Add CM packets missing and harden the proxying
Subsystems (2)
cover Add CM packets missing and harden the proxying
A high number of MAD packet drops are observed in the mlx4 MAD proxy system. These are fixed by separating the parameters for the tunnel vs. wire QPs and by introducing a separate worker-thread for the wire QPs.
Support for MRA and REJ with its reason being timeout is also added.
Dynamic debug prints adjusted and amended.
Håkon Bugge (5):
IB/mlx4: Add and improve logging
IB/mlx4: Add support for MRA
IB/mlx4: Separate tunnel and wire bufs parameters IB/mlx4: Fix starvation in paravirt mux/demux IB/mlx4: Add support for REJ due to timeout
drivers/infiniband/hw/mlx4/cm.c | 143 ++++++++++++++++++++++++++- drivers/infiniband/hw/mlx4/mad.c | 121 +++++++++++++---------- drivers/infiniband/hw/mlx4/mlx4_ib.h | 6 ++ 3 files changed, 211 insertions(+), 59 deletions(-)
--
2.20.1
PATCH 1/5 IB/mlx4: Add and improve logging
Add missing check for success after call to mlx4_ib_send_to_wire() in mlx4_ib_multiplex_mad().
Amended the existing pr_debug() in mlx4_ib_multiplex_cm_handler() and mlx4_ib_demux_cm_handler() with attr_id during a lookup failure.
Removed two noisy pr_debug() in mad.c
--- drivers/infiniband/hw/mlx4/cm.c | 7 +++--- drivers/infiniband/hw/mlx4/mad.c | 43 +++++++++----------------------- 2 files changed, 16 insertions(+), 34 deletions(-)
drivers/infiniband/hw/mlx4/cm.c +4 -3
@@ -314,8 +314,8 @@ int mlx4_ib_multiplex_cm_handler(struct ib_device *ibdev, int port, int slave_id }if (!id) { -pr_debug ("id{slave: %d, sl_cm_id: 0x%x} is NULL!\n" , - slave_id, sl_cm_id); +pr_debug ("id{slave: %d, sl_cm_id: 0x%x} is NULL! attr_id: 0x%x\n" , + slave_id, sl_cm_id,be16_to_cpu (mad->mad_hdr.attr_id));return -EINVAL ; }
@@ -354,7 +354,8 @@ int mlx4_ib_demux_cm_handler(struct ib_device *ibdev, int port, int *slave, id =id_map_get (ibdev, (int *)&pv_cm_id,-1 ,-1 );if (!id) { -pr_debug ("Couldn't find an entry for pv_cm_id 0x%x\n" , pv_cm_id); +pr_debug ("Couldn't find an entry for pv_cm_id 0x%x, attr_id 0x%x\n" , + pv_cm_id,be16_to_cpu (mad->mad_hdr.attr_id));return -ENOENT ; }
drivers/infiniband/hw/mlx4/mad.c +12 -31
67 lines not on this page · raw message
Signature · 1 line
2.20.1
PATCH 2/5 IB/mlx4: Add support for MRA
In reply to Håkon BuggeUsing CX-3 in virtualized mode, MAD packets are proxied through the PF driver. However, the handling lacks support of the MRA (Message Receipt Acknowledgment) packet. When having dynamic debug enabled, we see tons of:
mlx4_ib_multiplex_cm_handler: id{slave: 7, sl_cm_id: 0x8fcb45a0} is NULL! attr_id: 0x11
--- drivers/infiniband/hw/mlx4/cm.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)
drivers/infiniband/hw/mlx4/cm.c +3 -2
@@ -293,8 +293,9 @@ int mlx4_ib_multiplex_cm_handler(struct ib_device *ibdev, int port, int slave_idint pv_cm_id =-1 ;if (mad->mad_hdr.attr_id ==CM_REQ_ATTR_ID || - mad->mad_hdr.attr_id ==CM_REP_ATTR_ID || - mad->mad_hdr.attr_id ==CM_SIDR_REQ_ATTR_ID ) { + mad->mad_hdr.attr_id ==CM_REP_ATTR_ID || + mad->mad_hdr.attr_id ==CM_MRA_ATTR_ID || + mad->mad_hdr.attr_id ==CM_SIDR_REQ_ATTR_ID ) { sl_cm_id =get_local_comm_id (mad); id =id_map_get (ibdev, &pv_cm_id, slave_id, sl_cm_id);if (id)
Signature · 1 line
2.20.1
PATCH 3/5 IB/mlx4: Separate tunnel and wire bufs parameters
In reply to Håkon BuggeUsing CX-3 in virtualized mode, MAD packets are proxied through the PF driver. The feed is N tunnel QPs, and what is received from the VFs is multiplexed out on the wire QP. Since this is a many-to-one, it is better to have separate initialization parameters for the two usages.
Yanking the number of wire buffers up to 2K.
