linux-kernel - Re: [PATCH v5 2/3] tty: serial: use uart_port

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Tue, 22 Nov 2022 09:25:36 +0100
From:   Jiri Slaby <jirislaby@...nel.org>
To:     Michael Walle <michael@...le.cc>
Cc:     afaerber@...e.de, alexandre.belloni@...tlin.com,
        claudiu.beznea@...rochip.com, festevam@...il.com,
        gregkh@...uxfoundation.org, ilpo.jarvinen@...ux.intel.com,
        kernel@...gutronix.de, linux-imx@....com,
        linux-kernel@...r.kernel.org, linux-serial@...r.kernel.org,
        liviu.dudau@....com, lorenzo.pieralisi@....com, mani@...nel.org,
        nicolas.ferre@...rochip.com, richard.genoud@...il.com,
        s.hauer@...gutronix.de, shawnguo@...nel.org, sudeep.holla@....com,
        tklauser@...tanz.ch, vz@...ia.com
Subject: Re: [PATCH v5 2/3] tty: serial: use uart_port_tx() helper

On 22. 11. 22, 9:23, Jiri Slaby wrote:
> On 22. 11. 22, 9:18, Michael Walle wrote:
>> Am 2022-11-22 09:09, schrieb Michael Walle:
>>> Hi,
>>>
>>> Am 2022-11-22 08:02, schrieb Jiri Slaby:
>>>> On 21. 11. 22, 21:27, Michael Walle wrote:
>>>>> This will break serial output for the userspace on my board
>>>>> (arch/arm/boot/dts/lan966x-kontron-kswitch-d10-mmt*dts). The 
>>>>> uart_port_tx()
>>>>> helper will call __port->ops->stop_tx(__port) if 
>>>>> uart_circ_chars_pending()
>>>>> returns 0. But the code above, doesn't do that. In fact, removing the
>>>>> stop_tx() call in the helper macro, will fix the console output.
>>>>>
>>>>> Any ideas how to fix that?
>>>>
>>>> Hm, so ATMEL_US_TXRDY is removed from tx_done_mask in stop_tx, but not
>>>> added back in start_tx. So the tx interrupt is never handled (the tx
>>>> tasklet is not scheduled to send the queue chars) in
>>>> atmel_handle_transmit().
>>>>
>>>> Any chance, the below fixes it?
>>>>
>>>> diff --git a/drivers/tty/serial/atmel_serial.c
>>>> b/drivers/tty/serial/atmel_serial.c
>>>> index 11bf2466390e..395370e0c77b 100644
>>>> --- a/drivers/tty/serial/atmel_serial.c
>>>> +++ b/drivers/tty/serial/atmel_serial.c
>>>> @@ -596,6 +596,8 @@ static void atmel_start_tx(struct uart_port *port)
>>>>                 /* re-enable PDC transmit */
>>>>                 atmel_uart_writel(port, ATMEL_PDC_PTCR, 
>>>> ATMEL_PDC_TXTEN);
>>>>
>>>> +       atmel_port->tx_done_mask |= ATMEL_US_TXRDY;
>>>> +
>>>>         /* Enable interrupts */
>>>>         atmel_uart_writel(port, ATMEL_US_IER, 
>>>> atmel_port->tx_done_mask);
>>>>
>>>>
>>>> thanks,
>>>
>>> Unfortunately, that doesn't help. Btw, some characters are transmitted:
>>>
>>>
>>> [    6.219356] Key type dns_resolver registered
>>> [    6.223679] Registering SWP/SWPB emulation handler
>>> [    6.247530] Loading compiled-in X.509 certificates
>>> [    6.288467] Freeing unused kernel image (initmem) memory: 1024K
>>> [    6.297789] Run /init as init process
>>> WbSOROSOSOSOSOStarting linuxptp system clock synchronization: O
>>>
>>> -michael
>>
>> But you made me look at atmel_stop_tx() and there is this:
>>
>>         /*
>>      * Disable the transmitter.
>>      * This is mandatory when DMA is used, otherwise the DMA buffer
>>      * is fully transmitted.
>>      */
>>      atmel_uart_writel(port, ATMEL_US_CR, ATMEL_US_TXDIS);
>>
>> Removing that write, will also fix the problem. Could it be, that
>> the transmit is still active (via DMA) but the driver will call
>> tx_stop() and then stop the transmission in the background?
> 
> Yes, that was exactly the next step to try. The datasheet doesn't tell 
> much what happens when TXDIS is written while the characters are 
> transmitted.

Side note: your usart doesn't use dma. It's PIO (hence all that 
uart_tx_helper()). And the attached patch doesn't touch TXDIS for 
non-DMA case. I.e. it should transmit the final character (and nothing 
more).

-- 
js