[Qemu-devel] [PATCH v4 0/9] Misc fixes to pvrdma device

[Qemu-devel] [PATCH v4 0/9] Misc fixes to pvrdma device

[Yuval Shaia]

Hi,
Please review the following patch-set which consist of cosmetics fixes to
device's user interface (traces, error_report and monitor) and some bug
fixes.

Thanks Markus, Eric, Marcel and David for reviewing v0.

David, please see version notes below, since i restructured the HMP part
your second review is needed, thanks in advance.

v0 -> v1:
    * Explain why device attributes are exposed only in HMP interface.
    * Squash the 3 patches related to HMP interface into one.
    * Make monitor dump function simple.
    * Make HMP interface available only if pvrdma is included (detected by
      build robot).
    * Remove patch 03/10 ("Warn when too many consecutive poll CQ triggered
      on an empty CQ) and add the two counters to patch 0/7 (monitor).
    * Add Marcel's R-Bs.
    * Add mutex protection to cqe_ctx list.
    * Add two new patches.

v1 -> v2:
    * Rename locked-lists to protected-lists in patch 2 and patch 6.
    * Add Marcel's R-Bs.

v2 -> v3:
    * Address some 32 bit host compilation issues.

v3 -> v4:
    * Per suggestion from Markus rebase HMP report on
      object_child_foreach_recursive().
    * Strip off David's r-b from HMP patch (#4) because of the above.

Yuval Shaia (9):
  hw/rdma: Switch to generic error reporting way
  hw/rdma: Introduce protected qlist
  hw/rdma: Protect against concurrent execution of poll_cq
  {hmp,hw/pvrdma}: Expose device internals via monitor interface
  hw/rdma: Free all MAD receive buffers when device is closed
  hw/rdma: Free all receive buffers when QP is destroyed
  hw/pvrdma: Delete unneeded function argument
  hw/pvrdma: Delete pvrdma_exit function
  hw/pvrdma: Unregister from shutdown notifier when device goes down

 hmp-commands-info.hx          |  14 +
 hmp.c                         |  27 ++
 hmp.h                         |   1 +
 hw/rdma/Makefile.objs         |   2 +-
 hw/rdma/rdma_backend.c        | 483 +++++++++++++++++-----------------
 hw/rdma/rdma_backend.h        |   3 +-
 hw/rdma/rdma_backend_defs.h   |  10 +-
 hw/rdma/rdma_hmp.c            |  30 +++
 hw/rdma/rdma_rm.c             | 193 ++++++++------
 hw/rdma/rdma_rm.h             |   7 +-
 hw/rdma/rdma_rm_defs.h        |  28 +-
 hw/rdma/rdma_utils.c          |  83 +++++-
 hw/rdma/rdma_utils.h          |  61 ++---
 hw/rdma/trace-events          |  32 ++-
 hw/rdma/vmw/pvrdma.h          |   7 +-
 hw/rdma/vmw/pvrdma_cmd.c      | 113 ++------
 hw/rdma/vmw/pvrdma_dev_ring.c |  26 +-
 hw/rdma/vmw/pvrdma_main.c     | 161 +++++-------
 hw/rdma/vmw/pvrdma_qp_ops.c   |  52 +---
 hw/rdma/vmw/trace-events      |  16 +-
 include/hw/rdma/rdma_hmp.h    |  40 +++
 21 files changed, 779 insertions(+), 610 deletions(-)
 create mode 100644 hw/rdma/rdma_hmp.c
 create mode 100644 include/hw/rdma/rdma_hmp.h

-- 
2.17.2

[Qemu-devel] [PATCH v4 1/9] hw/rdma: Switch to generic error reporting way

[Yuval Shaia]

Utilize error_report for all pr_err calls and some pr_dbg that are
considered as errors.
For the remaining pr_dbg calls, the important ones were replaced by
trace points while other deleted.