--- drivers/infiniband/hw/mlx4/mad.c | 44 +++++++++++++++------------- drivers/infiniband/hw/mlx4/mlx4_ib.h | 1 + 2 files changed, 25 insertions(+), 20 deletions(-)
drivers/infiniband/hw/mlx4/mad.c +24 -20
@@ -1380,10 +1380,10 @@ int mlx4_ib_send_to_wire(struct mlx4_ib_dev *dev, int slave, u8 port,spin_lock (&sqp->tx_lock);if (sqp->tx_ix_head - sqp->tx_ix_tail >= - (MLX4_NUM_TUNNEL_BUFS -1 )) + (MLX4_NUM_WIRE_BUFS -1 )) ret = -EAGAIN ;else - wire_tx_ix = (++sqp->tx_ix_head) & (MLX4_NUM_TUNNEL_BUFS -1 ); + wire_tx_ix = (++sqp->tx_ix_head) & (MLX4_NUM_WIRE_BUFS -1 );spin_unlock (&sqp->tx_lock);if (ret) goto out;
@@ -1579,19 +1579,20 @@ static int mlx4_ib_alloc_pv_bufs(struct mlx4_ib_demux_pv_ctx *ctx,int i;struct mlx4_ib_demux_pv_qp *tun_qp;int rx_buf_size, tx_buf_size; +const int nmbr_bufs = is_tun ?MLX4_NUM_TUNNEL_BUFS :MLX4_NUM_WIRE_BUFS ;if (qp_type >IB_QPT_GSI )return -EINVAL ; tun_qp = &ctx->qp[qp_type]; - tun_qp->ring =kcalloc (MLX4_NUM_TUNNEL_BUFS , + tun_qp->ring =kcalloc (nmbr_bufs,sizeof (struct mlx4_ib_buf ),GFP_KERNEL );if (!tun_qp->ring)return -ENOMEM ; - tun_qp->tx_ring =kcalloc (MLX4_NUM_TUNNEL_BUFS , + tun_qp->tx_ring =kcalloc (nmbr_bufs,sizeof (struct mlx4_ib_tun_tx_buf ),GFP_KERNEL );if (!tun_qp->tx_ring) {
@@ -1608,7 +1609,7 @@ static int mlx4_ib_alloc_pv_bufs(struct mlx4_ib_demux_pv_ctx *ctx, tx_buf_size =sizeof (struct mlx4_mad_snd_buf ); } -for (i =0 ; i <MLX4_NUM_TUNNEL_BUFS ; i++) { +for (i =0 ; i < nmbr_bufs; i++) { tun_qp->ring[i].addr =kmalloc (rx_buf_size,GFP_KERNEL );if (!tun_qp->ring[i].addr) goto err;
@@ -1622,7 +1623,7 @@ static int mlx4_ib_alloc_pv_bufs(struct mlx4_ib_demux_pv_ctx *ctx, } } -for (i =0 ; i <MLX4_NUM_TUNNEL_BUFS ; i++) { +for (i =0 ; i < nmbr_bufs; i++) { tun_qp->tx_ring[i].buf.addr =kmalloc (tx_buf_size,GFP_KERNEL );if (!tun_qp->tx_ring[i].buf.addr)
@@ -1653,7 +1654,7 @@ static int mlx4_ib_alloc_pv_bufs(struct mlx4_ib_demux_pv_ctx *ctx, tx_buf_size, DMA_TO_DEVICE);kfree (tun_qp->tx_ring[i].buf.addr); } - i =MLX4_NUM_TUNNEL_BUFS ; + i = nmbr_bufs; err:while (i >0 ) { --i;
@@ -1674,6 +1675,7 @@ static void mlx4_ib_free_pv_qp_bufs(struct mlx4_ib_demux_pv_ctx *ctx,int i;struct mlx4_ib_demux_pv_qp *tun_qp;int rx_buf_size, tx_buf_size; +const int nmbr_bufs = is_tun ?MLX4_NUM_TUNNEL_BUFS :MLX4_NUM_WIRE_BUFS ;if (qp_type >IB_QPT_GSI )return ;
@@ -1688,13 +1690,13 @@ static void mlx4_ib_free_pv_qp_bufs(struct mlx4_ib_demux_pv_ctx *ctx, } -for (i =0 ; i <MLX4_NUM_TUNNEL_BUFS ; i++) { +for (i =0 ; i < nmbr_bufs; i++) {ib_dma_unmap_single (ctx->ib_dev, tun_qp->ring[i].map, rx_buf_size,DMA_FROM_DEVICE );kfree (tun_qp->ring[i].addr); } -for (i =0 ; i <MLX4_NUM_TUNNEL_BUFS ; i++) { +for (i =0 ; i < nmbr_bufs; i++) {ib_dma_unmap_single (ctx->ib_dev, tun_qp->tx_ring[i].buf.map, tx_buf_size,DMA_TO_DEVICE );kfree (tun_qp->tx_ring[i].buf.addr);
@@ -1774,6 +1776,7 @@ static int create_pv_sqp(struct mlx4_ib_demux_pv_ctx *ctx,struct mlx4_ib_qp_tunnel_init_attr qp_init_attr;struct ib_qp_attr attr;int qp_attr_mask_INIT; +const int nmbr_bufs = create_tun ?MLX4_NUM_TUNNEL_BUFS :MLX4_NUM_WIRE_BUFS ;if (qp_type >IB_QPT_GSI )return -EINVAL ;
@@ -1784,8 +1787,8 @@ static int create_pv_sqp(struct mlx4_ib_demux_pv_ctx *ctx, qp_init_attr.init_attr.send_cq = ctx->cq; qp_init_attr.init_attr.recv_cq = ctx->cq; qp_init_attr.init_attr.sq_sig_type =IB_SIGNAL_ALL_WR ; - qp_init_attr.init_attr.cap.max_send_wr =MLX4_NUM_TUNNEL_BUFS ; - qp_init_attr.init_attr.cap.max_recv_wr =MLX4_NUM_TUNNEL_BUFS ; + qp_init_attr.init_attr.cap.max_send_wr = nmbr_bufs; + qp_init_attr.init_attr.cap.max_recv_wr = nmbr_bufs; qp_init_attr.init_attr.cap.max_send_sge =1 ; qp_init_attr.init_attr.cap.max_recv_sge =1 ;if (create_tun) {
