RE: cdns3 uvc first ISO parkage lost problem

From: Pawel Laszczak
Date: Mon Nov 06 2023 - 06:12:47 EST


>
>On Mon, Oct 30, 2023 at 06:34:48AM +0000, Pawel Laszczak wrote:
>>
>> >
>> >hi Pawel Laszczak
>> >
>> >Recently, I met the problem when use uvc. UVC report jpg header error.
>> >
>> >Basic reproduce steps.
>> >Gadget side:
>> >1 -
>> > https://urldefense.com/v3/__https://gist.github.com/kbingham/c39c
>>
>>4cc7c20882a104c08df5206e2f9f?permalink_comment_id=3270713__;!!EHsc
>m
>>
>>S1ygiU1lA!H1h8GlLnbS6vqklXm_2qGyinP638O62Kk2eLB9zeMkjAGXUAPYyPXD
>L
>> >FasSqYt16xq0RGT0Ff-cP4A$
>> > uvc-gadget.sh start
>> >2 -
>> > https://urldefense.com/v3/__https://git.ideasonboard.org/uvc-
>>
>>gadget.git__;!!EHscmS1ygiU1lA!H1h8GlLnbS6vqklXm_2qGyinP638O62Kk2eLB
>9z
>> >eMkjAGXUAPYyPXDLFasSqYt16xq0RGT1ogOdRQA$
>> > uvc-gadget -i test.jpg
>> >
>> >
>> >Host side:
>> > https://urldefense.com/v3/__https://github.com/thekvs/uvccapture2
>>
>>__;!!EHscmS1ygiU1lA!H1h8GlLnbS6vqklXm_2qGyinP638O62Kk2eLB9zeMkjAG
>X
>> >UAPYyPXDLFasSqYt16xq0RGT1MNlKiXA$
>> > uvccapture2 --device /dev/video0 --resolution 640x360 --count 1 --
>> >result 8qxp.jpeg
>> >
>> > It will report jpeg header error.
>> >
>> >
>> >After debugs, I found two problem.
>> >
>> >Problem 1, sg is enabled. so uvc driver will use sg. each package
>> >include two trb, trb0 is 8bytes header, trb1 is 1016bytes. total 1024.
>> >
>> >num_trb here is wrong.
>> >it should be
>> > num_trb = priv_ep->interval * request->num_mapped_sgs.
>> >
>> >because priv_ep->interval is 1, I just simple set to
>> >request->num_mapped_sg as below patch. USB analyer show one whole
>> >1024 ISO package sent out as expectation although document said only
>> >support one TD when use ISO (Maybe my doc is too old).
>>
>> Support for sg in uvc has been added after upstreaming this driver,
>> so the driver needs some improvement.
>>
>> Calculating of num_trb probably will more complicated change.
>>
>> You can see how it is implemented in
>>
>https://urldefense.com/v3/__https://elixir.bootlin.com/linux/latest/source/d
>rivers/usb/gadget/udc/cdns2/cdns2-
>gadget.c*L412__;Iw!!EHscmS1ygiU1lA!EZS2StTKnSzdCT7N5B1-
>l8nGXQgS63KwgcGINcpBC8rnRJu2u8ryV1UjwQb6YfwYLPq9T_115KC5qQ$ .
>>
>> CDNS2 is different controller and support only HS but has borrowed the
>DMA part from CDNS3.
>> It was upsteamed after adding sg to UVC.
>>
>> Regarding TD, it is true that controller can support only one TD per
>> SOF but this TD can contain many TRBs
>
>Okay, great. I can work a patch if I can resolve problem 2.

At this moment I don't know how to resolve the problem 2.
I'm going to recreate this case on my side and try to find some solution.

Pawel

>
>>
>> >
>> >diff --git a/drivers/usb/cdns3/cdns3-gadget.c
>> >b/drivers/usb/cdns3/cdns3- gadget.c index
>> >69a44bd7e5d02..8cc99a885883f 100644
>> >--- a/drivers/usb/cdns3/cdns3-gadget.c
>> >+++ b/drivers/usb/cdns3/cdns3-gadget.c
>> >@@ -1125,10 +1125,7 @@ static int cdns3_ep_run_transfer(struct
>> >cdns3_endpoint *priv_ep,
>> > struct scatterlist *s = NULL;
>> > bool sg_supported = !!(request->num_mapped_sgs);
>> >
>> >- if (priv_ep->type == USB_ENDPOINT_XFER_ISOC)
>> >- num_trb = priv_ep->interval;
>> >- else
>> >- num_trb = sg_supported ? request->num_mapped_sgs : 1;
>> >+ num_trb = sg_supported ? request->num_mapped_sgs : 1;
>> >
>> > if (num_trb > priv_ep->free_trbs) {
>> > priv_ep->flags |= EP_RING_FULL;
>> >
>> >
>> >*** Problem 2 ***
>> >
>> >According to doc and my observation, looks like hardware fetch data
>> >into FIFO when get SOF, then transfer data when get IN token. Each
>> >SOF will increase TRB regardless it is ready or not.
>>
>> Yes, but this fetched data will be sent in next ITP.
>>
>> >
>> >When gadget complete equeue ISO data, so SOF will increase TRB
>> >regardless if there are IN token.
>> >
>> > SOF SOF SOF SOF IN SOF ....
>> > TRB0 TRB1 TRB2 TRB3 ...
>> >
>> >
>> >Host may start get data at some time after gadget queue data.
>> >
>> >So TRB0..2 data will be lost.
>> >
>> >If it is audio data, it should be okay. But for uvc, it is jpeg
>> >header, so host side report error.
>> >
>> >I checked dwc gadget driver, which start equeue ISO data only get NYET.
>> >
>> >I check cdns spec, there are ISOERR. But it is never happen.
>> >According to document, ISOERR should issue when IN token and FIFO no
>data.
>> >
>>
>> Current CDNS3 driver has disabled ISOERR. Did you enable it?
>
>Yes, I enabled all IRQ.
>
>+ if (priv_ep->type == USB_ENDPOINT_XFER_ISOC && priv_ep->dir) {
>+ priv_ep->flags |= EP_QUIRK_ISO_IN_NOST;
>+ reg |= 0xFFFF;
>+ }
>
>Supposed ISOERR should happen even DMA is disabled.
>But I also tried enable DMA, and using lenght 0 TRB and link to loop.
>
>Still no ISOERR happen. I can see TRBADDR changed, but still no ISOERR
>
>
> irq/447-5b13000-200 [000] d..1. 78.662729: cdns3_epx_irq: IRQ for ep2in:
>00000804 IOC , ep_traddr: c0086018 ep_last_sid: 00000000 use_streams: 0
>
> ^^^^^^^
> irq/447-5b13000-200 [000] d..1. 78.662851: cdns3_epx_irq: IRQ for ep2in:
>00000804 IOC , ep_traddr: c008600c ep_last_sid: 00000000 use_streams: 0
> irq/447-5b13000-200 [000] d..1. 78.662975: cdns3_epx_irq: IRQ for ep2in:
>00000804 IOC , ep_traddr: c0086018 ep_last_sid: 00000000 use_streams: 0
>
>STS is 0x804, only IOC set.
>
>Frank
>
>>
>> >I tried below method
>> > 1. Delay queue TRB, but no ISOERR.
>> > 2. queue a lenght 0 TRB,but no ISOERR
>> >
>> >My question is how to delay queue TRB to ISO IN token really happen
>> >to avoid lost JPEG header.
>> >
>> >Frank
>> >
>> >
>> >
>> >
>> >
>>