Fourth MPU series
cover Fourth MPU series
This series aims to further the ongoing work to introduce support for MPU systems in xen.
The patches in this series implement various memory functions and enable the hypervisor timer.
Luca Fancellu (3):
arm/mpu: Implement copy_from_paddr for MPU systems arm/mpu: Implement vmap functions for MPU arm/mpu: Introduce modify_after_init_mappings
Penny Zheng (3):
arm/mpu: Implement free_init_memory for MPU systems
arm: Use secure hypervisor timer in MPU system arm/mpu: Map domain page in AArch64 MPU systems
xen/arch/arm/Kconfig | 6 + xen/arch/arm/include/asm/arm64/sysregs.h | 12 ++ xen/arch/arm/include/asm/mpu/mm.h | 11 ++ xen/arch/arm/include/asm/setup.h | 5 + xen/arch/arm/mmu/setup.c | 15 +++ xen/arch/arm/mpu/Makefile | 1 + xen/arch/arm/mpu/domain-page.c | 53 ++++++++ xen/arch/arm/mpu/mm.c | 160 ++++++++++++++++++----- xen/arch/arm/mpu/setup.c | 54 +++++++- xen/arch/arm/mpu/vmap.c | 14 +- xen/arch/arm/setup.c | 15 +-- 11 files changed, 296 insertions(+), 50 deletions(-) create mode 100644 xen/arch/arm/mpu/domain-page.c
--
2.43.0
PATCH 1/6 arm/mpu: Implement copy_from_paddr for MPU systems
From: Luca Fancellu <[email protected]>
Implement the function copy_from_paddr variant for MPU systems, using the map_pages_to_xen/destroy_xen_mappings to temporarily map the memory range to be copied.
--- xen/arch/arm/mpu/setup.c | 14 +++++++++++++- 1 file changed, 13 insertions(+), 1 deletion(-)
xen/arch/arm/mpu/setup.c +13 -1
@@ -91,7 +91,19 @@ void * __init early_fdt_map(paddr_t fdt_paddr)*/ void __init copy_from_paddr (void *dst,paddr_t paddr,unsigned long len) { -BUG_ON ("unimplemented" ); +paddr_t start_pg =round_pgdown (paddr); +paddr_t end_pg =round_pgup (paddr + len); +unsigned long nr_mfns = (end_pg - start_pg) >>PAGE_SHIFT ; +mfn_t mfn =maddr_to_mfn (start_pg); + +if (map_pages_to_xen (start_pg, mfn, nr_mfns,PAGE_HYPERVISOR_WC ) ) +panic ("Unable to map range for copy_from_paddr\n" ); + +memcpy (dst,maddr_to_virt (paddr), len); +clean_dcache_va_range (dst, len); + +if (destroy_xen_mappings (start_pg, end_pg) ) +panic ("Unable to unmap range for copy_from_paddr\n" ); }void __init remove_early_mappings (void )
Signature · 1 line
2.43.0
Re: [PATCH 1/6] arm/mpu: Implement copy_from_paddr for MPU systems
Quoted 9 lines · Harry Ramsey
From: Luca Fancellu <[email protected]>
Implement the function
copy_from_paddrvariant for MPU systems, using the map_pages_to_xen/destroy_xen_mappings to temporarily map the memory range to be copied.Signed-off-by Luca Fancellu <[email protected]>Signed-off-by Hari Limaye <[email protected]>Signed-off-by Harry Ramsey <[email protected]>
~Michal
PATCH 2/6 arm/mpu: Implement vmap functions for MPU
In reply to Harry RamseyFrom: Luca Fancellu <[email protected]>
HAS_VMAP is not enabled on MPU systems, but the vmap functions are used in places across common code. In order to keep the existing code and maintain correct functionality, implement the vmap_contig and vunmap functions for MPU systems.
Introduce a helper function destroy_entire_xen_mapping to aid with unmapping an entire region when only the start address is known.
--- xen/arch/arm/include/asm/mpu/mm.h | 11 +++++ xen/arch/arm/mpu/mm.c | 67 +++++++++++++++++++++++++------ xen/arch/arm/mpu/vmap.c | 14 +++++-- 3 files changed, 77 insertions(+), 15 deletions(-)
xen/arch/arm/include/asm/mpu/mm.h +11 -0
@@ -111,6 +111,17 @@ pr_t pr_of_addr(paddr_t base, paddr_t limit, unsigned int flags);int mpumap_contains_region (pr_t *table,uint8_t nr_regions,paddr_t base,paddr_t limit,uint8_t *index); + +/* +* Destroys and frees (if reference count is 0) an entire xen mapping on MPU +* systems where only the start address is known. +* +* @param s Start address of memory region to be destroyed. +* +* @return: 0 on success, negative on error. +*/ +int destroy_entire_xen_mapping (paddr_t s); +#endif /* __ARM_MPU_MM_H__ */ /*
xen/arch/arm/mpu/mm.c +55 -12
85 lines not on this page · raw message
xen/arch/arm/mpu/vmap.c +11 -3
@@ -1,19 +1,27 @@/* SPDX-License-Identifier: GPL-2.0-only */ #include <xen/bug.h> +#include <xen/mm.h> #include <xen/mm-frame.h> #include <xen/types.h> #include <xen/vmap.h> void *vmap_contig (mfn_t mfn,unsigned int nr) { -BUG_ON ("unimplemented" ); -return NULL ; +paddr_t base =mfn_to_maddr (mfn); + +if (map_pages_to_xen (base, mfn, nr,PAGE_HYPERVISOR ) ) +return NULL ; + +return maddr_to_virt (base); }void vunmap (const void *va) { -BUG_ON ("unimplemented" ); +paddr_t base =virt_to_maddr (va); + +if (destroy_entire_xen_mapping (base) ) +panic ("Failed to vunmap region\n" ); }/*
Signature · 1 line
2.43.0
Re: [PATCH 2/6] arm/mpu: Implement vmap functions for MPU
Quoted 3 lines · Harry Ramsey
From: Luca Fancellu <[email protected]>
HAS_VMAPis not enabled on MPU systems, but the vmap functions are used
Just as a reminder, we don't intend to support VMAP on MPU? Asking because it would otherwise be a duplicate effort to implement only these two helpers.
Quoted 23 lines
in places across common code. In order to keep the existing code and maintain correct functionality, implement the
vmap_contigandvunmapfunctions for MPU systems.Introduce a helper function
destroy_entire_xen_mappingto aid with unmapping an entire region when only the start address is known.Signed-off-by Luca Fancellu <[email protected]>Signed-off-by Harry Ramsey <[email protected]>--- xen/arch/arm/include/asm/mpu/mm.h | 11 +++++ xen/arch/arm/mpu/mm.c | 67 +++++++++++++++++++++++++------ xen/arch/arm/mpu/vmap.c | 14 +++++-- 3 files changed, 77 insertions(+), 15 deletions(-)xen/arch/arm/include/asm/mpu/mm.h +0
-0@@ -111,6 +111,17 @@ pr_t pr_of_addr(paddr_t base, paddr_t limit, unsigned int flags);int mpumap_contains_region (pr_t *table,uint8_t nr_regions,paddr_t base,paddr_t limit,uint8_t *index);
xen/arch/arm/include/asm/mpu/mm.h pr_t pr_of_addr(paddr_t base, paddr_t limit, unsigned int flags); · line 123
Quoted 10 lines
xen/arch/arm/include/asm/mpu/mm.h +10
-0+ +/* +* Destroys and frees (if reference count is 0) an entire xen mapping on MPU +* systems where only the start address is known. +* +* @param s Start address of memory region to be destroyed. +* +* @return: 0 on success, negative on error. +*/ +int destroy_entire_xen_mapping (paddr_t s);
NIT: I read it as all the mappings which is a bit misleading :) Maybe something like: destroy_mapping_containing or alike?
