soc-camera : sh_mobile_ceu_camera race on free_buffer ?
Hi,
I am writing a soc camera driver, and I use sh_mobile_ceu_camera as an example.
But I don't understand how buffer are handled when the application is doing a streamoff :
streamoff will call videobuf_streamoff and then videobuf_queue_cancel. videobuf_queue_cancel will call free_buffer.
But we didn't do stop_capture, so as far I understand the controller is still writing data in memory. What prevent us to free the buffer we are writing.
Why doesn't we do a stop_capture before videobuf_streamoff ?
I saw that pxa_camera use videobuf_waiton, before freeing the buffer. That seem more safe, but that mean we need to wait that controller finish to write all the pending buffer.
Matthieu
--
video4linux-list mailing list
Unsubscribe mailto:[email protected]?subject=unsubscribe
https://www.redhat.com/mailman/listinfo/video4linux-list
Hi Matthieu,
Quoted 8 lines · Matthieu CASTET
Hi,
I am writing a soc camera driver, and I use
sh_mobile_ceu_cameraas an example.But I don't understand how buffer are handled when the application is doing a streamoff :
Quoted 6 lines
streamoff will call
videobuf_streamoffand thenvideobuf_queue_cancel.videobuf_queue_cancelwill callfree_buffer.But we didn't do
stop_capture, so as far I understand the controller is still writing data in memory. What prevent us to free the buffer we are writing.
I have not looked into this in great detail, but isn't this handled by the videobuf state? The videobuf has state VIDEOBUF_ACTIVE while it is in use. I don't think such a buffer is freed.
Quoted 5 lines
Why doesn't we do a
stop_capturebeforevideobuf_streamoff?I saw that
pxa_camerausevideobuf_waiton, before freeing the buffer. That seem more safe, but that mean we need to wait that controller finish to write all the pending buffer.
Hm, but vivi.c does not use videbuf_waiton(). I guess this depends on how the frames are queued in the driver.
Cheers,
/ magnus
--
video4linux-list mailing list
Unsubscribe mailto:[email protected]?subject=unsubscribe
https://www.redhat.com/mailman/listinfo/video4linux-list
Hi Magnus,
Quoted 10 lines · Magnus Damm
Hi Matthieu,
On Mon, Jan 19, 2009 at 11:02 PM, Matthieu CASTET
Quoted 3 lines
But we didn't do
stop_capture, so as far I understand the controller is still writing data in memory. What prevent us to free the buffer we are writing.I have not looked into this in great detail, but isn't this handled by the videobuf state? The videobuf has state
VIDEOBUF_ACTIVEwhile it is in use. I don't think such a buffer is freed.
Well from my understanding form videobuf_queue_cancel [1], we call buf_release on all buffer.
Quoted 6 lines
Quoted 3 lines
I saw that
pxa_camerausevideobuf_waiton, before freeing the buffer. That seem more safe, but that mean we need to wait that controller finish to write all the pending buffer.Hm, but
vivi.cdoes not usevidebuf_waiton(). I guess this depends on how the frames are queued in the driver.
May be.
Matthieu
[1]
void videobuf_queue_cancel(struct videobuf_queue *q)
{
unsigned long flags = 0;
int i;
q->streaming = 0;
q->reading = 0;
wake_up_interruptible_sync(&q->wait);
/* remove queued buffers from list */
spin_lock_irqsave(q->irqlock, flags);
for (i = 0; i < VIDEO_MAX_FRAME; i++) {
if (NULL == q->bufs[i])
continue;
if (q->bufs[i]->state == VIDEOBUF_QUEUED) {
list_del(&q->bufs[i]->queue);
q->bufs[i]->state = VIDEOBUF_ERROR;
wake_up_all(&q->bufs[i]->done);
}
}
spin_unlock_irqrestore(q->irqlock, flags);
/* free all buffers + clear queue */
for (i = 0; i < VIDEO_MAX_FRAME; i++) {
if (NULL == q->bufs[i])
continue;
q->ops->buf_release(q, q->bufs[i]);
}
INIT_LIST_HEAD(&q->stream);
}--
video4linux-list mailing list
Unsubscribe mailto:[email protected]?subject=unsubscribe
https://www.redhat.com/mailman/listinfo/video4linux-list
Hi Matthieu,
[CC Morimoto-san]
[Changed list to linux-media]
Quoted 11 lines · Matthieu CASTET
Quoted 8 lines · Magnus Damm
On Mon, Jan 19, 2009 at 11:02 PM, Matthieu CASTET
Quoted 3 lines
But we didn't do
stop_capture, so as far I understand the controller is still writing data in memory. What prevent us to free the buffer we are writing.I have not looked into this in great detail, but isn't this handled by the videobuf state? The videobuf has state
VIDEOBUF_ACTIVEwhile it is in use. I don't think such a buffer is freed.Well from my understanding form
videobuf_queue_cancel[1], we callbuf_releaseon all buffer.
