| Message ID | c447f9faf0283bc6b83bbfbf05acd7acca00762d.1727690180.git.federico.serafini@bugseng.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | xen: address violations of MISRA C Rule 13.6 | expand |
On 30.09.2024 14:49, Federico Serafini wrote: > guest_handle_ok()'s expansion contains a sizeof() involving its > first argument which is guest_handle_cast(). > The expansion of the latter, in turn, contains a variable > initialization. > > Since MISRA considers the initialization (even of a local variable) > a side effect, the chain of expansions mentioned above violates > MISRA C:2012 Rule 13.6 (The operand of the `sizeof' operator shall not > contain any expression which has potential side effect). I'm afraid I need to ask for clarification of terminology and alike here. While the Misra doc has a section on Persistent Side Effects in its Glossary appendix, what constitutes a side effect from its pov isn't really spelled out anywhere. Which in turn raises the question whether it is indeed Misra (and not just Eclair) which deems initialization a side effect. This is even more so relevant as 13.6 talks of only expressions, yet initializers fall under declarations (in turn involving an expression on the rhs of the equal sign). All the same of course affects patch 2 then, too. Jan
On 2024-09-30 15:07, Jan Beulich wrote: > On 30.09.2024 14:49, Federico Serafini wrote: >> guest_handle_ok()'s expansion contains a sizeof() involving its >> first argument which is guest_handle_cast(). >> The expansion of the latter, in turn, contains a variable >> initialization. >> >> Since MISRA considers the initialization (even of a local variable) >> a side effect, the chain of expansions mentioned above violates >> MISRA C:2012 Rule 13.6 (The operand of the `sizeof' operator shall not >> contain any expression which has potential side effect). > > I'm afraid I need to ask for clarification of terminology and alike here. > While the Misra doc has a section on Persistent Side Effects in its > Glossary appendix, what constitutes a side effect from its pov isn't > really spelled out anywhere. Which in turn raises the question whether it > is indeed Misra (and not just Eclair) which deems initialization a side > effect. This is even more so relevant as 13.6 talks of only expressions, > yet initializers fall under declarations (in turn involving an expression > on the rhs of the equal sign). > > All the same of course affects patch 2 then, too. MISRA C leaves the definition of "side effect" to the C Standard. E.g., C18 5.1.2.3p2: Accessing a volatile object, modifying an object, modifying a file, or calling a function that does any of those operations are all side effects,[omitted irrelevant footnote reference] which are changes in the state of the execution environment. The MISRA C:2012/2023 Glossary entry for "Persistent side effect" indirectly confirms that initialization is always a side effect. Kind regards, Roberto
On 01.10.2024 07:25, Roberto Bagnara wrote: > On 2024-09-30 15:07, Jan Beulich wrote: >> On 30.09.2024 14:49, Federico Serafini wrote: >>> guest_handle_ok()'s expansion contains a sizeof() involving its >>> first argument which is guest_handle_cast(). >>> The expansion of the latter, in turn, contains a variable >>> initialization. >>> >>> Since MISRA considers the initialization (even of a local variable) >>> a side effect, the chain of expansions mentioned above violates >>> MISRA C:2012 Rule 13.6 (The operand of the `sizeof' operator shall not >>> contain any expression which has potential side effect). >> >> I'm afraid I need to ask for clarification of terminology and alike here. >> While the Misra doc has a section on Persistent Side Effects in its >> Glossary appendix, what constitutes a side effect from its pov isn't >> really spelled out anywhere. Which in turn raises the question whether it >> is indeed Misra (and not just Eclair) which deems initialization a side >> effect. This is even more so relevant as 13.6 talks of only expressions, >> yet initializers fall under declarations (in turn involving an expression >> on the rhs of the equal sign). >> >> All the same of course affects patch 2 then, too. > > MISRA C leaves the definition of "side effect" to the C Standard. > E.g., C18 5.1.2.3p2: > > Accessing a volatile object, modifying an object, modifying a file, > or calling a function that does any of those operations are all > side effects,[omitted irrelevant footnote reference] which are > changes in the state of the execution environment. > > The MISRA C:2012/2023 Glossary entry for "Persistent side effect" > indirectly confirms that initialization is always a side effect. Hmm, that's interesting: There's indeed an example with an initializer there. Yet to me the text you quote from the C standard does not say that initialization is a side effect - it would be "modifying an object" aiui, yet ahead of initialization being complete the object doesn't "exist" imo, and hence can be "modified" only afterwards. Jan
On Tue, 1 Oct 2024, Jan Beulich wrote: > On 01.10.2024 07:25, Roberto Bagnara wrote: > > On 2024-09-30 15:07, Jan Beulich wrote: > >> On 30.09.2024 14:49, Federico Serafini wrote: > >>> guest_handle_ok()'s expansion contains a sizeof() involving its > >>> first argument which is guest_handle_cast(). > >>> The expansion of the latter, in turn, contains a variable > >>> initialization. > >>> > >>> Since MISRA considers the initialization (even of a local variable) > >>> a side effect, the chain of expansions mentioned above violates > >>> MISRA C:2012 Rule 13.6 (The operand of the `sizeof' operator shall not > >>> contain any expression which has potential side effect). > >> > >> I'm afraid I need to ask for clarification of terminology and alike here. > >> While the Misra doc has a section on Persistent Side Effects in its > >> Glossary appendix, what constitutes a side effect from its pov isn't > >> really spelled out anywhere. Which in turn raises the question whether it > >> is indeed Misra (and not just Eclair) which deems initialization a side > >> effect. This is even more so relevant as 13.6 talks of only expressions, > >> yet initializers fall under declarations (in turn involving an expression > >> on the rhs of the equal sign). > >> > >> All the same of course affects patch 2 then, too. > > > > MISRA C leaves the definition of "side effect" to the C Standard. > > E.g., C18 5.1.2.3p2: > > > > Accessing a volatile object, modifying an object, modifying a file, > > or calling a function that does any of those operations are all > > side effects,[omitted irrelevant footnote reference] which are > > changes in the state of the execution environment. > > > > The MISRA C:2012/2023 Glossary entry for "Persistent side effect" > > indirectly confirms that initialization is always a side effect. > > Hmm, that's interesting: There's indeed an example with an initializer > there. Yet to me the text you quote from the C standard does not say > that initialization is a side effect - it would be "modifying an > object" aiui, yet ahead of initialization being complete the object > doesn't "exist" imo, and hence can be "modified" only afterwards. Hi Jan, I feel it's becoming a bit too philosophical. Since there's some room for interpretation and only two violations left to address, I believe it's best to stick with the stricter interpretation of the definition. Therefore, I'd proceed with this series in its current form.
On 01.10.2024 23:36, Stefano Stabellini wrote: > On Tue, 1 Oct 2024, Jan Beulich wrote: >> On 01.10.2024 07:25, Roberto Bagnara wrote: >>> On 2024-09-30 15:07, Jan Beulich wrote: >>>> On 30.09.2024 14:49, Federico Serafini wrote: >>>>> guest_handle_ok()'s expansion contains a sizeof() involving its >>>>> first argument which is guest_handle_cast(). >>>>> The expansion of the latter, in turn, contains a variable >>>>> initialization. >>>>> >>>>> Since MISRA considers the initialization (even of a local variable) >>>>> a side effect, the chain of expansions mentioned above violates >>>>> MISRA C:2012 Rule 13.6 (The operand of the `sizeof' operator shall not >>>>> contain any expression which has potential side effect). >>>> >>>> I'm afraid I need to ask for clarification of terminology and alike here. >>>> While the Misra doc has a section on Persistent Side Effects in its >>>> Glossary appendix, what constitutes a side effect from its pov isn't >>>> really spelled out anywhere. Which in turn raises the question whether it >>>> is indeed Misra (and not just Eclair) which deems initialization a side >>>> effect. This is even more so relevant as 13.6 talks of only expressions, >>>> yet initializers fall under declarations (in turn involving an expression >>>> on the rhs of the equal sign). >>>> >>>> All the same of course affects patch 2 then, too. >>> >>> MISRA C leaves the definition of "side effect" to the C Standard. >>> E.g., C18 5.1.2.3p2: >>> >>> Accessing a volatile object, modifying an object, modifying a file, >>> or calling a function that does any of those operations are all >>> side effects,[omitted irrelevant footnote reference] which are >>> changes in the state of the execution environment. >>> >>> The MISRA C:2012/2023 Glossary entry for "Persistent side effect" >>> indirectly confirms that initialization is always a side effect. >> >> Hmm, that's interesting: There's indeed an example with an initializer >> there. Yet to me the text you quote from the C standard does not say >> that initialization is a side effect - it would be "modifying an >> object" aiui, yet ahead of initialization being complete the object >> doesn't "exist" imo, and hence can be "modified" only afterwards. > > I feel it's becoming a bit too philosophical. Since there's some room > for interpretation and only two violations left to address, I believe > it's best to stick with the stricter interpretation of the definition. > Therefore, I'd proceed with this series in its current form. Proceeding with the series in its current form may be okay (as you say, you view the changes as readability improvements anyway), but imo the interpretation needs settling on no matter what. In fact even for these two patches it may affect what their descriptions ought to say (would be nice imo to avoid permanently recording potentially misleading information by committing as is). And of course clarity would help dealing with future instances that might appear. I take it you realize that if someone had submitted a patch adding code similar to the original forms of what's being altered here, it would be relatively unlikely for a reviewer to spot the issue. IOW here we're making ourselves heavily dependent upon Eclair spotting (supposed) issues, adding extra work and delays for such changes to go in. Jan
On 2024-10-02 08:09, Jan Beulich wrote: > On 01.10.2024 23:36, Stefano Stabellini wrote: >> On Tue, 1 Oct 2024, Jan Beulich wrote: >>> On 01.10.2024 07:25, Roberto Bagnara wrote: >>>> On 2024-09-30 15:07, Jan Beulich wrote: >>>>> On 30.09.2024 14:49, Federico Serafini wrote: >>>>>> guest_handle_ok()'s expansion contains a sizeof() involving its >>>>>> first argument which is guest_handle_cast(). >>>>>> The expansion of the latter, in turn, contains a variable >>>>>> initialization. >>>>>> >>>>>> Since MISRA considers the initialization (even of a local variable) >>>>>> a side effect, the chain of expansions mentioned above violates >>>>>> MISRA C:2012 Rule 13.6 (The operand of the `sizeof' operator shall not >>>>>> contain any expression which has potential side effect). >>>>> >>>>> I'm afraid I need to ask for clarification of terminology and alike here. >>>>> While the Misra doc has a section on Persistent Side Effects in its >>>>> Glossary appendix, what constitutes a side effect from its pov isn't >>>>> really spelled out anywhere. Which in turn raises the question whether it >>>>> is indeed Misra (and not just Eclair) which deems initialization a side >>>>> effect. This is even more so relevant as 13.6 talks of only expressions, >>>>> yet initializers fall under declarations (in turn involving an expression >>>>> on the rhs of the equal sign). >>>>> >>>>> All the same of course affects patch 2 then, too. >>>> >>>> MISRA C leaves the definition of "side effect" to the C Standard. >>>> E.g., C18 5.1.2.3p2: >>>> >>>> Accessing a volatile object, modifying an object, modifying a file, >>>> or calling a function that does any of those operations are all >>>> side effects,[omitted irrelevant footnote reference] which are >>>> changes in the state of the execution environment. >>>> >>>> The MISRA C:2012/2023 Glossary entry for "Persistent side effect" >>>> indirectly confirms that initialization is always a side effect. >>> >>> Hmm, that's interesting: There's indeed an example with an initializer >>> there. Yet to me the text you quote from the C standard does not say >>> that initialization is a side effect - it would be "modifying an >>> object" aiui, yet ahead of initialization being complete the object >>> doesn't "exist" imo, and hence can be "modified" only afterwards. >> >> I feel it's becoming a bit too philosophical. Since there's some room >> for interpretation and only two violations left to address, I believe >> it's best to stick with the stricter interpretation of the definition. >> Therefore, I'd proceed with this series in its current form. > > Proceeding with the series in its current form may be okay (as you say, > you view the changes as readability improvements anyway), but imo the > interpretation needs settling on no matter what. In fact even for these > two patches it may affect what their descriptions ought to say (would > be nice imo to avoid permanently recording potentially misleading > information by committing as is). And of course clarity would help > dealing with future instances that might appear. I take it you realize > that if someone had submitted a patch adding code similar to the > original forms of what's being altered here, it would be relatively > unlikely for a reviewer to spot the issue. IOW here we're making > ourselves heavily dependent upon Eclair spotting (supposed) issues, > adding extra work and delays for such changes to go in. You can do two things to obtain a second opinion: 1) Use the MISRA forum (here is the link to the forum section devoted to the side-effect rules of MISRA C:2012 and MISRA C:2023 (https://forum.misra.org.uk/forumdisplay.php?fid=168). The MISRA C Working Group will, in due course, provide you with an official answer to your questions about what, for the interpretation of Rule 13.6, has to be considered a side effect. 2) Reach out to your ISO National Body and try to obtain an official answer from ISO/IEC JTC1/SC22/WG14 (the international standardization working group for the programming language C) to your questions about what the C Standard considers to be side effects. Kind regards, Roberto Roberto Bagnara, Ph.D. Software Verification Expert and Evangelist, BUGSENG (http://bugseng.com) Professor of Computer Science, University of Parma Member, ISO/IEC JTC1/SC22/WG14 - C Standardization Working Group Member, MISRA C Working Group
On Mon, 30 Sep 2024, Federico Serafini wrote: > guest_handle_ok()'s expansion contains a sizeof() involving its > first argument which is guest_handle_cast(). > The expansion of the latter, in turn, contains a variable > initialization. > > Since MISRA considers the initialization (even of a local variable) > a side effect, the chain of expansions mentioned above violates > MISRA C:2012 Rule 13.6 (The operand of the `sizeof' operator shall not > contain any expression which has potential side effect). > > Refactor the code to address the rule violation. > > Suggested-by: Andrew Cooper <andrew.cooper3@citrix.com> > Signed-off-by: Federico Serafini <federico.serafini@bugseng.com> Reviewed-by: Stefano Stabellini <sstabellini@kernel.org>
On Wed, Oct 02, 2024 at 05:36:47PM -0700, Stefano Stabellini wrote: > On Mon, 30 Sep 2024, Federico Serafini wrote: > > guest_handle_ok()'s expansion contains a sizeof() involving its > > first argument which is guest_handle_cast(). > > The expansion of the latter, in turn, contains a variable > > initialization. > > > > Since MISRA considers the initialization (even of a local variable) > > a side effect, the chain of expansions mentioned above violates > > MISRA C:2012 Rule 13.6 (The operand of the `sizeof' operator shall not > > contain any expression which has potential side effect). > > > > Refactor the code to address the rule violation. > > > > Suggested-by: Andrew Cooper <andrew.cooper3@citrix.com> > > Signed-off-by: Federico Serafini <federico.serafini@bugseng.com> > > Reviewed-by: Stefano Stabellini <sstabellini@kernel.org> Acked-by: Marek Marczykowski-Górecki <marmarek@invisiblethingslab.com>
On 02.10.2024 08:54, Roberto Bagnara wrote: > On 2024-10-02 08:09, Jan Beulich wrote: >> On 01.10.2024 23:36, Stefano Stabellini wrote: >>> On Tue, 1 Oct 2024, Jan Beulich wrote: >>>> On 01.10.2024 07:25, Roberto Bagnara wrote: >>>>> On 2024-09-30 15:07, Jan Beulich wrote: >>>>>> On 30.09.2024 14:49, Federico Serafini wrote: >>>>>>> guest_handle_ok()'s expansion contains a sizeof() involving its >>>>>>> first argument which is guest_handle_cast(). >>>>>>> The expansion of the latter, in turn, contains a variable >>>>>>> initialization. >>>>>>> >>>>>>> Since MISRA considers the initialization (even of a local variable) >>>>>>> a side effect, the chain of expansions mentioned above violates >>>>>>> MISRA C:2012 Rule 13.6 (The operand of the `sizeof' operator shall not >>>>>>> contain any expression which has potential side effect). >>>>>> >>>>>> I'm afraid I need to ask for clarification of terminology and alike here. >>>>>> While the Misra doc has a section on Persistent Side Effects in its >>>>>> Glossary appendix, what constitutes a side effect from its pov isn't >>>>>> really spelled out anywhere. Which in turn raises the question whether it >>>>>> is indeed Misra (and not just Eclair) which deems initialization a side >>>>>> effect. This is even more so relevant as 13.6 talks of only expressions, >>>>>> yet initializers fall under declarations (in turn involving an expression >>>>>> on the rhs of the equal sign). >>>>>> >>>>>> All the same of course affects patch 2 then, too. >>>>> >>>>> MISRA C leaves the definition of "side effect" to the C Standard. >>>>> E.g., C18 5.1.2.3p2: >>>>> >>>>> Accessing a volatile object, modifying an object, modifying a file, >>>>> or calling a function that does any of those operations are all >>>>> side effects,[omitted irrelevant footnote reference] which are >>>>> changes in the state of the execution environment. >>>>> >>>>> The MISRA C:2012/2023 Glossary entry for "Persistent side effect" >>>>> indirectly confirms that initialization is always a side effect. >>>> >>>> Hmm, that's interesting: There's indeed an example with an initializer >>>> there. Yet to me the text you quote from the C standard does not say >>>> that initialization is a side effect - it would be "modifying an >>>> object" aiui, yet ahead of initialization being complete