Quoted 9 lines
xen/arch/arm/include/asm/mpu/mm.h +1
-0+#endif /* __ARM_MPU_MM_H__ */ /* xen/arch/arm/mpu/mm.c +0
-0
xen/arch/arm/mpu/mm.c static void disable_mpu_region_from_index(uint8_t index) · line 298
Quoted 9 lines
xen/arch/arm/mpu/mm.c +6
-0@@ -290,6 +290,35 @@ static void disable_mpu_region_from_index(uint8_t index)write_protection_region (&xen_mpumap[index], index); } +/* +* Free a xen_mpumap entry given the index. A mpu region is actually disabled +* when the refcount is 0 and the region type is MPUMAP_REGION_FOUND. +* +* @param idx Index of the mpumap entry. +* @param region_found_type Either MPUMAP_REGION_FOUND or MPUMAP_REGION_INCLUSIVE.
Both of these are unsigned, so why the parameter is int?
Quoted 9 lines
xen/arch/arm/mpu/mm.c +9
-0+* @return 0 on success, otherwise negative on error. +*/ +static int xen_mpumap_free_entry (uint8_t idx,int region_found_type) +{ +ASSERT (spin_is_locked (&xen_mpumap_lock)); +ASSERT (idx !=INVALID_REGION_IDX ); + +if ( xen_mpumap[idx].refcount ==0 ) + {
xen/arch/arm/mpu/mm.c static void disable_mpu_region_from_index(uint8_t index) · line 317
Quoted 10 lines
xen/arch/arm/mpu/mm.c +10
-0+if (MPUMAP_REGION_FOUND == region_found_type ) +disable_mpu_region_from_index (idx); +else + { +printk (XENLOG_ERR "Cannot remove a partial region\n" ); +return -EINVAL ; + } + } +else + xen_mpumap[idx].refcount -=1 ;
You could avoid nesting if/else by doing:
if ( xen_mpumap[idx].refcount )
{
xen_mpumap[idx].refcount -= 1;
return 0;
}
if ( MPUMAP_REGION_FOUND == region_found_type )
disable_mpu_region_from_index(idx);
else
{
printk(XENLOG_ERR "Cannot remove a partial region\n");
return -EINVAL;
}Quoted 32 lines
xen/arch/arm/mpu/mm.c +6
-12+ +return 0 ; +} +/* * Update the entry in the MPU memory region mapping table (xen_mpumap) for the * given memory range and flags, creating one if none exists. @@ -357,18 +386,7 @@ static int xen_mpumap_update_entry(paddr_t base, paddr_t limit,return -EINVAL ; } -if ( xen_mpumap[idx].refcount ==0 ) - { -if (MPUMAP_REGION_FOUND == rc ) -disable_mpu_region_from_index (idx); -else - { -printk ("Cannot remove a partial region\n" ); -return -EINVAL ; - } - } -else - xen_mpumap[idx].refcount -=1 ; +return xen_mpumap_free_entry (idx, rc); }return 0 ;@@ -418,6 +436,31 @@ int destroy_xen_mappings(unsigned long s, unsigned long e)return xen_mpumap_update (s, e,0 ); } +int destroy_entire_xen_mapping (paddr_t s)
xen/arch/arm/mpu/mm.c int destroy_xen_mappings(unsigned long s, unsigned long e) · line 449
Quoted 10 lines
xen/arch/arm/mpu/mm.c +10
-0+{ +int rc; +uint8_t idx; + +ASSERT (IS_ALIGNED (s,PAGE_SIZE )); + +spin_lock (&xen_mpumap_lock); + + rc =mpumap_contains_region (xen_mpumap, max_mpu_regions, s, s +PAGE_SIZE , + &idx);
So here you are searching for a region that includes at least s + PAGE_SIZE.
xen/arch/arm/mpu/mm.c int destroy_xen_mappings(unsigned long s, unsigned long e) · line 453
Quoted 4 lines
xen/arch/arm/mpu/mm.c +4
-0+if ( rc ==MPUMAP_REGION_NOTFOUND ) + { +printk (XENLOG_ERR "Cannot remove entry that does not exist" ); + rc = -EINVAL ;
Here you assing rc but ...
xen/arch/arm/mpu/mm.c int destroy_xen_mappings(unsigned long s, unsigned long e) · line 457
Quoted 4 lines
xen/arch/arm/mpu/mm.c +4
-0+ } + +/* As we are unmapping entire region use MPUMAP_REGION_FOUND instead */ + rc =xen_mpumap_free_entry (idx,MPUMAP_REGION_FOUND );
here you would redefine it.
Quoted 31 lines
xen/arch/arm/mpu/mm.c +6
-0+ +spin_unlock (&xen_mpumap_lock); + +return rc; +} +int map_pages_to_xen (unsigned long virt,mfn_t mfn,unsigned long nr_mfns,unsigned int flags) {xen/arch/arm/mpu/vmap.c +6
-2@@ -1,19 +1,27 @@/* SPDX-License-Identifier: GPL-2.0-only */ #include <xen/bug.h> +#include <xen/mm.h> #include <xen/mm-frame.h> #include <xen/types.h> #include <xen/vmap.h> void *vmap_contig (mfn_t mfn,unsigned int nr) { -BUG_ON ("unimplemented" ); -return NULL ; +paddr_t base =mfn_to_maddr (mfn); + +if (map_pages_to_xen (base, mfn, nr,PAGE_HYPERVISOR ) ) +return NULL ; +
xen/arch/arm/mpu/vmap.c line 24
Quoted 10 lines
xen/arch/arm/mpu/vmap.c +5
-1+return maddr_to_virt (base); }void vunmap (const void *va) { -BUG_ON ("unimplemented" ); +paddr_t base =virt_to_maddr (va); + +if (destroy_entire_xen_mapping (base) ) +panic ("Failed to vunmap region\n" );
Looking at common vunmap() we ignore the return code from destroy_xen_mappings(). Is it ok to diverge?
~Michal
Quoted 3 lines
xen/arch/arm/mpu/vmap.c +0
-0}/*
Re: [PATCH 2/6] arm/mpu: Implement vmap functions for MPU
Hi Michal,
thanks for your review, I’ll answer only few of your points and I’ll let Harry take the rest.
Quoted 10 lines · Michal
Quoted 3 lines · Harry Ramsey
From: Luca Fancellu <[email protected]>
HAS_VMAPis not enabled on MPU systems, but the vmap functions are usedJust as a reminder, we don't intend to support VMAP on MPU? Asking because it would otherwise be a duplicate effort to implement only these two helpers.
Yes we are not going to support VMAP on MPU, here we want only to provide the implementation of these two helper so that we don’t touch the common code using these. Are you suggesting some rewording or was it only a curiosity on your side?
Quoted 8 lines
Quoted 7 lines
+/*
+ * Free axen_mpumapentry given the index. A mpu region is actually disabled
+ * when the refcount is 0 and the region type isMPUMAP_REGION_FOUND.
+ *
+ * @param idx Index of the mpumap entry.
+ * @paramregion_found_typeEitherMPUMAP_REGION_FOUNDorMPUMAP_REGION_INCLUSIVE.Both of these are unsigned, so why the parameter is int?
Uhm, good catch, I think this is a documentation issue because it might be also MPUMAP_REGION_OVERLAP which is -1, we should not mandate which value might be here, we should only say MPUMAP_REGION_* values.
Quoted 11 lines
xen/arch/arm/mpu/vmap.c line 24
Quoted 9 lines
void
vunmap(const void *va)
{xen/arch/arm/mpu/vmap.c +4
-1-BUG_ON ("unimplemented" ); +paddr_t base =virt_to_maddr (va); + +if (destroy_entire_xen_mapping (base) ) +panic ("Failed to vunmap region\n" );Looking at common
vunmap()we ignore the return code fromdestroy_xen_mappings(). Is it ok to diverge?
In our tests it’s ok, I asked Harry to have this so that we can catch any vunmap that is not balanced with a prior mapping. To be fair I’m not really sure why in the vmap.c implementation we are not reading the error codes from destroy_xen_mappings.
Cheers,
Luca
Re: [PATCH 2/6] arm/mpu: Implement vmap functions for MPU
Quoted 11 lines · Luca Fancellu
Hi Michal,
thanks for your review, I’ll answer only few of your points and I’ll let Harry take the rest.
