[2/2] xen/x86: hap: Clean-up and harden hap_enable()
diff mbox series

Message ID 20200204093411.15887-3-julien@xen.org
State Superseded
Headers show
Series
  • xen/x86: hap: Small clean-up/hardening in hap_enable()
Related show

Commit Message

Julien Grall Feb. 4, 2020, 9:34 a.m. UTC
From: Julien Grall <jgrall@amazon.com>

Unlike shadow_enable(), hap_enable() can only be called once during
domain creation and with the mode equal to mode equal to
PG_external | PG_translate | PG_refcounts.

If it were called twice, then we might have something interesting
problem as the p2m tables would be re-allocated (and therefore all the
mappings would be lost).

Add code to sanity check the mode and that the function is only called
once. Take the opportunity to an if checking that PG_translate is set.

Signed-off-by: Julien Grall <jgrall@amazon.com>

---

It is not entirely clear when PG_translate was enforced.
---
 xen/arch/x86/mm/hap/hap.c | 18 +++++++++++-------
 1 file changed, 11 insertions(+), 7 deletions(-)

Comments

Roger Pau Monné Feb. 4, 2020, 10:51 a.m. UTC | #1
On Tue, Feb 04, 2020 at 09:34:11AM +0000, Julien Grall wrote:
> From: Julien Grall <jgrall@amazon.com>
> 
> Unlike shadow_enable(), hap_enable() can only be called once during
> domain creation and with the mode equal to mode equal to
                                    ^ equals to
> PG_external | PG_translate | PG_refcounts.
> 
> If it were called twice, then we might have something interesting
                                              ^ a problem
> problem as the p2m tables would be re-allocated (and therefore all the
> mappings would be lost).
> 
> Add code to sanity check the mode and that the function is only called
> once. Take the opportunity to an if checking that PG_translate is set.
                                ^ add an if
> 
> Signed-off-by: Julien Grall <jgrall@amazon.com>
> 
> ---
> 
> It is not entirely clear when PG_translate was enforced.
> ---
>  xen/arch/x86/mm/hap/hap.c | 18 +++++++++++-------
>  1 file changed, 11 insertions(+), 7 deletions(-)
> 
> diff --git a/xen/arch/x86/mm/hap/hap.c b/xen/arch/x86/mm/hap/hap.c
> index 31362a31b6..b734e2e6d3 100644
> --- a/xen/arch/x86/mm/hap/hap.c
> +++ b/xen/arch/x86/mm/hap/hap.c
> @@ -445,6 +445,13 @@ int hap_enable(struct domain *d, u32 mode)
>      unsigned int i;
>      int rv = 0;
>  
> +    if ( mode != (PG_external | PG_translate | PG_refcounts) )
> +        return -EINVAL;
> +
> +    /* The function can only be called once */
> +    if ( d->arch.paging.mode != 0 )
> +        return -EINVAL;

If you want to return EINVAL for both they can be merged into a single
if. Also note that this would usually be written as
if ( d->arch.paging.mode ) to keep it shorter.

Albeit I think you might want to return EEXIST instead of EINVAL if
mode is already set.

> +
>      domain_pause(d);
>  
>      old_pages = d->arch.paging.hap.total_pages;
> @@ -465,13 +472,10 @@ int hap_enable(struct domain *d, u32 mode)
>      d->arch.paging.alloc_page = hap_alloc_p2m_page;
>      d->arch.paging.free_page = hap_free_p2m_page;
>  
> -    /* allocate P2m table */
> -    if ( mode & PG_translate )
> -    {
> -        rv = p2m_alloc_table(p2m_get_hostp2m(d));
> -        if ( rv != 0 )
> -            goto out;
> -    }
> +    /* allocate P2M table */
> +    rv = p2m_alloc_table(p2m_get_hostp2m(d));
> +    if ( rv != 0 )

I would also avoid comparing against 0 here.