Yeah, you are correct. I guess waiting for the buffer before freeing is the correct way to do this. I guess vivi doesn't have to do this since it's not using DMA.
Morimoto-san, can you check the attached patch? I've tested it on my Migo-R board together with mplayer and it seems to work well here. I don't think using mplayer triggers this error case though, so maybe we should try some other application.
Cheers,
/ magnus
drivers/media/video/sh_mobile_ceu_camera.c +1 -0
@@ -150,6 +150,7 @@ static void free_buffer(struct videobuf_if (in_interrupt ())BUG (); +videobuf_waiton (&buf->vb,0 ,0 );videobuf_dma_contig_free (vq, &buf->vb);dev_dbg (&icd->dev,"%s freed\n" , __func__); buf->vb.state =VIDEOBUF_NEEDS_INIT ;
Hi Magnus
Quoted 4 lines
Morimoto-san, can you check the attached patch? I've tested it on my Migo-R board together with mplayer and it seems to work well here. I don't think using mplayer triggers this error case though, so maybe we should try some other application.
I checked this patch with following case.
Migo-R + ov772x + mplayer Migo-R + tw9910 + mplayer AP325 + ov772x + mplayer
It works well on all cases.
And I can agree with your opinion
"so maybe we should try some other application."
Best regards
--
Kuninori Morimoto
Hi Morimoto-san
Quoted 8 lines · On Mon, 16 Feb 2009, [email protected] wrote:
Hi Magnus
Quoted 4 lines
Morimoto-san, can you check the attached patch? I've tested it on my Migo-R board together with mplayer and it seems to work well here. I don't think using mplayer triggers this error case though, so maybe we should try some other application.
Quoted 7 lines
I checked this patch with following case.
Migo-R + ov772x + mplayer
Migo-R + tw9910 + mplayer
AP325 + ov772x + mplayerIt works well on all cases.
So I can add your "Tested-by:"?
Quoted 2 lines
And I can agree with your opinion
"so maybe we should try some other application."
You can try to trigger the race with the capture.c example. Reduce the "count" variable in mainloop() and run capture.c in a loop for a while... Try without this patch and then with this patch. But I think this is a correct fix, so, unless you say, it crashes without it and with it, I'll apply it.
Thanks
Guennadi
Quoted 9 lines
Best regards
--
Kuninori Morimoto
--
To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
---
Guennadi Liakhovetski, Ph.D.
Freelance Open-Source Software Developer
Quoted 19 lines · Magnus Damm
Hi Matthieu,
[CC Morimoto-san]
[Changed list to linux-media]Quoted 11 lines · Matthieu CASTET
Quoted 8 lines · Magnus Damm
On Mon, Jan 19, 2009 at 11:02 PM, Matthieu CASTET
Quoted 3 lines
But we didn't do
stop_capture, so as far I understand the controller is still writing data in memory. What prevent us to free the buffer we are writing.I have not looked into this in great detail, but isn't this handled by the videobuf state? The videobuf has state
VIDEOBUF_ACTIVEwhile it is in use. I don't think such a buffer is freed.Well from my understanding form
videobuf_queue_cancel[1], we callbuf_releaseon all buffer.