the object >>>> doesn't "exist" imo, and hence can be "modified" only afterwards. >>> >>> I feel it's becoming a bit too philosophical. Since there's some room >>> for interpretation and only two violations left to address, I believe >>> it's best to stick with the stricter interpretation of the definition. >>> Therefore, I'd proceed with this series in its current form. >> >> Proceeding with the series in its current form may be okay (as you say, >> you view the changes as readability improvements anyway), but imo the >> interpretation needs settling on no matter what. In fact even for these >> two patches it may affect what their descriptions ought to say (would >> be nice imo to avoid permanently recording potentially misleading >> information by committing as is). And of course clarity would help >> dealing with future instances that might appear. I take it you realize >> that if someone had submitted a patch adding code similar to the >> original forms of what's being altered here, it would be relatively >> unlikely for a reviewer to spot the issue. IOW here we're making >> ourselves heavily dependent upon Eclair spotting (supposed) issues, >> adding extra work and delays for such changes to go in. > > You can do two things to obtain a second opinion: > > 1) Use the MISRA forum (here is the link to the forum > section devoted to the side-effect rules of MISRA C:2012 > and MISRA C:2023 (https://forum.misra.org.uk/forumdisplay.php?fid=168). > The MISRA C Working Group will, in due course, provide > you with an official answer to your questions about what, > for the interpretation of Rule 13.6, has to be considered > a side effect. > > 2) Reach out to your ISO National Body and try to obtain > an official answer from ISO/IEC JTC1/SC22/WG14 (the > international standardization working group for the > programming language C) to your questions about what the > C Standard considers to be side effects. I took the latter route, and to my (positive) surprise I got back an answer the same day. There was a request for someone to confirm, but so far I didn't see further replies. Since this is a German institution I raised the question in German and got back an answer in German (attached); I've tried my best to translate this without falsifying anything, but I've omitted non-technical parts: "Initialization of an object is never a side effect of the initialization by itself. Of course expressions used for initialization can themselves have side effects on other objects. @Andreas: Do you agree? C after all has a far simpler object model than C++. The (potential) change in object representation (i.e. the bytes underlying the object) is not a side effect of object initialization, but its primary purpose." Further for Misra she added a reference to a Swiss person, but I think with Bugseng we have sufficient expertise there. Jan Hallo zusammen, bitte nicht wundern, aber sobald ich "Ingenieur" rieche, bin ich sofort beim "Du" (bin ja selber eine). Mein Hauptfokus liegt zwar auf C++, und C ist für mich ein "Seiteneffekt", aber diese Frage kann ich vermutlich ebenfalls beantworten: Die Initialisierung eines Objekts ist niemals ein Seiteneffekt der Initialisierung von sich selbst. In den Expressions, welche zur Initialisierung evaluiert werden, können jedoch sehr wohl Seiteneffekte in anderen Objekten auftreten. @Andreas: kannst du dem zustimmen? C hat ja ein sehr viel einfacheres Objektmodell als C++. Die (potentielle) Änderung der Objektrepräsentation (d.h. die Bytes, welche das Objekt hinterlegen) ist ja kein Seiteneffekt der Objektinitialisierung, sondern sein primärer Zweck. Wenn es um MISRA geht, wäre sicherlich P. Sommerlad der richtige Ansprechpartner (https://sommerlad.ch/). Ansonsten tagt das WG14 Komittee gerade, große Teile der WG14 Tafelrunde sind versammelt, Robert Seacord ist sein Chair. lG Daniela > Wellhöfer, Johannes <johannes.wellhoefer@din.de> hat am 02.10.2024 15:02 CEST geschrieben: > > > Danke Manuela, fürs weiterleiten. > > Sehr geehrter Herr Beulich, > > vielen Dank für Ihre Anfrage. Ich habe Frau Engert hier mit in den Emailverlauf genommen. Sie arbeitet als deutsche Expertin in der JTC1/SC22/WG14 mit. > > @Daniela Engert können Sie Herrn Beulich mit seiner Frage (unten im Verlauf gelb markiert) weiterhelfen? > > Mit freundlichen Grüßen > > Johannes Wellhöfer > Er/Sein > Projektmanager > DIN-Normenausschuss Informationstechnik und Anwendungen (NIA) > DIN Deutsches Institut für Normung e. V. – Industrie und Informationstechnik (DIN – NuS IIT IuA) > > LinkedIn (https://www.linkedin.com/in/johannes-wellhoefer/) | T +49 30 2601-2455 | F +49 30 2601-4-2455 | M +49 174 2007820 > LinkedIn-Kanal des Normenausschusses für Informationstechnik und Anwendungen (https://www.linkedin.com/showcase/din-nia/) > > > > > Ab sofort kostenlos: Der Normungs-Monitor (https://www.din.de/de/din-und-seine-partner/presse/mitteilungen/ab-januar-kostenlos-der-normungs-monitor-983794) > Unkompliziert zu Normungsprojekten Ihrer Branche auf dem Laufenden bleiben > > > Folgen Sie uns auf https://twitter.com/DIN_Norm https://www.linkedin.com/company/din-ev https://www.youtube.com/user/DINBerlin > > DIN Deutsches Institut für Normung e. V., Am DIN-Platz, Burggrafenstraße 6, 10787 Berlin; www.din.de (http://www.din.de/)} Registergericht: AG Berlin-Charlottenburg, VR 288 B; Präsident: Ulrich B. Stoll; Vorstand: Christoph Winterhalter (Vorsitzender), Daniel Schmidt; > Der Inhalt dieser E-Mail (einschließlich Anhängen) ist vertraulich. Falls Sie diese E-Mail versehentlich erhalten haben, löschen Sie sie bitte und informieren den Absender. The contents of this e-mail (including attachments) are confidential. If you received this e-mail in error, please delete it and notify the sender. > -----Ursprüngliche Nachricht----- > Von: Tillack-Lübke, Manuela <Manuela.Tillack-luebke@din.de> > Gesendet: Mittwoch, 2. Oktober 2024 12:09 > An: Wellhöfer, Johannes <Johannes.Wellhoefer@din.de> > Betreff: WG: Klarstellung zu ISO/IEC JTC1/SC22/WG14 > > Lieber Johannes, > > zuständigkeitshalber bitte ich Dich um Beantwortung der unten angefügten Anfrage. Vielen Dank. > > > Mit besten Grüßen > > Manuela > AP A. 634 > - 2419 > > > > -----Ursprüngliche Nachricht----- > Von: info@din.de <info@din.de> > Gesendet: Mittwoch, 2. Oktober 2024 10:17 > An: DIN Info <DIN-Info@din.de> > Betreff: Klarstellung zu ISO/IEC JTC1/SC22/WG14 > > Guten Tag, > > folgende Nachricht wurde auf https://www.din.de/de an Sie geschickt. > > Guten Tag, > > im Zusammenhang mit Misra C:2012 ist eine Interpretationsfrage zum C99 Standard (ISO/IEC 9899:TC3) aufgetaucht. Zur Klärung wurde ich an den ISO National Body verwiesen, der - soweit ich feststellen kann - Sie sind. > > Die eigentliche Frage betrifft "side effects": Fällt die Initialisierung eines Objekts auch darunter? > > In der Hoffnung, dass Sie uns bei der Klärung behilflich sein können, möglicherweise auch durch Verweis an eine andere geeignete Stelle, vielen Dank im Voraus, > Jan Beulich > > Herr Jan Beulich > jbeulich@suse.com
On 2024-10-08 07:59, Jan Beulich wrote: > On 02.10.2024 08:54, Roberto Bagnara wrote: >> On 2024-10-02 08:09, Jan Beulich wrote: >>> On 01.10.2024 23:36, Stefano Stabellini wrote: >>>> On Tue, 1 Oct 2024, Jan Beulich wrote: >>>>> On 01.10.2024 07:25, Roberto Bagnara wrote: >>>>>> On 2024-09-30 15:07, Jan Beulich wrote: >>>>>>> On 30.09.2024 14:49, Federico Serafini wrote: >>>>>>>> guest_handle_ok()'s expansion contains a sizeof() involving its >>>>>>>> first argument which is guest_handle_cast(). >>>>>>>> The expansion of the latter, in turn, contains a variable >>>>>>>> initialization. >>>>>>>> >>>>>>>> Since MISRA considers the initialization (even of a local variable) >>>>>>>> a side effect, the chain of expansions mentioned above violates >>>>>>>> MISRA C:2012 Rule 13.6 (The operand of the `sizeof' operator shall not >>>>>>>> contain any expression which has potential side effect). >>>>>>> >>>>>>> I'm afraid I need to ask for clarification of terminology and alike here. >>>>>>> While the Misra doc has a section on Persistent Side Effects in its >>>>>>> Glossary appendix, what constitutes a side effect from its pov isn't >>>>>>> really spelled out anywhere. Which in turn raises the question whether it >>>>>>> is indeed Misra (and not just Eclair) which deems initialization a side >>>>>>> effect. This is even more so relevant as 13.6 talks of only expressions, >>>>>>> yet initializers fall under declarations (in turn involving an expression >>>>>>> on the rhs of the equal sign). >>>>>>> >>>>>>> All the same of course affects patch 2 then, too. >>>>>> >>>>>> MISRA C leaves the definition of "side effect" to the C Standard. >>>>>> E.g., C18 5.1.2.3p2: >>>>>> >>>>>> Accessing a volatile object, modifying an object, modifying a file, >>>>>> or calling a function that does any of those operations are all >>>>>> side effects,[omitted irrelevant footnote reference] which are >>>>>> changes in the state of the execution environment. >>>>>> >>>>>> The MISRA C:2012/2023 Glossary entry for "Persistent side effect" >>>>>> indirectly confirms that initialization is always a side effect. >>>>> >>>>> Hmm, that's interesting: There's indeed an example with an initializer >>>>> there. Yet to me the text you quote from the C standard does not say >>>>> that initialization is a side effect - it would be "modifying an >>>>> object" aiui, yet ahead of initialization being complete the object >>>>> doesn't "exist" imo, and hence can be "modified" only afterwards. >>>> >>>> I feel it's becoming a bit too philosophical. Since there's some room >>>> for interpretation and only two violations left to address, I believe >>>> it's best to stick with the stricter interpretation of the definition. >>>> Therefore, I'd proceed with this series in its current form. >>> >>> Proceeding with the series in its current form may be okay (as you say, >>> you view the changes as readability improvements anyway), but imo the >>> interpretation needs settling on no matter what. In fact even for these >>> two patches it may affect what their descriptions ought to say (would >>> be nice imo to avoid permanently recording potentially misleading >>> information by committing as is). And of course clarity would help >>> dealing with future instances that might appear. I take it you realize >>> that if someone had submitted a patch adding code similar to the >>> original forms of what's being altered here, it would be relatively >>> unlikely for a reviewer to spot the issue. IOW here we're making >>> ourselves heavily dependent upon Eclair spotting (supposed) issues, >>> adding extra work and delays for such changes to go in. >> >> You can do two things to obtain a second opinion: >> >> 1) Use the MISRA forum (here is the link to the forum >> section devoted to the side-effect rules of MISRA C:2012 >> and MISRA C:2023 (https://forum.misra.org.uk/forumdisplay.php?fid=168). >> The MISRA C Working Group will, in due course, provide >> you with an official answer to your questions about what, >> for the interpretation of Rule 13.6, has to be considered >> a side effect. >> >> 2) Reach out to your ISO National Body and try to obtain >> an official answer from ISO/IEC JTC1/SC22/WG14 (the >> international standardization working group for the >> programming language C) to your questions about what the >> C Standard considers to be side effects. > > I took the latter route, and to my (positive) surprise I got back an answer > the same day. There was a request for someone to confirm, but so far I didn't > see further replies. Since this is a German institution I raised the question > in German and got back an answer