@@ -1847,7 +1850,7 @@ static int create_pv_sqp(struct mlx4_ib_demux_pv_ctx *ctx, goto err_qp; } -for (i =0 ; i <MLX4_NUM_TUNNEL_BUFS ; i++) { +for (i =0 ; i < nmbr_bufs; i++) { ret =mlx4_ib_post_pv_qp_buf (ctx, tun_qp, i);if (ret) { pr_err(" mlx4_ib_post_pv_buf error"
@@ -1883,8 +1886,8 @@ static void mlx4_ib_sqp_comp_worker(struct work_struct *work)switch (wc.opcode) {case IB_WC_SEND :kfree (sqp->tx_ring[wc.wr_id & - (MLX4_NUM_TUNNEL_BUFS -1 )].ah); - sqp->tx_ring[wc.wr_id & (MLX4_NUM_TUNNEL_BUFS -1 )].ah + (MLX4_NUM_WIRE_BUFS -1 )].ah); + sqp->tx_ring[wc.wr_id & (MLX4_NUM_WIRE_BUFS -1 )].ah =NULL ;spin_lock (&sqp->tx_lock); sqp->tx_ix_tail++;
@@ -1893,13 +1896,13 @@ static void mlx4_ib_sqp_comp_worker(struct work_struct *work)case IB_WC_RECV : mad = (struct ib_mad *) &(((struct mlx4_mad_rcv_buf *) (sqp->ring[wc.wr_id & - (MLX4_NUM_TUNNEL_BUFS -1 )].addr))->payload); + (MLX4_NUM_WIRE_BUFS -1 )].addr))->payload); grh = &(((struct mlx4_mad_rcv_buf *) (sqp->ring[wc.wr_id & - (MLX4_NUM_TUNNEL_BUFS -1 )].addr))->grh); + (MLX4_NUM_WIRE_BUFS -1 )].addr))->grh);mlx4_ib_demux_mad (ctx->ib_dev, ctx->port, &wc, grh, mad);if (mlx4_ib_post_pv_qp_buf (ctx, sqp, wc.wr_id & - (MLX4_NUM_TUNNEL_BUFS -1 ))) + (MLX4_NUM_WIRE_BUFS -1 )))pr_err ("Failed reposting SQP " "buf:%lld\n" , wc.wr_id);break ;
@@ -1912,8 +1915,8 @@ static void mlx4_ib_sqp_comp_worker(struct work_struct *work) ctx->slave, wc.status, wc.wr_id);if (!MLX4_TUN_IS_RECV (wc.wr_id)) {kfree (sqp->tx_ring[wc.wr_id & - (MLX4_NUM_TUNNEL_BUFS -1 )].ah); - sqp->tx_ring[wc.wr_id & (MLX4_NUM_TUNNEL_BUFS -1 )].ah + (MLX4_NUM_WIRE_BUFS -1 )].ah); + sqp->tx_ring[wc.wr_id & (MLX4_NUM_WIRE_BUFS -1 )].ah =NULL ;spin_lock (&sqp->tx_lock); sqp->tx_ix_tail++;
@@ -1953,6 +1956,7 @@ static int create_pv_resources(struct ib_device *ibdev, int slave, int port, {int ret, cq_size;struct ib_cq_init_attr cq_attr = {}; +const int nmbr_bufs = create_tun ?MLX4_NUM_TUNNEL_BUFS :MLX4_NUM_WIRE_BUFS ;if (ctx->state !=DEMUX_PV_STATE_DOWN )return -EEXIST ;
@@ -1977,7 +1981,7 @@ static int create_pv_resources(struct ib_device *ibdev, int slave, int port, goto err_out_qp0; } - cq_size =2 *MLX4_NUM_TUNNEL_BUFS ; + cq_size =2 * nmbr_bufs;if (ctx->has_smi) cq_size *=2 ;
drivers/infiniband/hw/mlx4/mlx4_ib.h +1 -0
@@ -234,6 +234,7 @@ enum mlx4_ib_mad_ifc_flags {enum {MLX4_NUM_TUNNEL_BUFS =256 , +MLX4_NUM_WIRE_BUFS =2048 , };struct mlx4_ib_tunnel_header {
Signature · 1 line
2.20.1
PATCH 4/5 IB/mlx4: Fix starvation in paravirt mux/demux
In reply to Håkon BuggeThe mlx4 driver will proxy MAD packets through the PF driver. A VM or an instantiated VF will send its MAD packets to the PF driver using loop-back. The PF driver will be informed by an interrupt, but defer the handling and polling of CQEs to a worker thread running on an ordered work-queue.
Consider the following scenario: the VMs will in short proximity in time, for example due to a network event, send many MAD packets to the PF driver. Lets say there are K VMs, each sending N packets.
The interrupt from the first VM will start the worker thread, which will poll N CQEs. A common case here is where the PF driver will multiplex the packets received from the VMs out on the wire QP.
But before the wire QP has returned a send CQE and associated interrupt, the other K - 1 VMs has sent their N packets.
The PF driver will have to multiplex K * N packets out on the wire QP. But the send-queue on the wire QP has a finite capacity.
