Herbert Xu recently reported a problem concerning RCU and compiler
barriers. In the course of discussing the problem, he put forth a
litmus test which illustrated a serious defect in the Linux Kernel
Memory Model's data-race-detection code.
The defect was that the LKMM assumed visibility and executes-before
ordering of plain accesses had to be mediated by marked accesses. In
Herbert's litmus test this wasn't so, and the LKMM claimed the litmus
test was allowed and contained a data race although neither is true.
In fact, plain accesses can be ordered by fences even in the absence
of marked accesses. In most cases this doesn't matter, because most
fences only order accesses within a single thread. But the rcu-fence
relation is different; it can order (and induce visibility between)
accesses in different threads -- events which otherwise might be
concurrent. This makes it relevant to data-race detection.
This patch makes two changes to the memory model to incorporate the
new insight:
If a store is separated by a fence from another access,
the store is necessarily visible to the other access (as
reflected in the ww-vis and wr-vis relations). Similarly,
if a load is separated by a fence from another access then
the load necessarily executes before the other access (as
reflected in the rw-xbstar relation).
If a store is separated by a strong fence from a marked access
then it is necessarily visible to any access that executes
after the marked access (as reflected in the ww-vis and wr-vis
relations).
With these changes, the LKMM gives the desired result for Herbert's
litmus test and other related ones.
Signed-off-by: Alan Stern <[email protected]>
Reported-by: Herbert Xu <[email protected]>
---
tools/memory-model/linux-kernel.cat | 8 +++++---
1 file changed, 5 insertions(+), 3 deletions(-)
Index: usb-devel/tools/memory-model/linux-kernel.cat
===================================================================
--- usb-devel.orig/tools/memory-model/linux-kernel.cat
+++ usb-devel/tools/memory-model/linux-kernel.cat
@@ -181,9 +181,11 @@ let r-post-bounded = (nonrw-fence | ([~N
[Marked]
(* Visibility and executes-before for plain accesses *)
-let ww-vis = w-post-bounded ; vis ; w-pre-bounded
-let wr-vis = w-post-bounded ; vis ; r-pre-bounded
-let rw-xbstar = r-post-bounded ; xbstar ; w-pre-bounded
+let ww-vis = fence | (strong-fence ; xbstar ; w-pre-bounded) |
+ (w-post-bounded ; vis ; w-pre-bounded)
+let wr-vis = fence | (strong-fence ; xbstar ; r-pre-bounded) |
+ (w-post-bounded ; vis ; r-pre-bounded)
+let rw-xbstar = fence | (r-post-bounded ; xbstar ; w-pre-bounded)
(* Potential races *)
let pre-race = ext & ((Plain * M) | ((M \ IW) * Plain))
On Thu, Jun 20, 2019 at 11:55:58AM -0400, Alan Stern wrote:
> Herbert Xu recently reported a problem concerning RCU and compiler
> barriers. In the course of discussing the problem, he put forth a
> litmus test which illustrated a serious defect in the Linux Kernel
> Memory Model's data-race-detection code.
>
> The defect was that the LKMM assumed visibility and executes-before
> ordering of plain accesses had to be mediated by marked accesses. In
> Herbert's litmus test this wasn't so, and the LKMM claimed the litmus
> test was allowed and contained a data race although neither is true.
>
> In fact, plain accesses can be ordered by fences even in the absence
> of marked accesses. In most cases this doesn't matter, because most
> fences only order accesses within a single thread. But the rcu-fence
> relation is different; it can order (and induce visibility between)
> accesses in different threads -- events which otherwise might be
> concurrent. This makes it relevant to data-race detection.
>
> This patch makes two changes to the memory model to incorporate the
> new insight:
>
> If a store is separated by a fence from another access,
> the store is necessarily visible to the other access (as
> reflected in the ww-vis and wr-vis relations). Similarly,
> if a load is separated by a fence from another access then
> the load necessarily executes before the other access (as
> reflected in the rw-xbstar relation).
