On 6/25/2015 10:29 PM, Jason Gunthorpe wrote:
On Thu, Jun 25, 2015 at 02:25:49PM -0500, Steve Wise wrote:
To stage the changes we could introduce a new function that returns
the needed ib_access_flags value given the desired opcodes. Then
have a series that changes all the existing ULPs to make use of this
new function.
I wouldn't be afraid to add a new create_mr entry point that does the
right thing, we can unexport/delete the old one when all kernel users
are gone. Trivially the core can just have a default that translates
based on iwarp/!iwarp for now.
Note that we have ib_create_mr() that receives an easily extensible
ib_mr_init_attr (similar to qp creation). I think this is a good
opportunity to unite all the create routines to ib_create_mr.
ib_get_dma_mr will become:
ib_mr_init_attr.type = IB_MR_DMA;
ib_mr_init_attr.mr_access_flags = mr_access_flags;
ib_create_mr(pd, &ib_mr_init_attr);
ib_alloc_fast_reg_mr will become:
ib_mr_init_attr.type = IB_MR_FAST_REG;
ib_create_mr(pd, &ib_mr_init_attr);
ib_reg_phys_mr (if we don't want to kill it by now) will become:
ib_mr_init_attr.type = IB_MR_REG_PHYS;
ib_mr_init_attr.arr = phys_buf_array;
ib_mr_init_attr.max_reg_descriptors = num_phys_buf;
ib_mr_init_attr.mr_access_flags = mr_access_flags;
ib_mr_init_attr.iova_start = iova_start;
ib_create_mr(pd, &ib_mr_init_attr);
The core layer will spread these in the different driver routines
(although it would be nice if the drivers can have a single MR craetion
routine that can handle all MR types (including fail if unsupported).
Thoughts?
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
