From: Bjorn Helgaas <helgaas@kernel.org>
To: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Hari Vyas <hari.vyas@broadcom.com>,
bhelgaas@google.com, linux-pci@vger.kernel.org,
ray.jui@broadcom.com
Subject: Re: PCIe enable device races (Was: [PATCH v3] PCI: Data corruption happening due to race condition)
Date: Wed, 15 Aug 2018 13:50:28 -0500 [thread overview]
Message-ID: <20180815185027.GE28888@bhelgaas-glaptop.roam.corp.google.com> (raw)
In-Reply-To: <ecaed7664b46d73888d2494065905f8e108fc0f4.camel@kernel.crashing.org>
On Wed, Aug 15, 2018 at 01:35:16PM +1000, Benjamin Herrenschmidt wrote:
> On Tue, 2018-07-31 at 11:37 -0500, Bjorn Helgaas wrote:
> > On Tue, Jul 03, 2018 at 02:35:40PM +0530, Hari Vyas wrote:
> > > Changes in v3:
> > > As per review comments from Lukas Wunner <lukas@wunner.de>,
> > > squashed 3 commits to single commit. Without this build breaks.
> > > Also clubbed set and clear function for is_added bits to a
> > > single assign function. This optimizes code and reduce LoC.
> > > Removed one wrongly added blank line in pci.c
> > >
> > > Changes in v2:
> > > To avoid race condition while updating is_added and is_busmaster
> > > bits, is_added is moved to a private flag variable.
> > > is_added updation is handled in atomic manner also.
> > >
> > > Hari Vyas (1):
> > > PCI: Data corruption happening due to race condition
>
> Sooo .... I was chasing a different problem which makes me think we
> have a deeper problem here.
>
> In my case, I have a system with >70 nvme devices behind layers of
> switches.
>
> What I think is happening is all the nvme devices are probed in
> parallel (the machine has about 40 CPU cores).
>
> They all call pci_enable_device() around the same time.
>
> This will walk up the bridge/switch chain and try to enable every
> switch along the way. However there is *no* locking at the switch level
> at all that I can see. Or am I missing something subtle ?
>
> So here's an example simplified scenario:
>
> Bridge
> / \
> dev A dev B
>
> Both dev A and B hit pci_enable_device() simultaneously, thus both
> call pci_enable_bridge() at the same time: This does (simplified):
>
> if (pci_is_enabled(dev)) {
> if (!dev->is_busmaster)
> pci_set_master(dev);
> return;
> }
>
> retval = pci_enable_device(dev);
> if (retval)
> pci_err(dev, "Error enabling bridge (%d), continuing\n",
> retval);
> pci_set_master(dev);
>
> Now the pci_is_enabled() just checks dev->enable_cnt and pci_enable_device()
> increments it *before* enabling the device.
>
> So it's possible that pci_is_enabled() returns true for the bridge for dev B
> because dev A just did the atomic_inc_return(), but hasn't actually enabled
> the bridge yet (hasnt yet hit the config space).
>
> At that point, driver for dev B hits an MMIO and gets an UR response from
> the bridge.
>
> I need to setup a rig to verify my theory but I think this is racy. The same
> race is also present with dev->is_busmaster. Using bitmaps won't help.
>
> What's really needed is a per device mutex covering all those operations
> on a given device. (This would also allow to get rid of those games with
> atomics).