>
> If a store is separated by a strong fence from a marked access
> then it is necessarily visible to any access that executes
> after the marked access (as reflected in the ww-vis and wr-vis
> relations).
>
> With these changes, the LKMM gives the desired result for Herbert's
> litmus test and other related ones.
>
> Signed-off-by: Alan Stern <[email protected]>
> Reported-by: Herbert Xu <[email protected]>
For the entire series:
Acked-by: Andrea Parri <[email protected]>
Two nits, but up to Paul AFAIAC:
- This is a first time for "tools: memory-model:" in Subject; we were
kind of converging to "tools/memory-model:"...
- The report preceded the patch; we might as well reflect this in the
order of the tags.
Thanks,
Andrea
On Fri, 21 Jun 2019, Andrea Parri wrote:
> On Thu, Jun 20, 2019 at 11:55:58AM -0400, Alan Stern wrote:
> > Herbert Xu recently reported a problem concerning RCU and compiler
> > barriers. In the course of discussing the problem, he put forth a
> > litmus test which illustrated a serious defect in the Linux Kernel
> > Memory Model's data-race-detection code.
> >
> > The defect was that the LKMM assumed visibility and executes-before
> > ordering of plain accesses had to be mediated by marked accesses. In
> > Herbert's litmus test this wasn't so, and the LKMM claimed the litmus
> > test was allowed and contained a data race although neither is true.
> >
> > In fact, plain accesses can be ordered by fences even in the absence
> > of marked accesses. In most cases this doesn't matter, because most
> > fences only order accesses within a single thread. But the rcu-fence
> > relation is different; it can order (and induce visibility between)
> > accesses in different threads -- events which otherwise might be
> > concurrent. This makes it relevant to data-race detection.
> >
> > This patch makes two changes to the memory model to incorporate the
> > new insight:
> >
> > If a store is separated by a fence from another access,
> > the store is necessarily visible to the other access (as
> > reflected in the ww-vis and wr-vis relations). Similarly,
> > if a load is separated by a fence from another access then
> > the load necessarily executes before the other access (as
> > reflected in the rw-xbstar relation).
> >
> > If a store is separated by a strong fence from a marked access
> > then it is necessarily visible to any access that executes
> > after the marked access (as reflected in the ww-vis and wr-vis
> > relations).
> >
> > With these changes, the LKMM gives the desired result for Herbert's
> > litmus test and other related ones.
> >
> > Signed-off-by: Alan Stern <[email protected]>
> > Reported-by: Herbert Xu <[email protected]>
>
> For the entire series:
>
> Acked-by: Andrea Parri <[email protected]>
>
> Two nits, but up to Paul AFAIAC:
>
> - This is a first time for "tools: memory-model:" in Subject; we were
> kind of converging to "tools/memory-model:"...
Yeah, sure. That's the sort of detail I have a hard time remembering.
> - The report preceded the patch; we might as well reflect this in the
> order of the tags.
Either way is okay with me.
Alan
On Fri, Jun 21, 2019 at 10:25:23AM -0400, Alan Stern wrote:
> On Fri, 21 Jun 2019, Andrea Parri wrote:
>
> > On Thu, Jun 20, 2019 at 11:55:58AM -0400, Alan Stern wrote:
> > > Herbert Xu recently reported a problem concerning RCU and compiler
> > > barriers. In the course of discussing the problem, he put forth a
> > > litmus test which illustrated a serious defect in the Linux Kernel
> > > Memory Model's data-race-detection code.
> > >
> > > The defect was that the LKMM assumed visibility and executes-before
> > > ordering of plain accesses had to be mediated by marked accesses. In
> > > Herbert's litmus test this wasn't so, and the LKMM claimed the litmus
> > > test was allowed and contained a data race although neither is true.
> > >
> > > In fact, plain accesses can be ordered by fences even in the absence
> > > of marked accesses. In most cases this doesn't matter, because most
> > > fences only order accesses within a single thread. But the rcu-fence
> > > relation is different; it can order (and induce visibility between)
> > > accesses in different threads -- events which otherwise might be
> > > concurrent. This makes it relevant to data-race detection.
