* MISRA C:2012 D4.11 caution on staging @ 2023-10-10 9:31 Nicola Vetrini 2023-10-11 0:01 ` Stefano Stabellini 2023-10-11 0:15 ` Andrew Cooper 0 siblings, 2 replies; 17+ messages in thread From: Nicola Vetrini @ 2023-10-10 9:31 UTC (permalink / raw) To: Xen-devel Cc: consulting, Stefano Stabellini, Andrew Cooper, Jan Beulich, Julien Grall, Bertrand Marquis, roger.pau, George Dunlap, Wei Liu Hi, as you can see from [1], there's a MISRA C guideline, D4.11, that is supposed to be clean (i.e., have no reports), but has a caution on an argument to memcpy (the second argument might be null according to the checker, given a set of assumptions on the control flow). To access the report just click on the second link in the log, which should take you to a webpage with a list of MISRA guidelines. Click on D4.11 and you'll see the full report, which I pasted below for convenience. If the finding is genuine, then some countermeasure needs to be taken against this possible bug, otherwise it needs to be motivated why the field config->handle can't be null at that point. The finding is likely the result of an improvement made to the checker, because the first analysis I can see that spots it happened when rc1 has been tagged, but that commit does not touch the involved files. [1] https://gitlab.com/xen-project/xen/-/jobs/5251222578 caution for MC3R1.D4.11 untagged xen/common/domain.c:758.27-758.40: [59] null pointer passed as 2nd argument to memory copy function xen/common/domain.c:590.10-590.15: [1] assuming 'config' is null xen/common/domain.c:593.10-593.44: [2] assuming the condition is false; taking false branch <preprocessed xen/common/domain.c>:11332.10-11332.50: preprocessed tokens xen/include/xen/types.h:31.23: expanded from macro `NULL' xen/common/domain.c:603.12-603.27: [3] '?' condition is true <preprocessed xen/common/domain.c>:11336.37-11336.52: preprocessed tokens xen/include/xen/compiler.h:21.43: expanded from macro `unlikely' xen/include/xen/lib.h:26.15-26.28: expanded from macro `ASSERT' xen/include/xen/lib.h:26.26: expanded from macro `ASSERT' xen/common/domain.c:603.5-603.65: [4] taking false branch <preprocessed xen/common/domain.c>:11336.10-11336.11: preprocessed tokens xen/include/xen/lib.h:26.10-26.11: expanded from macro `ASSERT' xen/common/domain.c:603.5-603.65: [5] loop condition is false: exiting loop <preprocessed xen/common/domain.c>:11336.5-11336.6: preprocessed tokens xen/include/xen/lib.h:26.5-26.6: expanded from macro `ASSERT' xen/common/domain.c:605.10-605.15: [6] 'config' is null xen/common/domain.c:605.5-605.6: [7] taking false branch xen/common/domain.c:615.10-615.14: [8] 'domid' is not equal to 0 xen/common/domain.c:615.24-615.28: [9] 'domid' is not equal to hardware_domid xen/common/domain.c:615.5-615.6: [10] taking false branch xen/common/domain.c:624.5-624.44: [11] taking false branch <preprocessed xen/common/domain.c>:11356.10-11356.11: preprocessed tokens xen/include/xen/trace.h:86.9-86.10: expanded from macro `TRACE_varD' xen/include/xen/trace.h:94.5-94.22: expanded from macro `TRACE_1D' xen/common/domain.c:624.5-624.44: [12] loop condition is false: exiting loop <preprocessed xen/common/domain.c>:11356.5-11356.6: preprocessed tokens xen/include/xen/trace.h:85.5-85.6: expanded from macro `TRACE_varD' xen/include/xen/trace.h:94.5-94.22: expanded from macro `TRACE_1D' xen/common/domain.c:630.5-630.39: [13] assuming 'prof' is null; taking true branch <preprocessed xen/common/domain.c>:11362.111-11362.115: preprocessed tokens xen/include/xen/spinlock.h:114.14-114.18: expanded from macro `spin_lock_init_prof' xen/common/domain.c:630.5-630.39: [14] execution continues on line 631 <preprocessed xen/common/domain.c>:11363.104-11363.108: preprocessed tokens xen/include/xen/spinlock.h:119.13-119.17: expanded from macro `spin_lock_init_prof' xen/common/domain.c:631.5-631.43: [15] assuming 'prof' is null; taking true branch <preprocessed xen/common/domain.c>:11367.115-11367.119: preprocessed tokens xen/include/xen/spinlock.h:114.14-114.18: expanded from macro `spin_lock_init_prof' xen/common/domain.c:631.5-631.43: [16] execution continues on line 632 <preprocessed xen/common/domain.c>:11368.112-11368.116: preprocessed tokens xen/include/xen/spinlock.h:119.13-119.17: expanded from macro `spin_lock_init_prof' xen/common/domain.c:642.24-642.36: [17] '?' condition is true <preprocessed xen/common/domain.c>:11377.92: preprocessed tokens xen/include/xen/bitmap.h:97.9: expanded from macro `BITMAP_LAST_WORD_MASK' xen/include/xen/nodemask.h:238.29-238.63: expanded from macro `NODE_MASK_LAST_WORD' xen/include/xen/nodemask.h:244.43-244.61: expanded from macro `NODE_MASK_ALL' xen/common/domain.c:662.5-662.6: [18] taking false branch -- Nicola Vetrini, BSc Software Engineer, BUGSENG srl (https://bugseng.com) ^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: MISRA C:2012 D4.11 caution on staging 2023-10-10 9:31 MISRA C:2012 D4.11 caution on staging Nicola Vetrini @ 2023-10-11 0:01 ` Stefano Stabellini 2023-10-13 9:23 ` Julien Grall 2023-10-11 0:15 ` Andrew Cooper 1 sibling, 1 reply; 17+ messages in thread From: Stefano Stabellini @ 2023-10-11 0:01 UTC (permalink / raw) To: Nicola Vetrini, julien Cc: Xen-devel, consulting, Stefano Stabellini, Andrew Cooper, Jan Beulich, Julien