Signed-off-by: Yuval Shaia <yuval.shaia@oracle.com>
Reviewed-by: Marcel Apfelbaum <marcel.apfelbaum@gmail.com>
---
 hw/rdma/rdma_backend.c        | 336 ++++++++++++++--------------------
 hw/rdma/rdma_rm.c             | 127 ++++++-------
 hw/rdma/rdma_rm.h             |   6 +-
 hw/rdma/rdma_utils.c          |  15 +-
 hw/rdma/rdma_utils.h          |  45 +----
 hw/rdma/trace-events          |  32 +++-
 hw/rdma/vmw/pvrdma.h          |   2 +-
 hw/rdma/vmw/pvrdma_cmd.c      | 113 +++---------
 hw/rdma/vmw/pvrdma_dev_ring.c |  26 +--
 hw/rdma/vmw/pvrdma_main.c     | 132 +++++--------
 hw/rdma/vmw/pvrdma_qp_ops.c   |  49 ++---
 hw/rdma/vmw/trace-events      |  16 +-
 12 files changed, 343 insertions(+), 556 deletions(-)

diff --git a/hw/rdma/rdma_backend.c b/hw/rdma/rdma_backend.c
index fd571f21e5..24bac00a20 100644
--- a/hw/rdma/rdma_backend.c
+++ b/hw/rdma/rdma_backend.c
@@ -14,7 +14,6 @@
  */
 
 #include "qemu/osdep.h"
-#include "qemu/error-report.h"
 #include "sysemu/sysemu.h"
 #include "qapi/error.h"
 #include "qapi/qmp/qlist.h"
@@ -39,7 +38,6 @@
 
 typedef struct BackendCtx {
     void *up_ctx;
-    bool is_tx_req;
     struct ibv_sge sge; /* Used to save MAD recv buffer */
 } BackendCtx;
 
@@ -52,7 +50,7 @@ static void (*comp_handler)(void *ctx, struct ibv_wc *wc);
 
 static void dummy_comp_handler(void *ctx, struct ibv_wc *wc)
 {
-    pr_err("No completion handler is registered\n");
+    rdma_error_report("No completion handler is registered");
 }
 
 static inline void complete_work(enum ibv_wc_status status, uint32_t vendor_err,
@@ -66,29 +64,24 @@ static inline void complete_work(enum ibv_wc_status status, uint32_t vendor_err,
     comp_handler(ctx, &wc);
 }
 
-static void poll_cq(RdmaDeviceResources *rdma_dev_res, struct ibv_cq *ibcq)
+static void rdma_poll_cq(RdmaDeviceResources *rdma_dev_res, struct ibv_cq *ibcq)
 {
     int i, ne;
     BackendCtx *bctx;
     struct ibv_wc wc[2];
 
-    pr_dbg("Entering poll_cq loop on cq %p\n", ibcq);
     do {
         ne = ibv_poll_cq(ibcq, ARRAY_SIZE(wc), wc);
 
-        pr_dbg("Got %d completion(s) from cq %p\n", ne, ibcq);
+        trace_rdma_poll_cq(ne, ibcq);
 
         for (i = 0; i < ne; i++) {
-            pr_dbg("wr_id=0x%" PRIx64 "\n", wc[i].wr_id);
-            pr_dbg("status=%d\n", wc[i].status);
-
             bctx = rdma_rm_get_cqe_ctx(rdma_dev_res, wc[i].wr_id);
             if (unlikely(!bctx)) {
-                pr_dbg("Error: Failed to find ctx for req %" PRId64 "\n",
-                       wc[i].wr_id);
+                rdma_error_report("No matching ctx for req %"PRId64,
+                                  wc[i].wr_id);
                 continue;
             }
-            pr_dbg("Processing %s CQE\n", bctx->is_tx_req ? "send" : "recv");
 
             comp_handler(bctx->up_ctx, &wc[i]);
 
@@ -98,7 +91,7 @@ static void poll_cq(RdmaDeviceResources *rdma_dev_res, struct ibv_cq *ibcq)
     } while (ne > 0);
 
     if (ne < 0) {
-        pr_dbg("Got error %d from ibv_poll_cq\n", ne);
+        rdma_error_report("ibv_poll_cq fail, rc=%d, errno=%d", ne, errno);
     }
 }
 
@@ -115,12 +108,10 @@ static void *comp_handler_thread(void *arg)
     flags = fcntl(backend_dev->channel->fd, F_GETFL);
     rc = fcntl(backend_dev->channel->fd, F_SETFL, flags | O_NONBLOCK);
     if (rc < 0) {
-        pr_dbg("Fail to change to non-blocking mode\n");
+        rdma_error_report("Failed to change backend channel FD to non-blocking");
         return NULL;
     }
 
-    pr_dbg("Starting\n");
-
     pfds[0].fd = backend_dev->channel->fd;
     pfds[0].events = G_IO_IN | G_IO_HUP | G_IO_ERR;
 