> > >
> > > This patch makes two changes to the memory model to incorporate the
> > > new insight:
> > >
> > > If a store is separated by a fence from another access,
> > > the store is necessarily visible to the other access (as
> > > reflected in the ww-vis and wr-vis relations). Similarly,
> > > if a load is separated by a fence from another access then
> > > the load necessarily executes before the other access (as
> > > reflected in the rw-xbstar relation).
> > >
> > > If a store is separated by a strong fence from a marked access
> > > then it is necessarily visible to any access that executes
> > > after the marked access (as reflected in the ww-vis and wr-vis
> > > relations).
> > >
> > > With these changes, the LKMM gives the desired result for Herbert's
> > > litmus test and other related ones.
> > >
> > > Signed-off-by: Alan Stern <[email protected]>
> > > Reported-by: Herbert Xu <[email protected]>
> >
> > For the entire series:
> >
> > Acked-by: Andrea Parri <[email protected]>
> >
> > Two nits, but up to Paul AFAIAC:
> >
> > - This is a first time for "tools: memory-model:" in Subject; we were
> > kind of converging to "tools/memory-model:"...
>
> Yeah, sure. That's the sort of detail I have a hard time remembering.
>
> > - The report preceded the patch; we might as well reflect this in the
> > order of the tags.
>
> Either way is okay with me.
I applied Andrea's acks and edited as called out above, thank you both!
Thanx, Paul
Hi Paul and Alan,
On 2019/06/22 8:54, Paul E. McKenney wrote:
> On Fri, Jun 21, 2019 at 10:25:23AM -0400, Alan Stern wrote:
>> On Fri, 21 Jun 2019, Andrea Parri wrote:
>>
>>> On Thu, Jun 20, 2019 at 11:55:58AM -0400, Alan Stern wrote:
>>>> Herbert Xu recently reported a problem concerning RCU and compiler
>>>> barriers. In the course of discussing the problem, he put forth a
>>>> litmus test which illustrated a serious defect in the Linux Kernel
>>>> Memory Model's data-race-detection code.
I was not involved in the mail thread and wondering what the litmus test
looked like. Some searching of the archive has suggested that Alan presented
a properly formatted test based on Herbert's idea in [1].
[1]: https://lore.kernel.org/lkml/[email protected]/
If this is the case, adding the link (or message id) in the change
log would help people see the circumstances, I suppose.
Paul, can you amend the change log?
I ran herd7 on said litmus test at both "lkmm" and "dev" of -rcu and
confirmed that this patch fixes the result.
So,
Tested-by: Akira Yokosawa <[email protected]>
Thanks, Akira
>>>>
>>>> The defect was that the LKMM assumed visibility and executes-before
>>>> ordering of plain accesses had to be mediated by marked accesses. In
>>>> Herbert's litmus test this wasn't so, and the LKMM claimed the litmus
>>>> test was allowed and contained a data race although neither is true.
>>>>
>>>> In fact, plain accesses can be ordered by fences even in the absence
>>>> of marked accesses. In most cases this doesn't matter, because most
>>>> fences only order accesses within a single thread. But the rcu-fence
>>>> relation is different; it can order (and induce visibility between)
>>>> accesses in different threads -- events which otherwise might be
>>>> concurrent. This makes it relevant to data-race detection.
>>>>
>>>> This patch makes two changes to the memory model to incorporate the
>>>> new insight:
>>>>
>>>> If a store is separated by a fence from another access,
>>>> the store is necessarily visible to the other access (as
>>>> reflected in the ww-vis and wr-vis relations). Similarly,
>>>> if a load is separated by a fence from another access then
>>>> the load necessarily executes before the other access (as
>>>> reflected in the rw-xbstar relation).
