Re: Linux Kernel Regression: HAPS quirk breaks PCIe on i.MX6QP

[Lukas Hartmann]

Hi Thinh,

this is in a taped out i.MX6QP 1.2GHz SoC, I also confirmed it with the 1.0GHz version, it has the same 0xabcd device id integrated.

I will try your patch ASAP.

Cheers
Lukas

> On 1. Feb 2019, at 00:46, Thinh Nguyen <thinh.nguyen@xxxxxxxxxxxx> wrote:
> 
> Hi Lukas,
> 
> Thinh Nguyen wrote:
>> Bjorn Helgaas wrote:
>>> [+cc linux-pci, linux-kernel]
>>> 
>>>> On Thu, Jan 31, 2019 at 11:21 AM Lukas F. Hartmann <lukas@xxxxxxxxx> wrote:
>>>> Hi Thinh,
>>>> 
>>>> I'm writing you because you're the author in this commit:
>>>> https://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_torvalds_linux_commit_03e6742584af8866ba014874669d0440bed3a623&d=DwIBaQ&c=DPL6_X_6JkXFx7AXWqB0tg&r=VKipRJmm95P2RbIxyKKYrcUCOGNlQtjlV-5zhrVhIik&m=2sOrowYXlsC3rl0LfHQhIZzImag0jFZXGvR6NIQDsh8&s=oJoBWRE8_LGAYbX2alh6QnjYZTTmzcgLw4MtOMToNyo&e=
>>>> 
>>>> This quirk workaround breaks the PCIe on i.MX6QP, at least on my test
>>>> devices. The reason is because the Synposys PCIe IP in i.MX6QP has the
>>>> same device ID as the HAPS USB3 controller you are targeting in the
>>>> commit: 0xabcd.
>>>> 
>>>> Definition:
>>>> 
>>>> https://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_torvalds_linux_blob_master_include_linux_pci-5Fids.h-23L2364&d=DwIBaQ&c=DPL6_X_6JkXFx7AXWqB0tg&r=VKipRJmm95P2RbIxyKKYrcUCOGNlQtjlV-5zhrVhIik&m=2sOrowYXlsC3rl0LfHQhIZzImag0jFZXGvR6NIQDsh8&s=-FVCwe81XNjJsYHYk1w-kdAzSOLunTGeNc73azO2QYw&e=
>>>> 
>>>> Here is a bit of lspci output on my i.MX6QP machine (without the problem):
>>>> 
>>>> mntmn@reform:~/code/linux$ lspci -v
>>>> 00:00.0 PCI bridge: Synopsys, Inc. Device abcd (rev 01) (prog-if 00
>>>>                                          ^^^^
>>>> [Normal decode])
>>>>        Flags: bus master, fast devsel, latency 0, IRQ 307
>>>>        Memory at 01000000 (32-bit, non-prefetchable) [size=1M]
>>>>        Bus: primary=00, secondary=01, subordinate=ff, sec-latency=0
>>>> ...
>>>> 
>>>> The failure mode is that the PCIe controller shows up as a USB
>>>> controller and my ath9k wireless PCIe card cannot be assigned the
>>>> proper resources anymore (-ENOMEM even for very small BARs).
>>>> 
>>>> Reverting this commit fixes the problem for me. I suggest to make the
>>>> check more specific to the platform/chipset you are targeting.
>>> So Synopsys apparently re-used Device ID 0xabcd?  That's a pretty bad problem.
>>> 
>>> It looks like we merged 03e6742584af ("PCI: Override Synopsys USB 3.x
>>> HAPS device class") for v5.0, and v5.0-final hasn't been released yet,
>>> so if we don't hear from Thinh with a resolution, we can still revert
>>> it.
>>> 
>>> I set myself a reminder to revert this on Feb 11, but hopefully we'll
>>> have a better resolution before then.
>>> 
>>> Bjorn
>>> 
>> This is really odd that the PID for the PCIe controller in i.MX6QP is
>> the same as PID Synopsys use for USB controller. We use a different set
>> of PIDs for PCIe controllers and track VID and PID in our releases.
>> 
>> Is the Root Complex (00:00.0) part of the SoC? Or is this a Synopsys
>> prototype connected to your board? If it is the latter, then the FPGA
>> configuration needs to be updated to other PID.
>> 
>> We're investigating the workaround in case this is on a taped-out SoC.
> 
> Can you try to see if this will help?
> 
> diff --git a/drivers/pci/quirks.c b/drivers/pci/quirks.c
> index b0a413f3f7ca..64623fbdd1e5 100644
> --- a/drivers/pci/quirks.c
> +++ b/drivers/pci/quirks.c
> @@ -629,6 +629,9 @@ static void quirk_synopsys_haps(struct pci_dev *pdev)
> {
>        u32 class = pdev->class;
> 
> +       if (class != PCI_CLASS_SERIAL_USB_XHCI)
> +               return;
> +
>        switch (pdev->device) {
>        case PCI_DEVICE_ID_SYNOPSYS_HAPSUSB3:
>        case PCI_DEVICE_ID_SYNOPSYS_HAPSUSB3_AXI:
> 
> Thanks,
> Thinh
> 
>