Quoted 8 lines
Yeah, you are correct. I guess waiting for the buffer before freeing is the correct way to do this. I guess vivi doesn't have to do this since it's not using DMA.
Morimoto-san, can you check the attached patch? I've tested it on my Migo-R board together with mplayer and it seems to work well here. I don't think using mplayer triggers this error case though, so maybe we should try some other application.
Magnus, can you, please, submit it with an Sob, after Morimoto-san has tested it with capture.c?
Thanks
Guennadi
---
Guennadi Liakhovetski, Ph.D.
Freelance Open-Source Software Developer
Dear Guennadi
Quoted 2 lines
Quoted 1 line
It works well on all cases.
So I can add your "Tested-by:"?
Yes please.
Quoted 5 lines
You can try to trigger the race with the
capture.cexample. Reduce the "count" variable inmainloop()and runcapture.cin a loop for a while... Try without this patch and then with this patch. But I think this is a correct fix, so, unless you say, it crashes without it and with it, I'll apply it.
I tried with following command option
# capture_example -d /dev/videoX -f -c 1000
I used -f option, I think you already know the reason =). I think -c 1000 is enough.
# Because tw9910 can not use YUYV,
# I fixed capture_example to use VUYU for tw9910.
sh_mobile_ceu_camera didn't crashe with
and without Magnus patch.
MigoR + ov772x + capture_example
MigoR + tw9910 + capture_example (VUYU fix)
AP325 + ov772x + capture_example
Best regards
--
Kuninori Morimoto
Quoted 13 lines · On Thu, 19 Feb 2009, [email protected] wrote:
Dear Guennadi
Quoted 2 lines
Quoted 1 line
It works well on all cases.
So I can add your "Tested-by:"?
Yes please.
Quoted 5 lines
You can try to trigger the race with the
capture.cexample. Reduce the "count" variable inmainloop()and runcapture.cin a loop for a while... Try without this patch and then with this patch. But I think this is a correct fix, so, unless you say, it crashes without it and with it, I'll apply it.
Quoted 5 lines
I tried with following command option
# capture_example -d /dev/videoX -f -c 1000I used -f option, I think you already know the reason =).
Yes, and I'm working on that.
Quoted 1 line
I think -c 1000 is enough.
No, sorry, this is not the test I meant. "-c" doesn't really stress the path we need. You really have to execute capture_example multiple times completely. The race we're trying to catch happens on STREAMOFF, and for that you have to run the example completely through. So, I would suggest something like
for (( i=0; i<100; i++ )); do capture_example -d /dev/videoX -f -c 4; done
Thanks
Guennadi
---
Guennadi Liakhovetski, Ph.D.
Freelance Open-Source Software Developer
Dear Guennadi
Quoted 7 lines
No, sorry, this is not the test I meant. "-c" doesn't really stress the path we need. You really have to execute
capture_examplemultiple times completely. The race we're trying to catch happens on STREAMOFF, and for that you have to run the example completely through. So, I would suggest something likefor (( i=0; i<100; i++ )); do
capture_example-d/dev/videoX-f -c 4; done
wow !!
sorry miss understanding.
OK. I re-tried test with your script (300 times).
sh_mobile_ceu_camera didn't crashe with and without Magnus patch.
MigoR + ov772x + capture_example MigoR + tw9910 + capture_example (VUYU fix) AP325 + ov772x + capture_example
Best regards
--
Kuninori Morimoto
Quoted 14 lines · On Thu, 19 Feb 2009, [email protected] wrote:
Dear Guennadi
Quoted 7 lines
No, sorry, this is not the test I meant. "-c" doesn't really stress the path we need. You really have to execute
capture_examplemultiple times completely. The race we're trying to catch happens on STREAMOFF, and for that you have to run the example completely through. So, I would suggest something likefor (( i=0; i<100; i++ )); do
capture_example-d/dev/videoX-f -c 4; donewow !!
sorry miss understanding.
