[PATCH] drm/panfrost: fix power transition timeout warnings

[Christian Hewitt]

> On 25 Mar 2024, at 2:28 pm, Steven Price <steven.price at arm.com> wrote:
> 
> On 22/03/2024 16:45, Christian Hewitt wrote:
>> Increase the timeout value to prevent system logs on Amlogic boards flooding
>> with power transition warnings:
>> 
>> [   13.047638] panfrost ffe40000.gpu: shader power transition timeout
>> [   13.048674] panfrost ffe40000.gpu: l2 power transition timeout
>> [   13.937324] panfrost ffe40000.gpu: shader power transition timeout
>> [   13.938351] panfrost ffe40000.gpu: l2 power transition timeout
>> ...
>> [39829.506904] panfrost ffe40000.gpu: shader power transition timeout
>> [39829.507938] panfrost ffe40000.gpu: l2 power transition timeout
>> [39949.508369] panfrost ffe40000.gpu: shader power transition timeout
>> [39949.509405] panfrost ffe40000.gpu: l2 power transition timeout
>> 
>> The 2000 value has been found through trial and error testing with devices
>> using G52 and G31 GPUs.
> 
> How close to 2ms did you need in your trial and error testing? I'm
> wondering if we should increase it further in case this might still
> trigger occasionally?

I backed it off progressively but still saw occasional messages at 1.6ms
so padded it a little with 2ms, and those systems haven’t shown errors
since so I currently see it as a ’safe’ value. The one possible wildcard
is testing with older T820/T628 boards; but that needs to wait until I’m
back home from a long trip and able to test them. The possible theory
being that older/slower systems might require more time. Worst case I’ll
have to send another change.

> kbase seems to have a 5s (5000ms!) timeout before it will actually
> complain. But equally it doesn't busy wait on the registers in the same
> way as panfrost, so the impact to the rest of the system of a long wait
> is less.
> 
> But 2ms doesn't sound an unreasonable timeout so:
> 
> Reviewed-by: Steven Price <steven.price at arm.com>
> 
>> Fixes: 22aa1a209018 ("drm/panfrost: Really power off GPU cores in panfrost_gpu_power_off()")
>> Signed-off-by: Christian Hewitt <christianshewitt at gmail.com>> ---
>> drivers/gpu/drm/panfrost/panfrost_gpu.c | 6 +++---
>> 1 file changed, 3 insertions(+), 3 deletions(-)
>> 
>> diff --git a/drivers/gpu/drm/panfrost/panfrost_gpu.c b/drivers/gpu/drm/panfrost/panfrost_gpu.c
>> index 9063ce254642..fd8e44992184 100644
>> --- a/drivers/gpu/drm/panfrost/panfrost_gpu.c
>> +++ b/drivers/gpu/drm/panfrost/panfrost_gpu.c
>> @@ -441,19 +441,19 @@ void panfrost_gpu_power_off(struct panfrost_device *pfdev)
>> 
>> gpu_write(pfdev, SHADER_PWROFF_LO, pfdev->features.shader_present);
>> ret = readl_relaxed_poll_timeout(pfdev->iomem + SHADER_PWRTRANS_LO,
>> -  val, !val, 1, 1000);
>> +  val, !val, 1, 2000);
>> if (ret)
>> dev_err(pfdev->dev, "shader power transition timeout");
>> 
>> gpu_write(pfdev, TILER_PWROFF_LO, pfdev->features.tiler_present);
>> ret = readl_relaxed_poll_timeout(pfdev->iomem + TILER_PWRTRANS_LO,
>> -  val, !val, 1, 1000);
>> +  val, !val, 1, 2000);
>> if (ret)
>> dev_err(pfdev->dev, "tiler power transition timeout");
> 
> As Angelo points out the tiler probably doesn't need such a long
> timeout, but I can't see the harm in consistency so I'm happy with this
> change. If my memory of the hardware is correct then the tiler power off
> actually does very little and so I wouldn't expect it to take very long.

I’ve seen tiler timeouts once I think and thus included it, but not since
the values were increased. As long as it’s acceptable I won’t over-think
it but if more testing is needed I can look at it more.

> Steve
> 
>> gpu_write(pfdev, L2_PWROFF_LO, pfdev->features.l2_present);
>> ret = readl_poll_timeout(pfdev->iomem + L2_PWRTRANS_LO,
>> -  val, !val, 0, 1000);
>> +  val, !val, 0, 2000);
>> if (ret)
>> dev_err(pfdev->dev, "l2 power transition timeout");
>> }

Christian