in German (attached); I've tried my best to > translate this without falsifying anything, but I've omitted non-technical > parts: > > "Initialization of an object is never a side effect of the initialization > by itself. Of course expressions used for initialization can themselves have > side effects on other objects. > > @Andreas: Do you agree? C after all has a far simpler object model than C++. > The (potential) change in object representation (i.e. the bytes underlying > the object) is not a side effect of object initialization, but its primary > purpose." > > Further for Misra she added a reference to a Swiss person, but I think with > Bugseng we have sufficient expertise there. Unfortunately, the (translation of the) answer you received adds confusion to previous confusion: who answered has highlighted the "side" part of the term, which is indeed relevant in computer science, but not for the C standard. To the point that the same argument could be used to deny that ++i has a side effect because the increment is the "primary" effect... Part of the confusion is maybe in the the following paragraph Jan wrote earlier: > Hmm, that's interesting: There's indeed an example with an initializer > there. Yet to me the text you quote from the C standard does not say > that initialization is a side effect - it would be "modifying an > object" aiui, yet ahead of initialization being complete the object > doesn't "exist" imo, and hence can be "modified" only afterwards. In C, it is not true that the object does not exist ahead of initialization. Try the following: extern int f(int* p); int main() { int i = f(&i); } Kind regards, Roberto
On Tue, Oct 08, 2024 at 02:49:52PM +0200, Roberto Bagnara wrote: > On 2024-10-08 07:59, Jan Beulich wrote: > > On 02.10.2024 08:54, Roberto Bagnara wrote: > > > On 2024-10-02 08:09, Jan Beulich wrote: > > > > On 01.10.2024 23:36, Stefano Stabellini wrote: > > > > > On Tue, 1 Oct 2024, Jan Beulich wrote: > > > > > > On 01.10.2024 07:25, Roberto Bagnara wrote: > > > > > > > On 2024-09-30 15:07, Jan Beulich wrote: > > > > > > > > On 30.09.2024 14:49, Federico Serafini wrote: > > > > > > > > > guest_handle_ok()'s expansion contains a sizeof() involving its > > > > > > > > > first argument which is guest_handle_cast(). > > > > > > > > > The expansion of the latter, in turn, contains a variable > > > > > > > > > initialization. > > > > > > > > > > > > > > > > > > Since MISRA considers the initialization (even of a local variable) > > > > > > > > > a side effect, the chain of expansions mentioned above violates > > > > > > > > > MISRA C:2012 Rule 13.6 (The operand of the `sizeof' operator shall not > > > > > > > > > contain any expression which has potential side effect). > > > > > > > > > > > > > > > > I'm afraid I need to ask for clarification of terminology and alike here. > > > > > > > > While the Misra doc has a section on Persistent Side Effects in its > > > > > > > > Glossary appendix, what constitutes a side effect from its pov isn't > > > > > > > > really spelled out anywhere. Which in turn raises the question whether it > > > > > > > > is indeed Misra (and not just Eclair) which deems initialization a side > > > > > > > > effect. This is even more so relevant as 13.6 talks of only expressions, > > > > > > > > yet initializers fall under declarations (in turn involving an expression > > > > > > > > on the rhs of the equal sign). > > > > > > > > > > > > > > > > All the same of course affects patch 2 then, too. > > > > > > > > > > > > > > MISRA C leaves the definition of "side effect" to the C Standard. > > > > > > > E.g., C18 5.1.2.3p2: > > > > > > > > > > > > > > Accessing a volatile object, modifying an object, modifying a file, > > > > > > > or calling a function that does any of those operations are all > > > > > > > side effects,[omitted irrelevant footnote reference] which are > > > > > > > changes in the state of the execution environment. > > > > > > > > > > > > > > The MISRA C:2012/2023 Glossary entry for "Persistent side effect" > > > > > > > indirectly confirms that initialization is always a side effect. > > > > > > > > > > > > Hmm, that's interesting: There's indeed an example with an initializer > > > > > > there. Yet to me the text you quote from the C standard does not say > > > > > > that initialization is a side effect - it would be "modifying an > > > > > > object" aiui, yet ahead of initialization being complete the object > > > > > > doesn't "exist" imo, and hence can be "modified" only afterwards. > > > > > > > > > > I feel it's becoming a bit too philosophical. Since there's some room > > > > > for interpretation and only two violations left to address, I believe > > > > > it's best to stick with the stricter interpretation of the definition. > > > > > Therefore, I'd proceed with this series in its current form. > > > > > > > > Proceeding with the series in its current form may be okay (as you say, > > > > you view the changes as readability improvements anyway), but imo the > > > > interpretation needs settling on no matter what. In fact even for these > > > > two patches it may affect what their descriptions ought to say (would > > > > be nice imo to avoid permanently recording potentially misleading > > > > information by committing as is). And of course clarity would help > > > > dealing with future instances that might appear. I take it you realize > > > > that if someone had submitted a patch adding code similar to the > > > > original forms of what's being altered here, it would be relatively > > > > unlikely for a reviewer to spot the issue. IOW here we're making > > > > ourselves heavily dependent upon Eclair spotting (supposed) issues, > > > > adding extra work and delays for such changes to go in. > > > > > > You can do two things to obtain a second opinion: > > > > > > 