ASoC: Intel: Skylake: Fix IPC rx_list corruption

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 <[email protected]>
Signed-off-by: Subhransu S. Prusty <[email protected]>
Acked-by: Vinod Koul <[email protected]>
Signed-off-by: Mark Brown <[email protected]>
This commit is contained in:
Pardha Saradhi K 2017-05-15 19:44:29 +05:30 • committed by Mark Brown
commit 5cd1f5c321

View 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)