Re: 回复: 回复: [PATCH v13 0/21] Refactoring Microchip PCIe driver and add StarFive PCIe

[Emil Renner Berthing]

Kevin Xie wrote:
> > Kevin Xie <kevin.xie at starfivetech.com> writes:
> >
> > >> Minda Chen <minda.chen at starfivetech.com> writes:
> > >>
> > >> > This patchset final purpose is add PCIe driver for StarFive JH7110 SoC.
> > >> > JH7110 using PLDA XpressRICH PCIe IP. Microchip PolarFire Using the
> > >> > same IP and have commit their codes, which are mixed with PLDA
> > >> > controller codes and Microchip platform codes.
> > >>
> > >> Thank you for this series.
> > >>
> > >> I tested this on a VisionFive v2 board, and it seems to probe and
> > >> find my
> > >> M.2 NVMe SSD, but then gets timeouts when trying to use the NVMe (e.g.
> > >> 'blkid' command)
> > >>
> > >
> > > Hi, Kevin:
> > > Could you please provide the manufacturer and model of the M.2 NVMe
> > > SSD you tested?
> >
> > I have a 256 Gb Silicon Power P34A60 M.2 NVMe SSD (part number:
> > sp256gbp34a60m28)
> >
> Thanks, Kevin, we will buy one to test.
>
> Before doing this refactoring, we encountered the same bug with Kingston M.2 SSD,
> and we workaround the problem with the below patch, please have a try:
> diff --git a/drivers/nvme/host/pci.c b/drivers/nvme/host/pci.c
> index 507bc149046d..5be37f1ee150 100644
> --- a/drivers/nvme/host/pci.c
> +++ b/drivers/nvme/host/pci.c
> @@ -1059,6 +1059,16 @@ static inline int nvme_poll_cq(struct nvme_queue *nvmeq,
>  {
>         int found = 0;
>
> +       /*
> +        * In some cases, such as JH7110 SoC working with Kingston SSD,
> +        * the CQE status may update a little bit later than the MSI,
> +        * which cause an IRQ handle missing.
> +        * As a workaround, here we will check the status first, and wait
> +        * 1us if we get nothing.
> +        */
> +       if (!nvme_cqe_pending(nvmeq))
> +               udelay(1);
> +
>         while (nvme_cqe_pending(nvmeq)) {
>                 found++;
>                 /*
>

Hi Kevin,

Thanks, this fixes the same problem on my WD Blue SN570 250GB.

Before this patch I found that CONFIG_INIT_ON_ALLOC_DEFAULT_ON=y fixed the
problem, but it's now clear that this only slowed my kernel enough that this
wouldn't trigger as often.

/Emil

> > Also for reference, I tested the same SSD on another arm platform (Khadas
> > VIM3) and it works fine.
> >
> > Kevin
>
> Hi, Bjorn:
> Do you have any idea about the late CQE phase update condition as mentioned
> in the patch comments above?
> This is an issue that occurs with a small probability on individual devices in our
> platform.
> Thus, I suggest the refactoring patch set should go forward.
> Later we will try to find a more formal solution instead, and send a new patch.
> _______________________________________________
> linux-riscv mailing list
> linux-riscv at lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-riscv