1) Use the MISRA forum (here is the link to the forum > > > section devoted to the side-effect rules of MISRA C:2012 > > > and MISRA C:2023 (https://forum.misra.org.uk/forumdisplay.php?fid=168). > > > The MISRA C Working Group will, in due course, provide > > > you with an official answer to your questions about what, > > > for the interpretation of Rule 13.6, has to be considered > > > a side effect. > > > > > > 2) Reach out to your ISO National Body and try to obtain > > > an official answer from ISO/IEC JTC1/SC22/WG14 (the > > > international standardization working group for the > > > programming language C) to your questions about what the > > > C Standard considers to be side effects. > > > > I took the latter route, and to my (positive) surprise I got back an answer > > the same day. There was a request for someone to confirm, but so far I didn't > > see further replies. Since this is a German institution I raised the question > > in German and got back an answer in German (attached); I've tried my best to > > translate this without falsifying anything, but I've omitted non-technical > > parts: > > > > "Initialization of an object is never a side effect of the initialization > > by itself. Of course expressions used for initialization can themselves have > > side effects on other objects. > > > > @Andreas: Do you agree? C after all has a far simpler object model than C++. > > The (potential) change in object representation (i.e. the bytes underlying > > the object) is not a side effect of object initialization, but its primary > > purpose." > > > > Further for Misra she added a reference to a Swiss person, but I think with > > Bugseng we have sufficient expertise there. > > Unfortunately, the (translation of the) answer you received adds > confusion to previous confusion: who answered has highlighted the > "side" part of the term, which is indeed relevant in computer science, > but not for the C standard. To the point that the same argument could > be used to deny that ++i has a side effect because the increment is > the "primary" effect... > > Part of the confusion is maybe in the the following paragraph Jan > wrote earlier: > > > Hmm, that's interesting: There's indeed an example with an initializer > > there. Yet to me the text you quote from the C standard does not say > > that initialization is a side effect - it would be "modifying an > > object" aiui, yet ahead of initialization being complete the object > > doesn't "exist" imo, and hence can be "modified" only afterwards. > > In C, it is not true that the object does not exist ahead of > initialization. Try the following: > > extern int f(int* p); > > int main() { > int i = f(&i); > } This is interesting discussion, but I don't think it needs to block anything. The proposed change doesn't violate any other rule/code style, and I'd argue it's more readable. Taking more strict interpretation in this case doesn't really hurt. I already acked this patch.
On 08.10.2024 14:49, Roberto Bagnara wrote: > On 2024-10-08 07:59, Jan Beulich wrote: >> On 02.10.2024 08:54, Roberto Bagnara wrote: >>> On 2024-10-02 08:09, Jan Beulich wrote: >>>> On 01.10.2024 23:36, Stefano Stabellini wrote: >>>>> On Tue, 1 Oct 2024, Jan Beulich wrote: >>>>>> On 01.10.2024 07:25, Roberto Bagnara wrote: >>>>>>> On 2024-09-30 15:07, Jan Beulich wrote: >>>>>>>> On 30.09.2024 14:49, Federico Serafini wrote: >>>>>>>>> guest_handle_ok()'s expansion contains a sizeof() involving its >>>>>>>>> first argument which is guest_handle_cast(). >>>>>>>>> The expansion of the latter, in turn, contains a variable >>>>>>>>> initialization. >>>>>>>>> >>>>>>>>> Since MISRA considers the initialization (even of a local variable) >>>>>>>>> a side effect, the chain of expansions mentioned above violates >>>>>>>>> MISRA C:2012 Rule 13.6 (The operand of the `sizeof' operator shall not >>>>>>>>> contain any expression which has potential side effect). >>>>>>>> >>>>>>>> I'm afraid I need to ask for clarification of terminology and alike here. >>>>>>>> While the Misra doc has a section on Persistent Side Effects in its >>>>>>>> Glossary appendix, what constitutes a side effect from its pov isn't >>>>>>>> really spelled out anywhere. Which in turn raises the question whether it >>>>>>>> is indeed Misra (and not just Eclair) which deems initialization a side >>>>>>>> effect. This is even more so relevant as 13.6 talks of only expressions, >>>>>>>> yet initializers fall under declarations (in turn involving an expression >>>>>>>> on the rhs of the equal sign). >>>>>>>> >>>>>>>> All the same of course affects patch 2 then, too. >>>>>>> >>>>>>> MISRA C leaves the definition of "side effect" to the C Standard. >>>>>>> E.g., C18 5.1.2.3p2: >>>>>>> >>>>>>> Accessing a volatile object, modifying an object, modifying a file, >>>>>>> or calling a function that does any of those operations are all >>>>>>> side effects,[omitted irrelevant footnote reference] which are >>>>>>> changes in the state of the execution environment. >>>>>>> >>>>>>> The MISRA C:2012/2023 Glossary entry for "Persistent side effect" >>>>>>> indirectly confirms that initialization is always a side effect. >>>>>> >>>>>> Hmm, that's interesting: There's indeed an example with an initializer >>>>>> there. Yet to me the text you quote from the C standard does not say >>>>>> that initialization is a side effect - it would be "modifying an >>>>>> object" aiui, yet ahead of initialization being complete the object >>>>>> doesn't "exist" imo, and hence can be "modified" only afterwards. >>>>> >>>>> I feel it's becoming a bit too philosophical. Since there's some room >>>>> for interpretation and only two violations left to address, I believe >>>>> it's best to stick with the stricter interpretation of the definition. >>>>> Therefore, I'd proceed with this series in its current