public inbox for linux-kernel@vger.kernel.org
 help / color / mirror / Atom feed
From: Andrew Morton <akpm@linux-foundation.org>
To: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Cc: mingo@elte.hu, jens.axboe@oracle.com, nickpiggin@yahoo.com.au,
	peterz@infradead.org, rusty@rustcorp.com.au,
	linux-kernel@vger.kernel.org
Subject: Re: [PATCH 3/3 -mm] generic-ipi: fix the race between generic_smp_call_function_*() and hotplug_cfd()
Date: Wed, 29 Jul 2009 16:31:20 -0700	[thread overview]
Message-ID: <20090729163120.2e27be41.akpm@linux-foundation.org> (raw)
In-Reply-To: <4A7000FF.6040402@cn.fujitsu.com>

On Wed, 29 Jul 2009 15:57:51 +0800
Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> wrote:

> It have race between generic_smp_call_function_*() and hotplug_cfd()
> in many cases, see below examples:
> 
> 1: hotplug_cfd() can free cfd->cpumask, the system will crash if the
>    cpu's cfd still in the call_function list:
>   
>   
>       CPU A:                         CPU B
>    
>  smp_call_function_many()	    ......
>    cpu_down()                      ......
>   hotplug_cfd() ->                 ......
>  free_cpumask_var(cfd->cpumask)  (receive function IPI interrupte)
>                                 /* read cfd->cpumask */           
>                           generic_smp_call_function_interrupt() ->
>                          cpumask_test_and_clear_cpu(cpu, data->cpumask)
> 
>                          	CRASH!!!
> 
> 2: It's not handle call_function list when cpu down, It's will lead to
>    dead-wait if other path is waiting this cpu to execute function
>    
>     CPU A:                           CPU B
> 
>  smp_call_function_many(wait=0)					
>         ......			    CPU B down
>    smp_call_function_many() -->  (cpu down before recevie function
>     csd_lock(&data->csd);         IPI interrupte)
> 
>     DEAD-WAIT!!!!
>        
>   So, CPU A will dead-wait in csd_lock(), the same as
>   smp_call_function_single()
> 
> Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
> ---
>  kernel/smp.c |  140 ++++++++++++++++++++++++++++++++-------------------------
>  1 files changed, 79 insertions(+), 61 deletions(-)
> 

It was unfortunate that this patch moved a screenful of code around and
changed that code at the same time - it makes it hard to see what the
functional change was.

So I split this patch into two.  The first patch simply moves
hotplug_cfd() to the end of the file and the second makes the
functional changes.  The second patch is below, for easier review.

Do we think that this patch should be merged into 2.6.31?  2.6.30.x?



From: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>

There is a race between generic_smp_call_function_*() and hotplug_cfd() in
many cases, see below examples:

1: hotplug_cfd() can free cfd->cpumask, the system will crash if the
   cpu's cfd still in the call_function list:


      CPU A:                         CPU B

 smp_call_function_many()	    ......
   cpu_down()                      ......
  hotplug_cfd() ->                 ......
 free_cpumask_var(cfd->cpumask)  (receive function IPI interrupte)
                                /* read cfd->cpumask */
                          generic_smp_call_function_interrupt() ->
                         cpumask_test_and_clear_cpu(cpu, data->cpumask)

                         	CRASH!!!

2: It's not handle call_function list when cpu down, It's will lead to
   dead-wait if other path is waiting this cpu to execute function

    CPU A:                           CPU B

 smp_call_function_many(wait=0)
        ......			    CPU B down
   smp_call_function_many() -->  (cpu down before recevie function
    csd_lock(&data->csd);         IPI interrupte)

    DEAD-WAIT!!!!

  So, CPU A will dead-wait in csd_lock(), the same as
  smp_call_function_single()

Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
---

 kernel/smp.c |   38 ++++++++++++++++++++++++++++----------
 1 file changed, 28 insertions(+), 10 deletions(-)

diff -puN kernel/smp.c~generic-ipi-fix-the-race-between-generic_smp_call_function_-and-hotplug_cfd kernel/smp.c
--- a/kernel/smp.c~generic-ipi-fix-the-race-between-generic_smp_call_function_-and-hotplug_cfd
+++ a/kernel/smp.c
@@ -116,14 +116,10 @@ void generic_exec_single(int cpu, struct
 		csd_lock_wait(data);
 }
 
-/*
- * Invoked by arch to handle an IPI for call function. Must be called with
- * interrupts disabled.
- */
-void generic_smp_call_function_interrupt(void)
+static void
+__generic_smp_call_function_interrupt(int cpu, int run_callbacks)
 {
 	struct call_function_data *data;
-	int cpu = smp_processor_id();
 
 	/*
 	 * Ensure entry is visible on call_function_queue after we have
@@ -169,12 +165,18 @@ void generic_smp_call_function_interrupt
 }
 
 /*
- * Invoked by arch to handle an IPI for call function single. Must be
- * called from the arch with interrupts disabled.
+ * Invoked by arch to handle an IPI for call function. Must be called with
+ * interrupts disabled.
  */
-void generic_smp_call_function_single_interrupt(void)
+void generic_smp_call_function_interrupt(void)
+{
+	__generic_smp_call_function_interrupt(smp_processor_id(), 1);
+}
+
+static void
+__generic_smp_call_function_single_interrupt(int cpu, int run_callbacks)
 {
-	struct call_single_queue *q = &__get_cpu_var(call_single_queue);
+	struct call_single_queue *q = &per_cpu(call_single_queue, cpu);
 	unsigned int data_flags;
 	LIST_HEAD(list);
 
@@ -205,6 +207,15 @@ void generic_smp_call_function_single_in
 	}
 }
 
+/*
+ * Invoked by arch to handle an IPI for call function single. Must be
+ * called from the arch with interrupts disabled.
+ */
+void generic_smp_call_function_single_interrupt(void)
+{
+	__generic_smp_call_function_single_interrupt(smp_processor_id(), 1);
+}
+
 static DEFINE_PER_CPU(struct call_single_data, csd_data);
 
 /*
@@ -456,6 +467,7 @@ static int
 hotplug_cfd(struct notifier_block *nfb, unsigned long action, void *hcpu)
 {
 	long cpu = (long)hcpu;
+	unsigned long flags;
 	struct call_function_data *cfd = &per_cpu(cfd_data, cpu);
 
 	switch (action) {
@@ -472,6 +484,12 @@ hotplug_cfd(struct notifier_block *nfb, 
 
 	case CPU_DEAD:
 	case CPU_DEAD_FROZEN:
+		local_irq_save(flags);
+		__generic_smp_call_function_interrupt(cpu, 0);
+		__generic_smp_call_function_single_interrupt(cpu, 0);
+		local_irq_restore(flags);
+
+		csd_lock_wait(&cfd->csd);
 		free_cpumask_var(cfd->cpumask);
 		break;
 #endif
_


  reply	other threads:[~2009-07-29 23:31 UTC|newest]

Thread overview: 13+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2009-07-24  9:50 [PATCH] generic-ipi: make struct call_function_data lockless Xiao Guangrong
2009-07-27 22:00 ` Andrew Morton
2009-07-29  7:08 ` Jens Axboe
2009-07-29  7:53 ` [PATCH 1/3 -mm] generic-ipi: fix hotplug_cfd() Xiao Guangrong
2009-07-29  7:55   ` [PATCH 2/3 -mm] generic-ipi: cleanup for generic_smp_call_function_interrupt() Xiao Guangrong
2009-07-29  7:57   ` [PATCH 3/3 -mm] generic-ipi: fix the race between generic_smp_call_function_*() and hotplug_cfd() Xiao Guangrong
2009-07-29 23:31     ` Andrew Morton [this message]
2009-07-30  3:31       ` Xiao Guangrong
2009-07-30  6:50       ` Peter Zijlstra
2009-07-30  8:11         ` Xiao Guangrong
2009-07-30  8:23           ` Li Zefan
2009-07-29 23:27   ` [PATCH 1/3 -mm] generic-ipi: fix hotplug_cfd() Andrew Morton
2009-07-30  1:18     ` Li Zefan

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20090729163120.2e27be41.akpm@linux-foundation.org \
    --to=akpm@linux-foundation.org \
    --cc=jens.axboe@oracle.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=mingo@elte.hu \
    --cc=nickpiggin@yahoo.com.au \
    --cc=peterz@infradead.org \
    --cc=rusty@rustcorp.com.au \
    --cc=xiaoguangrong@cn.fujitsu.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox