From: Toshi Kani <toshi.kani@hpe.com>
To: Ingo Molnar <mingo@kernel.org>, "Luis R. Rodriguez" <mcgrof@kernel.org>
Cc: Toshi Kani <toshi.kani@hp.com>,
Paul McKenney <paulmck@linux.vnet.ibm.com>,
Dave Airlie <airlied@redhat.com>,
Benjamin Herrenschmidt <benh@kernel.crashing.org>,
"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
linux-arch@vger.kernel.org, X86 ML <x86@kernel.org>,
Daniel Vetter <daniel.vetter@intel.com>,
Thomas Gleixner <tglx@linutronix.de>,
"H. Peter Anvin" <hpa@zytor.com>,
Peter Zijlstra <a.p.zijlstra@chello.nl>,
Borislav Petkov <bp@alien8.de>,
Linus Torvalds <torvalds@linux-foundation.org>,
Andrew Morton <akpm@linux-foundation.org>,
Andy Lutomirski <luto@kernel.org>,
Brian Gerst <brgerst@gmail.com>
Subject: Re: Overlapping ioremap() calls, set_memory_*() semantics
Date: Fri, 04 Mar 2016 11:18:34 -0700 [thread overview]
Message-ID: <1457115514.15454.216.camel@hpe.com> (raw)
In-Reply-To: <20160304094424.GA16228@gmail.com>
On Fri, 2016-03-04 at 10:44 +0100, Ingo Molnar wrote:
> * Luis R. Rodriguez <mcgrof@kernel.org> wrote:
>
> > At kernel summit, during the semantics of ioremap() session, Paul
> > mentioned we'd write something up to help get some notes out on what
> > we need to do and help clarify things. I've run into an issue (just a
> > warning) with a user on some odd system that I suspect may be the
> > result of a driver using overlapping ioremap() calls on conflicting
> > memory regions, so I'm a bit interested to see a resolution to some of
> > these lingering discussions now.
> >
> > Although we spoke of quite a bit of things, I raised in particular the
> > 'semantics of overlapping ioremap()' calls as one item of interest we
> > should address across architectures. At least on x86 it seems we would
> > not get an error if this is done and in fact its expected behavior;
> > Toshi had determined we could not enable error'ing out on overlapping
> > ioremap() calls given we have a few users that use it intentionally,
> > for instance the /dev/mem setup code. I had suggested long ago then
> > that one possible resolution was for us to add an API that *enables*
> > overlapping ioremap() calls, and only use it on select locations in
> > the kernel. This means we only have to convert a few users to that
> > call to white list such semantics, and by default we'd disable
> > overlapping calls. To kick things off -- is this strategy agreeable
> > for all other architectures?
>
> So I'd say that since ioremap() in itself is fragile enough, we should
> work towards eliminating overlapping ranges.
>
> The thing is, the whole vmap_area logic is based around non-overlapping
> ranges, sorted into the vmap_area_root rbtree.
>
> Just check the logic in mm/vmalloc.c::alloc_vmap_area(): it's based on
> finding holes in the kernel-virtual allocations. 'Overlapping ranges' is
> very much not part of that logic, at least to my understanding.
>
> How are overlapping ioremap()s even possible with that logic? The
> allocator searches for holes, not allowing for overlaps. What am I
> missing?
>
> Could you outline a specific case where it's done intentionally - and the
> purpose behind that intention?
The term "overlapping" is a bit misleading. This is "alias" mapping -- a
physical address range is mapped to multiple virtual address ranges. There
is no overlapping in VMA.
Such alias mappings are used by multiple modules. For instance, a PMEM
range is mapped to the kernel and user spaces. /dev/mem is another example
that creates a user space mapping to a physical address where other
mappings may already exist.
Hence, alias mapping itself is a supported use-case. However, alias
mapping with different cache types is not as it causes undefined behavior.
Therefore, PAT module protects from this case by tracking cache types used
for mapping physical ranges. When a different cache type is requested,
is_new_memtype_allowed() checks if the request needs to be failed or can be
changed to the existing type.
I agree that the current implementation is fragile, and some interfaces
skip such check at all, ex. vm_insert_pfn().
> > The problem is that without this it remains up to the developer of the
> > driver to avoid overlapping calls, and a user may just get sporadic
> > errors if this happens. As another problem case, set_memor_*() will
> > not fail on MMIO even though set_memor_*() is designed only for RAM. If
> > the above strategy on avoiding overlapping is agreeable, could the next
> > step, or an orthogonal step be to error out on set_memory_*() on IO
> > memory?
>
> So how do drivers set up WC or WB MMIO areas? Does none of our upstream
> video drivers do that?
Drivers use ioremap family with a right cache type when mapping MMIO
ranges, ex. ioremap_wc(). They do not need to change the type to MMIO.
RAM is different since it's already mapped with WB at boot-time.
set_memory_*() allows us to change the type from WB, and put it back to
WB.
Thanks,
-Toshi
next prev parent reply other threads:[~2016-03-04 18:18 UTC|newest]
Thread overview: 28+ messages / expand[flat|nested] mbox.gz Atom feed top
2016-03-03 21:28 Overlapping ioremap() calls, set_memory_*() semantics Luis R. Rodriguez
2016-03-03 21:28 ` Luis R. Rodriguez
2016-03-04 9:44 ` Ingo Molnar
2016-03-04 18:18 ` Toshi Kani [this message]
2016-03-04 18:18 ` Toshi Kani
2016-03-04 18:51 ` Luis R. Rodriguez
2016-03-04 21:39 ` Toshi Kani
2016-03-05 11:42 ` Ingo Molnar
2016-03-05 11:40 ` Ingo Molnar
2016-03-07 17:03 ` Toshi Kani
2016-03-07 17:03 ` Toshi Kani
2016-03-08 12:16 ` Ingo Molnar
2016-03-09 0:29 ` Toshi Kani
2016-03-09 9:15 ` Ingo Molnar
2016-03-11 22:13 ` Toshi Kani
2016-03-16 1:45 ` Luis R. Rodriguez
2016-03-16 1:45 ` Luis R. Rodriguez
2016-03-17 22:44 ` Toshi Kani
2016-04-13 21:16 ` Luis R. Rodriguez
2016-04-15 14:47 ` Toshi Kani
2016-04-15 14:47 ` Toshi Kani
2016-04-16 9:20 ` Ingo Molnar
2016-04-16 9:20 ` Ingo Molnar
2016-03-21 17:38 ` Maciej W. Rozycki
2016-04-13 21:03 ` Luis R. Rodriguez
2016-03-11 6:47 ` Andy Lutomirski
2016-03-11 22:36 ` Toshi Kani
2016-03-13 1:02 ` Andy Lutomirski
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=1457115514.15454.216.camel@hpe.com \
--to=toshi.kani@hpe.com \
--cc=a.p.zijlstra@chello.nl \
--cc=airlied@redhat.com \
--cc=akpm@linux-foundation.org \
--cc=benh@kernel.crashing.org \
--cc=bp@alien8.de \
--cc=brgerst@gmail.com \
--cc=daniel.vetter@intel.com \
--cc=hpa@zytor.com \
--cc=linux-arch@vger.kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=luto@kernel.org \
--cc=mcgrof@kernel.org \
--cc=mingo@kernel.org \
--cc=paulmck@linux.vnet.ibm.com \
--cc=tglx@linutronix.de \
--cc=torvalds@linux-foundation.org \
--cc=toshi.kani@hp.com \
--cc=x86@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 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).