lkml.org 
[lkml]   [2011]   [Sep]   [28]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
    /
    From
    SubjectRe: [PATCH v3 08/24] C6X: memory management and DMA support
    Date
    On Tuesday 27 September 2011, Mark Salter wrote:

    > +dma_addr_t dma_map_single(struct device *dev, void *ptr, size_t size,
    > + enum dma_data_direction dir)
    > +{
    > + dma_addr_t addr = virt_to_phys(ptr);
    > +
    > + BUG_ON(!valid_dma_direction(dir));
    > +
    > + switch (dir) {
    > + case DMA_FROM_DEVICE:
    > + L2_cache_block_invalidate(addr, addr + size);
    > + break;
    > + case DMA_TO_DEVICE:
    > + L2_cache_block_writeback(addr, addr + size);
    > + break;
    > + case DMA_BIDIRECTIONAL:
    > + /*
    > + * If area will be used bidirectionally, syncing will
    > + * take place manually in the driver.
    > + */

    I think this comment is not strictly correct and you have to do a
    writeback+invalidate here. When you map an area, it initially
    belongs to the device, until you either do an unmap or sync_*_for_cpu.

    > +void dma_unmap_single(struct device *dev, dma_addr_t handle,
    > + size_t size, enum dma_data_direction dir)
    > +{
    > + /* nothing to do here */
    > + debug_dma_unmap_page(dev, handle, size, dir, true);
    > +}
    > +EXPORT_SYMBOL(dma_unmap_single);

    > +void dma_sync_single_for_cpu(struct device *dev, dma_addr_t handle,
    > + size_t size, enum dma_data_direction dir)
    > +{
    > + unsigned long paddr = handle;
    > +
    > + BUG_ON(!valid_dma_direction(dir));
    > +
    > + switch (dir) {
    > + case DMA_FROM_DEVICE:
    > + L2_cache_block_invalidate(paddr, paddr + size);
    > + break;
    > + case DMA_TO_DEVICE:
    > + L2_cache_block_writeback(paddr, paddr + size);
    > + break;
    > + case DMA_BIDIRECTIONAL:
    > + L2_cache_block_writeback_invalidate(paddr, paddr + size);
    > + break;
    > + default:
    > + break;
    > + }
    > +
    > + debug_dma_sync_single_for_cpu(dev, handle, size, dir);
    > +}
    > +EXPORT_SYMBOL(dma_sync_single_for_cpu);

    Normally, sync_*_for_cpu is the same as unmap regarding the cache.

    > +void dma_sync_single_for_device(struct device *dev, dma_addr_t handle,
    > + size_t size, enum dma_data_direction dir)
    > +{
    > + unsigned long paddr = handle;
    > +
    > + BUG_ON(!valid_dma_direction(dir));
    > +
    > + /* don't bother invalidating if DMA to device */
    > + if (dir != DMA_TO_DEVICE)
    > + L2_cache_block_invalidate(paddr, paddr + size);
    > +
    > + debug_dma_sync_single_for_device(dev, handle, size, dir);
    > +}
    > +EXPORT_SYMBOL(dma_sync_single_for_device);

    And this should do the same as map(). In particular it needs
    to writeback the cache when direction is TO_DEVICE or BIDIRECTIONAL.

    Arnd


    \
     
     \ /
      Last update: 2011-09-28 15:49    [W:4.417 / U:0.124 seconds]
    ©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site