summaryrefslogtreecommitdiffstats
path: root/drivers/staging
diff options
context:
space:
mode:
authorChristian Gromm <christian.gromm@microchip.com>2016-08-19 11:12:54 +0200
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>2016-08-21 16:53:57 +0200
commitbf9503f11dedf40158a8a5847f2d482bc15cd82b (patch)
tree7319082a573eb27c01ede556eed44ded6dd9ef2e /drivers/staging
parentcf05918370006fbdb6dc5708ca9951ee8093b8ef (diff)
downloadop-kernel-dev-bf9503f11dedf40158a8a5847f2d482bc15cd82b.zip
op-kernel-dev-bf9503f11dedf40158a8a5847f2d482bc15cd82b.tar.gz
staging: most: hdm-usb: fix race between enqueue and most_stop_enqueue
The "broken in pipe" handler of the USB-HDM calls most_stop_enqueue() to stop the MBO traffic before returning all MBOs back to the Mostcore. As the enqueue() call from the Mostcore may run in parallel with the most_stop_enqueue(), the HDM may run into the inconsistent state and crash the kernel. This patch synchronizes enqueue(), most_stop_enqueue() and most_resume_enqueue() with a mutex, hence avoiding the race condition. Signed-off-by: Andrey Shvetsov <andrey.shvetsov@k2l.de> Signed-off-by: Christian Gromm <christian.gromm@microchip.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Diffstat (limited to 'drivers/staging')
-rw-r--r--drivers/staging/most/hdm-usb/hdm_usb.c3
-rw-r--r--drivers/staging/most/mostcore/core.c53
2 files changed, 38 insertions, 18 deletions
diff --git a/drivers/staging/most/hdm-usb/hdm_usb.c b/drivers/staging/most/hdm-usb/hdm_usb.c
index 7f00aaf..8d8c72c 100644
--- a/drivers/staging/most/hdm-usb/hdm_usb.c
+++ b/drivers/staging/most/hdm-usb/hdm_usb.c
@@ -415,7 +415,6 @@ static void hdm_write_completion(struct urb *urb)
switch (urb->status) {
case -EPIPE:
dev_warn(dev, "Broken OUT pipe detected\n");
- most_stop_enqueue(&mdev->iface, channel);
mdev->is_channel_healthy[channel] = false;
mbo->status = MBO_E_INVAL;
mdev->clear_work[channel].pipe = urb->pipe;
@@ -578,7 +577,6 @@ static void hdm_read_completion(struct urb *urb)
switch (urb->status) {
case -EPIPE:
dev_warn(dev, "Broken IN pipe detected\n");
- most_stop_enqueue(&mdev->iface, channel);
mdev->is_channel_healthy[channel] = false;
mbo->status = MBO_E_INVAL;
mdev->clear_work[channel].pipe = urb->pipe;
@@ -928,6 +926,7 @@ static void wq_clear_halt(struct work_struct *wq_obj)
int pipe = clear_work->pipe;
mutex_lock(&mdev->io_mutex);
+ most_stop_enqueue(&mdev->iface, channel);
free_anchored_buffers(mdev, channel, MBO_E_INVAL);
if (usb_clear_halt(mdev->usb_device, pipe))
dev_warn(&mdev->usb_device->dev, "Failed to reset endpoint.\n");
diff --git a/drivers/staging/most/mostcore/core.c b/drivers/staging/most/mostcore/core.c
index b03cdc9..db0606ca 100644
--- a/drivers/staging/most/mostcore/core.c
+++ b/drivers/staging/most/mostcore/core.c
@@ -51,6 +51,7 @@ struct most_c_obj {