Thanks, Roger.
Julien Grall Feb. 4, 2020, 11:11 a.m. UTC | #2
On 04/02/2020 10:51, Roger Pau Monné wrote:
> On Tue, Feb 04, 2020 at 09:34:11AM +0000, Julien Grall wrote:
>> From: Julien Grall <jgrall@amazon.com>
>>
>> Unlike shadow_enable(), hap_enable() can only be called once during
>> domain creation and with the mode equal to mode equal to
>                                      ^ equals to

Will fix it.

>> PG_external | PG_translate | PG_refcounts.
>>
>> If it were called twice, then we might have something interesting
>                                                ^ a problem
>> problem as the p2m tables would be re-allocated (and therefore all the
>> mappings would be lost).
>>
>> Add code to sanity check the mode and that the function is only called
>> once. Take the opportunity to an if checking that PG_translate is set.
>                                  ^ add an if

Will fix it.

>>
>> Signed-off-by: Julien Grall <jgrall@amazon.com>
>>
>> ---
>>
>> It is not entirely clear when PG_translate was enforced.
>> ---
>>   xen/arch/x86/mm/hap/hap.c | 18 +++++++++++-------
>>   1 file changed, 11 insertions(+), 7 deletions(-)
>>
>> diff --git a/xen/arch/x86/mm/hap/hap.c b/xen/arch/x86/mm/hap/hap.c
>> index 31362a31b6..b734e2e6d3 100644
>> --- a/xen/arch/x86/mm/hap/hap.c
>> +++ b/xen/arch/x86/mm/hap/hap.c
>> @@ -445,6 +445,13 @@ int hap_enable(struct domain *d, u32 mode)
>>       unsigned int i;
>>       int rv = 0;
>>   
>> +    if ( mode != (PG_external | PG_translate | PG_refcounts) )
>> +        return -EINVAL;
>> +
>> +    /* The function can only be called once */
>> +    if ( d->arch.paging.mode != 0 )
>> +        return -EINVAL;
> 
> If you want to return EINVAL for both they can be merged into a single
> if. Also note that this would usually be written as
> if ( d->arch.paging.mode ) to keep it shorter.

To be honest, this is a matter of taste. There is also an argument that 
for MISRA, your suggestion is not compliant (see Rule 14.4).

> 
> Albeit I think you might want to return EEXIST instead of EINVAL if
> mode is already set.

I am happy with that.

> 
>> +
>>       domain_pause(d);
>>   
>>       old_pages = d->arch.paging.hap.total_pages;
>> @@ -465,13 +472,10 @@ int hap_enable(struct domain *d, u32 mode)
>>       d->arch.paging.alloc_page = hap_alloc_p2m_page;
>>       d->arch.paging.free_page = hap_free_p2m_page;
>>   
>> -    /* allocate P2m table */
>> -    if ( mode & PG_translate )
>> -    {
>> -        rv = p2m_alloc_table(p2m_get_hostp2m(d));
>> -        if ( rv != 0 )
>> -            goto out;
>> -    }
>> +    /* allocate P2M table */
>> +    rv = p2m_alloc_table(p2m_get_hostp2m(d));
>> +    if ( rv != 0 )
> 
> I would also avoid comparing against 0 here.

See above.