Quoted 7 lines · Michal
Quoted 2 lines · Harry Ramsey
From: Luca Fancellu <[email protected]>
Quoted 7 lines
Quoted 3 lines
Quoted 1 line
HAS_VMAPis not enabled on MPU systems, but the vmap functions are usedJust as a reminder, we don't intend to support VMAP on MPU? Asking because it would otherwise be a duplicate effort to implement only these two helpers.
Yes we are not going to support VMAP on MPU, here we want only to provide the implementation of these two helper so that we don’t touch the common code using these. Are you suggesting some rewording or was it only a curiosity on your side?
It was a question to check whether we are aligned.
Quoted 23 lines
Quoted 9 lines
Quoted 8 lines
+/*
+ * Free axen_mpumapentry given the index. A mpu region is actually disabled
+ * when the refcount is 0 and the region type isMPUMAP_REGION_FOUND.
+ *
+ * @param idx Index of the mpumap entry.
+ * @paramregion_found_typeEitherMPUMAP_REGION_FOUNDorMPUMAP_REGION_INCLUSIVE.Both of these are unsigned, so why the parameter is int?
Uhm, good catch, I think this is a documentation issue because it might be also
MPUMAP_REGION_OVERLAPwhich is -1, we should not mandate which value might be here, we should only say MPUMAP_REGION_* values.Quoted 9 lines
Quoted 9 lines
void
vunmap(const void *va)
{xen/arch/arm/mpu/vmap.c +4
-1-BUG_ON ("unimplemented" ); +paddr_t base =virt_to_maddr (va); + +if (destroy_entire_xen_mapping (base) ) +panic ("Failed to vunmap region\n" );
Quoted 6 lines
Quoted 2 lines
Looking at common
vunmap()we ignore the return code fromdestroy_xen_mappings(). Is it ok to diverge?In our tests it’s ok, I asked Harry to have this so that we can catch any vunmap that is not balanced with a prior mapping. To be fair I’m not really sure why in the
vmap.cimplementation we are not reading the error codes fromdestroy_xen_mappings.
I'm ok with that. I don't know why we don't check the rc there. If you look at x86 destroy_xen_mappings calls, none seem to be checked against rc. It can be that it's impossible scenario there.
~Michal
Quoted 4 lines
Cheers,
Luca
Re: [PATCH 2/6] arm/mpu: Implement vmap functions for MPU
In reply to Orzel, MichalQuoted 146 lines · Michal
Quoted 3 lines · Harry Ramsey
From: Luca Fancellu <[email protected]>
HAS_VMAPis not enabled on MPU systems, but the vmap functions are usedJust as a reminder, we don't intend to support VMAP on MPU? Asking because it would otherwise be a duplicate effort to implement only these two helpers.
Quoted 23 lines
in places across common code. In order to keep the existing code and maintain correct functionality, implement the
vmap_contigandvunmapfunctions for MPU systems.Introduce a helper function
destroy_entire_xen_mappingto aid with unmapping an entire region when only the start address is known.Signed-off-by Luca Fancellu <[email protected]>Signed-off-by Harry Ramsey <[email protected]>--- xen/arch/arm/include/asm/mpu/mm.h | 11 +++++ xen/arch/arm/mpu/mm.c | 67 +++++++++++++++++++++++++------ xen/arch/arm/mpu/vmap.c | 14 +++++-- 3 files changed, 77 insertions(+), 15 deletions(-)xen/arch/arm/include/asm/mpu/mm.h +0
-0@@ -111,6 +111,17 @@ pr_t pr_of_addr(paddr_t base, paddr_t limit, unsigned int flags);int mpumap_contains_region (pr_t *table,uint8_t nr_regions,paddr_t base,paddr_t limit,uint8_t *index);xen/arch/arm/include/asm/mpu/mm.h pr_t pr_of_addr(paddr_t base, paddr_t limit, unsigned int flags); · line 123
Quoted 10 lines
xen/arch/arm/include/asm/mpu/mm.h +10
-0+ +/* +* Destroys and frees (if reference count is 0) an entire xen mapping on MPU +* systems where only the start address is known. +* +* @param s Start address of memory region to be destroyed. +* +* @return: 0 on success, negative on error. +*/ +int destroy_entire_xen_mapping (paddr_t s);NIT: I read it as all the mappings which is a bit misleading :) Maybe something like:
destroy_mapping_containingor alike?Quoted 9 lines
xen/arch/arm/include/asm/mpu/mm.h +1
-0+#endif /* __ARM_MPU_MM_H__ */ /* xen/arch/arm/mpu/mm.c +0
-0xen/arch/arm/mpu/mm.c static void disable_mpu_region_from_index(uint8_t index) · line 298
Quoted 9 lines
xen/arch/arm/mpu/mm.c +6
-0@@ -290,6 +290,35 @@ static void disable_mpu_region_from_index(uint8_t index)write_protection_region (&xen_mpumap[index], index); } +/* +* Free a xen_mpumap entry given the index. A mpu region is actually disabled +* when the refcount is 0 and the region type is MPUMAP_REGION_FOUND. +* +* @param idx Index of the mpumap entry. +* @param region_found_type Either MPUMAP_REGION_FOUND or MPUMAP_REGION_INCLUSIVE. Both of these are unsigned, so why the parameter is int?
Quoted 9 lines
xen/arch/arm/mpu/mm.c +9
-0+* @return 0 on success, otherwise negative on error. +*/ +static int xen_mpumap_free_entry (uint8_t idx,int region_found_type) +{ +ASSERT (spin_is_locked (&xen_mpumap_lock)); +ASSERT (idx !=INVALID_REGION_IDX ); + +if ( xen_mpumap[idx].refcount ==0 ) + {xen/arch/arm/mpu/mm.c static void disable_mpu_region_from_index(uint8_t index) · line 317
Quoted 10 lines
xen/arch/arm/mpu/mm.c +10
-0+if (MPUMAP_REGION_FOUND == region_found_type ) +disable_mpu_region_from_index (idx); +else + { +printk (XENLOG_ERR "Cannot remove a partial region\n" ); +return -EINVAL ; + } + } +else + xen_mpumap[idx].refcount -=1 ;You could avoid nesting if/else by doing:
if ( xen_mpumap[idx].refcount ) { xen_mpumap[idx].refcount -= 1; return 0; } if ( MPUMAP_REGION_FOUND == region_found_type ) disable_mpu_region_from_index(idx); else { printk(XENLOG_ERR "Cannot remove a partial region\n"); return -EINVAL; }Quoted 32 lines
xen/arch/arm/mpu/mm.c +6
-12+ +return 0 ; +} +/* * Update the entry in the MPU memory region mapping table (xen_mpumap) for the * given memory range and flags, creating one if none exists. @@ -357,18 +386,7 @@ static int xen_mpumap_update_entry(paddr_t base, paddr_t limit,return -EINVAL ; } -if ( xen_mpumap[idx].refcount ==0 ) - { -if (MPUMAP_REGION_FOUND == rc ) -disable_mpu_region_from_index (idx); -else - { -printk ("Cannot remove a partial region\n" ); -return -EINVAL ; - } - } -else - xen_mpumap[idx].refcount -=1 ; +return xen_mpumap_free_entry (idx, rc); }return 0 ;@@ -418,6 +436,31 @@ int destroy_xen_mappings(unsigned long s, unsigned long e)return xen_mpumap_update (s, e,0 ); } +int destroy_entire_xen_mapping (paddr_t s)xen/arch/arm/mpu/mm.c int destroy_xen_mappings(unsigned long s, unsigned long e) · line 449
Quoted 10 lines
xen/arch/arm/mpu/mm.c +10
-0+{ +int rc; +uint8_t idx; + +ASSERT (IS_ALIGNED (s,PAGE_SIZE )); + +spin_lock (&xen_mpumap_lock); + + rc =mpumap_contains_region (xen_mpumap, max_mpu_regions, s, s +PAGE_SIZE , + &idx);So here you are searching for a region that includes at least s +
PAGE_SIZE.xen/arch/arm/mpu/mm.c int destroy_xen_mappings(unsigned long s, unsigned long e) · line 453
Quoted 4 lines
xen/arch/arm/mpu/mm.c +4
-0+if ( rc ==MPUMAP_REGION_NOTFOUND ) + { +printk (XENLOG_ERR "Cannot remove entry that does not exist" ); + rc = -EINVAL ;
Quoted 7 lines
Here you assing rc but ...
xen/arch/arm/mpu/mm.c int destroy_xen_mappings(unsigned long s, unsigned long e) · line 457
Quoted 4 lines
xen/arch/arm/mpu/mm.c +4
-0+ } + +/* As we are unmapping entire region use MPUMAP_REGION_FOUND instead */ + rc =xen_mpumap_free_entry (idx,MPUMAP_REGION_FOUND );here you would redefine it.