@@ -132,27 +123,25 @@ static void *comp_handler_thread(void *arg)
         } while (!rc && backend_dev->comp_thread.run);
 
         if (backend_dev->comp_thread.run) {
-            pr_dbg("Waiting for completion on channel %p\n", backend_dev->channel);
             rc = ibv_get_cq_event(backend_dev->channel, &ev_cq, &ev_ctx);
-            pr_dbg("ibv_get_cq_event=%d\n", rc);
             if (unlikely(rc)) {
-                pr_dbg("---> ibv_get_cq_event (%d)\n", rc);
+                rdma_error_report("ibv_get_cq_event fail, rc=%d, errno=%d", rc,
+                                  errno);
                 continue;
             }
 
             rc = ibv_req_notify_cq(ev_cq, 0);
             if (unlikely(rc)) {
-                pr_dbg("Error %d from ibv_req_notify_cq\n", rc);
+                rdma_error_report("ibv_req_notify_cq fail, rc=%d, errno=%d", rc,
+                                  errno);
             }
 
-            poll_cq(backend_dev->rdma_dev_res, ev_cq);
+            rdma_poll_cq(backend_dev->rdma_dev_res, ev_cq);
 
             ibv_ack_cq_events(ev_cq, 1);
         }
     }
 
-    pr_dbg("Going down\n");
-
     /* TODO: Post cqe for all remaining buffs that were posted */
 
     backend_dev->comp_thread.is_running = false;
@@ -177,55 +166,54 @@ static inline int rdmacm_mux_can_process_async(RdmaBackendDev *backend_dev)
     return atomic_read(&backend_dev->rdmacm_mux.can_receive);
 }
 
-static int check_mux_op_status(CharBackend *mad_chr_be)
+static int rdmacm_mux_check_op_status(CharBackend *mad_chr_be)
 {
     RdmaCmMuxMsg msg = {};
     int ret;
 
-    pr_dbg("Reading response\n");
     ret = qemu_chr_fe_read_all(mad_chr_be, (uint8_t *)&msg, sizeof(msg));
     if (ret != sizeof(msg)) {
-        pr_dbg("Invalid message size %d, expecting %ld\n", ret, sizeof(msg));
+        rdma_error_report("Got invalid message from mux: size %d, expecting %d",
+                          ret, (int)sizeof(msg));
         return -EIO;
     }
 
-    pr_dbg("msg_type=%d\n", msg.hdr.msg_type);
-    pr_dbg("op_code=%d\n", msg.hdr.op_code);
-    pr_dbg("err_code=%d\n", msg.hdr.err_code);
+    trace_rdmacm_mux_check_op_status(msg.hdr.msg_type, msg.hdr.op_code,
+                                     msg.hdr.err_code);
 
     if (msg.hdr.msg_type != RDMACM_MUX_MSG_TYPE_RESP) {
-        pr_dbg("Invalid message type %d\n", msg.hdr.msg_type);
+        rdma_error_report("Got invalid message type %d", msg.hdr.msg_type);
         return -EIO;
     }
 
     if (msg.hdr.err_code != RDMACM_MUX_ERR_CODE_OK) {
-        pr_dbg("Operation failed in mux, error code %d\n", msg.hdr.err_code);
+        rdma_error_report("Operation failed in mux, error code %d",
+                          msg.hdr.err_code);
         return -EIO;
     }
 
     return 0;
 }
 
-static int exec_rdmacm_mux_req(RdmaBackendDev *backend_dev, RdmaCmMuxMsg *msg)
+static int rdmacm_mux_send(RdmaBackendDev *backend_dev, RdmaCmMuxMsg *msg)
 {
     int rc = 0;
 
-    pr_dbg("Executing request %d\n", msg->hdr.op_code);
-
     msg->hdr.msg_type = RDMACM_MUX_MSG_TYPE_REQ;
+    trace_rdmacm_mux("send", msg->hdr.msg_type, msg->hdr.op_code);
     disable_rdmacm_mux_async(backend_dev);
     rc = qemu_chr_fe_write(backend_dev->rdmacm_mux.chr_be,
                            (const uint8_t *)msg, sizeof(*msg));
     if (rc != sizeof(*msg)) {
         enable_rdmacm_mux_async(backend_dev);
-        pr_dbg("Fail to send request to rdmacm_mux (rc=%d)\n", rc);
+        rdma_error_report("Failed to send request to rdmacm_mux (rc=%d)", rc);
         return -EIO;
     }
 