So, in this scenario, if K * N is larger that the send-queue capacity of the wire QP, we will get MAD packets dropped on the floor with this dynamic debug message:
mlx4_ib_multiplex_mad: failed sending GSI to wire on behalf of slave 2 (-11)
and this despite the fact that the wire send-queue could have capacity, but the PF driver isn't aware, because the wire send CQEs have not yet been polled.
We can also have a similar scenario inbound, with a wire recv-queue larger than the tunnel QP's recv queue. If many remote peers sends MAD packets to the very same VM, the tunnel send-queue destined to the VM could overflow.
This starvation is fixed by introducing separate work queues for the wire QPs vs. the tunnel QPs
With this fix, using a dual ported HCA, 8 VFs instantiated, we could run cmtime on each of the 18 interfaces towards a similar configured peer, each cmtime instance with 800 QPs (all in all 14400 QPs) without a single CM packet getting lost.
--- drivers/infiniband/hw/mlx4/mad.c | 34 +++++++++++++++++++++++++--- drivers/infiniband/hw/mlx4/mlx4_ib.h | 2 ++ 2 files changed, 33 insertions(+), 3 deletions(-)
drivers/infiniband/hw/mlx4/mad.c +31 -3
@@ -1278,6 +1278,18 @@ static void mlx4_ib_tunnel_comp_handler(struct ib_cq *cq, void *arg)spin_unlock_irqrestore (&dev->sriov.going_down_lock, flags); } +static void mlx4_ib_wire_comp_handler (struct ib_cq *cq,void *arg) +{ +unsigned long flags; +struct mlx4_ib_demux_pv_ctx *ctx = cq->cq_context; +struct mlx4_ib_dev *dev =to_mdev (ctx->ib_dev); + +spin_lock_irqsave (&dev->sriov.going_down_lock, flags); +if (!dev->sriov.is_going_down && ctx->state ==DEMUX_PV_STATE_ACTIVE ) +queue_work (ctx->wi_wq, &ctx->work); +spin_unlock_irqrestore (&dev->sriov.going_down_lock, flags); +} +static int mlx4_ib_post_pv_qp_buf (struct mlx4_ib_demux_pv_ctx *ctx,struct mlx4_ib_demux_pv_qp *tun_qp,int index)
@@ -1986,7 +1998,8 @@ static int create_pv_resources(struct ib_device *ibdev, int slave, int port, cq_size *=2 ; cq_attr.cqe = cq_size; - ctx->cq =ib_create_cq (ctx->ib_dev, mlx4_ib_tunnel_comp_handler, + ctx->cq =ib_create_cq (ctx->ib_dev, + create_tun ? mlx4_ib_tunnel_comp_handler : mlx4_ib_wire_comp_handler,NULL , ctx, &cq_attr);if (IS_ERR (ctx->cq)) { ret =PTR_ERR (ctx->cq);
@@ -2023,6 +2036,7 @@ static int create_pv_resources(struct ib_device *ibdev, int slave, int port,INIT_WORK (&ctx->work, mlx4_ib_sqp_comp_worker); ctx->wq =to_mdev (ibdev)->sriov.demux[port -1 ].wq; + ctx->wi_wq =to_mdev (ibdev)->sriov.demux[port -1 ].wi_wq; ret =ib_req_notify_cq (ctx->cq,IB_CQ_NEXT_COMP );if (ret) {
@@ -2166,7 +2180,7 @@ static int mlx4_ib_alloc_demux_ctx(struct mlx4_ib_dev *dev, goto err_mcg; } -snprintf (name,sizeof name,"mlx4_ibt%d" , port); +snprintf (name,sizeof (name),"mlx4_ibt%d" , port); ctx->wq =alloc_ordered_workqueue (name,WQ_MEM_RECLAIM );if (!ctx->wq) {pr_err ("Failed to create tunnelling WQ for port %d\n" , port);
@@ -2174,7 +2188,15 @@ static int mlx4_ib_alloc_demux_ctx(struct mlx4_ib_dev *dev, goto err_wq; } -snprintf (name,sizeof name,"mlx4_ibud%d" , port); +snprintf (name,sizeof (name),"mlx4_ibwi%d" , port); + ctx->wi_wq =alloc_ordered_workqueue (name,WQ_MEM_RECLAIM ); +if (!ctx->wi_wq) { +pr_err ("Failed to create wire WQ for port %d\n" , port); + ret = -ENOMEM ; + goto err_wiwq; + } + +snprintf (name,sizeof (name),"mlx4_ibud%d" , port); ctx->ud_wq =alloc_ordered_workqueue (name,WQ_MEM_RECLAIM );if (!ctx->ud_wq) {pr_err ("Failed to create up/down WQ for port %d\n" , port);
@@ -2185,6 +2207,10 @@ static int mlx4_ib_alloc_demux_ctx(struct mlx4_ib_dev *dev,return 0 ; err_udwq: +destroy_workqueue (ctx->wi_wq); + ctx->wi_wq =NULL ; + +err_wiwq:destroy_workqueue (ctx->wq); ctx->wq =NULL ;
@@ -2232,12 +2258,14 @@ static void mlx4_ib_free_demux_ctx(struct mlx4_ib_demux_ctx *ctx) ctx->tun[i]->state =DEMUX_PV_STATE_DOWNING ; }flush_workqueue (ctx->wq); +flush_workqueue (ctx->wi_wq);for (i =0 ; i < dev->dev->caps.sqp_demux; i++) {destroy_pv_resources (dev, i, ctx->port, ctx->tun[i],0 );free_pv_object (dev, i, ctx->port); }kfree (ctx->tun);destroy_workqueue (ctx->ud_wq); +destroy_workqueue (ctx->wi_wq);destroy_workqueue (ctx->wq); } }
drivers/infiniband/hw/mlx4/mlx4_ib.h +2 -0
@@ -455,6 +455,7 @@ struct mlx4_ib_demux_pv_ctx {struct ib_pd *pd;struct work_struct work;struct workqueue_struct *wq; +struct workqueue_struct *wi_wq;struct mlx4_ib_demux_pv_qp qp[2 ]; };
@@ -462,6 +463,7 @@ struct mlx4_ib_demux_ctx {struct ib_device *ib_dev;int port;struct workqueue_struct *wq; +struct workqueue_struct *wi_wq;struct workqueue_struct *ud_wq;spinlock_t ud_lock;atomic64_t subnet_prefix;
Signature · 1 line
2.20.1
PATCH 5/5 IB/mlx4: Add support for REJ due to timeout
In reply to Håkon BuggeA CM REJ packet with its reason equal to timeout is a special beast in the sense that it doesn't have a Remote Communication ID nor does it have a Remote Port GID.