Sorry, this is a mistake and we should just return early as the mapping doesn't exist and therefore will have no references to be decremented.
Quoted 44 lines
Quoted 31 lines
xen/arch/arm/mpu/mm.c +6
-0+ +spin_unlock (&xen_mpumap_lock); + +return rc; +} +int map_pages_to_xen (unsigned long virt,mfn_t mfn,unsigned long nr_mfns,unsigned int flags) {xen/arch/arm/mpu/vmap.c +6
-2@@ -1,19 +1,27 @@/* SPDX-License-Identifier: GPL-2.0-only */ #include <xen/bug.h> +#include <xen/mm.h> #include <xen/mm-frame.h> #include <xen/types.h> #include <xen/vmap.h> void *vmap_contig (mfn_t mfn,unsigned int nr) { -BUG_ON ("unimplemented" ); -return NULL ; +paddr_t base =mfn_to_maddr (mfn); + +if (map_pages_to_xen (base, mfn, nr,PAGE_HYPERVISOR ) ) +return NULL ; +xen/arch/arm/mpu/vmap.c line 24
Quoted 10 lines
xen/arch/arm/mpu/vmap.c +5
-1+return maddr_to_virt (base); }void vunmap (const void *va) { -BUG_ON ("unimplemented" ); +paddr_t base =virt_to_maddr (va); + +if (destroy_entire_xen_mapping (base) ) +panic ("Failed to vunmap region\n" );Looking at common
vunmap()we ignore the return code fromdestroy_xen_mappings(). Is it ok to diverge?
Quoted 5 lines
~Michal
Quoted 3 lines
xen/arch/arm/mpu/vmap.c +0
-0}/*
I will address the code style comments in the next version of these patches.
Thanks,
Harry Ramsey
PATCH 3/6 arm/mpu: Implement free_init_memory for MPU systems
In reply to Harry RamseyFrom: Penny Zheng <[email protected]>
Implement the function free_init_memory for MPU systems. In order to support this, the function modify_xen_mappings is implemented.
On MPU systems, we map the init text and init data sections using separate MPU memory regions. Therefore these are removed separately in free_init_memory.
Additionally remove warning messages from is_mm_attr_match as some permissions can now be updated by xen_mpumap_update_entry.
--- xen/arch/arm/include/asm/setup.h | 2 + xen/arch/arm/mpu/mm.c | 91 +++++++++++++++++++++++++------- 2 files changed, 73 insertions(+), 20 deletions(-)
xen/arch/arm/include/asm/setup.h +2 -0
@@ -65,6 +65,8 @@ int map_irq_to_domain(struct domain *d, unsigned int irq,int map_range_to_domain (const struct dt_device_node *dev,uint64_t addr,uint64_t len,void *data); +extern const char __init_data_begin[], __bss_start[], __bss_end[]; +struct init_info {/* Pointer to the stack, used by head.S when entering in C */
xen/arch/arm/mpu/mm.c +71 -20
122 lines not on this page · raw message
Signature · 1 line
2.43.0
Re: [PATCH 3/6] arm/mpu: Implement free_init_memory for MPU systems
Quoted 49 lines · Harry Ramsey
From: Penny Zheng <[email protected]>
Implement the function
free_init_memoryfor MPU systems. In order to support this, the functionmodify_xen_mappingsis implemented.On MPU systems, we map the init text and init data sections using separate MPU memory regions. Therefore these are removed separately in
free_init_memory.Additionally remove warning messages from
is_mm_attr_matchas some permissions can now be updated byxen_mpumap_update_entry.Signed-off-by Penny Zheng <[email protected]>Signed-off-by Wei Chen <[email protected]>Signed-off-by Luca Fancellu <[email protected]>Signed-off-by Hari Limaye <[email protected]>Signed-off-by Harry Ramsey <[email protected]>--- xen/arch/arm/include/asm/setup.h | 2 + xen/arch/arm/mpu/mm.c | 91 +++++++++++++++++++++++++------- 2 files changed, 73 insertions(+), 20 deletions(-)xen/arch/arm/include/asm/setup.h +2
-0@@ -65,6 +65,8 @@ int map_irq_to_domain(struct domain *d, unsigned int irq,int map_range_to_domain (const struct dt_device_node *dev,uint64_t addr,uint64_t len,void *data); +extern const char __init_data_begin[], __bss_start[], __bss_end[]; +struct init_info {/* Pointer to the stack, used by head.S when entering in C */ xen/arch/arm/mpu/mm.c +0
-5@@ -174,28 +174,13 @@ int mpumap_contains_region(pr_t *table, uint8_t nr_regions, paddr_t base,static bool is_mm_attr_match (pr_t *region,unsigned int attributes) {if ( region->prbar.reg.ro !=PAGE_RO_MASK (attributes) ) - { -printk (XENLOG_WARNING -"Mismatched Access Permission attributes (%#x instead of %#x)\n" , - region->prbar.reg.ro,PAGE_RO_MASK (attributes));return false; - }
xen/arch/arm/mpu/mm.c int mpumap_contains_region(pr_t *table, uint8_t nr_regions, paddr_t base, · line 182
Quoted 10 lines
xen/arch/arm/mpu/mm.c +0
-5if ( region->prbar.reg.xn !=PAGE_XN_MASK (attributes) ) - { -printk (XENLOG_WARNING -"Mismatched Execute Never attributes (%#x instead of %#x)\n" , - region->prbar.reg.xn,PAGE_XN_MASK (attributes));return false; - }if ( region->prlar.reg.ai !=PAGE_AI_MASK (attributes) )
This check and ...
Quoted 13 lines
xen/arch/arm/mpu/mm.c +0
-5- { -printk (XENLOG_WARNING -"Mismatched Memory Attribute Index (%#x instead of %#x)\n" , - region->prlar.reg.ai,PAGE_AI_MASK (attributes));return false; - }return true; }@@ -352,8 +337,33 @@ static int xen_mpumap_update_entry(paddr_t base, paddr_t limit, {if ( !is_mm_attr_match (&xen_mpumap[idx], flags) ) {
xen/arch/arm/mpu/mm.c static int xen_mpumap_update_entry(paddr_t base, paddr_t limit, · line 347
Quoted 10 lines
xen/arch/arm/mpu/mm.c +8
-2-printk ("Modifying an existing entry is not supported\n" ); -return -EINVAL ; +if ( rc ==MPUMAP_REGION_INCLUSIVE ) + { +printk (XENLOG_ERR +"Cannot modify partial region permissions\n" ); +return -EINVAL ; + } + +if ( xen_mpumap[idx].prlar.reg.ai !=PAGE_AI_MASK (flags) )
this one are identical. Why do we duplicate it here? If this is because is_mm_attr_match returns just bool, then maybe you want to introduce more logical return values and parse them here. What you eventually do here is you allow modifying regions for RO and XN. Therefore is_mm_attr_match could return true on RO and XN but false on AI.
