sb_mixer.c/save_mixer() awfully buggy?
Hi all,
with my ALS4000, I'm hitting this upon resume:
ALSA sound/isa/sb/sb_mixer.c:936: BUG? (num_regs > (sizeof(chip->saved_regs) / sizeof( (chip->saved_regs)[0]) + (sizeof(char[1 - 2 * !!(__builtin_types_compatible_p(typeof(c hip->saved_regs), typeof(&chip->saved_regs[0])))]) - 1)))
Pid: 4498, comm: my_sleep Not tainted 2.6.26-rc3 #1
[<e097ac8c>] restore_mixer+0x6c/0x70 [snd_sb_common] [<e097ad07>] snd_sbmixer_resume+0x77/0x80 [snd_sb_common] [<e098030a>] snd_als4000_resume+0x5a/0xd0 [snd_als4000] [<c0281a72>] pci_device_resume+0x22/0x60 [<c0303e76>] device_resume+0x86/0x100 [<c0145bb7>] hibernation_snapshot+0xc7/0x180 [<c0144f9f>] ? freeze_processes+0x3f/0x80 [<c0145d4c>] hibernate+0xdc/0x180 [<c0144980>] ? state_store+0x0/0xd0 [<c0144a3f>] state_store+0xbf/0xd0 [<c0144980>] ? state_store+0x0/0xd0 [<c02743d4>] kobj_attr_store+0x24/0x30 [<c01aaf5b>] sysfs_write_file+0xbb/0x110 [<c016ee56>] vfs_write+0x96/0x130 [<c01aaea0>] ? sysfs_write_file+0x0/0x110 [<c016f3cd>] sys_write+0x3d/0x70 [<c0103026>] syscall_call+0x7/0xb =======================
triggered by this line:
static void restore_mixer(struct snd_sb *chip, unsigned char *regs, int num_regs)
{
unsigned char *val = chip->saved_regs;
snd_assert(num_regs > ARRAY_SIZE(chip->saved_regs), return);
for (; num_regs; num_regs--)
snd_sbmixer_write(chip, *regs++, *val++);
}Uhmm, forgive my utterly dumb state of mind, but isn't this assert _exactly_ wrong?
We assert that the number of regs to copy is _higher_ than the array size of the content we want to copy, IOW we want to make _certain_ that we'll hit a nice array overflow or segmentation violation later. Doesn't make too much - well, exactly zero, zilch - sense to me.
Dito save_mixer().
Forgive me if this is already fixed in ALSA head (still exists in current 2.6.26-rc7).
Sounds like 2.6.26(!) material to me if it is indeed buggy.
Thanks,
Andreas Mohr
At Mon, 23 Jun 2008 00:46:35 +0200,
Quoted 38 lines · Andreas Mohr
Hi all,
with my ALS4000, I'm hitting this upon resume:
ALSA
sound/isa/sb/sb_mixer.c:936: BUG? (num_regs> (sizeof(chip->saved_regs)/sizeof( (chip->saved_regs)[0])+ (sizeof(char[1 - 2 * !!(__builtin_types_compatible_p(typeof(chip->saved_regs),typeof(&chip->saved_regs[0])))]) - 1)))
Pid: 4498, comm:my_sleepNot tainted 2.6.26-rc3 #1[<e097ac8c>] restore_mixer+0x6c/0x70 [snd_sb_common] [<e097ad07>] snd_sbmixer_resume+0x77/0x80 [snd_sb_common] [<e098030a>] snd_als4000_resume+0x5a/0xd0 [snd_als4000] [<c0281a72>] pci_device_resume+0x22/0x60 [<c0303e76>] device_resume+0x86/0x100 [<c0145bb7>] hibernation_snapshot+0xc7/0x180 [<c0144f9f>] ? freeze_processes+0x3f/0x80 [<c0145d4c>] hibernate+0xdc/0x180 [<c0144980>] ? state_store+0x0/0xd0 [<c0144a3f>] state_store+0xbf/0xd0 [<c0144980>] ? state_store+0x0/0xd0 [<c02743d4>] kobj_attr_store+0x24/0x30 [<c01aaf5b>] sysfs_write_file+0xbb/0x110 [<c016ee56>] vfs_write+0x96/0x130 [<c01aaea0>] ? sysfs_write_file+0x0/0x110 [<c016f3cd>] sys_write+0x3d/0x70 [<c0103026>] syscall_call+0x7/0xb =======================triggered by this line:
static void
restore_mixer(struct snd_sb *chip, unsigned char *regs, int num_regs){ unsigned char *val = chip->saved_regs; snd_assert(num_regs > ARRAY_SIZE(chip->saved_regs), return); for (; num_regs; num_regs--) snd_sbmixer_write(chip, *regs++, *val++); }
Quoted 8 lines
Uhmm, forgive my utterly dumb state of mind, but isn't this assert _exactly_ wrong?
We assert that the number of regs to copy is _higher_ than the array size of the content we want to copy, IOW we want to make _certain_ that we'll hit a nice array overflow or segmentation violation later. Doesn't make too much - well, exactly zero, zilch - sense to me.Dito
save_mixer().
Indeed, it's a silly bug. It just be num_regs <= ARRAY_SIZE(). The codepath wasn't tested with the debug environment. I fixed it on my tree now.
Quoted 4 lines
Forgive me if this is already fixed in ALSA head (still exists in current 2.6.26-rc7).
Sounds like 2.6.26(!) material to me if it is indeed buggy.
Yeah, will do a pull request.
thanks,
Takashi