On Thu, 7 Jul 2022 18:35:12 -0600 Yu Zhao <yuzhao@xxxxxxxxxx> wrote:
> > This relentless drive towards mm-stable: I for one cannot keep up.
> > I'd like to ask for slowing down a bit - my intention had been to
> > reach testing maple tree again (it's not yet what I'd call stable),
> > but this and a couple of other issues got in the way. More mails
> > to write.
>
> Sorry for not being clear (it doesn't seem confusing to me because I
> don't trust the bot):
>
> There were two reports, the first one [1] tested v4 without the fix
> [2]; the second one [3] tested v5 at patch 5/7 (not the whole series).
>
> v4 + the fix is good; v5 the whole series is good.
>
> Please do not drop patch 7/7 and do not add this fix.
As Vlastimil has sleuthed out that the
"BUG:sleeping_function_called_from_invalid_context_at_mm/gup.c"
reporter was using the v4 series, I've restored 7/7 ("mm/page_alloc:
replace local_lock with normal spinlock") for tomorrow's linux-next.
I retained this fix against 2/7 and reworked 7/7 appropriately. So it
is now
/* Lock and remove page from the per-cpu list */
static struct page *rmqueue_pcplist(struct zone *preferred_zone,
struct zone *zone, unsigned int order,
gfp_t gfp_flags, int migratetype,
unsigned int alloc_flags)
{
struct per_cpu_pages *pcp;
struct list_head *list;
struct page *page;
unsigned long flags;
unsigned long __maybe_unused UP_flags;
/*
* spin_trylock may fail due to a parallel drain. In the future, the
* trylock will also protect against IRQ reentrancy.
*/
pcp_trylock_prepare(UP_flags);
pcp = pcp_spin_trylock_irqsave(zone->per_cpu_pageset, flags);
if (!pcp) {
pcp_trylock_finish(UP_flags);
return NULL;
}
/*
* On allocation, reduce the number of pages that are batch freed.
* See nr_pcp_free() where free_factor is increased for subsequent
* frees.
*/
pcp->free_factor >>= 1;
list = &pcp->lists[order_to_pindex(migratetype, order)];
page = __rmqueue_pcplist(zone, order, migratetype, alloc_flags, pcp, list);
pcp_spin_unlock_irqrestore(pcp, flags);
pcp_trylock_finish(UP_flags);
if (page) {
__count_zid_vm_events(PGALLOC, page_zonenum(page), 1);
zone_statistics(preferred_zone, zone, 1);
}
return page;
}
btw, the leading comment implies (to me) that the page is to be locked.
The comment could do with a rethink.
