arm64: mm: phys_mem_access_prot() reports non-kernel memory as device memory
PATCH 1/1 arm64: mm: phys_mem_access_prot() reports non-kernel memory as device memory
Running 'acpidump' on an arm64 system causes the program to SIGSEGV, and yield the following trace:
[ 71.623572] acpidump[2159]: unhandled alignment fault (11) at 0xffff8a4a909c, esr 0x92000021 [ 71.632030] pgd = ffff8003ecf7d000 [ 71.635420] [ffff8a4a909c] *pgd=00000083ec370003, *pud=00000083e7225003, *pmd=00000083ec3f1003, *pte=01600083ff1c3fc3 [ 71.646042] [ 71.647524] CPU: 0 PID: 2159 Comm: acpidump Tainted: G W I 4.8.0-rc4 #5081 [ 71.655352] Hardware name: AMD Overdrive/Supercharger/Default string, BIOS ROD1002C 04/08/2016 [ 71.663957] task: ffff8003eac41900 task.stack: ffff8003e7210000 [ 71.669871] PC is at 0xffff8a3c77d4 [ 71.673353] LR is at 0x404934 [ 71.676314] pc : [<0000ffff8a3c77d4>] lr : [<0000000000404934>] pstate: 00000000 [ 71.683702] sp : 0000ffffd0e39af0
Acpidump digs through /sys/ to find the physical address of the ACPI tables, then reads them from /dev/mem. The code that provides /dev/mem uses phys_mem_access_prot() to determine the protection type it should use.
Arm64's phys_mem_access_prot() reports any memory region that isn't part of the linear map as device memory. This breaks Acpidump as the acpi tables are neither device memory nor part of the linear map.
Change this check to use memblock_is_memory() instead. On an EFI system any region the efi memory map described as any of WB/WC/WT appears in memblock.memory. On a non-EFI system, setup_machine_fdt() will cause memory nodes in the DT to be added to memblock.memory.
/dev/mem is the only user of phys_mem_access_prot()
---
Stable?
Acpi was introduced for v4.1 but hidden behind 'expert'. 'expert' was removed for v4.7.
arch/arm64/mm/mmu.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
arch/arm64/mm/mmu.c +1 -1
@@ -63,7 +63,7 @@ static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused;pgprot_t phys_mem_access_prot (struct file *file,unsigned long pfn,unsigned long size,pgprot_t vma_prot) { -if (!pfn_valid (pfn)) +if (!memblock_is_memory (pfn <<PAGE_SHIFT ))return pgprot_noncached (vma_prot);else if (file->f_flags &O_SYNC )return pgprot_writecombine (vma_prot);
Signature · 1 line
2.8.0.rc3
PATCH 1/1 arm64: Drop generic xlate_dev_mem_{k,}ptr()
The code that provides /dev/mem uses xlate_dev_mem_{k,}ptr() to avoid making a cachable mapping of a non-cachable area on ia64. On arm64 we do this via phys_mem_access_prot() instead, but provide dummy versions of xlate_dev_mem_{k,}ptr().