xen/arch/arm/mpu/mm.c static int xen_mpumap_update_entry(paddr_t base, paddr_t limit, · line 357
Quoted 10 lines
xen/arch/arm/mpu/mm.c +10
-0+ { +printk (XENLOG_ERR +"Modifying memory attribute is not supported\n" ); +return -EINVAL ; + } + +if ( xen_mpumap[idx].refcount !=0 ) + { +printk (XENLOG_ERR +"Cannot modify memory permissions for a region mapped multiple time\n" );
s/time/times/
Quoted 23 lines
xen/arch/arm/mpu/mm.c +10
-2+return -EINVAL ; + } + +/* Set new permission */ + xen_mpumap[idx].prbar.reg.ro =PAGE_RO_MASK (flags); + xen_mpumap[idx].prbar.reg.xn =PAGE_XN_MASK (flags); + +write_protection_region (&xen_mpumap[idx], idx); +return 0 ; }/* Check for overflow of refcount before incrementing. */ @@ -499,8 +509,7 @@ void __init setup_mm_helper(void)int modify_xen_mappings (unsigned long s,unsigned long e,unsigned int nf) { -BUG_ON ("unimplemented" ); -return -EINVAL ; +return xen_mpumap_update (s, e, nf); }void dump_hyp_walk (vaddr_t addr)
xen/arch/arm/mpu/mm.c void dump_hyp_walk(vaddr_t addr) · line 523
Quoted 5 lines
xen/arch/arm/mpu/mm.c +1
-1@@ -511,7 +520,49 @@ void dump_hyp_walk(vaddr_t addr)/* Release all __init and __initdata ranges to be reused */ void free_init_memory (void ) { -BUG_ON ("unimplemented" ); +unsigned long inittext_end =round_pgup ((unsigned long )__init_data_begin);
Looking at the linker script, __init_data_begin is already page aligned.
xen/arch/arm/mpu/mm.c void dump_hyp_walk(vaddr_t addr) · line 529
Quoted 6 lines
xen/arch/arm/mpu/mm.c +6
-0+unsigned long len = __init_end - __init_begin; +uint8_t idx; +int rc; + + rc =modify_xen_mappings ((unsigned long )__init_begin, inittext_end, +PAGE_HYPERVISOR_RW );
So here you modify mapping of text section but...
xen/arch/arm/mpu/mm.c void dump_hyp_walk(vaddr_t addr) · line 540
Quoted 11 lines
xen/arch/arm/mpu/mm.c +11
-0+if ( rc ) +panic ("Unable to map RW the init text section (rc = %d)\n" , rc); + +/* +* From now on, init will not be used for execution anymore, +* so nuke the instruction cache to remove entries related to init. +*/ +invalidate_icache_local (); + +/* Zeroing the memory before returning it */ +memset (__init_begin,0 , len);
here you zero the entire init. Is it because init.data is already RW, so you don't need to modify the mappings? If so, more informative comments would be welcome.
Quoted 18 lines
xen/arch/arm/mpu/mm.c +18
-0+ + rc =destroy_xen_mappings ((unsigned long )__init_begin, inittext_end); +if ( rc ) +panic ("Unable to remove init text section (rc = %d)\n" , rc); + +/* +* The initdata and bss sections are mapped using a single MPU region, so +* modify the start of this region to remove the initdata section. +*/ +spin_lock (&xen_mpumap_lock); + + rc =mpumap_contains_region (xen_mpumap, max_mpu_regions, + (unsigned long )__init_data_begin, + (unsigned long )__bss_end, + &idx); +if ( rc <MPUMAP_REGION_FOUND ) +panic ("Unable to find bss data section (rc = %d)\n" , rc); +
xen/arch/arm/mpu/mm.c void dump_hyp_walk(vaddr_t addr) · line 568
Quoted 10 lines
xen/arch/arm/mpu/mm.c +7
-0+/* bss data section is shrunk and now starts from __bss_start */ +pr_set_base (&xen_mpumap[idx], (unsigned long )__bss_start); + +write_protection_region (&xen_mpumap[idx], idx); +context_sync_mpu (); + +spin_unlock (&xen_mpumap_lock); }void __iomem *ioremap_attr (paddr_t start,size_t len,unsigned int flags)
~Michal
PATCH 4/6 arm/mpu: Introduce modify_after_init_mappings
In reply to Harry RamseyFrom: Luca Fancellu <[email protected]>
During init_done, Xen sets the permissions of all symbols marked with __ro_after_init to be read-only. Currently this is achieved by calling modify_xen_mappings and will shrink the RW mapping on one side and extend the RO mapping on the other.
This does not work on MPU systems at present because part-region modification is not supported. Therefore introduce the function modify_after_init_mappings for MMU and MPU, to handle the divergent approaches to setting permissions of __ro_after_init symbols.
As the new function is marked with __init, it needs to be called before free_init_memory.
--- xen/arch/arm/include/asm/setup.h | 3 +++ xen/arch/arm/mmu/setup.c | 15 ++++++++++++ xen/arch/arm/mpu/mm.c | 2 +- xen/arch/arm/mpu/setup.c | 40 ++++++++++++++++++++++++++++++++ xen/arch/arm/setup.c | 15 ++---------- 5 files changed, 61 insertions(+), 14 deletions(-)
xen/arch/arm/include/asm/setup.h +3 -0
@@ -78,6 +78,9 @@ struct init_infopaddr_t consider_modules (paddr_t s,paddr_t e,uint32_t size,paddr_t align,int first_mod); +/* Modify some mappings after the init is done */ +void modify_after_init_mappings (void ); +#endif /* * Local variables:
xen/arch/arm/mmu/setup.c +15 -0
@@ -213,6 +213,21 @@ void __init remove_early_mappings(void)BUG_ON (rc); } +void __init modify_after_init_mappings (void ) +{ +/* +* We have finished booting. Mark the section .data.ro_after_init +* read-only. +*/ +int rc =modify_xen_mappings ((unsigned long )&__ro_after_init_start, + (unsigned long )&__ro_after_init_end, +PAGE_HYPERVISOR_RO ); + +if ( rc ) +panic ("Unable to mark the .data.ro_after_init section read-only (rc = %d)\n" , + rc); +} +/* * After boot, Xen page-tables should not contain mapping that are both * Writable and eXecutables.
xen/arch/arm/mpu/mm.c +1 -1
@@ -32,7 +32,7 @@ DECLARE_BITMAP(xen_mpumap_mask, MAX_MPU_REGION_NR) \/* EL2 Xen MPU memory region mapping table. */ pr_t __cacheline_aligned __section(".data") xen_mpumap[MAX_MPU_REGION_NR]; -static DEFINE_SPINLOCK (xen_mpumap_lock); +DEFINE_SPINLOCK (xen_mpumap_lock);static void __init __maybe_unused build_assertions (void ) {
xen/arch/arm/mpu/setup.c +40 -0
@@ -8,11 +8,14 @@#include <xen/pfn.h> #include <xen/types.h> #include <xen/sizes.h> +#include <xen/spinlock.h> #include <asm/setup.h> static paddr_t __initdata mapped_fdt_base =INVALID_PADDR ;static paddr_t __initdata mapped_fdt_limit =INVALID_PADDR ; +extern spinlock_t xen_mpumap_lock; +void __init setup_pagetables (void ) {}void *__init early_fdt_map (paddr_t fdt_paddr)
@@ -106,6 +109,43 @@ void __init copy_from_paddr(void *dst, paddr_t paddr, unsigned long len)panic ("Unable to unmap range for copy_from_paddr\n" ); } +void __init modify_after_init_mappings (void ) +{ +int rc; +uint8_t idx_rodata; +uint8_t idx_rwdata; + +spin_lock (&xen_mpumap_lock); + + rc =mpumap_contains_region (xen_mpumap, max_mpu_regions, + (unsigned long )_srodata, + (unsigned long )_erodata, + &idx_rodata); + +if ( rc <MPUMAP_REGION_FOUND ) +panic ("Unable to find rodata section (rc = %d)\n" , rc); + + rc =mpumap_contains_region (xen_mpumap, max_mpu_regions, + (unsigned long )__ro_after_init_start, + (unsigned long )__init_begin, + &idx_rwdata); + +if ( rc <MPUMAP_REGION_FOUND ) +panic ("Unable to find rwdata section (rc = %d)\n" , rc); + +/* Shrink rwdata section to begin at __ro_after_init_end */ +pr_set_base (&xen_mpumap[idx_rwdata], (unsigned long )__ro_after_init_end); + +/* Extend rodata section to end at __ro_after_init_end */ +pr_set_limit (&xen_mpumap[idx_rodata], (unsigned long )__ro_after_init_end); + +write_protection_region (&xen_mpumap[idx_rwdata], idx_rwdata); +write_protection_region (&xen_mpumap[idx_rodata], idx_rodata); +context_sync_mpu (); + +spin_unlock (&xen_mpumap_lock); +} +void __init remove_early_mappings (void ) { int rc = destroy_xen_mappings(round_pgdown(mapped_fdt_base),
xen/arch/arm/setup.c +2 -13
@@ -66,23 +66,12 @@ domid_t __read_mostly max_init_domid;static __used void noreturninit_done (void ) { -int rc; -/* Must be done past setting system_state. */ unregister_init_virtual_region (); -free_init_memory (); +modify_after_init_mappings (); -/* -* We have finished booting. Mark the section .data.ro_after_init -* read-only. -*/ - rc =modify_xen_mappings ((unsigned long )&__ro_after_init_start, - (unsigned long )&__ro_after_init_end, -PAGE_HYPERVISOR_RO ); -if ( rc ) -panic ("Unable to mark the .data.ro_after_init section read-only (rc = %d)\n" , - rc); +free_init_memory ();startup_cpu_idle_loop (); }
Signature · 1 line
2.43.0
Re: [PATCH 4/6] arm/mpu: Introduce modify_after_init_mappings
Quoted 6 lines · Harry Ramsey
From: Luca Fancellu <[email protected]>
During
init_done, Xen sets the permissions of all symbols marked with__ro_after_initto be read-only. Currently this is achieved by callingmodify_xen_mappingsand will shrink the RW mapping on one side and extend the RO mapping on the other.
