Skip to contentSkip to conversation

USB Audio problems

DISCUSSION Karim Yaghmour 4 participants 11 messages Nov 19, 2003 – Nov 21, 2003

lore
Share & tools
lore
#0 lore raw

I've been trying to fix some problems I'm getting with the USB audio driver.

Basically, I get these types of messages over and over accompanied by some audio glitches:
ALSA usbaudio.c:688: cannot submit datapipe for urb 0, err = -6 usb-uhci.c: ENXIO 00038200, flags 2, urb cf74f5c0, burb cf74f5c0

I've been able to reproduce this problem on quite a few different systems, including different architectures. The problems seem to occur much more often on slower machines than on faster machines. Basically, I've been able to identify that this is due to the Linux USB layer checking whether the urb->hcpriv field being NULL in the *_submit_urb() code and returning -EINVAL if it's not. In other words, the URBs used by the USB audio driver are reused prior to being cosumed.

I've instrumented the usbaudio driver using the Linux Trace Toolkit to get some info about the behavior of start_urbs() and snd_complete_urb(). I've also added a trace point at the same place where the printk for "cannot submit datapipe for ubr" is called. Here's what I get (note where "Abusing URB" is happening):

Event creation         4,867,640,107     N/A     391     NEW EVENT TYPE : Start URB; CUSTOM EVENT ID : 24
Event creation         4,867,640,138     N/A     391     NEW EVENT TYPE : Complete URB; CUSTOM EVENT ID : 25
Event creation         4,867,640,260     N/A     391     NEW EVENT TYPE : Abuse URB; CUSTOM EVENT ID : 26
Start URB              4,869,110,362     90     33     Sending URB 0
Start URB              4,869,110,606     90     33     Sending URB 1
Start URB              4,869,110,728     90     33     Sending URB 2
Start URB              4,869,110,850     90     33     Sending URB 3
Start URB              4,869,110,972     90     33     Sending URB 4
Complete URB           4,869,124,063     0     36     Completing URB 0
Complete URB           4,869,128,000     0     36     Completing URB 1
Complete URB           4,869,132,058     0     36     Completing URB 2
Complete URB           4,869,135,995     0     36     Completing URB 3
Complete URB           4,869,140,023     0     36     Completing URB 4
Complete URB           4,869,144,051     0     36     Completing URB 0
Complete URB           4,869,147,987     0     36     Completing URB 1
Complete URB           4,869,152,015     0     36     Completing URB 2
Complete URB           4,869,156,043     0     36     Completing URB 3
Complete URB           4,869,160,010     0     36     Completing URB 4
Complete URB           4,869,164,038     0     36     Completing URB 0
Complete URB           4,869,168,066     0     36     Completing URB 1
Complete URB           4,869,172,003     0     36     Completing URB 2
Complete URB           4,869,176,031     0     36     Completing URB 3
Complete URB           4,869,180,059     0     36     Completing URB 4
Complete URB           4,869,183,995     0     36     Completing URB 0
Complete URB           4,869,188,054     0     36     Completing URB 1
Complete URB           4,869,192,021     0     36     Completing URB 2
Complete URB           4,869,196,018     0     36     Completing URB 3
Complete URB           4,869,200,046     0     36     Completing URB 4
Complete URB           4,869,204,013     0     36     Completing URB 0
Complete URB           4,869,208,011     0     36     Completing URB 1
Complete URB           4,869,212,039     0     36     Completing URB 2
Complete URB           4,869,216,006     0     36     Completing URB 3
Complete URB           4,869,220,003     0     36     Completing URB 4
Complete URB           4,869,224,001     0     36     Completing URB 0
Complete URB           4,869,227,998     0     36     Completing URB 1
Complete URB           4,869,231,996     0     36     Completing URB 2
Complete URB           4,869,235,993     0     36     Completing URB 3
Complete URB           4,869,240,021     0     36     Completing URB 4
Complete URB           4,869,244,019     0     36     Completing URB 0
....
Complete URB           4,870,112,025     0     36     Completing URB 3
Complete URB           4,870,116,022     0     36     Completing URB 4
Complete URB           4,870,120,020     0     36     Completing URB 0
Complete URB           4,870,124,017     0     36     Completing URB 1