These are the same as those in asm-generic/io.h, which we include from asm/io.h
--- arch/arm64/include/asm/io.h | 11 ----------- 1 file changed, 11 deletions(-)
arch/arm64/include/asm/io.h +0 -11
@@ -184,17 +184,6 @@ extern void __iomem *ioremap_cache(phys_addr_t phys_addr, size_t size);#define iowrite32be (v,p) ({ __iowmb(); __raw_writel((__force __u32)cpu_to_be32(v), p); })#define iowrite64be (v,p) ({ __iowmb(); __raw_writeq((__force __u64)cpu_to_be64(v), p); }) -/* -* Convert a physical pointer to a virtual kernel pointer for /dev/mem -* access -*/ -#define xlate_dev_mem_ptr (p) __va(p) - -/* -* Convert a virtual cached pointer to an uncached pointer -*/ -#define xlate_dev_kmem_ptr (p) p -#include <asm-generic/io.h> /*
Signature · 1 line
2.8.0.rc3
PATCH 1/1 arm64: mm: phys_mem_access_prot() reports non-kernel memory as device memory
In reply to James MorseQuoted 44 lines · James Morse
Running 'acpidump' on an arm64 system causes the program to SIGSEGV, and yield the following trace:
[ 71.623572] acpidump[2159]: unhandled alignment fault (11) at 0xffff8a4a909c, esr 0x92000021 [ 71.632030] pgd = ffff8003ecf7d000 [ 71.635420] [ffff8a4a909c] *pgd=00000083ec370003, *pud=00000083e7225003, *pmd=00000083ec3f1003, *pte=01600083ff1c3fc3 [ 71.646042] [ 71.647524] CPU: 0 PID: 2159 Comm: acpidump Tainted: G W I 4.8.0-rc4 #5081 [ 71.655352] Hardware name: AMD Overdrive/Supercharger/Default string, BIOS ROD1002C 04/08/2016 [ 71.663957] task: ffff8003eac41900 task.stack: ffff8003e7210000 [ 71.669871] PC is at 0xffff8a3c77d4 [ 71.673353] LR is at 0x404934 [ 71.676314] pc : [<0000ffff8a3c77d4>] lr : [<0000000000404934>] pstate: 00000000 [ 71.683702] sp : 0000ffffd0e39af0Acpidump digs through
/sys/to find the physical address of the ACPI tables, then reads them from/dev/mem. The code that provides/dev/memusesphys_mem_access_prot()to determine the protection type it should use.Arm64's
phys_mem_access_prot()reports any memory region that isn't part of the linear map as device memory. This breaks Acpidump as the acpi tables are neither device memory nor part of the linear map.Change this check to use
memblock_is_memory()instead. On an EFI system any region the efi memory map described as any of WB/WC/WT appears in memblock.memory. On a non-EFI system,setup_machine_fdt()will cause memory nodes in the DT to be added to memblock.memory.
/dev/memis the only user ofphys_mem_access_prot()Reported-by Lorenzo Pieralisi <[email protected]>Signed-off-by James Morse <[email protected]>---
Stable?
Acpi was introduced for v4.1 but hidden behind 'expert'. 'expert' was removed for v4.7.arch/arm64/mm/mmu.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)arch/arm64/mm/mmu.c +0
-0
arch/arm64/mm/mmu.c static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused; · line 66
Quoted 5 lines
arch/arm64/mm/mmu.c +1
-1@@ -63,7 +63,7 @@ static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused;pgprot_t phys_mem_access_prot (struct file *file,unsigned long pfn,unsigned long size,pgprot_t vma_prot) { -if (!pfn_valid (pfn)) +if (!memblock_is_memory (pfn <<PAGE_SHIFT ))
Hmm, this looks pretty weird to me. pfn_valid currently calls memblock_is_map_memory, so now we have this slightly paradoxical situation of having some memory marked NOMAP but wanting to map it to userspace.
So our abstractions don't seem to align with reality. Why are the ACPI tables getting marked as NOMAP to being with?
Will
PATCH 1/1 arm64: mm: phys_mem_access_prot() reports non-kernel memory as device memory
Hi Will,
(CC: Ard and Mark, get_maintainer.pl didn't apply common sense for me...)
Quoted 30 lines · Will Deacon
Quoted 28 lines · James Morse
Running 'acpidump' on an arm64 system causes the program to SIGSEGV, and yield the following trace:
[ 71.623572] acpidump[2159]: unhandled alignment fault (11) at 0xffff8a4a909c, esr 0x92000021 [ 71.632030] pgd = ffff8003ecf7d000 [ 71.635420] [ffff8a4a909c] *pgd=00000083ec370003, *pud=00000083e7225003, *pmd=00000083ec3f1003, *pte=01600083ff1c3fc3 [ 71.646042] [ 71.647524] CPU: 0 PID: 2159 Comm: acpidump Tainted: G W I 4.8.0-rc4 #5081 [ 71.655352] Hardware name: AMD Overdrive/Supercharger/Default string, BIOS ROD1002C 04/08/2016 [ 71.663957] task: ffff8003eac41900 task.stack: ffff8003e7210000 [ 71.669871] PC is at 0xffff8a3c77d4 [ 71.673353] LR is at 0x404934 [ 71.676314] pc : [<0000ffff8a3c77d4>] lr : [<0000000000404934>] pstate: 00000000 [ 71.683702] sp : 0000ffffd0e39af0Acpidump digs through
/sys/to find the physical address of the ACPI tables, then reads them from/dev/mem. The code that provides/dev/memusesphys_mem_access_prot()to determine the protection type it should use.Arm64's
phys_mem_access_prot()reports any memory region that isn't part of the linear map as device memory. This breaks Acpidump as the acpi tables are neither device memory nor part of the linear map.Change this check to use
memblock_is_memory()instead. On an EFI system any region the efi memory map described as any of WB/WC/WT appears in memblock.memory. On a non-EFI system,setup_machine_fdt()will cause memory nodes in the DT to be added to memblock.memory.