>>>>
>>>> If a store is separated by a strong fence from a marked access
>>>> then it is necessarily visible to any access that executes
>>>> after the marked access (as reflected in the ww-vis and wr-vis
>>>> relations).
>>>>
>>>> With these changes, the LKMM gives the desired result for Herbert's
>>>> litmus test and other related ones.
>>>>
>>>> Signed-off-by: Alan Stern <[email protected]>
>>>> Reported-by: Herbert Xu <[email protected]>
>>>
>>> For the entire series:
>>>
>>> Acked-by: Andrea Parri <[email protected]>
>>>
>>> Two nits, but up to Paul AFAIAC:
>>>
>>> - This is a first time for "tools: memory-model:" in Subject; we were
>>> kind of converging to "tools/memory-model:"...
>>
>> Yeah, sure. That's the sort of detail I have a hard time remembering.
>>
>>> - The report preceded the patch; we might as well reflect this in the
>>> order of the tags.
>>
>> Either way is okay with me.
>
> I applied Andrea's acks and edited as called out above, thank you both!
>
> Thanx, Paul
>
On Sun, 23 Jun 2019, Akira Yokosawa wrote:
> Hi Paul and Alan,
>
> On 2019/06/22 8:54, Paul E. McKenney wrote:
> > On Fri, Jun 21, 2019 at 10:25:23AM -0400, Alan Stern wrote:
> >> On Fri, 21 Jun 2019, Andrea Parri wrote:
> >>
> >>> On Thu, Jun 20, 2019 at 11:55:58AM -0400, Alan Stern wrote:
> >>>> Herbert Xu recently reported a problem concerning RCU and compiler
> >>>> barriers. In the course of discussing the problem, he put forth a
> >>>> litmus test which illustrated a serious defect in the Linux Kernel
> >>>> Memory Model's data-race-detection code.
>
> I was not involved in the mail thread and wondering what the litmus test
> looked like. Some searching of the archive has suggested that Alan presented
> a properly formatted test based on Herbert's idea in [1].
>
> [1]: https://lore.kernel.org/lkml/[email protected]/
Yes, that's it. The test is also available at:
https://github.com/paulmckrcu/litmus/blob/master/manual/plain/C-S-rcunoderef-2.litmus
Alan
> If this is the case, adding the link (or message id) in the change
> log would help people see the circumstances, I suppose.
> Paul, can you amend the change log?
>
> I ran herd7 on said litmus test at both "lkmm" and "dev" of -rcu and
> confirmed that this patch fixes the result.
>
> So,
>
> Tested-by: Akira Yokosawa <[email protected]>
>
> Thanks, Akira
On Sun, Jun 23, 2019 at 11:15:06AM -0400, Alan Stern wrote:
> On Sun, 23 Jun 2019, Akira Yokosawa wrote:
>
> > Hi Paul and Alan,
> >
> > On 2019/06/22 8:54, Paul E. McKenney wrote:
> > > On Fri, Jun 21, 2019 at 10:25:23AM -0400, Alan Stern wrote:
> > >> On Fri, 21 Jun 2019, Andrea Parri wrote:
> > >>
> > >>> On Thu, Jun 20, 2019 at 11:55:58AM -0400, Alan Stern wrote:
> > >>>> Herbert Xu recently reported a problem concerning RCU and compiler
> > >>>> barriers. In the course of discussing the problem, he put forth a
> > >>>> litmus test which illustrated a serious defect in the Linux Kernel
> > >>>> Memory Model's data-race-detection code.
> >
> > I was not involved in the mail thread and wondering what the litmus test
> > looked like. Some searching of the archive has suggested that Alan presented
> > a properly formatted test based on Herbert's idea in [1].
> >
> > [1]: https://lore.kernel.org/lkml/[email protected]/
>
> Yes, that's it. The test is also available at:
>
> https://github.com/paulmckrcu/litmus/blob/master/manual/plain/C-S-rcunoderef-2.litmus
>
> Alan
>
> > If this is the case, adding the link (or message id) in the change
> > log would help people see the circumstances, I suppose.
