Re: [PATCH v5 1/3] arm64/ras: support sea error recovery

[James Morse <james.morse@arm.com>]

Hi Xie XiuQi,

On 08/02/18 08:35, Xie XiuQi wrote:
> I am very glad that you are trying to solve the problem, which is very helpful.
> I agree with your proposal, and I'll test it on by box latter.
> 
> Indeed, we're in precess context when we are in sea handler. I was thought we
> can't call schedule() in the exception handler before.

While testing this I've come to the conclusion that the
memory_failure_queue_kick() approach I suggested makes arm64 behave slightly
differently with APEI, and would need re-inventing if we support kernel-first
too. The same race exists with memory-failure notifications signalled by SDEI,
and to a lesser extent IRQ. So by fixing this in arch-code, we actually making
our lives harder.

Instead, I have the patch below. This is smaller, and not arch specific. It also
saves the arch code secretly knowing that APEI calls memory_failure_queue().

I will post this as part of that series shortly...


Thanks,

James

---------------%<---------------
[PATCH] mm/memory-failure: increase queued recovery work's priority

arm64 can take an NMI-like error notification when user-space steps in
some corrupt memory. APEI's GHES code will call memory_failure_queue()
to schedule the recovery work. We then return to user-space, possibly
taking the fault again.

Currently the arch code unconditionally signals user-space from this
path, so we don't get stuck in this loop, but the affected process
never benefits from memory_failure()s recovery work. To fix this we
need to know the recovery work will run before we get back to user-space.

Increase the priority of the recovery work by scheduling it on the
system_highpri_wq, then try to bump the current task off this CPU
so that the recover work starts immediately.

Reported-by: Xie XiuQi <xiexiuqi@huawei.com>
Signed-off-by: James Morse <james.morse@arm.com>
CC: Xie XiuQi <xiexiuqi@huawei.com>
CC: gengdongjiu <gengdongjiu@huawei.com>
---
 mm/memory-failure.c | 11 ++++++++---
 1 file changed, 8 insertions(+), 3 deletions(-)

diff --git a/mm/memory-failure.c b/mm/memory-failure.c
index 4b80ccee4535..14f44d841e8b 100644
--- a/mm/memory-failure.c
+++ b/mm/memory-failure.c
@@ -55,6 +55,7 @@
 #include <linux/hugetlb.h>
 #include <linux/memory_hotplug.h>
 #include <linux/mm_inline.h>
+#include <linux/preempt.h>
 #include <linux/kfifo.h>
 #include <linux/ratelimit.h>
 #include "internal.h"
@@ -1319,6 +1320,7 @@ static DEFINE_PER_CPU(struct memory_failure_cpu,
memory_failure_cpu);
  */
 void memory_failure_queue(unsigned long pfn, int flags)
 {
+	int cpu = smp_processor_id();
 	struct memory_failure_cpu *mf_cpu;
 	unsigned long proc_flags;
 	struct memory_failure_entry entry = {
@@ -1328,11 +1330,14 @@ void memory_failure_queue(unsigned long pfn, int flags)

        mf_cpu = &get_cpu_var(memory_failure_cpu);
        spin_lock_irqsave(&mf_cpu->lock, proc_flags);
-       if (kfifo_put(&mf_cpu->fifo, entry))
-               schedule_work_on(smp_processor_id(), &mf_cpu->work);
-       else
+       if (kfifo_put(&mf_cpu->fifo, entry)) {
+               queue_work_on(cpu, system_highpri_wq, &mf_cpu->work);
+               set_tsk_need_resched(current);
+               preempt_set_need_resched();
+       } else {
                pr_err("Memory failure: buffer overflow when queuing memory
failure at %#lx\n",
                       pfn);
+       }
        spin_unlock_irqrestore(&mf_cpu->lock, proc_flags);
        put_cpu_var(memory_failure_cpu);
 }
---------------%<---------------