Skip to contentSkip to conversation

dmaengine: dw: resolve recursion lock when audio playback

PATCH In review yitian 4 participants 6 messages Sep 26, 2015 – Sep 28, 2015

Subsystems (2)
lore
Share & tools
lore

yitian

author
#0 lore raw

PATCH 1/1 dmaengine: dw: resolve recursion lock when audio playback

when audio playback is running, user space will call snd_pcm_lib_write1() function, it will first get pcm stream lock, after that it may call function snd_pcm_update_hw_ptr(), the call stack will be as below:
snd_pcm_lib_write1 <-- got pcm stream lock
--> snd_pcm_update_hw_ptr

     --> dwc_tx_status
      --> dwc_scan_descriptors
       --> callback
        --> dmaengine_pcm_dma_complete
         --> snd_pcm_period_elapsed <-- get stream lock again

recursion lock occurs in snd_pcm_period_elapsed function. remove dwc_scan_descriptors from dwc_tx_status can fix this issue.

---
 drivers/dma/dw/core.c | 6 +-----
 1 file changed, 1 insertion(+), 5 deletions(-)
drivers/dma/dw/core.c +1 -5
index cf1c87f..5341b44 100644
@@ -1078,11 +1078,7 @@ dwc_tx_status(struct dma_chan *chan,
 	ret = dma_cookie_status(chan, cookie, txstate);
 	if (ret == DMA_COMPLETE)
 		return ret;
-
-	dwc_scan_descriptors(to_dw_dma(chan->device), dwc);
-
-	ret = dma_cookie_status(chan, cookie, txstate);
-	if (ret != DMA_COMPLETE)
+	else
 		dma_set_residue(txstate, dwc_get_residue(dwc));
 
 	if (dwc->paused && ret == DMA_IN_PROGRESS)
Signature · 6 lines
1.7.12.4



Yitian
Best Regards
#1 lore raw
Quoted 11 lines · yitian

when audio playback is running, user space will call snd_pcm_lib_write1() function, it will first get pcm stream lock, after that it may call function snd_pcm_update_hw_ptr(), the call stack will be as below:
snd_pcm_lib_write1 <-- got pcm stream lock
--> snd_pcm_update_hw_ptr

     --> dwc_tx_status
      --> dwc_scan_descriptors
       --> callback
        --> dmaengine_pcm_dma_complete
         --> snd_pcm_period_elapsed <-- get stream lock again
Quoted 7 lines

recursion lock occurs in snd_pcm_period_elapsed function. remove dwc_scan_descriptors from dwc_tx_status can fix this issue.

---
 drivers/dma/dw/core.c | 6 +-----
 1 file changed, 1 insertion(+), 5 deletions(-)

I am not sure if this is a sane way of doing that, and we were scanning the descriptors for some valid reason..

Though, I am on vacation for few days now and will be able to look at the details later.

#2 lore raw
Quoted 15 lines · Viresh Kumar
Quoted 14 lines · yitian

when audio playback is running, user space will call snd_pcm_lib_write1() function, it will first get pcm stream lock, after that it may call function snd_pcm_update_hw_ptr(), the call stack will be as below:
snd_pcm_lib_write1 <-- got pcm stream lock
--> snd_pcm_update_hw_ptr

     --> dwc_tx_status
      --> dwc_scan_descriptors
       --> callback
        --> dmaengine_pcm_dma_complete
         --> snd_pcm_period_elapsed <-- get stream lock again

recursion lock occurs in snd_pcm_period_elapsed function. remove dwc_scan_descriptors from dwc_tx_status can fix this issue.

Quoted 7 lines
Quoted 4 lines
---
 drivers/dma/dw/core.c | 6 +-----
 1 file changed, 1 insertion(+), 5 deletions(-)

I am not sure if this is a sane way of doing that, and we were scanning the descriptors for some valid reason..

Actually one of the patches in a pile sitting in my private repo is also including similar change. In my case the reason is to support cyclic transfers natively.

Quoted 2 lines

Though, I am on vacation for few days now and will be able to look at the details later.

Have a good rest then!

