[PATCH v3 03/13] arm64: improve CONFIG_STRICT_DEVMEM handling

[Ard Biesheuvel]

On 26 November 2014 at 10:30, Dave Young <dyoung at redhat.com> wrote:
> On 11/18/14 at 01:57pm, Ard Biesheuvel wrote:
>> Improve the handling of /dev/mem mappings under CONFIG_STRICT_DEVMEM by:
>> - allowing read-only access to parts of System RAM that are not
>>   considered memory by the kernel, this is mainly intended for exposing
>>   UEFI Configuration tables to userland;
>
> Ard, can you elabrate a bit? Are this for the acpi case because no dtb in
> procfs?
>

No, that is a different issue.

> Anyway I also think using /dev/mem looks not a good way to expose
> infomation to userspace.
>

This is for compatibility with existing tools like dmidecode and lshw
that already use /dev/mem.
However, on x86, the tables they access are in iomem resource ranges
that are accessible under CONFIG_STRICT_DEVMEM, whereas on arm64,
those regions are inaccessible.

So while our position is that using /dev/mem for *anything* is an
awful idea, this particular patch just tries to bring arm64 in line
with what currently existing tools expect.

-- 
Ard.



>> - avoid using non-cached mappings for those parts of System RAM, as it
>>   may result in mismatched attributes.
>>
>> Signed-off-by: Ard Biesheuvel <ard.biesheuvel at linaro.org>
>> ---
>>  arch/arm64/mm/mmap.c |  5 +++--
>>  arch/arm64/mm/mmu.c  | 15 ++++++++++++++-
>>  2 files changed, 17 insertions(+), 3 deletions(-)
>>
>> diff --git a/arch/arm64/mm/mmap.c b/arch/arm64/mm/mmap.c
>> index 1d73662f00ff..802cdf0df921 100644
>> --- a/arch/arm64/mm/mmap.c
>> +++ b/arch/arm64/mm/mmap.c
>> @@ -127,14 +127,15 @@ int valid_mmap_phys_addr_range(unsigned long pfn, size_t size)
>>  /*
>>   * devmem_is_allowed() checks to see if /dev/mem access to a certain address
>>   * is valid. The argument is a physical page number.  We mimic x86 here by
>> - * disallowing access to system RAM as well as device-exclusive MMIO regions.
>> + * disallowing access to system RAM that is in active use by the kernel, as
>> + * well as device-exclusive MMIO regions.
>>   * This effectively disable read()/write() on /dev/mem.
>>   */
>>  int devmem_is_allowed(unsigned long pfn)
>>  {
>>       if (iomem_is_exclusive(pfn << PAGE_SHIFT))
>>               return 0;
>> -     if (!page_is_ram(pfn))
>> +     if (!pfn_valid(pfn))
>>               return 1;
>>       return 0;
>>  }
>> diff --git a/arch/arm64/mm/mmu.c b/arch/arm64/mm/mmu.c
>> index 4d583aa9ff4e..93ba10838359 100644
>> --- a/arch/arm64/mm/mmu.c
>> +++ b/arch/arm64/mm/mmu.c
>> @@ -121,7 +121,7 @@ early_param("cachepolicy", early_cachepolicy);
>>  pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
>>                             unsigned long size, pgprot_t vma_prot)
>>  {
>> -     if (!pfn_valid(pfn))
>> +     if (!page_is_ram(pfn))
>>               return pgprot_noncached(vma_prot);
>>       else if (file->f_flags & O_SYNC)
>>               return pgprot_writecombine(vma_prot);
>> @@ -129,6 +129,19 @@ pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
>>  }
>>  EXPORT_SYMBOL(phys_mem_access_prot);
>>
>> +/*
>> + * This definition of phys_mem_access_prot_allowed() overrides
>> + * the __weak definition in drivers/char/mem.c
>> + */
>> +int phys_mem_access_prot_allowed(struct file *file, unsigned long pfn,
>> +                              unsigned long size, pgprot_t *prot)
>> +{
>> +     /* Disallow read-write access to reserved system RAM */
>> +     if ((pgprot_val(*prot) & PTE_WRITE) && page_is_ram(pfn))
>> +             return 0;
>> +     return 1;
>> +}
>> +
>>  static void __init *early_alloc(unsigned long sz)
>>  {
>>       void *ptr = __va(memblock_alloc(sz, sz));
>> --
>> 1.8.3.2
>>