> > Paul, can you amend the change log?
> >
> > I ran herd7 on said litmus test at both "lkmm" and "dev" of -rcu and
> > confirmed that this patch fixes the result.
> >
> > So,
> >
> > Tested-by: Akira Yokosawa <[email protected]>
Thank you both! I will apply these changes tomorrow morning, Pacific Time.
Thanx, Paul
On Sun, Jun 23, 2019 at 09:34:55PM -0700, Paul E. McKenney wrote:
> On Sun, Jun 23, 2019 at 11:15:06AM -0400, Alan Stern wrote:
> > On Sun, 23 Jun 2019, Akira Yokosawa wrote:
> >
> > > Hi Paul and Alan,
> > >
> > > On 2019/06/22 8:54, Paul E. McKenney wrote:
> > > > On Fri, Jun 21, 2019 at 10:25:23AM -0400, Alan Stern wrote:
> > > >> On Fri, 21 Jun 2019, Andrea Parri wrote:
> > > >>
> > > >>> On Thu, Jun 20, 2019 at 11:55:58AM -0400, Alan Stern wrote:
> > > >>>> Herbert Xu recently reported a problem concerning RCU and compiler
> > > >>>> barriers. In the course of discussing the problem, he put forth a
> > > >>>> litmus test which illustrated a serious defect in the Linux Kernel
> > > >>>> Memory Model's data-race-detection code.
> > >
> > > I was not involved in the mail thread and wondering what the litmus test
> > > looked like. Some searching of the archive has suggested that Alan presented
> > > a properly formatted test based on Herbert's idea in [1].
> > >
> > > [1]: https://lore.kernel.org/lkml/[email protected]/
> >
> > Yes, that's it. The test is also available at:
> >
> > https://github.com/paulmckrcu/litmus/blob/master/manual/plain/C-S-rcunoderef-2.litmus
> >
> > Alan
> >
> > > If this is the case, adding the link (or message id) in the change
> > > log would help people see the circumstances, I suppose.
> > > Paul, can you amend the change log?
> > >
> > > I ran herd7 on said litmus test at both "lkmm" and "dev" of -rcu and
> > > confirmed that this patch fixes the result.
> > >
> > > So,
> > >
> > > Tested-by: Akira Yokosawa <[email protected]>
>
> Thank you both! I will apply these changes tomorrow morning, Pacific Time.
And done. Please see below for the updated commit.
Thanx, Paul
------------------------------------------------------------------------
commit 46a020e9464aff884df56e5fd483134c8801e39f
Author: Alan Stern <[email protected]>
Date: Thu Jun 20 11:55:58 2019 -0400
tools/memory-model: Improve data-race detection
Herbert Xu recently reported a problem concerning RCU and compiler
barriers. In the course of discussing the problem, he put forth a
litmus test which illustrated a serious defect in the Linux Kernel
Memory Model's data-race-detection code [1].
The defect was that the LKMM assumed visibility and executes-before
ordering of plain accesses had to be mediated by marked accesses. In
Herbert's litmus test this wasn't so, and the LKMM claimed the litmus
test was allowed and contained a data race although neither is true.
In fact, plain accesses can be ordered by fences even in the absence
of marked accesses. In most cases this doesn't matter, because most
fences only order accesses within a single thread. But the rcu-fence
relation is different; it can order (and induce visibility between)
accesses in different threads -- events which otherwise might be
concurrent. This makes it relevant to data-race detection.
This patch makes two changes to the memory model to incorporate the
new insight:
If a store is separated by a fence from another access,
the store is necessarily visible to the other access (as
reflected in the ww-vis and wr-vis relations). Similarly,
if a load is separated by a fence from another access then
the load necessarily executes before the other access (as
reflected in the rw-xbstar relation).
If a store is separated by a strong fence from a marked access
then it is necessarily visible to any access that executes
after the marked access (as reflected in the ww-vis and wr-vis
relations).
