mmc: dw_mmc: Add locking around cmd11 timer
authorDoug Anderson <dianders@chromium.org>
Fri, 3 Apr 2015 18:13:07 +0000 (11:13 -0700)
committerUlf Hansson <ulf.hansson@linaro.org>
Thu, 9 Apr 2015 07:08:32 +0000 (09:08 +0200)
It is possible for the cmd11 interrupt to fire and delete the
cmd11_timer before the cmd11_timer was actually setup.  Let's fix this
race by adding a few spinlocks.  Note that the race wasn't seen in
practice without adding some printk statements, but it still seems
wise to fix.

Fixes: 5c935165da79 ("mmc: dw_mmc: Add a timeout for sending CMD11")
Signed-off-by: Doug Anderson <dianders@chromium.org>
Signed-off-by: Jaehoon Chung <jh80.chung@samsung.com>
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
drivers/mmc/host/dw_mmc.c

index c2b568352ce321ff9a2b04c82ed8c767bdd7dd07..38b29265cc7c7625484db40b4d38f2148a2c1aa6 100644 (file)
@@ -1023,14 +1023,23 @@ static void __dw_mci_start_request(struct dw_mci *host,
        dw_mci_start_command(host, cmd, cmdflags);
 
        if (cmd->opcode == SD_SWITCH_VOLTAGE) {
+               unsigned long irqflags;
+
                /*
                 * Databook says to fail after 2ms w/ no response, but evidence
                 * shows that sometimes the cmd11 interrupt takes over 130ms.
                 * We'll set to 500ms, plus an extra jiffy just in case jiffies
                 * is just about to roll over.
+                *
+                * We do this whole thing under spinlock and only if the
+                * command hasn't already completed (indicating the the irq
+                * already ran so we don't want the timeout).
                 */
-               mod_timer(&host->cmd11_timer,
-                         jiffies + msecs_to_jiffies(500) + 1);
+               spin_lock_irqsave(&host->irq_lock, irqflags);
+               if (!test_bit(EVENT_CMD_COMPLETE, &host->pending_events))
+                       mod_timer(&host->cmd11_timer,
+                               jiffies + msecs_to_jiffies(500) + 1);
+               spin_unlock_irqrestore(&host->irq_lock, irqflags);
        }
 
        if (mrq->stop)
@@ -2160,11 +2169,20 @@ static irqreturn_t dw_mci_interrupt(int irq, void *dev_id)
                /* Check volt switch first, since it can look like an error */
                if ((host->state == STATE_SENDING_CMD11) &&
                    (pending & SDMMC_INT_VOLT_SWITCH)) {
-                       del_timer(&host->cmd11_timer);
+                       unsigned long irqflags;
 
                        mci_writel(host, RINTSTS, SDMMC_INT_VOLT_SWITCH);
                        pending &= ~SDMMC_INT_VOLT_SWITCH;
+
+                       /*
+                        * Hold the lock; we know cmd11_timer can't be kicked
+                        * off after the lock is released, so safe to delete.
+                        */
+                       spin_lock_irqsave(&host->irq_lock, irqflags);
                        dw_mci_cmd_interrupt(host, pending);
+                       spin_unlock_irqrestore(&host->irq_lock, irqflags);
+
+                       del_timer(&host->cmd11_timer);
                }
 
                if (pending & DW_MCI_CMD_ERROR_FLAGS) {