On Fri, Feb 11, 2022 at 12:51 PM Wolfram Sang <w...@kernel.org> wrote: > > On Fri, Feb 04, 2022 at 07:22:44PM +0000, Tareque Md.Hanif wrote: > > The issue still exists. It takes very long time to suspend (10-12s). > > `DRI_PRIME=1 glxgears` is a black window. > > > > journalctl attached > > Looking forward to any testing. > > Any new ideas which Tareque could test?
For some background, the GPU has multiple i2c buses attached to it which the driver uses for querying the EDID on monitors over i2c. Although in this case, the i2c buses are not used because there is no display controller on this particular GPU. I'm not even sure if the gpu driver exposes any i2c buses in this case. The i2c buses present are in a vbios table that the driver parses at load time. If there are none on your platform, then it's probably not the AMD GPU driver. Can you check if there are any i2c buses from the AMD GPU device in sysfs? I don't really see why this patch would change anything off hand on the GPU. Maybe the GPU is a red herring. No one reported any regressions on systems with AMD CPUs or even other Intel CPUs. This seems to be specific to a particular Intel CPU + AMD Topaz GPUs. The dGPU power is controlled by the platform via ACPI (it's usually a GPIO under the covers). I wonder if the issue is related to one of the i2c buses or devices on the intel platform? E.g., 00:15.0 Signal processing controller: Intel Corporation Sunrise Point-LP Serial IO I2C Controller #0 (rev 21) 00:15.1 Signal processing controller: Intel Corporation Sunrise Point-LP Serial IO I2C Controller #1 (rev 21) 00:1f.4 SMBus: Intel Corporation Sunrise Point-LP SMBus (rev 21) I'm not sure what those do, but maybe that is something to look at? Also there are some PCI AERs in the kernel log. Are those present without the patch? Alex
