DEBUG_PAGEALLOC
         bool "Debug page memory allocations"
         depends on DEBUG_KERNEL && !HIBERNATION
+	depends on ARCH_SUPPORTS_DEBUG_PAGEALLOC
         help
           Unmap pages from the kernel linear mapping after free_pages().
           This results in a large slowdown, but helps to find certain types
arch/s390/Kconfig.debug +1 -0
Index: 2.6-poison/arch/s390/Kconfig.debug
===================================================================
@@ -9,6 +9,7 @@ source "lib/Kconfig.debug"
 config DEBUG_PAGEALLOC
 	bool "Debug page memory allocations"
 	depends on DEBUG_KERNEL
+	depends on ARCH_SUPPORTS_DEBUG_PAGEALLOC
 	help
 	  Unmap pages from the kernel linear mapping after free_pages().
 	  This results in a slowdown, but helps to find certain types of
arch/sparc/Kconfig.debug +2 -1
Index: 2.6-poison/arch/sparc/Kconfig.debug
===================================================================
@@ -24,7 +24,8 @@ config STACK_DEBUG
 
 config DEBUG_PAGEALLOC
 	bool "Debug page memory allocations"
-	depends on SPARC64 && DEBUG_KERNEL && !HIBERNATION
+	depends on DEBUG_KERNEL && !HIBERNATION
+	depends on ARCH_SUPPORTS_DEBUG_PAGEALLOC
 	help
 	  Unmap pages from the kernel linear mapping after free_pages().
 	  This results in a large slowdown, but helps to find certain types
arch/x86/Kconfig +3 -0
Index: 2.6-poison/arch/x86/Kconfig
===================================================================
@@ -160,6 +160,9 @@ config AUDIT_ARCH
 config ARCH_SUPPORTS_OPTIMIZED_INLINING
 	def_bool y
 
+config ARCH_SUPPORTS_DEBUG_PAGEALLOC
+	def_bool y
+
 # Use the generic interrupt handling code in kernel/irq/:
 config GENERIC_HARDIRQS
 	bool
arch/x86/Kconfig.debug +1 -0
Index: 2.6-poison/arch/x86/Kconfig.debug
===================================================================
@@ -75,6 +75,7 @@ config DEBUG_STACK_USAGE
 config DEBUG_PAGEALLOC
 	bool "Debug page memory allocations"
 	depends on DEBUG_KERNEL
+	depends on ARCH_SUPPORTS_DEBUG_PAGEALLOC
 	help
 	  Unmap pages from the kernel linear mapping after free_pages().
 	  This results in a large slowdown, but helps to find certain types
arch/powerpc/Kconfig +3 -0
Index: 2.6-poison/arch/powerpc/Kconfig
===================================================================
@@ -227,6 +227,9 @@ config PPC_OF_PLATFORM_PCI
 	depends on PPC64 # not supported on 32 bits yet
 	default n
 
+config ARCH_SUPPORTS_DEBUG_PAGEALLOC
+	def_bool y
+
 source "init/Kconfig"
 
 source "kernel/Kconfig.freezer"
arch/sparc/Kconfig +3 -0
Index: 2.6-poison/arch/sparc/Kconfig
===================================================================
@@ -124,6 +124,9 @@ config ARCH_NO_VIRT_TO_BUS
 config OF
 	def_bool y
 
+config ARCH_SUPPORTS_DEBUG_PAGEALLOC
+	def_bool y if SPARC64
+
 source "init/Kconfig"
 
 source "kernel/Kconfig.freezer"
arch/avr32/mm/fault.c +0 -18
Index: 2.6-poison/arch/avr32/mm/fault.c
===================================================================
@@ -250,21 +250,3 @@ asmlinkage void do_bus_error(unsigned lo
 	dump_dtlb();
 	die("Bus Error", regs, SIGKILL);
 }
-
-/*
- * This functionality is currently not possible to implement because
- * we're using segmentation to ensure a fixed mapping of the kernel
- * virtual address space.
- *
- * It would be possible to implement this, but it would require us to
- * disable segmentation at startup and load the kernel mappings into
- * the TLB like any other pages. There will be lots of trickery to
- * avoid recursive invocation of the TLB miss handler, though...
- */
-#ifdef CONFIG_DEBUG_PAGEALLOC
-void kernel_map_pages(struct page *page, int numpages, int enable)
-{
-
-}
-EXPORT_SYMBOL(kernel_map_pages);
-#endif
arch/s390/Kconfig +3 -0
Index: 2.6-poison/arch/s390/Kconfig
===================================================================
@@ -72,6 +72,9 @@ config PGSTE
 config VIRT_CPU_ACCOUNTING
 	def_bool y
 