Quoted 8 lines
OK. I re-tried test with your script (300 times).
sh_mobile_ceu_cameradidn't crashe with and without Magnus patch.MigoR + ov772x +
capture_exampleMigoR + tw9910 +capture_example(VUYU fix) AP325 + ov772x +capture_example
Hm, ok, maybe I can ask you about one more test, if you don't mind. The thing is, you only see the problem, if after the ->active buffer has been freed in free_buffer(), your DMA engine continues writing to the freed memory, but you only notice this, if some other driver manages to grab and use it in this time, then its data is going to be overwritten.
To actually see, if the race occurs, please do
drivers/media/video/sh_mobile_ceu_camera.c +1 -0
@@ -326,6 +326,7 @@spin_lock_irqsave (&pcdev->lock, flags); vb = pcdev->active; +WARN_ON (vb->state ==VIDEOBUF_NEEDS_INIT );list_del_init (&vb->queue);if (!list_empty (&pcdev->capture))
Also, I think, even better than restarting capture-example completely would be to edit capture-example.c and put a loop around
init_device(); start_capturing(); mainloop(); stop_capturing(); uninit_device();
or maybe even only
start_capturing(); mainloop(); stop_capturing();
Best would be to first try all three loops without the patch from Magnus and see if any of them triggers. And use a larger frame (maximum that your sensor can deliver) to give the DMA engine more time per frame.
Thanks
Guennadi
---
Guennadi Liakhovetski, Ph.D.
Freelance Open-Source Software Developer
Dear Guennadi
Quoted 5 lines
Hm, ok, maybe I can ask you about one more test, if you don't mind. The thing is, you only see the problem, if after the ->active buffer has been freed in
free_buffer(), your DMA engine continues writing to the freed memory, but you only notice this, if some other driver manages to grab and use it in this time, then its data is going to be overwritten.
(snip)
Quoted 3 lines
Best would be to first try all three loops without the patch from Magnus and see if any of them triggers. And use a larger frame (maximum that your sensor can deliver) to give the DMA engine more time per frame.
Hmm. maybe I could understand...
OK.
I tested with MigoR + ov772x + following way.
1) apply patch to kernel
drivers/media/video/sh_mobile_ceu_camera.c line 332
Quoted 11 lines
diff -u
a/drivers/media/video/sh_mobile_ceu_camera.cb/drivers/media/video/sh_mobile_ceu_camera.cdrivers/media/video/sh_mobile_ceu_camera.c +1
-0@@ -326,6 +326,7 @@spin_lock_irqsave (&pcdev->lock, flags); vb = pcdev->active; +WARN_ON (vb->state ==VIDEOBUF_NEEDS_INIT );list_del_init (&vb->queue);if (!list_empty (&pcdev->capture))
2) apply patch to capture-example
I think your wanting to say is this...
capture_example.c.test renamed from capture_example.c +10 -5
@@ -590,6 +590,8 @@ int main(int argc, char **argv) { + int i; + dev_name = "/dev/video0"; for (;;) {
@@ -648,11 +650,14 @@ } open_device(); - init_device(); - start_capturing(); - mainloop(); - stop_capturing(); - uninit_device(); + for (i=0 ; i<100 ; i++) { + printf("%d ", i); + init_device(); + start_capturing(); + mainloop(); + stop_capturing(); + uninit_device(); + } close_device(); fprintf(stderr, "\n"); return 0;
3) test this capture_example.
