[RFC PATCH v3 0/5] ARM: Fix dma_alloc_coherent() and friends for NOMMU

[Vladimir Murzin]

On 10/01/17 13:13, Benjamin Gaignard wrote:
> 2017-01-09 14:47 GMT+01:00 Vladimir Murzin <vladimir.murzin at arm.com>:
>> Hi,
>>
>> It seem that addition of cache support for M-class cpus uncovered
>> latent bug in DMA usage. NOMMU memory model has been treated as being
>> always consistent; however, for R/M classes of cpu memory can be
>> covered by MPU which in turn might configure RAM as Normal
>> i.e. bufferable and cacheable. It breaks dma_alloc_coherent() and
>> friends, since data can stuck in caches now or be buffered.
>>
>> This patch set is trying to address the issue by providing region of
>> memory suitable for consistent DMA operations. It is supposed that
>> such region is marked by MPU as non-cacheable. Robin suggested to
>> advertise such memory as reserved shared-dma-pool, rather then using
>> homebrew command line option, and extend dma-coherent to provide
>> default DMA area in the similar way as it is done for CMA (PATCH
>> 2/5). It allows us to offload all bookkeeping on generic coherent DMA
>> framework, and it is seems that it might be reused by other
>> architectures like c6x and blackfin.
>>
>> Dedicated DMA region is required for cases other than:
>>  - MMU/MPU is off
>>  - cpu is v7m w/o cache support
>>  - device is coherent
>>
>> In case one of the above conditions is true dma operations are forced
>> to be coherent and wired with dma_noop_ops.
>>
>> To make life easier NOMMU dma operations are kept in separate
>> compilation unit.
>>
>> Since the issue was reported in the same time as Benjamin sent his
>> patch [1] to allow mmap for NOMMU, his case is also addressed in this
>> series (PATCH 1/5 and PATCH 3/5).
>>
>> @Benjamin I've tested that mmap is working with amba-clcd and following
>> hack on top:
>>
>> diff --git a/drivers/video/fbdev/core/fbmem.c b/drivers/video/fbdev/core/fbmem.c
>> index 76c1ad9..64465c9 100644
>> --- a/drivers/video/fbdev/core/fbmem.c
>> +++ b/drivers/video/fbdev/core/fbmem.c
>> @@ -1492,6 +1492,24 @@ __releases(&info->lock)
>>         return 0;
>>  }
>>
>> +static unsigned fb_capabilities(struct file *file)
>> +{
>> +       return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE;
>> +}
>> +
>> +static unsigned long get_fb_unmapped_area(struct file *filp,
>> +                                  unsigned long addr, unsigned long len,
>> +                                  unsigned long pgoff, unsigned long flags)
>> +{
>> +       struct fb_info * const info = filp->private_data;
>> +       unsigned long fb_size = PAGE_ALIGN(info->fix.smem_len);
>> +
>> +       if (pgoff > fb_size || len > fb_size - pgoff)
>> +               return -EINVAL;
>> +
>> +       return (unsigned long)info->screen_base + pgoff;
>> +}
>> +
>>  static const struct file_operations fb_fops = {
>>         .owner =        THIS_MODULE,
>>         .read =         fb_read,
>> @@ -1503,13 +1521,12 @@ static const struct file_operations fb_fops = {
>>         .mmap =         fb_mmap,
>>         .open =         fb_open,
>>         .release =      fb_release,
>> -#ifdef HAVE_ARCH_FB_UNMAPPED_AREA
>>         .get_unmapped_area = get_fb_unmapped_area,
>> -#endif
>>  #ifdef CONFIG_FB_DEFERRED_IO
>>         .fsync =        fb_deferred_io_fsync,
>>  #endif
>>         .llseek =       default_llseek,
>> +       .mmap_capabilities = fb_capabilities,
>>  };
>>
>>  struct class *fb_class;
>>
>> I can see that fb-test-app updates display and addresses returnend with
>> dma_alloc_cohernet() and mmap() are the same.
> 
> I have push get_fb_unmapped_area() function in fbmem last week:
> https://cgit.freedesktop.org/drm/drm-misc/commit/?id=82f42e4cc164ed486c9e2b1b74e65b176830d947
> and to the same in drm/kms part:
> https://cgit.freedesktop.org/drm/drm-misc/commit/?id=62a0d98a188cc4ebd8ea54b37d274ec20465e464
> all this work without need to change mmap_capabilities.
> 
> Other noMMU architectures don't need to change mmap_capabilities
> function to make fbmem works.
> I would like to understand how they do before push for this new function.
> It sound like it is an ARM Cortex M specific needs with your new implementation.
> 
> If we can't avoid using mmap_capabilities we will have to put it (at
> least) in fbmem, drm/kms and v4l2.

I've just tried with your fbmem patch only and it still working, display is
updated with what fb-test-app draws. Anyway, my point was that mmap
implementation introduced with these patches works for me and it'd be nice to
know if it works for you as well ;)

Meanwhile, Andras pointed out that fix from v2->v3 was missed, so I'm
resubmitting v4 shortly.

Cheers
Vladimir

> 
> Benjamin
> 
>> Thanks!
>>
>> [1] http://www.armlinux.org.uk/developer/patches/viewpatch.php?id=8633/1
>>
>> Vladimir Murzin (5):
>>   dma: Add simple dma_noop_mmap
>>   drivers: dma-coherent: Introduce default DMA pool
>>   ARM: NOMMU: Introduce dma operations for noMMU
>>   ARM: NOMMU: Set ARM_DMA_MEM_BUFFERABLE for M-class cpus
>>   ARM: dma-mapping: Remove traces of NOMMU code
>>
>>  .../bindings/reserved-memory/reserved-memory.txt   |   3 +
>>  arch/arm/include/asm/dma-mapping.h                 |   3 +-
>>  arch/arm/mm/Kconfig                                |   2 +-
>>  arch/arm/mm/Makefile                               |   5 +-
>>  arch/arm/mm/dma-mapping-nommu.c                    | 252 +++++++++++++++++++++
>>  arch/arm/mm/dma-mapping.c                          |  26 +--
>>  drivers/base/dma-coherent.c                        |  59 ++++-
>>  lib/dma-noop.c                                     |  21 ++
>>  8 files changed, 335 insertions(+), 36 deletions(-)
>>  create mode 100644 arch/arm/mm/dma-mapping-nommu.c
>>
>> --
>> 2.0.0
>>
> 
> 
>