+config ARCH_SUPPORTS_DEBUG_PAGEALLOC
+	def_bool y
+
 mainmenu "Linux Kernel Configuration"
 
 config S390
include/linux/mm_types.h +4 -0
Index: 2.6-poison/include/linux/mm_types.h
===================================================================
@@ -94,6 +94,10 @@ struct page {
 	void *virtual;			/* Kernel virtual address (NULL if
 					   not kmapped, ie. highmem) */
 #endif /* WANT_PAGE_VIRTUAL */
+
+#ifdef CONFIG_PAGE_POISONING
+	bool poison;
+#endif /* CONFIG_PAGE_POISONING */
 };
 
 /*

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#1 lore raw
Quoted 9 lines · Akinobu Mita

CONFIG_DEBUG_PAGEALLOC is now supported by x86, powerpc, sparc (64bit), and s390. This patch implements it for the rest of the architectures by filling the pages with poison byte patterns after free_pages() and verifying the poison patterns before alloc_pages().

This generic one cannot detect invalid read accesses and it can only detect invalid write accesses after a long delay. But it is an feasible way for nommu architectures.

if every architecture supports it now then i guess this config switch can go away:

Review

Quoted 2 lines
(quoted diff) +2 -0
+config ARCH_SUPPORTS_DEBUG_PAGEALLOC
+	def_bool y

Ingo

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#2 lore raw
Quoted 5 lines · Ingo Molnar

if every architecture supports it now then i guess this config switch can go away:

Review

Quoted 2 lines
(quoted diff) +2 -0
+config ARCH_SUPPORTS_DEBUG_PAGEALLOC
+	def_bool y

The generic debug pagealloc needs the poison flag for each struct page.

So I introduced CONFIG_ARCH_SUPPORTS_DEBUG_PAGEALLOC for x86, powerpc, sparc (64bit), and s390. If there is no CONFIG_ARCH_SUPPORTS_DEBUG_PAGEALLOC, make config prompts the generic debug pagealloc in mm/Kconfig.debug for the other architectures.

I was trying to add cleaner config dependency but CONFIG_ARCH_SUPPORTS_DEBUG_PAGEALLOC is my solution for now.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#3 lore raw
In reply to Akinobu Mita

On 3.3.2009 17:01, Akinobu Mita wrote:
...

Quoted 25 lines
mm/debug-pagealloc.c new file +25 -0
+static void dump_broken_mem(unsigned char *mem)
+{
+	int i;
+	int start = 0;
+	int end = PAGE_SIZE - 1;
+
+	for (i = 0; i<  PAGE_SIZE; i++) {
+		if (mem[i] != PAGE_POISON) {
+			start = i;
+			break;
+		}
+	}
+	for (i = PAGE_SIZE - 1; i>= start; i--) {
+		if (mem[i] != PAGE_POISON) {
+			end = i;
+			break;
+		}
+	}
+	printk(KERN_ERR "Page corruption: %p-%p\n", mem + start, mem + end);
+	print_hex_dump(KERN_ERR, "", DUMP_PREFIX_ADDRESS, 16, 1, mem + start,
+			end - start + 1, 1);
+}
+
+static void unpoison_page(struct page *page)
+{

mm/debug-pagealloc.c line 56

Quoted 10 lines
mm/debug-pagealloc.c new file +10 -0
+	unsigned char *mem;
+	int i;
+
+	if (!page->poison)
+		return;
+
+	mem = kmap_atomic(page, KM_USER0);
+	for (i = 0; i<  PAGE_SIZE; i++) {
+		if (mem[i] != PAGE_POISON) {
+			dump_broken_mem(mem);

Just an optimisation: pass the i to the dump_broken_mem as a start index.

mm/debug-pagealloc.c line 62

Quoted 6 lines
mm/debug-pagealloc.c new file +6 -0
+			break;
+		}
+	}
+	kunmap_atomic(mem, KM_USER0);
+	page->poison = false;
+}

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#4 lore raw
Quoted 1 line · Jiri Slaby

Just an optimisation: pass the i to the dump_broken_mem as a start index.