Grall, Bertrand Marquis, roger.pau, George Dunlap, Wei Liu Hey Julien, please double-check what I am writing below On Tue, 10 Oct 2023, Nicola Vetrini wrote: > Hi, > > as you can see from [1], there's a MISRA C guideline, D4.11, that is supposed > to be clean > (i.e., have no reports), but has a caution on an argument to memcpy > (the second argument might be null according to the checker, given a set of > assumptions on > the control flow). To access the report just click on the second link in the > log, which should take you to a webpage with a list of > MISRA guidelines. Click on D4.11 and you'll see the full report, which I > pasted below for convenience. > > If the finding is genuine, then some countermeasure needs to be taken against > this > possible bug, otherwise it needs to be motivated why the field config->handle > can't > be null at that point. > The finding is likely the result of an improvement made to the checker, > because the first > analysis I can see that spots it happened when rc1 has been tagged, but that > commit does not > touch the involved files. > > [1] https://gitlab.com/xen-project/xen/-/jobs/5251222578 > > caution for MC3R1.D4.11 untagged > xen/common/domain.c:758.27-758.40: > [59] null pointer passed as 2nd argument to memory copy function This looks like a genuine issue: in domain_create, config->handle could be uninitialized. For example, domain_create can be called from xen/arch/arm/domain_build.c:create_domUs, passing &d_cfg, and I don't see where we initialize d_cfg.handle. This was just by code inspection. Julien, did I miss anything? ^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: MISRA C:2012 D4.11 caution on staging 2023-10-11 0:01 ` Stefano Stabellini @ 2023-10-13 9:23 ` Julien Grall 0 siblings, 0 replies; 17+ messages in thread From: Julien Grall @ 2023-10-13 9:23 UTC (permalink / raw) To: Stefano Stabellini, Nicola Vetrini Cc: Xen-devel, consulting, Andrew Cooper, Jan Beulich, Bertrand Marquis, roger.pau, George Dunlap, Wei Liu On 11/10/2023 01:01, Stefano Stabellini wrote: > Hey Julien, please double-check what I am writing below Hi Stefano, > On Tue, 10 Oct 2023, Nicola Vetrini wrote: >> Hi, >> >> as you can see from [1], there's a MISRA C guideline, D4.11, that is supposed >> to be clean >> (i.e., have no reports), but has a caution on an argument to memcpy >> (the second argument might be null according to the checker, given a set of >> assumptions on >> the control flow). To access the report just click on the second link in the >> log, which should take you to a webpage with a list of >> MISRA guidelines. Click on D4.11 and you'll see the full report, which I >> pasted below for convenience. >> >> If the finding is genuine, then some countermeasure needs to be taken against >> this >> possible bug, otherwise it needs to be motivated why the field config->handle >> can't >> be null at that point. >> The finding is likely the result of an improvement made to the checker, >> because the first >> analysis I can see that spots it happened when rc1 has been tagged, but that >> commit does not >> touch the involved files. >> >> [1] https://gitlab.com/xen-project/xen/-/jobs/5251222578 >> >> caution for MC3R1.D4.11 untagged >> xen/common/domain.c:758.27-758.40: >> [59] null pointer passed as 2nd argument to memory copy function > > > This looks like a genuine issue: in domain_create, config->handle could > be uninitialized. For example, domain_create can be called > from xen/arch/arm/domain_build.c:create_domUs, passing &d_cfg, and I > don't see where we initialize d_cfg.handle. > > This was just by code inspection. Julien, did I miss anything? cfg->handle is an embedded array. It will get automatically initialized to zero as we are using: d_cfg = { ... }; Now, looking through the steps of the analysis [1], there is something odd. The code looks like this: if ( !is_idle_domain(d) ) ... if ( err = arch_domain_create(... ) != 0 ) ... init_status |= ... if ( !is_idle_domain() ) ... One the first if, Eclair assumes that the domain is equal to 32767 (e.g. idle domain): [27] field 'domain_id' is equal to 32767 But then on third if, we have: [30] assuming field 'domain_id' is not equal to 32767; taking true branch So the question is why did Eclair think 'domain_id' changed? Cheers, [1] https://saas.eclairit.com:3787/fs/var/local/eclair/xen-project.ecdf/xen-project/xen/ECLAIR_normal/4.18.0-rc2/ARM64/5251222578/PROJECT.ecd;/sources/xen/common/domain.c.html#{%22select%22:true,%22selection%22:{%22hiddenAreaKinds%22:[],%22hiddenSubareaKinds%22:[],%22show%22:false,%22selector%22:{%22enabled%22:true,%22negated%22:true,%22kind%22:0,%22domain%22:%22kind%22,%22inputs%22:[{%22enabled%22:true,%22text%22:%22caution%22}]}}} Here it detects -- Julien Grall ^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: MISRA C:2012 D4.11 caution on staging 2023-10-10 9:31 MISRA C:2012 D4.11 caution on staging Nicola Vetrini 2023-10-11 0:01 ` Stefano Stabellini @ 2023-10-11 0:15 ` Andrew Cooper 2023-10-11 7:47 ` Nicola Vetrini 1 sibling, 1 reply; 17+ messages in thread From: Andrew Cooper @ 2023-10-11 0:15 UTC (permalink / raw) To: Nicola Vetrini, Xen-devel Cc: consulting, Stefano Stabellini, Jan Beulich, Julien Grall, Bertrand Marquis, roger.pau, George Dunlap, Wei Liu On 10/10/2023 