HID: intel_ish-hid: clarify locking in client code
authorArnd Bergmann <arnd@arndb.de>
Thu, 18 May 2017 20:21:41 +0000 (22:21 +0200)
committerJiri Kosina <jkosina@suse.cz>
Tue, 30 May 2017 12:11:52 +0000 (14:11 +0200)
I was trying to understand this code while working on a warning
fix and the locking made no sense: spin_lock_irqsave() is pointless
when run inside of an interrupt handler or nested inside of another
spin_lock_irq() or spin_lock_irqsave().

Here it turned out that the comment above the function is wrong,
as both recv_ishtp_cl_msg_dma() and recv_ishtp_cl_msg() can in fact
be called from a work queue rather than an ISR, so we do have to
use the irqsave() version once.

This fixes the comments accordingly, removes the misleading 'dev_flags'
variable and modifies the inner spinlock to not use 'irqsave'.

No functional change is intended, this is just for readability and
it slightly simplifies the object code.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
drivers/hid/intel-ish-hid/ishtp/client.c

index 78d393e616a402cc5c866d88c14d1fb1aa8397d8..f54689ee67e1feae952c5bb2639e35d734db0e16 100644 (file)
@@ -803,7 +803,7 @@ void ishtp_cl_send_msg(struct ishtp_device *dev, struct ishtp_cl *cl)
  * @ishtp_hdr: Pointer to message header
  *
  * Receive and dispatch ISHTP client messages. This function executes in ISR
- * context
+ * or work queue context
  */
 void recv_ishtp_cl_msg(struct ishtp_device *dev,
                       struct ishtp_msg_hdr *ishtp_hdr)
@@ -813,7 +813,6 @@ void recv_ishtp_cl_msg(struct ishtp_device *dev,
        struct ishtp_cl_rb *new_rb;
        unsigned char *buffer = NULL;
        struct ishtp_cl_rb *complete_rb = NULL;
-       unsigned long   dev_flags;
        unsigned long   flags;
        int     rb_count;
 
@@ -828,7 +827,7 @@ void recv_ishtp_cl_msg(struct ishtp_device *dev,
                goto    eoi;
        }
 
-       spin_lock_irqsave(&dev->read_list_spinlock, dev_flags);
+       spin_lock_irqsave(&dev->read_list_spinlock, flags);
        rb_count = -1;
        list_for_each_entry(rb, &dev->read_list.list, list) {
                ++rb_count;
@@ -840,8 +839,7 @@ void recv_ishtp_cl_msg(struct ishtp_device *dev,
 
                 /* If no Rx buffer is allocated, disband the rb */
                if (rb->buffer.size == 0 || rb->buffer.data == NULL) {
-                       spin_unlock_irqrestore(&dev->read_list_spinlock,
-                               dev_flags);
+                       spin_unlock_irqrestore(&dev->read_list_spinlock, flags);
                        dev_err(&cl->device->dev,
                                "Rx buffer is not allocated.\n");
                        list_del(&rb->list);
@@ -857,8 +855,7 @@ void recv_ishtp_cl_msg(struct ishtp_device *dev,
                 * back FC, so communication will be stuck anyway)
                 */
                if (rb->buffer.size < ishtp_hdr->length + rb->buf_idx) {
-                       spin_unlock_irqrestore(&dev->read_list_spinlock,
-                               dev_flags);
+                       spin_unlock_irqrestore(&dev->read_list_spinlock, flags);
                        dev_err(&cl->device->dev,
                                "message overflow. size %d len %d idx %ld\n",
                                rb->buffer.size, ishtp_hdr->length,
@@ -884,14 +881,13 @@ void recv_ishtp_cl_msg(struct ishtp_device *dev,
                         * the whole msg arrived, send a new FC, and add a new
                         * rb buffer for the next coming msg
                         */
-                       spin_lock_irqsave(&cl->free_list_spinlock, flags);
+                       spin_lock(&cl->free_list_spinlock);
 
                        if (!list_empty(&cl->free_rb_list.list)) {
                                new_rb = list_entry(cl->free_rb_list.list.next,
                                        struct ishtp_cl_rb, list);
                                list_del_init(&new_rb->list);
-                               spin_unlock_irqrestore(&cl->free_list_spinlock,
-                                       flags);
+                               spin_unlock(&cl->free_list_spinlock);
                                new_rb->cl = cl;
                                new_rb->buf_idx = 0;
                                INIT_LIST_HEAD(&new_rb->list);
@@ -900,8 +896,7 @@ void recv_ishtp_cl_msg(struct ishtp_device *dev,
 
