via82xx calc_linear_pos bug
After playing about 10-20 seconds snd_pcm_status_get_avail returns too large value (1972895704, 1972895664 ..).
Tested kernels:
2.6.7-mm1 works 2.6.7-mm2 broken 2.6.8-rc1-mm1 broken
lspci -v
0000:00:11.5 Multimedia audio controller: VIA Technologies, Inc. VT8233/A/8235 AC97 Audio Controller (rev 50)
Subsystem: VIA Technologies, Inc.: Unknown device 4161
Flags: medium devsel, IRQ 22
I/O ports at e800
Capabilities: [c0] Power Management version 2
Attached patch (against 2.6.8-rc1-mm1) changes calc_linear_pos to the state it was in 2.6.7-mm1.
I don't understand the code so this might not be right solution to this problem but it seems to work.
pci/via82xx.c renamed from pci/via82xx.c.orig +1 -5
@@ -715,11 +715,7 @@ static inline unsigned int calc_linear_punsigned int size, res; size = viadev->idx_table[idx].size; -/* FIXME: is this always true? */ -if (count) - res = viadev->idx_table[idx].offset + size - count; -else - res = viadev->idx_table[idx].offset; + res = viadev->idx_table[idx].offset + size - count;/* check the validity of the calculated position */ if (size < count || (res < viadev->lastpos && (res >= viadev->bufsize2 || viadev->lastpos < viadev->bufsize2))) {
At Fri, 23 Jul 2004 18:47:30 +0300,
Quoted 3 lines · Timo Hirvonen
After playing about 10-20 seconds
snd_pcm_status_get_availreturns too large value (1972895704, 1972895664 ..).
Grrr, the empire strikes again.
The change in the recent version was a workaround to fix the similar bug on many buggy mobos.
Could you try the attached patch?
Takashi
Index: alsa-kernel/pci/via82xx.c
===================================================================
RCS file: /suse/tiwai/cvs/alsa/alsa-kernel/pci/via82xx.c,v retrieving revision 1.122
pci/via82xx.c +17 -10
@@ -707,29 +707,36 @@/* * calculate the linear position at the given sg-buffer index and the rest count */ + +#define check_invalid_pos (viadev,pos) \ + ((pos) < viadev->lastpos && ((pos) >= viadev->bufsize2 || viadev->lastpos < viadev->bufsize2)) +static inline unsigned int calc_linear_pos (viadev_t *viadev,unsigned int idx,unsigned int count) {unsigned int size, res; size = viadev->idx_table[idx].size; -/* FIXME: is this always true? */ -if (count) - res = viadev->idx_table[idx].offset + size - count; -else - res = viadev->idx_table[idx].offset; + res = viadev->idx_table[idx].offset + size - count;/* check the validity of the calculated position */ -if (size < count || (res < viadev->lastpos && (res >= viadev->bufsize2 || viadev->lastpos < viadev->bufsize2))) { +if (size < count) { +snd_printd (KERN_ERR "invalid via82xx_cur_ptr (size = %d, count = %d)\n" , (int )size, (int )count); + res = viadev->lastpos; + }else if (check_invalid_pos (viadev, res)) {#ifdef POINTER_DEBUG printk ("fail: idx = %i/%i, lastpos = 0x%x, bufsize2 = 0x%x, offsize = 0x%x, size = 0x%x, count = 0x%x\n" , idx, viadev->tbl_entries, viadev->lastpos, viadev->bufsize2, viadev->idx_table[idx].offset, viadev->idx_table[idx].size, count);#endif -/* count register returns full size when end of buffer is reached */ -if (size != count) { +if (count && size < count) {snd_printd (KERN_ERR "invalid via82xx_cur_ptr, using last valid pointer\n" ); res = viadev->lastpos; }else { - res = viadev->idx_table[idx].offset + size; -if (res < viadev->lastpos && (res >= viadev->bufsize2 || viadev->lastpos < viadev->bufsize2)) { +if (! count) +/* bogus count 0 on the DMA boundary? */ + res = viadev->idx_table[idx].offset; +else +/* count register returns full size when end of buffer is reached */ + res = viadev->idx_table[idx].offset + size; +if (check_invalid_pos (viadev, res)) {snd_printd (KERN_ERR "invalid via82xx_cur_ptr (2), using last valid pointer\n" ); res = viadev->lastpos; }
Quoted 11 lines · Takashi Iwai
At Fri, 23 Jul 2004 18:47:30 +0300,
Quoted 3 lines · Timo Hirvonen
After playing about 10-20 seconds
snd_pcm_status_get_availreturns too large value (1972895704, 1972895664 ..).Grrr, the empire strikes again.
The change in the recent version was a workaround to fix the similar bug on many buggy mobos.
Could you try the attached patch?
That patch works. snd_pcm_status_get_avail returns now sane values again. I don't see any "invalid via82xx_cur_ptr" messages so the code seems to be correct now.
thanks
-------------------------------------------------------
This SF.Net email is sponsored by BEA Weblogic Workshop FREE Java Enterprise J2EE developer tools!
Get your free copy of BEA WebLogic Workshop 8.1 today.
http://ads.osdn.com/?ad_id=4721&alloc_id=10040&op=click
At Tue, 27 Jul 2004 17:53:37 +0300,
Quoted 10 lines · Timo Hirvonen
Quoted 8 lines · Takashi Iwai
At Fri, 23 Jul 2004 18:47:30 +0300,
Quoted 3 lines · Timo Hirvonen
After playing about 10-20 seconds
snd_pcm_status_get_availreturns too large value (1972895704, 1972895664 ..).Grrr, the empire strikes again.
Quoted 7 lines
Quoted 3 lines
The change in the recent version was a workaround to fix the similar bug on many buggy mobos.
Could you try the attached patch?That patch works.
snd_pcm_status_get_availreturns now sane values again. I don't see any "invalidvia82xx_cur_ptr" messages so the code seems to be correct now.
Ok, I applied the patch to cvs.
thanks for testing.
Takashi
-------------------------------------------------------
This SF.Net email is sponsored by BEA Weblogic Workshop FREE Java Enterprise J2EE developer tools!
Get your free copy of BEA WebLogic Workshop 8.1 today.
http://ads.osdn.com/?ad_id=4721&alloc_id=10040&op=click