Re: [PATCH v2 4/6] arm/mpu: Destroy an existing entry in Xen MPU memory mapping table

["Orzel, Michal" <michal.orzel@xxxxxxx>]

On 02/07/2025 16:13, Hari Limaye wrote:
> From: Penny Zheng <Penny.Zheng@xxxxxxx>
> 
> This commit expands xen_mpumap_update/xen_mpumap_update_entry to include
> destroying an existing entry.
> 
> We define a new helper "disable_mpu_region_from_index" to disable the MPU
> region based on index. If region is within [0, 31], we could quickly
> disable the MPU region through PRENR_EL2 which provides direct access to the
> PRLAR_EL2.EN bits of EL2 MPU regions.
> 
> Rignt now, we only support destroying a *WHOLE* MPU memory region,
> part-region removing is not supported, as in worst case, it will
> leave two fragments behind.
> 
> Signed-off-by: Penny Zheng <penny.zheng@xxxxxxx>
> Signed-off-by: Wei Chen <wei.chen@xxxxxxx>
> Signed-off-by: Luca Fancellu <luca.fancellu@xxxxxxx>
> Signed-off-by: Hari Limaye <hari.limaye@xxxxxxx>
> ---
> Changes from v1:
> - Move check for part-region removal outside if condition
> - Use normal printk
> ---
>  xen/arch/arm/include/asm/mpu.h        |  2 +
>  xen/arch/arm/include/asm/mpu/cpregs.h |  4 ++
>  xen/arch/arm/mpu/mm.c                 | 69 ++++++++++++++++++++++++++-
>  3 files changed, 73 insertions(+), 2 deletions(-)
> 
> diff --git a/xen/arch/arm/include/asm/mpu.h b/xen/arch/arm/include/asm/mpu.h
> index 63560c613b..5053edaf63 100644
> --- a/xen/arch/arm/include/asm/mpu.h
> +++ b/xen/arch/arm/include/asm/mpu.h
> @@ -23,6 +23,8 @@
>  #define NUM_MPU_REGIONS_MASK    (NUM_MPU_REGIONS - 1)
>  #define MAX_MPU_REGION_NR       NUM_MPU_REGIONS_MASK
>  
> +#define PRENR_MASK  GENMASK(31, 0)
> +
>  #ifndef __ASSEMBLY__
>  
>  /*
> diff --git a/xen/arch/arm/include/asm/mpu/cpregs.h 
> b/xen/arch/arm/include/asm/mpu/cpregs.h
> index bb15e02df6..9f3b32acd7 100644
> --- a/xen/arch/arm/include/asm/mpu/cpregs.h
> +++ b/xen/arch/arm/include/asm/mpu/cpregs.h
> @@ -6,6 +6,9 @@
>  /* CP15 CR0: MPU Type Register */
>  #define HMPUIR          p15,4,c0,c0,4
>  
> +/* CP15 CR6: Protection Region Enable Register */
> +#define HPRENR          p15,4,c6,c1,1
> +
>  /* CP15 CR6: MPU Protection Region Base/Limit/Select Address Register */
>  #define HPRSELR         p15,4,c6,c2,1
>  #define HPRBAR          p15,4,c6,c3,0
> @@ -82,6 +85,7 @@
>  /* Alphabetically... */
>  #define MPUIR_EL2       HMPUIR
>  #define PRBAR_EL2       HPRBAR
> +#define PRENR_EL2       HPRENR
>  #define PRLAR_EL2       HPRLAR
>  #define PRSELR_EL2      HPRSELR
>  #endif /* CONFIG_ARM_32 */
> diff --git a/xen/arch/arm/mpu/mm.c b/xen/arch/arm/mpu/mm.c
> index dd54b66901..2e88c467d5 100644
> --- a/xen/arch/arm/mpu/mm.c
> +++ b/xen/arch/arm/mpu/mm.c
> @@ -185,6 +185,42 @@ static int xen_mpumap_alloc_entry(uint8_t *idx)
>      return 0;
>  }
>  
> +/*
> + * Disable and remove an MPU region from the data structure and MPU 
> registers.
> + *
> + * @param index Index of the MPU region to be disabled.
> + */
> +static void disable_mpu_region_from_index(uint8_t index)
> +{
> +    ASSERT(spin_is_locked(&xen_mpumap_lock));
> +    ASSERT(index != INVALID_REGION_IDX);
> +
> +    if ( !region_is_valid(&xen_mpumap[index]) )
> +    {
> +        printk(XENLOG_WARNING
> +               "mpu: MPU memory region[%u] is already disabled\n", index);
> +        return;
> +    }
> +
> +    /* Zeroing the region will also zero the region enable */
> +    memset(&xen_mpumap[index], 0, sizeof(pr_t));
Is it ok that for a fast case (i.e. 0-31) our representation of prbar/prlar will
be different from the HW i.e. xen_mpumap[index] is 0 vs only .en bit of prlar
being 0 in HW?