Complete URB           4,870,128,015     0     36     Completing URB 2
Complete URB           4,870,132,043     0     36     Completing URB 3
Complete URB           4,870,136,040     0     36     Completing URB 4
Complete URB           4,870,140,038     0     36     Completing URB 0
Complete URB           4,870,144,035     0     36     Completing URB 1
Start URB              4,870,145,378     90     33     Sending URB 0
Start URB              4,870,145,531     90     33     Sending URB 1
Start URB              4,870,145,653     90     33     Sending URB 2
Start URB              4,870,145,744     90     33     Sending URB 3
Start URB              4,870,145,866     90     33     Sending URB 4
Complete URB           4,870,148,063     0     36     Completing URB 2
Complete URB           4,870,152,061     0     36     Completing URB 3
Complete URB           4,870,156,058     0     36     Completing URB 4
Complete URB           4,870,159,049     0     36     Completing URB 0
Complete URB           4,870,160,056     0     36     Completing URB 0
Complete URB           4,870,163,046     0     36     Completing URB 1
Complete URB           4,870,164,053     0     36     Completing URB 1
Complete URB           4,870,167,044     0     36     Completing URB 2
Complete URB           4,870,168,020     0     36     Completing URB 2
Complete URB           4,870,171,041     0     36     Completing URB 3
...
Complete URB           4,871,171,087     0     36     Completing URB 3
Complete URB           4,871,172,063     0     36     Completing URB 0
Complete URB           4,871,175,054     0     36     Completing URB 4
Complete URB           4,871,176,061     0     36     Completing URB 1
Complete URB           4,871,179,051     0     36     Completing URB 0
Complete URB           4,871,180,089     0     36     Completing URB 2
Complete URB           4,871,180,364     0     36     Completing URB 0
Complete URB           4,871,181,127     0     36     Completing URB 1
Complete URB           4,871,182,103     0     36     Completing URB 3
Complete URB           4,871,183,079     0     36     Completing URB 1
Complete URB           4,871,183,202     0     36     Completing URB 4
Start URB              4,871,184,300     90     33     Sending URB 0
Start URB              4,871,184,483     90     33     Sending URB 1
Start URB              4,871,184,605     90     33     Sending URB 2
Start URB              4,871,184,697     90     33     Sending URB 3
Start URB              4,871,184,819     90     33     Sending URB 4
Start URB              4,871,186,833     90     33     Sending URB 0
Start URB              4,871,186,985     90     33     Sending URB 1
Start URB              4,871,187,138     90     33     Sending URB 2
Abuse URB              4,871,187,169     90     33     Abusing URB 2
Complete URB           4,871,204,410     90     36     Completing URB 2
Complete URB           4,871,205,173     0     36     Completing URB 3
Complete URB           4,871,205,295     0     36     Completing URB 4
Complete URB           4,871,205,447     0     36     Completing URB 0
Complete URB           4,871,205,722     0     36     Completing URB 0
Complete URB           4,871,205,935     0     36     Completing URB 1
Complete URB           4,871,206,180     0     36     Completing URB 1
Complete URB           4,871,207,095     90     36     Completing URB 2
Start URB              4,871,207,736     90     33     Sending URB 0
Start URB              4,871,207,858     90     33     Sending URB 1
Start URB              4,871,207,949     90     33     Sending URB 2
Start URB              4,871,208,071     90     33     Sending URB 3
Start URB              4,871,208,194     90     33     Sending URB 4
Complete URB           4,871,210,116     0     36     Completing URB 3
Complete URB           4,871,214,083     0     36     Completing URB 4
Complete URB           4,871,218,111     0     36     Completing URB 0
Complete URB           4,871,221,071     0     36     Completing URB 0
Complete URB           4,871,222,078     0     36     Completing URB 1
Complete URB           4,871,225,068     0     36     Completing URB 1
Complete URB           4,871,226,075     0     36     Completing URB 2
Complete URB           4,871,229,066     0     36     Completing URB 2
Complete URB           4,871,230,073     0     36     Completing URB 3
Complete URB           4,871,233,063     0     36     Completing URB 3
