[RFC] mn88472: reduce firmware download chunk size
diff mbox

Message ID 1424337200-6446-1-git-send-email-a.seppala@gmail.com
State New, archived
Headers show

Commit Message

Antti Seppälä Feb. 19, 2015, 9:13 a.m. UTC
It seems that currently the firmware download on the mn88472 is
somehow wrong for my Astrometa HD-901T2.

Reducing the download chunk size (mn88472_config.i2c_wr_max) to 2 
makes the firmware download consistently succeed.

Any larger value causes the download to always fail:

[    7.671482] mn88472 7-0018: downloading firmware from file 'dvb-demod-mn88472-02.fw'
[    8.206960] mn88472 7-0018: firmware download failed=-32
[    8.208610] rtl2832 7-0010: i2c reg write failed -32
[    8.208620] r820t 8-003a: r820t_write: i2c wr failed=-32 reg=05 len=1: 83
[    8.210459] rtl2832 7-0010: i2c reg write failed -32
[    8.212038] rtl2832 7-0010: i2c reg write failed -32

I'm obviously not too happy about this patch as it slows down the
firmware download but I have not found a way to keep larger chunks in
place and have a working firmware download at the same time.

Signed-off-by: Antti Seppälä <a.seppala@gmail.com>
---
 drivers/media/usb/dvb-usb-v2/rtl28xxu.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

Comments

Benjamin Larsson Feb. 19, 2015, 9:43 a.m. UTC | #1
On 2015-02-19 10:13, Antti Seppälä wrote:
> It seems that currently the firmware download on the mn88472 is
> somehow wrong for my Astrometa HD-901T2.
>
> Reducing the download chunk size (mn88472_config.i2c_wr_max) to 2
> makes the firmware download consistently succeed.
>


Hi, try adding the workaround patch I sent for this.

[PATCH 1/3] rtl28xxu: lower the rc poll time to mitigate i2c transfer errors

I now see that it hasn't been merged. But I have been running with this 
patch for a few months now without any major issues.

MvH
Benjamin Larsson
--
To unsubscribe from this list: send the line "unsubscribe linux-media" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Antti Seppälä Feb. 19, 2015, 10:21 a.m. UTC | #2
On 19 February 2015 at 11:43, Benjamin Larsson <benjamin@southpole.se> wrote:
> On 2015-02-19 10:13, Antti Seppälä wrote:
>>
>> It seems that currently the firmware download on the mn88472 is
>> somehow wrong for my Astrometa HD-901T2.
>>
>> Reducing the download chunk size (mn88472_config.i2c_wr_max) to 2
>> makes the firmware download consistently succeed.
>>
>
>
> Hi, try adding the workaround patch I sent for this.
>
> [PATCH 1/3] rtl28xxu: lower the rc poll time to mitigate i2c transfer errors
>
> I now see that it hasn't been merged. But I have been running with this
> patch for a few months now without any major issues.
>

The patch really did improve firmware loading. Weird...

Even with it I still get occasional i2c errors from r820t:

[   15.874402] r820t 8-003a: r820t_write: i2c wr failed=-32 reg=0a len=1: da
[   81.455517] r820t 8-003a: r820t_read: i2c rd failed=-32 reg=00
len=4: 69 74 e6 df
[   99.949702] r820t 8-003a: r820t_read: i2c rd failed=-32 reg=00
len=4: 69 74 e6 df

These errors seem to appear more often if I'm reading the signal
strength values using e.g. femon.