5:31 pm, Nicola Vetrini wrote: > Hi, > > as you can see from [1], there's a MISRA C guideline, D4.11, that is > supposed to be clean > (i.e., have no reports), but has a caution on an argument to memcpy > (the second argument might be null according to the checker, given a > set of assumptions on > the control flow). To access the report just click on the second link > in the log, which should take you to a webpage with a list of > MISRA guidelines. Click on D4.11 and you'll see the full report, which > I pasted below for convenience. > > If the finding is genuine, then some countermeasure needs to be taken > against this > possible bug, otherwise it needs to be motivated why the field > config->handle can't > be null at that point. > The finding is likely the result of an improvement made to the > checker, because the first > analysis I can see that spots it happened when rc1 has been tagged, > but that commit does not > touch the involved files. > > [1] https://gitlab.com/xen-project/xen/-/jobs/5251222578 That's a false positive, but I'm not entirely surprised that the checker struggled to see it. First, ASSERT(is_system_domain(d) ? config == NULL : config != NULL); All system domains (domid >= 0x7ff0, inc IDLE) pass a NULL config. All other domains pass a real config. Next, /* DOMID_{XEN,IO,etc} (other than IDLE) are sufficiently constructed. */ if ( is_system_domain(d) && !is_idle_domain(d) ) return d; So at this point we only have the IDLE domain and real domains. And finally, the complained-about construct is inside an: if ( !is_idle_domain(d) ) ... hence config cannot be NULL, but only because of the way in which is_{system,idle}_domain() interact. ~Andrew ^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: MISRA C:2012 D4.11 caution on staging 2023-10-11 0:15 ` Andrew Cooper @ 2023-10-11 7:47 ` Nicola Vetrini 2023-10-11 7:51 ` Andrew Cooper 0 siblings, 1 reply; 17+ messages in thread From: Nicola Vetrini @ 2023-10-11 7:47 UTC (permalink / raw) To: Andrew Cooper Cc: Xen-devel, consulting, Stefano Stabellini, Jan Beulich, Julien Grall, Bertrand Marquis, roger.pau, George Dunlap, Wei Liu On 11/10/2023 02:15, Andrew Cooper wrote: > On 10/10/2023 5:31 pm, Nicola Vetrini wrote: >> Hi, >> >> as you can see from [1], there's a MISRA C guideline, D4.11, that is >> supposed to be clean >> (i.e., have no reports), but has a caution on an argument to memcpy >> (the second argument might be null according to the checker, given a >> set of assumptions on >> the control flow). To access the report just click on the second link >> in the log, which should take you to a webpage with a list of >> MISRA guidelines. Click on D4.11 and you'll see the full report, which >> I pasted below for convenience. >> >> If the finding is genuine, then some countermeasure needs to be taken >> against this >> possible bug, otherwise it needs to be motivated why the field >> config->handle can't >> be null at that point. >> The finding is likely the result of an improvement made to the >> checker, because the first >> analysis I can see that spots it happened when rc1 has been tagged, >> but that commit does not >> touch the involved files. >> >> [1] https://gitlab.com/xen-project/xen/-/jobs/5251222578 > > That's a false positive, but I'm not entirely surprised that the > checker > struggled to see it. > > First, > > ASSERT(is_system_domain(d) ? config == NULL : config != NULL); > > All system domains (domid >= 0x7ff0, inc IDLE) pass a NULL config. All > other domains pass a real config. > > Next, > > /* DOMID_{XEN,IO,etc} (other than IDLE) are sufficiently constructed. > */ > if ( is_system_domain(d) && !is_idle_domain(d) ) > return d; > > So at this point we only have the IDLE domain and real domains. > > And finally, the complained-about construct is inside an: > > if ( !is_idle_domain(d) ) > ... > > hence config cannot be NULL, but only because of the way in which > is_{system,idle}_domain() interact. > > ~Andrew Ok. I think this should be documented as a false positive using the comment deviation mechanism in the false-positive-eclair.json file (once a name for the prefix is settled on in the other thread). -- Nicola Vetrini, BSc Software Engineer, BUGSENG srl (https://bugseng.com) ^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: MISRA C:2012 D4.11 caution on staging 2023-10-11 7:47 ` Nicola Vetrini @ 2023-10-11 7:51 ` Andrew Cooper 2023-10-13 9:27 ` Julien Grall 0 siblings, 1 reply; 17+ messages in thread From: Andrew Cooper @ 2023-10-11 7:51 UTC (permalink / raw) To: Nicola Vetrini Cc: Xen-devel, consulting, Stefano Stabellini, Jan Beulich, Julien Grall, Bertrand Marquis, roger.pau, George Dunlap, Wei Liu On 11/10/2023 3:47 pm, Nicola Vetrini wrote: > On 11/10/2023 02:15, Andrew Cooper wrote: >> On 10/10/2023 5:31 pm, Nicola Vetrini wrote: >>> Hi, >>> >>> as you can see from [1], there's a MISRA C guideline, D4.11, that is >>> supposed to be clean >>> (i.e., have no reports), but has a caution on an argument to memcpy >>> (the second argument might be null according to the checker, given a >>> set of assumptions on >>> the control flow). To access the report just click on the second link >>> in the log, which should take you to a webpage with a list of >>> MISRA guidelines. Click on D4.11 and you'll see the full report, which >>> I pasted below for convenience. >>> >>> If the finding is genuine, then some countermeasure needs to be taken >>> against this >>> possible bug, otherwise it needs to be motivated why the field >>> config->handle can't >>> be null at that point. >>> The finding is likely the result of an improvement made to the >>> checker, because the first >>> analysis I can see that spots it happened when rc1 has been tagged, >>> but that commit does not >>> touch the involved files. >>> >>> [1] https://gitlab.com/xen-project/xen/-/jobs/5251222578 >> >> That's a false positive, but I'm not entirely surprised that the checker >> struggled to see it. >> >> First, >> >> ASSERT(is_system_domain(d) ? config == NULL : config != NULL); >> >> All system domains (domid >= 0x7ff0, inc IDLE) pass a NULL config. All >> other domains pass a real config. >> >> Next, >> >> /* DOMID_{XEN,IO,etc} (other than IDLE) are sufficiently constructed. */ >> if ( is_system_domain(d) && !is_idle_domain(d) ) >> return d; >> >> So at this point we only have the IDLE domain and real domains. >> >> And finally, the complained-about construct is inside an: >> >> if ( !is_idle_domain(d) ) >> ... >> >> hence config cannot be NULL, but only because of the way in which >> is_{system,idle}_domain() interact. >> >> ~Andrew > > Ok. I think this should be documented as a false positive using the > comment deviation mechanism > in the false-positive-eclair.json file (once a name for the prefix is > settled on in the > other thread). > Yeah - I was expecting that. This code is mid-cleanup (in my copious free time, so not for several releases now...) but even when it is rearranged sufficiently for IDLE to have an earlier exit, I still wouldn't expect a static analyser to be able to recognise this as being safe. ~Andrew ^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: MISRA C:2012 D4.11 caution on staging 2023-10-11 7:51 ` Andrew Cooper @ 2023-10-13 9:27 ` Julien Grall 2023-10-13 15:11 ` Nicola Vetrini 0 siblings, 1 reply; 17+ messages in thread From: Julien Grall @ 2023-10-13 9:27 UTC (permalink / raw) To: Andrew Cooper, Nicola Vetrini Cc: Xen-devel, consulting, Stefano Stabellini, Jan Beulich, Bertrand Marquis, roger.pau, George Dunlap, Wei Liu Hi Andrew, On 11/10/2023 08:51, Andrew Cooper wrote: > On 11/10/2023 3:47 pm, Nicola Vetrini wrote: >> On 11/10/2023 02:15, Andrew Cooper wrote: >>> On 10/10/2023 5:31 pm, Nicola Vetrini wrote: >>>> Hi, >>>> >>>> as you can see from [1], there's a MISRA C guideline, D4.11, that is >>>> supposed to be clean >>>> (i.e., have no reports), but has a caution on an argument to memcpy >>>> (the second argument might be null according to the checker, given a >>>> set of assumptions on >>>> the control flow). To access the report just click on the second link >>>> in the log, which should take you to a webpage with a list of >>>> MISRA guidelines. Click on D4.11 and you'll see the full report, which >>>> I pasted below for convenience. >>>> >>>> If the finding is genuine, then some countermeasure needs to be taken >>>> against this >>>> possible bug, otherwise it needs to be motivated why the field >>>> config->handle can't >>>> be null at that point. >>>> The finding is likely the result of an improvement made to the >>>> checker, because the first >>>> analysis I can see that spots it happened when rc1 has been tagged, >>>> but that commit does not >>>> touch the involved files. >>>> >>>> [1] https://gitlab.com/xen-project/xen/-/jobs/5251222578 >>> >>> That's a false positive, but I'm not entirely surprised that the checker >>> struggled to see it. >>> >>> First, >>> >>> ASSERT(is_system_domain(d) ? config == NULL : config != NULL); >>> >>> All system domains (domid >= 0x7ff0, inc IDLE) pass a NULL config. All >>> other domains pass a real config. >>> >>> Next, >>> >>> /* DOMID_{XEN,IO,etc} (other than IDLE) are sufficiently constructed. */ >>> if ( is_system_domain(d) && !is_idle_domain(d) ) >>> return d; >>> >>> So at this point we only have the IDLE domain and real domains. >>> >>> And finally, the complained-about construct is inside an: >>> >>> if ( !is_idle_domain(d) ) >>> ... >>> >>> hence config cannot be NULL, but only because of the way in which >>> is_{system,idle}_domain() interact. >>> >>> ~Andrew >> >> Ok. I think this should be documented as a false positive using the >> comment deviation mechanism >> in the false-positive-eclair.json file (once a name for the prefix is >> settled on in the >> other thread). >> > > Yeah - I was expecting that. > > This code is mid-cleanup (in my copious free time, so not for several > releases now...) but even when it is rearranged sufficiently for IDLE to > have an earlier exit, I still wouldn't expect a static analyser to be > able to recognise this as being safe. Looking through the analysis, I think Eclair properly detect the IDLE domain. But it thinks the domain ID has changed mid function (see my other reply to Stefano). So we can return early for the IDLE domain, then this should silence Eclair. That said, it is unclear to me why it thinks the domain_id can change... A simple way to fix it would be to have a local boolean that will be used in place of is_idle_domain(d). This should help Eclair to detect that a domain