From mboxrd@z Thu Jan 1 00:00:00 1970 From: Boqun Feng Subject: Re: [PATCH v2] lkmm/docs: Correct ->prop example with additional rfe link Date: Sun, 28 Jul 2019 23:28:06 +0800 Message-ID: <20190728152806.GB26905@tardis> References: <20190728031303.164545-1-joel@joelfernandes.org> <20190728151959.GA82871@google.com> Mime-Version: 1.0 Content-Type: multipart/signed; micalg=pgp-sha256; protocol="application/pgp-signature"; boundary="E39vaYmALEf/7YXx" Return-path: Content-Disposition: inline In-Reply-To: <20190728151959.GA82871@google.com> Sender: linux-kernel-owner@vger.kernel.org To: Joel Fernandes Cc: Alan Stern , linux-kernel@vger.kernel.org, kernel-team@android.com, Akira Yokosawa , Andrea Parri , Daniel Lustig , David Howells , Ingo Molnar , Jade Alglave , linux-arch@vger.kernel.org, Luc Maranget , Nicholas Piggin , "Paul E. McKenney" , Peter Zijlstra , Will Deacon List-Id: linux-arch.vger.kernel.org --E39vaYmALEf/7YXx Content-Type: text/plain; charset=us-ascii Content-Disposition: inline Content-Transfer-Encoding: quoted-printable On Sun, Jul 28, 2019 at 11:19:59AM -0400, Joel Fernandes wrote: > On Sun, Jul 28, 2019 at 10:48:51AM -0400, Alan Stern wrote: > > On Sat, 27 Jul 2019, Joel Fernandes (Google) wrote: > >=20 > > > The lkmm example about ->prop relation should describe an additional = rfe > > > link between P1's store to y and P2's load of y, which should be > > > critical to establishing the ordering resulting in the ->prop ordering > > > on P0. IOW, there are 2 rfe links, not one. > > >=20 > > > Correct these in the docs to make the ->prop ordering on P0 more clea= r. > > >=20 > > > Cc: kernel-team@android.com > > > Reviewed-by: Boqun Feng > > > Signed-off-by: Joel Fernandes (Google) > > > --- > >=20 > > This is not a good update. See below... >=20 > No problem, thanks for the feedback. I am new to the LKMM so please bear > with me.. I should have tagged this RFC. >=20 > > > .../memory-model/Documentation/explanation.txt | 17 ++++++++++-----= -- > > > 1 file changed, 10 insertions(+), 7 deletions(-) > > >=20 > > > diff --git a/tools/memory-model/Documentation/explanation.txt b/tools= /memory-model/Documentation/explanation.txt > > > index 68caa9a976d0..aa84fce854cc 100644 > > > --- a/tools/memory-model/Documentation/explanation.txt > > > +++ b/tools/memory-model/Documentation/explanation.txt > > > @@ -1302,8 +1302,8 @@ followed by an arbitrary number of cumul-fence = links, ending with an > > > rfe link. You can concoct more exotic examples, containing more than > > > one fence, although this quickly leads to diminishing returns in ter= ms > > > of complexity. For instance, here's an example containing a coe link > > > -followed by two fences and an rfe link, utilizing the fact that > > > -release fences are A-cumulative: > > > +followed by a fence, an rfe link, another fence and and a final rfe = link, > > ^---^ > > > +utilizing the fact that release fences are A-cumulative: > >=20 > > I don't like this, for two reasons. First is the repeated "and" typo. >=20 > Will fix the trivial typo, sorry about that. >=20 > > More importantly, it's not necessary to go into this level of detail; a > > better revision would be: > >=20 > > +followed by two cumul-fences and an rfe link, utilizing the fact that > >=20 > > This is appropriate because the cumul-fence relation is defined to=20 > > contain the rfe link which you noticed wasn't mentioned explicitly. >=20 > No, I am talking about the P1's store to Y and P2's load of Y. That is not > through