Cheers,
Roger Pau Monné Feb. 4, 2020, 11:28 a.m. UTC | #3
On Tue, Feb 04, 2020 at 11:11:11AM +0000, Julien Grall wrote:
> 
> 
> On 04/02/2020 10:51, Roger Pau Monné wrote:
> > On Tue, Feb 04, 2020 at 09:34:11AM +0000, Julien Grall wrote:
> > > From: Julien Grall <jgrall@amazon.com>
> > > 
> > > Unlike shadow_enable(), hap_enable() can only be called once during
> > > domain creation and with the mode equal to mode equal to
> >                                      ^ equals to
> 
> Will fix it.
> 
> > > PG_external | PG_translate | PG_refcounts.
> > > 
> > > If it were called twice, then we might have something interesting
> >                                                ^ a problem
> > > problem as the p2m tables would be re-allocated (and therefore all the
> > > mappings would be lost).
> > > 
> > > Add code to sanity check the mode and that the function is only called
> > > once. Take the opportunity to an if checking that PG_translate is set.
> >                                  ^ add an if
> 
> Will fix it.
> 
> > > 
> > > Signed-off-by: Julien Grall <jgrall@amazon.com>
> > > 
> > > ---
> > > 
> > > It is not entirely clear when PG_translate was enforced.
> > > ---
> > >   xen/arch/x86/mm/hap/hap.c | 18 +++++++++++-------
> > >   1 file changed, 11 insertions(+), 7 deletions(-)
> > > 
> > > diff --git a/xen/arch/x86/mm/hap/hap.c b/xen/arch/x86/mm/hap/hap.c
> > > index 31362a31b6..b734e2e6d3 100644
> > > --- a/xen/arch/x86/mm/hap/hap.c
> > > +++ b/xen/arch/x86/mm/hap/hap.c
> > > @@ -445,6 +445,13 @@ int hap_enable(struct domain *d, u32 mode)
> > >       unsigned int i;
> > >       int rv = 0;
> > > +    if ( mode != (PG_external | PG_translate | PG_refcounts) )
> > > +        return -EINVAL;
> > > +
> > > +    /* The function can only be called once */
> > > +    if ( d->arch.paging.mode != 0 )
> > > +        return -EINVAL;
> > 
> > If you want to return EINVAL for both they can be merged into a single
> > if. Also note that this would usually be written as
> > if ( d->arch.paging.mode ) to keep it shorter.
> 
> To be honest, this is a matter of taste. There is also an argument that for
> MISRA, your suggestion is not compliant (see Rule 14.4).

Oh, then we should add those rules to CODING_STYLE if they are to be
enforced.

So far the style of most of the hypervisor code is to omit the value
when comparing against 0 or NULL AFAIK.

I don't have an issue with requiring explicit comparisons, but it
needs to be documented so we can aim to have an homogeneous style,
because so far I've been recommending the other way around.

Thanks, Roger.
George Dunlap Feb. 4, 2020, 11:33 a.m. UTC | #4
On 2/4/20 11:28 AM, Roger Pau Monné wrote:
> On Tue, Feb 04, 2020 at 11:11:11AM +0000, Julien Grall wrote:
>>
>>
>> On 04/02/2020 10:51, Roger Pau Monné wrote:
>>> On Tue, Feb 04, 2020 at 09:34:11AM +0000, Julien Grall wrote:
>>>> From: Julien Grall <jgrall@amazon.com>
>>>>
>>>> Unlike shadow_enable(), hap_enable() can only be called once during
>>>> domain creation and with the mode equal to mode equal to
>>>                                      ^ equals to
>>
>> Will fix it.
>>
>>>> PG_external | PG_translate | PG_refcounts.
>>>>
>>>> If it were called twice, then we might have something interesting
>>>                                                ^ a problem
>>>> problem as the p2m tables would be re-allocated (and therefore all the
>>>> mappings would be lost).
>>>>
>>>> Add code to sanity check the mode and that the function is only called
>>>> once. Take the opportunity to an if checking that PG_translate is set.
>>>                                  ^ add an if
>>
>> Will fix it.
>>
>>>>
>>>> Signed-off-by: Julien Grall <jgrall@amazon.com>
>>>>
>>>> ---
>>>>
>>>> It is not entirely clear when PG_translate was enforced.
>>>> ---
>>>>   xen/arch/x86/mm/hap/hap.c | 18 +++++++++++-------
>>>>   1 file changed, 11 insertions(+), 7 deletions(-)
>>>>
>>>> diff --git a/xen/arch/x86/mm/hap/hap.c b/xen/arch/x86/mm/hap/hap.c
>>>> index 31362a31b6..b734e2e6d3 100644
>>>> --- a/xen/arch/x86/mm/hap/hap.c
>>>> +++ b/xen/arch/x86/mm/hap/hap.c
>>>> @@ -445,6 +445,13 @@ int hap_enable(struct domain *d, u32 mode)
>>>>       unsigned int i;
>>>>       int rv = 0;
>>>> +    if ( mode != (PG_external | PG_translate | PG_refcounts) )
>>>> +        return -EINVAL;
>>>> +
>>>> +    /* The function can only be called once */
>>>> +    if ( d->arch.paging.mode != 0 )
>>>> +        return -EINVAL;
>>>
>>> If you want to return EINVAL for both they can be merged into a single
>>> if. Also note that this would usually be written as
>>> if ( d->arch.paging.mode ) to keep it shorter.
>>
>> To be honest, this is a matter of taste. There is also an argument that for
>> MISRA, your suggestion is not compliant (see Rule 14.4).
> 
> Oh, then we should add those rules to CODING_STYLE if they are to be
> enforced.
> 
> So far the style of most of the hypervisor code is to omit the value
> when comparing against 0 or NULL AFAIK.
> 
> I don't have an issue with requiring explicit comparisons, but it
> needs to be documented so we can aim to have an homogeneous style,
> because so far I've been recommending the other way around.