Can you be more specific about the sides you mention? How did you deduce it? I assume you are talking about MMU part.
Quoted 3 lines
This does not work on MPU systems at present because part-region modification is not supported. Therefore introduce the function
What else is in that region?
Wouldn't it be better to have one region for this __ro_after_init so that we don't need to shrink/extend the mappings? Is it done because of number of regions limitation?
~Michal
Quoted 163 lines
modify_after_init_mappingsfor MMU and MPU, to handle the divergent approaches to setting permissions of__ro_after_initsymbols.As the new function is marked with
__init, it needs to be called beforefree_init_memory.Signed-off-by Luca Fancellu <[email protected]>Signed-off-by Hari Limaye <[email protected]>Signed-off-by Harry Ramsey <[email protected]>--- xen/arch/arm/include/asm/setup.h | 3 +++ xen/arch/arm/mmu/setup.c | 15 ++++++++++++ xen/arch/arm/mpu/mm.c | 2 +- xen/arch/arm/mpu/setup.c | 40 ++++++++++++++++++++++++++++++++ xen/arch/arm/setup.c | 15 ++---------- 5 files changed, 61 insertions(+), 14 deletions(-)xen/arch/arm/include/asm/setup.h +3
-0@@ -78,6 +78,9 @@ struct init_infopaddr_t consider_modules (paddr_t s,paddr_t e,uint32_t size,paddr_t align,int first_mod); +/* Modify some mappings after the init is done */ +void modify_after_init_mappings (void ); +#endif /* * Local variables: xen/arch/arm/mmu/setup.c +15
-0@@ -213,6 +213,21 @@ void __init remove_early_mappings(void)BUG_ON (rc); } +void __init modify_after_init_mappings (void ) +{ +/* +* We have finished booting. Mark the section .data.ro_after_init +* read-only. +*/ +int rc =modify_xen_mappings ((unsigned long )&__ro_after_init_start, + (unsigned long )&__ro_after_init_end, +PAGE_HYPERVISOR_RO ); + +if ( rc ) +panic ("Unable to mark the .data.ro_after_init section read-only (rc = %d)\n" , + rc); +} +/* * After boot, Xen page-tables should not contain mapping that are both * Writable and eXecutables. xen/arch/arm/mpu/mm.c +1
-1@@ -32,7 +32,7 @@ DECLARE_BITMAP(xen_mpumap_mask, MAX_MPU_REGION_NR) \/* EL2 Xen MPU memory region mapping table. */ pr_t __cacheline_aligned __section(".data") xen_mpumap[MAX_MPU_REGION_NR]; -static DEFINE_SPINLOCK (xen_mpumap_lock); +DEFINE_SPINLOCK (xen_mpumap_lock);static void __init __maybe_unused build_assertions (void ) {xen/arch/arm/mpu/setup.c +40
-0@@ -8,11 +8,14 @@#include <xen/pfn.h> #include <xen/types.h> #include <xen/sizes.h> +#include <xen/spinlock.h> #include <asm/setup.h> static paddr_t __initdata mapped_fdt_base =INVALID_PADDR ;static paddr_t __initdata mapped_fdt_limit =INVALID_PADDR ; +extern spinlock_t xen_mpumap_lock; +void __init setup_pagetables (void ) {}void *__init early_fdt_map (paddr_t fdt_paddr)@@ -106,6 +109,43 @@ void __init copy_from_paddr(void *dst, paddr_t paddr, unsigned long len)panic ("Unable to unmap range for copy_from_paddr\n" ); } +void __init modify_after_init_mappings (void ) +{ +int rc; +uint8_t idx_rodata; +uint8_t idx_rwdata; + +spin_lock (&xen_mpumap_lock); + + rc =mpumap_contains_region (xen_mpumap, max_mpu_regions, + (unsigned long )_srodata, + (unsigned long )_erodata, + &idx_rodata); + +if ( rc <MPUMAP_REGION_FOUND ) +panic ("Unable to find rodata section (rc = %d)\n" , rc); + + rc =mpumap_contains_region (xen_mpumap, max_mpu_regions, + (unsigned long )__ro_after_init_start, + (unsigned long )__init_begin, + &idx_rwdata); + +if ( rc <MPUMAP_REGION_FOUND ) +panic ("Unable to find rwdata section (rc = %d)\n" , rc); + +/* Shrink rwdata section to begin at __ro_after_init_end */ +pr_set_base (&xen_mpumap[idx_rwdata], (unsigned long )__ro_after_init_end); + +/* Extend rodata section to end at __ro_after_init_end */ +pr_set_limit (&xen_mpumap[idx_rodata], (unsigned long )__ro_after_init_end); + +write_protection_region (&xen_mpumap[idx_rwdata], idx_rwdata); +write_protection_region (&xen_mpumap[idx_rodata], idx_rodata); +context_sync_mpu (); + +spin_unlock (&xen_mpumap_lock); +} +void __init remove_early_mappings (void ) { int rc = destroy_xen_mappings(round_pgdown(mapped_fdt_base),xen/arch/arm/setup.c +2
-13@@ -66,23 +66,12 @@ domid_t __read_mostly max_init_domid;static __used void noreturninit_done (void ) { -int rc; -/* Must be done past setting system_state. */ unregister_init_virtual_region (); -free_init_memory (); +modify_after_init_mappings (); -/* -* We have finished booting. Mark the section .data.ro_after_init -* read-only. -*/ - rc =modify_xen_mappings ((unsigned long )&__ro_after_init_start, - (unsigned long )&__ro_after_init_end, -PAGE_HYPERVISOR_RO ); -if ( rc ) -panic ("Unable to mark the .data.ro_after_init section read-only (rc = %d)\n" , - rc); +free_init_memory ();startup_cpu_idle_loop (); }
Re: [PATCH 4/6] arm/mpu: Introduce modify_after_init_mappings
Hi Michael,
Quoted 5 lines · Michal
On 28/11/2025 10:58, Harry Ramsey wrote:
Quoted 8 lines
Quoted 6 lines
From: Luca Fancellu <[email protected]>
During
init_done, Xen sets the permissions of all symbols marked with__ro_after_initto be read-only. Currently this is achieved by callingmodify_xen_mappingsand will shrink the RW mapping on one side and extend the RO mapping on the other.Can you be more specific about the sides you mention? How did you deduce it? I assume you are talking about MMU part.
You are right, this sentence is a bit misleading.