cannot change its ID in the middle of domain construction. Cheers, -- Julien Grall ^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: MISRA C:2012 D4.11 caution on staging 2023-10-13 9:27 ` Julien Grall @ 2023-10-13 15:11 ` Nicola Vetrini 2023-10-16 18:06 ` Julien Grall 0 siblings, 1 reply; 17+ messages in thread From: Nicola Vetrini @ 2023-10-13 15:11 UTC (permalink / raw) To: Julien Grall Cc: Andrew Cooper, Xen-devel, consulting, Stefano Stabellini, Jan Beulich, Bertrand Marquis, roger.pau, George Dunlap, Wei Liu On 13/10/2023 11:27, Julien Grall wrote: > Hi Andrew, > > On 11/10/2023 08:51, Andrew Cooper wrote: >> On 11/10/2023 3:47 pm, Nicola Vetrini wrote: >>> On 11/10/2023 02:15, Andrew Cooper wrote: >>>> On 10/10/2023 5:31 pm, Nicola Vetrini wrote: >>>>> Hi, >>>>> >>>>> as you can see from [1], there's a MISRA C guideline, D4.11, that >>>>> is >>>>> supposed to be clean >>>>> (i.e., have no reports), but has a caution on an argument to memcpy >>>>> (the second argument might be null according to the checker, given >>>>> a >>>>> set of assumptions on >>>>> the control flow). To access the report just click on the second >>>>> link >>>>> in the log, which should take you to a webpage with a list of >>>>> MISRA guidelines. Click on D4.11 and you'll see the full report, >>>>> which >>>>> I pasted below for convenience. >>>>> >>>>> If the finding is genuine, then some countermeasure needs to be >>>>> taken >>>>> against this >>>>> possible bug, otherwise it needs to be motivated why the field >>>>> config->handle can't >>>>> be null at that point. >>>>> The finding is likely the result of an improvement made to the >>>>> checker, because the first >>>>> analysis I can see that spots it happened when rc1 has been tagged, >>>>> but that commit does not >>>>> touch the involved files. >>>>> >>>>> [1] https://gitlab.com/xen-project/xen/-/jobs/5251222578 >>>> >>>> That's a false positive, but I'm not entirely surprised that the >>>> checker >>>> struggled to see it. >>>> >>>> First, >>>> >>>> ASSERT(is_system_domain(d) ? config == NULL : config != NULL); >>>> >>>> All system domains (domid >= 0x7ff0, inc IDLE) pass a NULL config. >>>> All >>>> other domains pass a real config. >>>> >>>> Next, >>>> >>>> /* DOMID_{XEN,IO,etc} (other than IDLE) are sufficiently >>>> constructed. */ >>>> if ( is_system_domain(d) && !is_idle_domain(d) ) >>>> return d; >>>> >>>> So at this point we only have the IDLE domain and real domains. >>>> >>>> And finally, the complained-about construct is inside an: >>>> >>>> if ( !is_idle_domain(d) ) >>>> ... >>>> >>>> hence config cannot be NULL, but only because of the way in which >>>> is_{system,idle}_domain() interact. >>>> >>>> ~Andrew >>> >>> Ok. I think this should be documented as a false positive using the >>> comment deviation mechanism >>> in the false-positive-eclair.json file (once a name for the prefix is >>> settled on in the >>> other thread). >>> >> >> Yeah - I was expecting that. >> >> This code is mid-cleanup (in my copious free time, so not for several >> releases now...) but even when it is rearranged sufficiently for IDLE >> to >> have an earlier exit, I still wouldn't expect a static analyser to be >> able to recognise this as being safe. > > Looking through the analysis, I think Eclair properly detect the IDLE > domain. But it thinks the domain ID has changed mid function (see my > other reply to Stefano). > > So we can return early for the IDLE domain, then this should silence > Eclair. That said, it is unclear to me why it thinks the domain_id can > change... > Well, the implementation of the directive has best-effort precision, therefore false positives like this one are possible. Since Andrew argued that the path is indeed safe, I think it's best to deviate this as such, since other minimal changes could also make this one resurface in the future. > A simple way to fix it would be to have a local boolean that will be > used in place of is_idle_domain(d). This should help Eclair to detect > that a domain cannot change its ID in the middle of domain > construction. > > Cheers, I think only conditions are checked to get the possible code paths, to have a reasonable tradeoff between speed and analysis precision. Therefore, it's quite possible that it would give the same caution. -- Nicola Vetrini, BSc Software Engineer, BUGSENG srl (https://bugseng.com) ^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: MISRA C:2012 D4.11 caution on staging 2023-10-13 15:11 ` Nicola Vetrini @ 2023-10-16 18:06 ` Julien Grall 2023-10-17 6:11 ` Jan Beulich 0 siblings, 1 reply; 17+ messages in thread From: Julien Grall @ 2023-10-16 18:06 UTC (permalink / raw) To: Nicola Vetrini Cc: Andrew Cooper, Xen-devel, consulting, Stefano Stabellini, Jan Beulich, Bertrand Marquis, roger.pau, George Dunlap, Wei Liu Hi Nicola, On 13/10/2023 16:11, Nicola Vetrini wrote: > On 13/10/2023 11:27, Julien Grall wrote: >> Hi Andrew, >> >> On 11/10/2023 08:51, Andrew Cooper wrote: >>> On 11/10/2023 3:47 pm, Nicola Vetrini wrote: >>>> On 11/10/2023 02:15, Andrew Cooper wrote: >>>>> On 10/10/2023 5:31 pm, Nicola Vetrini wrote: >>>>>> Hi, >>>>>> >>>>>> as you can see from [1], there's a MISRA C guideline, D4.11, that is >>>>>> supposed to be clean >>>>>> (i.e., have no reports), but