search

sophgo / linux-riscv

Linux kernel stable tree
https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/
Other
23 stars 47 forks source link

Rebase sg2042-master on release v6.8 #108

Closed silvanshade closed 5 months ago

silvanshade commented 5 months ago

This PR just rebases the sg2042-master branch on to v6.8 release upstream.

xingxg2022 commented 5 months ago

"This branch cannot be rebased due to too many changes"

I will rebase it

kth5 commented 5 months ago

I does work, mostly anyway. Compiles cleanly with my .config but fails to bring up any but - seemingly - the first CPU core on a Milk-V Pioneer. The rest of the cores are detected but stay marked offline.

skeuchel commented 5 months ago

I does work, mostly anyway. Compiles cleanly with my .config but fails to bring up any but - seemingly - the first CPU core on a Milk-V Pioneer. The rest of the cores are detected but stay marked offline.

I have the same issue with 128G of ram, it brings up all cores with 64G though. I am currently doing that since I find this branch still more stable than the sg2042-dev branch.

kth5 commented 5 months ago

@skeuchel sg2042-dev is unstable indeed. Brings up my box just fine but when under load (compiling a kernel with make -j48 for instance), GCC is randomly segfaulting. I haven't given this one an indepth try yet to see about general stability since a single core is fairly useless to me.

I did however try the dtbs built with this as well as the one from sg2042-dev but it made no difference.

Generally speaking, I would love to see the upgrade to 6.8.x to get the Intel Xe driver without massively patching my kernel sources to get i915 going with my ARC380 instead.

silvanshade commented 5 months ago

I does work, mostly anyway. Compiles cleanly with my .config but fails to bring up any but - seemingly - the first CPU core on a Milk-V Pioneer. The rest of the cores are detected but stay marked offline.

I have the same issue with 128G of ram, it brings up all cores with 64G though. I am currently doing that since I find this branch still more stable than the sg2042-dev branch.

@kth5 @skeuchel Can either of you check, when you encounter this issue when booting with 128GB of RAM, whether you can re-enable the offline harts?

You should be able to re-enable the offline harts through CPU hotplug interface:

for i in {0..63}; do echo 1 > /sys/devices/system/cpu/cpu$i/online; done

I haven't had a chance to test this yet running a proper installation but it at least seems to work in the linuxboot u-root shell (on 6.8 kernel with 128GB RAM on a Pioneer V1.2 board). Before running the above command, only 1 hart shows up in /proc/cpuinfo, but after running it, all 64 are listed.

kth5 commented 5 months ago

@silvanshade I'll have another go at it this evening (CET).

I am running a V1.3 afaik and yes, with the full compliment of 128GB RAM.

Also, all cores were listed but marked offline much like hotplugging on ppc64le when disabling cores on run-time.

Will report back once I've tried enabling them manually after boot.

silvanshade commented 5 months ago

Thanks.

One thing to note is that I am using the kernel from the branch here, which is based on this PR.

The current sg2042-master has been updated with a different non-overlapping set of commits including a number of changes to the device tree and I haven't experimented with that yet.

kth5 commented 5 months ago

I'll head over to the milkv-community repo and file the issue there if it doesn't yet exist. Queued a build just now.