From: poza@codeaurora.org
To: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Thomas Tai <thomas.tai@oracle.com>,
bhelgaas@google.com, keith.busch@intel.com,
linux-pci@vger.kernel.org, linux-pci-owner@vger.kernel.org
Subject: Re: [PATCH 1/1] PCI/AER: prevent pcie_do_fatal_recovery from using device after it is removed
Date: Fri, 17 Aug 2018 16:14:17 +0530 [thread overview]
Message-ID: <a27b3466349bbed29653d4b2394ce715@codeaurora.org> (raw)
In-Reply-To: <65ec0c5b4682c0f0215114ac09d46cf5@codeaurora.org>
On 2018-08-17 15:59, poza@codeaurora.org wrote:
> On 2018-08-17 05:00, Benjamin Herrenschmidt wrote:
>> On Thu, 2018-08-16 at 19:41 +0530, poza@codeaurora.org wrote:
>>>
>>> > See above.
>>> > >
>>> > > okay, so here is what current pcie_do_nonfatal_recovery() doe.
>>> > >
>>> > > pcie_do_nonfatal_recovery
>>> > > report_error_detected() >> calls driver callbacks
>>> > > report_mmio_enabled()
>>> > > report_slot_reset() >> if PCI_ERS_RESULT_NEED_RESET
>>> >
>>> > Above if the driver returned "NEED RESET" we should not just "report" a
>>> > slot reset, we should *perform* one :-) Unless the AER code does it in
>>> > a place I missed...
>>>
>>> I am willing work on this if Bjorn agrees.
>>> but I am still trying to figure out missing piece.
>>>
>>> so Ben,
>>> you are suggesting
>>>
>>> ERR_NONFATAL handling
>>> pcie_do_nonfatal_recovery
>>> report_error_detected() >> calls driver callbacks
>>> report_mmio_enabled()
>>> report_slot_reset() >> if PCI_ERS_RESULT_NEED_RESET
>>> Here along with calling slot_reset, you are suggesting to do
>>> Secondary
>>> Bus Reset ?
>>>
>>> but this is ERR_NONFATAL and according to definition the link is
>>> still
>>> good, so that the transcriptions on PCIe link can happen.
>>> so my question is why do we want to reset the link ?
>>
>> The driver responded to you from error_detected() or mmio_enabled()
>> that it wants the slot to be reset. So you should do so. This comes
>> from the driver deciding that whatever happens, it doesn't trust the
>> device state anymore.
>>
>> report_slot_reset() iirc is about telling the driver that the reset
>> has
>> been performed and it can reconfigure the device.
>>
>> To be frank I haven't looked at this in a while, so we should refer to
>> the document before ou patched it :-) But the basic design we created
>> back then was precisely that the driver would have the ultimate say in
>> the matter.
>
> The existing design is; framework dictating drivers what it should do
> rather than driver deciding.
> let me explain.
>
> aer_isr
> aer_isr_one_error
> get_device_error_info >> this checks
> {
> /* Link is still healthy for IO reads */ >> Bjorn
> this looks like a very strange thing to me, if link is not healthy we
> are not even going for pcie_do_fatal_recovery()
I misread mask as severity, this code is ok, it just does not process
masked errors.
> pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS,
> &info->status);
> pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_MASK,
> &info->mask);
> if (!(info->status & ~info->mask))
> return 0;
>
> {
> handle_error_source
> pcie_do_nonfatal_recovery or pcie_do_fatal_recovery
>
> now it does not seem to me that driver has some way of knowing if it
> has to return PCI_ERS_RESULT_CAN_RECOVER ? or
> PCI_ERS_RESULT_NEED_RESET?
>
> let us take an example of storage driver here e.g. NVME
> nvme_error_detected() relies heavily on pci_channel_state_t which is
> passed by error framework (err.c)
>
> I understand your point of view and original intention of design that
> let driver dictate whether it wants slot_reset ?
> but driver relies on framework to pass that information in order to
> take decision.
>
> In case of pci_channel_io_frozen (which is ERR_FATAL), most of the
> drivers demand link reset.
