Re: Regression in 3.0: oops with module RO/NX in snd_printk whenloading snd-hda-intel (gold linker bug?)

From: Maarten Lankhorst
Date: Fri Jun 10 2011 - 08:57:10 EST

Op 10-06-11 14:23, Takashi Iwai schreef:
> At Fri, 10 Jun 2011 12:14:14 +0200,
> Maarten Lankhorst wrote:
>>
>> Since 3.0.0rc1 I get an oops in alc_auto_parse_customize_define, when it's calling snd_printk, I have module NX/RO enabled, and linked the kernel with gold. the only command I do is 'modprobe snd debug=8; modprobe snd-hda-intel' after init=/bin/bash
>>
>> ld --version: GNU gold (GNU Binutils for Ubuntu 2.21.0.20110327) 1.11
>>
>> [ 43.327003] HDA Intel 0000:00:1b.0: enabling device (0000 -> 0002)
>> [ 43.327012] HDA Intel 0000:00:1b.0: PCI INT A -> GSI 22 (level, low) -> IRQ 22
>> [ 43.327063] HDA Intel 0000:00:1b.0: irq 46 for MSI/MSI-X
>> [ 43.327090] HDA Intel 0000:00:1b.0: setting latency timer to 64
>> [ 43.327097] BUG: unable to handle kernel paging request at ffffffffa003c80a
>> [ 43.327253] IP: [<ffffffffa002f791>] __snd_printk+0x71/0x110 [snd]
>> [ 43.327392] PGD 1a25067 PUD 1a29063 PMD 231bce067 PTE 0
>> [ 43.327523] Oops: 0000 [#1] PREEMPT SMP
>> [ 43.327620] CPU 1
>> [ 43.327660] Modules linked in: snd_hda_intel(+) snd_hda_codec snd_hwdep snd_pcm_oss snd_pcm snd_page_alloc snd_seq_midi snd_seq_midi_event snd_rawmidi snd_seq snd_seq_device
>> snd_timer snd soundcore
>> [ 43.328104]
>> [ 43.328134] Pid: 149, comm: modprobe Not tainted 3.0.0-rc2-patser+ #34 To Be Filled By O.E.M. To Be Filled By O.E.M./P67 Pro3
>> [ 43.328393] RIP: 0010:[<ffffffffa002f791>] [<ffffffffa002f791>] __snd_printk+0x71/0x110 [snd]
>> [ 43.328581] RSP: 0018:ffff8802310a9c38 EFLAGS: 00010216
>> [ 43.328696] RAX: ffff8802310a9c78 RBX: 0000000000000002 RCX: ffffffffa00bea30
>> [ 43.328851] RDX: 0000000000000a1a RSI: ffffffffa00be00a RDI: ffffffffa00be00a
>> [ 43.329006] RBP: ffff8802310a9cb8 R08: 0000000000004401 R09: 0000000000000040
>> [ 43.329161] R10: 00000000000000d8 R11: 0000000000000002 R12: ffffffffa00bea30
>> [ 43.329316] R13: ffff88023110e800 R14: ffff88023110e800 R15: 0000000000004401
>> [ 43.329472] FS: 00007fc62e05c720(0000) GS:ffff88023f480000(0000) knlGS:0000000000000000
>> [ 43.329647] CS: 0010 DS: 0000 ES: 0000 CR0: 000000008005003b
>> [ 43.329772] CR2: ffffffffa003c80a CR3: 00000002310b6000 CR4: 00000000000406e0
>> [ 43.329927] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