OK. I'll remove that redundancy in next verserion.

maintainer
#5 lore raw
In reply to Akinobu Mita
Quoted 27 lines · On Wed, 4 Mar 2009 01:01:04 +0900 Akinobu Mita <[email protected]> wrote:
mm/debug-pagealloc.c new file +27 -0
+static void unpoison_page(struct page *page)
+{
+	unsigned char *mem;
+	int i;
+
+	if (!page->poison)
+		return;
+
+	mem = kmap_atomic(page, KM_USER0);
+	for (i = 0; i < PAGE_SIZE; i++) {
+		if (mem[i] != PAGE_POISON) {
+			dump_broken_mem(mem);
+			break;
+		}
+	}
+	kunmap_atomic(mem, KM_USER0);
+	page->poison = false;
+}
+
+static void unpoison_pages(struct page *page, int n)
+{
+	int i;
+
+	for (i = 0; i < n; i++)
+		unpoison_page(page + i);
+}
+

mm/debug-pagealloc.c line 81

Quoted 10 lines
mm/debug-pagealloc.c new file +10 -0
+void kernel_map_pages(struct page *page, int numpages, int enable)
+{
+	if (!debug_pagealloc_enabled)
+		return;
+
+	if (enable)
+		unpoison_pages(page, numpages);
+	else
+		poison_pages(page, numpages);
+}

kernel_map_pages() is called from the memory-allocation and memory-freeing paths. Hence it can be called from interrupt contexts.

KM_USER0 must not be used from interrupt context - it will corrupt the non-interrupt context's pte, causing unpleasing very hard to track down memory corruption. Often memory which is getting written to the user's disk. This makes users unhappy.

We could use KM_IRQ0 here. The code should disable local interrupts when holding a KM_IRQ0 kmap.

If this code were to switch to using KM_IRQ0 then we still have a problem - if any other place in the kernel does a memory allcoation or free while holding a KM_IRQ0 kmap, then this new code will corrupt the caller's pte.

So I guess we'll need to create a new kmap_atomic slot for this application. It will need interrupt protection - the page allocator can be called from interrupt context while it is already running in non-interrupt context.

Alternatively, we could just not do the kmap_atomic() at all. i386 won't be using this code and IIRC the only other highmem architecture is powerpc32, and ppc32 appears to also have its own DEBUG_PAGEALLOC implementation. So you could remove the kmap_atomic() stuff and put

#ifdef CONFIG_HIGHMEM
#error i goofed
#endif

in there.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#6 lore raw
Quoted 4 lines · Andrew Morton

Alternatively, we could just not do the kmap_atomic() at all. i386 won't be using this code and IIRC the only other highmem architecture is powerpc32, and ppc32 appears to also have its own DEBUG_PAGEALLOC implementation. So you could remove the kmap_atomic() stuff and put

ARM will also be joining the highmem club in due course, maybe during the next merge window depending on how things pan out.

The biggest issue we have is with kmaps interacting with ARM's DMA API code. Nicolas Pitre currently has a work-around for it, but I believe it can be better handled by extending the generic kmap infrastructure to be able to grab a reference on an already kmapped page. I've asked Nicolas to discuss this aspect of his patch set on lkml.

Signature · 9 lines
Russell King
 Linux kernel    2.6 ARM Linux   - http://www.arm.linux.org.uk/
 maintainer of:

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to [email protected].  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
#7 lore raw
In reply to Andrew Morton
Quoted 10 lines · Andrew Morton

Alternatively, we could just not do the kmap_atomic() at all. i386 won't be using this code and IIRC the only other highmem architecture is powerpc32, and ppc32 appears to also have its own DEBUG_PAGEALLOC implementation. So you could remove the kmap_atomic() stuff and put

#ifdef CONFIG_HIGHMEM
#error i goofed
#endif

in there.

I'll take the variant of this. Then poison_page() will be

static void poison_page(struct page *page)

{
        void *addr;

        if (PageHighmem(page))
                return; // i goofed

       page->poison = true;
       addr = page_address(page);
       memset(addr, PAGE_POISON, PAGE_SIZE);
}

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#8 lore raw
In reply to Andrew Morton
Quoted 5 lines · Andrew Morton

Alternatively, we could just not do the kmap_atomic() at all. i386 won't be using this code and IIRC the only other highmem architecture is powerpc32, and ppc32 appears to also have its own DEBUG_PAGEALLOC implementation. So you could remove the kmap_atomic() stuff and put

Actually, ppc32 DEBUG_PAGEALLOC is busted in several ways and probably unfixable (though this is still being debated).

Cheers,
Ben.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>