has a caution on an argument to memcpy >>>>>> (the second argument might be null according to the checker, given a >>>>>> set of assumptions on >>>>>> the control flow). To access the report just click on the second link >>>>>> in the log, which should take you to a webpage with a list of >>>>>> MISRA guidelines. Click on D4.11 and you'll see the full report, >>>>>> which >>>>>> I pasted below for convenience. >>>>>> >>>>>> If the finding is genuine, then some countermeasure needs to be taken >>>>>> against this >>>>>> possible bug, otherwise it needs to be motivated why the field >>>>>> config->handle can't >>>>>> be null at that point. >>>>>> The finding is likely the result of an improvement made to the >>>>>> checker, because the first >>>>>> analysis I can see that spots it happened when rc1 has been tagged, >>>>>> but that commit does not >>>>>> touch the involved files. >>>>>> >>>>>> [1] https://gitlab.com/xen-project/xen/-/jobs/5251222578 >>>>> >>>>> That's a false positive, but I'm not entirely surprised that the >>>>> checker >>>>> struggled to see it. >>>>> >>>>> First, >>>>> >>>>> ASSERT(is_system_domain(d) ? config == NULL : config != NULL); >>>>> >>>>> All system domains (domid >= 0x7ff0, inc IDLE) pass a NULL config. All >>>>> other domains pass a real config. >>>>> >>>>> Next, >>>>> >>>>> /* DOMID_{XEN,IO,etc} (other than IDLE) are sufficiently >>>>> constructed. */ >>>>> if ( is_system_domain(d) && !is_idle_domain(d) ) >>>>> return d; >>>>> >>>>> So at this point we only have the IDLE domain and real domains. >>>>> >>>>> And finally, the complained-about construct is inside an: >>>>> >>>>> if ( !is_idle_domain(d) ) >>>>> ... >>>>> >>>>> hence config cannot be NULL, but only because of the way in which >>>>> is_{system,idle}_domain() interact. >>>>> >>>>> ~Andrew >>>> >>>> Ok. I think this should be documented as a false positive using the >>>> comment deviation mechanism >>>> in the false-positive-eclair.json file (once a name for the prefix is >>>> settled on in the >>>> other thread). >>>> >>> >>> Yeah - I was expecting that. >>> >>> This code is mid-cleanup (in my copious free time, so not for several >>> releases now...) but even when it is rearranged sufficiently for IDLE to >>> have an earlier exit, I still wouldn't expect a static analyser to be >>> able to recognise this as being safe. >> >> Looking through the analysis, I think Eclair properly detect the IDLE >> domain. But it thinks the domain ID has changed mid function (see my >> other reply to Stefano). >> >> So we can return early for the IDLE domain, then this should silence >> Eclair. That said, it is unclear to me why it thinks the domain_id can >> change... >> > > Well, the implementation of the directive has best-effort precision, > therefore false positives > like this one are possible. Since Andrew argued that the path is indeed > safe, I think > it's best to deviate this as such, since other minimal changes could > also make this one > resurface in the future. If Eclair always reported the false positive, then I would have agree that deviating would make sense. But you said this was due to an improvement in the checker. I expect improvement in the checker to reduce the number of false positive, not introducing new ones. So I think we need to understand what changed. > >> A simple way to fix it would be to have a local boolean that will be >> used in place of is_idle_domain(d). This should help Eclair to detect >> that a domain cannot change its ID in the middle of domain >> construction. >> >> Cheers, > > I think only conditions are checked to get the possible code paths, to > have a reasonable > tradeoff between speed and analysis precision. Therefore, it's quite > possible that it would > give the same caution. What I want to avoid is adding a SAF-* every time there are an update to Eclair that reports more false positive. This will make the code more difficult to read. Instead, it would be best to find a way to help Eclair to detect this is not an issue and also improve readability. Would the following help Eclair? diff --git a/xen/common/domain.c b/xen/common/domain.c index 30c227967345..ab16124eabd6 100644 --- a/xen/common/domain.c +++ b/xen/common/domain.c @@ -671,6 +671,8 @@ struct domain *domain_create(domid_t domid, if ( !is_idle_domain(d) ) { + ASSERT(config); + watchdog_domain_init(d); init_status |= INIT_watchdog; Cheers, -- Julien Grall ^ permalink raw reply related [flat|nested] 17+ messages in thread
* Re: MISRA C:2012 D4.11 caution on staging 2023-10-16 18:06 ` Julien Grall @ 2023-10-17 6:11 ` Jan Beulich 2023-10-17 13:28 ` Julien Grall 0 siblings, 1 reply; 17+ messages in thread From: Jan Beulich @ 2023-10-17 6:11 UTC (permalink / raw) To: Julien Grall, Nicola Vetrini Cc: Andrew Cooper, Xen-devel, consulting, Stefano Stabellini, Bertrand Marquis, roger.pau, George Dunlap, Wei Liu On 16.10.2023 20:06, Julien Grall wrote: > Instead, it would be best to find a way to help Eclair to detect this is > not an issue and also improve readability. Would the following help Eclair? > > diff --git a/xen/common/domain.c b/xen/common/domain.c > index 30c227967345..ab16124eabd6 100644 > --- a/xen/common/domain.c > +++ b/xen/common/domain.c > @@ -671,6 +671,8 @@ struct domain *domain_create(domid_t