> On 02/27/2016 03:28 PM, John Paul Adrian Glaubitz wrote: > > Hi Hans! > > > > On 02/27/2016 03:20 PM, Geert Uytterhoeven wrote: > >>> I noticed that 4.1 was ok and v4.2 wasn't, so I did a git bisect and > >>> ended up with > >>> commit 8222dbe21e79338de92d5e1956cd1e3994cc9f93 (sched/preempt, mm/fault: > >>> Decouple > >>> preemption from the page fault logic) as the culprit. > > > > If you like, please report this the kernel bug tracker and set > > "Hardware" to "SuperH". This way we are able to keep track of > > the issue. > > > > I think Yoshinori Sato and Rich Felker will have a look at this. > > > > Adrian > > > > Done: https://bugzilla.kernel.org/show_bug.cgi?id=113321 > > Regards, > > Hans >
Looks like the code then requires to not be allowed to sleep/preempt in these
sections (just like kmap_atomic, or kmap_coherent on mips).
So the fix should be simple as (untested):
>From d29f0d0cb670175c688cdd181c7a693d28b27a33 Mon Sep 17 00:00:00 2001
>
From: David Hildenbrand <d...@linux.vnet.ibm.com>
Date: Mon, 29 Feb 2016 09:19:24 +0100
Subject: [PATCH] sched/preempt, sh: kmap_coherent relies on disabled
preemption
kmap_coherent needs disabled preemption to not schedule in the critical
section, just like kmap_coherent on mips and kmap_atomic in general.
Fixes: 8222dbe21e79 "sched/preempt, mm/fault: Decouple preemption from the page
fault logic"
Reported-by: Hans Verkuil <hverk...@xs4all.nl>
Signed-off-by: David Hildenbrand <d...@linux.vnet.ibm.com>
---
arch/sh/mm/kmap.c | 2 ++
1 file changed, 2 insertions(+)
diff --git a/arch/sh/mm/kmap.c b/arch/sh/mm/kmap.c
index ec29e14..bf25d7c 100644
--- a/arch/sh/mm/kmap.c
+++ b/arch/sh/mm/kmap.c
@@ -36,6 +36,7 @@ void *kmap_coherent(struct page *page, unsigned long addr)
BUG_ON(!test_bit(PG_dcache_clean, &page->flags));
+ preempt_disable();
pagefault_disable();
idx = FIX_CMAP_END -
@@ -64,4 +65,5 @@ void kunmap_coherent(void *kvaddr)
}
pagefault_enable();
+ preempt_enable();
}
--
2.3.9
David
