ntb_transport: fix qp count bug
authorLogan Gunthorpe <logang@deltatee.com>
Mon, 5 Jun 2017 20:00:52 +0000 (14:00 -0600)
committerJon Mason <jdmason@kudzu.us>
Mon, 19 Jun 2017 18:24:41 +0000 (14:24 -0400)
In cases where there are more mw's than spads/2-2, the mw count gets
reduced to match the limitation. ntb_transport also tries to ensure that
there are fewer qps than mws but uses the full mw count instead of
the reduced one. When this happens, the math in
'ntb_transport_setup_qp_mw' will get confused and result in a kernel
paging request bug.

This patch fixes the bug by reducing qp_count to the reduced mw count
instead of the full mw count.

Signed-off-by: Logan Gunthorpe <logang@deltatee.com>
Fixes: e26a5843f7f5 ("NTB: Split ntb_hw_intel and ntb_transport drivers")
Acked-by: Allen Hubbe <Allen.Hubbe@dell.com>
Signed-off-by: Jon Mason <jdmason@kudzu.us>
drivers/ntb/ntb_transport.c

index 02ca45fdd89203f31246f552811b1264059232d6..0a778d2cab94f9d2658f66568b92423b9a0dd329 100644 (file)
@@ -1128,8 +1128,8 @@ static int ntb_transport_probe(struct ntb_client *self, struct ntb_dev *ndev)
        qp_count = ilog2(qp_bitmap);
        if (max_num_clients && max_num_clients < qp_count)
                qp_count = max_num_clients;
-       else if (mw_count < qp_count)
-               qp_count = mw_count;
+       else if (nt->mw_count < qp_count)
+               qp_count = nt->mw_count;
 
        qp_bitmap &= BIT_ULL(qp_count) - 1;