/dev/memis the only user ofphys_mem_access_prot()
Quoted 14 lines
arch/arm64/mm/mmu.c static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused; · line 66
Quoted 10 lines
arch/arm64/mm/mmu.c +1
-1@@ -63,7 +63,7 @@ static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused;pgprot_t phys_mem_access_prot (struct file *file,unsigned long pfn,unsigned long size,pgprot_t vma_prot) { -if (!pfn_valid (pfn)) +if (!memblock_is_memory (pfn <<PAGE_SHIFT ))Hmm, this looks pretty weird to me.
pfn_validcurrently callsmemblock_is_map_memory, so now we have this slightly paradoxical situation of having some memory marked NOMAP but wanting to map it to userspace.
Yup, crazy as it looks, that is what is happening.
Quoted 2 lines
So our abstractions don't seem to align with reality. Why are the ACPI tables getting marked as NOMAP to being with?
EFI adds any region we can map as normal memory to memblock.memory, it also adds anything it considers reserved to the memblock.nomap, e.g. the ACPI tables. This causes these regions be left out of the linear map.
(I don't know why exactly, but assume its so that the attributes and permissions can be tweaked easily)
Nomap here means 'not in the linear map'. When we call phys_mem_access_prot(), or acpi_os_ioremap() we are accessing these pages and want to know the attributes to map with. For phys_mem_access_prot() 'not in the linear map' means 'device memory', which causes acpidump to crash as it makes unaligned accesses. (We had similar problems with acpica in the kernel and acpi_os_ioremap(), which
[0] fixes).
Thanks,
James
PATCH 1/1 arm64: mm: phys_mem_access_prot() reports non-kernel memory as device memory
Quoted 50 lines · James Morse
Hi Will,
(CC: Ard and Mark, get_maintainer.pl didn't apply common sense for me...)
Quoted 30 lines · Will Deacon
Quoted 28 lines · James Morse
Running 'acpidump' on an arm64 system causes the program to SIGSEGV, and yield the following trace:
[ 71.623572] acpidump[2159]: unhandled alignment fault (11) at 0xffff8a4a909c, esr 0x92000021 [ 71.632030] pgd = ffff8003ecf7d000 [ 71.635420] [ffff8a4a909c] *pgd=00000083ec370003, *pud=00000083e7225003, *pmd=00000083ec3f1003, *pte=01600083ff1c3fc3 [ 71.646042] [ 71.647524] CPU: 0 PID: 2159 Comm: acpidump Tainted: G W I 4.8.0-rc4 #5081 [ 71.655352] Hardware name: AMD Overdrive/Supercharger/Default string, BIOS ROD1002C 04/08/2016 [ 71.663957] task: ffff8003eac41900 task.stack: ffff8003e7210000 [ 71.669871] PC is at 0xffff8a3c77d4 [ 71.673353] LR is at 0x404934 [ 71.676314] pc : [<0000ffff8a3c77d4>] lr : [<0000000000404934>] pstate: 00000000 [ 71.683702] sp : 0000ffffd0e39af0Acpidump digs through
/sys/to find the physical address of the ACPI tables, then reads them from/dev/mem. The code that provides/dev/memusesphys_mem_access_prot()to determine the protection type it should use.Arm64's
phys_mem_access_prot()reports any memory region that isn't part of the linear map as device memory. This breaks Acpidump as the acpi tables are neither device memory nor part of the linear map.Change this check to use
memblock_is_memory()instead. On an EFI system any region the efi memory map described as any of WB/WC/WT appears in memblock.memory. On a non-EFI system,setup_machine_fdt()will cause memory nodes in the DT to be added to memblock.memory.
