From: Benjamin Herrenschmidt <benh@kernel.crashing.org>
To: Jason Gunthorpe <jgg@nvidia.com>,
Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
Cc: Clint Sbisa <csbisa@amazon.com>,
linux-pci@vger.kernel.org, Bjorn Helgaas <helgaas@kernel.org>,
linux-arm-kernel@lists.infradead.org, will@kernel.org,
catalin.marinas@arm.com, Leon Romanovsky <leon@kernel.org>
Subject: Re: [PATCH] arm64: Enable PCI write-combine resources under sysfs
Date: Fri, 11 Sep 2020 07:46:47 +1000 [thread overview]
Message-ID: <44acc22377958a57c738f5139c5b5df2841c2544.camel@kernel.crashing.org> (raw)
In-Reply-To: <20200910171033.GG904879@nvidia.com>
On Thu, 2020-09-10 at 14:10 -0300, Jason Gunthorpe wrote:
> Can you explain what this actually does on ARM?
>
> Can it ever speculate loads across page boundaries, or speculate
> loads
> that never exist in the program? ie will we get random unpredicable
> MemRds?
Probably, at least on powerpc you will as well, that's the only way to
get write combine.
> Does it/could it "combine writes"?
I assume so for ARM, definitely for powerpc.
> > > If the CPU fails to generate a 64 byte TLP then the device will
> > > still
> > > operate correctly but does a different, slower, flow.
> >
> > Side note: on ARM that TLP is not a native interconnect
> > transaction,
> > reworded, it depends on what the system-bus->PCI logic does in
> > this respect.
>
> I think the issue is that ARM never defined what the bits set by
> pgprot_writecombine() do at a system level so we see implementations
> that do not cause write combining at the PCI-E interface for those
> bits. (I assume from what I've heard)
Nobody did. I think only x86 has a real "write combine" attribute. I
tried to untangled that mess years ago and didnt' get to the bottom of
it, but basically, on non-x86 archs, pgprot_writecombine will give you
what you asked ... and more.
> > That's why I looped you in - that's what worries me about
> > "enabling"
> > arch_can_pci_mmap_wc() on arm64. If we enable it and we have perf
> > regressions that's not OK.
> >
> > Or we *can* enable arch_can_pci_mmap_wc() but force the mellanox
> > driver (or more broadly all drivers following this message push
> > semantics) to use "something else" for WC detection.
>
> arch_can_pci_mmap_wc() really only controls the sysfs resource file
> and it seems very unclear who in userspace uses that these days.
dpdk under some circumstances afaik.
> vfio is now the right way to do that stuff. I don't see an obvious
> way to get WC memory in VFIO though...
Which would be a performance issue on a number of things I suppose...
Cheers,
Ben.
WARNING: multiple messages have this Message-ID (diff)
From: Benjamin Herrenschmidt <benh@kernel.crashing.org>
To: Jason Gunthorpe <jgg@nvidia.com>,
Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
Cc: Leon Romanovsky <leon@kernel.org>,
linux-pci@vger.kernel.org, Bjorn Helgaas <helgaas@kernel.org>,
catalin.marinas@arm.com, will@kernel.org,
linux-arm-kernel@lists.infradead.org,
Clint Sbisa <csbisa@amazon.com>
Subject: Re: [PATCH] arm64: Enable PCI write-combine resources under sysfs
Date: Fri, 11 Sep 2020 07:46:47 +1000 [thread overview]
Message-ID: <44acc22377958a57c738f5139c5b5df2841c2544.camel@kernel.crashing.org> (raw)
In-Reply-To: <20200910171033.GG904879@nvidia.com>
On Thu, 2020-09-10 at 14:10 -0300, Jason Gunthorpe wrote:
> Can you explain what this actually does on ARM?
>
> Can it ever speculate loads across page boundaries, or speculate
> loads
> that never exist in the program? ie will we get random unpredicable
> MemRds?
Probably, at least on powerpc you will as well, that's the only way to
get write combine.
> Does it/could it "combine writes"?
I assume so for ARM, definitely for powerpc.
> > > If the CPU fails to generate a 64 byte TLP then the device will
> > > still
> > > operate correctly but does a different, slower, flow.
> >
> > Side note: on ARM that TLP is not a native interconnect
> > transaction,
> > reworded, it depends on what the system-bus->PCI logic does in
> > this respect.
