Re: [PATCH] percpu: Sync vmalloc mappings in pcpu_alloc() and free_percpu()

From: Steven Rostedt
Date: Mon May 04 2020 - 13:04:57 EST


On Mon, 4 May 2020 11:38:43 -0400 (EDT)
Mathieu Desnoyers <mathieu.desnoyers@xxxxxxxxxxxx> wrote:

> ----- On May 4, 2020, at 11:31 AM, Joerg Roedel jroedel@xxxxxxx wrote:
>
> > On Mon, May 04, 2020 at 11:28:46AM -0400, Mathieu Desnoyers wrote:
> >> ----- On May 4, 2020, at 11:12 AM, Joerg Roedel jroedel@xxxxxxx wrote:
> >> Placing this here is inefficient. It syncs mappings for each percpu allocation.
> >> I would recommend moving it right after __vmalloc() is called to allocate the
> >> underlying memory chunk instead:
> >>
> >> static void *pcpu_mem_zalloc(size_t size, gfp_t gfp)
> >> {
> >
> > Tried this before, actually I put it into the caller of
> > pcpu_mem_zalloc(), but that didn't fix the problem for me. Stevens
> > test-case still hangs the machine.
>
> That's unexpected.
>
> Did you confirm that those hangs were also caused by percpu allocations ?
>
> Maybe adding the vmalloc_sync_mappings() at each percpu allocation happens
> to luckily sync mappings after some other vmalloc.
>

It doesn't surprise me because my alloc_percpu() call never gets to that
path. But systemd does hit it for me earlier on.

-- Steve