/dev/memis the only user ofphys_mem_access_prot()Quoted 14 lines
arch/arm64/mm/mmu.c static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused; · line 66
Quoted 10 lines
arch/arm64/mm/mmu.c +1
-1@@ -63,7 +63,7 @@ static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused;pgprot_t phys_mem_access_prot (struct file *file,unsigned long pfn,unsigned long size,pgprot_t vma_prot) { -if (!pfn_valid (pfn)) +if (!memblock_is_memory (pfn <<PAGE_SHIFT ))Hmm, this looks pretty weird to me.
pfn_validcurrently callsmemblock_is_map_memory, so now we have this slightly paradoxical situation of having some memory marked NOMAP but wanting to map it to userspace.
Quoted 3 lines
Yup, crazy as it looks, that is what is happening.
Indeed. Formerly, we would remove this from memblock entirely, resulting in the same situation, i.e., that the region is omitted from the linear mapping. However, by doing that, we also lose the annotation that the region was memory to begin with, resulting in the ACPI layer using ioremap() rather than ioremap_cache() or memremap() to map it.
So the whole point of introducing MEMBLOCK_NOMAP was to allow memory to be kept in the memblock array, but with an annotation that it should not be covered by the linear mapping.
Quoted 9 lines
Quoted 2 lines
So our abstractions don't seem to align with reality. Why are the ACPI tables getting marked as NOMAP to being with?
EFI adds any region we can map as normal memory to memblock.memory, it also adds anything it considers reserved to the memblock.nomap, e.g. the ACPI tables. This causes these regions be left out of the linear map.
(I don't know why exactly, but assume its so that the attributes and permissions can be tweaked easily)
To prevent mismatched attributes between the linear mapping and whatever mapping the firmware and/or ACPI are using. Also, going through the trouble of mapping runtime services executable code with R-X permissions and then having a writable alias is not great in terms of security.
Quoted 7 lines
Nomap here means 'not in the linear map'. When we call
phys_mem_access_prot(), oracpi_os_ioremap()we are accessing these pages and want to know the attributes to map with. Forphys_mem_access_prot()'not in the linear map' means 'device memory', which causes acpidump to crash as it makes unaligned accesses. (We had similar problems with acpica in the kernel andacpi_os_ioremap(), which
[0] fixes).
Indeed. We need to classify memory regions into 3 groups: { linearly mapped memory, other memory, not memory }
PATCH 1/1 arm64: mm: phys_mem_access_prot() reports non-kernel memory as device memory
Quoted 42 lines · Ard Biesheuvel
Quoted 17 lines · James Morse
Quoted 15 lines · Will Deacon
arch/arm64/mm/mmu.c static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused; · line 66
Quoted 10 lines · James Morse
arch/arm64/mm/mmu.c +1
-1@@ -63,7 +63,7 @@ static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused;pgprot_t phys_mem_access_prot (struct file *file,unsigned long pfn,unsigned long size,pgprot_t vma_prot) { -if (!pfn_valid (pfn)) +if (!memblock_is_memory (pfn <<PAGE_SHIFT ))Hmm, this looks pretty weird to me.
pfn_validcurrently callsmemblock_is_map_memory, so now we have this slightly paradoxical situation of having some memory marked NOMAP but wanting to map it to userspace.Quoted 2 lines
Yup, crazy as it looks, that is what is happening.
Indeed. Formerly, we would remove this from memblock entirely, resulting in the same situation, i.e., that the region is omitted from the linear mapping. However, by doing that, we also lose the annotation that the region was memory to begin with, resulting in the ACPI layer using
ioremap()rather thanioremap_cache()ormemremap()to map it.So the whole point of introducing
MEMBLOCK_NOMAPwas to allow memory to be kept in the memblock array, but with an annotation that it should not be covered by the linear mapping.Quoted 9 lines
Quoted 2 lines
So our abstractions don't seem to align with reality. Why are the ACPI tables getting marked as NOMAP to being with?
EFI adds any region we can map as normal memory to memblock.memory, it also adds anything it considers reserved to the memblock.nomap, e.g. the ACPI tables. This causes these regions be left out of the linear map.
