ARM: 7216/1: mmc: mmci: Do not release spinlock in request_end

The patch "mmc: core: move ->request() call from atomic context",
is the reason to why this change is possible. This simplifies the
error handling code execution path quite a lot and potentially also
fixes some error handling hang problems.

Tested-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Ulf Hansson <ulf.hansson@stericsson.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>

drivers/mmc/host/mmci.c

@@ -166,14 +166,8 @@ mmci_request_end(struct mmci_host *host, struct mmc_request *mrq)
 	host->mrq = NULL;
 	host->cmd = NULL;
 
-	/*
-	 * Need to drop the host lock here; mmc_request_done may call
-	 * back into the driver...
-	 */
-	spin_unlock(&host->lock);
 	pm_runtime_put(mmc_dev(host->mmc));
 	mmc_request_done(host->mmc, mrq);
-	spin_lock(&host->lock);
 }
 
 static void mmci_set_mask1(struct mmci_host *host, unsigned int mask)