-    rc = check_mux_op_status(backend_dev->rdmacm_mux.chr_be);
+    rc = rdmacm_mux_check_op_status(backend_dev->rdmacm_mux.chr_be);
     if (rc) {
-        pr_dbg("Fail to execute rdmacm_mux request %d (rc=%d)\n",
-               msg->hdr.op_code, rc);
+        rdma_error_report("Failed to execute rdmacm_mux request %d (rc=%d)",
+                          msg->hdr.op_code, rc);
     }
 
     enable_rdmacm_mux_async(backend_dev);
@@ -237,7 +225,6 @@ static void stop_backend_thread(RdmaBackendThread *thread)
 {
     thread->run = false;
     while (thread->is_running) {
-        pr_dbg("Waiting for thread to complete\n");
         sleep(THR_POLL_TO / SCALE_US / 2);
     }
 }
@@ -273,7 +260,7 @@ int rdma_backend_query_port(RdmaBackendDev *backend_dev,
 
     rc = ibv_query_port(backend_dev->context, backend_dev->port_num, port_attr);
     if (rc) {
-        pr_dbg("Error %d from ibv_query_port\n", rc);
+        rdma_error_report("ibv_query_port fail, rc=%d, errno=%d", rc, errno);
         return -EIO;
     }
 
@@ -282,7 +269,7 @@ int rdma_backend_query_port(RdmaBackendDev *backend_dev,
 
 void rdma_backend_poll_cq(RdmaDeviceResources *rdma_dev_res, RdmaBackendCQ *cq)
 {
-    poll_cq(rdma_dev_res, cq->ibcq);
+    rdma_poll_cq(rdma_dev_res, cq->ibcq);
 }
 
 static GHashTable *ah_hash;
@@ -294,8 +281,8 @@ static struct ibv_ah *create_ah(RdmaBackendDev *backend_dev, struct ibv_pd *pd,
     struct ibv_ah *ah = g_hash_table_lookup(ah_hash, ah_key);
 
     if (ah) {
-        trace_create_ah_cache_hit(be64_to_cpu(dgid->global.subnet_prefix),
-                                  be64_to_cpu(dgid->global.interface_id));
+        trace_rdma_create_ah_cache_hit(be64_to_cpu(dgid->global.subnet_prefix),
+                                       be64_to_cpu(dgid->global.interface_id));
         g_bytes_unref(ah_key);
     } else {
         struct ibv_ah_attr ah_attr = {
@@ -312,13 +299,13 @@ static struct ibv_ah *create_ah(RdmaBackendDev *backend_dev, struct ibv_pd *pd,
             g_hash_table_insert(ah_hash, ah_key, ah);
         } else {
             g_bytes_unref(ah_key);
-            pr_dbg("Fail to create AH for gid <0x%" PRIx64 ", 0x%" PRIx64 ">\n",
-                    be64_to_cpu(dgid->global.subnet_prefix),
-                    be64_to_cpu(dgid->global.interface_id));
+            rdma_error_report("Failed to create AH for gid <0x%" PRIx64", 0x%"PRIx64">",
+                              be64_to_cpu(dgid->global.subnet_prefix),
+                              be64_to_cpu(dgid->global.interface_id));
         }
 
-        trace_create_ah_cache_miss(be64_to_cpu(dgid->global.subnet_prefix),
-                                   be64_to_cpu(dgid->global.interface_id));
+        trace_rdma_create_ah_cache_miss(be64_to_cpu(dgid->global.subnet_prefix),
+                                        be64_to_cpu(dgid->global.interface_id));
     }
 
     return ah;
@@ -349,12 +336,10 @@ static int build_host_sge_array(RdmaDeviceResources *rdma_dev_res,
     RdmaRmMR *mr;
     int ssge_idx;
 
-    pr_dbg("num_sge=%d\n", num_sge);
-
     for (ssge_idx = 0; ssge_idx < num_sge; ssge_idx++) {
         mr = rdma_rm_get_mr(rdma_dev_res, ssge[ssge_idx].lkey);
         if (unlikely(!mr)) {
-            pr_dbg("Invalid lkey 0x%x\n", ssge[ssge_idx].lkey);
+            rdma_error_report("Invalid lkey 0x%x", ssge[ssge_idx].lkey);
             return VENDOR_ERR_INVLKEY | ssge[ssge_idx].lkey;
         }
 
@@ -362,17 +347,28 @@ static int build_host_sge_array(RdmaDeviceResources *rdma_dev_res,
         dsge->length = ssge[ssge_idx].length;
         dsge->lkey = rdma_backend_mr_lkey(&mr->backend_mr);
 
-        pr_dbg("ssge->addr=0x%" PRIx64 "\n", ssge[ssge_idx].addr);
-        pr_dbg("dsge->addr=0x%" PRIx64 "\n", dsge->addr);
-        pr_dbg("dsge->length=%d\n", dsge->length);
-        pr_dbg("dsge->lkey=0x%x\n", dsge->lkey);
-
         dsge++;
     }
 
     return 0;
 }
 
+static void trace_mad_message(const char *title, char *buf, int len)
+{
+    int i;
+    char *b = g_malloc0(len * 3 + 1);
+    char b1[4];
+
+    for (i = 0; i < len; i++) {
+        sprintf(b1, "%.2X ", buf[i] & 0x000000FF);
+        strcat(b, b1);
+    }
+
+    trace_rdma_mad_message(title, len, b);
+
+    g_free(b);
+}
+
 static int mad_send(RdmaBackendDev *backend_dev, uint8_t sgid_idx,
                     union ibv_gid *sgid, struct ibv_sge *sge, uint32_t num_sge)
 {
@@ -380,8 +376,6 @@ static int mad_send(RdmaBackendDev *backend_dev, uint8_t sgid_idx,
     char *hdr, *data;
     int ret;
 
-    pr_dbg("num_sge=%d\n", num_sge);
-
     if (num_sge != 2) {
         return -EINVAL;
     }
@@ -390,7 +384,6 @@ static int mad_send(RdmaBackendDev *backend_dev, uint8_t sgid_idx,
     memcpy(msg.hdr.sgid.raw, sgid->raw, sizeof(msg.hdr.sgid));
 
     msg.umad_len = sge[0].length + sge[1].length;
-    pr_dbg("umad_len=%d\n", msg.umad_len);
 
     if (msg.umad_len > sizeof(msg.umad.mad)) {
         return -ENOMEM;
@@ -398,36 +391,31 @@ static int mad_send(RdmaBackendDev *backend_dev, uint8_t sgid_idx,
 
     msg.umad.hdr.addr.qpn = htobe32(1);
     msg.umad.hdr.addr.grh_present = 1;
-    pr_dbg("sgid_idx=%d\n", sgid_idx);
-    pr_dbg("sgid=0x%llx\n", sgid->global.interface_id);
     msg.umad.hdr.addr.gid_index = sgid_idx;
     memcpy(msg.umad.hdr.addr.gid, sgid->raw, sizeof(msg.umad.hdr.addr.gid));
     msg.umad.hdr.addr.hop_limit = 0xFF;
 
     hdr = rdma_pci_dma_map(backend_dev->dev, sge[0].addr, sge[0].length);
     if (!hdr) {
-        pr_dbg("Fail to map to sge[0]\n");
         return -ENOMEM;
     }
     data = rdma_pci_dma_map(backend_dev->dev, sge[1].addr, sge[1].length);
     if (!data) {
-        pr_dbg("Fail to map to sge[1]\n");
         rdma_pci_dma_unmap(backend_dev->dev, hdr, sge[0].length);
         return -ENOMEM;
     }
 
-    pr_dbg_buf("mad_hdr", hdr, sge[0].length);
-    pr_dbg_buf("mad_data", data, sge[1].length);
-
     memcpy(&msg.umad.mad[0], hdr, sge[0].length);
     memcpy(&msg.umad.mad[sge[0].length], data, sge[1].length);
 
     rdma_pci_dma_unmap(backend_dev->dev, data, sge[1].length);
     rdma_pci_dma_unmap(backend_dev->dev, hdr, sge[0].length);
 
-    ret = exec_rdmacm_mux_req(backend_dev, &msg);
+    trace_mad_message("send", msg.umad.mad, msg.umad_len);
+
+    ret = rdmacm_mux_send(backend_dev, &msg);
     if (ret) {
-        pr_dbg("Fail to send MAD to rdma_umadmux (%d)\n", ret);
+        rdma_error_report("Failed to send MAD to rdma_umadmux (%d)", ret);
         return -EIO;
     }
 
@@ -447,12 +435,11 @@ void rdma_backend_post_send(RdmaBackendDev *backend_dev,
     int rc;
     struct ibv_send_wr wr = {0}, *bad_wr;
 
