x86/mm: Fix limit mmap() of /dev/mem to valid physical
cover x86/mm: Fix limit mmap() of /dev/mem to valid physical
From: Ralph Campbell <[email protected]>
I was debugging with v5.1.0-rc1 and while booting I hit a kernel BUG at arch/x86/mm/physaddr.c:27
which I fixed with the following patch but now I can't seem to reproduce the exact setup that triggered it. Still, it seems like a valid problem and maybe my difficulty in reproducing it explains why others haven't seen it earlier.
Ralph Campbell (1):
x86/mm: Fix limit mmap() of /dev/mem to valid physical addresses
arch/x86/mm/mmap.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
Signature · 1 line
2.20.1
PATCH 1/1 x86/mm: Fix limit mmap() of /dev/mem to valid physical addresses
From: Ralph Campbell <[email protected]>
If CONFIG_DEBUG_VIRTUAL is enabled, a read or write to /dev/mem can trigger a VIRTUAL_BUG_ON() depending on the value of high_memory. For example:
read_mem()
valid_phys_addr_range(p=401f1550, count=8)
__pa(high_memory)
__phys_addr(x=ffffc88000000000)
// __START_KERNEL_map = ffffffff80000000
// y = ffffc88000000000 - ffffffff80000000
VIRTUAL_BUG_ON(phys_addr_valid(400000000000))
// boot_cpu_data.x86_phys_bits=46Since by design high_memory is outside the range of valid physical addresses, use the non-error checking version __pa_nodebug(high_memory).
/dev/mem to valid physical addresses")
--- arch/x86/mm/mmap.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
arch/x86/mm/mmap.c +1 -1
@@ -230,7 +230,7 @@ bool mmap_address_hint_valid(unsigned long addr, unsigned long len)/* Can we access it for direct reading/writing? Must be RAM: */ int valid_phys_addr_range (phys_addr_t addr,size_t count) { -return addr + count <=__pa (high_memory); +return addr + count <=__pa_nodebug (high_memory); }/* Can we access it through mmap? Must be a valid physical address: */
Signature · 1 line
2.20.1
Re: [PATCH 1/1] x86/mm: Fix limit mmap() of /dev/mem to valid physical addresses
Ralph,
Quoted 6 lines · On Mon, 18 Mar 2019, [email protected] wrote:
From: Ralph Campbell <[email protected]>
If
CONFIG_DEBUG_VIRTUALis enabled, a read or write to/dev/memcan trigger aVIRTUAL_BUG_ON()depending on the value ofhigh_memory. For example:
Quoted 8 lines
read_mem()
valid_phys_addr_range(p=401f1550, count=8)__pa(high_memory) __phys_addr(x=ffffc88000000000) // __START_KERNEL_map = ffffffff80000000 // y = ffffc88000000000 - ffffffff80000000 VIRTUAL_BUG_ON(phys_addr_valid(400000000000)) // boot_cpu_data.x86_phys_bits=46
I have no idea why all the irrelevant information in this example would be helpful, but after extracting the meat I think I know what you want to say.
Quoted 2 lines
Since by design
high_memoryis outside the range of valid physical addresses, use the non-error checking version__pa_nodebug(high_memory).
high_memory is not outside the range of valid physical addresses by design. It's only outside when memory is populated right at the end of the physical address space.
So what you really want to say in the changelog is:
valid_phys_addr_range() is used to sanity check the physical address range of an operation, e.g. access to /dev/mem. It uses __pa(high_memory) internally.
If memory is populated at the end of the physical address space, then __pa(high_memory) is outside of the physical address space because:
high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
For the comparison in valid_phys_addr_range() this is not an issue, but if CONFIG_DEBUG_VIRTUAL is enabled, __pa() maps to __phys_addr(), which verifies that the resulting physical address is within the valid physical address space of the CPU. So in the case that memory is populated at the end of the physical address space, this is not true and triggers a VIRTUAL_BUG_ON().
Use ... instead, because ...
Quoted 1 line
Fixes be62a32044061cb4a3b70a10598e093f1319102e ("x86/mm: Limit mmap() of
Please limit the sha1 to the first 12 characters.
Quoted 2 lines
/dev/memto valid physical addresses")
No newline between Fixes and the rest please.
arch/x86/mm/mmap.c bool mmap_address_hint_valid(unsigned long addr, unsigned long len) · line 233
Quoted 10 lines
Signed-off-by Ralph Campbell <[email protected]>arch/x86/mm/mmap.c +1
-1@@ -230,7 +230,7 @@ bool mmap_address_hint_valid(unsigned long addr, unsigned long len)/* Can we access it for direct reading/writing? Must be RAM: */ int valid_phys_addr_range (phys_addr_t addr,size_t count) { -return addr + count <=__pa (high_memory); +return addr + count <=__pa_nodebug (high_memory);
This lacks a comment. Aside of that I think there is no point in using __pa(high_memory) here. This is all about the physical address range. So this can be simply expressed via:
return addr + count <= max_pfn * PAGE_SIZE;
which is much more obvious.
