lists.openwall.net   lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  linux-hardening  linux-cve-announce  PHC 
Open Source and information security mailing list archives
 
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date:	Sat, 11 Jan 2014 10:35:33 +0100
From:	Takashi Iwai <tiwai@...e.de>
To:	Liam Girdwood <liam.r.girdwood@...ux.intel.com>
Cc:	"alsa-devel@...a-project.org" <alsa-devel@...a-project.org>,
	Mark Brown <broonie@...nel.org>,
	"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
	Nenghua Cao <nhcao@...vell.com>
Subject: Re: [alsa-devel] [PATCH] ASoC: dpcm: don't do hw_param when BE has done	hw_param

At Fri, 10 Jan 2014 18:43:09 +0000,
Liam Girdwood wrote:
> 
> On Fri, 2014-01-10 at 14:46 +0100, Takashi Iwai wrote:
> > At Fri, 10 Jan 2014 12:29:08 +0000,
> > Liam Girdwood wrote:
> > > 
> > > On Fri, 2014-01-10 at 13:01 +0100, Takashi Iwai wrote:
> > > > At Fri, 10 Jan 2014 19:59:42 +0800,
> > > > Nenghua Cao wrote:
> > > > > 
> > > > > On 01/10/2014 07:47 PM, Liam Girdwood wrote:
> > > > > > On Fri, 2014-01-10 at 19:21 +0800, Nenghua Cao wrote:
> > > > > >> On 01/10/2014 06:55 PM, Takashi Iwai wrote:
> > > > > >>> [Corrected mail addresses of both Mark and Liam]
> > > > > >>>
> > > > > >> Hi, Takashi:
> > > > > >> Thanks for correcting my mistake.
> > > > > >>> At Fri, 10 Jan 2014 13:36:35 +0800,
> > > > > >>> Nenghua Cao wrote:
> > > > > >>>>
> > > > > >>>> From: Nenghua Cao <nhcao@...vell.com>
> > > > > >>>>
> > > > > >>>>     It fixes the following case:
> > > > > >>>>     Two FEs connects the same BE; FE1 & BE path has been opened and hw_paramed.
> > > > > >>>> At this momment, FE2 & BE path is being opened and hw_paramed. The BE
> > > > > >>>> dai will do hw_param again even if it has done hw_param. It is not
> > > > > >>>> reasonable.
> > > > > >>>> FE1------------>BE
> > > > > >>>> FE2-------------^
> > > > > >>>
> > > > > >>> What happens if FE2 tries to set up an incompatible hw_params?
> > > > > >>> (Through a quick glance, it won't work properly well, too, though...)
> > > > > >>>
> > > > > > 
> > > > > > The intention in this case would be for the DSP FE driver to determine
> > > > > > if it can perform format conversion or SRC to the running BE. If the DSP
> > > > > > cant do the conversion then it should fail.
> > > > > > 
> > > > > >> If FE2 uses an incompatible param, it will make FE1 doesn't work. Maybe
> > > > > >> FE2 works well.
> > > > > >> If FE2 uses the same param, BE hw_param function will be called twice
> > > > > >> (This is the most happening case).
> > > > > >> So we can't get benefits from it.
> > > > > > 
> > > > > > We shouldn't be calling the hw_params() on the BE when it's already
> > > > > > configured in this case. So this seems like a bug. However :-
> > > > > > 
> > > > > > /* only allow hw_params() if no connected FEs are running */
> > > > > > 		if (!snd_soc_dpcm_can_be_params(fe, be, stream))
> > > > > > 			continue;
> > > > > > 
> > > > > > 		if ((be->dpcm[stream].state != SND_SOC_DPCM_STATE_OPEN) &&
> > > > > > 		    (be->dpcm[stream].state != SND_SOC_DPCM_STATE_HW_PARAMS) &&
> > > > > > 		    (be->dpcm[stream].state != SND_SOC_DPCM_STATE_HW_FREE))
> > > > > > 			continue;
> > > > > > 
> > > > > > We do do a test to check if any connected FEs are running (i.e.
> > > > > > triggered) prior to calling hw_params() on the BE. Can you confirm if
> > > > > > the FE was running in your case ?
> > > > > > 
> > > > > Hi, Liam:
> > > > >    I am so glad to hear from you. In my case, FE1 has called hw_param,
> > > > > and before FE1 calls prepare/trigger function, the scheduler switches to
> > > > > do FE2 open, hw_param. So hw_param is called twice.
> > > > 
> > > > So basically the current implementation is racy about this.
> > > > 
> > > 
> > > This is a valid use case from the userspace perspective too. The BE in
> > > this case is a shared resource (whether userspace is aware or not) and
> > > I'd expect it to take the the last configured hw_params (in this case
> > > FE2 hw_params) before it is triggered.
> > 
> > Yes, it's how the current code works.  But, what if FE1 didn't know
> > that it's shared? (Actually how it can be informed explicitly?)
> 
> It cant atm, although this is not a problem on Android. It can be fixed
> when we export the graph to userspace though.

I see.

> > FE1 will still try to feed data in wrong formats/rates/etc, won't it?
> > 
> 
> No, FE1 wont change it's params in this state. It will be up to the DSP
> driver to perform the conversion to then new formats or fail if the new
> format cannot be supported.

Well, my concern is that FE1 is never notified about the hw_params
change done by FE2 in the current situation.  So, FE1 will feed the
data to BE as if the old hw_params is used.  The rest behavior is
certainly all up to hardware.

But, the point is that basically we already know that something is
wrong at the point BE2 setting up an incompatible hw_params; then it
should be notified properly to FE1, or the incompatible change must be
handled as an error.  This is the missing piece in the current
implementation.  The skip of redundant BE hw_params call can be
implemented as an optimization in this compatibility check, too.

> > At best, it should return an error when an incompatible hw_params
> > setup is done by FE2, IMO.  The re-setup by FE1 should be available
> > freely.  So, BE needs to remember who has set it up, then allows only
> > the further re-setup by that FE, for example.
> > 
> > > Fwiw, a similar topic came up the conference this year wrt compressed
> > > streams. The question was about configuring the BE format and rate
> > > directly from userspace. This should be possible without too much effort
> > > since the BE is essentially a PCM. e.g. from userspace
> > > 
> > > 1) configure FE1 hw_params
> > > 
> > > 2) configure FE2 hw_params
> > > 
> > > 3) optional - configure BE1 hw_params
> > > 
> > > If step 3 is not performed then the values from step2 are used.
> > 
> > I forgot about this discussion -- so how was the proposal to allow
> > BE's hw_params?  A new API, or a flag in hw_params?
> 
> The intention was to use the existing alsa-lib/tinyalsa PCM hw_params
> APIs. The BE would just export itself to usespace as a PCM (but without
> the capability for direct playback/capture - just format, rate setting)

Does it mean that, from kernel perspective, a BE creates a dedicated
(virtual) PCM device and expose it to user-space?  Or just through
special API?


thanks,

Takashi
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists