]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/commitdiff
mmc: queue: prevent soft lockups on PREEMPT=n
authorRabin Vincent <rabin.vincent@axis.com>
Sun, 14 Jun 2015 17:26:11 +0000 (19:26 +0200)
committerUlf Hansson <ulf.hansson@linaro.org>
Thu, 18 Jun 2015 07:21:04 +0000 (09:21 +0200)
On systems with CONFIG_PREEMPT=n, under certain circumstances, mmcqd
can continuously process requests for several seconds without blocking,
triggering the soft lockup watchdog.  For example, this can happen if
mmcqd runs on the CPU which services the controller's interrupt, and
a process on a different CPU continuously writes to the MMC block
device.

 NMI watchdog: BUG: soft lockup - CPU#0 stuck for 22s! [mmcqd/0:664]
 CPU: 0 PID: 664 Comm: mmcqd/0 Not tainted 4.1.0-rc7+ #4
 PC is at _raw_spin_unlock_irqrestore+0x24/0x28
 LR is at mmc_start_request+0x104/0x134
 ...
 [<805112a8>] (_raw_spin_unlock_irqrestore) from [<803db664>] (mmc_start_request+0x104/0x134)
 [<803db664>] (mmc_start_request) from [<803dc008>] (mmc_start_req+0x274/0x394)
 [<803dc008>] (mmc_start_req) from [<803eb2c4>] (mmc_blk_issue_rw_rq+0xd0/0xb98)
 [<803eb2c4>] (mmc_blk_issue_rw_rq) from [<803ebe8c>] (mmc_blk_issue_rq+0x100/0x470)
 [<803ebe8c>] (mmc_blk_issue_rq) from [<803ecab8>] (mmc_queue_thread+0xd0/0x170)
 [<803ecab8>] (mmc_queue_thread) from [<8003fd14>] (kthread+0xe0/0xfc)
 [<8003fd14>] (kthread) from [<8000f768>] (ret_from_fork+0x14/0x2c)

Fix it by adding a cond_resched() in the request handling loop so that
other processes get a chance to run.

Signed-off-by: Rabin Vincent <rabin.vincent@axis.com>
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
drivers/mmc/card/queue.c

index d34e09b4161e66c1d757e9eabc6b9964c7a6fc14..b5a2b145d89f6b32c42c67e25cc029c1ef6744ea 100644 (file)
@@ -68,6 +68,7 @@ static int mmc_queue_thread(void *d)
                        set_current_state(TASK_RUNNING);
                        cmd_flags = req ? req->cmd_flags : 0;
                        mq->issue_fn(mq, req);
+                       cond_resched();
                        if (mq->flags & MMC_QUEUE_NEW_REQUEST) {
                                mq->flags &= ~MMC_QUEUE_NEW_REQUEST;
                                continue; /* fetch again */