Br,
-Antti
--
To unsubscribe from this list: send the line "unsubscribe linux-media" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Antti Palosaari Feb. 19, 2015, 3:38 p.m. UTC | #3
On 02/19/2015 12:21 PM, Antti Seppälä wrote:
> On 19 February 2015 at 11:43, Benjamin Larsson <benjamin@southpole.se> wrote:
>> On 2015-02-19 10:13, Antti Seppälä wrote:
>>>
>>> It seems that currently the firmware download on the mn88472 is
>>> somehow wrong for my Astrometa HD-901T2.
>>>
>>> Reducing the download chunk size (mn88472_config.i2c_wr_max) to 2
>>> makes the firmware download consistently succeed.
>>>
>>
>>
>> Hi, try adding the workaround patch I sent for this.
>>
>> [PATCH 1/3] rtl28xxu: lower the rc poll time to mitigate i2c transfer errors
>>
>> I now see that it hasn't been merged. But I have been running with this
>> patch for a few months now without any major issues.
>>
>
> The patch really did improve firmware loading. Weird...
>
> Even with it I still get occasional i2c errors from r820t:
>
> [   15.874402] r820t 8-003a: r820t_write: i2c wr failed=-32 reg=0a len=1: da
> [   81.455517] r820t 8-003a: r820t_read: i2c rd failed=-32 reg=00
> len=4: 69 74 e6 df
> [   99.949702] r820t 8-003a: r820t_read: i2c rd failed=-32 reg=00
> len=4: 69 74 e6 df
>
> These errors seem to appear more often if I'm reading the signal
> strength values using e.g. femon.

Could you disable whole IR polling and test
modprobe dvb_usb_v2 disable_rc_polling=1

It is funny that *increasing* RC polling makes things better, though...

regards
Antti
Antti Seppälä Feb. 19, 2015, 4:01 p.m. UTC | #4
On 19 February 2015 at 17:38, Antti Palosaari <crope@iki.fi> wrote:
>
>
> On 02/19/2015 12:21 PM, Antti Seppälä wrote:
>>
>> On 19 February 2015 at 11:43, Benjamin Larsson <benjamin@southpole.se>
>> wrote:
>>>
>>> On 2015-02-19 10:13, Antti Seppälä wrote:
>>>>
>>>>
>>>> It seems that currently the firmware download on the mn88472 is
>>>> somehow wrong for my Astrometa HD-901T2.
>>>>
>>>> Reducing the download chunk size (mn88472_config.i2c_wr_max) to 2
>>>> makes the firmware download consistently succeed.
>>>>
>>>
>>>
>>> Hi, try adding the workaround patch I sent for this.
>>>
>>> [PATCH 1/3] rtl28xxu: lower the rc poll time to mitigate i2c transfer
>>> errors
>>>
>>> I now see that it hasn't been merged. But I have been running with this
>>> patch for a few months now without any major issues.
>>>
>>
>> The patch really did improve firmware loading. Weird...
>>
>> Even with it I still get occasional i2c errors from r820t:
>>
>> [   15.874402] r820t 8-003a: r820t_write: i2c wr failed=-32 reg=0a len=1:
>> da
>> [   81.455517] r820t 8-003a: r820t_read: i2c rd failed=-32 reg=00
>> len=4: 69 74 e6 df
>> [   99.949702] r820t 8-003a: r820t_read: i2c rd failed=-32 reg=00
>> len=4: 69 74 e6 df
>>
>> These errors seem to appear more often if I'm reading the signal
>> strength values using e.g. femon.
>
>
> Could you disable whole IR polling and test
> modprobe dvb_usb_v2 disable_rc_polling=1
>
> It is funny that *increasing* RC polling makes things better, though...
>

Hi.

I tried loading the driver with polling disabled and it fails completely:

[ 5526.693563] mn88472 7-0018: downloading firmware from file
'dvb-demod-mn88472-02.fw'
[ 5527.032209] mn88472 7-0018: firmware download failed=-32
[ 5527.033864] rtl2832 7-0010: i2c reg write failed -32
[ 5527.033874] r820t 8-003a: r820t_write: i2c wr failed=-32 reg=05 len=1: 83
[ 5527.036014] rtl2832 7-0010: i2c reg write failed -32

I have no idea why the device behaves so counter-intuitively. Is there
maybe some sorf of internal power-save mode the device enters when
there is no i2c traffic for a while or something?

-Antti
--
To unsubscribe from this list: send the line "unsubscribe linux-media" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Antti Palosaari Feb. 19, 2015, 4:14 p.m. UTC | #5
On 02/19/2015 06:01 PM, Antti Seppälä wrote:
> On 19 February 2015 at 17:38, Antti Palosaari <crope@iki.fi> wrote:
>> On 02/19/2015 12:21 PM, Antti Seppälä wrote:
>>> On 19 February 2015 at 11:43, Benjamin Larsson <benjamin@southpole.se>
>>> wrote:
>>>>
>>>> On 2015-02-19 10:13, Antti Seppälä wrote:
>>>>>
>>>>>
>>>>> It seems that currently the firmware download on the mn88472 is
>>>>> somehow wrong for my Astrometa HD-901T2.
>>>>>
>>>>> Reducing the download chunk size (mn88472_config.i2c_wr_max) to 2
>>>>> makes the firmware download consistently succeed.
>>>>>
>>>>
>>>>
>>>> Hi, try adding the workaround patch I sent for this.
>>>>
>>>> [PATCH 1/3] rtl28xxu: lower the rc poll time to mitigate i2c transfer
>>>> errors
>>>>
>>>> I now see that it hasn't been merged. But I have been running with this
>>>> patch for a few months now without any major issues.
>>>>
>>>
>>> The patch really did improve firmware loading. Weird...
>>>
>>> Even with it I still get occasional i2c errors from r820t:
>>>
>>> [   15.874402] r820t 8-003a: r820t_write: i2c wr failed=-32 reg=0a len=1:
>>> da
>>> [   81.455517] r820t 8-003a: r820t_read: i2c rd failed=-32 reg=00
>>> len=4: 69 74 e6 df
>>> [   99.949702] r820t 8-003a: r820t_read: i2c rd failed=-32 reg=00
>>> len=4: 69 74 e6 df
>>>
>>> These errors seem to appear more often if I'm reading the signal
>>> strength values using e.g. femon.
>>
>>
>> Could you disable whole IR polling and test
>> modprobe dvb_usb_v2 disable_rc_polling=1
>>
>> It is funny that *increasing* RC polling makes things better, though...
>>
>
> Hi.
>
> I tried loading the driver with polling disabled and it fails completely:
>
> [ 5526.693563] mn88472 7-0018: downloading firmware from file
> 'dvb-demod-mn88472-02.fw'
> [ 5527.032209] mn88472 7-0018: firmware download failed=-32
> [ 5527.033864] rtl2832 7-0010: i2c reg write failed -32
> [ 5527.033874] r820t 8-003a: r820t_write: i2c wr failed=-32 reg=05 len=1: 83
> [ 5527.036014] rtl2832 7-0010: i2c reg write failed -32
>
> I have no idea why the device behaves so counter-intuitively. Is there
> maybe some sorf of internal power-save mode the device enters when
> there is no i2c traffic for a while or something?

IR polling does not use I2C but some own commands. Could you make more 
tests. Use rtl28xxu module parameter to disable IR and test. It will 
disable both IR interrupts and polling. Then make some tests with 
different IR polling intervals to see how it behaves.

I have 3 mn88472 and 1 mn88473 device and all those seems to work fine 
for me. I don't care to buy anymore devices to find out one which does 
not work. Somehow root of cause should be find - it is not proper fix to 
repeat or break I2C messages to multiple smaller ones.

Antti
Steven Toth Feb. 19, 2015, 4:25 p.m. UTC | #6
>> I tried loading the driver with polling disabled and it fails completely:
>>
>> [ 5526.693563] mn88472 7-0018: downloading firmware from file
>> 'dvb-demod-mn88472-02.fw'
>> [ 5527.032209] mn88472 7-0018: firmware download failed=-32
>> [ 5527.033864] rtl2832 7-0010: i2c reg write failed -32
>> [ 5527.033874] r820t 8-003a: r820t_write: i2c wr failed=-32 reg=05 len=1:
>> 83
>> [ 5527.036014] rtl2832 7-0010: i2c reg write failed -32
>>
>> I have no idea why the device behaves so counter-intuitively. Is there
>> maybe some sorf of internal power-save mode the device enters when
>> there is no i2c traffic for a while or something?
>
>
> IR polling does not use I2C but some own commands. Could you make more
> tests. Use rtl28xxu module parameter to disable IR and test. It will disable
> both IR interrupts and polling. Then make some tests with different IR
> polling intervals to see how it behaves.
>
> I have 3 mn88472 and 1 mn88473 device and all those seems to work fine for
> me. I don't care to buy anymore devices to find out one which does not work.
> Somehow root of cause should be find - it is not proper fix to repeat or
> break I2C messages to multiple smaller ones.