With these changes, the LKMM gives the desired result for Herbert's
litmus test and other related ones [2].
[1] https://lore.kernel.org/lkml/[email protected]/
[2] https://github.com/paulmckrcu/litmus/blob/master/manual/plain/C-S-rcunoderef-1.litmus
https://github.com/paulmckrcu/litmus/blob/master/manual/plain/C-S-rcunoderef-2.litmus
https://github.com/paulmckrcu/litmus/blob/master/manual/plain/C-S-rcunoderef-3.litmus
https://github.com/paulmckrcu/litmus/blob/master/manual/plain/C-S-rcunoderef-4.litmus
Reported-by: Herbert Xu <[email protected]>
Signed-off-by: Alan Stern <[email protected]>
Acked-by: Andrea Parri <[email protected]>
Signed-off-by: Paul E. McKenney <[email protected]>
Tested-by: Akira Yokosawa <[email protected]>
diff --git a/tools/memory-model/linux-kernel.cat b/tools/memory-model/linux-kernel.cat
index ca2f4297b4e6..ea2ff4b94074 100644
--- a/tools/memory-model/linux-kernel.cat
+++ b/tools/memory-model/linux-kernel.cat
@@ -179,9 +179,11 @@ let r-post-bounded = (nonrw-fence | ([~Noreturn] ; fencerel(Rmb) ; [R4rmb]))? ;
[Marked]
(* Visibility and executes-before for plain accesses *)
-let ww-vis = w-post-bounded ; vis ; w-pre-bounded
-let wr-vis = w-post-bounded ; vis ; r-pre-bounded
-let rw-xbstar = r-post-bounded ; xbstar ; w-pre-bounded
+let ww-vis = fence | (strong-fence ; xbstar ; w-pre-bounded) |
+ (w-post-bounded ; vis ; w-pre-bounded)
+let wr-vis = fence | (strong-fence ; xbstar ; r-pre-bounded) |
+ (w-post-bounded ; vis ; r-pre-bounded)
+let rw-xbstar = fence | (r-post-bounded ; xbstar ; w-pre-bounded)
(* Potential races *)
let pre-race = ext & ((Plain * M) | ((M \ IW) * Plain))
On Mon, 24 Jun 2019, Paul E. McKenney wrote:
> On Sun, Jun 23, 2019 at 09:34:55PM -0700, Paul E. McKenney wrote:
> > On Sun, Jun 23, 2019 at 11:15:06AM -0400, Alan Stern wrote:
> > > On Sun, 23 Jun 2019, Akira Yokosawa wrote:
> > >
> > > > Hi Paul and Alan,
> > > >
> > > > On 2019/06/22 8:54, Paul E. McKenney wrote:
> > > > > On Fri, Jun 21, 2019 at 10:25:23AM -0400, Alan Stern wrote:
> > > > >> On Fri, 21 Jun 2019, Andrea Parri wrote:
> > > > >>
> > > > >>> On Thu, Jun 20, 2019 at 11:55:58AM -0400, Alan Stern wrote:
> > > > >>>> Herbert Xu recently reported a problem concerning RCU and compiler
> > > > >>>> barriers. In the course of discussing the problem, he put forth a
> > > > >>>> litmus test which illustrated a serious defect in the Linux Kernel
> > > > >>>> Memory Model's data-race-detection code.
> > > >
> > > > I was not involved in the mail thread and wondering what the litmus test
> > > > looked like. Some searching of the archive has suggested that Alan presented
> > > > a properly formatted test based on Herbert's idea in [1].
> > > >
> > > > [1]: https://lore.kernel.org/lkml/[email protected]/
> > >
> > > Yes, that's it. The test is also available at:
> > >
> > > https://github.com/paulmckrcu/litmus/blob/master/manual/plain/C-S-rcunoderef-2.litmus
> > >
> > > Alan
> > >
> > > > If this is the case, adding the link (or message id) in the change
> > > > log would help people see the circumstances, I suppose.
