Re: [PATCH 1/2] dma-mapping: add a dma_ops_bypass flag to, struct device

Hi All,

The DMA mapping works great on our PowerPC machines currently. It was a 
long way to get the new DMA mapping code to work successfully on our 
PowerPC machines.

P L E A S E  don't modify the good working DMA mapping code. There are 
many other topics which needs improvements. For us (first level + second 
level support) it is really laborious to find your problematic code and 
patch it. It takes a long time to find the problematic code because we 
have to do it besides our main work.

P L E A S E test your code on PowerPC machines before you add it to the 
mainline vanilla kernel.

Thanks,
Christian


On Tue, Mar 24, 2020 at 12:00:09PM +0530, Aneesh Kumar K.V wrote:
 > dma_addr_t dma_direct_map_page(struct device *dev, struct page *page,
 >         unsigned long offset, size_t size, enum dma_data_direction dir,
 >         unsigned long attrs)
 > {
 >     phys_addr_t phys = page_to_phys(page) + offset;
 >     dma_addr_t dma_addr = phys_to_dma(dev, phys);
 >
 >     if (unlikely(!dma_capable(dev, dma_addr, size, true))) {
 >             return iommu_map(dev, phys, size, dir, attrs);
 >
 >         return DMA_MAPPING_ERROR;

If powerpc hardware / firmware people really come up with crap that
stupid you'll have to handle it yourself and will always pay the
indirect call penality.