omap-mcbsp: TX Buffer Overflow

omap-mcbsp: TX Buffer Overflow

[Tomas Novotny]

Hi,

we have AM3703 based board similar to BeagleBoard. I'm hitting this error
after upgrade to latest LTS 4.19.71 (upgraded from 4.1):

omap-mcbsp 49022000.mcbsp: TX Buffer Overflow!

This appears during or after playing of short (~2s) ding-dong wav. That error
exists for longer time, because handling of tx buffer overflow irq was
introduced in 2016: 4e85e7776eba ("ASoC: omap-mcbsp: Enable TX/RX under and
overflow interrupts"). I've cherry-picked it to 4.1 and I see the error there also.
The sound seems clear and ok to me, but we are using low quality speaker.

There are two workarounds to get rid of the message:
1) Change 'dma_op_mode' sysfs attribute from 'element' to 'threshold'. I
found that just by coincidence when checking sysfs attributes.
2) Compile kernel with CONFIG_VIDEO_OMAP3=y. Found on Logic PD forum [1].

Does anybody have any idea what's going wrong? Or why these (somehow)
unrelated workarounds help?

Thanks,

Tomas

[1] https://support.logicpd.com/TDGForum/tabid/124/aft/2277/Default.aspx

Re: omap-mcbsp: TX Buffer Overflow

[Ladislav Michl]

On Fri, Sep 06, 2019 at 04:51:09PM +0200, Tomas Novotny wrote:
> Hi,
> 
> we have AM3703 based board similar to BeagleBoard. I'm hitting this error
> after upgrade to latest LTS 4.19.71 (upgraded from 4.1):
> 
> omap-mcbsp 49022000.mcbsp: TX Buffer Overflow!
> 
> This appears during or after playing of short (~2s) ding-dong wav. That error
> exists for longer time, because handling of tx buffer overflow irq was
> introduced in 2016: 4e85e7776eba ("ASoC: omap-mcbsp: Enable TX/RX under and
> overflow interrupts"). I've cherry-picked it to 4.1 and I see the error there also.
> The sound seems clear and ok to me, but we are using low quality speaker.

Just FYI, for stream capture there's
omap-mcbsp 49022000.mcbsp: RX Buffer Underflow!

As far as I remember all stable kernels we have in production - 4.9.x, 4.14.x and
4.19.x - are affected. IGEPv2 with both DM3730 and OMAP3530 are affected
(headless machines, CONFIG_VIDEO_OMAP3=n).

And DT is probably worth updating:
omap_hwmod: mcbsp2_sidetone using broken dt data from mcbsp
omap_hwmod: mcbsp3_sidetone using broken dt data from mcbsp

I never motivated myself to dig deeper as catured stream looks pretty normal.

	l.

> There are two workarounds to get rid of the message:
> 1) Change 'dma_op_mode' sysfs attribute from 'element' to 'threshold'. I
> found that just by coincidence when checking sysfs attributes.
> 2) Compile kernel with CONFIG_VIDEO_OMAP3=y. Found on Logic PD forum [1].
> 
> Does anybody have any idea what's going wrong? Or why these (somehow)
> unrelated workarounds help?
> 
> Thanks,
> 
> Tomas
> 
> [1] https://support.logicpd.com/TDGForum/tabid/124/aft/2277/Default.aspx

Re: omap-mcbsp: TX Buffer Overflow

[Tony Lindgren]

* Ladislav Michl <ladis@linux-mips.org> [190907 09:14]:
> On Fri, Sep 06, 2019 at 04:51:09PM +0200, Tomas Novotny wrote:
> > Hi,
> > 
> > we have AM3703 based board similar to BeagleBoard. I'm hitting this error
> > after upgrade to latest LTS 4.19.71 (upgraded from 4.1):
> > 
> > omap-mcbsp 49022000.mcbsp: TX Buffer Overflow!
> > 
> > This appears during or after playing of short (~2s) ding-dong wav. That error
> > exists for longer time, because handling of tx buffer overflow irq was
> > introduced in 2016: 4e85e7776eba ("ASoC: omap-mcbsp: Enable TX/RX under and
> > overflow interrupts"). I've cherry-picked it to 4.1 and I see the error there also.
> > The sound seems clear and ok to me, but we are using low quality speaker.
> 
> Just FYI, for stream capture there's
> omap-mcbsp 49022000.mcbsp: RX Buffer Underflow!
> 
> As far as I remember all stable kernels we have in production - 4.9.x, 4.14.x and
> 4.19.x - are affected. IGEPv2 with both DM3730 and OMAP3530 are affected
> (headless machines, CONFIG_VIDEO_OMAP3=n).