>
> I think the issue is that ARM never defined what the bits set by
> pgprot_writecombine() do at a system level so we see implementations
> that do not cause write combining at the PCI-E interface for those
> bits. (I assume from what I've heard)
Nobody did. I think only x86 has a real "write combine" attribute. I
tried to untangled that mess years ago and didnt' get to the bottom of
it, but basically, on non-x86 archs, pgprot_writecombine will give you
what you asked ... and more.
> > That's why I looped you in - that's what worries me about
> > "enabling"
> > arch_can_pci_mmap_wc() on arm64. If we enable it and we have perf
> > regressions that's not OK.
> >
> > Or we *can* enable arch_can_pci_mmap_wc() but force the mellanox
> > driver (or more broadly all drivers following this message push
> > semantics) to use "something else" for WC detection.
>
> arch_can_pci_mmap_wc() really only controls the sysfs resource file
> and it seems very unclear who in userspace uses that these days.
dpdk under some circumstances afaik.
> vfio is now the right way to do that stuff. I don't see an obvious
> way to get WC memory in VFIO though...
Which would be a performance issue on a number of things I suppose...
Cheers,
Ben.
_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
next prev parent reply other threads:[~2020-09-10 22:00 UTC|newest]
Thread overview: 117+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-08-31 15:18 [PATCH] arm64: Enable PCI write-combine resources under sysfs Clint Sbisa
2020-08-31 23:24 ` Benjamin Herrenschmidt
2020-09-01 18:37 ` Bjorn Helgaas
2020-09-01 23:22 ` Benjamin Herrenschmidt
2020-09-02 8:57 ` Will Deacon
2020-09-02 11:32 ` Lorenzo Pieralisi
2020-09-02 14:29 ` Clint Sbisa
2020-09-02 16:47 ` Lorenzo Pieralisi
2020-09-02 16:47 ` Lorenzo Pieralisi
2020-09-02 17:21 ` Catalin Marinas
2020-09-02 17:21 ` Catalin Marinas
2020-09-02 17:54 ` Lorenzo Pieralisi
2020-09-02 17:54 ` Lorenzo Pieralisi
2020-09-02 23:03 ` Benjamin Herrenschmidt
2020-09-02 23:03 ` Benjamin Herrenschmidt
2020-09-02 23:08 ` Benjamin Herrenschmidt
2020-09-02 23:08 ` Benjamin Herrenschmidt
2020-09-02 23:08 ` Benjamin Herrenschmidt
2020-09-02 23:08 ` Benjamin Herrenschmidt
2020-09-02 23:07 ` Benjamin Herrenschmidt
2020-09-02 23:07 ` Benjamin Herrenschmidt
2020-09-03 11:08 ` Lorenzo Pieralisi
2020-09-03 11:08 ` Lorenzo Pieralisi
2020-09-03 14:36 ` Clint Sbisa
2020-09-03 14:36 ` Clint Sbisa
2020-09-03 22:26 ` Benjamin Herrenschmidt
2020-09-03 22:26 ` Benjamin Herrenschmidt
2020-09-07 23:33 ` Benjamin Herrenschmidt
2020-09-07 23:33 ` Benjamin Herrenschmidt
2020-09-10 9:46 ` Lorenzo Pieralisi
2020-09-10 9:46 ` Lorenzo Pieralisi
2020-09-10 10:54 ` Leon Romanovsky
2020-09-10 10:54 ` Leon Romanovsky
2020-09-10 12:37 ` Jason Gunthorpe
2020-09-10 12:37 ` Jason Gunthorpe
2020-09-10 15:17 ` Lorenzo Pieralisi
2020-09-10 15:17 ` Lorenzo Pieralisi
2020-09-10 17:10 ` Jason Gunthorpe
2020-09-10 17:10 ` Jason Gunthorpe
2020-09-10 21:46 ` Benjamin Herrenschmidt [this message]
2020-09-10 21:46 ` Benjamin Herrenschmidt
2020-09-10 23:29 ` Jason Gunthorpe
2020-09-10 23:29 ` Jason Gunthorpe
2020-09-11 0:39 ` Benjamin Herrenschmidt
2020-09-11 0:39 ` Benjamin Herrenschmidt
