Re: kernel: tracepoints: add support for relative references

[Ard Biesheuvel]

On 5 September 2017 at 13:48, Jessica Yu <jeyu@xxxxxxxxxx> wrote:
> +++ Ard Biesheuvel [19/08/17 19:10 +0100]:
>>
>> To avoid the need for relocating absolute references to tracepoint
>> structures at boot time when running relocatable kernels (which may
>> take a disproportionate amount of space), add the option to emit
>> these tables as relative references instead.
>>
>> Cc: Steven Rostedt <rostedt@xxxxxxxxxxx>
>> Cc: Ingo Molnar <mingo@xxxxxxxxxx>
>> Signed-off-by: Ard Biesheuvel <ard.biesheuvel@xxxxxxxxxx>
>
>
> Hi Ard,
>
> While testing this patchset I ran into the following general protection
> fault when removing a module with tracepoints:
>
> # rmmod kvm
> [   25.279049] general protection fault: 0000 [#1] SMP
> [   25.279385] Modules linked in: kvm(-) irqbypass
> [   25.279829] CPU: 0 PID: 125 Comm: rmmod Not tainted 4.13.0-rc2+ #1
> [   25.280007] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS
> 1.10.2-1.fc26 04/01/2014
> [   25.280315] task: ffff9697db7c0000 task.stack: ffffa907802dc000
> [   25.280624] RIP: 0010:tracepoint_module_notify+0xd5/0x1d0
> [   25.280782] RSP: 0018:ffffa907802dfe68 EFLAGS: 00000287
> [   25.280963] RAX: ffffffffc02b6fb0 RBX: ffff9697ddb33440 RCX:
> ffffffffc02b70f0
> [   25.281168] RDX: 000163cc00016410 RSI: ffff9697df21d860 RDI:
> ffff9697de001900
> [   25.281371] RBP: ffffa907802dfe88 R08: 000000000001d860 R09:
> ffffffff941692b4
> [   25.281543] R10: fffff0ec4076ccc0 R11: 0000000000000300 R12:
> ffffffffc02ce740
> [   25.281716] R13: 0000000000000002 R14: ffffffffc02ce740 R15:
> 0000000000000000
> [   25.281917] FS:  0000000000a6e880(0000) GS:ffff9697df200000(0000)
> knlGS:0000000000000000
> [   25.282107] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> [   25.282254] CR2: 00000000004adfc0 CR3: 000000001f838000 CR4:
> 00000000000006f0
> [   25.282517] Call Trace:
> [   25.283019]  notifier_call_chain+0x4a/0x70
> [   25.283287]  blocking_notifier_call_chain+0x43/0x60
> [   25.283422]  SyS_delete_module+0x1b8/0x240
> [   25.283542]  ? SyS_chdir+0xc4/0xe0
> [   25.283648]  entry_SYSCALL_64_fastpath+0x1a/0xa5
> [   25.283863] RIP: 0033:0x4ba267
> [   25.283947] RSP: 002b:00007ffc3316bdf8 EFLAGS: 00000246 ORIG_RAX:
> 00000000000000b0
> [   25.284134] RAX: ffffffffffffffda RBX: 0000000000000000 RCX:
> 00000000004ba267
> [   25.284310] RDX: 0000000000000000 RSI: 0000000000000880 RDI:
> 00007ffc3316be00
> [   25.284511] RBP: 00007ffc3316be40 R08: 0000000000000003 R09:
> 0000000000000000
> [   25.284699] R10: 0000000000000000 R11: 0000000000000246 R12:
> 0000000000637672
> [   25.284916] R13: 00007ffc3316c1b0 R14: 0000000000000000 R15:
> 0000000000000000
> [   25.285226] Code: 00 48 89 43 08 e8 9c b4 0d 00 49 8b 84 24 90 02 00 00
> 48 85 c0 74 28 41 8b 94 24 8c 02 00 00 48 8d 0c d0 48 39 c8 73 17 48 8b 10
> <48> 83 7a 28 00 0f 85 e8 00 00 00 48 83 c0 08 48 39 c1 77 e9 48 [
> 25.286199] RIP: tracepoint_module_notify+0xd5/0x1d0 RSP: ffffa907802dfe68
> [   25.286708] ---[ end trace 04f35cade1663177 ]---
>
> The offending line is in tp_module_going_check_quiescent() (shown inlined):
>
> (gdb) list *(tracepoint_module_notify+0xd5)
> 0xffffffff811692d5 is in tracepoint_module_notify
> (/home/jeyu/work/jeyu-linux/kernel/tracepoint.c:402).
> 397             struct tracepoint * const *iter;
> 398
> 399             if (!begin)
> 400                     return;
> 401             for (iter = begin; iter < end; iter++)
> 402                     WARN_ON_ONCE((*iter)->funcs);
> 403     }
> 404
> 405     static int tracepoint_module_coming(struct module *mod)
> 406     {
>
> The relative reference calculations need to be done here too.
>

OK, thanks for spotting that.

I will fix and resend after the merge window closes.