Complete URB           4,871,234,070     0     36     Completing URB 4
Complete URB           4,871,237,061     0     36     Completing URB 4
Complete URB           4,871,238,068     0     36     Completing URB 0
Complete URB           4,871,241,058     0     36     Completing URB 0
...
Complete URB           4,871,716,089     0     36     Completing URB 0
Complete URB           4,871,717,096     0     36     Completing URB 4
Complete URB           4,871,720,087     0     36     Completing URB 1
Complete URB           4,871,721,094     0     36     Completing URB 0
Complete URB           4,871,724,084     0     36     Completing URB 2
Complete URB           4,871,725,091     0     36     Completing URB 1
Complete URB           4,871,728,082     0     36     Completing URB 3
Complete URB           4,871,729,089     0     36     Completing URB 2
Complete URB           4,871,730,187     0     36     Completing URB 2
Start URB              4,871,731,377     90     33     Sending URB 0
Abuse URB              4,871,731,408     90     33     Abusing URB 0
Complete URB           4,871,748,741     90     36     Completing URB 4
Start URB              4,871,749,015     90     33     Sending URB 0
Abuse URB              4,871,749,046     90     33     Abusing URB 0
Complete URB           4,871,766,439     90     36     Completing URB 3
Complete URB           4,871,766,561     90     36     Completing URB 0
Complete URB           4,871,766,684     90     36     Completing URB 4
Complete URB           4,871,766,806     90     36     Completing URB 1
Complete URB           4,871,766,928     90     36     Completing URB 0
Complete URB           4,871,767,019     90     36     Completing URB 1
Complete URB           4,871,767,141     90     36     Completing URB 3
Start URB              4,871,767,294     90     33     Sending URB 0
Start URB              4,871,767,538     90     33     Sending URB 1
Start URB              4,871,767,629     90     33     Sending URB 2
Start URB              4,871,767,752     90     33     Sending URB 3
Start URB              4,871,767,843     90     33     Sending URB 4
Start URB              4,871,769,766     90     33     Sending URB 0
Start URB              4,871,769,918     90     33     Sending URB 1
Start URB              4,871,770,040     90     33     Sending URB 2
Start URB              4,871,770,162     90     33     Sending URB 3
Start URB              4,871,770,254     90     33     Sending URB 4
Complete URB           4,871,772,176     0     36     Completing URB 1
Complete URB           4,871,773,122     0     36     Completing URB 2
Complete URB           4,871,774,099     0     36     Completing URB 3
Complete URB           4,871,775,106     0     36     Completing URB 4
Start URB              4,871,775,899     90     33     Sending URB 0
Abuse URB              4,871,775,960     90     33     Abusing URB 0
Start URB              4,871,793,323     90     33     Sending URB 0
Abuse URB              4,871,793,384     90     33     Abusing URB 0
Complete URB           4,871,810,800     90     36     Completing URB 0
Complete URB           4,871,811,257     90     36     Completing URB 0
Complete URB           4,871,811,379     90     36     Completing URB 1
Complete URB           4,871,811,654     90     36     Completing URB 2
Complete URB           4,871,811,898     90     36     Completing URB 3
Complete URB           4,871,813,128     90     36     Completing URB 4
Start URB              4,871,819,109     90     33     Sending URB 0
Start URB              4,871,819,292     90     33     Sending URB 1
Start URB              4,871,819,414     90     33     Sending URB 2
Start URB              4,871,819,505     90     33     Sending URB 3
Start URB              4,871,819,627     90     33     Sending URB 4
Complete URB           4,871,824,113     0     36     Completing URB 0
Complete URB           4,871,828,080     0     36     Completing URB 1
Complete URB           4,871,832,078     0     36     Completing URB 2
Complete URB           4,871,833,085     0     36     Completing URB 0
Complete URB           4,871,836,075     0     36     Completing URB 3
Complete URB           4,871,837,082     0     36     Completing URB 1
Complete URB           4,871,840,073     0     36     Completing URB 4
Complete URB           4,871,841,080     0     36     Completing URB 2
Complete URB           4,871,844,070     0     36     Completing URB 0