2020-09-11 14:21 ` Jason Gunthorpe
2020-09-11 14:21 ` Jason Gunthorpe
2020-09-11 21:42 ` Clint Sbisa
2020-09-11 21:42 ` Clint Sbisa
2020-09-14 14:17 ` Jason Gunthorpe
2020-09-14 14:17 ` Jason Gunthorpe
2020-09-14 14:24 ` Clint Sbisa
2020-09-14 14:24 ` Clint Sbisa
2020-09-14 14:38 ` Jason Gunthorpe
2020-09-14 14:38 ` Jason Gunthorpe
2020-09-14 21:42 ` Benjamin Herrenschmidt
2020-09-14 21:42 ` Benjamin Herrenschmidt
2020-09-14 22:00 ` Benjamin Herrenschmidt
2020-09-14 22:00 ` Benjamin Herrenschmidt
2020-09-14 22:32 ` Clint Sbisa
2020-09-14 22:32 ` Clint Sbisa
2020-09-14 22:57 ` Jason Gunthorpe
2020-09-14 22:57 ` Jason Gunthorpe
2020-09-14 23:25 ` Benjamin Herrenschmidt
2020-09-14 23:25 ` Benjamin Herrenschmidt
2020-09-15 10:18 ` Lorenzo Pieralisi
2020-09-15 10:18 ` Lorenzo Pieralisi
2020-09-15 11:05 ` Jason Gunthorpe
2020-09-15 11:05 ` Jason Gunthorpe
2020-09-15 23:17 ` Benjamin Herrenschmidt
2020-09-15 23:17 ` Benjamin Herrenschmidt
2020-09-15 23:40 ` Jason Gunthorpe
2020-09-15 23:40 ` Jason Gunthorpe
2020-09-16 7:59 ` Benjamin Herrenschmidt
2020-09-16 7:59 ` Benjamin Herrenschmidt
2020-09-16 12:12 ` Jason Gunthorpe
2020-09-16 12:12 ` Jason Gunthorpe
2020-09-16 14:09 ` Lorenzo Pieralisi
2020-09-16 14:09 ` Lorenzo Pieralisi
2020-09-16 14:14 ` Jason Gunthorpe
2020-09-16 14:14 ` Jason Gunthorpe
2020-09-16 23:59 ` Benjamin Herrenschmidt
2020-09-16 23:59 ` Benjamin Herrenschmidt
2020-09-17 10:28 ` Lorenzo Pieralisi
2020-09-17 10:28 ` Lorenzo Pieralisi
2020-09-17 11:32 ` Jason Gunthorpe
2020-09-17 11:32 ` Jason Gunthorpe
2020-09-17 14:01 ` Lorenzo Pieralisi
2020-09-17 14:01 ` Lorenzo Pieralisi
2020-09-17 16:08 ` Will Deacon
2020-09-17 16:08 ` Will Deacon
2020-09-16 12:48 ` Leon Romanovsky
2020-09-16 12:48 ` Leon Romanovsky
2020-09-16 8:33 ` Will Deacon
2020-09-16 8:33 ` Will Deacon
2020-09-16 8:48 ` Catalin Marinas
2020-09-16 8:48 ` Catalin Marinas
2020-09-16 14:15 ` Lorenzo Pieralisi
2020-09-16 14:15 ` Lorenzo Pieralisi
2020-09-16 17:00 ` Catalin Marinas
2020-09-16 17:00 ` Catalin Marinas
2020-09-16 21:29 ` Benjamin Herrenschmidt
2020-09-16 21:29 ` Benjamin Herrenschmidt
2020-09-16 12:08 ` Jason Gunthorpe
2020-09-16 12:08 ` Jason Gunthorpe
2020-09-15 23:00 ` Benjamin Herrenschmidt
2020-09-15 23:00 ` Benjamin Herrenschmidt
2020-09-15 23:12 ` Clint Sbisa
2020-09-15 23:12 ` Clint Sbisa
2020-09-14 21:41 ` Benjamin Herrenschmidt
2020-09-14 21:41 ` Benjamin Herrenschmidt
-- strict thread matches above, loose matches on Subject: below --
2020-08-21 15:51 Clint Sbisa
2020-08-21 15:51 ` Clint Sbisa
2020-08-27 14:41 ` Clint Sbisa
2020-08-27 14:41 ` Clint Sbisa
2020-08-31 15:22 ` Clint Sbisa
2020-08-31 15:22 ` Clint Sbisa
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=44acc22377958a57c738f5139c5b5df2841c2544.camel@kernel.crashing.org \
--to=benh@kernel.crashing.org \
--cc=catalin.marinas@arm.com \
--cc=csbisa@amazon.com \
--cc=helgaas@kernel.org \
--cc=jgg@nvidia.com \
--cc=leon@kernel.org \
--cc=linux-arm-kernel@lists.infradead.org \
--cc=linux-pci@vger.kernel.org \
--cc=lorenzo.pieralisi@arm.com \
--cc=will@kernel.org \
/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 an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.