Re: [PATCH 2/2] usb: host: xhci: Handle the right timeout command

[OGAWA Hirofumi]

Mathias Nyman <mathias.nyman@xxxxxxxxxxxxxxx> writes:

>> Below is the latest code. I put my comments in line.
>>
>>   322 static int xhci_abort_cmd_ring(struct xhci_hcd *xhci)
>>   323 {
>>   324         u64 temp_64;
>>   325         int ret;
>>   326
>>   327         xhci_dbg(xhci, "Abort command ring\n");
>>   328
>>   329         reinit_completion(&xhci->cmd_ring_stop_completion);
>>   330
>>   331         temp_64 = xhci_read_64(xhci, &xhci->op_regs->cmd_ring);
>>   332         xhci_write_64(xhci, temp_64 | CMD_RING_ABORT,
>>   333                         &xhci->op_regs->cmd_ring);
>>
>> We should hold xhci->lock when we are modifying xhci registers
>> at runtime.
>>
>
> Makes sense, but we need to unlock it before sleeping or waiting for
> completion.  I need to look into that in more detail.
>
> But this was an issue already before these changes.

We set CMD_RING_STATE_ABORTED state under locking. I'm not checking what
is for taking lock for register though, I guess it should be enough just
lock around of read=>write of ->cmd_ring if need lock.

[Rather ->cmd_ring user should check CMD_RING_STATE_ABORTED state.]

> But then again I really like OGAWA Hiroumi's solution that separates the
> command ring stopping from aborting commands and restarting the ring.
>
> The current way of always restarting the command ring as a response to
> a stop command ring command really limits its usage.
>
> So, with this in mind most reasonable would be to
> 1. fix the lock to cover abort+CRR check, and send it to usb-linus +stable
> 2. rebase OGAWA Hirofumi's changes on top of that, and send to usb-linus only
> 3. remove unnecessary second abort try as a separate patch, send to usb-next
> 4. remove polling for the Command ring running (CRR), waiting for completion
>     is enough, if completion times out then we can check CRR. for usb-next

I think we should check both of CRR and even of stop completion. Because
CRR and stop completion was not same time (both can be first).

Thanks.
-- 
OGAWA Hirofumi <hirofumi@xxxxxxxxxxxxxxxxxx>