cx23885 - regression between 4.17.x and 4.18.x

cx23885 - regression between 4.17.x and 4.18.x

[David R]

Hi

Just a heads up. I'm having to revert cx23885-core.c to the 4.17 version
to obtain stability with my old AMD Phenom/ASUS M4A785TD and Hauppauge
WinTV-HVR-5525. The latest code drops out and refuses to return video
streams in hours or a few days max. The 4.17 version is fine and stable
over weeks/months.

04:00.0 Multimedia video controller: Conexant Systems, Inc. CX23887/8
PCIe Broadcast Audio and Video Decoder with 3D Comb (rev 04)
    Subsystem: Hauppauge computer works Inc. Device f038
    Control: I/O- Mem+ BusMaster+ SpecCycle- MemWINV- VGASnoop- ParErr-
Stepping- SERR+ FastB2B- DisINTx-
    Status: Cap+ 66MHz- UDF- FastB2B- ParErr- DEVSEL=fast >TAbort-
<TAbort- <MAbort- >SERR- <PERR- INTx-
    Latency: 0, Cache Line Size: 64 bytes
    Interrupt: pin A routed to IRQ 17
    Region 0: Memory at fe800000 (64-bit, non-prefetchable) [size=2M]
    Capabilities: <access denied>
    Kernel driver in use: cx23885

Cheers
David

Re: cx23885 - regression between 4.17.x and 4.18.x

[Vincent McIntyre]

On Thu, Sep 13, 2018 at 06:39:57PM +0100, David R wrote:
> Hi
> 
> Just a heads up. I'm having to revert cx23885-core.c to the 4.17 version
> to obtain stability with my old AMD Phenom/ASUS M4A785TD and Hauppauge
> WinTV-HVR-5525. The latest code drops out and refuses to return video
> streams in hours or a few days max. The 4.17 version is fine and stable
> over weeks/months.
> 
> 04:00.0 Multimedia video controller: Conexant Systems, Inc. CX23887/8
> PCIe Broadcast Audio and Video Decoder with 3D Comb (rev 04)
>     Subsystem: Hauppauge computer works Inc. Device f038
>     Control: I/O- Mem+ BusMaster+ SpecCycle- MemWINV- VGASnoop- ParErr-
> Stepping- SERR+ FastB2B- DisINTx-
>     Status: Cap+ 66MHz- UDF- FastB2B- ParErr- DEVSEL=fast >TAbort-
> <TAbort- <MAbort- >SERR- <PERR- INTx-
>     Latency: 0, Cache Line Size: 64 bytes
>     Interrupt: pin A routed to IRQ 17
>     Region 0: Memory at fe800000 (64-bit, non-prefetchable) [size=2M]
>     Capabilities: <access denied>
>     Kernel driver in use: cx23885

Interesting. cx88 also seems to have regressed.
Are you able to give a git commit range for what works & doesn't work?

I asked Adam to try building the modules with media-build
but have not heard anything further.

    Date: Sat, 8 Sep 2018 20:49:22 -0400
    From: Adam Stylinski <kungfujesus06@gmail.com>
    To: linux-media@vger.kernel.org
    Subject: 4.18 regression

    Hello,

    I haven't done a thorough bisection of kernel revisions, but moving from kernel 4.17.19
    to 4.18.6 results in mythtv being unable to tune in any channel with a pic hdtv 5500
    tuner (cx88 based device with an LG frontend). I get an error back from the channel
    scanner about not being able to measure signal strength and getting back an error unknown
    (errno 254).

    I was able to use dvbtools with get-atsc to get a channel, but I don't think any of the
    forward error correction was applied to it.

    Let me know if you need more details.

Vince

Re: cx23885 - regression between 4.17.x and 4.18.x

[David R]

Any bisection would take ages, as I need to run for approaching a week
to be sure. There are only 3 commits that are possible.

commit 95f408bbc4e4d46c75da3d2558dd835e56ac7720
Author: Brad Love <brad@nextdimension.cc>
Date:   Tue May 8 17:20:18 2018 -0400

    media: cx23885: Ryzen DMA related RiSC engine stall fixes

commit 3b8315f37d6eaa36f0f2e484eabaef03a7bd1eff
Author: Brad Love <brad@nextdimension.cc>
Date:   Tue May 8 17:20:17 2018 -0400

    media: cx23885: Use PCI and TS masks in irq functions

commit 9a7dc2b064ef7477d4c3a477f4de0a44b3a40cbd
Author: Brad Love <brad@nextdimension.cc>
Date:   Tue May 8 17:20:16 2018 -0400

