lkml.org 
[lkml]   [2013]   [Jan]   [30]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: [PATCH] cpsw: Fix interrupt storm among other things
From
Date
Hi Mugunthan,

On Jan 30, 2013, at 12:55 PM, Mugunthan V N wrote:

> On 1/30/2013 3:06 PM, Pantelis Antoniou wrote:
>> Hi,
>>
>> On Jan 30, 2013, at 11:03 AM, Mugunthan V N wrote:
>>
>>> On 1/30/2013 2:06 PM, Pantelis Antoniou wrote:
>>>> Hi Mugunthan,
>>>>
>>>> On Jan 29, 2013, at 1:45 PM, Mugunthan V N wrote:
>>>>
>>>>> On 1/28/2013 6:41 PM, Pantelis Antoniou wrote:
>>>>>> Fix interrupt storm on bone A4 cause by non-by-the-book interrupt handling.
>>>>>> While at it, added a non-NAPI mode (which is easier to debug), plus
>>>>>> some general fixes.
>>>>>>
>>>>>> Signed-off-by: Pantelis Antoniou <panto@antoniou-consulting.com>
>>>>>> ---
>>>>>> Documentation/devicetree/bindings/net/cpsw.txt | 1 +
>>>>>> drivers/net/ethernet/ti/cpsw.c | 222 +++++++++++++++++++++----
>>>>>> drivers/net/ethernet/ti/davinci_cpdma.c | 4 +-
>>>>>> drivers/net/ethernet/ti/davinci_cpdma.h | 2 +-
>>>>>> include/linux/platform_data/cpsw.h | 1 +
>>>>>> 5 files changed, 194 insertions(+), 36 deletions(-)
>>>>> I have tested CPSW on AM335x EVM 1.5A with flood ping and i am not
>>>>> seeing any interrupt storm.
>>>>> Can you provide more details on how to reproduce the issue.
>>>>>
>>>> A beaglebone prototype with the new silicon version, with the ethernet errata
>>>> fixed displays this. You can't trigger it on old silicon.
>>>>
>>>> The TI people on the CC list can confirm.
>>> But i have the same silicon revision (PG2.0) in my EVM and I am not seeing any issues. Can you
>>> point me to the ethernet errata which you are mentioning?
>>>
>>> Regards
>>> Mugunthan V N
>> What kernel version are you using? This is only triggered on the mainline driver.
>>
>> The advisory in question: From http://www.ti.com/lit/er/sprz360c/sprz360c.pdf
>>
>> Advisory 1.0.9: "Ethernet Media Access Controller and Switch Subsystem: C0_TX_PEND
>> and C0_RX_PEND Interrupts Not Connected to ARM Cortex-A8"
>>
>> I bet you're using an old kernel driver with the workarounds with the timers.
>>
>> If I had to guess (although I didn't use a probe or anything) is that the
>> interrupts are now proper level interrupts, instead of working in edge
>> triggered mode due to the workaround.
>>
>> Apparently the interrupt was never acked properly in the original driver
>> (the sequence described in the TRM is not followed).
>>
>> Looking at the TRM (spruh73g.pdf) 14.3.1.3 Interrupts in particular, the
>> the status registers are not read, and more damning the proper values to the
>> CPDMA_EOI_VECTOR register are not written.
>>
>> The original driver blindly wrote zero (cpdma_ctlr_eoi), while you have to
>> write different values according to the interrupt you ack.
>>
>> What happened was that on the first interrupt, the interrupt was never acked,
>> and we had an irq storm...
>>
>> Regards
>>
>> -- Pantelis
> The above mentioned advisory is for PG1.0 and not for PG2.0
> I am booting net-next kernel.
>
> [ 0.000000] Booting Linux on physical CPU 0x0
> [ 0.000000] Linux version 3.8.0-rc5-01248-gd2ed273 (a0131834@a0131834-linux) (gcc version 4.5.3 20110311 (prerelease) (GCC) ) #21 SMP Wed Jan 30 163
> [ 0.000000] CPU: ARMv7 Processor [413fc082] revision 2 (ARMv7), cr=10c53c7d
> [ 0.000000] CPU: PIPT / VIPT nonaliasing data cache, VIPT aliasing instruction cache
>
> [root@arago /]# uname -a
> Linux arago 3.8.0-rc5-01248-gd2ed273 #21 SMP Wed Jan 30 16:13:26 IST 2013 armv7l GNU/Linux
>
> In theory what you are mentioning is correct. I have a beagle bone black and yet to try it.
>

I don't know what kind of silicon revision you have there.

FWIW both the original bone and the black show the exact same CPU: ARMv7 lines:

bone-original:
[ 0.000000] CPU: ARMv7 Processor [413fc082] revision 2 (ARMv7), cr=10c5387d

bone-black: (with the known silicon version that has the fix)
[ 0.000000] CPU: ARMv7 Processor [413fc082] revision 2 (ARMv7), cr=10c5387d

TBH I haven't found a simple way to print out the silicon revision number.
Anyone on the list know a quick and dirty method?

> Regards
> Mugunthan V N

Regards

-- Pantelis




\
 
 \ /
  Last update: 2013-01-30 12:44    [W:0.059 / U:1.244 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site