2.6.36-rc5-mmotm0928 BUG while trying to ptrace a process

[Valdis . Kletnieks]

Dell Latitude E6500,  Intel Core2 Duo P8700 CPU.

Have seen this three times now while trying to attach gdb to a hung Firefox
process.

[20669.480674] BUG: unable to handle kernel NULL pointer dereference at 0000000000000100
[20669.480682] IP: [<ffffffff810b8275>] __reserve_bp_slot+0x56/0x17d

Here's the stack traceback:

[20669.480761] Process gdb (pid: 60320, threadinfo ffff8800a0e4c000, task ffff8800c62acc80)
[20669.480763] Stack:
[20669.480765]  ffff8801188c2eac ffffffff81a40750 0000000100000000 ffffffff81a406b0
[20669.480771] <0> ffffffff8100b469 ffff8800a0d6c400 0000000000000000 ffffffff81a406b0
[20669.480777] <0> ffffffff8100b469 0000000000000000 ffff8800a0e4dcf8 ffffffff810b83e5
[20669.480784] Call Trace:
[20669.480789]  [<ffffffff8100b469>] ? ptrace_triggered+0x0/0x39
[20669.480793]  [<ffffffff8100b469>] ? ptrace_triggered+0x0/0x39
[20669.480797]  [<ffffffff810b83e5>] reserve_bp_slot+0x22/0x3b
[20669.480800]  [<ffffffff810b6805>] ? perf_init_event+0x0/0xc9
[20669.480804]  [<ffffffff810b8487>] register_perf_hw_breakpoint+0x11/0x36
[20669.480808]  [<ffffffff810b84cb>] hw_breakpoint_event_init+0x1f/0x34
[20669.480811]  [<ffffffff8100b469>] ? ptrace_triggered+0x0/0x39
[20669.480815]  [<ffffffff810b685a>] perf_init_event+0x55/0xc9
[20669.480818]  [<ffffffff810b6805>] ? perf_init_event+0x0/0xc9
[20669.480822]  [<ffffffff810b6a74>] perf_event_alloc+0x1a6/0x3ef
[20669.480826]  [<ffffffff810b6f92>] perf_event_create_kernel_counter+0x22/0x185
[20669.480830]  [<ffffffff810b7d9b>] register_user_hw_breakpoint+0xf/0x11
[20669.480834]  [<ffffffff8100c7d3>] ptrace_set_debugreg+0xaa/0x257
[20669.480838]  [<ffffffff8156bfab>] ? sub_preempt_count+0x35/0x48
[20669.480844]  [<ffffffff81568957>] ? _raw_spin_unlock_irqrestore+0x72/0x80
[20669.480848]  [<ffffffff8100cd48>] arch_ptrace+0x126/0x214
[20669.480852]  [<ffffffff81045609>] sys_ptrace+0x7e/0x99
[20669.480857]  [<ffffffff810027eb>] system_call_fastpath+0x16/0x1b

Any thoughts? I haven't gotten any further than "the big honking commits
against perf in linux-next".  Probably bisectable with some effort - I've got
a fairly replicable Firefox hang that seems to trigger the gdb indigestion.,

Attachment: pgp00000.pgp
Description: PGP signature