Re: [PATCH V7 00/11] Support for generic ACPI based PCI host controller

[Jon Masters <jcm@redhat.com>]

Bjorn,

Out walking so sorry about top posting. Quick reply though:

1. I checked with the Windows team. They usually avoid quirks entirely but when it has happened, it has been done via the MCFG/FADT not DSDT.

2. They would be ok if we were to key off the OEM  name and revision for the IP in the MCFG table.

3. I have already verified existing shipping ARMv8 systems provide enough unique data in that entry, and have asked that vendors guarantee to rev it in future IP (which I will verify on models pre tapeout and certainly in early firmware builds). One vendor has a platform that isn't public yet that uses a non-public name in the MCFG but I spoke with them on Friday and they will shortly update their firmware so that a quirk could be posted.

4. I have requested (and Linaro are investigating) that Linaro (with ARM's assistance) begin to drive a separate thread around upstreaming (independent of this core effort) quirks that use the OEM fields in the MCFG as a more scalable approach than one per platform via DMI.

5. I will drive a clarification to the SBBR that does not encourage or endorse quirks but does merely reinforce that data must be unique in such tables. I am driving a separate series of conversations with vendors to ensure that this is the case on all future platforms - though just generally, there is no more high end top shelf "Xeon class" silicon needing common quirks in the pipeline.

More later.

Jon.