So what was written here was meant to say that on MPU modify_xen_mappings should shrink the RW mapping region and extend the RO mapping region because as of now they are declared like this in xen.lds.S:
read only data:
+------------------+ | _srodata | | _erodata | +-------------------+
RW data:
+---------------------------+ | __ro_after_init_start | | __ro_after_init_end | | __init_begin | +---------------------------+
And in head.S we map like this:
/* Xen read-only data section. */
ldr x1, =_srodata
ldr x2, =_erodataprepare_xen_region x0, x1, x2, x3, x4, x5, attr_prbar=REGION_RO_PRBAR
/* Xen read-only after init and data section. (RW data) */ ldr x1, =__ro_after_init_start
ldr x2, =__init_beginprepare_xen_region x0, x1, x2, x3, x4, x5
Now, because (__ro_after_init_start, __ro_after_init_end) needs to become RO, it means that RO section will be extended to (_srodata, __ro_after_init_end) and RW section will be shrinked to (__ro_after_init_end, __init_begin):
read only data region:
+--------------------------+ | _srodata | | __ro_after_init_end | +--------------------------+
RW data region:
+---------------------------+ | __ro_after_init_end | | __init_begin | +---------------------------+
So what we’ve done is taking (__ro_after_init_start, __ro_after_init_end) from the RW region and attach it to the RO region.
Quoted 8 lines
Quoted 4 lines
This does not work on MPU systems at present because part-region modification is not supported. Therefore introduce the function
What else is in that region?
Wouldn't it be better to have one region for this__ro_after_initso that we don't need to shrink/extend the mappings? Is it done because of number of regions limitation?
So if we do in this way we waste one region, because we will have 2 region declared RO that are also contiguous, so easily mergeable, like how we are doing above by Extending/shrinking.
Cheers,
Luca
Re: [PATCH 4/6] arm/mpu: Introduce modify_after_init_mappings
Quoted 69 lines · Luca Fancellu
Hi Michael,
Quoted 5 lines · Michal
On 28/11/2025 10:58, Harry Ramsey wrote:
Quoted 8 lines
Quoted 6 lines
From: Luca Fancellu <[email protected]>
During
init_done, Xen sets the permissions of all symbols marked with__ro_after_initto be read-only. Currently this is achieved by callingmodify_xen_mappingsand will shrink the RW mapping on one side and extend the RO mapping on the other.Can you be more specific about the sides you mention? How did you deduce it? I assume you are talking about MMU part.
You are right, this sentence is a bit misleading.
So what was written here was meant to say that on MPUmodify_xen_mappingsshould shrink the RW mapping region and extend the RO mapping region because as of now they are declared like this inxen.lds.S:read only data:
+------------------+ | _srodata | | _erodata | +-------------------+RW data:
+---------------------------+ | __ro_after_init_start | | __ro_after_init_end | | __init_begin | +---------------------------+And in
head.Swe map like this:/* Xen read-only data section. */
ldr x1, =_srodata
ldr x2, =_erodataprepare_xen_regionx0, x1, x2, x3, x4, x5, attr_prbar=REGION_RO_PRBAR/* Xen read-only after init and data section. (RW data) */ ldr x1, =__ro_after_init_start
ldr x2, =__init_beginprepare_xen_regionx0, x1, x2, x3, x4, x5Now, because (
__ro_after_init_start,__ro_after_init_end) needs to become RO, it means that RO section will be extended to (_srodata,__ro_after_init_end) and RW section will be shrinked to (__ro_after_init_end,__init_begin):read only data region:
+--------------------------+ | _srodata | | __ro_after_init_end | +--------------------------+RW data region:
+---------------------------+ | __ro_after_init_end | | __init_begin | +---------------------------+So what we’ve done is taking (
__ro_after_init_start,__ro_after_init_end) from the RW region and attach it to the RO region.Quoted 4 lines
Quoted 4 lines
This does not work on MPU systems at present because part-region modification is not supported. Therefore introduce the function
Quoted 8 lines
Quoted 4 lines
What else is in that region?
Wouldn't it be better to have one region for this__ro_after_initso that we don't need to shrink/extend the mappings? Is it done because of number of regions limitation?So if we do in this way we waste one region, because we will have 2 region declared RO that are also contiguous, so easily mergeable, like how we are doing above by Extending/shrinking.
Ok, that makes more sense. I thought the descrption in commit msg was somehow about MMU. It's not ideal to depend on the regions layout but I guess it's ok if we don't want to waste regions.
~Michal
PATCH 5/6 arm: Use secure hypervisor timer in MPU system
In reply to Harry RamseyFrom: Penny Zheng <[email protected]>
As MPU systems only have one secure state, we have to use secure EL2 hypervisor timer for Xen in secure EL2.
In this patch, we introduce a new Kconfig option ARM_SECURE_STATE and a set of secure hypervisor timer registers CNTHPS_*_EL2. We alias CNTHP_*_EL2 to CNTHPS_*_EL2 to keep the timer code flow unchanged.
--- xen/arch/arm/Kconfig | 5 +++++ xen/arch/arm/include/asm/arm64/sysregs.h | 12 ++++++++++++ 2 files changed, 17 insertions(+)
xen/arch/arm/Kconfig +5 -0
@@ -91,6 +91,7 @@ config MMU config MPU bool "MPU" if UNSUPPORTED + select ARM_SECURE_STATE if ARM_64 select STATIC_MEMORY help Memory Protection Unit (MPU). Select if you plan to run Xen on ARMv8-R
@@ -223,6 +224,10 @@ config HARDEN_BRANCH_PREDICTOR If unsure, say Y. +config ARM_SECURE_STATE + bool "Xen will run in Arm Secure State" + default n + config ARM64_PTR_AUTH def_bool n depends on ARM_64
xen/arch/arm/include/asm/arm64/sysregs.h +12 -0
@@ -462,6 +462,18 @@#define ZCR_ELx_LEN_SIZE 9#define ZCR_ELx_LEN_MASK 0x1ff +#ifdef CONFIG_ARM_SECURE_STATE +/* +* The Armv8-R AArch64 architecture always executes code in Secure +* state with EL2 as the highest Exception. +* +* Hypervisor timer registers for Secure EL2. +*/ +#define CNTHP_TVAL_EL2 CNTHPS_TVAL_EL2 +#define CNTHP_CTL_EL2 CNTHPS_CTL_EL2 +#define CNTHP_CVAL_EL2 CNTHPS_CVAL_EL2 +#endif +#define REGION_TEXT_PRBAR 0x38/* SH=11 AP=10 XN=00 */ #define REGION_RO_PRBAR 0x3A/* SH=11 AP=10 XN=10 */ #define REGION_DATA_PRBAR 0x32/* SH=11 AP=00 XN=10 */
Signature · 1 line
2.43.0
Re: [PATCH 5/6] arm: Use secure hypervisor timer in MPU system
Quoted 32 lines · Harry Ramsey
From: Penny Zheng <[email protected]>
As MPU systems only have one secure state, we have to use secure EL2 hypervisor timer for Xen in secure EL2.
In this patch, we introduce a new
KconfigoptionARM_SECURE_STATEand a set of secure hypervisor timer registers CNTHPS_*_EL2. We alias CNTHP_*_EL2 to CNTHPS_*_EL2 to keep the timer code flow unchanged.Signed-off-by Penny Zheng <[email protected]>Signed-off-by Wei Chen <[email protected]>Signed-off-by Luca Fancellu <[email protected]>Signed-off-by Harry Ramsey <[email protected]>--- xen/arch/arm/Kconfig | 5 +++++ xen/arch/arm/include/asm/arm64/sysregs.h | 12 ++++++++++++ 2 files changed, 17 insertions(+)xen/arch/arm/Kconfig +1
-0@@ -91,6 +91,7 @@ config MMU config MPU bool "MPU" if UNSUPPORTED + select ARM_SECURE_STATE if ARM_64 select STATIC_MEMORY help Memory Protection Unit (MPU). Select if you plan to run Xen on ARMv8-R
xen/arch/arm/Kconfig config HARDEN_BRANCH_PREDICTOR · line 229
Quoted 6 lines
xen/arch/arm/Kconfig +3
-0@@ -223,6 +224,10 @@ config HARDEN_BRANCH_PREDICTOR If unsure, say Y. +config ARM_SECURE_STATE + bool "Xen will run in Arm Secure State" + default n
No need, n is a default. I don't think this should be a selectable option.