>> [ 43.330082] DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
>> [ 43.330237] Process modprobe (pid: 149, threadinfo ffff8802310a8000, task ffff8802311216f0)
>> [ 43.330418] Stack:
>> [ 43.330462] ffff8802310a9c58 ffffffff81a05080 ffff880200000020 ffff8802310a9cc8
>> [ 43.330640] ffff8802310a9c78 ffffffff810bab78 ffff8802310a9c98 0000000005411735
>> [ 43.330815] ffff88023110e800 40ff880232ad3000 ffff880232ad3000 ffff880232ad3000
>> [ 43.330991] Call Trace:
>> [ 43.331044] [<ffffffff810bab78>] ? synchronize_irq+0x58/0xf0
>> [ 43.336434] [<ffffffffa00bd191>] azx_probe+0x48b/0x9fb [snd_hda_intel]
>> [ 43.341867] [<ffffffff81317e8f>] local_pci_probe+0x5f/0xd0
>> [ 43.347300] [<ffffffff813185f9>] pci_device_probe+0x119/0x120
>> [ 43.352744] [<ffffffff813b27aa>] ? driver_sysfs_add+0x7a/0xb0
>> [ 43.358193] [<ffffffff813b2b23>] driver_probe_device+0xa3/0x2a0
>> [ 43.363649] [<ffffffff813b2dcb>] __driver_attach+0xab/0xb0
>> [ 43.369038] [<ffffffff813b2d20>] ? driver_probe_device+0x2a0/0x2a0
>> [ 43.374387] [<ffffffff813b193e>] bus_for_each_dev+0x5e/0x90
>> [ 43.379725] [<ffffffff813b272e>] driver_attach+0x1e/0x20
>> [ 43.385029] [<ffffffff813b22d0>] bus_add_driver+0x160/0x2b0
>> [ 43.390172] [<ffffffffa0021000>] ? 0xffffffffa0020fff
>> [ 43.395151] [<ffffffff813b3394>] driver_register+0x74/0x130
>> [ 43.400133] [<ffffffffa0021000>] ? 0xffffffffa0020fff
>> [ 43.405010] [<ffffffff81318876>] __pci_register_driver+0x56/0xd0
>> [ 43.409827] [<ffffffff8102b653>] ? set_memory_nx+0x43/0x50
>> [ 43.414645] [<ffffffffa002101e>] alsa_card_azx_init+0x1e/0x1000 [snd_hda_intel]
>> [ 43.419550] [<ffffffff810002d5>] do_one_initcall+0x45/0x170
>> [ 43.424489] [<ffffffff8109505a>] sys_init_module+0xfa/0x250
>> [ 43.429420] [<ffffffff8157c52b>] system_call_fastpath+0x16/0x1b
>> [ 43.434403] Code: 00 00 00 75 6b 48 8b 5d f0 4c 8b 65 f8 c9 c3 48 8d 45 10 c7 45 90 20 00 00 00 48 89 45 98 48 8d 45 c0 48 89 45 a0 80 3e 2f 74 78 <8b> 05 73 d0 00 00 85 db 89 45 b0 48 c7 c6 b3 23 03 a0 48 c7 c0
>> [ 43.444891] RIP [<ffffffffa002f791>] __snd_printk+0x71/0x110 [snd]
>> [ 43.450012] RSP <ffff8802310a9c38>
>> [ 43.450013] CR2: ffffffffa003c80a
>> [ 43.450014] ---[ end trace 890ffb2c861edf22 ]---
>>
>> everything sound loaded modular, all hda codecs too.
>>
>> CONFIG_SND_VERBOSE_PRINTK=y
>> CONFIG_SND_DEBUG=y
>> CONFIG_SND_DEBUG_VERBOSE=y
>>
>> module RO/NX is set, which I assume is causing the problem.
