Re: [RFC v1 2/2] PCI: tegra: Use readl_poll_timeout() for link status polling

[Mikko Perttunen <mperttunen@nvidia.com>]

On Wednesday, September 17, 2025 4:45 PM Anand Moon wrote:
> Hi Mikko,
> 
> Thanks for your review comments.
> 
> On Wed, 17 Sept 2025 at 08:51, Mikko Perttunen <mperttunen@nvidia.com> wrote:
> >
> > On Monday, September 1, 2025 4:00 AM Anand Moon wrote:
> > > Replace the manual `do-while` polling loops with the readl_poll_timeout()
> > > helper when checking the link DL_UP and DL_LINK_ACTIVE status bits
> > > during link bring-up. This simplifies the code by removing the open-coded
> > > timeout logic in favor of the standard, more robust iopoll framework.
> > > The change improves readability and reduces code duplication.
> > >
> > > Cc: Thierry Reding <thierry.reding@gmail.com>
> > > Signed-off-by: Anand Moon <linux.amoon@gmail.com>
> > > ---
> > >  drivers/pci/controller/pci-tegra.c | 38 ++++++++++++------------------
> > >  1 file changed, 15 insertions(+), 23 deletions(-)
> > >
> > > diff --git a/drivers/pci/controller/pci-tegra.c b/drivers/pci/controller/pci-tegra.c
> > > index 3841489198b64..8e850f7c84e40 100644
> > > --- a/drivers/pci/controller/pci-tegra.c
> > > +++ b/drivers/pci/controller/pci-tegra.c
> > > @@ -24,6 +24,7 @@
> > >  #include <linux/irqchip/chained_irq.h>
> > >  #include <linux/irqchip/irq-msi-lib.h>
> > >  #include <linux/irqdomain.h>
> > > +#include <linux/iopoll.h>
> >
> > There is already an iopoll.h include in this file, so this adds a duplicate.
> >
> Opps, I missed this in rebasing my code.
> 
> > >  #include <linux/kernel.h>
> > >  #include <linux/init.h>
> > >  #include <linux/module.h>
> > > @@ -2157,37 +2158,28 @@ static bool tegra_pcie_port_check_link(struct tegra_pcie_port *port)
> > >       value |= RP_PRIV_MISC_PRSNT_MAP_EP_PRSNT;
> > >       writel(value, port->base + RP_PRIV_MISC);
> > >
> > > -     do {
> > > -             unsigned int timeout = TEGRA_PCIE_LINKUP_TIMEOUT;
> > > -
> > > -             do {
> > > -                     value = readl(port->base + RP_VEND_XP);
> > > -
> > > -                     if (value & RP_VEND_XP_DL_UP)
> > > -                             break;
> > > -
> > > -                     usleep_range(1000, 2000);
> > > -             } while (--timeout);
> > > +     while (retries--) {
> > > +             int err;
> > >
> > > -             if (!timeout) {
> > > +             err = readl_poll_timeout(port->base + RP_VEND_XP, value,
> > > +                                      value & RP_VEND_XP_DL_UP,
> > > +                                      1000,
> > > +                                      TEGRA_PCIE_LINKUP_TIMEOUT * 1000);
> >
> > The logic change here looks OK to me. This makes the timeout 200ms (TEGRA_PCIE_LINKUP_TIMEOUT is 200). Previously, the code looped 200 times with a 1 to 2ms sleep on each iteration. So the timeout could have been longer than 200ms previously, but not in a way that could be relied on.
> 
> You're right; the original usleep_range(1000, 2000) had a variable sleep time.
> To replicate the worst-case behavior of the old loop, the
> readl_poll_timeout should
> use a delay_us of 1000 and a timeout_us that matches the original
> maximum duration.
> Since the previous code looped 200 times with a maximum 2ms sleep,
> the correct timeout is 400ms, so update (TEGRA_PCIE_LINKUP_TIMEOUT * 2000).
> or increase TEGRA_PCIE_LINKUP_TIMEOUT to 400.
> 
> Are these changes ok with you?

I think the code is fine as is. Before, the shortest the timeout could be was 200ms, i.e. there should be no situation where we need a timeout longer than that, or otherwise that would fail randomly depending on the sleep duration. So I think the 200ms is correct here and the only change necessary is the removal of the second iopoll.h

Cheers,
Mikko

> 
> Thank
> -Anand