Ack.

Its the job of the I2C controller to manage the I2C bus
implementation, including any fragmentation needs, not the
tuner/demod/other driver.

Find and fix the resource contention bug in the bridge and the mn88472
will work as is. I suspect something is broken with I2C locking.
Antti Palosaari Feb. 19, 2015, 4:44 p.m. UTC | #7
On 02/19/2015 06:25 PM, Steven Toth wrote:
>>> I tried loading the driver with polling disabled and it fails completely:
>>>
>>> [ 5526.693563] mn88472 7-0018: downloading firmware from file
>>> 'dvb-demod-mn88472-02.fw'
>>> [ 5527.032209] mn88472 7-0018: firmware download failed=-32
>>> [ 5527.033864] rtl2832 7-0010: i2c reg write failed -32
>>> [ 5527.033874] r820t 8-003a: r820t_write: i2c wr failed=-32 reg=05 len=1:
>>> 83
>>> [ 5527.036014] rtl2832 7-0010: i2c reg write failed -32
>>>
>>> I have no idea why the device behaves so counter-intuitively. Is there
>>> maybe some sorf of internal power-save mode the device enters when
>>> there is no i2c traffic for a while or something?
>>
>>
>> IR polling does not use I2C but some own commands. Could you make more
>> tests. Use rtl28xxu module parameter to disable IR and test. It will disable
>> both IR interrupts and polling. Then make some tests with different IR
>> polling intervals to see how it behaves.
>>
>> I have 3 mn88472 and 1 mn88473 device and all those seems to work fine for
>> me. I don't care to buy anymore devices to find out one which does not work.
>> Somehow root of cause should be find - it is not proper fix to repeat or
>> break I2C messages to multiple smaller ones.
>
> Ack.
>
> Its the job of the I2C controller to manage the I2C bus
> implementation, including any fragmentation needs, not the
> tuner/demod/other driver.
>
> Find and fix the resource contention bug in the bridge and the mn88472
> will work as is. I suspect something is broken with I2C locking.

It uses i2c control messages - single message per single operation.

You will need lock "control message" when it is done using multiple 
messages, very typically two bulk message one for request and one for 
reply - but that's not the case. Anyhow, it is possible firmware 
rtl2832u firmware does not like some "large (larger than single 
message)" operation is interrupted...

