Re: [PATCH] PCI: dw-rockchip: Fix GPIO initialization flag

[Niklas Cassel <cassel@kernel.org>]

On Tue, Apr 16, 2024 at 04:58:07PM +0530, Manivannan Sadhasivam wrote:
> On Wed, Mar 27, 2024 at 04:25:31PM +0100, Niklas Cassel wrote:
> > PERST is active low according to the PCIe specification.
> > 
> > However, the existing pcie-dw-rockchip.c driver does:
> > gpiod_set_value(..., 0); msleep(100); gpiod_set_value(..., 1);
> > When asserting + deasserting PERST.
> > 
> > This is of course wrong, but because all the device trees for this
> > compatible string have also incorrectly marked this GPIO as ACTIVE_HIGH:
> > $ git grep -B 10 reset-gpios arch/arm64/boot/dts/rockchip/rk3568*
> > $ git grep -B 10 reset-gpios arch/arm64/boot/dts/rockchip/rk3588*
> > 
> > The actual toggling of PERST is correct.
> > (And we cannot change it anyway, since that would break device tree
> > compatibility.)
> > 
> > However, this driver does request the GPIO to be initialized as
> > GPIOD_OUT_HIGH, which does cause a silly sequence where PERST gets
> > toggled back and forth for no good reason.
> > 
> > Fix this by requesting the GPIO to be initialized as GPIOD_OUT_LOW
> > (which for this driver means PERST asserted).
> > 
> > This will avoid an unnecessary signal change where PERST gets deasserted
> > (by devm_gpiod_get_optional()) and then gets asserted
> > (by rockchip_pcie_start_link()) just a few instructions later.
> > 
> > Before patch, debug prints on EP side, when booting RC:
> > [  845.606810] pci: PERST asserted by host!
> > [  852.483985] pci: PERST de-asserted by host!
> > [  852.503041] pci: PERST asserted by host!
> > [  852.610318] pci: PERST de-asserted by host!
> > 
> > After patch, debug prints on EP side, when booting RC:
> > [  125.107921] pci: PERST asserted by host!
> > [  132.111429] pci: PERST de-asserted by host!
> > 
> > Without this change, there is no guarantee that PERST will be asserted
> > while the core is performing a fundamental reset.

> There is no 'core' here, are you referring to the device?

If you at pcie-qcom.c, it does:
1) PERST# assert
2) core reset (using reset_control_assert() in ops->init())
3) PERST# deassert

So the EP is held in reset while the RC resets.

Right now, for dw-rockchip driver, there is no guarantee that
EP is held in reset when the core on RC side resets, which seems bad.


> 
> > (E.g. if the bootloader would leave PERST deasserted.)
> > 
> 
> I don't follow this last sentence. But even without that, the commit message
> itself is descriptive enough.

When I flash u-boot on my board, and boot using TFTP,
I can see that when u-boot jumps to Linux, PERST# is not asserted
and will not be asserted when PCIe RC driver loads.

So the scenario mentioned above can happen.

(If I don't boot using TFTP, PERST# is deasserted when PCIe RC driver loads.)

Anyway, I will remove or clarify this sentence.


> 
> > Signed-off-by: Niklas Cassel <cassel@kernel.org>
> 
> This is a legitimate bug fix. So you should add the fixes tag and CC stable to
> get it backported.

Ok, will do in V2.


> 
> But the patch itself looks fine to me.
> 
> Reviewed-by: Manivannan Sadhasivam <manivannan.sadhasivam@linaro.org>

Thank you.


Kind regards,
Niklas