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.
This proxying doesn't not handle said REJ packets.
If the active side abandons its connection attempt after having sent a REQ, it will send a REJ with the reason being timeout. This example can be provoked by a simple user-verbs program, which ends up doing:
rdma_connect(cm_id, &conn_param);
rdma_destroy_id(cm_id);using the async librdmacm API.
Having dynamic debug prints enabled in the mlx4_ib driver, we will then see:
mlx4_ib_demux_cm_handler: Couldn't find an entry for pv_cm_id 0x0, attr_id 0x12
The solution is to introduce a radix-tree. When a REQ packet is received and handled in mlx4_ib_demux_cm_handler(), we know the connecting peer's para-virtual cm_id and the destination slave. We then insert an entry into the tree with said information. We also schedule work to remove this entry from the tree and free it, in order to avoid memory leak.
When a REJ packet with reason timeout is received, we can look up the slave in the tree, and deliver the packet to the correct slave.
When cleaning up, we simply traverse the tree and modify any delayed work to use a zero delay. A subsequent flush of the system_wq will ensure all entries being wiped out.
--- drivers/infiniband/hw/mlx4/cm.c | 133 ++++++++++++++++++++++++++- drivers/infiniband/hw/mlx4/mlx4_ib.h | 3 + 2 files changed, 135 insertions(+), 1 deletion(-)
drivers/infiniband/hw/mlx4/cm.c +132 -1
@@ -54,11 +54,22 @@ struct id_map_entry {struct delayed_work timeout; }; +struct rej_tmout_entry { +int slave; +u32 rem_pv_cm_id; +struct delayed_work timeout; +struct radix_tree_root *rej_tmout_root; +/* Points to the mutex protecting this radix-tree */ +struct mutex *lock; +}; +struct cm_generic_msg {struct ib_mad_hdr hdr;__be32 local_comm_id;__be32 remote_comm_id; +unsigned char unused[2 ]; +__be16 rej_reason; };struct cm_sidr_generic_msg {
@@ -285,6 +296,7 @@ static void schedule_delayed(struct ib_device *ibdev, struct id_map_entry *id)spin_unlock (&sriov->id_map_lock); } +#define REJ_REASON (m) be16_to_cpu(((struct cm_generic_msg *)(m))->rej_reason)int mlx4_ib_multiplex_cm_handler (struct ib_device *ibdev,int port,int slave_id,struct ib_mad *mad) {
@@ -295,7 +307,8 @@ int mlx4_ib_multiplex_cm_handler(struct ib_device *ibdev, int port, int slave_idif (mad->mad_hdr.attr_id ==CM_REQ_ATTR_ID || mad->mad_hdr.attr_id ==CM_REP_ATTR_ID || mad->mad_hdr.attr_id ==CM_MRA_ATTR_ID || - mad->mad_hdr.attr_id ==CM_SIDR_REQ_ATTR_ID ) { + mad->mad_hdr.attr_id ==CM_SIDR_REQ_ATTR_ID || + (mad->mad_hdr.attr_id ==CM_REJ_ATTR_ID &&REJ_REASON (mad) ==IB_CM_REJ_TIMEOUT )) { sl_cm_id =get_local_comm_id (mad); id =id_map_get (ibdev, &pv_cm_id, slave_id, sl_cm_id);if (id)
@@ -328,11 +341,87 @@ int mlx4_ib_multiplex_cm_handler(struct ib_device *ibdev, int port, int slave_idreturn 0 ; } +static void rej_tmout_timeout (struct work_struct *work) +{ +struct delayed_work *delay =to_delayed_work (work); + struct rej_tmout_entry *item = container_of(delay, struct rej_tmout_entry, timeout); +struct rej_tmout_entry *deleted; + +mutex_lock (item->lock); + deleted =radix_tree_delete_item (item->rej_tmout_root, item->rem_pv_cm_id,NULL ); +mutex_unlock (item->lock); + +if (deleted != item) +pr_debug ("deleted(%p) != item(%p)\n" , deleted, item); + +pr_debug ("rej_tmout entry, rem_pv_cm_id 0x%x, slave %d deleted\n" , + item->rem_pv_cm_id, item->slave); +kfree (item); +} + +static int alloc_rej_tmout (struct mlx4_ib_sriov *sriov,u32 rem_pv_cm_id,int slave) +{ +struct rej_tmout_entry *item; +int sts; + +mutex_lock (&sriov->rej_tmout_lock); + item =radix_tree_lookup (&sriov->rej_tmout_root, (unsigned long )rem_pv_cm_id); +mutex_unlock (&sriov->rej_tmout_lock); +if (item) +return PTR_ERR (item); + + item =kmalloc (sizeof (*item),GFP_KERNEL ); +if (!item) +return -ENOMEM ; + +INIT_DELAYED_WORK (&item->timeout, rej_tmout_timeout); + item->slave = slave; + item->rem_pv_cm_id = rem_pv_cm_id; + item->rej_tmout_root = &sriov->rej_tmout_root; + item->lock = &sriov->rej_tmout_lock; + +mutex_lock (&sriov->rej_tmout_lock); + sts =radix_tree_insert (&sriov->rej_tmout_root, (unsigned long )rem_pv_cm_id, item); +mutex_unlock (&sriov->rej_tmout_lock); +if (sts) + goto err_insert; + +pr_debug ("Inserted rem_pv_cm_id 0x%x slave %d\n" , rem_pv_cm_id, slave); +schedule_delayed_work (&item->timeout,CM_CLEANUP_CACHE_TIMEOUT ); + +return 0 ; + +err_insert: +kfree (item); +return sts; +} + +static int lookup_rej_tmout_slave (struct mlx4_ib_sriov *sriov,u32 rem_pv_cm_id) +{ +struct rej_tmout_entry *item; + +mutex_lock (&sriov->rej_tmout_lock); + item =radix_tree_lookup (&sriov->rej_tmout_root, (unsigned long )rem_pv_cm_id); +mutex_unlock (&sriov->rej_tmout_lock); + +if (!item ||IS_ERR (item)) { +pr_debug ("Could not find rem_pv_cm_id 0x%x error: %d\n" , + rem_pv_cm_id, (int )PTR_ERR (item)); +return !item ? -ENOENT :PTR_ERR (item); + } +pr_debug ("Found rem_pv_cm_id 0x%x slave: %d\n" , rem_pv_cm_id, item->slave); + +return item->slave; +} +int mlx4_ib_demux_cm_handler (struct ib_device *ibdev,int port,int *slave,struct ib_mad *mad) { +struct mlx4_ib_sriov *sriov = &to_mdev (ibdev)->sriov; +u32 rem_pv_cm_id =get_local_comm_id (mad);u32 pv_cm_id;struct id_map_entry *id; +int sts;if (mad->mad_hdr.attr_id ==CM_REQ_ATTR_ID || mad->mad_hdr.attr_id ==CM_SIDR_REQ_ATTR_ID ) {
@@ -348,7 +437,18 @@ int mlx4_ib_demux_cm_handler(struct ib_device *ibdev, int port, int *slave, be64_to_cpu(gid.global.interface_id));return -ENOENT ; } + + sts =alloc_rej_tmout (sriov, rem_pv_cm_id, *slave); +if (sts) +/* Even if this fails, we pass on the REQ to the slave */ +pr_debug ("Could not allocate rej_tmout entry. rem_pv_cm_id 0x%x slave %d status %d\n" , + rem_pv_cm_id, *slave, sts); +return 0 ; + }else if (mad->mad_hdr.attr_id ==CM_REJ_ATTR_ID &&REJ_REASON (mad) ==IB_CM_REJ_TIMEOUT ) { + *slave =lookup_rej_tmout_slave (sriov, rem_pv_cm_id); + +return *slave <0 ? *slave :0 ; } pv_cm_id =get_remote_comm_id (mad);
@@ -377,6 +477,35 @@ void mlx4_ib_cm_paravirt_init(struct mlx4_ib_dev *dev)INIT_LIST_HEAD (&dev->sriov.cm_list); dev->sriov.sl_id_map =RB_ROOT ;xa_init_flags (&dev->sriov.pv_id_table,XA_FLAGS_ALLOC ); +mutex_init (&dev->sriov.rej_tmout_lock); +INIT_RADIX_TREE (&dev->sriov.rej_tmout_root,GFP_KERNEL ); +} + +static void rej_tmout_tree_cleanup (struct mlx4_ib_sriov *sriov,int slave) +{ +struct radix_tree_iter iter; +bool flush_needed = false; +void **slot; +int cnt =0 ; + +mutex_lock (&sriov->rej_tmout_lock); +radix_tree_for_each_slot (slot, &sriov->rej_tmout_root, &iter,0 ) { +struct rej_tmout_entry *item = *slot; + +if (slave <0 || slave == item->slave) { +mod_delayed_work (system_wq, &item->timeout,0 ); + flush_needed = true; + ++cnt; + } + } +mutex_unlock (&sriov->rej_tmout_lock); + +if (flush_needed) { +flush_scheduled_work (); + +pr_debug ("%d entries in radix_tree for slave %d during cleanup\n" , + slave, cnt); + } }/* slave = -1 ==> all slaves */
@@ -446,4 +575,6 @@ void mlx4_ib_cm_paravirt_clean(struct mlx4_ib_dev *dev, int slave)list_del (&map->list);kfree (map); } + +rej_tmout_tree_cleanup (sriov, slave); }
drivers/infiniband/hw/mlx4/mlx4_ib.h +3 -0
@@ -495,6 +495,9 @@ struct mlx4_ib_sriov {spinlock_t id_map_lock;struct rb_root sl_id_map;struct list_head cm_list; +/* Protects the radix-tree */ +struct mutex rej_tmout_lock; +struct radix_tree_root rej_tmout_root; };struct gid_cache_context {
Signature · 1 line
2.20.1
Re: [PATCH for-rc 5/5] IB/mlx4: Add support for REJ due to timeout
Quoted 118 lines · Håkon Bugge
A CM REJ packet with its reason equal to timeout is a special beast in the sense that it doesn't have a Remote Communication ID nor does it have a Remote Port GID.