(I don't know why exactly, but assume its so that the attributes and permissions can be tweaked easily)
Quoted 5 lines
To prevent mismatched attributes between the linear mapping and whatever mapping the firmware and/or ACPI are using. Also, going through the trouble of mapping runtime services executable code with R-X permissions and then having a writable alias is not great in terms of security.
Ok, but with this patch we potentially have both of those problems (mismatched attributes and RWX permission) with the userspace mapping :/
Is it worth trying to avoid any of this, or do we just throw our hands in the air and give up when /dev/mem is being used?
Will
PATCH 1/1 arm64: mm: phys_mem_access_prot() reports non-kernel memory as device memory
Quoted 49 lines · Will Deacon
Quoted 42 lines · Ard Biesheuvel
Quoted 17 lines · James Morse
Quoted 15 lines · Will Deacon
arch/arm64/mm/mmu.c static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused; · line 66
Quoted 10 lines · James Morse
arch/arm64/mm/mmu.c +1
-1@@ -63,7 +63,7 @@ static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused;pgprot_t phys_mem_access_prot (struct file *file,unsigned long pfn,unsigned long size,pgprot_t vma_prot) { -if (!pfn_valid (pfn)) +if (!memblock_is_memory (pfn <<PAGE_SHIFT ))Hmm, this looks pretty weird to me.
pfn_validcurrently callsmemblock_is_map_memory, so now we have this slightly paradoxical situation of having some memory marked NOMAP but wanting to map it to userspace.Quoted 2 lines
Yup, crazy as it looks, that is what is happening.
Indeed. Formerly, we would remove this from memblock entirely, resulting in the same situation, i.e., that the region is omitted from the linear mapping. However, by doing that, we also lose the annotation that the region was memory to begin with, resulting in the ACPI layer using
ioremap()rather thanioremap_cache()ormemremap()to map it.So the whole point of introducing
MEMBLOCK_NOMAPwas to allow memory to be kept in the memblock array, but with an annotation that it should not be covered by the linear mapping.Quoted 9 lines
Quoted 2 lines
So our abstractions don't seem to align with reality. Why are the ACPI tables getting marked as NOMAP to being with?
EFI adds any region we can map as normal memory to memblock.memory, it also adds anything it considers reserved to the memblock.nomap, e.g. the ACPI tables. This causes these regions be left out of the linear map.
(I don't know why exactly, but assume its so that the attributes and permissions can be tweaked easily)Quoted 5 lines
To prevent mismatched attributes between the linear mapping and whatever mapping the firmware and/or ACPI are using. Also, going through the trouble of mapping runtime services executable code with R-X permissions and then having a writable alias is not great in terms of security.
Quoted 6 lines
Ok, but with this patch we potentially have both of those problems (mismatched attributes and RWX permission) with the userspace mapping :/
Is it worth trying to avoid any of this, or do we just throw our hands in the air and give up when
/dev/memis being used?
Well, I do think /dev/mem is a ridiculous hack, but it is arguably an improvement to not map a region as device if we know it is backed by memory.
Leif already did a lot of work on dmidecode and related tools to get rid of /dev/mem dependencies, especially because of the fact that it used 'known' physical addresses to look for magic numbers identifying SMBIOS tables etc. I am not sure if acpidump was on our radar yet, but in the long term, we should be able to reduce the dependency on /dev/mem for anything other than development use.
Would it help at all if we restricted these uses to read only? I am not aware of any tools that require read/write access to memory in this way.
PATCH 1/1 arm64: mm: phys_mem_access_prot() reports non-kernel memory as device memory
Quoted 56 lines · Ard Biesheuvel
Quoted 49 lines · Will Deacon
Quoted 42 lines · Ard Biesheuvel
Quoted 17 lines · James Morse
Quoted 15 lines · Will Deacon
arch/arm64/mm/mmu.c static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused; · line 66
Quoted 10 lines · James Morse
arch/arm64/mm/mmu.c +1
-1@@ -63,7 +63,7 @@ static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused;pgprot_t phys_mem_access_prot (struct file *file,unsigned long pfn,unsigned long size,pgprot_t vma_prot) { -if (!pfn_valid (pfn)) +if (!memblock_is_memory (pfn <<PAGE_SHIFT ))Hmm, this looks pretty weird to me.
pfn_validcurrently callsmemblock_is_map_memory, so now we have this slightly paradoxical situation of having some memory marked NOMAP but wanting to map it to userspace.Quoted 2 lines
Yup, crazy as it looks, that is what is happening.
