Re: [XEN PATCH 1/4] xen/arm: address violations of MISRA C:2012 Rule 13.1

[Simone Ballarin <simone.ballarin@xxxxxxxxxxx>]

On 19/10/23 10:19, Julien Grall wrote:
> Hi Simone,
>
> On 19/10/2023 08:34, Simone Ballarin wrote:
> > On 18/10/23 17:03, Julien Grall wrote:
> > > Hi,
> > >
> > > On 18/10/2023 15:18, Simone Ballarin wrote:
> > > > Rule 13.1: Initializer lists shall not contain persistent side effects
> > > >
> > > > This patch moves expressions with side-effects into new variables 
> > > > before
> > > > the initializer lists.
> > > Looking at the code, I feel the commit message should be a bit more 
> > > verbose because they are no apparent side-effects.
> > > > Function calls do not necessarily have side-effects, in these cases the
> > > > GCC pure or const attributes are added when possible.
> > > You are only adding pure in this patch.
> > >
> > > > No functional changes.
> > > >
> > > > Signed-off-by: Simone Ballarin <simone.ballarin@xxxxxxxxxxx>
> > > >
> > > > ---
> > > > Function attributes pure and const do not need to be added as GCC
> > > > attributes, they can be added using ECLAIR configurations.
> > > > ---
> > > >   xen/arch/arm/device.c              |  6 +++---
> > > >   xen/arch/arm/guestcopy.c           | 12 ++++++++----
> > > >   xen/arch/arm/include/asm/current.h |  2 +-
> > > >   3 files changed, 12 insertions(+), 8 deletions(-)
> > > >
> > > > diff --git a/xen/arch/arm/device.c b/xen/arch/arm/device.c
> > > > index 1f631d3274..e9be078415 100644
> > > > --- a/xen/arch/arm/device.c
> > > > +++ b/xen/arch/arm/device.c
> > > > @@ -319,6 +319,8 @@ int handle_device(struct domain *d, struct 
> > > > dt_device_node *dev, p2m_type_t p2mt,
> > > >       int res;
> > > >       paddr_t addr, size;
> > > >       bool own_device = !dt_device_for_passthrough(dev);
> > > > +    bool dev_is_hostbridge = is_pci_passthrough_enabled() &&
> > > > +                             device_get_class(dev) == 
> > > > DEVICE_PCI_HOSTBRIDGE;
> > > The commit message suggests that the code is moved because there are 
> > > side-effects. But none of them should have any side-effects.
> > device_get_class contains an 'ASSERT(dev != NULL)' which is definitely
> > a side-effect.
> Just to confirm my understanding, the side-effect here would be the fact 
> that it traps and then panic(). IOW, if the trap was hypothetically 
> replaced by a while (1), then it would not be an issue. is it correct? >
No, it isn't. A infinite loop is a side effect.

> I can see two solutions:
>    1) Remove the ASSERT(). It is only here to make the NULL dereference 
> obvious in debug build. That said, the stack trace for a NULL 
> dereference would still be as clear.
Removing the ASSERT just to make MISRA happy does not sound good to me.

>    2) Replace the ASSERT with a proper check if ( !dev ) return 
> DEVICE_UNKONWN. AFAIU, we would not be able to add a 
> ASSERT_UNREACHABLE() because this would be again a perceived side-effect.
Replacing it with a proper check can be a solution, but I still prefer 
to add a deviation or move the invocation outside the initializer list.
> The former feels a bit circumventing MISRA as the side effect is 
> technically still present. Just hidden. But I do prefer over 2).
> > > >       /*
> > > >        * We want to avoid mapping the MMIO in dom0 for the following 
> > > > cases:
> > > >        *   - The device is owned by dom0 (i.e. it has been flagged for
> > > > @@ -329,9 +331,7 @@ int handle_device(struct domain *d, struct 
> > > > dt_device_node *dev, p2m_type_t p2mt,
> > > >       struct map_range_data mr_data = {
> > > >           .d = d,
> > > >           .p2mt = p2mt,
> > > > -        .skip_mapping = !own_device ||
> > > > -                        (is_pci_passthrough_enabled() &&
> > > > -                        (device_get_class(dev) == 
> > > > DEVICE_PCI_HOSTBRIDGE)),
> > > > +        .skip_mapping = !own_device || dev_is_hostbridge,
> > > >           .iomem_ranges = iomem_ranges,
> > > >           .irq_ranges = irq_ranges
> > > >       };
> > > > diff --git a/xen/arch/arm/guestcopy.c b/xen/arch/arm/guestcopy.c
> > > > index 6716b03561..3ec6743bf6 100644
> > > > --- a/xen/arch/arm/guestcopy.c
> > > > +++ b/xen/arch/arm/guestcopy.c
> > > > @@ -109,27 +109,31 @@ static unsigned long copy_guest(void *buf, 
> > > > uint64_t addr, unsigned int len,
> > > >   unsigned long raw_copy_to_guest(void *to, const void *from, 
> > > > unsigned int len)
> > > >   {
> > > > +    struct vcpu *current_vcpu = current;
> > > It is not clear to me what is the perceived side effect here and the 
> > > others below. Can you clarify?
> > I will use the pure attribute.
> So x86 is using a function to define current. But AFAICT this is not the 
> case on Arm. So how would you add the pure?
> If it is by adding a function, then I would first like to understand 
> which part 'current' is currently perceived as a side-effect.
Yes, sorry I was looking to the wrong definition.
In ARM the problem is the presence of a *volatile* ASM.
Please take a look here:

https://saas.eclairit.com:3787/fs/var/local/eclair/XEN.ecdf/ECLAIR_normal/arm/for-4.19/ARM64-Set2/latest/PROJECT.ecd;/by_service/MC3R1.R13.1.html#{"select":true,"selection":{"hiddenAreaKinds":[],"hiddenSubareaKinds":[],"show":true,"selector":{"enabled":true,"negated":false,"kind":0,"domain":"fingerprint","inputs":[{"enabled":true,"text":"0da7f0c9aea5491eba343618f965c81f5d7aed3c"}]}}}

> Cheers,
--
Simone Ballarin, M.Sc.

Field Application Engineer, BUGSENG (https://bugseng.com)