Thanks,
tglx
Re: [PATCH 1/1] x86/mm: Fix limit mmap() of /dev/mem to valid physical addresses
Quoted 48 lines · Thomas Gleixner
Ralph,
Quoted 6 lines · On Mon, 18 Mar 2019, [email protected] wrote:
From: Ralph Campbell <[email protected]>
If
CONFIG_DEBUG_VIRTUALis enabled, a read or write to/dev/memcan trigger aVIRTUAL_BUG_ON()depending on the value ofhigh_memory. For example:Quoted 8 lines
read_mem()
valid_phys_addr_range(p=401f1550, count=8)__pa(high_memory) __phys_addr(x=ffffc88000000000) // __START_KERNEL_map = ffffffff80000000 // y = ffffc88000000000 - ffffffff80000000 VIRTUAL_BUG_ON(phys_addr_valid(400000000000)) // boot_cpu_data.x86_phys_bits=46I have no idea why all the irrelevant information in this example would be helpful, but after extracting the meat I think I know what you want to say.
Quoted 2 lines
Since by design
high_memoryis outside the range of valid physical addresses, use the non-error checking version__pa_nodebug(high_memory).
high_memoryis not outside the range of valid physical addresses by design. It's only outside when memory is populated right at the end of the physical address space.So what you really want to say in the changelog is:
valid_phys_addr_range()is used to sanity check the physical address range of an operation, e.g. access to/dev/mem. It uses__pa(high_memory)internally.If memory is populated at the end of the physical address space, then
__pa(high_memory)is outside of the physical address space because:high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;For the comparison in
valid_phys_addr_range()this is not an issue, but ifCONFIG_DEBUG_VIRTUALis enabled,__pa()maps to__phys_addr(), which verifies that the resulting physical address is within the valid physical address space of the CPU. So in the case that memory is populated at the end of the physical address space, this is not true and triggers aVIRTUAL_BUG_ON().Use ... instead, because ...
Quoted 11 lines
Quoted 1 line
Fixes be62a32044061cb4a3b70a10598e093f1319102e ("x86/mm: Limit mmap() ofPlease limit the sha1 to the first 12 characters.
Quoted 2 lines
/dev/memto valid physical addresses")No newline between Fixes and the rest please.
Quoted 1 line
Signed-off-by Ralph Campbell <[email protected]>
Thanks for the comments. I'll apply them and send a v2 when ready.
Quoted 13 lines
arch/x86/mm/mmap.c bool mmap_address_hint_valid(unsigned long addr, unsigned long len) · line 233
Quoted 8 lines
arch/x86/mm/mmap.c +1
-1@@ -230,7 +230,7 @@ bool mmap_address_hint_valid(unsigned long addr, unsigned long len)/* Can we access it for direct reading/writing? Must be RAM: */ int valid_phys_addr_range (phys_addr_t addr,size_t count) { -return addr + count <=__pa (high_memory); +return addr + count <=__pa_nodebug (high_memory);This lacks a comment. Aside of that I think there is no point in using
__pa(high_memory)here. This is all about the physical address range. So this can be simply expressed via:
Quoted 7 lines
return addr + count <= max_pfn * PAGE_SIZE;which is much more obvious.
Thanks,
tglx
This looks OK to me for x86_64 but looking at arch/x86/mm/init_32.c, initmem_init() sets high_memory based on highstart_pfn or max_low_pfn depending on CONFIG_HIGHMEM. Would using max_pfn in this case work?
Re: [PATCH 1/1] x86/mm: Fix limit mmap() of /dev/mem to valid physical addresses
Ralph,
Quoted 15 lines · Ralph Campbell
Quoted 14 lines · Thomas Gleixner
arch/x86/mm/mmap.c bool mmap_address_hint_valid(unsigned long addr, · line 234
Quoted 9 lines
arch/x86/mm/mmap.c +1
-1@@ -230,7 +230,7 @@ bool mmap_address_hint_valid(unsigned long addr, unsigned long len)/* Can we access it for direct reading/writing? Must be RAM: */ int valid_phys_addr_range (phys_addr_t addr,size_t count) { -return addr + count <=__pa (high_memory); +return addr + count <=__pa_nodebug (high_memory);This lacks a comment. Aside of that I think there is no point in using
__pa(high_memory)here. This is all about the physical address range. So this can be simply expressed via:
Quoted 7 lines
Quoted 3 lines
return addr + count <= max_pfn * PAGE_SIZE;which is much more obvious.
This looks OK to me for
x86_64but looking atarch/x86/mm/init_32.c,initmem_init()setshigh_memorybased onhighstart_pfnormax_low_pfndepending onCONFIG_HIGHMEM. Would usingmax_pfnin this case work?
Aargh. There is also numa_32.c ...
So __pa_nodebug(high_memory) should be fine, but looking at the other places which do __pa() on high_memory they all use __pa(high_memory - 1) which is more obvious that the nodebug thing.
Thanks,
tglx