Then, It say
======================================================================
camera 0-0: SuperH Mobile CEU driver attached to camera 0
camera_host0: Format 0 not found
.0 ... camera_host0: Format 0 not found
------------[ cut here ]------------ Badness at 8c141b9e [verbose debug info unavailable]
Pid : 869, Comm: c2
CPU : 0 Not tainted (2.6.29-rc4-00197-g41480ae-dirty #801)
PC is at sh_mobile_ceu_irq+0x2a/0xc8
PR is at handle_IRQ_event+0x2e/0x68
PC : 8c141b9e SP : 8fe15e40 SR : 400001f1 TEA : 2961ba40 R0 : 8ff05afc R1 : 00000000 R2 : 8fe14000 R3 : 00000000 R4 : 8ff05a00 R5 : 8ff05a00 R6 : 00000010 R7 : 00000000 R8 : 8ff416e0 R9 : 000000f0 R10 : 00000000 R11 : 00000034 R12 : 8ff414d0 R13 : 8ff3a1e0 R14 : 8ff89e74
MACH: 00000000 MACL: 00000c30 GBR : 29686450 PR : 8c030e0a
Call trace: [<8c030e0a>] handle_IRQ_event+0x2e/0x68 [<8c0326a8>] handle_level_irq+0x6c/0x100 [<8c00356c>] do_IRQ+0x34/0x5c [<8c007118>] ret_from_irq+0x0/0x10 [<8c003538>] do_IRQ+0x0/0x5c [<8c0f2c44>] memset+0x30/0x4c [<8c00b25a>] dma_alloc_coherent+0x5e/0x148 [<8c141742>] __videobuf_mmap_mapper+0xbe/0x1a4 [<8c13fb88>] videobuf_mmap_mapper+0x3c/0x70 [<8c050ad8>] mmap_region+0x184/0x34c [<8c005234>] old_mmap+0x60/0xa8 [<8c007266>] syscall_call+0xc/0x10 [<8c0051d4>] old_mmap+0x0/0xa8
camera 0-0: SuperH Mobile CEU driver detached from camera 0 Trace/breakpoint trap
======================================================================
4) apply Magnus patch
5) and test it again
then It say...
========================================================
-bash-3.00# ./c2 -d /dev/video0 -f -c 4
camera 0-0: SuperH Mobile CEU driver attached to camera 0
camera_host0: Format 0 not found
.0 ... camera_host0: Format 0 not found
.1 ... camera_host0: Format 0 not found
.2 ... camera_host0: Format 0 not found
.3 ... camera_host0: Format 0 not found
(snip)
.96 ... camera_host0: Format 0 not found
.97 ... camera_host0: Format 0 not found
.98 ... camera_host0: Format 0 not found
.99 ...camera 0-0: SuperH Mobile CEU driver detached from camera 0
========================================================
Best regards
--
Kuninori Morimoto
Quoted 92 lines · On Thu, 19 Feb 2009, [email protected] wrote:
Dear Guennadi
Quoted 5 lines
Hm, ok, maybe I can ask you about one more test, if you don't mind. The thing is, you only see the problem, if after the ->active buffer has been freed in
free_buffer(), your DMA engine continues writing to the freed memory, but you only notice this, if some other driver manages to grab and use it in this time, then its data is going to be overwritten.(snip)
Quoted 3 lines
Best would be to first try all three loops without the patch from Magnus and see if any of them triggers. And use a larger frame (maximum that your sensor can deliver) to give the DMA engine more time per frame.
Hmm. maybe I could understand...
OK.
I tested with MigoR + ov772x + following way.1) apply patch to kernel
drivers/media/video/sh_mobile_ceu_camera.c line 332
Quoted 11 lines
diff -u
a/drivers/media/video/sh_mobile_ceu_camera.cb/drivers/media/video/sh_mobile_ceu_camera.cdrivers/media/video/sh_mobile_ceu_camera.c +1
-0@@ -326,6 +326,7 @@spin_lock_irqsave (&pcdev->lock, flags); vb = pcdev->active; +WARN_ON (vb->state ==VIDEOBUF_NEEDS_INIT );list_del_init (&vb->queue);if (!list_empty (&pcdev->capture))2) apply patch to capture-example
I think your wanting to say is this...
