I work with zhangzhansheng(from h3c) find that some situation
may casue deadlock when we use kmap_high and kmap_XXX or kumap_xxx
between differt cores at the same time, kmap_high->map_new_virtual->
flush_all_zero_pkmaps->flush_tlb_kernel_range->on_each_cpu. On this
condition, kmap_high hold the kmap_lock，wait the other cores respond
to ipi. But the others may disable irq and wait kmap_lock,
this is some deadlock condition. I think it's necessary to give
kmap_lock before call flush_tlb_kernel_range.
Like this:
spin_unlock(&kmap_lock);
flush_tlb_kernel_range(xxx);
spin_lock(&kmap_lock);
CPU 0: cpu 1:
kmap_xxx() {
xxx irq_disable();
kmap_high(); spin_lock(&kmap_lock)
xxx yyyyyyy
spin_unlock(&kmap_lock)
irq_enable();
}
kmap_high detail:
kmap_high() {
zzz
spin_lock(&kmap_lock)
map_new_virtual->
flush_all_zero_pkmaps->
flush_tlb_kernel_range->
on_each_cpu
/*
if cpu 1 irq_disabled, the cpu 1
cannot ack, then cpu 0 and cpu 1 may hangup.
*/
spin_unlock(&kmap_lock)
zzz
}
Signed-off-by: zhangchun <zhang.chuna@xxxxxxx>
Reviewed-by: zhangzhengming <zhang.zhengming@xxxxxxx>
---
mm/highmem.c | 5 ++++-
1 file changed, 4 insertions(+), 1 deletion(-)
diff --git a/mm/highmem.c b/mm/highmem.c
index bd48ba4..841b370 100644
--- a/mm/highmem.c
+++ b/mm/highmem.c
@@ -220,8 +220,11 @@ static void flush_all_zero_pkmaps(void)
set_page_address(page, NULL);
need_flush = 1;
}
- if (need_flush)
+ if (need_flush) {
+ spin_unlock(&kmap_lock);
flush_tlb_kernel_range(PKMAP_ADDR(0), PKMAP_ADDR(LAST_PKMAP));
+ spin_lock(&kmap_lock);
+ }
}
void __kmap_flush_unused(void)
--
1.8.3.1