Indeed. Formerly, we would remove this from memblock entirely, resulting in the same situation, i.e., that the region is omitted from the linear mapping. However, by doing that, we also lose the annotation that the region was memory to begin with, resulting in the ACPI layer using
ioremap()rather thanioremap_cache()ormemremap()to map it.So the whole point of introducing
MEMBLOCK_NOMAPwas to allow memory to be kept in the memblock array, but with an annotation that it should not be covered by the linear mapping.Quoted 9 lines
Quoted 2 lines
So our abstractions don't seem to align with reality. Why are the ACPI tables getting marked as NOMAP to being with?
EFI adds any region we can map as normal memory to memblock.memory, it also adds anything it considers reserved to the memblock.nomap, e.g. the ACPI tables. This causes these regions be left out of the linear map.
(I don't know why exactly, but assume its so that the attributes and permissions can be tweaked easily)Quoted 5 lines
To prevent mismatched attributes between the linear mapping and whatever mapping the firmware and/or ACPI are using. Also, going through the trouble of mapping runtime services executable code with R-X permissions and then having a writable alias is not great in terms of security.
Quoted 5 lines
Ok, but with this patch we potentially have both of those problems (mismatched attributes and RWX permission) with the userspace mapping :/
Is it worth trying to avoid any of this, or do we just throw our hands in the air and give up when
/dev/memis being used?
Quoted 8 lines
Well, I do think
/dev/memis a ridiculous hack, but it is arguably an improvement to not map a region as device if we know it is backed by memory.Leif already did a lot of work on dmidecode and related tools to get rid of
/dev/memdependencies, especially because of the fact that it used 'known' physical addresses to look for magic numbers identifying SMBIOS tables etc. I am not sure if acpidump was on our radar yet, but
Yeah, Roy did acpidump last year:
https://lists.linaro.org/pipermail/linaro-uefi/2015-June/000906.html
Quoted 2 lines
in the long term, we should be able to reduce the dependency on
/dev/memfor anything other than development use.
On x86 there are certain legacy reasons why they need to keep the interface enabled (for now) - but on ARM the only use for it is not wanting to bother with creating a driver for poking various controllers from userland. It has no business being enabled in a distro kernel. It is the opposite of what an operating system is for.
Quoted 3 lines
Would it help at all if we restricted these uses to read only? I am not aware of any tools that require read/write access to memory in this way.
If any such utilities remain, we need to fix them.
/
Leif
PATCH 1/1 arm64: mm: phys_mem_access_prot() reports non-kernel memory as device memory
In reply to Ard BiesheuvelQuoted 31 lines · Ard Biesheuvel
Quoted 19 lines · Will Deacon
Quoted 12 lines · Ard Biesheuvel
Quoted 10 lines · James Morse
Quoted 2 lines · Will Deacon
So our abstractions don't seem to align with reality. Why are the ACPI tables getting marked as NOMAP to being with?
EFI adds any region we can map as normal memory to memblock.memory, it also adds anything it considers reserved to the memblock.nomap, e.g. the ACPI tables. This causes these regions be left out of the linear map.
(I don't know why exactly, but assume its so that the attributes and permissions can be tweaked easily)Quoted 5 lines
To prevent mismatched attributes between the linear mapping and whatever mapping the firmware and/or ACPI are using. Also, going through the trouble of mapping runtime services executable code with R-X permissions and then having a writable alias is not great in terms of security.
Quoted 6 lines
Ok, but with this patch we potentially have both of those problems (mismatched attributes and RWX permission) with the userspace mapping :/
Is it worth trying to avoid any of this, or do we just throw our hands in the air and give up when
/dev/memis being used?Well, I do think
/dev/memis a ridiculous hack, but it is arguably an improvement to not map a region as device if we know it is backed by memory.
