|
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] Re: [PATCH v4] xen/arm: validate non-common page-table ranges
On 14-Nov-26 02:53, Gabriel Quintáns Souto wrote:
>> There's no need to check for TEMPORARY area. It's only ever used in boot
>> pgtables, not in runtime tables. Therefore what we need is just the
>> following:
>> virt < DOMHEAP_VIRT_START && size <= DOMHEAP_VIRT_START - virt
>> without any helper.
>>
>> BTW. This patch breaks Arm32 MPU build. If at all, the helper should be
>> placed
>> under mmu/.
>
> Ok, thanks for the review. I was not really sure whether TEMPORARY_AREA was
> needed or not, so thanks for making that clear. One thing, though:
> `virt < DOMHEAP_VIRT_START && size <= DOMHEAP_VIRT_START - virt`
>
> To do that validation in pt.c, I would need to use `#ifdef CONFIG_ARM_32`
> instead of `IS_ENABLED()` (which would let DCE drop it on arm64), otherwise it
> doesn't build on arm64. That's why I made the xen_pt_range_is_common helper.
> How
> should I approach this then? I'll drop `ranges_overlap()` either way, but
> should I keep `xen_pt_range_is_common()` and move it under `mmu/`, or just
> drop
> all the helpers and use `#ifdef CONFIG_ARM_32` directly in `pt.c`? I assume a
> single `#ifdef` in `pt.c` is fine, so I'd go with the second option, as you
> suggested. Let me know what you think.
I'm ok with the #ifdef in xen_pt_update() together with a comment explaining why
this is needed for Arm32 and not Arm64.
~Michal
>
>>> +#include <asm/mmu/layout.h>
>> Is this needed?
>
> I think I added it while trying to get the helper to build, but I think
> it's already pulled in through asm/config.h, so I'll likely drop it in v5.
>
> Best regards,
> Gabriel
>
> On Thu Nov 12, 2026 at 3:26 PM CET, Michal Orzel wrote:
>>
>>
>> On 10-Nov-26 15:52, Gabriel Quintáns Souto wrote:
>>> On arm32, page-tables are different on each CPUs, although they share some
>>> common mappings.
>> Please say that the non-common area is the per-CPU domheap mapping.
>>
>>>
>>> The xen_pt_update function assumed that only said mappings would be
>>> modified, but did not check this assumption.
>>>
>>> Add a range check to reject modifications to non-common mappings.
>>>
>>> Signed-off-by: Gabriel Quintáns Souto <gabi.qs.mail@xxxxxxxxx>
>>> ---
>>> v4:
>>> - Rebased onto staging (v3 was diffed against v2).
>>>
>>> v3:
>>> - Dropped accidentally tracked patch file from commit.
>>>
>>> v2:
>>> - Inverted polarity of xen_pt_range_is_non_common() as suggested.
>>> - Moved xen_pt_range_is_common() implementation to arch-specific headers
>>> (arm32 vs arm64) instead of keeping #ifdef CONFIG_ARM_32 in pt.c (tried
>>> IS_ENABLED() for DCE, but ARM64 lacks definitions for
>>> DOMHEAP_VIRT_START/TEMPORARY_AREA_ADDR, breaking the build).
>>> - Added ASSERT_UNREACHABLE() on failure path in xen_pt_update().
>>> ---
>>> xen/arch/arm/include/asm/arm32/mm.h | 24 ++++++++++++++++++++++++
>>> xen/arch/arm/include/asm/arm64/mm.h | 10 ++++++++++
>>> xen/arch/arm/mmu/pt.c | 14 +++++++-------
>>> 3 files changed, 41 insertions(+), 7 deletions(-)
>>>
>>> diff --git a/xen/arch/arm/include/asm/arm32/mm.h
>>> b/xen/arch/arm/include/asm/arm32/mm.h
>>> index 856f2db..c7d09d3 100644
>>> --- a/xen/arch/arm/include/asm/arm32/mm.h
>>> +++ b/xen/arch/arm/include/asm/arm32/mm.h
>>> @@ -16,6 +16,30 @@ static inline bool arch_mfns_in_directmap(unsigned long
>>> mfn, unsigned long nr)
>>> return false;
>>> }
>>>
>>> +static inline bool ranges_overlap(unsigned long start1, unsigned long
>>> size1,
>>> + unsigned long start2, unsigned long size2)
>>> +{
>>> + if ( start1 < start2 )
>>> + return start2 - start1 < size1;
>>> +
>>> + return start1 - start2 < size2;
>>> +}
>>> +
>>> +/*
>>> + * On ARM32, page-tables differ between CPUs, so not all mappings are
>>> common.
>>> + */
>>> +static inline bool xen_pt_range_is_common(unsigned long virt,
>>> + unsigned long size)
>>> +{
>>> + unsigned long temporary_start = TEMPORARY_AREA_ADDR(0);
>> There's no need to check for TEMPORARY area. It's only ever used in boot
>> pgtables, not in runtime tables. Therefore what we need is just the
>> following:
>> virt < DOMHEAP_VIRT_START && size <= DOMHEAP_VIRT_START - virt
>> without any helper.
>>
>> BTW. This patch breaks Arm32 MPU build. If at all, the helper should be
>> placed
>> under mmu/.
>>
>>
>>> + unsigned long temporary_size = XEN_PT_LEVEL_SIZE(1);
>>> +
>>> + return !ranges_overlap(virt, size,
>>> + DOMHEAP_VIRT_START, DOMHEAP_VIRT_SIZE) &&
>>> + !ranges_overlap(virt, size,
>>> + temporary_start, temporary_size);
>>> +}
>>> +
>>> bool init_domheap_mappings(unsigned int cpu);
>>>
>>> static inline void arch_setup_page_tables(void)
>>> diff --git a/xen/arch/arm/include/asm/arm64/mm.h
>>> b/xen/arch/arm/include/asm/arm64/mm.h
>>> index b4f7545..a3cb56d 100644
>>> --- a/xen/arch/arm/include/asm/arm64/mm.h
>>> +++ b/xen/arch/arm/include/asm/arm64/mm.h
>>> @@ -12,6 +12,16 @@ static inline bool arch_mfns_in_directmap(unsigned long
>>> mfn, unsigned long nr)
>>> return true;
>>> }
>>>
>>> +/*
>>> + * On ARM64, the Xen page-tables are shared by all the CPUs,
>>> + * any range is therefore common.
>>> + */
>>> +static inline bool xen_pt_range_is_common(unsigned long virt,
>>> + unsigned long size)
>>> +{
>>> + return true;
>>> +}
>>> +
>>> void arch_setup_page_tables(void);
>>>
>>> void update_boot_mapping(bool enable);
>>> diff --git a/xen/arch/arm/mmu/pt.c b/xen/arch/arm/mmu/pt.c
>>> index 81fe121..42fa79a 100644
>>> --- a/xen/arch/arm/mmu/pt.c
>>> +++ b/xen/arch/arm/mmu/pt.c
>>> @@ -13,6 +13,7 @@
>>>
>>> #include <asm/current.h>
>>> #include <asm/fixmap.h>
>>> +#include <asm/mmu/layout.h>
>> Is this needed?
>>
>> ~Michal
>
|
![]() |
Lists.xenproject.org is hosted with RackSpace, monitoring our |