Re: [PATCH v5 00/13] riscv: improve boot time isa extensions handling

[Conor Dooley]

On Sun, Feb 12, 2023 at 12:27:10PM -0800, Guenter Roeck wrote:
> On 2/12/23 10:45, Conor Dooley wrote:
> ...
> > 
> > > However, I still see that the patch series
> > > results in boot hangs with the sifive_u qemu emulation, where
> > > the log ends with "Oops - illegal instruction". Is that problem
> > > being addressed as well ?
> > 
> > Hmm, if it died on the last commit in this series, then I am not sure.
> > If you meant with riscv/for-next or linux-next that's fixed by a patch
> > from Samuel:
> > https://patchwork.kernel.org/project/linux-riscv/patch/20230212021534.59121-3-samuel@xxxxxxxxxxxx/
> > 
> 
> It failed after the merge, so it looks like it may have been merge damage.
> 
> Anyway, I applied
> 
> RISC-V: Don't check text_mutex during stop_machine

That being:
https://lore.kernel.org/all/20220322022331.32136-1-palmer@xxxxxxxxxxxx/
Which handles the lockdep assertion during stop_machine...

> riscv: Fix early alternative patching
> riscv: Fix Zbb alternative IDs
> 
> and the sifive_u emulation no longer crashes. However, I still get
> 
> [    0.000000] ------------[ cut here ]------------
> [    0.000000] WARNING: CPU: 0 PID: 0 at arch/riscv/kernel/patch.c:71 patch_insn_write+0x222/0x2f6

...but doesn't prevent the early "spam" of assertion failures from the
code patching for alternatives. I sent a patch to take the lock during
the alternative patching which should get rid of them for you. It did
for me at least!
https://lore.kernel.org/all/20230212194735.491785-1-conor@xxxxxxxxxx

> repeated several times.
> 
> I then also tested
> 
> riscv: patch: Fixup lockdep warning in stop_machine

This one just deletes the lockdep check, so I would expect it to remove
the complaints.

> riscv: Fix early alternative patching
> riscv: Fix Zbb alternative IDs
> 
> which works fine (no warning backtrace) for sifive_u, but gives me
> 
> WARNING: CPU: 0 PID: 0 at kernel/trace/trace_events.c:433 trace_event_raw_init+0xde/0x642

Hmm, do you have the full splat for this one handy?

> and a whole lot of
> 
> event btrfs_clear_extent_bit has unsafe dereference of argument 1
> 
> and similar messages when running the "virt" emulation. That was there before,
> but drowned in the noise. Ok, guess I'll need another round of bisect.

Thanks for all of your testing :)

Attachment: signature.asc
Description: PGP signature