> > > > Paul, can you amend the change log?
> > > >
> > > > I ran herd7 on said litmus test at both "lkmm" and "dev" of -rcu and
> > > > confirmed that this patch fixes the result.
> > > >
> > > > So,
> > > >
> > > > Tested-by: Akira Yokosawa <[email protected]>
> >
> > Thank you both! I will apply these changes tomorrow morning, Pacific Time.
>
> And done. Please see below for the updated commit.
>
> Thanx, Paul
>
> ------------------------------------------------------------------------
>
> commit 46a020e9464aff884df56e5fd483134c8801e39f
> Author: Alan Stern <[email protected]>
> Date: Thu Jun 20 11:55:58 2019 -0400
>
> tools/memory-model: Improve data-race detection
>
> Herbert Xu recently reported a problem concerning RCU and compiler
> barriers. In the course of discussing the problem, he put forth a
> litmus test which illustrated a serious defect in the Linux Kernel
> Memory Model's data-race-detection code [1].
>
> The defect was that the LKMM assumed visibility and executes-before
> ordering of plain accesses had to be mediated by marked accesses. In
> Herbert's litmus test this wasn't so, and the LKMM claimed the litmus
> test was allowed and contained a data race although neither is true.
>
> In fact, plain accesses can be ordered by fences even in the absence
> of marked accesses. In most cases this doesn't matter, because most
> fences only order accesses within a single thread. But the rcu-fence
> relation is different; it can order (and induce visibility between)
> accesses in different threads -- events which otherwise might be
> concurrent. This makes it relevant to data-race detection.
>
> This patch makes two changes to the memory model to incorporate the
> new insight:
>
> If a store is separated by a fence from another access,
> the store is necessarily visible to the other access (as
> reflected in the ww-vis and wr-vis relations). Similarly,
> if a load is separated by a fence from another access then
> the load necessarily executes before the other access (as
> reflected in the rw-xbstar relation).
>
> If a store is separated by a strong fence from a marked access
> then it is necessarily visible to any access that executes
> after the marked access (as reflected in the ww-vis and wr-vis
> relations).
>
> With these changes, the LKMM gives the desired result for Herbert's
> litmus test and other related ones [2].
>
> [1] https://lore.kernel.org/lkml/[email protected]/
>
> [2] https://github.com/paulmckrcu/litmus/blob/master/manual/plain/C-S-rcunoderef-1.litmus
> https://github.com/paulmckrcu/litmus/blob/master/manual/plain/C-S-rcunoderef-2.litmus
> https://github.com/paulmckrcu/litmus/blob/master/manual/plain/C-S-rcunoderef-3.litmus
> https://github.com/paulmckrcu/litmus/blob/master/manual/plain/C-S-rcunoderef-4.litmus
Please add:
https://github.com/paulmckrcu/litmus/blob/master/manual/plain/strong-vis.litmus
Alan
On Mon, Jun 24, 2019 at 11:39:23AM -0400, Alan Stern wrote:
> On Mon, 24 Jun 2019, Paul E. McKenney wrote:
>
> > On Sun, Jun 23, 2019 at 09:34:55PM -0700, Paul E. McKenney wrote:
> > > On Sun, Jun 23, 2019 at 11:15:06AM -0400, Alan Stern wrote:
> > > > On Sun, 23 Jun 2019, Akira Yokosawa wrote:
> > > >
> > > > > Hi Paul and Alan,
> > > > >
> > > > > On 2019/06/22 8:54, Paul E. McKenney wrote:
> > > > > > On Fri, Jun 21, 2019 at 10:25:23AM -0400, Alan Stern wrote:
> > > > > >> On Fri, 21 Jun 2019, Andrea Parri wrote:
> > > > > >>
> > > > > >>> On Thu, Jun 20, 2019 at 11:55:58AM -0400, Alan Stern wrote:
> > > > > >>>> Herbert Xu recently reported a problem concerning RCU and compiler
> > > > > >>>> barriers. In the course of discussing the problem, he put forth a
> > > > > >>>> litmus test which illustrated a serious defect in the Linux Kernel
> > > > > >>>> Memory Model's data-race-detection code.
