[PATCH 2/2] mt76: mt7921s: fix the device cannot sleep deeply in suspend

[Lorenzo Bianconi]

> From: Sean Wang <sean.wang at mediatek.com>
> 
> >> From: Sean Wang <sean.wang at mediatek.com>
> >>
> >>
> >> <snip>
> >>
> >> >> >>
> >> >> >> -		if (test_bit(MT76_MCU_RESET, &dev->phy.state)) {
> >> >> >> -			if (!mt76s_txqs_empty(dev))
> >> >> >> -				continue;
> >> >> >> -			else
> >> >> >> -				wake_up(&sdio->wait);
> >> >> >> -		}
> >> >> >>	} while (nframes > 0);
> >> >> >>
> >> >> >> +	if (test_bit(MT76_MCU_RESET, &dev->phy.state) &&
> >> >> >> +	    mt76s_txqs_empty(dev))
> >> >> >> +		wake_up(&sdio->wait);
> >> >> >> +
> >> >>
> >> >> If doing so, mt76s_txqs_empty may not always be true because
> >> >> enqueuing packets to q_tx or MCU command to q_mcu simultanenously
> >> >> from the other contexts in different cpu is possible.
> >> >>
> >> >> It seemed to me we should check it for each iteration to guarantee
> >> >> that we can wake up the one that is waiting for the all the queues are empty at some time.
> >> >
> >> >IIUC what we are interested here is there are no queued frames into the hw queues during suspend or reset, right?
> >>
> >> That is not completely true. Take the suspend procedure on mt7921s as an example.
> >>
> >> That should be "There are no queued frames into the hw queues right after mt76_connac_mcu_set_hif_suspend."
> >>
> >> The MCU data and WiFi are all handled in mt76s_txrx_worker so we
> >> should synchronize all of the Tx queues are all empty and then handle
> >> mt76_connac_mcu_set_hif_suspend to guarantee mt76_connac_mcu_set_hif_suspend is the last one to access the SDIO bus and there is no frame that accesses SDIO bus afterhand.
> >
> >ack, correct, "there are no queued frames into the hw queues right after mt76_connac_mcu_set_hif_suspend."
> >What I mean is we are not really checking there are no frames in the hw queue here, but mt76 sdio has processed all the frames, got my point? maybe it is what we are looking for..
> 
> It seemed to me there is no a way to check if no frames in the hw queue for mt7921s
> 
> I think we can discuss the topic "it is ok move the check out of the loop" with another patch in the future
> 
> because it seemed not related to the patch.

sure, fine to me.

Regards,
Lorenzo

> 
> >
> >Regards,
> >Lorenzo
> >
> >>
> >> >
> >> >>
> >> >> >>	/* enable interrupt */
> >> >> >>	sdio_writel(sdio->func, WHLPCR_INT_EN_SET, MCR_WHLPCR, NULL);
> >> >> >>	sdio_release_host(sdio->func);
> >> >> >>
> >> >> >> Regards,
> >> >> >> Lorenzo
> >> >> >>
> >> >> >> > --
> >> >> >> > 2.25.1
> >> >> >> >
> >> >> >
> >> >> >
> >> >> >
> >> >>
> >> >
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 228 bytes
Desc: not available
URL: <http://lists.infradead.org/pipermail/linux-mediatek/attachments/20211122/1250f3ec/attachment.sig>