Indeed, the general preference of the codebase as a whole is to favor
conciseness in this case; there's value in being consistent.

I don't want to be annoying about this.  I don't agree with the MISRA
rule here; but I do think that MISRA is important.  OTOH this is in x86
code, which I don't think anyone has suggested become MISRA compliant.
And if we're going to start making these sorts of changes, I agree that
we should have a discussion about it, rather than implicitly do things
sometimes one way and sometimes another.

 -George
Julien Grall Feb. 4, 2020, 11:44 a.m. UTC | #5
On 04/02/2020 11:28, Roger Pau Monné wrote:
> On Tue, Feb 04, 2020 at 11:11:11AM +0000, Julien Grall wrote:
>>
>>
>> On 04/02/2020 10:51, Roger Pau Monné wrote:
>>> On Tue, Feb 04, 2020 at 09:34:11AM +0000, Julien Grall wrote:
>>>> From: Julien Grall <jgrall@amazon.com>
>>>>
>>>> Unlike shadow_enable(), hap_enable() can only be called once during
>>>> domain creation and with the mode equal to mode equal to
>>>                                       ^ equals to
>>
>> Will fix it.
>>
>>>> PG_external | PG_translate | PG_refcounts.
>>>>
>>>> If it were called twice, then we might have something interesting
>>>                                                 ^ a problem
>>>> problem as the p2m tables would be re-allocated (and therefore all the
>>>> mappings would be lost).
>>>>
>>>> Add code to sanity check the mode and that the function is only called
>>>> once. Take the opportunity to an if checking that PG_translate is set.
>>>                                   ^ add an if
>>
>> Will fix it.
>>
>>>>
>>>> Signed-off-by: Julien Grall <jgrall@amazon.com>
>>>>
>>>> ---
>>>>
>>>> It is not entirely clear when PG_translate was enforced.
>>>> ---
>>>>    xen/arch/x86/mm/hap/hap.c | 18 +++++++++++-------
>>>>    1 file changed, 11 insertions(+), 7 deletions(-)
>>>>
>>>> diff --git a/xen/arch/x86/mm/hap/hap.c b/xen/arch/x86/mm/hap/hap.c
>>>> index 31362a31b6..b734e2e6d3 100644
>>>> --- a/xen/arch/x86/mm/hap/hap.c
>>>> +++ b/xen/arch/x86/mm/hap/hap.c
>>>> @@ -445,6 +445,13 @@ int hap_enable(struct domain *d, u32 mode)
>>>>        unsigned int i;
>>>>        int rv = 0;
>>>> +    if ( mode != (PG_external | PG_translate | PG_refcounts) )
>>>> +        return -EINVAL;
>>>> +
>>>> +    /* The function can only be called once */
>>>> +    if ( d->arch.paging.mode != 0 )
>>>> +        return -EINVAL;
>>>
>>> If you want to return EINVAL for both they can be merged into a single
>>> if. Also note that this would usually be written as
>>> if ( d->arch.paging.mode ) to keep it shorter.
>>
>> To be honest, this is a matter of taste. There is also an argument that for
>> MISRA, your suggestion is not compliant (see Rule 14.4).
> 
> Oh, then we should add those rules to CODING_STYLE if they are to be
> enforced.