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.
This proxying doesn't not handle said REJ packets.
If the active side abandons its connection attempt after having sent a REQ, it will send a REJ with the reason being timeout. This example can be provoked by a simple user-verbs program, which ends up doing:
rdma_connect(cm_id, &conn_param); rdma_destroy_id(cm_id);using the async librdmacm API.
Having dynamic debug prints enabled in the
mlx4_ibdriver, we will then see:
mlx4_ib_demux_cm_handler: Couldn't find an entry forpv_cm_id0x0,attr_id0x12The solution is to introduce a radix-tree. When a REQ packet is received and handled in
mlx4_ib_demux_cm_handler(), we know the connecting peer's para-virtualcm_idand the destination slave. We then insert an entry into the tree with said information. We also schedule work to remove this entry from the tree and free it, in order to avoid memory leak.When a REJ packet with reason timeout is received, we can look up the slave in the tree, and deliver the packet to the correct slave.
When cleaning up, we simply traverse the tree and modify any delayed work to use a zero delay. A subsequent flush of the
system_wqwill ensure all entries being wiped out.Signed-off-by Håkon Bugge <[email protected]>---
drivers/infiniband/hw/mlx4/cm.c| 133 ++++++++++++++++++++++++++-drivers/infiniband/hw/mlx4/mlx4_ib.h| 3 +2 files changed, 135 insertions(+), 1 deletion(-)drivers/infiniband/hw/mlx4/cm.c +32
-1@@ -54,11 +54,22 @@ struct id_map_entry {struct delayed_work timeout; }; +struct rej_tmout_entry { +int slave; +u32 rem_pv_cm_id; +struct delayed_work timeout; +struct radix_tree_root *rej_tmout_root; +/* Points to the mutex protecting this radix-tree */ +struct mutex *lock; +}; +struct cm_generic_msg {struct ib_mad_hdr hdr;__be32 local_comm_id;__be32 remote_comm_id; +unsigned char unused[2 ]; +__be16 rej_reason; };struct cm_sidr_generic_msg {@@ -285,6 +296,7 @@ static void schedule_delayed(struct ib_device *ibdev, struct id_map_entry *id)spin_unlock (&sriov->id_map_lock); } +#define REJ_REASON (m) be16_to_cpu(((struct cm_generic_msg *)(m))->rej_reason)int mlx4_ib_multiplex_cm_handler (struct ib_device *ibdev,int port,int slave_id,struct ib_mad *mad) {@@ -295,7 +307,8 @@ int mlx4_ib_multiplex_cm_handler(struct ib_device *ibdev, int port, int slave_idif (mad->mad_hdr.attr_id ==CM_REQ_ATTR_ID || mad->mad_hdr.attr_id ==CM_REP_ATTR_ID || mad->mad_hdr.attr_id ==CM_MRA_ATTR_ID || - mad->mad_hdr.attr_id ==CM_SIDR_REQ_ATTR_ID ) { + mad->mad_hdr.attr_id ==CM_SIDR_REQ_ATTR_ID || + (mad->mad_hdr.attr_id ==CM_REJ_ATTR_ID &&REJ_REASON (mad) ==IB_CM_REJ_TIMEOUT )) { sl_cm_id =get_local_comm_id (mad); id =id_map_get (ibdev, &pv_cm_id, slave_id, sl_cm_id);if (id)@@ -328,11 +341,87 @@ int mlx4_ib_multiplex_cm_handler(struct ib_device *ibdev, int port, int slave_idreturn 0 ; } +static void rej_tmout_timeout (struct work_struct *work) +{ +struct delayed_work *delay =to_delayed_work (work); + struct rej_tmout_entry *item = container_of(delay, struct rej_tmout_entry, timeout); +struct rej_tmout_entry *deleted; + +mutex_lock (item->lock); + deleted =radix_tree_delete_item (item->rej_tmout_root, item->rem_pv_cm_id,NULL ); +mutex_unlock (item->lock); + +if (deleted != item) +pr_debug ("deleted(%p) != item(%p)\n" , deleted, item); + +pr_debug ("rej_tmout entry, rem_pv_cm_id 0x%x, slave %d deleted\n" , + item->rem_pv_cm_id, item->slave); +kfree (item); +} +
drivers/infiniband/hw/mlx4/cm.c int mlx4_ib_multiplex_cm_handler(struct ib_device *ibdev, int port, int slave_id · line 371
Quoted 10 lines
drivers/infiniband/hw/mlx4/cm.c +10
-0+static int alloc_rej_tmout (struct mlx4_ib_sriov *sriov,u32 rem_pv_cm_id,int slave) +{ +struct rej_tmout_entry *item; +int sts; + +mutex_lock (&sriov->rej_tmout_lock); + item =radix_tree_lookup (&sriov->rej_tmout_root, (unsigned long )rem_pv_cm_id); +mutex_unlock (&sriov->rej_tmout_lock); +if (item) +return PTR_ERR (item);
Hmm, this shall read:
return IS_ERR(item) ? PTR_ERR(item) : 0;
I'll also remove the noisy pr_debug()s in this commit.
