Re: [PATCH] Fix Oops in crash_shrink_memory

[Pavan Naregundi]

On Tue, 2010-06-08 at 15:59 +0800, AmÃrico Wang wrote:
> On Tue, Jun 08, 2010 at 12:37:37PM +0530, Pavan Naregundi wrote:
> >Adding CC's..
> >
> >
> >On Mon, 2010-06-07 at 12:58 +0530, Pavan Naregundi wrote:
> >> Hi Everyone,
> >> 
> >> Please add me to CC in your reply..
> >> 
> >> When crashkernel is not enabled, "echo 0 > /sys/kernel/kexec_crash_size"
> >> will generate OOPS message in the kernel. Below is the OOPS message and
> >> other details,
> >> 
> >> # cat /proc/cmdline 
> >> ro  LANG=en_US.UTF-8 SYSFONT=latarcyrheb-sun16 KEYTABLE=us console=hvc0
> >> rhgb root=UUID=eafd9874-010c-46f9-a0c6-ea8db7c61ac3
> >> # uname -a
> >> Linux XXXXXXXX 2.6.35-rc1 #1 SMP Mon Jun 7 18:04:53 IST 2010 ppc64 ppc64
> >> ppc64 GNU/Linux
> >> # cd /sys/kernel/
> >> # ls
> >> debug               kexec_loaded  profiling      uevent_seqnum
> >> kexec_crash_loaded  mm            security       vmcoreinfo
> >> kexec_crash_size    notes         uevent_helper
> >> # cat kexec_crash_loaded
> >> 0
> >> # cat kexec_loaded
> >> 0
> >> # cat kexec_crash_size
> >> 1
> >> # echo 0 > kexec_crash_size
> >> Unable to handle kernel paging request for data at address 0x00000030
> >> Faulting instruction address: 0xc0000000000930b4
> >> Oops: Kernel access of bad area, sig: 11 [#1]
> >> SMP NR_CPUS=1024 NUMA pSeries
> >> last sysfs file: /sys/kernel/kexec_crash_size
> >> Modules linked in: sunrpc ipv6 ext3 jbd dm_mirror dm_region_hash dm_log
> >> dm_multipath dm_mod uinput sg ehea sr_mod ibmveth cdrom ext4 jbd2
> >> mbcache sd_mod crc_t10dif ibmvscsic scsi_transport_srp scsi_tgt [last
> >> unloaded: scsi_wait_scan]
> >> NIP: c0000000000930b4 LR: c0000000000930ac CTR: c0000000000b7ce0
> >> REGS: c0000000b7803750 TRAP: 0300   Not tainted  (2.6.35-rc1)
> >> MSR: 8000000000009032 <EE,ME,IR,DR>  CR: 28242482  XER: 20000000
> >> DAR: 0000000000000030, DSISR: 0000000040000000
> >> TASK = c0000000b9cc3dc0[1381] 'bash' THREAD: c0000000b7800000 CPU: 12
> >> GPR00: c0000000000930ac c0000000b78039d0 c000000000e885a8
> >> c000000000f42950 
> >> GPR04: c0000000b7803af0 0000000000000008 0000000000000002
> >> c0000000005c6438 
> >> GPR08: 0000000000000000 000000008000000c 0000000000000000
> >> 0000000000000000 
> >> GPR12: 0000000040242448 c000000007441e00 00000000100f6210
> >> 0000000000000000 
> >> GPR16: 00000000100f4a38 00000000100cfb98 00000000100f4bdc
> >> 00000000100f4b4c 
> >> GPR20: 00000000103f4de8 0000000000000000 0000000000000000
> >> 00000000100f0000 
> >> GPR24: c0000000005c65c0 0000000000000000 c000000000da83e0
> >> c0000000bc67f780 
> >> GPR28: c0000000bc684ae0 c000000000f42950 c000000000e1e3f8
> >> c000000000da8408 
> >> NIP [c0000000000930b4] .release_resource+0x34/0xe0
> >> LR [c0000000000930ac] .release_resource+0x2c/0xe0
> >> Call Trace:
> >> [c0000000b78039d0] [c0000000000930ac] .release_resource+0x2c/0xe0
> >> (unreliable)
> >> [c0000000b7803a60] [c0000000000d4fc8] .crash_shrink_memory+0x1c8/0x1f0
> >> [c0000000b7803b30] [c0000000000b7d38] .kexec_crash_size_store+0x58/0x90
> >> [c0000000b7803bc0] [c0000000002b0bb4] .kobj_attr_store+0x34/0x50
> >> [c0000000b7803c30] [c000000000226d5c] .sysfs_write_file+0xec/0x1f0
> >> [c0000000b7803ce0] [c00000000019e0bc] .vfs_write+0xec/0x1f0
> >> [c0000000b7803d80] [c00000000019e2e8] .SyS_write+0x58/0xb0
> >> [c0000000b7803e30] [c00000000000852c] syscall_exit+0x0/0x40
> >> Instruction dump:
> >> fba1ffe8 fbc1fff0 fbe1fff8 ebc2b228 7c7f1b78 f8010010 f821ff71 ebbe8000 
> >> 7fa3eb78 484e35d9 60000000 e97f0020 <e92b0030> 2fa90000 419e002c
> >> 7fbf4800 
> >> ---[ end trace afbc780462c9bf4e ]---
> >> 
> >> When crashkernel is not enabled, crashk_res resource have not been
> >> reserved. Hence crashk_res.parent will be NULL.
> >> 
> >> Attaching a simple patch to this problem. Patch is tested and resolves this bug.
> 
> 
> Ouch...
> 
> The patch indeed addresses the problem, looks good to me.
> Please add your Signed-off-by and my:
> 
> Reviewed-by: WANG Cong <xiyou.wangcong@xxxxxxxxx>
> 
> Another problem is that you should get 0 instead of 1 when you don't
> reserve any memory.

We get 1 here because, crash_get_memory_size() adds 1 as below, 

size = crashk_res.end - crashk_res.start + 1;

We cant remove this addition, as it is required to display correct size
in case if we reserve the crash memory.

Coming back to issue.. Attaching the patch again.

Signed-off-by: Pavan Naregundi <pavan@xxxxxxxxxxxxxxxxxx>
Reviewed-by: WANG Cong <xiyou.wangcong@xxxxxxxxx>
---
diff -Naur a/kernel/kexec.c b/kernel/kexec.c
--- a/kernel/kexec.c	2010-06-07 18:45:55.050000000 +0530
+++ b/kernel/kexec.c	2010-06-07 18:50:28.070000004 +0530
@@ -1134,7 +1134,7 @@
 
 	free_reserved_phys_range(end, crashk_res.end);
 
-	if (start == end)
+	if ((start == end) && (crashk_res.parent != NULL))
 		release_resource(&crashk_res);
 	crashk_res.end = end - 1;