u16 channel_id;
bool is_poisoned;
struct mutex start_mutex;
+ struct mutex nq_mutex; /* nq thread synchronization */
int is_starving;
struct most_interface *iface;
struct most_inst_obj *inst;
@@ -1131,18 +1132,18 @@ static inline void trash_mbo(struct mbo *mbo)
spin_unlock_irqrestore(&c->fifo_lock, flags);
}
-static struct mbo *get_hdm_mbo(struct most_c_obj *c)
+static bool hdm_mbo_ready(struct most_c_obj *c)
{
- unsigned long flags;
- struct mbo *mbo;
+ bool empty;
- spin_lock_irqsave(&c->fifo_lock, flags);
- if (c->enqueue_halt || list_empty(&c->halt_fifo))
- mbo = NULL;
- else
- mbo = list_pop_mbo(&c->halt_fifo);
- spin_unlock_irqrestore(&c->fifo_lock, flags);
- return mbo;
+ if (c->enqueue_halt)
+ return false;
+
+ spin_lock_irq(&c->fifo_lock);
+ empty = list_empty(&c->halt_fifo);
+ spin_unlock_irq(&c->fifo_lock);
+
+ return !empty;
}
static void nq_hdm_mbo(struct mbo *mbo)
@@ -1160,20 +1161,32 @@ static int hdm_enqueue_thread(void *data)
{
struct most_c_obj *c = data;
struct mbo *mbo;
+ int ret;
typeof(c->iface->enqueue) enqueue = c->iface->enqueue;
while (likely(!kthread_should_stop())) {
wait_event_interruptible(c->hdm_fifo_wq,
- (mbo = get_hdm_mbo(c)) ||
+ hdm_mbo_ready(c) ||
kthread_should_stop());
- if (unlikely(!mbo))
+ mutex_lock(&c->nq_mutex);
+ spin_lock_irq(&c->fifo_lock);
+ if (unlikely(c->enqueue_halt || list_empty(&c->halt_fifo))) {
+ spin_unlock_irq(&c->fifo_lock);
+ mutex_unlock(&c->nq_mutex);
continue;
+ }
+
+ mbo = list_pop_mbo(&c->halt_fifo);
+ spin_unlock_irq(&c->fifo_lock);
if (c->cfg.direction == MOST_CH_RX)
mbo->buffer_length = c->cfg.buffer_size;
- if (unlikely(enqueue(mbo->ifp, mbo->hdm_channel_id, mbo))) {
+ ret = enqueue(mbo->ifp, mbo->hdm_channel_id, mbo);
+ mutex_unlock(&c->nq_mutex);
+
+ if (unlikely(ret)) {
pr_err("hdm enqueue failed\n");
nq_hdm_mbo(mbo);
c->hdm_enqueue_task = NULL;
@@ -1759,6 +1772,7 @@ struct kobject *most_register_interface(struct most_interface *iface)
init_completion(&c->cleanup);
atomic_set(&c->mbo_ref, 0);
mutex_init(&c->start_mutex);
+ mutex_init(&c->nq_mutex);
list_add_tail(&c->list, &inst->channel_list);
}
pr_info("registered new MOST device mdev%d (%s)\n",
@@ -1824,8 +1838,12 @@ void most_stop_enqueue(struct most_interface *iface, int id)
{
struct most_c_obj *c = get_channel_by_iface(iface, id);
- if (likely(c))
- c->enqueue_halt = true;
+ if (!c)
+ return;
+
+ mutex_lock(&c->nq_mutex);
+ c->enqueue_halt = true;
+ mutex_unlock(&c->nq_mutex);
}
EXPORT_SYMBOL_GPL(most_stop_enqueue);
@@ -1841,9 +1859,12 @@ void most_resume_enqueue(struct most_interface *iface, int id)
{
struct most_c_obj *c = get_channel_by_iface(iface, id);
- if (unlikely(!c))
+ if (!c)
return;
+
+ mutex_lock(&c->nq_mutex);
c->enqueue_halt = false;
+ mutex_unlock(&c->nq_mutex);
wake_up_interruptible(&c->hdm_fifo_wq);
}
OpenPOWER on IntegriCloud