Re: [PATCH 2/2] Bluetooth: hci_qca: wcn3990: Drop baudrate change vendor event

[Matthias Kaehlcke]

On Tue, Apr 02, 2019 at 11:05:01AM -0700, Matthias Kaehlcke wrote:
> On Tue, Apr 02, 2019 at 05:32:54PM +0530, Balakrishna Godavarthi wrote:
> > Hi Matthias,
> > 
> > On 2019-04-01 22:42, Matthias Kaehlcke wrote:
> > > On Mon, Apr 01, 2019 at 01:48:23PM +0530, Balakrishna Godavarthi wrote:
> > > > Hi Matthias,
> > > > 
> > > > On 2019-04-01 13:29, Balakrishna Godavarthi wrote:
> > > > > Hi Matthias,
> > > > >
> > > > > Sorry for the late reply i was on vacation.
> > > > >
> > > > > On 2019-03-08 05:00, Matthias Kaehlcke wrote:
> > > > > > On Thu, Mar 07, 2019 at 10:20:09AM -0800, Matthias Kaehlcke wrote:
> > > > > > > Hi Balakrishna,
> > > > > > >
> > > > > > > On Thu, Mar 07, 2019 at 10:35:08AM +0530, Balakrishna Godavarthi
> > > > > > > wrote:
> > > > > > > > hi Matthias,
> > > > > > > >
> > > > > > > > On 2019-03-07 06:10, Matthias Kaehlcke wrote:
> > > > > > > > > Firmware download to the WCN3990 often fails with a 'TLV response size
> > > > > > > > > mismatch' error:
> > > > > > > > >
> > > > > > > > > [  133.064659] Bluetooth: hci0: setting up wcn3990
> > > > > > > > > [  133.489150] Bluetooth: hci0: QCA controller version 0x02140201
> > > > > > > > > [  133.495245] Bluetooth: hci0: QCA Downloading qca/crbtfw21.tlv
> > > > > > > > > [  133.507214] Bluetooth: hci0: QCA TLV response size mismatch
> > > > > > > > > [  133.513265] Bluetooth: hci0: QCA Failed to download patch (-84)
> > > > > > > > >
> > > > > > > > > This is caused by a vendor event that corresponds to an earlier command
> > > > > > > > > to change the baudrate. The event is not processed in the context of the
> > > > > > > > > baudrate change and later interpreted as response to the firmware
> > > > > > > > > download command (which is also a vendor command), but the driver
> > > > > > > > > detects
> > > > > > > > > that the event doesn't have the expected amount of associated data.
> > > > > > > > >
> > > > > > > > > More details:
> > > > > > > > >
> > > > > > > > > For the WCN3990 the vendor command for a baudrate change isn't sent as
> > > > > > > > > synchronous HCI command, because the controller sends the corresponding
> > > > > > > > > vendor event with the new baudrate. The event is received and decoded
> > > > > > > > > after the baudrate change of the host port.
> > > > > > > > >
> > > > > > > > > Identify the 'unused' event when it is received and don't add it to
> > > > > > > > > the queue of RX frames.
> > > > > > > > >
> > > > > > > > > Signed-off-by: Matthias Kaehlcke <mka@xxxxxxxxxxxx>
> > > > > > > > > ---
> > > > > > > >
> > > > > > > > ...
> > > > > > > >
> > > > > > > > Can you test by reverting this change "94d6671473924".
> > > > > > >
> > > > > > > The issue is still reproducible.
> > > > > > >
> > > > > > > > We need at least 15ms minimum delay for the soc to change its baud rate and
> > > > > > > > respond to the with command complete event.
> > > > > > >
> > > > > > > The baudrate change has clearly been successful when the problem is
> > > > > > > observed, since the host receives the vendor event with the new
> > > > > > > baudrate.
> > > > > >
> > > > > > I forgot to mention this earlier: the controller doesn't send a
> > > > > > command complete event for the command, or at least not a correct
> > > > > > one.
> > > > > >
> > > > > > That's the data that is received:
> > > > > >
> > > > > > 04 0e 04 01 00 00 00
> > > > > >             ~~ ~~
> > > > > >
> > > > > [Bala]: can you share me the command sent and event recevied.
> > > > >  I see that we receive a command complete event for the baud rate
> > > > > change command.
> > > > >
> > > > > command sent: 01 48 fc 01 11
> > > > > vendor specific event: 04 ff 02 92 01
> > > > > command complete event: 04 0e 04 01 00 00 00.
> > > > >
> > > > >
> > > > >
> > > > > > This is *a* command complete event, but the opcode is 0x0000 instead
> > > > > > of the earlier command. The same happens for the firmware
> > > > > > download/read version command, which is the reason why the command
> > > > > > complete injection mess
> > > > > > (https://lore.kernel.org/patchwork/patch/1027955/) is needed in one
> > > > > > way or another.
> > > > > >
> > > > > [Bala]: fw download approach is different where we use
> > > > > __hci_cmd_sync() where as here we use hci_uart_tx_wakeup()
> > > > >         which directly calls the hci_uart_write_work(). so even we
> > > > > send an valid opcode or not for baudrate change will bot matter.
> > > > >
> > > > [Bala]: i miss understood the comment. Yes your true. in the all
> > > > vendor
> > > > commands SoC responds with an 0x0000 opcode.
> > > 
> > > And IIUC this is not compliant with the spec, or at least the BT core
> > > expects the actual opcode to consider the command to be completed.
> > 
> > [Bala]: Did you try increasing the the baud rate change timeout to 50ms
> > instead of 10ms.
> 
> It is even reproducible with the ROME timeout of 300ms.
> 
> The timeout doesn't help here. With 78e8fa2972e5 ("Bluetooth: hci_qca:
> Deassert RTS while baudrate change command") RTS is deasserted during
> the baudrate change, hence the controller only sends the response when
> RTS is asserted again. Before the event resulted in a frame reassembly
> error and the data was discarded.
> 
> >         i suspect it is an timing issue.
> 
> Timing is certainly also a factor here (the problem isn't seen
> always), but I don't think 'timing issue' is a proper description of
> this issue. The problem is an event that the Bluetooth core doesn't
> expect due to the hack of sending a raw command behind the core's back
> to work around the firmware 'feature' of sending the command response
> with the new baudrate.
> 
> Maybe a delay after re-asserting RTS can address is reliably, I don't
> remember if I already experimented with that in the past. If a delay
> can 'fix' the issue reliably I'm open to consider it for the sake of
> simplicity, but only with a detailed comment that describes the
> problem.

A short delay after re-asserting RTS indeed seems to 'fix' this, the
problem didn't reproduce in 5000+ iterations of binding/unbinding the
HCI.

I'm fine with either solution, the delay is slightly less code, then
again this patch (or v2:
https://lore.kernel.org/patchwork/patch/1050247/) also isn't overly
complex. Just noticed though that my post of v2 is messed up and I
would have to re-send it with proper format.

Marcel, please let me know if you prefer an msleep + comment or
actively dropping the event.

Thanks

Matthias