AMD Lands Displayable DCC Support For Raven APUs In Mesa 19.1's RadeonSI

[andrei_me]

Wow, nice feedback in this thread, thanks agd5f finalzone chithanh for the feedback/clarifications/info

[chithanh]

If you are having IOMMU problems, iommu=pt or iommu=off may help.

Also iommu=soft is a workaround for some issues.

This sounds like an important thing to know. Which boards/OEMs have requested Linux support? Seems logical that we should be buying their hardware in this case.

I have had my hands on a few Ryzen laptops now, and I can tell the situation is pretty bad. It is very apparent that these OEMs did not attempt to run Linux at least once on their systems. BIOS/UEFI bugs galore, sometimes worked around in Linux, sometimes not.

1. Huawei Matebook D 14" Ryzen Edition. This is the only one that I can say works well in Linux. Everything just works OOTB in Ubuntu. For correct touchpad/touchscreen function, other distros may need kernel patch for AMD MP2 I2C controller (scheduled for mainline inclusion in kernel 5.2, A WHOLE 1.5 YEARS AFTER RAVEN RIDGE LAPTOPS LAUNCHED WTF AMD).
2. Dell Latitude 5495. It works ok. Fingerprint reader is some Broadcom 5880 which is unsupported in Linux. amdkfd doesn't load because of buggy ACPI CRAT[0].
3. Lenovo Thinkpad E485. Meh. Its poor battery life is even slightly worse in Linux. It is affected by both broken ACPI IVRS[1] which prevents boot (workaround iommu=soft or ivrs_ioapic=... parameter), and broken ACPI CRAT[0] that prevents amdkfd from working.
4. HP Envy x360 15z. Stay away from this one. The fan goes crazy in Linux (need NBFC or similar to make it behave). The BIOS is extremely buggy, but partially worked around in modern kernels[2]. It has an Achilles heel/hinge as mentioned above. Sadly it is the only one with a FreeSync capable display besides the chunky gaming laptops.

From Ryzen mobos for PCs, these usually work better but also not perfect. On my ASRock B350 mobo for instance I cannot use the TPM because of buggy youguessedit.

[0] https://bugs.freedesktop.org/show_bug.cgi?id=107898
[1] https://forums.lenovo.com/t5/ThinkPa...e/td-p/4191484
[2] https://bugzilla.kernel.org/show_bug.cgi?id=198715

[stiiixy]

Sell your weak laptop before hinges break. My kid did not and the laptop is garbage now. Fixing hinges broke the display back panel.
https://www.notebookcheck.net/HP-Env....266614.0.html

"
Perhaps the worst physical aspect of the new design is its dual hinges. They feel weaker than the hinges of the Yoga 720 series and are more susceptible to movement when lifting and transporting the notebook
"

I just turn laptops in to desktops when the display panels break. Unless it's an expensive model worth fixing.

[Go_Vulkan]

If you go to an OEM site and select Linux as the OS, that should give you the list of platforms that come with Linux (usually Ubuntu) preloaded. .....

Thanks for your comments, @agd5f!

Talking about commercial mainboards from ASRock, MSI, Asus, Gigabyte and Biostar .... what is your personal feeling where I could expect the best (or better than elsewhere) BIOS support?

[agd5f]

That does not yet translate into specific products one could buy.

If you go to an OEM site and select Linux as the OS, that should give you the list of platforms that come with Linux (usually Ubuntu) preloaded. We work with the OEMs to validate those parts on Linux just like happens with windows on other OEM platforms.

If you've got a specific recommendation for kernel parameters or ".config" options to use to disable such problematic features, please let us know. I for one certainly would not miss a few percent performance if it was that what it would take to achieve reasonably long uptimes while using amdgpu.

