[alsa-devel] Issues w/ Creative Labs [SB X-Fi Xtreme Audio] CA0110-IBG

Takashi Iwai tiwai at suse.de
Thu Mar 12 07:19:12 CET 2015


At Wed, 11 Mar 2015 20:59:58 -0700,
Alnie wrote:
> 
> 
> 
> On 03/09/2015 09:00 AM, Takashi Iwai wrote:
> > At Mon, 09 Mar 2015 08:22:34 -0700,
> > Alnie wrote:
> >>
> >> On 03/07/2015 10:06 AM, Alnie wrote:
> >>> On 03/07/2015 08:32 AM, Takashi Iwai wrote:
> >>>> At Sat, 07 Mar 2015 08:28:29 -0800,
> >>>> Alnie wrote:
> >>>>>
> >>>>> On 03/06/2015 11:56 PM, Takashi Iwai wrote:
> >>>>>> At Fri, 06 Mar 2015 15:31:32 -0800,
> >>>>>> Alnie wrote:
> >>>>>>>
> >>>>>>> On 03/04/2015 01:35 AM, Takashi Iwai wrote:
> >>>>>>>> At Wed, 04 Mar 2015 01:01:32 -0800,
> >>>>>>>> Alnie wrote:
> >>>>>>>>>
> >>>>>>>>>
> >>>>>>>>>
> >>>>>>>>> On 03/04/2015 12:00 AM, Takashi Iwai wrote:
> >>>>>>>>>> At Tue, 03 Mar 2015 18:12:31 -0800,
> >>>>>>>>>> Alnie wrote:
> >>>>>>>>>>>
> >>>>>>>>>>>> My suggestion isn't about a compile option but that you add
> >>>>>>>>>>>> some debug
> >>>>>>>>>>>> printk() calls manually around some codes.  We need to know
> >>>>>>>>>>>> the value
> >>>>>>>>>>>> written and read by azx_write*() and azx_read*() calls.
> >>>>>>>>>>>> Especially
> >>>>>>>>>>>> the value read in pci_azx_read*() is more interesting. You can
> >>>>>>>>>>>> try to
> >>>>>>>>>>>> modify sound/pci/hda/hda_intel.c and add a printk() to each
> >>>>>>>>>>>> pci_azx_read*() function for printing the value to be returned.
> >>>>>>>>>>>> Beware that this will likely flood many messages, so just try
> >>>>>>>>>>>> once.
> >>>>>>>>>>>>
> >>>>>>>>>>>>
> >>>>>>>>>>>> Takashi
> >>>>>>>>>>>>
> >>>>>>>>>>>
> >>>>>>>>>>> I can not find any reference to pci_azx_read in hda_intel.c
> >>>>>>>>>>
> >>>>>>>>>> You must be using a too old kernel, then.  Please use the latest
> >>>>>>>>>> kernel for debugging.
> >>>>>>>>>>
> >>>>>>>>>>
> >>>>>>>>>> Takashi
> >>>>>>>>>>
> >>>>>>>>>
> >>>>>>>>> Ok. I now have latest kernel.
> >>>>>>>>>
> >>>>>>>>> Here is a small portion...
> >>>>>>>>>
> >>>>>>>>> /* PCI register access. */
> >>>>>>>>> static void pci_azx_writel(u32 value, u32 __iomem *addr)
> >>>>>>>>> {
> >>>>>>>>>      writel(value, addr);
> >>>>>>>>> }
> >>>>>>>>>
> >>>>>>>>> static u32 pci_azx_readl(u32 __iomem *addr)
> >>>>>>>>> {
> >>>>>>>>>      return readl(addr);
> >>>>>>>>> }
> >>>>>>>>>
> >>>>>>>>> Can you show me how I can properly place printk without breaking
> >>>>>>>>> things
> >>>>>>>>> and produce relevant messages?
> >>>>>>>>
> >>>>>>>> Something like:
> >>>>>>>>
> >>>>>>>> static u32 pci_azx_readl(u32 __iomem *addr)
> >>>>>>>> {
> >>>>>>>>      u32 val = readl(addr);
> >>>>>>>>      pr_info("XXX readl %p %x\n", addr, val);
> >>>>>>>>      return val;
> >>>>>>>> }
> >>>>>>>>
> >>>>>>>> But since there are quite lots of accesses, it might be safer to use
> >>>>>>>> the ratelimited version.  Use like the following:
> >>>>>>>>      pr_info_ratelimited("XXX readl %p %x\n", addr, val);
> >>>>>>>>
> >>>>>>>> Also there are variants pci_azx_readw() and _readb().
> >>>>>>>>
> >>>>>>>>
> >>>>>>>> Takashi
> >>>>>>>>
> >>>>>>>
> >>>>>>>
> >>>>>>> Ok, I was only able to guess through the code for the read values. If
> >>>>>>> you need write information too please show me how.
> >>>>>>> here are the mods I made...
> >>>>>>>
> >>>>>>> /* PCI register access. */
> >>>>>>> static void pci_azx_writel(u32 value, u32 __iomem *addr)
> >>>>>>> {
> >>>>>>>      writel(value, addr);
> >>>>>>> }
> >>>>>>>
> >>>>>>> static u32 pci_azx_readl(u32 __iomem *addr)
> >>>>>>> {
> >>>>>>>         u32 val = readl(addr);
> >>>>>>>         pr_info_ratelimited("XXX readl %p %x\n", addr, val);
> >>>>>>>         return val;
> >>>>>>> }
> >>>>>>>
> >>>>>>> static void pci_azx_writew(u16 value, u16 __iomem *addr)
> >>>>>>> {
> >>>>>>>      writew(value, addr);
> >>>>>>> }
