On Thursday 24 September 2020 10:11:06 Bjorn Helgaas wrote: > On Thu, Sep 24, 2020 at 10:46:18AM +0200, Pali Rohár wrote: > > It is not HW bug or workaround for some cards but it is requirement by PCI > > Express spec. After fundamental reset is needed 100ms delay prior enabling > > link training. So update comment in code to reflect this requirement. > > > > Signed-off-by: Pali Rohár <[email protected]> > > --- > > drivers/pci/controller/pci-aardvark.c | 7 ++++++- > > 1 file changed, 6 insertions(+), 1 deletion(-) > > > > diff --git a/drivers/pci/controller/pci-aardvark.c > > b/drivers/pci/controller/pci-aardvark.c > > index 50ab6d7519ae..19b9b79226e5 100644 > > --- a/drivers/pci/controller/pci-aardvark.c > > +++ b/drivers/pci/controller/pci-aardvark.c > > @@ -259,7 +259,12 @@ static void advk_pcie_issue_perst(struct advk_pcie > > *pcie) > > if (!pcie->reset_gpio) > > return; > > > > - /* PERST does not work for some cards when link training is enabled */ > > + /* > > + * As required by PCI Express spec a delay for at least 100ms after > > + * de-asserting PERST# signal is needed before link training is enabled. > > + * So ensure that link training is disabled prior de-asserting PERST# > > + * signal to fulfill that PCI Express spec requirement. > > Can you please include the spec citation here? In the PCIe base spec, > PERST# is only mentioned in PCIe r5.0, sec 6.6.1, and I don't see the > connection there to 100ms between de-assert of PERST# and enabling > link training.
Hello! I copied this "comment" from other place in pci-aardvark.c where that timeout 100ms is already applied. Timeout with explanation comment was introduced in following commit: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=f4c7d053d7f7 Here are links to discussions about that patch: https://lore.kernel.org/linux-pci/[email protected]/T/#u https://lore.kernel.org/linux-pci/[email protected]/T/#u > Sec 6.1.1 does talk about 100ms before sending config requests (for > ports that support <= 5 GT/s), and 100ms after link training completes > (for ports that support > 5 GT/s). > > Maybe there's more language in a form-factor spec or something? > > > + */ > > reg = advk_readl(pcie, PCIE_CORE_CTRL0_REG); > > reg &= ~LINK_TRAINING_EN; > > advk_writel(pcie, reg, PCIE_CORE_CTRL0_REG); > > -- > > 2.20.1 > >