regards
Antti
Antti Seppälä Feb. 19, 2015, 6:42 p.m. UTC | #8
On 19 February 2015 at 18:14, Antti Palosaari <crope@iki.fi> wrote:
> On 02/19/2015 06:01 PM, Antti Seppälä wrote:
>>
>> On 19 February 2015 at 17:38, Antti Palosaari <crope@iki.fi> wrote:
>>>
>>> On 02/19/2015 12:21 PM, Antti Seppälä wrote:
>>>>
>>>> On 19 February 2015 at 11:43, Benjamin Larsson <benjamin@southpole.se>
>>>> wrote:
>>>>>
>>>>>
>>>>> On 2015-02-19 10:13, Antti Seppälä wrote:
>>>>>>
>>>>>>
>>>>>>
>>>>>> It seems that currently the firmware download on the mn88472 is
>>>>>> somehow wrong for my Astrometa HD-901T2.
>>>>>>
>>>>>> Reducing the download chunk size (mn88472_config.i2c_wr_max) to 2
>>>>>> makes the firmware download consistently succeed.
>>>>>>
>>>>>
>>>>>
>>>>> Hi, try adding the workaround patch I sent for this.
>>>>>
>>>>> [PATCH 1/3] rtl28xxu: lower the rc poll time to mitigate i2c transfer
>>>>> errors
>>>>>
>>>>> I now see that it hasn't been merged. But I have been running with this
>>>>> patch for a few months now without any major issues.
>>>>>
>>>>
>>>> The patch really did improve firmware loading. Weird...
>>>>
>>>> Even with it I still get occasional i2c errors from r820t:
>>>>
>>>> [   15.874402] r820t 8-003a: r820t_write: i2c wr failed=-32 reg=0a
>>>> len=1:
>>>> da
>>>> [   81.455517] r820t 8-003a: r820t_read: i2c rd failed=-32 reg=00
>>>> len=4: 69 74 e6 df
>>>> [   99.949702] r820t 8-003a: r820t_read: i2c rd failed=-32 reg=00
>>>> len=4: 69 74 e6 df
>>>>
>>>> These errors seem to appear more often if I'm reading the signal
>>>> strength values using e.g. femon.
>>>
>>>
>>>
>>> Could you disable whole IR polling and test
>>> modprobe dvb_usb_v2 disable_rc_polling=1
>>>
>>> It is funny that *increasing* RC polling makes things better, though...
>>>
>>
>> Hi.
>>
>> I tried loading the driver with polling disabled and it fails completely:
>>
>> [ 5526.693563] mn88472 7-0018: downloading firmware from file
>> 'dvb-demod-mn88472-02.fw'
>> [ 5527.032209] mn88472 7-0018: firmware download failed=-32
>> [ 5527.033864] rtl2832 7-0010: i2c reg write failed -32
>> [ 5527.033874] r820t 8-003a: r820t_write: i2c wr failed=-32 reg=05 len=1:
>> 83
>> [ 5527.036014] rtl2832 7-0010: i2c reg write failed -32
>>
>> I have no idea why the device behaves so counter-intuitively. Is there
>> maybe some sorf of internal power-save mode the device enters when
>> there is no i2c traffic for a while or something?
>
>
> IR polling does not use I2C but some own commands. Could you make more
> tests. Use rtl28xxu module parameter to disable IR and test. It will disable
> both IR interrupts and polling. Then make some tests with different IR
> polling intervals to see how it behaves.
>

Hi Antti.

I made some further tests for you. Here are the results:

dvb_usb_v2 disable_rc_polling=1: firmware download FAILED

dvb_usb_rtl28xxu disable_rc=1: firmware download FAILED

Then I restored the module parameters to default values and tested
with various rc->interval values:

interval = 800: firmware download FAILED
interval = 600: firmware download FAILED
interval = 400: firmware download FAILED
interval = 300: firmware download SUCCESS but I2C errors from tuner
could be sometimes observed
interval = 200: firmware download SUCCESS
interval = 100: firmware download SUCCESS

So somehow higher rc polling rate makes the firmware download succeed.
This could indeed be some locking/timing related bug.

Please let me know if there is something else I can test.