Quoted 6 lines
Leif already did a lot of work on dmidecode and related tools to get rid of
/dev/memdependencies, especially because of the fact that it used 'known' physical addresses to look for magic numbers identifying SMBIOS tables etc. I am not sure if acpidump was on our radar yet, but in the long term, we should be able to reduce the dependency on/dev/memfor anything other than development use.
Judging by Leif's reply, acpidump has already been fixed too.
Quoted 3 lines
Would it help at all if we restricted these uses to read only? I am not aware of any tools that require read/write access to memory in this way.
I'd certainly be happier if we dropped the write permission and derived the memory type from EFI for the NOMAP regions (it looks like ia64 tries to do something like this, whereas we currently return non-cacheable if O_SYNC is set on the fd).
Will
PATCH 1/1 arm64: mm: phys_mem_access_prot() reports non-kernel memory as device memory
Quoted 32 lines · Will Deacon
Quoted 31 lines · Ard Biesheuvel
Quoted 19 lines · Will Deacon
Quoted 12 lines · Ard Biesheuvel
Quoted 10 lines · James Morse
Quoted 2 lines · Will Deacon
So our abstractions don't seem to align with reality. Why are the ACPI tables getting marked as NOMAP to being with?
EFI adds any region we can map as normal memory to memblock.memory, it also adds anything it considers reserved to the memblock.nomap, e.g. the ACPI tables. This causes these regions be left out of the linear map.
(I don't know why exactly, but assume its so that the attributes and permissions can be tweaked easily)Quoted 5 lines
To prevent mismatched attributes between the linear mapping and whatever mapping the firmware and/or ACPI are using. Also, going through the trouble of mapping runtime services executable code with R-X permissions and then having a writable alias is not great in terms of security.
Quoted 6 lines
Ok, but with this patch we potentially have both of those problems (mismatched attributes and RWX permission) with the userspace mapping :/
Is it worth trying to avoid any of this, or do we just throw our hands in the air and give up when
/dev/memis being used?Well, I do think
/dev/memis a ridiculous hack, but it is arguably an improvement to not map a region as device if we know it is backed by memory.
Quoted 9 lines
Quoted 6 lines
Leif already did a lot of work on dmidecode and related tools to get rid of
/dev/memdependencies, especially because of the fact that it used 'known' physical addresses to look for magic numbers identifying SMBIOS tables etc. I am not sure if acpidump was on our radar yet, but in the long term, we should be able to reduce the dependency on/dev/memfor anything other than development use.Judging by Leif's reply, acpidump has already been fixed too.
Fortunately, yes.
Quoted 8 lines
Quoted 3 lines
Would it help at all if we restricted these uses to read only? I am not aware of any tools that require read/write access to memory in this way.
Quoted 5 lines
I'd certainly be happier if we dropped the write permission and derived the memory type from EFI for the NOMAP regions (it looks like ia64 tries to do something like this, whereas we currently return non-cacheable if
O_SYNCis set on the fd).
The UEFI memory map describes the supported mapping types. I.e, most regions are listed as
[Conventional Memory| | | | | | | | |WB|WT|WC|UC]
which means that it is legal to map it as
WB:
Normal Memory
Outer Write-back non-transient
Inner Write-back non-transient
WT:
Normal Memory
Outer Write-through non-transient
Inner Write-through non-transient
WC:
Normal Memory
Outer non-cacheable
Inner non-cacheable
UC:
Device-nGnRnE
(this is straight from the UEFI spec)
So while WB is used in the majority of cases, this is not universally true, and the UEFI or ACPI layers may decide to use any of those for reserved regions.
We already had a separate discussion whether it makes sense to include the UC attribute in ACPI memory regions. The problem here is that it is not unambiguously defined whether the attributes describe the nature of the memory regions (i.e, the backing) or the nature of the /contents/ of the memory regions. This is further complicated by the fact that occupied and available regions are described in the same way.
The bottom line is that, in general, we cannot infer from the UEFI memory map how a region is currently mapped if it is not covered by the linear mapping. What we /could/ do is generally disallow device mappings for regions that can also be mapped as memory (i.e, not only for /dev/mem) (although I think ACPI RAS stipulates Device-nGnRnE in some cases.)