On 9/21/2020 6:26 PM, Jason Gunthorpe wrote:
On Mon, Sep 21, 2020 at 06:17:51PM +0300, Yishai Hadas wrote:
On 9/21/2020 5:23 PM, Jason Gunthorpe wrote:
On Thu, Sep 17, 2020 at 02:21:49PM +0300, Leon Romanovsky wrote:
-out:
- put_page(page);
- return ret;
+ *dma_addr &= ODP_DMA_ADDR_MASK;
I thought we agreed to get rid of this because it must be zero?
Jason
At the end we agreed to not rely on as potentially dma_addr may be zero, the
access flags are used to mark an entry as valid in this case (see also
ib_umem_odp_unmap_dma_pages).
I mean the mask, that masking is always a NOP or it is really broken
Jason
I tend to agree that this is redundant as the invalidation flow should
to the cleanup in case was some change in access from previous setting.
However,
Don't we want to have it for clearness at least as of the non-faulting
mode (i.e. snapshot) that we just add in this series which get current
CPU page access and sets it ?
Yishai
