[PATCH 00/16] STM32 configure UART nodes for DMA

[Uwe Kleine-König]

On Fri, Feb 04, 2022 at 04:41:55PM +0100, Erwan LE RAY wrote:
> On 2/4/22 2:22 PM, Alexandre TORGUE wrote:
> > Hi Ahmad
> > 
> > On 2/3/22 18:25, Ahmad Fatoum wrote:
> > > Hello Erwan,
> > > 
> > > On 03.02.22 18:10, Erwan Le Ray wrote:
> > > > Add DMA configuration to UART nodes in stm32mp15x (SOC level) and
> > > > remove it at board level to keep current PIO behavior when needed.
> > > > For stm32-ed1 and stm32-dkx boards, UART4 (console) and UART7
> > > > (no HW flow control pin available) are kept in PIO mode, while USART3
> > > > is now configured in DMA mode.
> > > > UART4 (console UART) has to be kept in irq mode, as DMA support for
> > > > console has been removed from the driver by commit e359b4411c28
> > > > ("serial: stm32: fix threaded interrupt handling").
> > > 
> > > Do I understand correctly that your first patch breaks consoles of
> > > most/all boards, because they will briefly use DMA, which is refused
> > > by the stm32-usart driver and then you add a patch for each board
> > > to fix that breakage?
> > 
> > We have two solutions and both have pro/drawbacks. The first one (Erwan
> > ones, can break the boot if the patch is taken "alone". Your proposition
> > avoids this breakage but deletes a non define property (which is a bit
> > weird). However I prefer to keep a functional behavior, and keep Ahmad
> > proposition. Ahmad, just one question, dt-bindings check doesn't
> > complain about it ?
> > 
> > Cheers
> > Alex
> > 
> > > 
> > > Such intermittent breakage makes bisection a hassle. /delete-property/
> > > is a no-op when the property doesn't exist, so you could move the first
> > > patch to the very end to avoid intermittent breakage.
> > > 
> > > I also think that the driver's behavior is a bit harsh. I think it would
> > > be better for the UART driver to print a warning and fall back to
> > > PIO for console instead of outright refusing and rendering the system
> > > silent. That's not mutually exclusive with your patch series here,
> > > of course.
> > > 
> > > Cheers,
> > > Ahmad
> > > 
> 
> The driver implementation will consider the request to probe the UART
> console in DMA mode as an error (-ENODEV), and will fallback this UART probe
> in irq mode.

> Whatever the patch ordering, the boot will never be broken. The board dt
> patches aim to get a "proper" implementation, but from functional
> perspective the driver will manage a request to probe an UART console in DMA
> mode as an error and fall it back in irq mode.

I didn't debug this further yet, but my machine (with an out-of-tree
dts) fails to boot 6.1-rc4 without removing the dma properties from the
console UART. This is a bug isn't it? The same dts created a working
setup with stm32mp157.dtsi from 5.15 + kernel 5.15.

I can debug this further, but maybe you know off-hand what the problem
is?

Best regards
Uwe

-- 
Pengutronix e.K.                           | Uwe Kleine-König            |
Industrial Linux Solutions                 | https://www.pengutronix.de/ |
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 488 bytes
Desc: not available
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20221108/4d01d8a7/attachment.sig>