ASoC: Intel: Skylake: Fix IPC rx_list corruption
authorPardha Saradhi K <pardha.saradhi.kesapragada@intel.com>
Mon, 15 May 2017 14:14:29 +0000 (19:44 +0530)
committerMark Brown <broonie@kernel.org>
Wed, 24 May 2017 17:25:12 +0000 (18:25 +0100)
In SKL+ platforms, all IPC commands are serialised, i.e. the driver sends
a new IPC to DSP, only after receiving a reply from the firmware for the
current IPC.

Hence it seems apparent that there is only a single modifier of the IPC RX
List. However, during an IPC timeout case in a multithreaded environment,
there is a possibility of the list element being deleted two times if not
properly protected.

So, use spin lock save/restore to prevent rx_list corruption.

Signed-off-by: Pardha Saradhi K <pardha.saradhi.kesapragada@intel.com>
Signed-off-by: Subhransu S. Prusty <subhransu.s.prusty@intel.com>
Acked-by: Vinod Koul <vinod.koul@intel.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
sound/soc/intel/skylake/skl-sst-ipc.c

index 58c525096a7cbcd6ea4fd833d06e03a0127201fe..498b15345b1a657d608a3fcff773dae206e819b0 100644 (file)
@@ -413,8 +413,11 @@ static void skl_ipc_process_reply(struct sst_generic_ipc *ipc,
        u32 reply = header.primary & IPC_GLB_REPLY_STATUS_MASK;
        u64 *ipc_header = (u64 *)(&header);
        struct skl_sst *skl = container_of(ipc, struct skl_sst, ipc);
+       unsigned long flags;
 
+       spin_lock_irqsave(&ipc->dsp->spinlock, flags);
        msg = skl_ipc_reply_get_msg(ipc, *ipc_header);
+       spin_unlock_irqrestore(&ipc->dsp->spinlock, flags);
        if (msg == NULL) {
                dev_dbg(ipc->dev, "ipc: rx list is empty\n");
                return;
@@ -456,8 +459,10 @@ static void skl_ipc_process_reply(struct sst_generic_ipc *ipc,
                }
        }
 
+       spin_lock_irqsave(&ipc->dsp->spinlock, flags);
        list_del(&msg->list);
        sst_ipc_tx_msg_reply_complete(ipc, msg);
+       spin_unlock_irqrestore(&ipc->dsp->spinlock, flags);
 }
 
 irqreturn_t skl_dsp_irq_thread_handler(int irq, void *context)