Hi Philipp. Hi Dragan,
Am 10.07.24 um 17:14 schrieb Philipp Puschmann:
Hi Dragan,
Am 10.07.24 um 16:56 schrieb Dragan Simic:
Hello Philipp,
On 2024-07-10 12:20, Philipp Puschmann wrote:
Am 10.07.24 um 12:02 schrieb Diederik de Haas:
On Wednesday, 10 July 2024 11:33:56 CEST Philipp Puschmann wrote:
DMA names are required by of_dma_request_slave_channel function that is
called during uart probe. So to enable DMA for uarts add the names as in
the RK3568 TRM.
Setting it on channels without flow control apparently causes issues. See
https://lore.kernel.org/linux-rockchip/20240628120130.24076-1-didi.debian@xxxxxxxxx/
Ah is see. The only problem that i have is to enable/disable dmas by
having or not having
dma-names properties, where the latter case is followed by kernel
error messages. That
is very counterintuitive. Maybe a explicit boolean like dma-broken
would be better. That
could be set on dtsi level as default and deleted on board dts if
wanted. With such
a boolean we could also prevent the misleading "dma-names property of"
error message
and replace it with a hint that dma is disabled on purpose.
From what I've read in the prior discussions, this seems like a driver
issue, so the driver should be fixed instead.
I would tend to disagree. The serial driver just uses the generic dma API. The error
message comes from of_dma_request_slave_channel() in drivers/dma/of-dma.c
and is called from dma_request_chan() inn drivers/dma/dmaengine.c.
The first function expects a device tree node and "dmas" and "dma-names" properties.
And "dma-names" is misused as "enable" switch and if not present (aka disabled) it
dumps "dma-names property of node X missing or empty". For me it's clear that
a clean way to disable or enable using dma via dts would be better to tell the
of_dma_request_slave_channel function that dma is disabled on purpose, so it
could return ENODEV but without printing a misleading error level message.
Regards,
Philipp
I'm with Philipp here. Topmost because DT should have to do nothing with a
(speculative) driver issue, it represents hardware. I would go further and
say boards using a (non-kernel console) uart, which can't use dma for
whatever reason should disable it. It will never be an issue for the kernel
console (i.e. "stdout-path = "serialX..." in DT or "console=ttySX" in
cmdline) as the kernel will not use dma for this console. It's by the way
even worse for RK3399 Soc DT, which just doesn't expose the dma channels
for the uarts and for RK3368 which does not a expose a single dma channel
of the peripheral dmac (I tent to speculate for the very same reason).
Alex
Signed-off-by: Philipp Puschmann <p.puschmann@xxxxxxxxxxx>
---
arch/arm64/boot/dts/rockchip/rk356x.dtsi | 10 ++++++++++
1 file changed, 10 insertions(+)
diff --git a/arch/arm64/boot/dts/rockchip/rk356x.dtsi
b/arch/arm64/boot/dts/rockchip/rk356x.dtsi index d8543b5557ee..4ae40661ca6a
100644
--- a/arch/arm64/boot/dts/rockchip/rk356x.dtsi
+++ b/arch/arm64/boot/dts/rockchip/rk356x.dtsi
@@ -489,6 +489,7 @@ uart0: serial@fdd50000 {
clocks = <&pmucru SCLK_UART0>, <&pmucru PCLK_UART0>;
clock-names = "baudclk", "apb_pclk";
dmas = <&dmac0 0>, <&dmac0 1>;
+ dma-names = "tx", "rx";
pinctrl-0 = <&uart0_xfer>;
pinctrl-names = "default";
reg-io-width = <4>;
@@ -1389,6 +1390,7 @@ uart1: serial@fe650000 {
clocks = <&cru SCLK_UART1>, <&cru PCLK_UART1>;
clock-names = "baudclk", "apb_pclk";
dmas = <&dmac0 2>, <&dmac0 3>;
+ dma-names = "tx", "rx";
pinctrl-0 = <&uart1m0_xfer>;
pinctrl-names = "default";
reg-io-width = <4>;
...
_______________________________________________
Linux-rockchip mailing list
Linux-rockchip@xxxxxxxxxxxxxxxxxxx
http://lists.infradead.org/mailman/listinfo/linux-rockchip
_______________________________________________
Linux-rockchip mailing list
Linux-rockchip@xxxxxxxxxxxxxxxxxxx
http://lists.infradead.org/mailman/listinfo/linux-rockchip