>>
>> I don't understand the oops, but disassembly of the specific offset is below, from objdump -S:
>>
>> 7de1 is the location of the crash
>>
>> #ifdef CONFIG_SND_VERBOSE_PRINTK
>> /* strip the leading path if the given path is absolute */
>> static const char *sanity_file_name(const char *path)
>> {
>> if (*path == '/')
>> 7ddc: 80 3e 2f cmpb $0x2f,(%rsi)
>> 7ddf: 74 78 je 7e59 <__snd_printk+0xe9>
>> /* print file and line with a certain printk prefix */
>> static int print_snd_pfx(unsigned int level, const char *path, int line,
>> const char *format)
>> {
>> const char *file = sanity_file_name(path);
>> char tmp[] = "<0>";
>> 7de1: 8b 05 00 00 00 00 mov 0x0(%rip),%eax # 7de7 <__snd_printk+0x77>
>> const char *pfx = level ? KERN_DEBUG : KERN_DEFAULT;
>> 7de7: 85 db test %ebx,%ebx
>> /* print file and line with a certain printk prefix */
>> static int print_snd_pfx(unsigned int level, const char *path, int line,
>> const char *format)
>> {
>> const char *file = sanity_file_name(path);
>> char tmp[] = "<0>";
>> 7de9: 89 45 b0 mov %eax,-0x50(%rbp)
>> const char *pfx = level ? KERN_DEBUG : KERN_DEFAULT;
>> 7dec: 48 c7 c6 00 00 00 00 mov $0x0,%rsi
>> 7df3: 48 c7 c0 00 00 00 00 mov $0x0,%rax
>> 7dfa: 48 0f 44 f0 cmove %rax,%rsi
>>
>> ---
>> Seems to me like it's segfaulting when loading <0>\0 (4 bytes, so fits in an integer) to %eax. building all modules with the normal binutils works fine, and makes this crash disappear.
>>
>> Is this bug interesting enough that I'll see if I can find the commit which caused this regression?
> The snd_printk() stuff can be nowadays better presented with %pV.
> Does the patch below work for you?
>
>
> thanks,
>
> Takashi
>
> ---
> diff --git a/sound/core/misc.c b/sound/core/misc.c
> index 2c41825..095e423 100644
> --- a/sound/core/misc.c
> +++ b/sound/core/misc.c
> @@ -59,25 +59,8 @@ static const char *sanity_file_name(const char *path)
> return path;
> }
>
> -/* print file and line with a certain printk prefix */
> -static int print_snd_pfx(unsigned int level, const char *path, int line,
> - const char *format)
> -{
> - const char *file = sanity_file_name(path);
> - char tmp[] = "<0>";
> - const char *pfx = level ? KERN_DEBUG : KERN_DEFAULT;
> - int ret = 0;
> -
> - if (format[0] == '<' && format[2] == '>') {
> - tmp[1] = format[1];
> - pfx = tmp;
> - ret = 1;
> - }
> - printk("%sALSA %s:%d: ", pfx, file, line);
> - return ret;
> -}
> #else
> -#define print_snd_pfx(level, path, line, format) 0
> +#define verbose_printk(level, path, line, fmt, args) printk(fmt, args)
> #endif
#define verbose_printk makes no sense?
>
> #if defined(CONFIG_SND_DEBUG) || defined(CONFIG_SND_VERBOSE_PRINTK)
> @@ -85,15 +68,29 @@ void __snd_printk(unsigned int level, const char *path, int line,
> const char *format, ...)
> {
> va_list args;
> -
> +#ifdef CONFIG_SND_VERBOSE_PRINTK
> + struct va_format vaf;
> + char verbose_fmt[] = KERN_DEFAULT "ALSA %s:%d %pV";
> +#endif
> +
> #ifdef CONFIG_SND_DEBUG
> if (debug < level)
> return;
> #endif
> +
> va_start(args, format);
> - if (print_snd_pfx(level, path, line, format))
> - format += 3; /* skip the printk level-prefix */
> +#ifdef CONFIG_SND_VERBOSE_PRINTK
> + vaf.fmt = format;
> + vaf.va = &args;
> + if (format[0] == '<' && format[2] == '>') {
> + memcpy(verbose_fmt, format, 3);
> + vaf.fmt = format + 3;
> + } else if (level)
> + memcpy(verbose_fmt, KERN_DEBUG, 3);
> + printk(verbose_fmt, sanity_file_name(path), line, &vaf);
> +#else
> vprintk(format, args);
> +#endif
> va_end(args);
> }
> EXPORT_SYMBOL_GPL(__snd_printk);
It fixes the oops.

Tested-by: Maarten Lankhorst <m.b.lankhorst@xxxxxxxxx>

~Maarten
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/