It depends on the sbios on your system. If you are having IOMMU problems, iommu=pt or iommu=off may help. Adding idle=nomwait also seems to fix stability issues on some systems. That seems to be related to certain combinations of ACPI tables and our CPU team has been working on trying to narrow down the issue. For the interrupt remapping, I'm not sure how or if you can disable it in the kernel, but you can override the mappings on the kernel command line (e.g., ivrs_ioapic[4]=00:14.0 ivrs_ioapic[5]=00:00.2). Sometimes you can also work around the issue by changing settings in the sbios (e.g., changing some settings from "auto" to "enabled" for example, or enabling SVM mode, etc.). I suppose in some cases, not all possible combinations of bios settings were tested by the OEM.

[finalzone]

The Linux ACPI driver is made by intel and they use latest intel motherboards for testing. Reporting acpi bugs is useless, they will will answer that the motherboard manufacturer uses old bios compiler. You can comment out annoying ACPI errors.

Maybe assigning the issue to bios i.e. HP do the trick.

[finalzone]

The autorotation works fine here on this very laptop since summer. The drivers seem to be in place, but the desktops do not directly make use of it.
Instead you need to install a daemon which exposes the functionality to them. Just installing iio-sensor-proxy worked for me on Cinnamon and Gnome.

Not in my case with iio-sensor-proxy installed by default despite the sensor presence.

3:00.7 Non-VGA unclassified device [0000]: Advanced Micro Devices, Inc. [AMD] Raven/Raven2/Renoir Sensor Fusion Hub [1022:15e4]

It seems something is missing. The laptop model in question is a newer cp0xxx I got on October of last year.

[Mani]

Not to mention the still missing driver for auto-rotation screen for year since Raven Ridge launch

Nevertheless, good to hear a further improvement for Raven Ridge hardware assuming the mobile version benefits as well.

The autorotation works fine here on this very laptop since summer. The drivers seem to be in place, but the desktops do not directly make use of it.
Instead you need to install a daemon which exposes the functionality to them. Just installing iio-sensor-proxy worked for me on Cinnamon and Gnome.

[finalzone]

The most problematic ones are IOMMU and interrupt remapping (IVRS tables). The boards usually have the ACPI tables, but they tend to be buggy since windows doesn't use them. Getting a sbios that works properly with Linux is the best bet unfortunately.

ACPI still issue on majority of AMD laptops as noticed on boot as seen below from ENVY x360 Raven Ridge powered;

ACPI Error: Field [D128] at bit offset/length 128/1024 exceeds size of target Buffer (160 bits) (20181213/dsopcode-197)
ACPI Error: Method parse/execution failed \HWMC, AE_AML_BUFFER_LIMIT (20181213/psparse-529)
 ACPI Error: Method parse/execution failed \_SB.WMID.WMAA, AE_AML_BUFFER_LIMIT (20181213/psparse-529)
ACPI Error: Field [D128] at bit offset/length 128/1024 exceeds size of target Buffer (160 bits) (20181213/dsopcode-197)

Apparently, such problem occurred for years on other AMD powered HP laptops and left unresolved. It would be great one of AMD division addresses that problem.


Another issue still outstanding is this message

[drm:construct [amdgpu]] *ERROR* construct: Invalid Connector ObjectID from Adapter Service for connector index:2! type 0 expected 3

Not to mention the still missing driver for auto-rotation screen for year since Raven Ridge launch

Nevertheless, good to hear a further improvement for Raven Ridge hardware assuming the mobile version benefits as well.

[dwagner]

On the graphics side, we use a few commercial boards, but mostly AMD engineering boards and OEM platforms where Linux support was requested.

That does not yet translate into specific products one could buy.

This biggest issues generally come down to UEFI and ACPI. Most OEMs only validate windows and windows doesn't enable a bunch of features that Linux does. The most problematic ones are IOMMU and interrupt remapping (IVRS tables). The boards usually have the ACPI tables, but they tend to be buggy since windows doesn't use them.

If you've got a specific recommendation for kernel parameters or ".config" options to use to disable such problematic features, please let us know. I for one certainly would not miss a few percent performance if it was that what it would take to achieve reasonably long uptimes while using amdgpu.