                                ishtp_hbm_cl_flow_control_req(dev, cl);
                        } else {
-                               spin_unlock_irqrestore(&cl->free_list_spinlock,
-                                       flags);
+                               spin_unlock(&cl->free_list_spinlock);
                        }
                }
                /* One more fragment in message (even if this was last) */
@@ -914,7 +909,7 @@ void recv_ishtp_cl_msg(struct ishtp_device *dev,
                break;
        }
 
-       spin_unlock_irqrestore(&dev->read_list_spinlock, dev_flags);
+       spin_unlock_irqrestore(&dev->read_list_spinlock, flags);
        /* If it's nobody's message, just read and discard it */
        if (!buffer) {
                uint8_t rd_msg_buf[ISHTP_RD_MSG_BUF_SIZE];
@@ -941,7 +936,7 @@ eoi:
  * @hbm: hbm buffer
  *
  * Receive and dispatch ISHTP client messages using DMA. This function executes
- * in ISR context
+ * in ISR or work queue context
  */
 void recv_ishtp_cl_msg_dma(struct ishtp_device *dev, void *msg,
                           struct dma_xfer_hbm *hbm)
@@ -951,10 +946,10 @@ void recv_ishtp_cl_msg_dma(struct ishtp_device *dev, void *msg,
        struct ishtp_cl_rb *new_rb;
        unsigned char *buffer = NULL;
        struct ishtp_cl_rb *complete_rb = NULL;
-       unsigned long   dev_flags;
        unsigned long   flags;
 
-       spin_lock_irqsave(&dev->read_list_spinlock, dev_flags);
+       spin_lock_irqsave(&dev->read_list_spinlock, flags);
+
        list_for_each_entry(rb, &dev->read_list.list, list) {
                cl = rb->cl;
                if (!cl || !(cl->host_client_id == hbm->host_client_id &&
@@ -966,8 +961,7 @@ void recv_ishtp_cl_msg_dma(struct ishtp_device *dev, void *msg,
                 * If no Rx buffer is allocated, disband the rb
                 */
                if (rb->buffer.size == 0 || rb->buffer.data == NULL) {
-                       spin_unlock_irqrestore(&dev->read_list_spinlock,
-                               dev_flags);
+                       spin_unlock_irqrestore(&dev->read_list_spinlock, flags);
                        dev_err(&cl->device->dev,
                                "response buffer is not allocated.\n");
                        list_del(&rb->list);
@@ -983,8 +977,7 @@ void recv_ishtp_cl_msg_dma(struct ishtp_device *dev, void *msg,
                 * back FC, so communication will be stuck anyway)
                 */
                if (rb->buffer.size < hbm->msg_length) {
-                       spin_unlock_irqrestore(&dev->read_list_spinlock,
-                               dev_flags);
+                       spin_unlock_irqrestore(&dev->read_list_spinlock, flags);
                        dev_err(&cl->device->dev,
                                "message overflow. size %d len %d idx %ld\n",
                                rb->buffer.size, hbm->msg_length, rb->buf_idx);
@@ -1008,14 +1001,13 @@ void recv_ishtp_cl_msg_dma(struct ishtp_device *dev, void *msg,
                 * the whole msg arrived, send a new FC, and add a new
                 * rb buffer for the next coming msg
                 */
-               spin_lock_irqsave(&cl->free_list_spinlock, flags);
+               spin_lock(&cl->free_list_spinlock);
 
                if (!list_empty(&cl->free_rb_list.list)) {
                        new_rb = list_entry(cl->free_rb_list.list.next,
                                struct ishtp_cl_rb, list);
                        list_del_init(&new_rb->list);
-                       spin_unlock_irqrestore(&cl->free_list_spinlock,
-                               flags);
+                       spin_unlock(&cl->free_list_spinlock);
                        new_rb->cl = cl;
                        new_rb->buf_idx = 0;
                        INIT_LIST_HEAD(&new_rb->list);
@@ -1024,8 +1016,7 @@ void recv_ishtp_cl_msg_dma(struct ishtp_device *dev, void *msg,
 
                        ishtp_hbm_cl_flow_control_req(dev, cl);
                } else {
-                       spin_unlock_irqrestore(&cl->free_list_spinlock,
-                               flags);
+                       spin_unlock(&cl->free_list_spinlock);
                }
 
                /* One more fragment in message (this is always last) */
@@ -1038,7 +1029,7 @@ void recv_ishtp_cl_msg_dma(struct ishtp_device *dev, void *msg,
                break;
        }
 
-       spin_unlock_irqrestore(&dev->read_list_spinlock, dev_flags);
+       spin_unlock_irqrestore(&dev->read_list_spinlock, flags);
        /* If it's nobody's message, just read and discard it */
        if (!buffer) {
                dev_err(dev->devc, "Dropped Rx (DMA) msg - no request\n");