...

[All the parts with "..." are events where there's only "Complete URB" which happens.]

I'm using ALSA 0.9.6, but the problems do occur on later versiosns still. This particular trace was collected on an ARM board.

Does anyone have any hints as to what I should be looking at.

Karim

Signature · 11 lines
Author, Speaker, Developer, Consultant
Pushing Embedded and Real-Time Linux Systems Beyond the Limits
http://www.opersys.com || [email protected] || 514-812-4145



-------------------------------------------------------
This SF.net email is sponsored by: SF.net Giveback Program.
Does SourceForge.net help you be more productive?  Does it
help you create better code?  SHARE THE LOVE, and help us help
YOU!  Click Here: http://sourceforge.net/donate/
#1 lore raw

Well, it seems that I'm going to have to answer my own self ... :)

The following is what I've been able to find using additional tracing info. Also there's a fix for usbaudio.c.

Basically, it's as I said before: the usbaudio driver uses URBs without even checking if they're in use or not. Surprisingly, it does this with one channel, but not the other. Here's what I get in the traces when I separated the events by channel:
Channel A (the hex data is the value of *subs):
-----------------------------------------------------------------------------------
...

Complete URB           11,242,322,467     0     48     Completing URB 3: 0xC1F6040C
Complete URB           11,242,326,465     0     48     Completing URB 4: 0xC1F6040C
Complete URB           11,242,330,493     0     48     Completing URB 0: 0xC1F6040C
Complete URB           11,242,334,460     0     48     Completing URB 1: 0xC1F6040C
Complete URB           11,242,338,640     126     48     Completing URB 2: 0xC1F6040C
Start URB              11,242,338,945     126     45     Sending URB 0: 0xC1F6040C
Start URB              11,242,339,159     126     45     Sending URB 1: 0xC1F6040C
Start URB              11,242,339,342     126     45     Sending URB 2: 0xC1F6040C
Start URB              11,242,339,495     126     45     Sending URB 3: 0xC1F6040C
Abuse URB              11,242,339,617     126     45     Abusing URB 3: 0xC1F6040C
Deactivate URB         11,242,356,553     126     50     Deactivating URB 0: 0xC1F6040C
Complete URB           11,242,357,621     0     48     Completing URB 3: 0xC1F6040C
Complete URB           11,242,357,834     0     48     Completing URB 4: 0xC1F6040C
Complete URB           11,242,358,323     0     48     Completing URB 0: 0xC1F6040C
Complete URB           11,242,358,933     0     48     Completing URB 1: 0xC1F6040C
URB Active Mask        11,242,359,787     126     57     Active Mask NOT set for 1: 0xC1F6040C
Deactivate URB         11,242,359,970     126     50     Deactivating URB 2: 0xC1F6040C
Complete URB           11,242,360,611     0     48     Completing URB 2: 0xC1F6040C
URB Active Mask        11,242,361,008     126     57     Active Mask NOT set for 3: 0xC1F6040C
URB Active Mask        11,242,361,100     126     57     Active Mask NOT set for 4: 0xC1F6040C
Start URB              11,242,362,442     126     45     Sending URB 0: 0xC1F6040C
Start URB              11,242,362,656     126     45     Sending URB 1: 0xC1F6040C
Start URB              11,242,362,839     126     45     Sending URB 2: 0xC1F6040C
Start URB              11,242,362,991     126     45     Sending URB 3: 0xC1F6040C
Start URB              11,242,363,144     126     45     Sending URB 4: 0xC1F6040C
Complete URB           11,242,376,479     0     48     Completing URB 0: 0xC1F6040C
Complete URB           11,242,380,477     0     48     Completing URB 1: 0xC1F6040C
Complete URB           11,242,384,474     0     48     Completing URB 2: 0xC1F6040C
Complete URB           11,242,388,472     0     48     Completing URB 3: 0xC1F6040C
Complete URB           11,242,392,469     0     48     Completing URB 4: 0xC1F6040C

...
-----------------------------------------------------------------------------------

In this first channel, the deactivation function was never called at any point prior to sending being initiated. Hence, a URB already in use was reused without proper deactivation. Looking at the timestamps above, it is clear that when sending is initiated on URB 3, this one is in its 4ms playing period and would be interupted.

Channel B:
-----------------------------------------------------------------------------------
...
In the second channel, I had this:

Complete URB           11,242,319,446     0     48     Completing URB 4: 0xC1F6058C
Complete URB           11,242,323,474     0     48     Completing URB 0: 0xC1F6058C
Complete URB           11,242,327,472     0     48     Completing URB 1: 0xC1F6058C
Deactivate URB         11,242,330,767     126     50     Deactivating URB 0: 0xC1F6058C
Complete URB           11,242,331,500     0     48     Completing URB 2: 0xC1F6058C
Complete URB           11,242,331,713     0     48     Completing URB 0: 0xC1F6058C
Deactivate URB         11,242,331,896     126     50     Deactivating URB 1: 0xC1F6058C
Complete URB           11,242,332,537     0     48     Completing URB 1: 0xC1F6058C
URB Active Mask        11,242,332,720     126     57     Active Mask NOT set for 2: 0xC1F6058C
Deactivate URB         11,242,332,812     126     50     Deactivating URB 3: 0xC1F6058C
Complete URB           11,242,333,514     0     48     Completing URB 3: 0xC1F6058C
Deactivate URB         11,242,333,697     126     50     Deactivating URB 4: 0xC1F6058C
Complete URB           11,242,334,643     0     48     Completing URB 4: 0xC1F6058C
Start URB              11,242,335,833     126     45     Sending URB 0: 0xC1F6058C
Start URB              11,242,336,047     126     45     Sending URB 1: 0xC1F6058C
Start URB              11,242,336,199     126     45     Sending URB 2: 0xC1F6058C
Start URB              11,242,336,382     126     45     Sending URB 3: 0xC1F6058C
Start URB              11,242,336,535     126     45     Sending URB 4: 0xC1F6058C
Complete URB           11,242,357,987     0     48     Completing URB 0: 0xC1F6058C
Complete URB           11,242,358,628     0     48     Completing URB 1: 0xC1F6058C
Complete URB           11,242,359,168     0     48     Completing URB 2: 0xC1F6058C
Complete URB           11,242,361,557     126     48     Completing URB 3: 0xC1F6058C
Complete URB           11,242,365,555     0     48     Completing URB 4: 0xC1F6058C
Complete URB           11,242,369,522     0     48     Completing URB 0: 0xC1F6058C