domid, > > if ( !is_idle_domain(d) ) > { > + ASSERT(config); > + > watchdog_domain_init(d); > init_status |= INIT_watchdog; Just to mention it: Even if right now it turned out to help, it wouldn't once release builds are also checked. Jan ^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: MISRA C:2012 D4.11 caution on staging 2023-10-17 6:11 ` Jan Beulich @ 2023-10-17 13:28 ` Julien Grall 2023-10-18 12:30 ` Nicola Vetrini 0 siblings, 1 reply; 17+ messages in thread From: Julien Grall @ 2023-10-17 13:28 UTC (permalink / raw) To: Jan Beulich, Nicola Vetrini Cc: Andrew Cooper, Xen-devel, consulting, Stefano Stabellini, Bertrand Marquis, roger.pau, George Dunlap, Wei Liu Hi Jan, On 17/10/2023 07:11, Jan Beulich wrote: > On 16.10.2023 20:06, Julien Grall wrote: >> Instead, it would be best to find a way to help Eclair to detect this is >> not an issue and also improve readability. Would the following help Eclair? >> >> diff --git a/xen/common/domain.c b/xen/common/domain.c >> index 30c227967345..ab16124eabd6 100644 >> --- a/xen/common/domain.c >> +++ b/xen/common/domain.c >> @@ -671,6 +671,8 @@ struct domain *domain_create(domid_t domid, >> >> if ( !is_idle_domain(d) ) >> { >> + ASSERT(config); >> + >> watchdog_domain_init(d); >> init_status |= INIT_watchdog; > > Just to mention it: Even if right now it turned out to help, it wouldn't > once release builds are also checked. Indeed. I thought about it when writing the e-mail yesterday. I have the feeling that we are not getting many similar report today thanks to the various ASSERT(). This may mean that the ASSERT() will have to be kept during static analysis or we deviate/add proper error checking. Cheers, -- Julien Grall ^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: MISRA C:2012 D4.11 caution on staging 2023-10-17 13:28 ` Julien Grall @ 2023-10-18 12:30 ` Nicola Vetrini 2023-10-18 12:38 ` Julien Grall 0 siblings, 1 reply; 17+ messages in thread From: Nicola Vetrini @ 2023-10-18 12:30 UTC (permalink / raw) To: Julien Grall Cc: Jan Beulich, Andrew Cooper, Xen-devel, consulting, Stefano Stabellini, Bertrand Marquis, roger.pau, George Dunlap, Wei Liu On 17/10/2023 15:28, Julien Grall wrote: > Hi Jan, > > On 17/10/2023 07:11, Jan Beulich wrote: >> On 16.10.2023 20:06, Julien Grall wrote: >>> Instead, it would be best to find a way to help Eclair to detect this >>> is >>> not an issue and also improve readability. Would the following help >>> Eclair? >>> >>> diff --git a/xen/common/domain.c b/xen/common/domain.c >>> index 30c227967345..ab16124eabd6 100644 >>> --- a/xen/common/domain.c >>> +++ b/xen/common/domain.c >>> @@ -671,6 +671,8 @@ struct domain *domain_create(domid_t domid, >>> >>> if ( !is_idle_domain(d) ) >>> { >>> + ASSERT(config); >>> + >>> watchdog_domain_init(d); >>> init_status |= INIT_watchdog; >> >> Just to mention it: Even if right now it turned out to help, it >> wouldn't >> once release builds are also checked. > > Indeed. I thought about it when writing the e-mail yesterday. I have > the feeling that we are not getting many similar report today thanks > to the various ASSERT(). > > This may mean that the ASSERT() will have to be kept during static > analysis or we deviate/add proper error checking. > > Cheers, I tested this, and the report is prevented by the ASSERT. It's up to the maintainers to decide how do you want to proceed: my suggestion is deviating it, and then when someone has the time to fix this with proper error handling remove the deviation. -- Nicola Vetrini, BSc Software Engineer, BUGSENG srl (https://bugseng.com) ^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: MISRA C:2012 D4.11 caution on staging 2023-10-18 12:30 ` Nicola Vetrini @ 2023-10-18 12:38 ` Julien Grall 2023-10-18 12:52 ` Nicola Vetrini 0 siblings, 1 reply; 17+ messages in thread From: Julien Grall @ 2023-10-18 12:38 UTC (permalink / raw) To: Nicola Vetrini Cc: Jan Beulich, Andrew Cooper, Xen-devel, consulting, Stefano Stabellini, Bertrand Marquis, roger.pau, George Dunlap, Wei Liu Hi Nicola, On 18/10/2023 13:30, Nicola Vetrini wrote: > On 17/10/2023 15:28, Julien Grall wrote: > I tested this, and the report is prevented by the ASSERT. It's up to the > maintainers to > decide how do you want to proceed: my suggestion is deviating it, It is not clear to me what would you deviate. Can you clarify? Cheers, -- Julien Grall ^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: MISRA C:2012 D4.11 caution on staging 2023-10-18 12:38 ` Julien Grall @ 2023-10-18 12:52 ` Nicola Vetrini 2023-10-18 13:02 ` Julien Grall 0 siblings, 1 reply; 17+ messages in thread From: Nicola Vetrini @ 2023-10-18 12:52 UTC (permalink / raw) To: Julien Grall Cc: Jan Beulich, Andrew Cooper, Xen-devel, consulting, Stefano Stabellini, Bertrand Marquis, roger.pau, George Dunlap, Wei Liu On 18/10/2023 14:38, Julien Grall wrote: > Hi Nicola, > > On 18/10/2023 13:30, Nicola Vetrini wrote: >> On 17/10/2023 15:28, Julien Grall wrote: >> I tested this, and the report is prevented by the ASSERT. It's up to >> the maintainers to >> decide how do you want to proceed: my suggestion is deviating it, > > It is not clear to me what would you deviate. Can you clarify? > > Cheers, The memcpy call, as in: /* SAF-x-false-positive-eclair */ memcpy(d->handle, config->handle, sizeof(d->handle)); -- Nicola