> > > > >
> > > > > I was not involved in the mail thread and wondering what the litmus test
> > > > > looked like. Some searching of the archive has suggested that Alan presented
> > > > > a properly formatted test based on Herbert's idea in [1].
> > > > >
> > > > > [1]: https://lore.kernel.org/lkml/[email protected]/
> > > >
> > > > Yes, that's it. The test is also available at:
> > > >
> > > > https://github.com/paulmckrcu/litmus/blob/master/manual/plain/C-S-rcunoderef-2.litmus
> > > >
> > > > Alan
> > > >
> > > > > If this is the case, adding the link (or message id) in the change
> > > > > log would help people see the circumstances, I suppose.
> > > > > Paul, can you amend the change log?
> > > > >
> > > > > I ran herd7 on said litmus test at both "lkmm" and "dev" of -rcu and
> > > > > confirmed that this patch fixes the result.
> > > > >
> > > > > So,
> > > > >
> > > > > Tested-by: Akira Yokosawa <[email protected]>
> > >
> > > Thank you both! I will apply these changes tomorrow morning, Pacific Time.
> >
> > And done. Please see below for the updated commit.
> >
> > Thanx, Paul
> >
> > ------------------------------------------------------------------------
> >
> > commit 46a020e9464aff884df56e5fd483134c8801e39f
> > Author: Alan Stern <[email protected]>
> > Date: Thu Jun 20 11:55:58 2019 -0400
> >
> > tools/memory-model: Improve data-race detection
> >
> > Herbert Xu recently reported a problem concerning RCU and compiler
> > barriers. In the course of discussing the problem, he put forth a
> > litmus test which illustrated a serious defect in the Linux Kernel
> > Memory Model's data-race-detection code [1].
> >
> > The defect was that the LKMM assumed visibility and executes-before
> > ordering of plain accesses had to be mediated by marked accesses. In
> > Herbert's litmus test this wasn't so, and the LKMM claimed the litmus
> > test was allowed and contained a data race although neither is true.
> >
> > In fact, plain accesses can be ordered by fences even in the absence
> > of marked accesses. In most cases this doesn't matter, because most
> > fences only order accesses within a single thread. But the rcu-fence
> > relation is different; it can order (and induce visibility between)
> > accesses in different threads -- events which otherwise might be
> > concurrent. This makes it relevant to data-race detection.
> >
> > This patch makes two changes to the memory model to incorporate the
> > new insight:
> >
> > If a store is separated by a fence from another access,
> > the store is necessarily visible to the other access (as
> > reflected in the ww-vis and wr-vis relations). Similarly,
> > if a load is separated by a fence from another access then
> > the load necessarily executes before the other access (as
> > reflected in the rw-xbstar relation).
> >
> > If a store is separated by a strong fence from a marked access
> > then it is necessarily visible to any access that executes
> > after the marked access (as reflected in the ww-vis and wr-vis
> > relations).
> >
> > With these changes, the LKMM gives the desired result for Herbert's
> > litmus test and other related ones [2].
> >
> > [1] https://lore.kernel.org/lkml/[email protected]/
> >
> > [2] https://github.com/paulmckrcu/litmus/blob/master/manual/plain/C-S-rcunoderef-1.litmus
> > https://github.com/paulmckrcu/litmus/blob/master/manual/plain/C-S-rcunoderef-2.litmus
> > https://github.com/paulmckrcu/litmus/blob/master/manual/plain/C-S-rcunoderef-3.litmus
> > https://github.com/paulmckrcu/litmus/blob/master/manual/plain/C-S-rcunoderef-4.litmus
>
> Please add:
>
> https://github.com/paulmckrcu/litmus/blob/master/manual/plain/strong-vis.litmus
Done, and calling this version final. Thank you all again!
Thanx, Paul