>
> but in case of ERR_NONFATAL, the link is supposed to be functional and
> there is no need for link reset.
> and exactly all the PCI based drivers returns
> PCI_ERS_RESULT_CAN_RECOVER, which is valid.
>
>
> so I think to conclude, you are referring more of your views towards
> pcie_do_fatal_recovery()
>
>
> which does
>
>> remove_devices from the downstream link
>> reset_link() (Secondary bus reset)
>> re-enumerate.
>
> and this is what DPC defined, and AER is just following that.
>
> Regards,
> Oza.
>
>>
>> Also because multiple devices, at least on power, can share a domain,
>> we can get into a situation where one device requires a reset, so all
>> will get reset, and their respective drivers need to be notified.
>>
>> Note: it's not so much about resetting the *link* than about resetting
>> the *device*.
>>
>>> although
>>> I see following note in the code as well.
>>> /*
>>> * TODO: Should call platform-specific
>>> * functions to reset slot before calling
>>> * drivers' slot_reset callbacks?
>>> */
>>
>> Yup :-)
>>
>> Cheers,
>> Ben.
>>
>>> Regards,
>>> Oza.
>>>
>>> >
>>> > Also we should do a hot reset at least, not just a link reset.
>>> >
>>> > > report_resume()
>>> > >
>>> > > If you suggest how it is broken, it will help me to understand.
>>> > > probably you might want to point out what are the calls need to be
>>> > > added
>>> > > or removed or differently handled, specially storage point of view.
>>> >
>>> >
>>> > > Regards,
>>> > > Oza.
>>> > >
>>> > > >
>>> > > > Keep in mind that those callbacks were designed originally for EEH
>>> > > > (which predates AER), and so was the spec written.
>>> > > >
>>> > > > We don't actually use the AER code on POWER today, so we didn't notice
>>> > > > how broken the implementation was :-)
>>> > > >
>>> > > > We should fix that.
>>> > > >
>>> > > > Either we can sort all that out by email, or we should plan some kind
>>> > > > of catch-up, either at Plumbers (provided I can go there) or maybe a
>>> > > > webex call.
>>> > > >
>>> > > > Cheers,
>>> > > > Ben.
next prev parent reply other threads:[~2018-08-17 13:47 UTC|newest]
Thread overview: 91+ messages / expand[flat|nested] mbox.gz Atom feed top
2018-08-13 16:51 [PATCH 0/1] PCI/AER: prevent pcie_do_fatal_recovery from using device after it is removed Thomas Tai
2018-08-13 16:51 ` [PATCH 1/1] " Thomas Tai
2018-08-14 9:16 ` poza
2018-08-14 9:22 ` poza
2018-08-14 13:51 ` Thomas Tai
2018-08-15 14:57 ` poza
2018-08-15 15:02 ` Thomas Tai
2018-08-15 15:26 ` poza
2018-08-15 15:43 ` Thomas Tai
2018-08-15 15:59 ` poza
2018-08-15 16:04 ` Thomas Tai
2018-08-15 21:55 ` Benjamin Herrenschmidt
2018-08-15 21:56 ` Benjamin Herrenschmidt
2018-08-16 6:36 ` poza
2018-08-16 6:51 ` Benjamin Herrenschmidt
2018-08-16 6:59 ` Benjamin Herrenschmidt
2018-08-16 8:07 ` poza
2018-08-16 8:12 ` Benjamin Herrenschmidt
2018-08-16 9:03 ` poza