Signature · 2 lines
With Best Regards,
Andy Shevchenko
author
#3 lore raw
Quoted 10 lines

From: Andy Shevchenko [mailto:[email protected]]
Sent: Monday, September 28, 2015 3:06 PM
To: Viresh Kumar <[email protected]>
Cc: yitian <[email protected]>; Viresh Kumar <[email protected]>; Andy Shevchenko <[email protected]>; Vinod Koul <[email protected]>; Dan Williams <[email protected]>; dmaengine <[email protected]>; Linux Kernel Mailing List <[email protected]>
Subject: Re: [PATCH 1/1] dmaengine: dw: resolve recursion

Quoted 7 lines
Quoted 2 lines

I am not sure if this is a sane way of doing that, and we were scanning the descriptors for some valid reason..

Actually one of the patches in a pile sitting in my private repo is also including similar change. In my case the reason is to support cyclic transfers natively.

Yes, i am using DW DMAC to support cyclic transfer. Currently it is very easy to get recursion lock crash, but after this patch everything is fine on my device. I am waiting for your further guide.
Thanks.

#4 lore raw
Quoted 16 lines · yitian
Quoted 16 lines

From: Andy Shevchenko [mailto:[email protected]]
Sent: Monday, September 28, 2015 3:06 PM
To: Viresh Kumar <[email protected]>
Cc: yitian <[email protected]>; Viresh Kumar <[email protected]>; Andy Shevchenko <[email protected]>; Vinod Koul < [email protected]>;
Dan Williams <[email protected]>; dmaengine <[email protected]>; Linux Kernel Mailing List <[email protected]>
Subject: Re: [PATCH 1/1] dmaengine: dw: resolve recursion

Quoted 3 lines

I am not sure if this is a sane way of doing that, and we were scanning
the descriptors for some valid reason..

Quoted 8 lines
Quoted 4 lines

Actually one of the patches in a pile sitting in my private repo is also including similar change. In my case the reason is to support cyclic transfers natively.

Yes, i am using DW DMAC to support cyclic transfer. Currently it is very easy to get recursion lock crash, but after this patch everything is fine on my device.

What is an actual hardware you are running kernel on?

Signature · 2 lines
Andy Shevchenko <[email protected]>
Intel Finland Oy
author
#5 lore raw
Quoted 33 lines

From: Andy Shevchenko [mailto:[email protected]]
Sent: Monday, September 28, 2015 4:46 PM
To: yitian <[email protected]>; 'Andy Shevchenko' <[email protected]>; 'Viresh Kumar' <[email protected]>
Cc: 'Viresh Kumar' <[email protected]>; 'Vinod Koul' <[email protected]>; 'Dan Williams' <[email protected]>; 'dmaengine' <[email protected]>; 'Linux Kernel Mailing List' <[email protected]>
Subject: Re: [PATCH 1/1] dmaengine: dw: resolve recursion lock when audio playback

Quoted 21 lines · yitian
Quoted 20 lines

From: Andy Shevchenko [mailto:[email protected]]
Sent: Monday, September 28, 2015 3:06 PM
To: Viresh Kumar <[email protected]>
Cc: yitian <[email protected]>; Viresh Kumar <[email protected]>; Andy Shevchenko <[email protected]>; Vinod Koul < [email protected]>;
Dan Williams <[email protected]>; dmaengine <[email protected]>; Linux Kernel Mailing List <[email protected]>
Subject: Re: [PATCH 1/1] dmaengine: dw: resolve recursion

Quoted 3 lines

I am not sure if this is a sane way of doing that, and we were scanning
the descriptors for some valid reason..

Actually one of the patches in a pile sitting in my private repo is also including similar change. In my case the reason is to support cyclic transfers natively.

Quoted 6 lines
Quoted 3 lines

Yes, i am using DW DMAC to support cyclic transfer. Currently it is very easy to get recursion lock crash, but after this patch everything is fine on my device.

What is an actual hardware you are running kernel on?

Hi Andy:

I am using a FPGA, with Cortex-A5 core, Designware I2S IP, Designware DMAC IP. What I was done is to run tinyplay and tinycap to test the playback and capture function on the FPGA. With my change, both of them are okay now. I didn't push the patch which added cyclic DMA support for dw DMAC because I didn't make it decent enough yet.