Re: [PATCH 6/9] drm/panthor: Reset GPU after L2 cache power off

[Boris Brezillon <boris.brezillon@collabora.com>]

On Mon, 3 Mar 2025 13:35:23 +0100
Boris Brezillon <boris.brezillon@collabora.com> wrote:

> On Mon, 3 Mar 2025 12:08:51 +0000
> Liviu Dudau <liviu.dudau@arm.com> wrote:
> 
> > On Fri, Feb 28, 2025 at 06:05:36PM +0100, Marek Vasut wrote:  
> > > On 2/28/25 12:06 PM, Liviu Dudau wrote:    
> > > > Hi Marek,    
> > > 
> > > Hi,
> > >     
> > > > On Thu, Feb 27, 2025 at 05:58:06PM +0100, Marek Vasut wrote:    
> > > > > This seems necessary on Freescale i.MX95 Mali G310 to reliably resume
> > > > > from runtime PM suspend. Without this, if only the L2 is powered down
> > > > > on RPM entry, the GPU gets stuck and does not indicate the firmware is
> > > > > booted after RPM resume.    
> > > > 
> > > > That doesn't sound right. Can you tell me what GPU firmware are you running
> > > > (we are now printing the git sha of the image at boot time).    
> > > Please see below. It could be some sort of NXP firmware fork ?
> > > 
> > > It comes from the NXP firmware repo , mali-imx-r50.2-710cfb6 .
> > > 
> > > panthor 4d900000.gpu: [drm] clock rate = 1000000000
> > > panthor 4d900000.gpu: EM: created perf domain
> > > panthor 4d900000.gpu: [drm] mali-unknown id 0xac74 major 0x0 minor 0x0
> > > status 0x1
> > > panthor 4d900000.gpu: [drm] Features: L2:0x7110306 Tiler:0x809 Mem:0x1
> > > MMU:0x2830 AS:0xff
> > > panthor 4d900000.gpu: [drm] shader_present=0x1 l2_present=0x1
> > > tiler_present=0x1
> > > panthor 4d900000.gpu: [drm] Firmware protected mode entry not be supported,
> > > ignoring
> > > panthor 4d900000.gpu: [drm] Firmware git sha:
> > > 8e5cfcfec20cc8aff8509d37e72babc935d34a3b    
> > 
> > This looks like it has been part of a R50 release of the DDK, which is recent
> > enough to consider it up-to-date. The issues you're seeing with fast resume are
> > probably due to some integration issues or other quirks.
> > 
> > Boris has the most recent experience with playing with fast resume, maybe
> > he can share some tips on where to add messages in Panthor to try to debug
> > your problem.  
> 
> If you end up with fast_reset=true, that means the FW claims it entered
> a suspend state from which is can resume quickly (fast-reset), and in
> that case, we're only supposed to power on the L2 block at resume time
> AFAIK.

Can you try deasserting the reset lines at probe time and keeping them
deasserted until the device is removed instead of re-asserting at
suspend time? If that doesn't work, try doing the same with power
domains.

> If there's a component that can tell us more, it's probably the
> FW, unfortunately, it doesn't tell us much through the FW interface,
> especially if it's not booting properly. Do you hit this error [1]
> without this patch, and if you do, what's the status reported there?
> 
> [1]https://elixir.bootlin.com/linux/v6.14-rc4/source/drivers/gpu/drm/panthor/panthor_fw.c#L1051