[openib-general] [PATCH v2 1/7] IB/core - Add DMA mapping functions to allow device drivers to interpose

[Or Gerlitz]

Ralph Campbell wrote:
> Basically, use a hash table to store the kmap result.
> See attached for 90% of the code.
> static u64 ipath_dma_map_page(struct ib_device *dev,
> 			      struct page *page,
> 			      unsigned long offset,
> 			      size_t size,
> 			      enum dma_data_direction direction)
> {
> 	u64 addr;
> 
> 	BUG_ON(!valid_dma_direction(direction));
> 
> 	if (offset + size > PAGE_SIZE) {
> 		addr = BAD_DMA_ADDRESS;
> 		goto done;
> 	}
> 
> #ifdef CONFIG_HIGHMEM
> 	/* handle highmem pages */
> 	if (PageHighMem(page)) {
> 		void *v = kmap(page);

another comment we have got on iser, is that this code can be called 
context that requires kmap_atomic (and xxx_dma_unmap_page in a context 
that requires kunmap_atomic). This imposes another problem, since the 
kmap_atomic slots are somehow limited and with this patch the ipath 
driver would hold those mapping for relatively long time (ie it does not 
kmap/copy/kunmap).

> 
> 		if (!v)
> 			addr = BAD_DMA_ADDRESS;
> 		else {
> 			addr = (u64) v + offset;
> 			hash_insert(dev, v + offset, page);
> 		}
> 		goto done;
> 	}
> #endif
> 	addr = (u64) page_address(page);
> 	if (addr)
> 		addr += offset;
> 
> done:
> 	return addr;
> }