...
-----------------------------------------------------------------------------------

In no occurence did the second channel reuse a URB prior to deactivating it. Hence, there are no errors that ever occur on this channel.

To make a long story short, the most important issue here is to check whether the URB is active before using it. If it is, then it must be unlinked first.

Here's my modified start_urbs() function from usbaudio.c:
static int start_urbs(snd_usb_substream_t *subs, snd_pcm_runtime_t *runtime)

{
    unsigned int i;
    int err;

    for (i = 0; i < subs->nurbs; i++) {
        snd_assert(subs->dataurb[i].urb, return -EINVAL);
        /* Deactivate URB prior to using it. K.Y. */
        if (test_and_clear_bit(i, &subs->active_mask)) {
            set_bit(i, &subs->unlink_mask);
            usb_unlink_urb(subs->dataurb[i].urb);
        }
        if (subs->ops.prepare(subs, runtime, subs->dataurb[i].urb) < 0) {
            snd_printk(KERN_ERR "cannot prepare datapipe for urb %d\n", i);
            goto __error;
        }
    }
    if (subs->syncpipe) {
        for (i = 0; i < SYNC_URBS; i++) {
            snd_assert(subs->syncurb[i].urb, return -EINVAL);
            if (subs->ops.prepare_sync(subs, runtime, subs->syncurb[i].urb) < 0) {
                snd_printk(KERN_ERR "cannot prepare syncpipe for urb %d\n", i);
                goto __error;
            }
        }
    }

    subs->running = 1;
    for (i = 0; i < subs->nurbs; i++) {
        trace_std_formatted_event(event_start_urb, i,  (unsigned int) subs);
        if ((err = usb_submit_urb(subs->dataurb[i].urb, GFP_KERNEL)) < 0) {
            trace_std_formatted_event(event_abuse_urb, i,  (unsigned int) subs);
            snd_printk(KERN_ERR "cannot submit datapipe for urb %d, err = %d\n", i, err);
            goto __error;
        }
        set_bit(i, &subs->active_mask);
        clear_bit(i, &subs->unlink_mask);
    }
    if (subs->syncpipe) {
        for (i = 0; i < SYNC_URBS; i++) {
            if ((err = usb_submit_urb(subs->syncurb[i].urb, GFP_KERNEL)) < 0) {
                snd_printk(KERN_ERR "cannot submit syncpipe for urb %d, err = %d\n", i, err);
                goto __error;
            }
            set_bit(i + 16, &subs->active_mask);
        }
    }
    return 0;

__error:

    // snd_pcm_stop(subs->pcm_substream, SNDRV_PCM_STATE_XRUN);
    deactivate_urbs(subs, 0);
    return -EPIPE;
}

Using this code, I haven't had any more errors at all.

In order to use this, just remove all the trace_* functions. This new function just checks if a URB is already in use prior to using it and clears it. Also, it clears the unlink_mask after the call to usb_submit_urb in order for it to be "clearable" by deactivate_urbs(). Without this last clear_bit(), URBs can only be cleared once by deactivate_urbs() and it takes a new init_substream_urbs() to set the entire mask back to 0. I don't see how this makes any sense.

Feedback is welcome.

