From: Håkon Bugge haakon.bugge@oracle.com
[ Upstream commit 2612d723aadcf8281f9bf8305657129bd9f3cd57 ]
Using CX-3 virtual functions, either from a bare-metal machine or pass-through from a VM, MAD packets are proxied through the PF driver.
Since the VF drivers have separate name spaces for MAD Transaction Ids (TIDs), the PF driver has to re-map the TIDs and keep the book keeping in a cache.
Following the RDMA Connection Manager (CM) protocol, it is clear when an entry has to evicted form the cache. But life is not perfect, remote peers may die or be rebooted. Hence, it's a timeout to wipe out a cache entry, when the PF driver assumes the remote peer has gone.
During workloads where a high number of QPs are destroyed concurrently, excessive amount of CM DREQ retries has been observed
The problem can be demonstrated in a bare-metal environment, where two nodes have instantiated 8 VFs each. This using dual ported HCAs, so we have 16 vPorts per physical server.
64 processes are associated with each vPort and creates and destroys one QP for each of the remote 64 processes. That is, 1024 QPs per vPort, all in all 16K QPs. The QPs are created/destroyed using the CM.
When tearing down these 16K QPs, excessive CM DREQ retries (and duplicates) are observed. With some cat/paste/awk wizardry on the infiniband_cm sysfs, we observe as sum of the 16 vPorts on one of the nodes:
cm_rx_duplicates: dreq 2102 cm_rx_msgs: drep 1989 dreq 6195 rep 3968 req 4224 rtu 4224 cm_tx_msgs: drep 4093 dreq 27568 rep 4224 req 3968 rtu 3968 cm_tx_retries: dreq 23469
Note that the active/passive side is equally distributed between the two nodes.
Enabling pr_debug in cm.c gives tons of:
[171778.814239] <mlx4_ib> mlx4_ib_multiplex_cm_handler: id{slave: 1,sl_cm_id: 0xd393089f} is NULL!
By increasing the CM_CLEANUP_CACHE_TIMEOUT from 5 to 30 seconds, the tear-down phase of the application is reduced from approximately 90 to 50 seconds. Retries/duplicates are also significantly reduced:
cm_rx_duplicates: dreq 2460 [] cm_tx_retries: dreq 3010 req 47
Increasing the timeout further didn't help, as these duplicates and retries stems from a too short CMA timeout, which was 20 (~4 seconds) on the systems. By increasing the CMA timeout to 22 (~17 seconds), the numbers fell down to about 10 for both of them.
Adjustment of the CMA timeout is not part of this commit.
Signed-off-by: Håkon Bugge haakon.bugge@oracle.com Acked-by: Jack Morgenstein jackm@dev.mellanox.co.il Signed-off-by: Jason Gunthorpe jgg@mellanox.com Signed-off-by: Sasha Levin sashal@kernel.org --- drivers/infiniband/hw/mlx4/cm.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/infiniband/hw/mlx4/cm.c b/drivers/infiniband/hw/mlx4/cm.c index 56a593e0ae5d..2aedbd18ffbc 100644 --- a/drivers/infiniband/hw/mlx4/cm.c +++ b/drivers/infiniband/hw/mlx4/cm.c @@ -39,7 +39,7 @@
#include "mlx4_ib.h"
-#define CM_CLEANUP_CACHE_TIMEOUT (5 * HZ) +#define CM_CLEANUP_CACHE_TIMEOUT (30 * HZ)
struct id_map_entry { struct rb_node node;