-- 
Computer Architect | Sent from my 64-bit #ARM Powered phone

> On May 23, 2016, at 19:39, Bjorn Helgaas <helgaas@kernel.org> wrote:
> 
>> On Mon, May 23, 2016 at 03:16:01PM +0000, Gabriele Paoloni wrote:
>> Hi Lorenzo
>> 
>>> -----Original Message-----
>>> From: Lorenzo Pieralisi [mailto:lorenzo.pieralisi@arm.com]
>>> Sent: 23 May 2016 11:57
>>> To: Ard Biesheuvel
>>> Cc: Gabriele Paoloni; Jon Masters; Tomasz Nowicki; helgaas@kernel.org;
>>> arnd@arndb.de; will.deacon@arm.com; catalin.marinas@arm.com;
>>> rafael@kernel.org; hanjun.guo@linaro.org; okaya@codeaurora.org;
>>> jchandra@broadcom.com; linaro-acpi@lists.linaro.org; linux-
>>> pci@vger.kernel.org; dhdang@apm.com; Liviu.Dudau@arm.com;
>>> ddaney@caviumnetworks.com; jeremy.linton@arm.com; linux-
>>> kernel@vger.kernel.org; linux-acpi@vger.kernel.org;
>>> robert.richter@caviumnetworks.com; Suravee.Suthikulpanit@amd.com;
>>> msalter@redhat.com; Wangyijing; mw@semihalf.com;
>>> andrea.gallo@linaro.org; linux-arm-kernel@lists.infradead.org
>>> Subject: Re: [PATCH V7 00/11] Support for generic ACPI based PCI host
>>> controller
>>> 
>>>> On Fri, May 20, 2016 at 11:14:03AM +0200, Ard Biesheuvel wrote:
>>>> On 20 May 2016 at 10:40, Gabriele Paoloni
>>> <gabriele.paoloni@huawei.com> wrote:
>>>>> Hi Ard
>>>>> 
>>>>>> -----Original Message-----
>>>>>> From: Ard Biesheuvel [mailto:ard.biesheuvel@linaro.org]
>>>> [...]
>>>>>> 
>>>>>> Is the PCIe root complex so special that you cannot simply
>>> describe an
>>>>>> implementation that is not PNP0408 compatible as something else,
>>> under
>>>>>> its own unique HID? If everybody is onboard with using ACPI, how
>>> is
>>>>>> this any different from describing other parts of the platform
>>>>>> topology? Even if the SBSA mandates generic PCI, they already
>>> deviated
>>>>>> from that when they built the hardware, so pretending that it is a
>>>>>> PNP0408 with quirks really does not buy us anything.
>>>>> 
>>>>> From my understanding we want to avoid this as this would allow
>>> each
>>>>> vendor to come up with his own code and it would be much more
>>> effort
>>>>> for the PCI maintainer to rework the PCI framework to accommodate
>>> X86
>>>>> and "all" ARM64 Host Controllers...
>>>>> 
>>>>> I guess this approach is too risky and we want to avoid this.
>>> Through
>>>>> standardization we can more easily maintain the code and scale it
>>> to
>>>>> multiple SoCs...
>>>>> 
>>>>> So this is my understanding; maybe Jon, Tomasz or Lorenzo can give
>>>>> a bit more explanation...
>>>> 
>>>> OK, so that boils down to recommending to vendors to represent known
>>>> non-compliant hardware as compliant, just so that we don't have to
>>>> change the code to support additional flavors of ECAM ? It's fine to
>>>> be pragmatic, but that sucks.
>>>> 
>>>> We keep confusing the x86 case with the ARM case here: for x86, they
>>>> needed to deal with broken hardware *after* the fact, and all they
>>>> could do is find /some/ distinguishing feature in order to guess
>>> which
>>>> exact hardware they might be running on. For arm64, it is the
>>> opposite
>>>> case. We are currently in a position where we can demand vendors to
>>>> comply with the standards they endorsed themselves, and (ab)using
>>> ACPI
>>>> + DMI as a de facto platform description rather than plain ACPI makes
>>>> me think the DT crowd were actually right from the beginning. It
>>>> *directly* violates the standardization principle, since it requires
>>> a
>>>> priori knowledge inside the OS that a certain 'generic' device must
>>> be
>>>> driven in a special way.
>>>> 
>>>> So can anyone comment on the feasibility of adding support for
>>> devices
>>>> with vendor specific HIDs (and no generic CIDs) to the current ACPI
>>>> ECAM driver in Linux?
> 
> I don't think of ECAM support itself as a "driver".  It's just a
> service available to drivers, similar to OF resource parsing.
> 
> Per PCI Firmware r3.2, sec 4.1.5, "PNP0A03" means a PCI/PCI-X/PCIe
> host bridge.  "PNP0A08" means a PCI-X Mode 2 or PCIe bridge that
> supports extended config space.  It doesn't specify how we access that
> config space, so I think hardware with non-standard ECAM should still
> have PNP0A03 and PNP0A08 in _CID or _HID.
> 
> "ECAM" as used in the specs (PCIe r3.0, sec 7.2.2, and PCI Firmware
> r3.2, sec 4.1) means:
> 
>  (a) a memory-mapped model for config space access, and
>  (b) a specific mapping of address bits to bus/device/function/
>      register
> 
> MCFG and _CBA assume both (a) and (b), so I think a device with
> non-standard ECAM mappings should not be described in MCFG or _CBA.
> 
> If a bridge has ECAM with non-standard mappings, I think either a
> vendor-specific _HID or a device-specific method, e.g., _DSM, could
> communicate that.
> 
> Jon, I agree that we should avoid describing non-standardized hardware
