[PATCH v2 09/28] ARM: mmu: Specify size in bytes in create_sections()

[Andrey Smirnov]

On Thu, May 17, 2018 at 12:08 AM, Sascha Hauer <s.hauer at pengutronix.de> wrote:
> On Thu, May 17, 2018 at 12:01:34AM -0700, Andrey Smirnov wrote:
>> On Wed, May 16, 2018 at 11:55 PM, Sascha Hauer <s.hauer at pengutronix.de> wrote:
>> > On Wed, May 16, 2018 at 01:00:17PM -0700, Andrey Smirnov wrote:
>> >> Seeing
>> >>
>> >>         create_sections(ttb, 0, PAGE_SIZE, ...);
>> >>
>> >> as the code the creates initial flat 4 GiB mapping is a bit less
>> >> intuitive then
>> >>
>> >>         create_sections(ttb, 0, SZ_4G, ...);
>> >>
>> >> so, for the sake of clarification, convert create_sections() to accept
>> >> size in bytes and do bytes -> MiB converstion as a part of the
>> >> function.
>> >>
>> >> NOTE: To keep all of the arguments of create_sections() 32-bit:
>> >>
>> >>     - Move all of the real code into a helper function accepting first
>> >>       and last addresses of the region (e.g passing 0 and U32_MAX
>> >>       means all 4GiB of address space)
>> >>
>> >>     - Convert create_section() into a macro that does necessary size
>> >>       -> last addres conversion under the hood to preserve original API
>> >>
>> >> Signed-off-by: Andrey Smirnov <andrew.smirnov at gmail.com>
>> >> ---
>> >>  arch/arm/cpu/mmu-early.c |  4 ++--
>> >>  arch/arm/cpu/mmu.c       |  4 ++--
>> >>  arch/arm/cpu/mmu.h       | 22 ++++++++++++++++------
>> >>  3 files changed, 20 insertions(+), 10 deletions(-)
>> >>
>> >> diff --git a/arch/arm/cpu/mmu-early.c b/arch/arm/cpu/mmu-early.c
>> >> index 70ece0d2f..136b33c3a 100644
>> >> --- a/arch/arm/cpu/mmu-early.c
>> >> +++ b/arch/arm/cpu/mmu-early.c
>> >> @@ -16,7 +16,7 @@ static void map_cachable(unsigned long start, unsigned long size)
>> >>       start = ALIGN_DOWN(start, SZ_1M);
>> >>       size  = ALIGN(size, SZ_1M);
>> >>
>> >> -     create_sections(ttb, start, size >> 20, PMD_SECT_AP_WRITE |
>> >> +     create_sections(ttb, start, size, PMD_SECT_AP_WRITE |
>> >>                       PMD_SECT_AP_READ | PMD_TYPE_SECT | PMD_SECT_WB);
>> >>  }
>> >>
>> >> @@ -30,7 +30,7 @@ void mmu_early_enable(unsigned long membase, unsigned long memsize,
>> >>       set_ttbr(ttb);
>> >>       set_domain(DOMAIN_MANAGER);
>> >>
>> >> -     create_sections(ttb, 0, 4096, PMD_SECT_AP_WRITE |
>> >> +     create_sections(ttb, 0, SZ_4G, PMD_SECT_AP_WRITE |
>> >>                       PMD_SECT_AP_READ | PMD_TYPE_SECT);
>> >>
>> >>       map_cachable(membase, memsize);
>> >> diff --git a/arch/arm/cpu/mmu.c b/arch/arm/cpu/mmu.c
>> >> index 0c367e47c..f02c99f65 100644
>> >> --- a/arch/arm/cpu/mmu.c
>> >> +++ b/arch/arm/cpu/mmu.c
>> >> @@ -460,7 +460,7 @@ static int mmu_init(void)
>> >>       set_domain(DOMAIN_MANAGER);
>> >>
>> >>       /* create a flat mapping using 1MiB sections */
>> >> -     create_sections(ttb, 0, PAGE_SIZE, PMD_SECT_AP_WRITE | PMD_SECT_AP_READ |
>> >> +     create_sections(ttb, 0, SZ_4G, PMD_SECT_AP_WRITE | PMD_SECT_AP_READ |
>> >>                       PMD_TYPE_SECT);
>> >>       __mmu_cache_flush();
>> >>
>> >> @@ -472,7 +472,7 @@ static int mmu_init(void)
>> >>        * below
>> >>        */
>> >>       for_each_memory_bank(bank) {
>> >> -             create_sections(ttb, bank->start, bank->size >> 20,
>> >> +             create_sections(ttb, bank->start, bank->size,
>> >>                               PMD_SECT_DEF_CACHED);
>> >>               __mmu_cache_flush();
>> >>       }
>> >> diff --git a/arch/arm/cpu/mmu.h b/arch/arm/cpu/mmu.h
>> >> index d71cd7e38..52689359a 100644
>> >> --- a/arch/arm/cpu/mmu.h
>> >> +++ b/arch/arm/cpu/mmu.h
>> >> @@ -27,16 +27,26 @@ static inline void set_domain(unsigned val)
>> >>  }
>> >>
>> >>  static inline void
>> >> -create_sections(uint32_t *ttb, unsigned long addr,
>> >> -             int size_m, unsigned int flags)
>> >> +__create_sections(uint32_t *ttb, unsigned long first,
>> >> +               unsigned long last, unsigned int flags)
>> >>  {
>> >> -     unsigned long ttb_start = addr >> 20;
>> >> -     unsigned long ttb_end   = ttb_start + size_m;
>> >> -     unsigned int i;
>> >> +     unsigned long ttb_start = first >> 20;
>> >> +     unsigned long ttb_end   = (last >> 20) + 1;
>> >> +     unsigned int i, addr;
>> >>
>> >> -     for (i = ttb_start; i < ttb_end; i++, addr += SZ_1M)
>> >> +     for (i = ttb_start, addr = first; i < ttb_end; i++, addr += SZ_1M)
>> >>               ttb[i] = addr | flags;
>> >>  }
>> >>
>> >> +#define create_sections(ttb, addr, size, flags)                              \
>> >> +     ({                                                              \
>> >> +             typeof(addr) __addr = addr;                             \
>> >> +             typeof(size) __size = size;                             \
>> >> +             /* Check for overflow */                                \
>> >> +             BUG_ON(__addr > ULONG_MAX - __size + 1);                \
>> >> +             __create_sections(ttb, __addr, __addr + (size)  - 1,    \
>> >> +                               flags);                               \
>> >> +     })
>> >
>> > Why do you preserve the original API of create_sections()? I would just
>> > change it. We have only a few callers and they are easy to change.
>> >
>>
>> Mostly because it keeps "addr + size - 1" arithmetic in one place
>> instead of spreading it to every caller. If you'd rather have that
>> instead of the macro above, I can change that in v3.
>
> I agree that the end address calculation often is a source for
> off-by-one errors, still I would prefer not using macros to hide that.
>

I can't think of a way to keep arithmetic in one place and not to make
one of the arguments 64-bit at the same time without resorting to a
macro. AFAICT the options are either to drop the macro and change the
API or keep the API/arithmetic in one place and keep the macro.

I am happy with either, so let me know what you'd prefer.

Thanks,
Andrey Smirnov