Quoted 9 lines
xen/arch/arm/Kconfig +1
-0+ config ARM64_PTR_AUTH def_bool n depends on ARM_64xen/arch/arm/include/asm/arm64/sysregs.h +0
-0
xen/arch/arm/include/asm/arm64/sysregs.h line 468
Quoted 7 lines
xen/arch/arm/include/asm/arm64/sysregs.h +4
-0@@ -462,6 +462,18 @@#define ZCR_ELx_LEN_SIZE 9#define ZCR_ELx_LEN_MASK 0x1ff +#ifdef CONFIG_ARM_SECURE_STATE +/* +* The Armv8-R AArch64 architecture always executes code in Secure +* state with EL2 as the highest Exception.
s/Exception/exception level/
xen/arch/arm/include/asm/arm64/sysregs.h line 472
Quoted 4 lines
xen/arch/arm/include/asm/arm64/sysregs.h +4
-0+* +* Hypervisor timer registers for Secure EL2. +*/ +#define CNTHP_TVAL_EL2 CNTHPS_TVAL_EL2
TVAL is not used in Xen, so you can drop it
xen/arch/arm/include/asm/arm64/sysregs.h line 474
Quoted 2 lines
xen/arch/arm/include/asm/arm64/sysregs.h +2
-0+#define CNTHP_CTL_EL2 CNTHPS_CTL_EL2 +#define CNTHP_CVAL_EL2 CNTHPS_CVAL_EL2
EL1 will still use the NS EL1 timer (CNTP)?
xen/arch/arm/include/asm/arm64/sysregs.h line 479
Quoted 5 lines
xen/arch/arm/include/asm/arm64/sysregs.h +2
-0+#endif +#define REGION_TEXT_PRBAR 0x38/* SH=11 AP=10 XN=00 */ #define REGION_RO_PRBAR 0x3A/* SH=11 AP=10 XN=10 */ #define REGION_DATA_PRBAR 0x32/* SH=11 AP=00 XN=10 */
~Michal
PATCH 6/6 arm/mpu: Map domain page in AArch64 MPU systems
In reply to Harry RamseyFrom: Penny Zheng <[email protected]>
In MPU systems, we implement map_domain_page()/unmap_domain_page() through mapping the domain page with a MPU region on demand.
--- xen/arch/arm/Kconfig | 1 + xen/arch/arm/mpu/Makefile | 1 + xen/arch/arm/mpu/domain-page.c | 53 ++++++++++++++++++++++++++++++++++ 3 files changed, 55 insertions(+) create mode 100644 xen/arch/arm/mpu/domain-page.c
xen/arch/arm/Kconfig +1 -0
@@ -91,6 +91,7 @@ config MMU config MPU bool "MPU" if UNSUPPORTED + select ARCH_MAP_DOMAIN_PAGE if ARM_64 select ARM_SECURE_STATE if ARM_64 select STATIC_MEMORY help
xen/arch/arm/mpu/Makefile +1 -0
@@ -1,5 +1,6 @@ obj-$(CONFIG_ARM_32) += arm32/ obj-$(CONFIG_ARM_64) += arm64/ +obj-$(CONFIG_ARM_64) += domain-page.o obj-y += mm.o obj-y += p2m.o obj-y += setup.init.o
xen/arch/arm/mpu/domain-page.c new file +53 -0
53 lines not on this page · raw message
Signature · 1 line
2.43.0
Re: [PATCH 6/6] arm/mpu: Map domain page in AArch64 MPU systems
Quoted 4 lines · Harry Ramsey
From: Penny Zheng <[email protected]>
In MPU systems, we implement map_domain_page()/unmap_domain_page() through mapping the domain page with a MPU region on demand.
What prevents you from implementing the remaining few helpers?
~Michal
Quoted 94 lines
Signed-off-by Penny Zheng <[email protected]>Signed-off-by Wei Chen <[email protected]>Signed-off-by Luca Fancellu <[email protected]>Signed-off-by Harry Ramsey <[email protected]>--- xen/arch/arm/Kconfig | 1 + xen/arch/arm/mpu/Makefile | 1 + xen/arch/arm/mpu/domain-page.c | 53 ++++++++++++++++++++++++++++++++++ 3 files changed, 55 insertions(+) create mode 100644 xen/arch/arm/mpu/domain-page.cxen/arch/arm/Kconfig +1
-0@@ -91,6 +91,7 @@ config MMU config MPU bool "MPU" if UNSUPPORTED + select ARCH_MAP_DOMAIN_PAGE if ARM_64 select ARM_SECURE_STATE if ARM_64 select STATIC_MEMORY helpxen/arch/arm/mpu/Makefile +1
-0@@ -1,5 +1,6 @@ obj-$(CONFIG_ARM_32) += arm32/ obj-$(CONFIG_ARM_64) += arm64/ +obj-$(CONFIG_ARM_64) += domain-page.o obj-y += mm.o obj-y += p2m.o obj-y += setup.init.oxen/arch/arm/mpu/domain-page.c new file +53
-0@@ -0,0 +1,53 @@ +/* SPDX-License-Identifier: GPL-2.0-only */ + +#include <xen/bug.h> +#include <xen/domain_page.h> +#include <xen/mm.h> +#include <xen/mm-frame.h> +#include <xen/types.h> + +void *map_domain_page_global (mfn_t mfn) +{ +BUG_ON ("unimplemented" ); +return NULL ; +} + +/* Map a page of domheap memory */ +void *map_domain_page (mfn_t mfn) +{ +paddr_t pa =mfn_to_maddr (mfn); + +if (map_pages_to_xen ((unsigned long )pa, mfn,1 ,PAGE_HYPERVISOR_RW ) ) +return NULL ; + +return maddr_to_virt (pa); +} + +/* Release a mapping taken with map_domain_page() */ +void unmap_domain_page (const void *ptr) +{ +paddr_t base =virt_to_maddr (ptr); + +if (destroy_entire_xen_mapping (base) ) +panic ("Failed to unmap domain page\n" ); +} + +mfn_t domain_page_map_to_mfn (const void *ptr) +{ +BUG_ON ("unimplemented" ); +return INVALID_MFN ; +} + +void unmap_domain_page_global (const void *va) +{ +BUG_ON ("unimplemented" ); +} + +/* +* Local variables: +* mode: C +* c-file-style: "BSD" +* c-basic-offset: 4 +* indent-tabs-mode: nil +* End: +*/
Re: [PATCH 6/6] arm/mpu: Map domain page in AArch64 MPU systems
Hi Michael,
Quoted 10 lines · Michal
Quoted 4 lines · Harry Ramsey
From: Penny Zheng <[email protected]>
In MPU systems, we implement map_domain_page()/unmap_domain_page() through mapping the domain page with a MPU region on demand.
What prevents you from implementing the remaining few helpers?
Only the fact that they are not used at this stage, otherwise we would have seen a panic while running Linux.
Cheers,
Luca
Re: [PATCH 6/6] arm/mpu: Map domain page in AArch64 MPU systems
Quoted 7 lines · Luca Fancellu
Hi Michael,
Quoted 5 lines · Michal
On 28/11/2025 10:58, Harry Ramsey wrote:
Quoted 8 lines
Quoted 5 lines
Quoted 4 lines
From: Penny Zheng <[email protected]>
In MPU systems, we implement map_domain_page()/unmap_domain_page() through mapping the domain page with a MPU region on demand.
What prevents you from implementing the remaining few helpers?
Only the fact that they are not used at this stage, otherwise we would have seen a panic while running Linux.
Sure but it looks like that they would also be a few-liners hence there is a feeling that they could all be done in one go for completeness sake.
~Michal
Re: [PATCH 6/6] arm/mpu: Map domain page in AArch64 MPU systems
Quoted 16 lines · Michal
Quoted 9 lines · Luca Fancellu
Hi Michael,
Quoted 7 lines · Michal
Quoted 2 lines · Harry Ramsey
From: Penny Zheng <[email protected]>
Quoted 8 lines
Quoted 6 lines
Quoted 3 lines
Quoted 2 lines
In MPU systems, we implement map_domain_page()/unmap_domain_page() through mapping the domain page with a MPU region on demand.
What prevents you from implementing the remaining few helpers?
Only the fact that they are not used at this stage, otherwise we would have seen a panic while running Linux.
Sure but it looks like that they would also be a few-liners hence there is a feeling that they could all be done in one go for completeness sake.
yeah but we don’t have the confidence it works because we can’t test
Cheers,
Luca