diff -u
capture_example.ccapture_example.c.testcapture_example.c.test renamed from capture_example.c +10
-5@@ -590,6 +590,8 @@ int main(int argc, char **argv) { + int i; + dev_name = "/dev/video0"; for (;;) {@@ -648,11 +650,14 @@ } open_device(); - init_device(); - start_capturing(); - mainloop(); - stop_capturing(); - uninit_device(); + for (i=0 ; i<100 ; i++) { + printf("%d ", i); + init_device(); + start_capturing(); + mainloop(); + stop_capturing(); + uninit_device(); + } close_device(); fprintf(stderr, "\n"); return 0;3) test this
capture_example.Then, It say
======================================================================
camera 0-0: SuperH Mobile CEU driver attached to camera 0
camera_host0: Format 0 not found
.0 ...camera_host0: Format 0 not found------------[ cut here ]------------ Badness at 8c141b9e [verbose debug info unavailable]Pid : 869, Comm: c2
CPU : 0 Not tainted (2.6.29-rc4-00197-g41480ae-dirty #801)PC is at
sh_mobile_ceu_irq+0x2a/0xc8
PR is athandle_IRQ_event+0x2e/0x68
PC :8c141b9eSP :8fe15e40SR :400001f1TEA :2961ba40R0 :8ff05afcR1 : 00000000 R2 :8fe14000R3 : 00000000 R4 :8ff05a00R5 :8ff05a00R6 : 00000010 R7 : 00000000 R8 :8ff416e0R9 :000000f0R10 : 00000000 R11 : 00000034 R12 :8ff414d0R13 :8ff3a1e0R14 :8ff89e74
MACH: 00000000 MACL:00000c30GBR : 29686450 PR :8c030e0a
Quoted 14 lines
Call trace: [<8c030e0a>] handle_IRQ_event+0x2e/0x68 [<8c0326a8>] handle_level_irq+0x6c/0x100 [<8c00356c>] do_IRQ+0x34/0x5c [<8c007118>] ret_from_irq+0x0/0x10 [<8c003538>] do_IRQ+0x0/0x5c [<8c0f2c44>] memset+0x30/0x4c [<8c00b25a>] dma_alloc_coherent+0x5e/0x148 [<8c141742>] __videobuf_mmap_mapper+0xbe/0x1a4 [<8c13fb88>] videobuf_mmap_mapper+0x3c/0x70 [<8c050ad8>] mmap_region+0x184/0x34c [<8c005234>] old_mmap+0x60/0xa8 [<8c007266>] syscall_call+0xc/0x10 [<8c0051d4>] old_mmap+0x0/0xa8
Yesss! There you go - this is the race we are hunting.
Quoted 9 lines
camera 0-0: SuperH Mobile CEU driver detached from camera 0 Trace/breakpoint trap
======================================================================4) apply Magnus patch
5) and test it againthen It say...
Quoted 14 lines
========================================================
-bash-3.00# ./c2 -d/dev/video0-f -c 4
camera 0-0: SuperH Mobile CEU driver attached to camera 0
camera_host0: Format 0 not found
.0 ...camera_host0: Format 0 not found
.1 ...camera_host0: Format 0 not found
.2 ...camera_host0: Format 0 not found
.3 ...camera_host0: Format 0 not found
(snip)
.96 ...camera_host0: Format 0 not found
.97 ...camera_host0: Format 0 not found
.98 ...camera_host0: Format 0 not found
.99 ...camera 0-0: SuperH Mobile CEU driver detached from camera 0
========================================================
Nice and clean. Good, now I just need a formal patch submission with a "Tested-by" string, please.
Thanks
Guennadi
---
Guennadi Liakhovetski, Ph.D.
Freelance Open-Source Software Developer
[email protected]
Tested-byDear Guennadi
Quoted 5 lines
Quoted 5 lines
------------[ cut here ]------------ Badness at 8c141b9e [verbose debug info unavailable]Pid : 869, Comm: c2
CPU : 0 Not tainted (2.6.29-rc4-00197-g41480ae-dirty #801)
(snip)
Quoted 1 line
Yesss! There you go - this is the race we are hunting.
(snip)
Quoted 2 lines
Nice and clean. Good, now I just need a formal patch submission with a "Tested-by" string, please.
Fuuu. good !! Thank you.
Best regards
--
Kuninori Morimoto