base-defconfig: enable DRM_XE #91
Conversation
Enable DRM for Intel Xe series GPUs. Signed-off-by: Fred Oh <[email protected]>
|
@kv2019i do we also need and there are tons of other DRM_XE options.... |
|
Device Testings are going on,
|
|
@plbossart wrote:
Here's the kconfig for Fedora test kernel for XE driver: So yes definitely for DRM_XE_DISPLAY=m |
| @@ -237,6 +237,7 @@ CONFIG_SND_SOC_ACPI_INTEL_MATCH=m | |||
| # DRM | |||
| CONFIG_DRM=m | |||
| CONFIG_DRM_I915_ALPHA_SUPPORT=y | |||
| CONFIG_DRM_XE=m | |||
There was a problem hiding this comment.
This should be good. We also need "CONFIG_DRM_XE_DISPLAY=y" but that's anyways the default.
There was a problem hiding this comment.
best to not add something that's default :-)
There was a problem hiding this comment.
confirmed, ONFIG_DRM_XE_DISPLAY=y is in generated .config file
There was a problem hiding this comment.
Looks CONFIG_DRM_I915_ALPHA_SUPPORT is removed from i915 driver.
|
@fredoh9 let us know what the test results are, and we can merge. Thanks for starting this! |
|
|
re-run for stable-v2.2 |
fredoh9
changed the title
|
stable-v2.2 looks good too |
|
Will merge this soon to be included in tomorrow's daily build. |
|
I tried to test this. I don't understand the i915 part of the advice. After adding both force_probe: After adding the GuC firmware, and re-enabling gdm3, wayland fails to start with "no GPU found" + the Ubuntu 22 infinite loop documented in thesofproject/sof-test#998
Maybe it's missing some other firmware? The kernel logs don't have any error anymore. To be continued. |
Searching for "(dis)connected" in i915_display_info shows useful connector info. Also show /sys/class/drm/ and /sys/kernel/debug/dri/0/, this will be useful for DRM_XE, see thesofproject/kconfig#91 Signed-off-by: Marc Herbert <[email protected]>
Searching for "(dis)connected" in i915_display_info shows useful connector info. Also show /sys/class/drm/ and /sys/kernel/debug/dri/0/, this will be useful for DRM_XE, see thesofproject/kconfig#91 Signed-off-by: Marc Herbert <[email protected]>
|
Smoking gun Still no idea why, to be continued. |
|
I got used to misnamed Wow! The undocumented default for |
|
Intel only supports a Schroedinger display which can be enabled and disabled at the same time. |
|
Thanks Clint and RK for root causing this. And the issue is... LNL display is simply not enabled upstream yet! This one-line patch is missing: https://cgit.freedesktop.org/drm-tip/commit/?id=79263e4b3f0ed5928a1622300d32ed35f7d8fc24 --- a/[drivers/gpu/drm/xe/xe_pci.c](https://cgit.freedesktop.org/drm-tip/tree/drivers/gpu/drm/xe/xe_pci.c?id=bf1a72ab5a446e383682e34347237ee5317c2185)
+++ b/[drivers/gpu/drm/xe/xe_pci.c](https://cgit.freedesktop.org/drm-tip/tree/drivers/gpu/drm/xe/xe_pci.c?id=79263e4b3f0ed5928a1622300d32ed35f7d8fc24)
@@ -333,6 +333,7 @@ static const struct xe_device_desc mtl_desc = {
static const struct xe_device_desc lnl_desc = {
PLATFORM(XE_LUNARLAKE),
+ .has_display = true,
.require_force_probe = true,
};Now why would you hold this line back when an explicit The answer is: display code tends to lag behind GPU code for various reasons. So there is a time window where you want users to beta test the GPU with force_probe but NOT the display yet. Lesson learned: next time I'll just... wait. |
|
so what's the conclusion? revert this PR, or add something on LNL test devices? |
yes, that is correct. We had the core of LNL upstream around November/2023, but the display part is more recent. It is already upstream, but not in a released kernel. You are missing this commit: |
|
Searching for "(dis)connected" in i915_display_info shows useful connector info. Also show /sys/class/drm/ and /sys/kernel/debug/dri/0/, this will be useful for DRM_XE, see thesofproject/kconfig#91 Signed-off-by: Marc Herbert <[email protected]>
|
Wow, the transition is actually documented! I don't think anyone ever pointed me to that page... |
|
New |
Enable DRM for Intel Xe series GPUs.