a cumul-fence so I don't understand what you meant? That _is_ mis= sing > the rfe link I am referring to, that is left out. >=20 > The example says r2 =3D 1 and then we work backwards from that. r2 could = very > well have been 0, there's no fence or anything involved, it just happens = to > be the executation pattern causing r2 =3D 1 and hence the rfe link. Right? >=20 > > > int x, y, z; > > > =20 > > > @@ -1334,11 +1334,14 @@ If x =3D 2, r0 =3D 1, and r2 =3D 1 after this= code runs then there is a prop > > > link from P0's store to its load. This is because P0's store gets > > > overwritten by P1's store since x =3D 2 at the end (a coe link), the > > > smp_wmb() ensures that P1's store to x propagates to P2 before the > > > -store to y does (the first fence), the store to y propagates to P2 > > > -before P2's load and store execute, P2's smp_store_release() > > > -guarantees that the stores to x and y both propagate to P0 before the > > > -store to z does (the second fence), and P0's load executes after the > > > -store to z has propagated to P0 (an rfe link). > > > +store to y does (the first fence), P2's store to y happens before P2= 's > > ---------------------------------------^ > >=20 > > This makes no sense, since P2 doesn't store to y. You meant P1's store > > to y. Also, the use of "happens before" is here unnecessarily > > ambiguous (is it an informal usage or does it refer to the formal > > happens-before relation?). The original "propagates to" is better. >=20 > Will reword this. >=20 > > > +load of y (rfe link), P2's smp_store_release() ensures that P2's load > > > +of y executes before P2's store to z (second fence), which implies t= hat > > > +that stores to x and y propagate to P2 before the smp_store_release(= ), which > > > +means that P2's smp_store_release() will propagate stores to x and y= to all > > > +CPUs before the store to z propagates (A-cumulative property of this= fence). > > > +Finally P0's load of z executes after P2's store to z has propagated= to > > > +P0 (rfe link). > >=20 > > Again, a better change would be simply to replace the two instances of > > "fence" in the original text with "cumul-fence". >=20 > Ok that's fine. But I still feel the rfe is not a part of the cumul-fence. > The fences have nothing to do with the rfe. Or, I am missing something qu= ite > badly. >=20 > Boqun, did you understand what Alan is saying? >=20 I think 'cumul-fence' that Alan mentioned is not a fence, but a relation, which could be the result of combining a rfe relation and a A-cumulative fence relation. Please see the section "PROPAGATION ORDER RELATION: cumul-fence" or the definition of cumul-fence in linux-kernel.cat. Did I get you right, Alan? If so, your suggestion is indeed a better change. Regards, Boqun > thanks, >=20 > - Joel >=20 --E39vaYmALEf/7YXx Content-Type: application/pgp-signature; name="signature.asc" -----BEGIN PGP SIGNATURE----- iQEzBAABCAAdFiEEj5IosQTPz8XU1wRHSXnow7UH+rgFAl09vwIACgkQSXnow7UH +rhTkQf+P27OfVcqDn7RAZlp6KBxXcTwxUccGXSnJcXP3XYvPCkOSkFyBLkZeZbF LM/uL1Ec0Cc6xSIl4Bsqh+y6C11SV5HibIaLRrzuuC2A9Hfzm0eU+6BhzX9ppLfb 0YVlO0aiwUhN4Dp3LLeJVCEkMkICgvW4sJtRNqGTf8f9SyjLrmQWClxui+PifO8O ZHj5Lo03at/Qz5MCDTRAqfpfXY0h0/fI0tkLVlTn7115hUSessr5VQFbZCSjAe6L Nmp8tNbqSgEGWiLwzaju7EOIGk3ffaHfPihl9V5PTg8mtXf5Bp9Ml+7WDnAr4uKO Qj9ctOJd8qAjbRBoBeHLhhQtBvoblw== =LDZp -----END PGP SIGNATURE----- --E39vaYmALEf/7YXx-- From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail-qt1-f196.google.com ([209.85.160.196]:36753 "EHLO mail-qt1-f196.