Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1753715AbcCHM2z (ORCPT ); Tue, 8 Mar 2016 07:28:55 -0500 Received: from hqemgate16.nvidia.com ([216.228.121.65]:6510 "EHLO hqemgate16.nvidia.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S933012AbcCHM2k (ORCPT ); Tue, 8 Mar 2016 07:28:40 -0500 X-PGP-Universal: processed; by hqnvupgp08.nvidia.com on Tue, 08 Mar 2016 04:28:07 -0800 From: Jon Hunter To: Mark Brown , CC: linux-spi@vger.kernel.org, linux-kernel@vger.kernel.org, linux-tegra@vger.kernel.org, Jon Hunter Subject: [PATCH V2] spi: core: Fix deadlock when sending messages Date: Tue, 8 Mar 2016 12:28:20 +0000 Message-ID: <1457440100-31546-1-git-send-email-jonathanh@nvidia.com> X-Mailer: git-send-email 2.1.4 In-Reply-To: <1457430543-15179-1-git-send-email-jonathanh@nvidia.com> References: <1457430543-15179-1-git-send-email-jonathanh@nvidia.com> X-NVConfidentiality: public MIME-Version: 1.0 Content-Type: text/plain Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 4524 Lines: 125 The function __spi_pump_messages() is called by spi_pump_messages() and __spi_sync(). The function __spi_sync() has an argument 'bus_locked' that indicates if it is called with the SPI bus mutex held or not. If 'bus_locked' is false then __spi_sync() will acquire the mutex itself. Commit 556351f14e74 ("spi: introduce accelerated read support for spi flash devices") made a change to acquire the SPI bus mutex within __spi_pump_messages(). However, this change did not check to see if the mutex is already held. If __spi_sync() is called with the mutex held (ie. 'bus_locked' is true), then a deadlock occurs when __spi_pump_messages() is called. Fix this deadlock by passing the 'bus_locked' state from __spi_sync() to __spi_pump_messages() and only acquire the mutex if not already held. In the case where __spi_pump_messages() is called from spi_pump_messages() it is assumed that the mutex is not held and so call __spi_pump_messages() with 'bus_locked' set to false. Finally, move the unlocking of the mutex to the end of the __spi_pump_messages() function to simplify the code and only call cond_resched() if there are no errors. Fixes: 556351f14e74 ("spi: introduce accelerated read support for spi flash devices") Signed-off-by: Jon Hunter --- Changes v1-v2: - Updated kernel-doc for new parameter added to __spi_pump_messages() drivers/spi/spi.c | 29 +++++++++++++++++------------ 1 file changed, 17 insertions(+), 12 deletions(-) diff --git a/drivers/spi/spi.c b/drivers/spi/spi.c index fe0196328aa0..78b7e899ce76 100644 --- a/drivers/spi/spi.c +++ b/drivers/spi/spi.c @@ -1053,6 +1053,7 @@ EXPORT_SYMBOL_GPL(spi_finalize_current_transfer); * __spi_pump_messages - function which processes spi message queue * @master: master to process queue for * @in_kthread: true if we are in the context of the message pump thread + * @bus_locked: true if the bus mutex is held when calling this function * * This function checks if there is any spi message in the queue that * needs processing and if so call out to the driver to initialize hardware @@ -1062,7 +1063,8 @@ EXPORT_SYMBOL_GPL(spi_finalize_current_transfer); * inside spi_sync(); the queue extraction handling at the top of the * function should deal with this safely. */ -static void __spi_pump_messages(struct spi_master *master, bool in_kthread) +static void __spi_pump_messages(struct spi_master *master, bool in_kthread, + bool bus_locked) { unsigned long flags; bool was_busy = false; @@ -1158,7 +1160,9 @@ static void __spi_pump_messages(struct spi_master *master, bool in_kthread) } } - mutex_lock(&master->bus_lock_mutex); + if (!bus_locked) + mutex_lock(&master->bus_lock_mutex); + trace_spi_message_start(master->cur_msg); if (master->prepare_message) { @@ -1168,8 +1172,7 @@ static void __spi_pump_messages(struct spi_master *master, bool in_kthread) "failed to prepare message: %d\n", ret); master->cur_msg->status = ret; spi_finalize_current_message(master); - mutex_unlock(&master->bus_lock_mutex); - return; + goto out; } master->cur_msg_prepared = true; } @@ -1178,21 +1181,23 @@ static void __spi_pump_messages(struct spi_master *master, bool in_kthread) if (ret) { master->cur_msg->status = ret; spi_finalize_current_message(master); - mutex_unlock(&master->bus_lock_mutex); - return; + goto out; } ret = master->transfer_one_message(master, master->cur_msg); if (ret) { dev_err(&master->dev, "failed to transfer one message from queue\n"); - mutex_unlock(&master->bus_lock_mutex); - return; + goto out; } - mutex_unlock(&master->bus_lock_mutex); + +out: + if (!bus_locked) + mutex_unlock(&master->bus_lock_mutex); /* Prod the scheduler in case transfer_one() was busy waiting */ - cond_resched(); + if (!ret) + cond_resched(); } /** @@ -1204,7 +1209,7 @@ static void spi_pump_messages(struct kthread_work *work) struct spi_master *master = container_of(work, struct spi_master, pump_messages); - __spi_pump_messages(master, true); + __spi_pump_messages(master, true, false); } static int spi_init_queue(struct spi_master *master) @@ -2814,7 +2819,7 @@ static int __spi_sync(struct spi_device *spi, struct spi_message *message, spi_sync_immediate); SPI_STATISTICS_INCREMENT_FIELD(&spi->statistics, spi_sync_immediate); - __spi_pump_messages(master, false); + __spi_pump_messages(master, false, bus_locked); } wait_for_completion(&done); -- 2.1.4