Re: [PATCH fixes v4] pci: loongson: Workaround MIPS firmware MRRS settings

[Jiaxun Yang]

在2023年11月1日十一月 下午10:02，Bjorn Helgaas写道：
> On Wed, Nov 01, 2023 at 11:49:56AM +0000, Jiaxun Yang wrote:
>> This is a partial revert of commit 8b3517f88ff2 ("PCI:
>> loongson: Prevent LS7A MRRS increases") for MIPS based Loongson.
>
> Thanks for this patch.  We're in the v6.7 merge window, so we won't
> start merging v6.8 content until v6.7-rc1 (probably Nov 12).
>
>> There are many MIPS based Loongson systems in wild that
>> shipped with firmware which does not set maximum MRRS properly.
>
> As far as I know, there's no requirement for firmware to set MRRS at
> all *except* for the "no_inc_mrrs" hack added by 8b3517f88ff2.  That
> hack treats the current MRRS value as a limit to work around the
> Loongson bug that read requests larger than the limit cause a
> Completer Abort instead of multiple completions.

Yep, it happens that we can't trust MRRS left by firmware on MIPS
Loongson.
>
>> Limiting MRRS to 256 for all as MIPS Loongson comes with higher
>> MRRS support is considered rare.
>> 
>> It must be done at device enablement stage because hardware will
>> reset MRRS to inavlid value if a device got disabled.
>
> s/inavlid/invalid/
>
> This part isn't clear to me, though.  What exactly does "device got
> disabled" mean?  The device got reset?  Power cycled?
> PCI_COMMAND_MASTER was cleared?
>
> PCI_FIXUP_ENABLE quirks are run during pci_enable_device(), which
> basically just turns on PCI_COMMAND_MEMORY and/or PCI_COMMAND_IO.
>
> If MRRS gets reset when PCI_COMMAND_MASTER is set or cleared, we don't
> have a quirk phase that runs during pci_set_master(), which is where
> PCI_COMMAND_MASTER gets set, so it's not clear that
> pci_enable_device() is the right place.

I should make myself clear that MRRS reset actually happens when the
PCIe port (i.e. the PCI to PCI bridge) lose it's PCI_COMMAND_MASTER.

Since pci_enable_bridge is called by pci_enable_device it is still the
best place for such quirk.

I'll amend the commit message to make it clear.

>
>> Cc: stable@xxxxxxxxxxxxxxx
>> Closes: https://bugzilla.kernel.org/show_bug.cgi?id=217680
>> Fixes: 8b3517f88ff2 ("PCI: loongson: Prevent LS7A MRRS increases")
>> Signed-off-by: Jiaxun Yang <jiaxun.yang@xxxxxxxxxxx>
>
> We'll look for an ack from the maintainer.  Maybe that's you, since
> you added the driver in the first place?  Or maybe it's Huacai?

I'm dumb to ACPI enabled Loongson so I think Huacai must be the maintainer
but I'd like to take care of this driver as well, perhaps having two M entry
or give me an R entry?

Huacai, what's your opinion?

>
> MAINTAINERS currently doesn't list anybody for
> drivers/pci/controller/pci-loongson.c, and it should.  That should be
> a separate patch.
>
>> ---
>> v4: Improve commit message
>> 
>> This is a partial revert of the origin quirk so there shouldn't
>> be any drama.
>> ---
>>  drivers/pci/controller/pci-loongson.c | 38 +++++++++++++++++++++++++++
>>  1 file changed, 38 insertions(+)
>> 
>> diff --git a/drivers/pci/controller/pci-loongson.c b/drivers/pci/controller/pci-loongson.c
>> index d45e7b8dc530..d184d7b97e54 100644
>> --- a/drivers/pci/controller/pci-loongson.c
>> +++ b/drivers/pci/controller/pci-loongson.c
>> @@ -108,6 +108,44 @@ DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_LOONGSON,
>>  DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_LOONGSON,
>>  			DEV_LS7A_PCIE_PORT6, loongson_mrrs_quirk);
>>  
>> +#ifdef CONFIG_MIPS
>> +static void loongson_old_mrrs_quirk(struct pci_dev *pdev)
>> +{
>> +	struct pci_bus *bus = pdev->bus;
>> +	struct pci_dev *bridge;
>> +	static const struct pci_device_id bridge_devids[] = {
>> +		{ PCI_VDEVICE(LOONGSON, DEV_LS2K_PCIE_PORT0) },
>> +		{ PCI_VDEVICE(LOONGSON, DEV_LS7A_PCIE_PORT0) },
>> +		{ PCI_VDEVICE(LOONGSON, DEV_LS7A_PCIE_PORT1) },
>> +		{ PCI_VDEVICE(LOONGSON, DEV_LS7A_PCIE_PORT2) },
>> +		{ PCI_VDEVICE(LOONGSON, DEV_LS7A_PCIE_PORT3) },
>> +		{ PCI_VDEVICE(LOONGSON, DEV_LS7A_PCIE_PORT4) },
>> +		{ PCI_VDEVICE(LOONGSON, DEV_LS7A_PCIE_PORT5) },
>> +		{ PCI_VDEVICE(LOONGSON, DEV_LS7A_PCIE_PORT6) },
>
> This looks like the same list of devices as for loongson_mrrs_quirk().
> So I guess the idea is that we need loongson_mrrs_quirk() for
> Loongarch-based systems, and this loongson_old_mrrs_quirk() for
> MIPS-based systems?
>
> If so, maybe they could be #ifdef'd to show that, e.g., so that only
> one or the other is compiled?

I think no_inc_mrrs still needs to be set for those devices so I left
the "new" quirk enabled for MIPS as well here.

But we can also do it in "old" quirk with an extra match, will do in next
version.

[...]

-- 
- Jiaxun