Re: vchiq: Performance regression since 5.18-rc1

[Phil Elwell <phil@raspberrypi.com>]

On 23/05/2022 12:15, Stefan Wahren wrote:
> Hi Phil,
> 
> Am 23.05.22 um 13:01 schrieb Phil Elwell:
>> Hi Stefan,
>>
>> On 23/05/2022 11:48, Stefan Wahren wrote:
>>> Hi Phil,
>>>
>>> Am 23.05.22 um 11:29 schrieb Phil Elwell:
>>>> Hi Stefan,
>>>>
>>>> On 23/05/2022 07:19, Stefan Wahren wrote:
>>>>> Hi Paul,
>>>>>
>>>>> Am 23.05.22 um 06:48 schrieb Paul E. McKenney:
>>>>>> On Sun, May 22, 2022 at 05:11:36PM +0200, Stefan Wahren wrote:
>>>>>>> Hi Paul,
>>>>>>>
>>>>>>> Am 22.05.22 um 01:46 schrieb Paul E. McKenney:
>>>>>>>> On Sun, May 22, 2022 at 01:22:00AM +0200, Stefan Wahren wrote:
>>>>>>>>> Hi,
>>>>>>>>>
>>>>>>>>> while testing the staging/vc04_services/interface/vchiq_arm driver with my
>>>>>>>>> Raspberry Pi 3 B+ (multi_v7_defconfig) i noticed a huge performance
>>>>>>>>> regression since [ff042f4a9b050895a42cae893cc01fa2ca81b95c] mm:
>>>>>>>>> lru_cache_disable: replace work queue synchronization with synchronize_rcu
>>>>>>>>>
>>>>>>>>> Usually i run "vchiq_test -f 1" to see the driver is still working [1].
>>>>>>>>>
>>>>>>>>> Before commit:
>>>>>>>>>
>>>>>>>>> real    0m1,500s
>>>>>>>>> user    0m0,068s
>>>>>>>>> sys    0m0,846s
>>>>>>>>>
>>>>>>>>> After commit:
>>>>>>>>>
>>>>>>>>> real    7m11,449s
>>>>>>>>> user    0m2,049s
>>>>>>>>> sys    0m0,023s
>>>>>>>>>
>>>>>>>>> Best regards
>>>>>>>>>
>>>>>>>>> [1] - https://github.com/raspberrypi/userland
>>>>>>>> Please feel free to try the patch shown below.  Or the pair of patches
>>>>>>>> from Rik here:
>>>>>>>>
>>>>>>>> https://lore.kernel.org/lkml/20220218183114.2867528-2-riel@surriel.com/
>>>>>>>> https://lore.kernel.org/lkml/20220218183114.2867528-3-riel@surriel.com/
>>>>>>> I tried your patch and Rik's patches but in both cases vchiq_test runs 7
>>>>>>> minutes instead of ~ 1 second.
>>>>>> That is surprising.  Do you boot with rcupdate.rcu_normal=1?
>>>>> No, not explicit.
>>>>>>    That would
>>>>>> nullify my patch, but I would expect that Rik's patch would still provide
>>>>>> increased performance even in that case.
>>>>> I will retest with a fresh SD card image.
>>>>>>
>>>>>> Could you please characterize where the slowdown is occurring?
>>>>>
>>>>> Unfortunately i don't have a deep insight into driver and vchiq_test tool. 
>>>>> Just a user view.
>>>>>
>>>>> Do you think an strace would be a good starting point?
>>>>>
>>>>> @Phil Any advices to analyse this issue?
>>>>
>>>> Sending many small control packets:
>>>>
>>>>    vchiq_test -c 1 10000
>>>>
>>>> essentially tests interrupt latency. Using a small number of large bulk 
>>>> transfers:
>>>>
>>>>    vchiq_test -b 10000 1
>>>>
>>>> becomes a test of how long it takes to lock down pages. It also tests DMA 
>>>> transfer speeds, but since the DMA is run by the firmware (which you aren't 
>>>> changing), I think you can rule that.
>>> Thanks i will try.
>>>>
>>>> You may also find it helpful to include "force_turbo=1" in config.txt for 
>>>> more predictable results.
>>>>
>>>> By the way, running our 5.18-rc7-based branch on a 3B+ I'm not seeing any 
>>>> performance problems:
>>> I assume you are using arm/bcm2709_defconfig and not arm/multi_v7_defconfig 
>>> as me?
>>
>> That's correct. Simply switching to multi_v7_defconfig breaks vchiq 
>> completely, presumably because it doesn't define CONFIG_BCM2835_VCHIQ.
> sorry, forgot to mention. I that i enable VCHIQ as module on top of 
> multi_v7_defconfig.

Downstream tree with multi_v7_defconfig + CONFIG_BCM2835_VCHIQ:

pi@raspberrypi:~$ time vchiq_test -f 1
Functional test - iters:1
======== iteration 1 ========
Testing bulk transfer for alignment.
Testing bulk transfer at PAGE_SIZE.

real    0m0.566s
user    0m0.037s
sys     0m0.166s

Phil