Hmm I wonder if this is still related to the SoC idling?
See commit 9834ffd1ecc3 ("ASoC: omap-mcbsp: Add PM QoS support for McBSP
to prevent glitches"), maybe something still needs to be fixed in that
area.

> And DT is probably worth updating:
> omap_hwmod: mcbsp2_sidetone using broken dt data from mcbsp
> omap_hwmod: mcbsp3_sidetone using broken dt data from mcbsp
> 
> I never motivated myself to dig deeper as catured stream looks pretty normal.

These mean the devices should really have separate nodes
in the dts rather than combining multiple devices into a
single node with multiple reg entries.

The issue with combining multiple devices into a single device
is that flushing posted write with a read back to one register
range will not flush it for the other which can cause mysterious
bugs.

Regards,

Tony

Re: omap-mcbsp: TX Buffer Overflow

[Tomas Novotny]

Hi Tony,

On Mon, 9 Sep 2019 09:24:07 -0700
Tony Lindgren <tony@atomide.com> wrote:

> * Ladislav Michl <ladis@linux-mips.org> [190907 09:14]:
> > On Fri, Sep 06, 2019 at 04:51:09PM +0200, Tomas Novotny wrote:  
> > > Hi,
> > > 
> > > we have AM3703 based board similar to BeagleBoard. I'm hitting this error
> > > after upgrade to latest LTS 4.19.71 (upgraded from 4.1):
> > > 
> > > omap-mcbsp 49022000.mcbsp: TX Buffer Overflow!
> > > 
> > > This appears during or after playing of short (~2s) ding-dong wav. That error
> > > exists for longer time, because handling of tx buffer overflow irq was
> > > introduced in 2016: 4e85e7776eba ("ASoC: omap-mcbsp: Enable TX/RX under and
> > > overflow interrupts"). I've cherry-picked it to 4.1 and I see the error there also.
> > > The sound seems clear and ok to me, but we are using low quality speaker.  
> > 
> > Just FYI, for stream capture there's
> > omap-mcbsp 49022000.mcbsp: RX Buffer Underflow!
> > 
> > As far as I remember all stable kernels we have in production - 4.9.x, 4.14.x and
> > 4.19.x - are affected. IGEPv2 with both DM3730 and OMAP3530 are affected
> > (headless machines, CONFIG_VIDEO_OMAP3=n).  
> 
> Hmm I wonder if this is still related to the SoC idling?
> See commit 9834ffd1ecc3 ("ASoC: omap-mcbsp: Add PM QoS support for McBSP
> to prevent glitches"), maybe something still needs to be fixed in that
> area.

I also found the cpuidle information somewhere, but CPU_IDLE=n (and let PM=y)
didn't help me. I'm speaking just for playback. I can do some better test if
you want.

Thanks,

Tomas

> > And DT is probably worth updating:
> > omap_hwmod: mcbsp2_sidetone using broken dt data from mcbsp
> > omap_hwmod: mcbsp3_sidetone using broken dt data from mcbsp
> > 
> > I never motivated myself to dig deeper as catured stream looks pretty normal.  
> 
> These mean the devices should really have separate nodes
> in the dts rather than combining multiple devices into a
> single node with multiple reg entries.
> 
> The issue with combining multiple devices into a single device
> is that flushing posted write with a read back to one register
> range will not flush it for the other which can cause mysterious
> bugs.
> 
> Regards,
> 
> Tony
>