I am not looking to enforce anything at the moment. My main point here 
is this is pretty much as matter of taste. But there might be concern 
with your suggestion if go forward with MISRA (this is not the only one 
though ;)).

> 
> So far the style of most of the hypervisor code is to omit the value
> when comparing against 0 or NULL AFAIK.
> 
> I don't have an issue with requiring explicit comparisons, but it
> needs to be documented so we can aim to have an homogeneous style,
> because so far I've been recommending the other way around.

Aside the MISRA, there are some cases where I feel the explicit 
comparisons make sense. But I don't have any rational for them and view 
this as a matter of taste. So I would leave it to the author of the 
patch the choice.

Cheers,
Jan Beulich Feb. 4, 2020, 12:36 p.m. UTC | #6
On 04.02.2020 12:44, Julien Grall wrote:
> Aside the MISRA, there are some cases where I feel the explicit 
> comparisons make sense. But I don't have any rational for them and view 
> this as a matter of taste. So I would leave it to the author of the 
> patch the choice.

FWIW, I disagree on this aspect. Consistency of the code base
is, I think, more important an aspect. Yes, we likely never
won't reach a fully consistent state, as goals shift, but
anyway.

Jan
Julien Grall Feb. 4, 2020, 12:51 p.m. UTC | #7
On 04/02/2020 12:36, Jan Beulich wrote:
> On 04.02.2020 12:44, Julien Grall wrote:
>> Aside the MISRA, there are some cases where I feel the explicit
>> comparisons make sense. But I don't have any rational for them and view
>> this as a matter of taste. So I would leave it to the author of the
>> patch the choice.
> 
> FWIW, I disagree on this aspect. Consistency of the code base
> is, I think, more important an aspect. Yes, we likely never
> won't reach a fully consistent state, as goals shift, but
> anyway.

The meaning of consistency is quite broad. What you view as consistent 
may not be for me (and vice versa). So we are down to the "matter of 
taste" territory.

We could suggest to be consistent with the code surrounding, but I think 
you will not be happy with it as some code does not suit the coding 
style (see the recent discussion about using __).

Cheers,

Patch
diff mbox series

diff --git a/xen/arch/x86/mm/hap/hap.c b/xen/arch/x86/mm/hap/hap.c
index 31362a31b6..b734e2e6d3 100644
--- a/xen/arch/x86/mm/hap/hap.c
+++ b/xen/arch/x86/mm/hap/hap.c
@@ -445,6 +445,13 @@  int hap_enable(struct domain *d, u32 mode)
     unsigned int i;
     int rv = 0;
 
+    if ( mode != (PG_external | PG_translate | PG_refcounts) )
+        return -EINVAL;
+
+    /* The function can only be called once */
+    if ( d->arch.paging.mode != 0 )
+        return -EINVAL;
+
     domain_pause(d);
 
     old_pages = d->arch.paging.hap.total_pages;
@@ -465,13 +472,10 @@  int hap_enable(struct domain *d, u32 mode)
     d->arch.paging.alloc_page = hap_alloc_p2m_page;
     d->arch.paging.free_page = hap_free_p2m_page;
 
-    /* allocate P2m table */
-    if ( mode & PG_translate )
-    {
-        rv = p2m_alloc_table(p2m_get_hostp2m(d));
-        if ( rv != 0 )
-            goto out;
-    }
+    /* allocate P2M table */
+    rv = p2m_alloc_table(p2m_get_hostp2m(d));
+    if ( rv != 0 )
+        goto out;
 
     for ( i = 0; i < MAX_NESTEDP2M; i++ )
     {