Karim

Signature · 12 lines
Author, Speaker, Developer, Consultant
Pushing Embedded and Real-Time Linux Systems Beyond the Limits
http://www.opersys.com || [email protected] || 514-812-4145




-------------------------------------------------------
This SF.net email is sponsored by: SF.net Giveback Program.
Does SourceForge.net help you be more productive?  Does it
help you create better code?  SHARE THE LOVE, and help us help
YOU!  Click Here: http://sourceforge.net/donate/
#2 lore raw
Quoted 2 lines · Karim Yaghmour

Well, it seems that I'm going to have to answer my own self ... :)

Yes, usb-audio seems to be a bit forgotten... (Takashi, those mplayer-plughw-segfaults still persist, even on x86 and even with kernel 2.4 (current cvs of drivers/lib, of course)

Quoted 4 lines

The following is what I've been able to find using additional tracing info. Also there's a fix for usbaudio.c.

hmmm, the submit_urb-error is gone, but random lockups and usb-device-disconnect until reboot have come... (bitkeeper-2.6 from just one hour ago...) with your function.

Have fun*

Niklas

-------------------------------------------------------
This SF.net email is sponsored by: SF.net Giveback Program. Does SourceForge.net help you be more productive? Does it help you create better code? SHARE THE LOVE, and help us help YOU! Click Here: http://sourceforge.net/donate/

#3 lore raw

At Thu, 20 Nov 2003 11:04:52 +0100,

Quoted 9 lines · Niklas Werner
Quoted 2 lines · Karim Yaghmour

Well, it seems that I'm going to have to answer my own self ... :)

Yes, usb-audio seems to be a bit forgotten... (Takashi, those mplayer-plughw-segfaults still persist, even on x86 and even with kernel 2.4 (current cvs of drivers/lib, of course)

not forgotten, but it's difficult to debug.
the problem is the behavior of submit_urb() and unlink_urb() is quite dependent on the hub module and kernel version.

Quoted 7 lines
Quoted 4 lines

The following is what I've been able to find using additional tracing info. Also there's a fix for usbaudio.c.

hmmm, the submit_urb-error is gone, but random lockups and usb-device-disconnect until reboot have come... (bitkeeper-2.6 from just one hour ago...) with your function.

try async_unlink option. it's experimental but better for 2.6. on 2.4 usb-uhci, it caused oops, though...

Takashi

-------------------------------------------------------
This SF.net email is sponsored by: SF.net Giveback Program. Does SourceForge.net help you be more productive? Does it help you create better code? SHARE THE LOVE, and help us help YOU! Click Here: http://sourceforge.net/donate/

#4 lore raw
In reply to Niklas Werner
Quoted 1 line · Niklas Werner

hmmm, the submit_urb-error is gone, but random lockups and usb-device-disconnect until reboot have come... (bitkeeper-2.6 from just one hour ago...) with your function.

Interesting. I haven't seen any of these.

Karim

Signature · 11 lines
Author, Speaker, Developer, Consultant
Pushing Embedded and Real-Time Linux Systems Beyond the Limits
http://www.opersys.com || [email protected] || 514-812-4145



-------------------------------------------------------
This SF.net email is sponsored by: SF.net Giveback Program.
Does SourceForge.net help you be more productive?  Does it
help you create better code?  SHARE THE LOVE, and help us help
YOU!  Click Here: http://sourceforge.net/donate/
#5 lore raw
In reply to Karim Yaghmour

At Thu, 20 Nov 2003 00:50:03 -0500,

Quoted 13 lines · Karim Yaghmour

Well, it seems that I'm going to have to answer my own self ... :)

The following is what I've been able to find using additional tracing info. Also there's a fix for usbaudio.c.

Basically, it's as I said before: the usbaudio driver uses URBs without even checking if they're in use or not. Surprisingly, it does this with one channel, but not the other. Here's what I get in the traces when I separated the events by channel:
Channel A (the hex data is the value of *subs):
-----------------------------------------------------------------------------------

(snip)

Quoted 61 lines

To make a long story short, the most important issue here is to check whether the URB is active before using it. If it is, then it must be unlinked first.

Here's my modified start_urbs() function from usbaudio.c:
static int start_urbs(snd_usb_substream_t *subs, snd_pcm_runtime_t *runtime)

{
     unsigned int i;
     int err;

     for (i = 0; i < subs->nurbs; i++) {
         snd_assert(subs->dataurb[i].urb, return -EINVAL);
         /* Deactivate URB prior to using it. K.Y. */
         if (test_and_clear_bit(i, &subs->active_mask)) {
             set_bit(i, &subs->unlink_mask);
             usb_unlink_urb(subs->dataurb[i].urb);
         }
         if (subs->ops.prepare(subs, runtime, subs->dataurb[i].urb) < 0) {
             snd_printk(KERN_ERR "cannot prepare datapipe for urb %d\n", i);
             goto __error;
         }
     }
     if (subs->syncpipe) {
         for (i = 0; i < SYNC_URBS; i++) {
             snd_assert(subs->syncurb[i].urb, return -EINVAL);
             if (subs->ops.prepare_sync(subs, runtime, subs->syncurb[i].urb) < 0) {
                 snd_printk(KERN_ERR "cannot prepare syncpipe for urb %d\n", i);
                 goto __error;
             }
         }
     }