> >>>>>>>
> >>>>>>> static u16 pci_azx_readw(u16 __iomem *addr)
> >>>>>>> {
> >>>>>>>      u16 val = readw(addr);
> >>>>>>>      pr_info_ratelimited("XXX readw %p %x\n", addr, val);
> >>>>>>>      return val;
> >>>>>>> }
> >>>>>>>
> >>>>>>> static void pci_azx_writeb(u8 value, u8 __iomem *addr)
> >>>>>>> {
> >>>>>>>      writeb(value, addr);
> >>>>>>> }
> >>>>>>>
> >>>>>>> static u8 pci_azx_readb(u8 __iomem *addr)
> >>>>>>> {
> >>>>>>>      u8 val = readb(addr);
> >>>>>>>      pr_info_ratelimited("XXX readb %p %x\n", addr, val);
> >>>>>>>      return val;
> >>>>>>> }
> >>>>>>>
> >>>>>>> and i was able to produce this...
> >>>>>>>
> >>>>>>> dmesg | grep -i read
> >>>>>>> [    0.764102] tpm_tis 00:05: A TPM error (7) occurred attempting to
> >>>>>>> read a pcr value
> >>>>>>> [    0.769674] Write protecting the kernel read-only data: 12288k
> >>>>>>> [    1.729952] sd 0:0:0:0: [sda] Write cache: enabled, read cache:
> >>>>>>> enabled, doesn't support DPO or FUA
> >>>>>>> [    2.071100] sd 1:0:0:0: [sdb] Write cache: enabled, read cache:
> >>>>>>> enabled, doesn't support DPO or FUA
> >>>>>>> [    2.778875] EXT4-fs (sda5): INFO: recovery required on readonly
> >>>>>>> filesystem
> >>>>>>> [    2.817982] EXT4-fs (sda5): orphan cleanup on readonly fs
> >>>>>>> [    4.662442] XXX readw ffffc90004f30000 4401
> >>>>>>
> >>>>>> This reads good, but I thought you have two sound devices (onboard and
> >>>>>> Creative)?  If so, disable the onboard one via enable=0 option (if the
> >>>>>> onboard one is assigned first).  Otherwise the all good and bad
> >>>>>> results are mixed up.
> >>>>>>
> >>>>>>
> >>>>>> Takashi
> >>>>>>
> >>>>>
> >>>>> I included options snd-hda-intel enable=0 index=0
> >>>>>
> >>>>> This is dmesg output (1b is onboard)...
> >>>>> dmesg | grep -i hda
> >>>>> [    3.700213] snd_hda_intel: probe of 0000:00:1b.0 failed with error -2
> >>>>> [   37.729507] snd_hda_intel 0000:06:00.0: enabling device (0000 ->
> >>>>> 0002)
> >>>>> [   37.729703] snd_hda_intel 0000:06:00.0: enabling bus mastering
> >>>>> [   37.736026] snd_hda_intel 0000:06:00.0: CORB reset timeout#1,
> >>>>> CORBRP = 0
> >>>>> [   38.740090] snd_hda_intel 0000:06:00.0: Codec #1 probe error;
> >>>>> disabling it...
> >>>>> [   38.747292] snd_hda_intel 0000:06:00.0: CORB reset timeout#1,
> >>>>> CORBRP = 0
> >>>>> [   43.776089] snd_hda_intel 0000:06:00.0: no AFG or MFG node found
> >>>>> [   43.776111] snd_hda_intel 0000:06:00.0: no codecs initialized
> >>>>>
> >>>>> dmesg | grep -i XXX
> >>>>> [   37.729723] XXX readw ffffc90005188000 3300
> >>>>> [   37.729768] XXX readl ffffc90005188008 0
> >>>>
> >>>> OK, so all read look fine.  Right now we saw a bug report with git
> >>>> bisection showing a bad commit in HD-audio controller code.  This
> >>>> might be your issue, too.  Could you try the following oneliner?
> >>>>
> >>>>
> >>>> thanks,
> >>>>
> >>>> Takashi
> >>>>
> >>>> diff --git a/sound/pci/hda/hda_controller.c
> >>>> b/sound/pci/hda/hda_controller.c
> >>>> index 7b4377265b25..57c4575fac26 100644
> >>>> --- a/sound/pci/hda/hda_controller.c
> >>>> +++ b/sound/pci/hda/hda_controller.c
> >>>> @@ -1194,7 +1194,7 @@ static unsigned int azx_rirb_get_response(struct
> >>>> hda_bus *bus,
> >>>>            }
> >>>>        }
> >>>>
> >>>> -    if (!bus->no_response_fallback)
> >>>> +    if (bus->no_response_fallback)
> >>>>            return -1;
> >>>>
> >>>>        if (!chip->polling_mode && chip->poll_count < 2) {
> >>>>
> >>>
> >>>
> >>> I believe I applied the patch correctly... if you would like to double
> >>> check here is the hda_controller.c file (http://pastebin.com/NA7AV2n1)
> >>>
> >>> Same complaint comes up, after reinsert and force-reload as well...
> >>>
> >>> [  419.344572] snd_hda_intel 0000:06:00.0: enabling bus mastering