Yes, this is definitely broken. Some folks have tried to fix it in
the past, but it hasn't quite happened yet. We actually merged one
patch, 40f11adc7cd9 ("PCI: Avoid race while enabling upstream
bridges"), but had to revert it after we found issues:
https://lkml.kernel.org/r/1501858648-22228-1-git-send-email-srinath.mannam@broadcom.com
https://lkml.kernel.org/r/20170915072352.10453.31977.stgit@bhelgaas-glaptop.roam.corp.google.com
next prev parent reply other threads:[~2018-08-15 21:43 UTC|newest]
Thread overview: 51+ messages / expand[flat|nested] mbox.gz Atom feed top
2018-07-03 9:05 [PATCH v3] PCI: Data corruption happening due to race condition Hari Vyas
2018-07-03 9:05 ` Hari Vyas
2018-07-03 9:13 ` Lukas Wunner
2018-07-18 23:29 ` Bjorn Helgaas
2018-07-19 4:18 ` Benjamin Herrenschmidt
2018-07-19 14:04 ` Hari Vyas
2018-07-19 18:55 ` Lukas Wunner
2018-07-20 4:27 ` Benjamin Herrenschmidt
2018-07-27 22:25 ` Bjorn Helgaas
2018-07-28 0:45 ` Benjamin Herrenschmidt
2018-07-31 11:21 ` Michael Ellerman
2018-07-19 17:41 ` Bjorn Helgaas
2018-07-20 9:16 ` Hari Vyas
2018-07-20 12:20 ` Bjorn Helgaas
2018-07-31 16:37 ` Bjorn Helgaas
2018-08-15 3:35 ` PCIe enable device races (Was: [PATCH v3] PCI: Data corruption happening due to race condition) Benjamin Herrenschmidt
2018-08-15 4:16 ` Benjamin Herrenschmidt
2018-08-15 4:44 ` Benjamin Herrenschmidt
2018-08-15 5:21 ` [RFC PATCH] pci: Proof of concept at fixing pci_enable_device/bridge races Benjamin Herrenschmidt
2018-08-15 19:09 ` PCIe enable device races (Was: [PATCH v3] PCI: Data corruption happening due to race condition) Bjorn Helgaas
2018-08-15 21:50 ` [RFC PATCH] pci: Proof of concept at fixing pci_enable_device/bridge races Benjamin Herrenschmidt
2018-08-15 22:40 ` Guenter Roeck
2018-08-15 23:38 ` Benjamin Herrenschmidt
2018-08-20 1:31 ` Guenter Roeck
2018-08-17 3:07 ` Bjorn Helgaas
2018-08-17 3:42 ` Benjamin Herrenschmidt
2018-08-15 18:50 ` Bjorn Helgaas [this message]
2018-08-15 21:52 ` PCIe enable device races (Was: [PATCH v3] PCI: Data corruption happening due to race condition) Benjamin Herrenschmidt
2018-08-15 23:23 ` Benjamin Herrenschmidt
2018-08-16 7:58 ` Konstantin Khlebnikov
2018-08-16 8:02 ` Benjamin Herrenschmidt
2018-08-16 9:22 ` Hari Vyas
2018-08-16 10:10 ` Benjamin Herrenschmidt
2018-08-16 10:11 ` Benjamin Herrenschmidt
2018-08-16 10:26 ` Lukas Wunner
2018-08-16 10:47 ` Hari Vyas
2018-08-16 23:20 ` Benjamin Herrenschmidt
2018-08-16 23:17 ` Benjamin Herrenschmidt
2018-08-17 0:43 ` Benjamin Herrenschmidt
2018-08-16 19:43 ` Jens Axboe
2018-08-16 21:37 ` Benjamin Herrenschmidt
2018-08-16 21:56 ` Jens Axboe
2018-08-16 23:09 ` Benjamin Herrenschmidt
2018-08-17 0:14 ` Jens Axboe
2018-08-16 12:28 ` Lukas Wunner
2018-08-16 23:25 ` Benjamin Herrenschmidt
2018-08-17 1:12 ` Benjamin Herrenschmidt
2018-08-17 16:39 ` Lukas Wunner
2018-08-18 3:37 ` Benjamin Herrenschmidt
2018-08-18 9:22 ` Lukas Wunner
2018-08-18 13:11 ` Benjamin Herrenschmidt
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=20180815185027.GE28888@bhelgaas-glaptop.roam.corp.google.com \
--to=helgaas@kernel.org \
--cc=benh@kernel.crashing.org \
--cc=bhelgaas@google.com \
--cc=hari.vyas@broadcom.com \
--cc=linux-pci@vger.kernel.org \
--cc=ray.jui@broadcom.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).