Vetrini, BSc Software Engineer, BUGSENG srl (https://bugseng.com) ^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: MISRA C:2012 D4.11 caution on staging 2023-10-18 12:52 ` Nicola Vetrini @ 2023-10-18 13:02 ` Julien Grall 2023-10-18 21:33 ` Stefano Stabellini 0 siblings, 1 reply; 17+ messages in thread From: Julien Grall @ 2023-10-18 13:02 UTC (permalink / raw) To: Nicola Vetrini Cc: Jan Beulich, Andrew Cooper, Xen-devel, consulting, Stefano Stabellini, Bertrand Marquis, roger.pau, George Dunlap, Wei Liu Hi, On 18/10/2023 13:52, Nicola Vetrini wrote: > On 18/10/2023 14:38, Julien Grall wrote: >> Hi Nicola, >> >> On 18/10/2023 13:30, Nicola Vetrini wrote: >>> On 17/10/2023 15:28, Julien Grall wrote: >>> I tested this, and the report is prevented by the ASSERT. It's up to >>> the maintainers to >>> decide how do you want to proceed: my suggestion is deviating it, >> >> It is not clear to me what would you deviate. Can you clarify? >> >> Cheers, > > The memcpy call, as in: > > /* SAF-x-false-positive-eclair */ > memcpy(d->handle, config->handle, sizeof(d->handle)); I am not in favor of this deviation. It could be a false positive today, but it may not be in the future. I much prefer the ASSERT() version or the rework. Cheers, -- Julien Grall ^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: MISRA C:2012 D4.11 caution on staging 2023-10-18 13:02 ` Julien Grall @ 2023-10-18 21:33 ` Stefano Stabellini 2023-10-23 10:34 ` Nicola Vetrini 0 siblings, 1 reply; 17+ messages in thread From: Stefano Stabellini @ 2023-10-18 21:33 UTC (permalink / raw) To: Julien Grall Cc: Nicola Vetrini, Jan Beulich, Andrew Cooper, Xen-devel, consulting, Stefano Stabellini, Bertrand Marquis, roger.pau, George Dunlap, Wei Liu On Wed, 18 Oct 2023, Julien Grall wrote: > On 18/10/2023 13:52, Nicola Vetrini wrote: > > On 18/10/2023 14:38, Julien Grall wrote: > > > Hi Nicola, > > > > > > On 18/10/2023 13:30, Nicola Vetrini wrote: > > > > On 17/10/2023 15:28, Julien Grall wrote: > > > > I tested this, and the report is prevented by the ASSERT. It's up to the > > > > maintainers to > > > > decide how do you want to proceed: my suggestion is deviating it, > > > > > > It is not clear to me what would you deviate. Can you clarify? > > > > > > Cheers, > > > > The memcpy call, as in: > > > > /* SAF-x-false-positive-eclair */ > > memcpy(d->handle, config->handle, sizeof(d->handle)); > > I am not in favor of this deviation. It could be a false positive today, but > it may not be in the future. > > I much prefer the ASSERT() version or the rework. Just to be clear about the next steps, Nicola are you OK with sending a patch with the ASSERT or would you prefer Julien or someone else to do it? ^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: MISRA C:2012 D4.11 caution on staging 2023-10-18 21:33 ` Stefano Stabellini @ 2023-10-23 10:34 ` Nicola Vetrini 0 siblings, 0 replies; 17+ messages in thread From: Nicola Vetrini @ 2023-10-23 10:34 UTC (permalink / raw) To: Stefano Stabellini Cc: Julien Grall, Jan Beulich, Andrew Cooper, Xen-devel, consulting, Bertrand Marquis, roger.pau, George Dunlap, Wei Liu On 18/10/2023 23:33, Stefano Stabellini wrote: > On Wed, 18 Oct 2023, Julien Grall wrote: >> On 18/10/2023 13:52, Nicola Vetrini wrote: >> > On 18/10/2023 14:38, Julien Grall wrote: >> > > Hi Nicola, >> > > >> > > On 18/10/2023 13:30, Nicola Vetrini wrote: >> > > > On 17/10/2023 15:28, Julien Grall wrote: >> > > > I tested this, and the report is prevented by the ASSERT. It's up to the >> > > > maintainers to >> > > > decide how do you want to proceed: my suggestion is deviating it, >> > > >> > > It is not clear to me what would you deviate. Can you clarify? >> > > >> > > Cheers, >> > >> > The memcpy call, as in: >> > >> > /* SAF-x-false-positive-eclair */ >> > memcpy(d->handle, config->handle, sizeof(d->handle)); >> >> I am not in favor of this deviation. It could be a false positive >> today, but >> it may not be in the future. >> >> I much prefer the ASSERT() version or the rework. > > Just to be clear about the next steps, Nicola are you OK with sending a > patch with the ASSERT or would you prefer Julien or someone else to do > it? Anyone can pick this up. -- Nicola Vetrini, BSc Software Engineer, BUGSENG srl (https://bugseng.com) ^ permalink raw reply [flat|nested] 17+ messages in thread
end of thread, other threads:[~2023-10-23 10:34 UTC | newest] Thread overview: 17+ messages (download: mbox.gz follow: Atom feed -- links below jump to the message on this page -- 2023-10-10 9:31 MISRA C:2012 D4.11 caution on staging Nicola Vetrini 2023-10-11 0:01 ` Stefano Stabellini 2023-10-13 9:23 ` Julien Grall 2023-10-11 0:15 ` Andrew Cooper 2023-10-11 7:47 ` Nicola Vetrini 2023-10-11 7:51 ` Andrew Cooper 2023-10-13 9:27 ` Julien Grall 2023-10-13 15:11 ` Nicola Vetrini 2023-10-16 18:06 ` Julien Grall 2023-10-17 6:11 ` Jan Beulich 2023-10-17 13:28 ` Julien Grall 2023-10-18 12:30 ` Nicola Vetrini 2023-10-18 12:38 ` Julien Grall 2023-10-18 12:52 ` Nicola Vetrini 2023-10-18 13:02 ` Julien Grall 2023-10-18 21:33 ` Stefano Stabellini 2023-10-23 10:34 ` Nicola Vetrini
This is an external index of several public inboxes, see mirroring instructions on how to clone and mirror all data and code used by this external index.