> +    clear_bit(index, xen_mpumap_mask);
> +
> +    /*
> +     * Both Armv8-R AArch64 and AArch32 have direct access to the enable bit 
> for
> +     * MPU regions numbered from 0 to 31.
> +     */
> +    if ( (index & PRENR_MASK) != 0 )
> +    {
> +        /* Clear respective bit */
> +        uint64_t val = READ_SYSREG(PRENR_EL2) & (~(1UL << index));
On AArch32 the register is 32bit, so I think you should use register_t type.

> +
> +        WRITE_SYSREG(val, PRENR_EL2);
> +    }
> +    else
> +        write_protection_region(&xen_mpumap[index], index);
> +}
> +
>  /*
>   * Update the entry in the MPU memory region mapping table (xen_mpumap) for 
> the
>   * given memory range and flags, creating one if none exists.
> @@ -203,11 +239,11 @@ static int xen_mpumap_update_entry(paddr_t base, 
> paddr_t limit,
>      ASSERT(spin_is_locked(&xen_mpumap_lock));
>  
>      rc = mpumap_contains_region(xen_mpumap, max_mpu_regions, base, limit, 
> &idx);
> -    if ( !(rc == MPUMAP_REGION_NOTFOUND) )
> +    if ( rc < 0 )
>          return -EINVAL;
>  
>      /* We are inserting a mapping => Create new region. */
> -    if ( flags & _PAGE_PRESENT )
> +    if ( (flags & _PAGE_PRESENT) && (MPUMAP_REGION_NOTFOUND == rc) )
I think we need more sanity checking. What if flags has _PAGE_PRESENT but rc !=
MPUMAP_REGION_NOTFOUND, e.g. function called to modify existing entry? You will
silently return success. Maybe a similar function as for MMU is needed to
perform some sanity checks depending on the reason of the call?

>      {
>          rc = xen_mpumap_alloc_entry(&idx);
>          if ( rc )
> @@ -218,6 +254,20 @@ static int xen_mpumap_update_entry(paddr_t base, paddr_t 
> limit,
>          write_protection_region(&xen_mpumap[idx], idx);
>      }
>  
> +    /*
> +     * Currently, we only support destroying a *WHOLE* MPU memory region.
> +     * Part-region removal is not supported as in the worst case it will 
> leave
> +     * two fragments behind.
> +     */
> +    if ( rc == MPUMAP_REGION_INCLUSIVE )
> +    {
> +        printk("mpu: part-region removal is not supported\n");
You mention removal but why do you limit this place to removal only? You don't
have any checks making sure that flags is 0 at this point.

> +        return -EINVAL;
> +    }
> +
> +    if ( !(flags & _PAGE_PRESENT) && (rc >= MPUMAP_REGION_FOUND) )
> +        disable_mpu_region_from_index(idx);
> +
>      return 0;
>  }
>  
> @@ -251,6 +301,21 @@ int xen_mpumap_update(paddr_t base, paddr_t limit, 
> unsigned int flags)
>      return rc;
>  }
>  
> +int destroy_xen_mappings(unsigned long s, unsigned long e)
> +{
> +    int rc;
> +
> +    ASSERT(IS_ALIGNED(s, PAGE_SIZE));
> +    ASSERT(IS_ALIGNED(e, PAGE_SIZE));
> +    ASSERT(s <= e);
> +
> +    rc = xen_mpumap_update(virt_to_maddr(s), virt_to_maddr(e), 0);
> +    if ( !rc )
> +        context_sync_mpu();
> +
> +    return rc;
> +}
> +
>  int map_pages_to_xen(unsigned long virt, mfn_t mfn, unsigned long nr_mfns,
>                       unsigned int flags)
>  {

~Michal