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1726046AbfG1P2S (ORCPT ); Sun, 28 Jul 2019 11:28:18 -0400 Date: Sun, 28 Jul 2019 23:28:06 +0800 From: Boqun Feng Subject: Re: [PATCH v2] lkmm/docs: Correct ->prop example with additional rfe link Message-ID: <20190728152806.GB26905@tardis> References: <20190728031303.164545-1-joel@joelfernandes.org> <20190728151959.GA82871@google.com> MIME-Version: 1.0 Content-Type: multipart/signed; micalg=pgp-sha256; protocol="application/pgp-signature"; boundary="E39vaYmALEf/7YXx" Content-Disposition: inline In-Reply-To: <20190728151959.GA82871@google.com> Sender: linux-arch-owner@vger.kernel.org List-ID: To: Joel Fernandes Cc: Alan Stern , linux-kernel@vger.kernel.org, kernel-team@android.com, Akira Yokosawa , Andrea Parri , Daniel Lustig , David Howells , Ingo Molnar , Jade Alglave , linux-arch@vger.kernel.org, Luc Maranget , Nicholas Piggin , "Paul E. McKenney" , Peter Zijlstra , Will Deacon Message-ID: <20190728152806.W_CqwKlDBdbZ1RYasCfvhwrbaxEn1C_lDdfzz79pAfY@z> --E39vaYmALEf/7YXx Content-Type: text/plain; charset=us-ascii Content-Disposition: inline Content-Transfer-Encoding: quoted-printable On Sun, Jul 28, 2019 at 11:19:59AM -0400, Joel Fernandes wrote: > On Sun, Jul 28, 2019 at 10:48:51AM -0400, Alan Stern wrote: > > On Sat, 27 Jul 2019, Joel Fernandes (Google) wrote: > >=20 > > > The lkmm example about ->prop relation should describe an additional = rfe > > > link between P1's store to y and P2's load of y, which should be > > > critical to establishing the ordering resulting in the ->prop ordering > > > on P0. IOW, there are 2 rfe links, not one. > > >=20 > > > Correct these in the docs to make the ->prop ordering on P0 more clea= r. > > >=20 > > > Cc: kernel-team@android.com > > > Reviewed-by: Boqun Feng > > > Signed-off-by: Joel Fernandes (Google) > > > --- > >=20 > > This is not a good update. See below... >=20 > No problem, thanks for the feedback. I am new to the LKMM so please bear > with me.. I should have tagged this RFC. >=20 > > > .../memory-model/Documentation/explanation.txt | 17 ++++++++++-----= -- > > > 1 file changed, 10 insertions(+), 7 deletions(-) > > >=20 > > > diff --git a/tools/memory-model/Documentation/explanation.txt b/tools= /memory-model/Documentation/explanation.txt > > > index 68caa9a976d0..aa84fce854cc 100644 > > > --- a/tools/memory-model/Documentation/explanation.txt > > > +++ b/tools/memory-model/Documentation/explanation.txt > > > @@ -1302,8 +1302,8 @@ followed by an arbitrary number of cumul-fence = links, ending with an > > > rfe link. You can concoct more exotic examples, containing more than > > > one fence, although this quickly leads to diminishing returns in ter= ms > > > of complexity. For instance, here's an example containing a coe link > > > -followed by two fences and an rfe link, utilizing the fact that > > > -release fences are A-cumulative: > > > +followed by a fence, an rfe link, another fence and and a final rfe = link, > > ^---^ > > > +utilizing the fact that release fences are A-cumulative: > >=20 > > I don't like this, for two reasons. First is the repeated "and" typo. >=20 > Will fix the trivial typo, sorry about that. >=20 > > More importantly, it's not necessary to go into this level of detail; a > > better revision would be: > >=20 > > +followed by two cumul-fences and an rfe link, utilizing the fact that > >=20 > > This is appropriate because the cumul-fence relation is defined to=20 > > contain the rfe link which you noticed wasn't