Will wait before sending out a v2.
Thxs, Håkon
Quoted 135 lines
drivers/infiniband/hw/mlx4/cm.c +90
-0+ + item =kmalloc (sizeof (*item),GFP_KERNEL ); +if (!item) +return -ENOMEM ; + +INIT_DELAYED_WORK (&item->timeout, rej_tmout_timeout); + item->slave = slave; + item->rem_pv_cm_id = rem_pv_cm_id; + item->rej_tmout_root = &sriov->rej_tmout_root; + item->lock = &sriov->rej_tmout_lock; + +mutex_lock (&sriov->rej_tmout_lock); + sts =radix_tree_insert (&sriov->rej_tmout_root, (unsigned long )rem_pv_cm_id, item); +mutex_unlock (&sriov->rej_tmout_lock); +if (sts) + goto err_insert; + +pr_debug ("Inserted rem_pv_cm_id 0x%x slave %d\n" , rem_pv_cm_id, slave); +schedule_delayed_work (&item->timeout,CM_CLEANUP_CACHE_TIMEOUT ); + +return 0 ; + +err_insert: +kfree (item); +return sts; +} + +static int lookup_rej_tmout_slave (struct mlx4_ib_sriov *sriov,u32 rem_pv_cm_id) +{ +struct rej_tmout_entry *item; + +mutex_lock (&sriov->rej_tmout_lock); + item =radix_tree_lookup (&sriov->rej_tmout_root, (unsigned long )rem_pv_cm_id); +mutex_unlock (&sriov->rej_tmout_lock); + +if (!item ||IS_ERR (item)) { +pr_debug ("Could not find rem_pv_cm_id 0x%x error: %d\n" , + rem_pv_cm_id, (int )PTR_ERR (item)); +return !item ? -ENOENT :PTR_ERR (item); + } +pr_debug ("Found rem_pv_cm_id 0x%x slave: %d\n" , rem_pv_cm_id, item->slave); + +return item->slave; +} +int mlx4_ib_demux_cm_handler (struct ib_device *ibdev,int port,int *slave,struct ib_mad *mad) { +struct mlx4_ib_sriov *sriov = &to_mdev (ibdev)->sriov; +u32 rem_pv_cm_id =get_local_comm_id (mad);u32 pv_cm_id;struct id_map_entry *id; +int sts;if (mad->mad_hdr.attr_id ==CM_REQ_ATTR_ID || mad->mad_hdr.attr_id ==CM_SIDR_REQ_ATTR_ID ) {@@ -348,7 +437,18 @@ int mlx4_ib_demux_cm_handler(struct ib_device *ibdev, int port, int *slave, be64_to_cpu(gid.global.interface_id));return -ENOENT ; } + + sts =alloc_rej_tmout (sriov, rem_pv_cm_id, *slave); +if (sts) +/* Even if this fails, we pass on the REQ to the slave */ +pr_debug ("Could not allocate rej_tmout entry. rem_pv_cm_id 0x%x slave %d status %d\n" , + rem_pv_cm_id, *slave, sts); +return 0 ; + }else if (mad->mad_hdr.attr_id ==CM_REJ_ATTR_ID &&REJ_REASON (mad) ==IB_CM_REJ_TIMEOUT ) { + *slave =lookup_rej_tmout_slave (sriov, rem_pv_cm_id); + +return *slave <0 ? *slave :0 ; } pv_cm_id =get_remote_comm_id (mad);@@ -377,6 +477,35 @@ void mlx4_ib_cm_paravirt_init(struct mlx4_ib_dev *dev)INIT_LIST_HEAD (&dev->sriov.cm_list); dev->sriov.sl_id_map =RB_ROOT ;xa_init_flags (&dev->sriov.pv_id_table,XA_FLAGS_ALLOC ); +mutex_init (&dev->sriov.rej_tmout_lock); +INIT_RADIX_TREE (&dev->sriov.rej_tmout_root,GFP_KERNEL ); +} + +static void rej_tmout_tree_cleanup (struct mlx4_ib_sriov *sriov,int slave) +{ +struct radix_tree_iter iter; +bool flush_needed = false; +void **slot; +int cnt =0 ; + +mutex_lock (&sriov->rej_tmout_lock); +radix_tree_for_each_slot (slot, &sriov->rej_tmout_root, &iter,0 ) { +struct rej_tmout_entry *item = *slot; + +if (slave <0 || slave == item->slave) { +mod_delayed_work (system_wq, &item->timeout,0 ); + flush_needed = true; + ++cnt; + } + } +mutex_unlock (&sriov->rej_tmout_lock); + +if (flush_needed) { +flush_scheduled_work (); + +pr_debug ("%d entries in radix_tree for slave %d during cleanup\n" , + slave, cnt); + } }/* slave = -1 ==> all slaves */ @@ -446,4 +575,6 @@ void mlx4_ib_cm_paravirt_clean(struct mlx4_ib_dev *dev, int slave)list_del (&map->list);kfree (map); } + +rej_tmout_tree_cleanup (sriov, slave); }drivers/infiniband/hw/mlx4/mlx4_ib.h +3
-0@@ -495,6 +495,9 @@ struct mlx4_ib_sriov {spinlock_t id_map_lock;struct rb_root sl_id_map;struct list_head cm_list; +/* Protects the radix-tree */ +struct mutex rej_tmout_lock; +struct radix_tree_root rej_tmout_root; };struct gid_cache_context {Signature · 1 line
2.20.1