Message-ID: <5AF2E91E.3030804@kontron.com>
Date:   Wed, 9 May 2018 12:27:13 +0000
From:   Gilles Buloz <Gilles.Buloz@...tron.com>
To:     Bjorn Helgaas <helgaas@...nel.org>
CC:     Bjorn Helgaas <bhelgaas@...gle.com>,
        linux-pci <linux-pci@...r.kernel.org>,
        "Minghuan.Lian@....com" <Minghuan.Lian@....com>,
        "linux-arm-kernel@...ts.infradead.org" 
        <linux-arm-kernel@...ts.infradead.org>,
        Ard Biesheuvel <ard.biesheuvel@...aro.org>,
        "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
        Frederick Lawler <fred@...dlawl.com>,
        "Sinan Kaya" <okaya@...eaurora.org>
Subject: Re: [PATCH] PCI: Check whether bridges allow access to extended
 config space

Le 07/05/2018 23:56, Bjorn Helgaas a écrit :
> On Fri, May 04, 2018 at 03:06:00PM -0500, Bjorn Helgaas wrote:
>> On Fri, May 04, 2018 at 03:45:07PM +0000, Gilles Buloz wrote:
>>> Le 04/05/2018 00:31, Bjorn Helgaas a écrit :
>>>> [+cc LKML]
>>>>
>>>> On Thu, May 03, 2018 at 12:40:27PM +0000, Gilles Buloz wrote:
>>>>> Subject:    [PATCH] For exception at PCI probe due to bridge reporting UR
>>>>>
>>>>> Even if a device supports extended config access, no such access must be
>>>>> done to this device If there's a bridge not supporting that in the path
>>>>> to this device. Doing such access with UR reporting enabled on the root
>>>>> bridge leads to an exception.
>>>>>
>>>>> This is the case on a LS1043A CPU (NXP QorIQ Layerscape) platform with
>>>>> the following bus topology :
>>>>>     LS1043 PCIe root
>>>>>       -> PEX8112 PCIe-to-PCI bridge (not supporting ext cfg on PCI side)
>>>>>         -> PMC slot connector (for legacy PMC modules)
>>>>> With a PMC module topology as follows :
>>>>>     PMC connector
>>>>>       -> PCI-to-PCIe bridge
>>>>>         -> PCIe switch (4 ports)
>>>>>           -> 4 PCIe devices (one on each port)
>>>>> In this case all devices behind the PEX8112 are supporting extended config
>>>>> access but this is prohibited by the PEX8112. Without this patch, an
>>>>> exception (synchronous abort) occurs in pci_cfg_space_size_ext().
>>>>>
>>>>> This patch checks the parent bridge of each allocated child bus to know if
>>>>> extended config access is supported on the child bus, and sets a flag in
>>>>> child->bus_flags if not supported. This  flag is inherited by all children
>>>>> buses of this child bus and then is checked to avoid this unsupported
>>>>> accesses to every device on these buses.
>>>> Hi Gilles,
>>>>
>>>> Thanks for the patch!  I reworked it a little bit to simplify the code
>>>> in pci_alloc_child_bus().  Can you test it and make sure I didn't
>>>> break anything?
>>>>
>>> Hi Bjorn,
>>>
>>> Your rework works as expected. Tested on LS1043A platform with kernel
>>> 4.17-rc1, and with some backport on kernel 4.1.35
> Thanks for testing it!  I applied it to pci/enumeration for v4.18.
>
> I think the ASPM issue below is unrelated.  But I would like to figure out
> what's going on there, too, if you have any more information.
Hi Bjorn,