2018-08-16 10:07 ` Benjamin Herrenschmidt
2018-08-16 14:11 ` poza
2018-08-16 23:30 ` Benjamin Herrenschmidt
2018-08-17 10:29 ` poza
2018-08-17 10:44 ` poza [this message]
2018-08-18 7:38 ` Benjamin Herrenschmidt
2018-08-16 7:05 ` Benjamin Herrenschmidt
2018-08-16 7:15 ` Benjamin Herrenschmidt
2018-08-16 7:56 ` poza
2018-08-16 8:10 ` Benjamin Herrenschmidt
2018-08-16 8:05 ` poza
2018-08-16 8:15 ` Benjamin Herrenschmidt
2018-08-16 8:22 ` poza
2018-08-16 8:28 ` Benjamin Herrenschmidt
2018-08-16 13:30 ` Thomas Tai
2018-08-16 13:46 ` Sinan Kaya
2018-08-16 23:27 ` Benjamin Herrenschmidt
2018-08-17 6:35 ` poza
2018-08-19 2:24 ` Bjorn Helgaas
2018-08-20 5:09 ` poza
2018-08-20 5:15 ` Benjamin Herrenschmidt
2018-08-20 13:02 ` Thomas Tai
2018-08-20 13:27 ` Benjamin Herrenschmidt
2018-08-19 2:19 ` Bjorn Helgaas
2018-08-19 21:41 ` Sinan Kaya
2018-08-20 2:03 ` Benjamin Herrenschmidt
2018-08-20 5:19 ` poza
2018-08-20 5:33 ` Benjamin Herrenschmidt
2018-08-20 7:56 ` poza
2018-08-20 11:22 ` Benjamin Herrenschmidt
2018-08-20 13:26 ` poza
2018-08-20 21:02 ` Benjamin Herrenschmidt
2018-08-21 5:14 ` poza
2018-08-21 6:06 ` Benjamin Herrenschmidt
2018-08-21 14:37 ` Keith Busch
2018-08-21 15:07 ` Sinan Kaya
2018-08-21 15:29 ` Keith Busch
2018-08-21 15:50 ` Sinan Kaya
2018-08-21 15:55 ` Sinan Kaya
2018-08-21 16:44 ` Keith Busch
2018-08-21 15:30 ` poza
2018-08-21 21:14 ` Benjamin Herrenschmidt
2018-08-21 22:04 ` Keith Busch
2018-08-21 22:33 ` Keith Busch
2018-08-21 23:06 ` Benjamin Herrenschmidt
2018-08-21 23:13 ` Keith Busch
2018-08-22 0:36 ` Benjamin Herrenschmidt
2018-08-30 0:01 ` Keith Busch
2018-08-30 0:10 ` Sinan Kaya
2018-08-30 0:46 ` Keith Busch
2018-08-30 4:26 ` Benjamin Herrenschmidt
2018-08-20 15:53 ` Keith Busch
2018-08-20 16:13 ` poza
2018-08-20 16:32 ` Keith Busch
2018-08-20 21:05 ` Benjamin Herrenschmidt
2018-08-20 21:21 ` Sinan Kaya
2018-08-20 21:35 ` Keith Busch
2018-08-20 21:53 ` Benjamin Herrenschmidt
2018-08-20 22:02 ` Sinan Kaya
2018-08-20 22:04 ` Benjamin Herrenschmidt
2018-08-20 22:13 ` Sinan Kaya
2018-08-20 22:19 ` Benjamin Herrenschmidt
2018-08-22 9:13 ` Lukas Wunner
2018-08-22 14:38 ` Keith Busch
2018-08-22 14:51 ` Sinan Kaya
2018-08-20 22:13 ` Keith Busch
2018-08-20 22:19 ` Benjamin Herrenschmidt
2018-08-21 1:30 ` Keith Busch
2018-08-20 4:37 ` Benjamin Herrenschmidt
2018-08-20 4:39 ` PATCH] Partial revert of "PCI/AER: Handle ERR_FATAL with removal and re-enumeration of devices" Benjamin Herrenschmidt
2018-08-21 19:50 ` Bjorn Helgaas
2018-08-22 4:35 ` poza
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=a27b3466349bbed29653d4b2394ce715@codeaurora.org \
--to=poza@codeaurora.org \
--cc=benh@kernel.crashing.org \
--cc=bhelgaas@google.com \
--cc=keith.busch@intel.com \
--cc=linux-pci-owner@vger.kernel.org \
--cc=linux-pci@vger.kernel.org \
--cc=thomas.tai@oracle.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).