On Tuesday 21 April 2020 13:16:56 Marek Behún wrote:
> From: Pali Rohár <pali@xxxxxxxxxx>
>
> Add support for issuing PERST via GPIO specified in 'reset-gpios'
> property (as described in PCI device tree bindings).
>
> Some buggy cards (e.g. Compex WLE900VX or WLE1216) are not detected
> after reboot when PERST is not issued during driver initialization.
>
> Tested on Turris MOX.
>
> Signed-off-by: Pali Rohár <pali@xxxxxxxxxx>
> ---
> drivers/pci/controller/pci-aardvark.c | 32 +++++++++++++++++++++++++++
> 1 file changed, 32 insertions(+)
>
> diff --git a/drivers/pci/controller/pci-aardvark.c b/drivers/pci/controller/pci-aardvark.c
> index 606bae1e7a88..e2d18094d8ca 100644
> --- a/drivers/pci/controller/pci-aardvark.c
> +++ b/drivers/pci/controller/pci-aardvark.c
...
> +static void advk_pcie_issue_perst(struct advk_pcie *pcie)
> +{
> + if (!pcie->reset_gpio)
> + return;
> +
> + dev_info(&pcie->pdev->dev, "issuing PERST via reset GPIO for 1ms\n");
> + gpiod_set_value_cansleep(pcie->reset_gpio, 1);
> + usleep_range(1000, 2000);
> + gpiod_set_value_cansleep(pcie->reset_gpio, 0);
> +}
After more testing we will have to increase this timeout to 10ms as some
Compex cards are sometimes not detected with current 1ms timeout. I will
do it in V3 patch series.
Bjorn, do you know if there is a defined timeout in PCIE specification
how long should be card in PERST?
I looked into others pci kernel drivers and basically every driver is
using its own timeout.
pcie-kirin.c --> usleep_range(20000, 25000);
pcie-qcom.c --> usleep_range(1000, 1000 + 500); msleep(100);
pci-mvebu.c --> udelay(100);
pci-tegra.c --> usleep_range(1000, 2000);
pcie-iproc.c --> udelay(250);
pcie-mediatek.c --> no delay
pci-imx6.c --> msleep(100);
But I guess that this timeout should not depend on driver or pci
controller, but rather on connected card or on some recommended value if
defined by PCIE specification.