Here are full dmesg + lspci -vv + ... for the following cases :
- 4.17-rc1_probe_failed_without_pcieaspmoff.txt : I get a broken PCI config at most boots, without pcie_aspm=off
- 4.17-rc1_probe_ok_without_pcieaspmoff.txt : I get a correct PCI config at some boots, even without pcie_aspm=off
- 4.17-rc1_probe_always_ok_with_pcieaspmoff.txt : I get a correct PCI config at all boots when using pcie_aspm=off
- 4.1.35_probe_always_ok_without_pcieaspmoff.txt : I get a correct PCI config at all boots when using kernel 4.1.35-rt41 (from NXP 
Yocto BSP), even without pcie_aspm=off

I noticed these strange things :
- with 4.17-rc1, without pcie_aspm=off, the kernel is hanging during ~1 second :
   - before line "ASPM: Could not configure common clock" when PCI config is OK
   - or before line "bridge configuration invalid ([bus ff-ff]), reconfiguring" when PCI config is NOT OK
     (the FFs in "[bus ff-ff]" seems side effects of read returning ~0 because device is off)
- no such hang when pcie_aspm=off is used
- applying reverse patch for "PCI/ASPM: Don't warn if already in common clock mode"
   (commit 04875177dbe034055f23960981ecf6fb8ea1d638) does not give more message

>>> Info : with kernel 4.17-rc1, it turns out I need pcie_aspm=off to
>>> have the PMC devices behind the PCI-to-PCIe bridge of the PMC safely
>>> detected/configured. But this is not caused by the patch.
>>> Without pcie_aspm=off I saw this at one boot :
>>>      "pci 0000:02:0e.0: ASPM: Could not configure common clock" for this bridge, but devices
>>>      correctly detected/configured
>>> but at most boots I get :
>>>      no ASPM message but "pci 0000:04:02.0: bridge configuration invalid ([bus ff-ff]), reconfiguring "
>>>      instead, and some devices are missing. Also lspci show "rev ff" for some devices.
>>> I don't see this problem on 4.1.35 with the same backported patch.
>> This is interesting, especially since you have this unusual topology
>> of a path to the device that is PCIe, then conventional PCI, then PCIe
>> again.  We *should* be able to use ASPM on the PCIe links, but it's
>> definitely not a well-tested scenario.
>>
>> Can you tell if something is actually broken?  Sinan's recent change,
>> 04875177dbe0 ("PCI/ASPM: Don't warn if already in common clock mode"),
>> which appeared in v4.17-rc1, turns off the message in some cases.
>>
>> The "bridge configuration invalid" message just means the firmware
>> didn't configure the bridge.  We *should* still set it up correctly,
>> but please report a bug if we don't.
>>
>> lspci showing "ff" for some devices might be a symptom of the devices
>> being powered off.  In that case config reads normally return ~0 data
>> (though on your platform maybe it would cause exceptions).  I've seen
>> this in other situations and wondered if it would be worth adding a
>> hint to lspci so it could say "device may be powered off".
>>
>> Anyway, if you are seeing something broken (more than just the
>> messages), please start a new thread about each one.  If you do, could
>> you please:
>>
>>    - open a report at https://bugzilla.kernel.org/, in the Drivers/PCI
>>      component (open a separate bug for each issue you see)
>>
>>    - use kernel version 4.17-rc1 and mark it as a regression if
>>      appropriate
>>
>>    - attach (don't paste inline) the complete dmesg log and "lspci -vv"
>>      output (as root) to the bug
>>
>>    - post a note to linux-pci@...r.kernel.org, cc Fred, Sinan, and me,
>>      and include the link to the bugzilla
>>
>> Bjorn
> .
>


View attachment "4.1.35_probe_always_ok_without_pcieaspmoff.txt" of type "text/plain" (43211 bytes)

View attachment "4.17-rc1_probe_always_ok_with_pcieaspmoff.txt" of type "text/plain" (70092 bytes)

View attachment "4.17-rc1_probe_failed_without_pcieaspmoff.txt" of type "text/plain" (46121 bytes)

View attachment "4.17-rc1_probe_ok_without_pcieaspmoff.txt" of type "text/plain" (70260 bytes)

Powered by blists - more mailing lists