     subs->running = 1;
     for (i = 0; i < subs->nurbs; i++) {
         trace_std_formatted_event(event_start_urb, i,  (unsigned int) subs);
         if ((err = usb_submit_urb(subs->dataurb[i].urb, GFP_KERNEL)) < 0) {
             trace_std_formatted_event(event_abuse_urb, i,  (unsigned int) subs);
             snd_printk(KERN_ERR "cannot submit datapipe for urb %d, err = %d\n", i, err);
             goto __error;
         }
         set_bit(i, &subs->active_mask);
         clear_bit(i, &subs->unlink_mask);
     }
     if (subs->syncpipe) {
         for (i = 0; i < SYNC_URBS; i++) {
             if ((err = usb_submit_urb(subs->syncurb[i].urb, GFP_KERNEL)) < 0) {
                 snd_printk(KERN_ERR "cannot submit syncpipe for urb %d, err = %d\n", i, err);
                 goto __error;
             }
             set_bit(i + 16, &subs->active_mask);
         }
     }
     return 0;

__error:

     // snd_pcm_stop(subs->pcm_substream, SNDRV_PCM_STATE_XRUN);
     deactivate_urbs(subs, 0);
     return -EPIPE;
}

Using this code, I haven't had any more errors at all.

Quoted 7 lines

In order to use this, just remove all the trace_* functions. This new function just checks if a URB is already in use prior to using it and clears it. Also, it clears the unlink_mask after the call to usb_submit_urb in order for it to be "clearable" by deactivate_urbs(). Without this last clear_bit(), URBs can only be cleared once by deactivate_urbs() and it takes a new init_substream_urbs() to set the entire mask back to 0. I don't see how this makes any sense.

it'd be better to clean unlink_mask in the complete callback for the case you use async unlink mode (see below).
and, the check of active_mask should be done in prepare callback, not in the trigger callback. the trigger callback must be as short as possible. we can put deactivate_urbs() in prepre callback so that the urbs become clean before starting streams.
(but still we have a problem of async unlink because prepare callback
is also in the spinlocked context.)

the current code is surely buggy, because it issues sync unlink inside the spinlocked context. it's problematic on 2.6 kernel or SMP system, and may result in kernel oops. i added async_unlink module option to change the behavior in the new version.

but it's still disabled as default, because unlinking multiple urbs asynchronouly on 2.4 kernel causes kernel panic. sigh. unfortunately, there is no perfect solution to satisfy all versions yet. perhaps we need to redesign the linked-pcm streams to make it possible to call prepare callback without spinlocks.

Takashi

-------------------------------------------------------
This SF.net email is sponsored by: SF.net Giveback Program. Does SourceForge.net help you be more productive? Does it help you create better code? SHARE THE LOVE, and help us help YOU! Click Here: http://sourceforge.net/donate/

#6 lore raw
Quoted 6 lines · Takashi Iwai

the current code is surely buggy, because it issues sync unlink inside the spinlocked context. it's problematic on 2.6 kernel or SMP system, and may result in kernel oops. i added async_unlink module option to change the behavior in the new version.

Quoted 7 lines

but it's still disabled as default, because unlinking multiple urbs asynchronouly on 2.4 kernel causes kernel panic. sigh. unfortunately, there is no perfect solution to satisfy all versions yet. perhaps we need to redesign the linked-pcm streams to make it possible to call prepare callback without spinlocks.

Ahhh, Finally an explanaiton for why my root /var/ partition is always being filled up with usb errors.

Would this also have a detrimental effect on latency? If so then I can let other usb-audio user know why.

Signature · 17 lines
Patrick Shirkey - Boost Hardware Ltd.
Http://www.boosthardware.com
Http://www.djcj.org - The Linux Audio Users guide
========================================

Being on stage with the band in front of crowds shouting, "Get off! No! We want normal music!", I think that was more like acting than anything I've ever done.

