On Mon, Mar 8, 2021 at 8:44 AM Jason Gunthorpe [off-list ref] wrote:
On Sun, Mar 07, 2021 at 10:28:33PM +0800, Zhu Yanjun wrote:
quoted
From: Zhu Yanjun <zyjzyj2000@gmail.com>
After the commit ("RDMA/umem: Move to allocate SG table from pages"),
the sg list from ib_ume_get is like the following:
If you are not getting huge pages then you need to fix
__sg_alloc_table_from_pages()
But as far as I know it is not buggy like you are describing.
In the same host and the same test, both __sg_alloc_table_from_pages
and ib_umem_add_sg_table
are called at the same time, but the returned sg lists are different.
It is weird.
Zhu Yanjun
quoted
The new function is ib_umem_get to ib_umem_hugepage_get that calls
ib_umem_add_sg_table.
Nope
Jason