> in Linux-specific ways.  Is there a mechanism in use already?  How
> does Windows handle this?  DMI is a poor long-term solution because it
> requires ongoing maintenance for new platforms, but I think it's OK
> for getting started with platforms already shipping.
> 
> A _DSM has the advantage that once it is defined and supported, OEMs
> can ship new platforms without requiring a new quirk or a new _HID to
> be added to a driver.
> 
> There would still be the problem of config access before the namespace
> is available, i.e., the MCFG use case.  I don't know how important
> that is.  Defining an MCFG extension seems like the most obvious
> solution.
> 
> If we only expect a few non-standard devices, maybe it's enough to
> have DMI quirks to statically set up ECAM and just live with the
> inconvenience of requiring a kernel change for every new non-standard
> device.
> 
>>> Host bridges in ACPI are handled through PNP0A08/PNP0A03 ids, and
>>> most of the arch specific code is handled in the respective arch
>>> directories (X86 and IA64, even though IA64 does not rely on ECAM/MCFG
>>> for
>>> PCI ops), it is not a driver per-se, PNP0A08/PNP0A03 are detected
>>> through
>>> ACPI scan handlers and the respective arch code (ie pci_acpi_scan_root)
>>> sets-up resources AND config space on an arch specific basis.
>>> 
>>> X86 deals with that with code in arch/x86 that sets-up the pci_raw_ops
>>> on a platform specific basis (and it is not nice, but it works because
>>> as you all know the number of platforms in X86 world is contained).
>>> 
>>> Will this happen for ARM64 in arch/arm64 based on vendor specific
>>> HIDs ?
>>> 
>>> No.
>>> 
>>> So given the current state of play (we were requested to move the
>>> arch/arm64 specific ACPI PCI bits to arch/arm64), we would end up
>>> with arch/arm64 code requiring code in /drivers to set-up pci_ops
>>> in a platform specific way, it is horrible, if feasible at all.
>>> 
>>> The only way this can be implemented is by pretending that the
>>> ACPI/PCI arch/arm64 implementation is generic code (that's what this
>>> series does), move it to /drivers (where it is in this series), and
>>> implement _DSD vendor specific bindings (per HID) to set-up the pci
>>> operations; whether this solution should go upstream, given that it
>>> is just a short-term solution for early platforms bugs, it is another
>>> story and my personal answer is no.
> 
> It seems like there should be a way to look for a _DSM before we call
> acpi_pci_root_get_mcfg_addr() to look for _CBA.
> 
> Currently we call acpi_pci_root_get_mcfg_addr() (to read _CBA) from
> the generic acpi_pci_root_add(), but the result (root->mcfg_addr) is
> only used in x86-specific code.  I think it would be nicer if the
> lookup and the use were together.  Then it would be easier to override
> it because the mapping assumptions would all be in one place.
> 
>> I think it shouldn't be too bad to move quirk handling mechanism to
>> arch/arm64. Effectively we would not move platform specific code into
>> arch/arm64 but just the mechanism checking if there is any quirk that
>> is defined.
>> 
>> i.e.:
>> 
>> extern struct pci_cfg_fixup __start_acpi_mcfg_fixups[];
>> extern struct pci_cfg_fixup __end_acpi_mcfg_fixups[];
>> 
>> static struct pci_ecam_ops *pci_acpi_get_ops(struct acpi_pci_root *root)
>> {
>>        int bus_num = root->secondary.start;
>>        int domain = root->segment;
>>        struct pci_cfg_fixup *f;
>> 
>>        /*
>>         * Match against platform specific quirks and return corresponding
>>         * CAM ops.
>>         *
>>         * First match against PCI topology <domain:bus> then use DMI or
>>         * custom match handler.
>>         */
>>        for (f = __start_acpi_mcfg_fixups; f < __end_acpi_mcfg_fixups; f++) {
>>                if ((f->domain == domain || f->domain == PCI_MCFG_DOMAIN_ANY) &&
>>                    (f->bus_num == bus_num || f->bus_num == PCI_MCFG_BUS_ANY) &&
>>                    (f->system ? dmi_check_system(f->system) : 1) &&
>>                    (f->match ? f->match(f, root) : 1))
>>                        return f->ops;
>>        }
>>        /* No quirks, use ECAM */
>>        return &pci_generic_ecam_ops;
>> }
>> 
>> Such quirks will be defined anyway in drivers/pci/host/ in the vendor
>> specific quirk implementations.
>> 
>> e.g. in HiSilicon case we would have
>> 
>> DECLARE_ACPI_MCFG_FIXUP(NULL, hisi_pcie_match, &hisi_pcie_ecam_ops,
>>            PCI_MCFG_DOMAIN_ANY, PCI_MCFG_BUS_ANY);
>> 
>> in "drivers/pci/host/pcie-hisi-acpi.c "
>> 
>> Thanks
>> 
>> Gab
> 
> Sorry Gab, I guess I was really responding to earlier messages :)
> 
> I don't really have much to say here, except that it doesn't seem
> right to have an MCFG that describes a non-standard ECAM mapping.
> I suppose there's already firmware in the field that does that,
> though?
> 
> Bjorn