[v2,2/2] PCI: histb: Correct PCIe reset operation

The PCIe reset via GPIO in the driver never worked as expected.  Per
"Power Sequencing and Reset Signal Timings" table in
PCI EXPRESS CARD ELECTROMECHANICAL SPECIFICATION, the PERST# should be
deasserted after minimum of 100us once REFCLK is stable.

The assertion has been done when the GPIO is being requested, and
deassertion should be done in host enabling rather than disabling. Also
a bit wait is added to ensure device get ready after reset.

Signed-off-by: Shawn Guo <shawn.guo@linaro.org>
---
 drivers/pci/controller/dwc/pcie-histb.c | 20 ++++++++++++++------
 1 file changed, 14 insertions(+), 6 deletions(-)

On Thu, Jan 09, 2020 at 11:28:51AM +0800, Shawn Guo wrote:
> The PCIe reset via GPIO in the driver never worked as expected.  Per
> "Power Sequencing and Reset Signal Timings" table in
> PCI EXPRESS CARD ELECTROMECHANICAL SPECIFICATION, the PERST# should be
> deasserted after minimum of 100us once REFCLK is stable.
> 
> The assertion has been done when the GPIO is being requested, and
> deassertion should be done in host enabling rather than disabling. Also
> a bit wait is added to ensure device get ready after reset.
> 
> Signed-off-by: Shawn Guo <shawn.guo@linaro.org>
> ---
>  drivers/pci/controller/dwc/pcie-histb.c | 20 ++++++++++++++------
>  1 file changed, 14 insertions(+), 6 deletions(-)

Shawn,

this looks like a fix, please tag it as such and let me know if
it has to be backported, in which case also the previous patch
should I assume.

Thanks,
Lorenzo

> diff --git a/drivers/pci/controller/dwc/pcie-histb.c b/drivers/pci/controller/dwc/pcie-histb.c
> index 112254619ed0..67c27a8036c7 100644
> --- a/drivers/pci/controller/dwc/pcie-histb.c
> +++ b/drivers/pci/controller/dwc/pcie-histb.c
> @@ -219,9 +219,6 @@ static void histb_pcie_host_disable(struct histb_pcie *hipcie)
>  	clk_disable_unprepare(hipcie->sys_clk);
>  	clk_disable_unprepare(hipcie->bus_clk);
>  
> -	if (hipcie->reset_gpio)
> -		gpiod_set_value_cansleep(hipcie->reset_gpio, 0);
> -
>  	if (hipcie->vpcie)
>  		regulator_disable(hipcie->vpcie);
>  }
> @@ -242,9 +239,6 @@ static int histb_pcie_host_enable(struct pcie_port *pp)
>  		}
>  	}
>  
> -	if (hipcie->reset_gpio)
> -		gpiod_set_value_cansleep(hipcie->reset_gpio, 1);
> -
>  	ret = clk_prepare_enable(hipcie->bus_clk);
>  	if (ret) {
>  		dev_err(dev, "cannot prepare/enable bus clk\n");
> @@ -278,6 +272,20 @@ static int histb_pcie_host_enable(struct pcie_port *pp)
>  	reset_control_assert(hipcie->bus_reset);
>  	reset_control_deassert(hipcie->bus_reset);
>  
> +	if (hipcie->reset_gpio) {
> +		/*
> +		 * "Power Sequencing and Reset Signal Timings" table in
> +		 * PCI EXPRESS CARD ELECTROMECHANICAL SPECIFICATION, indicates
> +		 * PERST# should be deasserted after minimum of 100us
> +		 * once REFCLK is stable.
> +		 */
> +		usleep_range(100, 200);
> +		gpiod_set_value_cansleep(hipcie->reset_gpio, 0);
> +
> +		/* wait 1ms for device to be ready */
> +		usleep_range(1000, 2000);
> +	}
> +
>  	return 0;
>  
>  err_aux_clk:
> -- 
> 2.17.1
>

On Wed, Feb 26, 2020 at 7:31 PM Lorenzo Pieralisi
<lorenzo.pieralisi@arm.com> wrote:
>
> On Thu, Jan 09, 2020 at 11:28:51AM +0800, Shawn Guo wrote:
> > The PCIe reset via GPIO in the driver never worked as expected.  Per
> > "Power Sequencing and Reset Signal Timings" table in
> > PCI EXPRESS CARD ELECTROMECHANICAL SPECIFICATION, the PERST# should be
> > deasserted after minimum of 100us once REFCLK is stable.
> >
> > The assertion has been done when the GPIO is being requested, and
> > deassertion should be done in host enabling rather than disabling. Also
> > a bit wait is added to ensure device get ready after reset.
> >
> > Signed-off-by: Shawn Guo <shawn.guo@linaro.org>
> > ---
> >  drivers/pci/controller/dwc/pcie-histb.c | 20 ++++++++++++++------
> >  1 file changed, 14 insertions(+), 6 deletions(-)
>
> Shawn,
>
> this looks like a fix, please tag it as such and let me know if
> it has to be backported, in which case also the previous patch
> should I assume.

Hi Lorenzo,

It is a fix, but we recently realized that the problem needs to be
fixed in a way that does not break existing DTB.  So please ignore the
series for now, and we will try to work out a better one.  Sorry that
we did not update the thread in time.

Shawn