form. >>>> >>>> Proceeding with the series in its current form may be okay (as you say, >>>> you view the changes as readability improvements anyway), but imo the >>>> interpretation needs settling on no matter what. In fact even for these >>>> two patches it may affect what their descriptions ought to say (would >>>> be nice imo to avoid permanently recording potentially misleading >>>> information by committing as is). And of course clarity would help >>>> dealing with future instances that might appear. I take it you realize >>>> that if someone had submitted a patch adding code similar to the >>>> original forms of what's being altered here, it would be relatively >>>> unlikely for a reviewer to spot the issue. IOW here we're making >>>> ourselves heavily dependent upon Eclair spotting (supposed) issues, >>>> adding extra work and delays for such changes to go in. >>> >>> You can do two things to obtain a second opinion: >>> >>> 1) Use the MISRA forum (here is the link to the forum >>> section devoted to the side-effect rules of MISRA C:2012 >>> and MISRA C:2023 (https://forum.misra.org.uk/forumdisplay.php?fid=168). >>> The MISRA C Working Group will, in due course, provide >>> you with an official answer to your questions about what, >>> for the interpretation of Rule 13.6, has to be considered >>> a side effect. >>> >>> 2) Reach out to your ISO National Body and try to obtain >>> an official answer from ISO/IEC JTC1/SC22/WG14 (the >>> international standardization working group for the >>> programming language C) to your questions about what the >>> C Standard considers to be side effects. >> >> I took the latter route, and to my (positive) surprise I got back an answer >> the same day. There was a request for someone to confirm, but so far I didn't >> see further replies. Since this is a German institution I raised the question >> in German and got back an answer in German (attached); I've tried my best to >> translate this without falsifying anything, but I've omitted non-technical >> parts: >> >> "Initialization of an object is never a side effect of the initialization >> by itself. Of course expressions used for initialization can themselves have >> side effects on other objects. >> >> @Andreas: Do you agree? C after all has a far simpler object model than C++. >> The (potential) change in object representation (i.e. the bytes underlying >> the object) is not a side effect of object initialization, but its primary >> purpose." >> >> Further for Misra she added a reference to a Swiss person, but I think with >> Bugseng we have sufficient expertise there. > > Unfortunately, the (translation of the) answer you received adds > confusion to previous confusion: who answered has highlighted the > "side" part of the term, which is indeed relevant in computer science, > but not for the C standard. I can't see any highlighting in the original reply I received. > To the point that the same argument could > be used to deny that ++i has a side effect because the increment is > the "primary" effect... Well, if it's just "++i;" there's no side effect, just a primary one. In "n = ++i + j--;" there are side effects (the increment and decrement). > Part of the confusion is maybe in the the following paragraph Jan > wrote earlier: > > > Hmm, that's interesting: There's indeed an example with an initializer > > there. Yet to me the text you quote from the C standard does not say > > that initialization is a side effect - it would be "modifying an > > object" aiui, yet ahead of initialization being complete the object > > doesn't "exist" imo, and hence can be "modified" only afterwards. > > In C, it is not true that the object does not exist ahead of > initialization. I quoted "exist" for that reason. Of course the object's lifetime starts with its declaration. It just has indeterminate value at that point. > Try the following: > > extern int f(int* p); > > int main() { > int i = f(&i); > } Which to me falls under "Of course expressions used for initialization can themselves have side effects on other objects." Just that "other" isn't quite right here. Jan
diff --git a/xen/common/efi/runtime.c b/xen/common/efi/runtime.c index d03e5c90ce..acf08dcaa3 100644 --- a/xen/common/efi/runtime.c +++ b/xen/common/efi/runtime.c @@ -250,14 +250,20 @@ int efi_get_info(uint32_t idx, union xenpf_efi_info *info) info->cfg.addr = __pa(efi_ct); info->cfg.nent = efi_num_ct; break; + case XEN_FW_EFI_VENDOR: + { + XEN_GUEST_HANDLE_PARAM(CHAR16) vendor_name = + guest_handle_cast(info->vendor.name, CHAR16); + if ( !efi_fw_vendor ) return -EOPNOTSUPP; + info->vendor.revision = efi_fw_revision; n = info->vendor.bufsz / sizeof(*efi_fw_vendor); - if ( !guest_handle_okay(guest_handle_cast(info->vendor.name, - CHAR16), n) ) + if ( !guest_handle_okay(vendor_name, n) ) return -EFAULT; + for ( i = 0; i < n; ++i ) { if ( __copy_to_guest_offset(info->vendor.name, i, @@ -267,6 +273,8 @@ int efi_get_info(uint32_t idx, union xenpf_efi_info *info) break; } break; + } + case XEN_FW_EFI_MEM_INFO: for ( i = 0; i < efi_memmap_size; i += efi_mdesc_size ) {
guest_handle_ok()'s expansion contains a sizeof() involving its first argument which is guest_handle_cast(). The expansion of the latter, in turn, contains a variable initialization. Since MISRA considers the initialization (even of a local variable) a side effect, the chain of expansions mentioned above violates MISRA C:2012 Rule 13.6 (The operand of the `sizeof' operator shall not contain any expression which has potential side effect). Refactor the code to address the rule violation. Suggested-by: Andrew Cooper <andrew.cooper3@citrix.com> Signed-off-by: Federico Serafini <federico.serafini@bugseng.com> --- Changes in v2: - better description. --- xen/common/efi/runtime.c | 12 ++++++++++-- 1 file changed, 10 insertions(+), 2 deletions(-)