Chun-Yeow-

Thank you for the suggestion. Unfortunately, there is no /sys/kernel/debug/crashlog. Perhaps, because the system is out of memory when it crashes, it is unable to write the crashlog.

What I see in the kmemleak file is that there are a large number of entries like these:

unreferenced object 0x8158c000 (size 4096):
  comm "softirq", pid 0, jiffies 10618 (age 2200.850s)
  hex dump (first 32 bytes):
    00 00 00 14 00 03 00 02 51 8b c8 5a 00 00 0e 27 ........Q..Z...'
    00 00 00 00 00 08 00 03 00 00 00 08 00 0a 00 06 ................
  backtrace:
    [<800bf9d0>] __kmalloc+0x11c/0x16c
    [<801c7b28>] __alloc_skb+0x74/0x140
    [<80d8c0c0>] _13+0x38/0xb8 [ath]
    [<80dcf848>] _113+0x104/0x1034 [ath5k]
    [<80dd127c>] _14+0xb04/0xb14 [ath5k]
unreferenced object 0x8158d000 (size 4096):
  comm "softirq", pid 0, jiffies 10623 (age 2200.800s)
  hex dump (first 32 bytes):
    00 00 00 14 00 03 00 02 51 8b c8 2f 00 00 0c 9e ........Q../....
    00 00 00 00 00 0b 00 02 6e 6c 63 74 72 6c 00 00 ........nlctrl..
  backtrace:
    [<800bf9d0>] __kmalloc+0x11c/0x16c
    [<801c7b28>] __alloc_skb+0x74/0x140
    [<80d8c0c0>] _13+0x38/0xb8 [ath]
    [<80dcf848>] _113+0x104/0x1034 [ath5k]
    [<80dd127c>] _14+0xb04/0xb14 [ath5k]
unreferenced object 0x815ec000 (size 4096):
  comm "softirq", pid 0, jiffies 10639 (age 2200.640s)
  hex dump (first 32 bytes):
    00 00 00 14 00 03 00 02 51 8b c8 57 00 00 0e 0b ........Q..W....
    00 00 00 00 00 08 00 03 00 00 00 08 00 0a 00 06 ................
  backtrace:
    [<800bf9d0>] __kmalloc+0x11c/0x16c
    [<801c7b28>] __alloc_skb+0x74/0x140
    [<80d8c0c0>] _13+0x38/0xb8 [ath]
    [<80dcf848>] _113+0x104/0x1034 [ath5k]
    [<80dd127c>] _14+0xb04/0xb14 [ath5k]

So it appears there are a large number of 4096-byte chunks that are being abandoned by the ath5k driver.

Unfortunately, I don't know how to track down these locations in the executable, and, frankly, I don't think I have enough knowledge of the kernel and drivers to actually do any good if I could.

I'm currently waiting for the node to crash again to see if there is a crashlog.

--
Bill Moffitt

_______________________________________________
Devel mailing list
[email protected]
http://lists.open80211s.org/cgi-bin/mailman/listinfo/devel

Reply via email to