mentioned explicitly. >=20 > No, I am talking about the P1's store to Y and P2's load of Y. That is not > through a cumul-fence so I don't understand what you meant? That _is_ mis= sing > the rfe link I am referring to, that is left out. >=20 > The example says r2 =3D 1 and then we work backwards from that. r2 could = very > well have been 0, there's no fence or anything involved, it just happens = to > be the executation pattern causing r2 =3D 1 and hence the rfe link. Right? >=20 > > > int x, y, z; > > > =20 > > > @@ -1334,11 +1334,14 @@ If x =3D 2, r0 =3D 1, and r2 =3D 1 after this= code runs then there is a prop > > > link from P0's store to its load. This is because P0's store gets > > > overwritten by P1's store since x =3D 2 at the end (a coe link), the > > > smp_wmb() ensures that P1's store to x propagates to P2 before the > > > -store to y does (the first fence), the store to y propagates to P2 > > > -before P2's load and store execute, P2's smp_store_release() > > > -guarantees that the stores to x and y both propagate to P0 before the > > > -store to z does (the second fence), and P0's load executes after the > > > -store to z has propagated to P0 (an rfe link). > > > +store to y does (the first fence), P2's store to y happens before P2= 's > > ---------------------------------------^ > >=20 > > This makes no sense, since P2 doesn't store to y. You meant P1's store > > to y. Also, the use of "happens before" is here unnecessarily > > ambiguous (is it an informal usage or does it refer to the formal > > happens-before relation?). The original "propagates to" is better. >=20 > Will reword this. >=20 > > > +load of y (rfe link), P2's smp_store_release() ensures that P2's load > > > +of y executes before P2's store to z (second fence), which implies t= hat > > > +that stores to x and y propagate to P2 before the smp_store_release(= ), which > > > +means that P2's smp_store_release() will propagate stores to x and y= to all > > > +CPUs before the store to z propagates (A-cumulative property of this= fence). > > > +Finally P0's load of z executes after P2's store to z has propagated= to > > > +P0 (rfe link). > >=20 > > Again, a better change would be simply to replace the two instances of > > "fence" in the original text with "cumul-fence". >=20 > Ok that's fine. But I still feel the rfe is not a part of the cumul-fence. > The fences have nothing to do with the rfe. Or, I am missing something qu= ite > badly. >=20 > Boqun, did you understand what Alan is saying? >=20 I think 'cumul-fence' that Alan mentioned is not a fence, but a relation, which could be the result of combining a rfe relation and a A-cumulative fence relation. Please see the section "PROPAGATION ORDER RELATION: cumul-fence" or the definition of cumul-fence in linux-kernel.cat. Did I get you right, Alan? If so, your suggestion is indeed a better change. Regards, Boqun > thanks, >=20 > - Joel >=20 --E39vaYmALEf/7YXx Content-Type: application/pgp-signature; name="signature.asc" -----BEGIN PGP SIGNATURE----- iQEzBAABCAAdFiEEj5IosQTPz8XU1wRHSXnow7UH+rgFAl09vwIACgkQSXnow7UH +rhTkQf+P27OfVcqDn7RAZlp6KBxXcTwxUccGXSnJcXP3XYvPCkOSkFyBLkZeZbF LM/uL1Ec0Cc6xSIl4Bsqh+y6C11SV5HibIaLRrzuuC2A9Hfzm0eU+6BhzX9ppLfb 0YVlO0aiwUhN4Dp3LLeJVCEkMkICgvW4sJtRNqGTf8f9SyjLrmQWClxui+PifO8O ZHj5Lo03at/Qz5MCDTRAqfpfXY0h0/fI0tkLVlTn7115hUSessr5VQFbZCSjAe6L Nmp8tNbqSgEGWiLwzaju7EOIGk3ffaHfPihl9V5PTg8mtXf5Bp9Ml+7WDnAr4uKO Qj9ctOJd8qAjbRBoBeHLhhQtBvoblw== =LDZp -----END PGP SIGNATURE----- --E39vaYmALEf/7YXx--