mmc: tmio: always unmap DMA before waiting for interrupt
authorWolfram Sang <wsa+renesas@sang-engineering.com>
Thu, 16 Mar 2017 10:56:02 +0000 (11:56 +0100)
committerUlf Hansson <ulf.hansson@linaro.org>
Mon, 24 Apr 2017 19:41:40 +0000 (21:41 +0200)
In the (maybe academical) case, we don't get a DATAEND interrupt after
DMA completed, we will wait endlessly for the completion to complete.
This is not bad per se, since we have a more generic completion tracking
a timeout. In that rare case, however, the DMA buffer will not get
unmapped and we have a leak. Reorder the code, so unmapping will always
take place.

Signed-off-by: Wolfram Sang <wsa+renesas@sang-engineering.com>
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
drivers/mmc/host/tmio_mmc_dma.c

index c7684fa91f1f9c46a2cc043ad9f7df9a4b052723..e2093db2b7ffce5fc326a615be59c869d63cf845 100644 (file)
@@ -47,8 +47,6 @@ static void tmio_mmc_dma_callback(void *arg)
 {
        struct tmio_mmc_host *host = arg;
 
-       wait_for_completion(&host->dma_dataend);
-
        spin_lock_irq(&host->lock);
 
        if (!host->data)
@@ -63,6 +61,11 @@ static void tmio_mmc_dma_callback(void *arg)
                             host->sg_ptr, host->sg_len,
                             DMA_TO_DEVICE);
 
+       spin_unlock_irq(&host->lock);
+
+       wait_for_completion(&host->dma_dataend);
+
+       spin_lock_irq(&host->lock);
        tmio_mmc_do_data_irq(host);
 out:
        spin_unlock_irq(&host->lock);