    media: cx23885: Handle additional bufs on interrupt

Cheers
David

On 15/09/18 01:42, Vincent McIntyre wrote:
> On Thu, Sep 13, 2018 at 06:39:57PM +0100, David R wrote:
>> Hi
>>
>> Just a heads up. I'm having to revert cx23885-core.c to the 4.17 version
>> to obtain stability with my old AMD Phenom/ASUS M4A785TD and Hauppauge
>> WinTV-HVR-5525. The latest code drops out and refuses to return video
>> streams in hours or a few days max. The 4.17 version is fine and stable
>> over weeks/months.
>>
>> 04:00.0 Multimedia video controller: Conexant Systems, Inc. CX23887/8
>> PCIe Broadcast Audio and Video Decoder with 3D Comb (rev 04)
>>     Subsystem: Hauppauge computer works Inc. Device f038
>>     Control: I/O- Mem+ BusMaster+ SpecCycle- MemWINV- VGASnoop- ParErr-
>> Stepping- SERR+ FastB2B- DisINTx-
>>     Status: Cap+ 66MHz- UDF- FastB2B- ParErr- DEVSEL=fast >TAbort-
>> <TAbort- <MAbort- >SERR- <PERR- INTx-
>>     Latency: 0, Cache Line Size: 64 bytes
>>     Interrupt: pin A routed to IRQ 17
>>     Region 0: Memory at fe800000 (64-bit, non-prefetchable) [size=2M]
>>     Capabilities: <access denied>
>>     Kernel driver in use: cx23885
> Interesting. cx88 also seems to have regressed.
> Are you able to give a git commit range for what works & doesn't work?
>
> I asked Adam to try building the modules with media-build
> but have not heard anything further.
>
>     Date: Sat, 8 Sep 2018 20:49:22 -0400
>     From: Adam Stylinski <kungfujesus06@gmail.com>
>     To: linux-media@vger.kernel.org
>     Subject: 4.18 regression
>
>     Hello,
>
>     I haven't done a thorough bisection of kernel revisions, but moving from kernel 4.17.19
>     to 4.18.6 results in mythtv being unable to tune in any channel with a pic hdtv 5500
>     tuner (cx88 based device with an LG frontend). I get an error back from the channel
>     scanner about not being able to measure signal strength and getting back an error unknown
>     (errno 254).
>
>     I was able to use dvbtools with get-atsc to get a channel, but I don't think any of the
>     forward error correction was applied to it.
>
>     Let me know if you need more details.
>
> Vince

Re: cx23885 - regression between 4.17.x and 4.18.x

[Adam Stylinski]

[-- Attachment #1: Type: text/plain, Size: 3620 bytes --]

On Tue, Sep 18, 2018 at 09:13:04AM -0400, Adam Stylinski wrote:
> On Sat, Sep 15, 2018 at 09:30:30AM +0100, David R wrote:
> > Any bisection would take ages, as I need to run for approaching a week
> > to be sure. There are only 3 commits that are possible.
> > 
> > commit 95f408bbc4e4d46c75da3d2558dd835e56ac7720
> > Author: Brad Love <brad@nextdimension.cc>
> > Date:   Tue May 8 17:20:18 2018 -0400
> > 
> >     media: cx23885: Ryzen DMA related RiSC engine stall fixes
> > 
> > commit 3b8315f37d6eaa36f0f2e484eabaef03a7bd1eff
> > Author: Brad Love <brad@nextdimension.cc>
> > Date:   Tue May 8 17:20:17 2018 -0400
> > 
> >     media: cx23885: Use PCI and TS masks in irq functions
> > 
> > commit 9a7dc2b064ef7477d4c3a477f4de0a44b3a40cbd
> > Author: Brad Love <brad@nextdimension.cc>
> > Date:   Tue May 8 17:20:16 2018 -0400
> > 
> >     media: cx23885: Handle additional bufs on interrupt
> > 
> > Cheers
> > David
> > 
> > On 15/09/18 01:42, Vincent McIntyre wrote:
> > > On Thu, Sep 13, 2018 at 06:39:57PM +0100, David R wrote:
> > >> Hi
> > >>
> > >> Just a heads up. I'm having to revert cx23885-core.c to the 4.17 version
> > >> to obtain stability with my old AMD Phenom/ASUS M4A785TD and Hauppauge
> > >> WinTV-HVR-5525. The latest code drops out and refuses to return video
> > >> streams in hours or a few days max. The 4.17 version is fine and stable
> > >> over weeks/months.
> > >>
> > >> 04:00.0 Multimedia video controller: Conexant Systems, Inc. CX23887/8
> > >> PCIe Broadcast Audio and Video Decoder with 3D Comb (rev 04)
> > >>     Subsystem: Hauppauge computer works Inc. Device f038
> > >>     Control: I/O- Mem+ BusMaster+ SpecCycle- MemWINV- VGASnoop- ParErr-
> > >> Stepping- SERR+ FastB2B- DisINTx-
> > >>     Status: Cap+ 66MHz- UDF- FastB2B- ParErr- DEVSEL=fast >TAbort-
> > >> <TAbort- <MAbort- >SERR- <PERR- INTx-
> > >>     Latency: 0, Cache Line Size: 64 bytes
> > >>     Interrupt: pin A routed to IRQ 17
> > >>     Region 0: Memory at fe800000 (64-bit, non-prefetchable) [size=2M]
> > >>     Capabilities: <access denied>
> > >>     Kernel driver in use: cx23885
> > > Interesting. cx88 also seems to have regressed.
> > > Are you able to give a git commit range for what works & doesn't work?
> > >
> > > I asked Adam to try building the modules with media-build
> > > but have not heard anything further.
> > >
> > >     Date: Sat, 8 Sep 2018 20:49:22 -0400
> > >     From: Adam Stylinski <kungfujesus06@gmail.com>
> > >     To: linux-media@vger.kernel.org
> > >     Subject: 4.18 regression
> > >
> > >     Hello,
> > >
> > >     I haven't done a thorough bisection of kernel revisions, but moving from kernel 4.17.19
> > >     to 4.18.6 results in mythtv being unable to tune in any channel with a pic hdtv 5500
> > >     tuner (cx88 based device with an LG frontend). I get an error back from the channel
> > >     scanner about not being able to measure signal strength and getting back an error unknown
> > >     (errno 254).
> > >
> > >     I was able to use dvbtools with get-atsc to get a channel, but I don't think any of the
> > >     forward error correction was applied to it.
> > >
> > >     Let me know if you need more details.
> > >
> > > Vince
> > 
> > 
> 

For me the issue with my cx88 card happens immediately, most of the commands to the card seem to fail.  Unfortunately I build my kernel with this drivers builtin as opposed to modules, so for me to use the 4.17 media branch with 4.18 I'll need to change my kernel config.  Any instructions on how I should bisect with this?


[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

cx23885 - regression between 4.17.x and 4.18.x

[Markus Dobel]

Hi,

First of all, sorry for not correctly replying to the thread. I just 
subscribed to the list yesterday, so I do not have the message in my 
inbox.

I am referring to this message: 
https://www.spinics.net/lists/linux-media/msg140286.html

My system also suffered from the problem that, with Kernel 4.18 my DVB 
card became unresponsive after a few hours. I am using tvheadend as a 
video recorder, and with Kernel 4.18, tvheadend started missing 
recordings, with log messages like the following:

   linuxdvb: Silicon Labs Si2168 : DVB-C #1 - poll TIMEOUT
   subscription: 0163: service instance is bad, reason: No input detected
   subscription: 0162: service instance is bad, reason: No input detected
   subscription: 0163: No input source available for subscription "DVR: 
..." to channel "..."

In this state, the DVB card was not usable until unloading the cx23885 
module and reloading it -- just to fail again few hours later.

Also, I noticed log messages like these, not directly correlatable to 
the failures (not at the same time as the failed recordings, and 
sometimes also working recordings after such a log message).

   kernel: cx23885 0000:02:00.0: dma in progress detected 0x00000001 
0x00000001, clearing

Nevertheless, the patch introducing the log message seems also to cause 
the failure:

   [PATCH 3/5] cx23885: Ryzen DMA related RiSC engine stall fixes
   https://www.spinics.net/lists/linux-media/msg133899.html

I never suffered the problem the patch is trying to solve (card used to 
work flawlessly for months). To see if that patch is not only the 
messenger, but also the culprit, I built a custom kernel based on the 
kernel-4.18.9-200.fc28.x86_64 source RPM from Fedora 28, only applying 
the patch above in reverse -- and did not have a single failure for > 48 
hours now.

Since the author tried to solve a problem, and maybe even did for his 
system, I guess reversing the patch alone is not a proper solution for 
everyone. And since the author did submit this patch, I guess he also 
did not suffer from the problem I have right now... so a 
fits-for-all-solution might be harder to find, also because the failure 
is not immediate.

Unfortunately I don't know enough about the code in question (or kernel 
development as whole), so I have no clue how to analyze (or even fix) 
this any further. What I can offer is to test modified versions of the 
patch above on my computer, as it does not seem to affect everyone.

Some information on my system:

- HP Microserver G7 N54L with AMD Turion II Neo CPU
- DVBSky T982 PCI-E card with dual DVB-T/C tuners, identifying as
   02:00.0 Multimedia video controller: Conexant Systems, Inc. CX23885 
PCI Video and Audio Decoder (rev 04)
	Subsystem: DVBSky T982
	Flags: bus master, fast devsel, latency 0, IRQ 16, NUMA node 0
	Memory at fe600000 (64-bit, non-prefetchable) [size=2M]
	Capabilities: <access denied>
	Kernel driver in use: cx23885
	Kernel modules: cx23885
- Fedora 28, currently with a custom kernel, usually with the most 
recent "stock" fedora kernel.
- kernel-4.18.9-200.fixdvb.fc28.x86_64 ("fixdvb" being my local 
extraversion, only difference to the original kernel: the reversed patch 
above)
- tvheadend-4.2.6-1.fc28.x86_64

Regards,
Markus