-Antti
--
To unsubscribe from this list: send the line "unsubscribe linux-media" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Antti Palosaari Feb. 19, 2015, 8:32 p.m. UTC | #9
On 02/19/2015 08:42 PM, Antti Seppälä wrote:
> On 19 February 2015 at 18:14, Antti Palosaari <crope@iki.fi> wrote:
>> On 02/19/2015 06:01 PM, Antti Seppälä wrote:
>>>
>>> On 19 February 2015 at 17:38, Antti Palosaari <crope@iki.fi> wrote:
>>>>
>>>> On 02/19/2015 12:21 PM, Antti Seppälä wrote:
>>>>>
>>>>> On 19 February 2015 at 11:43, Benjamin Larsson <benjamin@southpole.se>
>>>>> wrote:
>>>>>>
>>>>>>
>>>>>> On 2015-02-19 10:13, Antti Seppälä wrote:
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> It seems that currently the firmware download on the mn88472 is
>>>>>>> somehow wrong for my Astrometa HD-901T2.
>>>>>>>
>>>>>>> Reducing the download chunk size (mn88472_config.i2c_wr_max) to 2
>>>>>>> makes the firmware download consistently succeed.
>>>>>>>
>>>>>>
>>>>>>
>>>>>> Hi, try adding the workaround patch I sent for this.
>>>>>>
>>>>>> [PATCH 1/3] rtl28xxu: lower the rc poll time to mitigate i2c transfer
>>>>>> errors
>>>>>>
>>>>>> I now see that it hasn't been merged. But I have been running with this
>>>>>> patch for a few months now without any major issues.
>>>>>>
>>>>>
>>>>> The patch really did improve firmware loading. Weird...
>>>>>
>>>>> Even with it I still get occasional i2c errors from r820t:
>>>>>
>>>>> [   15.874402] r820t 8-003a: r820t_write: i2c wr failed=-32 reg=0a
>>>>> len=1:
>>>>> da
>>>>> [   81.455517] r820t 8-003a: r820t_read: i2c rd failed=-32 reg=00
>>>>> len=4: 69 74 e6 df
>>>>> [   99.949702] r820t 8-003a: r820t_read: i2c rd failed=-32 reg=00
>>>>> len=4: 69 74 e6 df
>>>>>
>>>>> These errors seem to appear more often if I'm reading the signal
>>>>> strength values using e.g. femon.
>>>>
>>>>
>>>>
>>>> Could you disable whole IR polling and test
>>>> modprobe dvb_usb_v2 disable_rc_polling=1
>>>>
>>>> It is funny that *increasing* RC polling makes things better, though...
>>>>
>>>
>>> Hi.
>>>
>>> I tried loading the driver with polling disabled and it fails completely:
>>>
>>> [ 5526.693563] mn88472 7-0018: downloading firmware from file
>>> 'dvb-demod-mn88472-02.fw'
>>> [ 5527.032209] mn88472 7-0018: firmware download failed=-32
>>> [ 5527.033864] rtl2832 7-0010: i2c reg write failed -32
>>> [ 5527.033874] r820t 8-003a: r820t_write: i2c wr failed=-32 reg=05 len=1:
>>> 83
>>> [ 5527.036014] rtl2832 7-0010: i2c reg write failed -32
>>>
>>> I have no idea why the device behaves so counter-intuitively. Is there
>>> maybe some sorf of internal power-save mode the device enters when
>>> there is no i2c traffic for a while or something?
>>
>>
>> IR polling does not use I2C but some own commands. Could you make more
>> tests. Use rtl28xxu module parameter to disable IR and test. It will disable
>> both IR interrupts and polling. Then make some tests with different IR
>> polling intervals to see how it behaves.
>>
>
> Hi Antti.
>
> I made some further tests for you. Here are the results:
>
> dvb_usb_v2 disable_rc_polling=1: firmware download FAILED
>
> dvb_usb_rtl28xxu disable_rc=1: firmware download FAILED
>
> Then I restored the module parameters to default values and tested
> with various rc->interval values:
>
> interval = 800: firmware download FAILED
> interval = 600: firmware download FAILED
> interval = 400: firmware download FAILED
> interval = 300: firmware download SUCCESS but I2C errors from tuner
> could be sometimes observed
> interval = 200: firmware download SUCCESS
> interval = 100: firmware download SUCCESS
>
> So somehow higher rc polling rate makes the firmware download succeed.
> This could indeed be some locking/timing related bug.
>
> Please let me know if there is something else I can test.

Sure you could do. Play with I2C settings. Test few values to I2C bus 
speed / clock is good start. Put oscilloscope to I2C bus and look what 
there happens on error cases. There is also 2 different I2C commands, 
test if there is any difference. And so. Increasing polling interval to 
something between 250-300 does not sound very bad, though.

regards
Antti

Patch
diff mbox

diff --git a/drivers/media/usb/dvb-usb-v2/rtl28xxu.c b/drivers/media/usb/dvb-usb-v2/rtl28xxu.c
index d88f799..3c5c6f9 100644
--- a/drivers/media/usb/dvb-usb-v2/rtl28xxu.c
+++ b/drivers/media/usb/dvb-usb-v2/rtl28xxu.c
@@ -865,7 +865,7 @@  static int rtl2832u_frontend_attach(struct dvb_usb_adapter *adap)
 			struct mn88472_config mn88472_config = {};
 
 			mn88472_config.fe = &adap->fe[1];
-			mn88472_config.i2c_wr_max = 22,
+			mn88472_config.i2c_wr_max = 2,
 			strlcpy(info.type, "mn88472", I2C_NAME_SIZE);
 			mn88472_config.xtal = 20500000;
 			info.addr = 0x18;