mali_kbase_device_hw_csf gpu crash #129
Replies: 5 comments 75 replies
-
|
Since Armbian does not ship with GPU drivers installed, I can't be sure it's this distro. Unless you know all the GPU drivers were installed correctly, including performance governer tweaks. I can look at the kernel logs. If you can attach them to a text file and upload them here, use the command |
Beta Was this translation helpful? Give feedback.
-
|
i beleive i started experimenting with your build around version 1.8, maybe 1.7, maybe even 1.6, cant remember, but i am sure i had crashes since 1.9-1.10, but that maybe also when i started noticing or leaving it on. but for me only once or twice system froze, then chromium crashed and system recovered, most of the time its just frozen and thats it(with chroimum open). i also dont care if chromium chrashes but system stays alive. i can just leave it on fora few days without chromium loaded just to make sure it doesnt crash this way. |
Beta Was this translation helpful? Give feedback.
-
|
@Joshua-Riek Totally, agree. |
Beta Was this translation helpful? Give feedback.
-
|
I was searching for this bug, and found another patch that was in the android kernel but not patched in the linux kernel: |
Beta Was this translation helpful? Give feedback.
-
|
Interesting, thanks for the find. |
Beta Was this translation helpful? Give feedback.
-
|
I would recommend to disable this systemd service. It will set the GPU governor to performance on boot. |
Beta Was this translation helpful? Give feedback.
-
|
Definitely, since I installed version 1.13, OrangePi 5 has been freezing a lot. |
Beta Was this translation helpful? Give feedback.
-
|
I may have found a solution. The original error can be seen below for future reference. After looking at the file From c35b602e2caade4071361ac0070eeb036bdfed52 Mon Sep 17 00:00:00 2001
From: Joshua Riek <[email protected]>
Date: Sun, 7 May 2023 17:53:31 -0400
Subject: [PATCH] mali_kbase_device_hw_csf: remove kbase debug assert
---
.../device/backend/mali_kbase_device_hw_csf.c | 14 ++++++++++----
1 file changed, 10 insertions(+), 4 deletions(-)
diff --git a/drivers/gpu/arm/bifrost/device/backend/mali_kbase_device_hw_csf.c b/drivers/gpu/arm/bifrost/device/backend/mali_kbase_device_hw_csf.c
index 7939bfd8e..6289cb4c1 100644
--- a/drivers/gpu/arm/bifrost/device/backend/mali_kbase_device_hw_csf.c
+++ b/drivers/gpu/arm/bifrost/device/backend/mali_kbase_device_hw_csf.c
@@ -201,8 +201,11 @@ static bool kbase_is_register_accessible(u32 offset)
void kbase_reg_write(struct kbase_device *kbdev, u32 offset, u32 value)
{
- KBASE_DEBUG_ASSERT(kbdev->pm.backend.gpu_powered);
- KBASE_DEBUG_ASSERT(kbdev->dev != NULL);
+ if (WARN_ON(!kbdev->pm.backend.gpu_powered))
+ return;
+
+ if (WARN_ON(kbdev->dev == NULL))
+ return;
if (!kbase_is_register_accessible(offset))
return;
@@ -222,8 +225,11 @@ u32 kbase_reg_read(struct kbase_device *kbdev, u32 offset)
{
u32 val;
- KBASE_DEBUG_ASSERT(kbdev->pm.backend.gpu_powered);
- KBASE_DEBUG_ASSERT(kbdev->dev != NULL);
+ if (WARN_ON(!kbdev->pm.backend.gpu_powered))
+ return 0;
+
+ if (WARN_ON(kbdev->dev == NULL))
+ return 0;
if (!kbase_is_register_accessible(offset))
return 0;
--
2.25.1
|
Beta Was this translation helpful? Give feedback.
-
|
Yeah, I noticed the same result on my end. I asked monka where I could find the source for their os and got an ambiguous response, saying it's an Armbian mod build from armbian-next. I have the armbian-next repo, but I don't know what changes they made from armbian-next. |
Beta Was this translation helpful? Give feedback.
-
|
I did a fresh install and set cma=256M After some time it freezes gnome and I have to ssh into the box and kill the glmark. Then everything was ok and I could start the glmarktest again Is there a way I can provoke chromium to crash with a 1920x1080 screen? Hste |
Beta Was this translation helpful? Give feedback.
-
|
I might have just found the issue. By default, I have Removing |
Beta Was this translation helpful? Give feedback.
-
|
I have been testing for a few hours now, and it looks like |
Beta Was this translation helpful? Give feedback.
-
|
@HSTe for me it crashes in any resolution by just leaving youtube on, paused. |
Beta Was this translation helpful? Give feedback.
-
|
I saw that mali was missing some value from devicetree power-off-delay-ms so I search for if it was set in any of the rockchip's dts files. And found it in rk3399.dtsi I merged the value from there and added some other setting from that part. I also added that security patch that was mention for android: I sat PAN_MESA_DEBUG=gofaster,gl3 So far it looks more stable. The glmark2-es2-wayland still get out of resources after some time, but its just to terminate it with ctrl-cand thing are back to normal. |
Beta Was this translation helpful? Give feedback.
-
|
I have been running youtube 4k in chromium for many hours and it doesn't crash for me with the changes I mentioned. I have attached the dmesg from the boot |
Beta Was this translation helpful? Give feedback.
-
|
I had a talk with some other devs working on similar projects, and they recommend not using PAN_MESA_DEBUG=gofaster as a global env because it's a hack. Currently my plan is to remove PAN_MESA_DEBUG=gofaster in the next release for this reason, i will need to do more testing with that device tree modification you have. I have been working on a new kernel based on linux-5.10-gen-rkr3.6 from Rockchip. With support for the rock5, nanopi r6, and orangepi 5 plus, so I have not had the time to test this further. If you want to take a look at the new kernel i have the repo below, please note that I might rebase and come commits will change. |
Beta Was this translation helpful? Give feedback.
-
|
has there been any luck on this issue? i skipped 2 version updates :) |
Beta Was this translation helpful? Give feedback.
-
|
The system will not crash any more due to GPU-related tasks. I removed PAN_MESA_DEBUG=gofaster because it's an experimental feature. |
Beta Was this translation helpful? Give feedback.
-
|
thanks, ill go download and try :) |
Beta Was this translation helpful? Give feedback.
-
|
Thanks. I think its a good approach to use a common kernel. I'm still using PAN_MESA_DEBUG=gofaster,gl3 but I think it's a good idea not to drop gofaster as default and let people add it if they want the performance on the cost of stability |
Beta Was this translation helpful? Give feedback.
-
it just freezes, mostly when using chrome but sometimes chrome is not on. at first i thought it was a weak psu but then i got 5v 5amp brick. then i thought maybe something is faulty, like memory or whatever so i tried armbian. no crashes so far on armbian. but it gets really hot bcos cpu does heavy work even on 1080p youtube.
sorry for me being a noob but is there something/somewhere i can look at, some log or whatever after the restart, to start with fixing this crashing cos it seems its related to this distro. when its frozen i can just push and hold power button or pull the cable, thats the only thing that helps.
thanks!
Beta Was this translation helpful? Give feedback.
All reactions