> >>> [  419.350967] snd_hda_intel 0000:06:00.0: CORB reset timeout#1, CORBRP = 0
> >>> [  422.360107] snd_hda_intel 0000:06:00.0: azx_get_response timeout,
> >>> switching to polling mode: last cmd=0x100f0000
> >>> [  423.364094] snd_hda_intel 0000:06:00.0: Codec #1 probe error;
> >>> disabling it...
> >>> [  423.371128] snd_hda_intel 0000:06:00.0: CORB reset timeout#1, CORBRP = 0
> >>> [  424.376098] snd_hda_intel 0000:06:00.0: azx_get_response timeout,
> >>> switching to single_cmd mode: last cmd=0x100f0000
> >>> [  424.377052] snd_hda_intel 0000:06:00.0: no AFG or MFG node found
> >>> [  424.377067] snd_hda_intel 0000:06:00.0: no codecs initialized
> >>>
> >>
> >> I'm assuming we've exhausted our options for now. It seems probable this
> >> model card (ExpressCard/54) never worked in Linux to begin with. I
> >> looked over the forum posts from those claiming they had it operational
> >> (in 2.6 kernel), but it was for the PCI-E version. I greatly appreciate
> >> your assistance with this. Thank you.
> >
> > Yes, some PCI-e version has worked, I personally tested it once.
> > But yours is apparently a different one, and we finally go back to the
> > starting point again.
> >
> > What you need to do from now on is to analyze why the codec probe
> > fails.  It failed even with single_cmd mode.  This is strange, but it
> > might be that the board is never 100% compliant, so it might not
> > support that mode.
> >
> > There are some workarounds done specific to the Creative devices.  For
> > example, azx_init_cmd_io() in hda_controller.c has a special handling
> > to set RIRB interrupt count to 0xc0.  I guess this needs to be set
> > normally.  That is, disable the following if () check, and run always
> > azx_writew(chip, RINTCNT, 1) instead:
> >
> > 	if (chip->driver_caps & AZX_DCAPS_CTX_WORKAROUND)
> > 		azx_writew(chip, RINTCNT, 0xc0);
> > 	else
> > 		azx_writew(chip, RINTCNT, 1);
> 
> This modification alone produced the same complaint. These are the edits 
> I made...
> 
> 	/* set N=1, get RIRB response interrupt for new entry */
> 	//if (chip->driver_caps & AZX_DCAPS_CTX_WORKAROUND)
> 		//azx_writew(chip, RINTCNT, 0xc0);
> 	//else
> 	azx_writew(chip, RINTCNT, 1);
> 
> >
> > There is another workaround in azx_pcm_prepare() for the capture
> > streams, but this is irrelevant as of now -- we're concentrating only
> > on the codec probe issue.
> >
> > If modifying above still doesn't work, you may try to force continuing
> > the probe by removing -EIO return from probe_codec() in
> > hda_controller.c.  That is, remove the following line from that
> > function:
> >
> > 	if (res == -1)
> > 		return -EIO;
> 
> This is the edit I made...
> 
> 	mutex_unlock(&chip->bus->cmd_mutex);
> 	//if (res == -1)
> 		//return -EIO;
> 	dev_dbg(chip->card->dev, "codec #%d probed OK\n", addr);
> 
> I assume I did this correctly, but here is the full hda_controller.c I 
> compiled if you would like to double check http://pastebin.com/fvVVA1Ew
> 
> >
> > Of course, it doesn't mean this would work, but we can see whether the
> > board reacts somehow better.
> >
> 
> Same complaint comes up.
> 
> [    7.600051] snd_hda_intel 0000:06:00.0: azx_get_response timeout, 
> switching to polling mode: last cmd=0x100f0000
> [    9.608050] snd_hda_intel 0000:06:00.0: azx_get_response timeout, 
> switching to single_cmd mode: last cmd=0x100f0000
> [    9.609976] snd_hda_intel 0000:06:00.0: no AFG or MFG node found
> [    9.611023] snd_hda_intel 0000:06:00.0: no codecs initialized
> [   10.776445] init: alsa-restore main process (1461) terminated with 
> status 19

Then there is little hope at this point.  We can analyze more, but
it's difficult without the hardware locally, or you'll need to have a
deeper knowledge of HD-audio.


Takashi


More information about the Alsa-devel mailing list