Re: ACPI: kmemcheck: Caught 16-bit read from freed memory (f7c12ec6)

[Pekka Enberg <penberg@cs.helsinki.fi>]

Vegard Nossum wrote:
> On Tue, May 6, 2008 at 10:38 PM, Pekka Enberg <penberg@cs.helsinki.fi> wrote:
>> On Tue, May 6, 2008 at 7:09 PM, Vegard Nossum <vegard.nossum@gmail.com> wrote:
>>  >  Running kmemcheck on top of v2.6.26-rc1 gives the following (never
>>  >  before seen) warning:
>>  >
>>  >  Linux Plug and Play Support v0.97 (c) Adam Belay
>>  >  pnp: PnP ACPI init
>>  >  ACPI: bus type pnp registered
>>  >  kmemcheck: Caught 16-bit read from freed memory (f7c12ec6)
>>  >
>>  >  Pid: 1, comm: swapper Not tainted (2.6.26-rc1-00010-g7966e04 #2)
>>  >  EIP: 0060:[<c027ecaa>] EFLAGS: 00010286 CPU: 0
>>  >  EIP is at acpi_ps_get_next_arg+0x1b8/0x262
>>  >  EAX: f7c12ec0 EBX: f7d20428 ECX: f7ca1b00 EDX: 00000001
>>  >  ESI: 00000049 EDI: f7d20400 EBP: f7c61e38 ESP: c06c9dc8
>>  >   DS: 007b ES: 007b FS: 00d8 GS: 0000 SS: 0068
>>  >  CR0: 8005003b CR2: f7c46456 CR3: 006ba000 CR4: 000006c0
>>  >  DR0: 00000000 DR1: 00000000 DR2: 00000000 DR3: 00000000
>>  >  DR6: ffff4ff0 DR7: 00000400
>>  >   [<c0119192>] kmemcheck_read+0xe2/0x140
>>  >   [<c0119326>] kmemcheck_access+0x136/0x1a0
>>  >   [<c04bd286>] do_page_fault+0x5e6/0x690
>>  >   [<c04bb2da>] error_code+0x72/0x78
>>  >   [<c027f991>] acpi_ps_parse_loop+0x4dd/0x7f8
>>  >   [<c027ee36>] acpi_ps_parse_aml+0xb4/0x332
>>  >   [<c02802c0>] acpi_ps_execute_method+0x13d/0x20d
>>  >   [<c027c8a2>] acpi_ns_evaluate+0x10e/0x1b0
>>  >   [<c0283210>] acpi_ut_evaluate_object+0x51/0x18d
>>  >   [<c0283406>] acpi_ut_execute_STA+0x22/0x7b
>>  >   [<c027c18b>] acpi_ns_get_device_callback+0x5a/0x121
>>  >   [<c027e410>] acpi_ns_walk_namespace+0xf0/0x10c
>>  >   [<c027c0b5>] acpi_get_devices+0x47/0x5d
>>  >   [<c068db45>] pnpacpi_init+0x65/0xa0
>>  >   [<c06735c7>] kernel_init+0x127/0x290
>>  >   [<c0104cc7>] kernel_thread_helper+0x7/0x10
>>  >   [<ffffffff>] 0xffffffff
>>  >  pnp: PnP ACPI: found 17 devices
>>  >  ACPI: ACPI bus type pnp unregistered
>>  >
>>  >  This faulting instruction comes from
>>  >
>>  >  $ addr2line -e vmlinux -i c027ecaa
>>  >  drivers/acpi/parser/psargs.c:694
>>
>>  (That's some seriously hairy code in ACPI btw.)
>>
>>  Vegard, can you do disassembly for the faulting instruction? I *think*
>>  it's the "walk_state->op" bit that is hanging to an object that was
>>  already deleted in the strange loop in acpi_ps_parse_loop() but it
>>  would be good to have some more data on this.
> 
> Of course. This is in fact another image, but the EIP (and indeed
> EAX...EDX) are exactly the same. I hope this doesn't get mangled too
> much by gmail. It's a lot, though :-)

[snip]

> c027ec9b:	89 45 e4             	mov    %eax,-0x1c(%ebp)
> c027ec9e:	0f 84 9e 00 00 00    	je     c027ed42 <acpi_ps_get_next_arg+0x250>
> c027eca4:	8b 87 d4 01 00 00    	mov    0x1d4(%edi),%eax
> c027ecaa:	66 81 78 06 2a 5b    	cmpw   $0x5b2a,0x6(%eax)
> 
>        ^--- EIP here
> 
> Well, at least kmemcheck correctly identifies it as a 16-bit read...

Aah, I didn't notice it was a 16-bit read. But yeah at offset 6 we have 
16-bit aml_opcode:

#define ACPI_PARSE_COMMON \
         union acpi_parse_object         *parent; \
         u8                              descriptor_type; \
         u8                              flags; \
         u16                             aml_opcode; \

So walk_state->op probably points to an object that was already free'd. Len?

			Pekka