Goldie, 8 Nov, 2002
The Scotsman



-------------------------------------------------------
This SF.net email is sponsored by: SF.net Giveback Program.
Does SourceForge.net help you be more productive?  Does it
help you create better code?  SHARE THE LOVE, and help us help
YOU!  Click Here: http://sourceforge.net/donate/
#7 lore raw

At Thu, 20 Nov 2003 23:33:52 +0900,

Quoted 16 lines · Patrick Shirkey
Quoted 6 lines · Takashi Iwai

the current code is surely buggy, because it issues sync unlink inside the spinlocked context. it's problematic on 2.6 kernel or SMP system, and may result in kernel oops. i added async_unlink module option to change the behavior in the new version.

Quoted 7 lines

but it's still disabled as default, because unlinking multiple urbs asynchronouly on 2.4 kernel causes kernel panic. sigh. unfortunately, there is no perfect solution to satisfy all versions yet. perhaps we need to redesign the linked-pcm streams to make it possible to call prepare callback without spinlocks.

Quoted 5 lines

Ahhh, Finally an explanaiton for why my root /var/ partition is always being filled up with usb errors.

Would this also have a detrimental effect on latency? If so then I can let other usb-audio user know why.

no, it has nothing to do with the latency but oopsen / lock-up. the problem exists only at stopping the streams. if the everything is ok, it's not a big issue. but, when the device runs with small latency, which may result in xrun often, has the higher probability to hit this bug, because ALSA tries to stop the stream if xrun happens.

Takashi

-------------------------------------------------------
This SF.net email is sponsored by: SF.net Giveback Program. Does SourceForge.net help you be more productive? Does it help you create better code? SHARE THE LOVE, and help us help YOU! Click Here: http://sourceforge.net/donate/

#8 lore raw
Quoted 8 lines · Takashi Iwai

no, it has nothing to do with the latency but oopsen / lock-up. the problem exists only at stopping the streams. if the everything is ok, it's not a big issue. but, when the device runs with small latency, which may result in xrun often, has the higher probability to hit this bug, because ALSA tries to stop the stream if xrun happens.

Thanks. There should be a disclaimer somewhere for this problem.

Signature · 17 lines
Patrick Shirkey - Boost Hardware Ltd.
Http://www.boosthardware.com
Http://www.djcj.org - The Linux Audio Users guide
========================================

Being on stage with the band in front of crowds shouting, "Get off! No! We want normal music!", I think that was more like acting than anything I've ever done.

Goldie, 8 Nov, 2002
The Scotsman



-------------------------------------------------------
This SF.net email is sponsored by: SF.net Giveback Program.
Does SourceForge.net help you be more productive?  Does it
help you create better code?  SHARE THE LOVE, and help us help
YOU!  Click Here: http://sourceforge.net/donate/
#9 lore raw
In reply to Takashi Iwai
Quoted 6 lines · Takashi Iwai

it'd be better to clean unlink_mask in the complete callback for the case you use async unlink mode (see below).
and, the check of active_mask should be done in prepare callback, not in the trigger callback. the trigger callback must be as short as possible. we can put deactivate_urbs() in prepre callback so that the urbs become clean before starting streams.

Sounds right.

Quoted 2 lines

(but still we have a problem of async unlink because prepare callback
is also in the spinlocked context.)

Would it be fair to say that this driver requires some major rework?

Karim

Signature · 11 lines
Author, Speaker, Developer, Consultant
Pushing Embedded and Real-Time Linux Systems Beyond the Limits
http://www.opersys.com || [email protected] || 514-812-4145



-------------------------------------------------------
This SF.net email is sponsored by: SF.net Giveback Program.
Does SourceForge.net help you be more productive?  Does it
help you create better code?  SHARE THE LOVE, and help us help
YOU!  Click Here: http://sourceforge.net/donate/
#10 lore raw

At Thu, 20 Nov 2003 15:28:25 -0500,

Quoted 11 lines · Karim Yaghmour
Quoted 6 lines · Takashi Iwai

it'd be better to clean unlink_mask in the complete callback for the case you use async unlink mode (see below).
and, the check of active_mask should be done in prepare callback, not in the trigger callback. the trigger callback must be as short as possible. we can put deactivate_urbs() in prepre callback so that the urbs become clean before starting streams.

Sounds right.

already changed on cvs :) will be reflected to 1.0.0-test2.

Quoted 4 lines
Quoted 2 lines

(but still we have a problem of async unlink because prepare callback
is also in the spinlocked context.)

Would it be fair to say that this driver requires some major rework?

yes. not much in usbaudio.c itself but in the core PCM part.

Takashi

-------------------------------------------------------
This SF.net email is sponsored by: SF.net Giveback Program. Does SourceForge.net help you be more productive? Does it help you create better code? SHARE THE LOVE, and help us help YOU! Click Here: http://sourceforge.net/donate/