-    if (!qp->ibqp) { /* This field does not get initialized for QP0 and QP1 */
+    if (!qp->ibqp) { /* This field is not initialized for QP0 and QP1 */
         if (qp_type == IBV_QPT_SMI) {
-            pr_dbg("QP0 unsupported\n");
+            rdma_error_report("Got QP0 request");
             complete_work(IBV_WC_GENERAL_ERR, VENDOR_ERR_QP0, ctx);
         } else if (qp_type == IBV_QPT_GSI) {
-            pr_dbg("QP1\n");
             rc = mad_send(backend_dev, sgid_idx, sgid, sge, num_sge);
             if (rc) {
                 complete_work(IBV_WC_GENERAL_ERR, VENDOR_ERR_MAD_SEND, ctx);
@@ -463,22 +450,17 @@ void rdma_backend_post_send(RdmaBackendDev *backend_dev,
         return;
     }
 
-    pr_dbg("num_sge=%d\n", num_sge);
-
     bctx = g_malloc0(sizeof(*bctx));
     bctx->up_ctx = ctx;
-    bctx->is_tx_req = 1;
 
     rc = rdma_rm_alloc_cqe_ctx(backend_dev->rdma_dev_res, &bctx_id, bctx);
     if (unlikely(rc)) {
-        pr_dbg("Failed to allocate cqe_ctx\n");
         complete_work(IBV_WC_GENERAL_ERR, VENDOR_ERR_NOMEM, ctx);
         goto out_free_bctx;
     }
 
     rc = build_host_sge_array(backend_dev->rdma_dev_res, new_sge, sge, num_sge);
     if (rc) {
-        pr_dbg("Error: Failed to build host SGE array\n");
         complete_work(IBV_WC_GENERAL_ERR, rc, ctx);
         goto out_dealloc_cqe_ctx;
     }
@@ -500,10 +482,9 @@ void rdma_backend_post_send(RdmaBackendDev *backend_dev,
     wr.wr_id = bctx_id;
 
     rc = ibv_post_send(qp->ibqp, &wr, &bad_wr);
-    pr_dbg("ibv_post_send=%d\n", rc);
     if (rc) {
-        pr_dbg("Fail (%d, %d) to post send WQE to qpn %d\n", rc, errno,
-                qp->ibqp->qp_num);
+        rdma_error_report("ibv_post_send fail, qpn=0x%x, rc=%d, errno=%d",
+                          qp->ibqp->qp_num, rc, errno);
         complete_work(IBV_WC_GENERAL_ERR, VENDOR_ERR_FAIL_BACKEND, ctx);
         goto out_dealloc_cqe_ctx;
     }
@@ -526,29 +507,23 @@ static unsigned int save_mad_recv_buffer(RdmaBackendDev *backend_dev,
     uint32_t bctx_id;
 
     if (num_sge != 1) {
-        pr_dbg("Invalid num_sge (%d), expecting 1\n", num_sge);
+        rdma_error_report("Invalid num_sge (%d), expecting 1", num_sge);
         return VENDOR_ERR_INV_NUM_SGE;
     }
 
     if (sge[0].length < RDMA_MAX_PRIVATE_DATA + sizeof(struct ibv_grh)) {
-        pr_dbg("Too small buffer for MAD\n");
+        rdma_error_report("Too small buffer for MAD");
         return VENDOR_ERR_INV_MAD_BUFF;
     }
 
-    pr_dbg("addr=0x%" PRIx64"\n", sge[0].addr);
-    pr_dbg("length=%d\n", sge[0].length);
-    pr_dbg("lkey=%d\n", sge[0].lkey);
-
     bctx = g_malloc0(sizeof(*bctx));
 
     rc = rdma_rm_alloc_cqe_ctx(backend_dev->rdma_dev_res, &bctx_id, bctx);
     if (unlikely(rc)) {
         g_free(bctx);
-        pr_dbg("Fail to allocate cqe_ctx\n");
         return VENDOR_ERR_NOMEM;
     }
 
-    pr_dbg("bctx_id %d, bctx %p, ctx %p\n", bctx_id, bctx, ctx);
     bctx->up_ctx = ctx;
     bctx->sge = *sge;
 
@@ -572,11 +547,10 @@ void rdma_backend_post_recv(RdmaBackendDev *backend_dev,
 
     if (!qp->ibqp) { /* This field does not get initialized for QP0 and QP1 */
         if (qp_type == IBV_QPT_SMI) {
-            pr_dbg("QP0 unsupported\n");
+            rdma_error_report("Got QP0 request");
             complete_work(IBV_WC_GENERAL_ERR, VENDOR_ERR_QP0, ctx);
         }
         if (qp_type == IBV_QPT_GSI) {
-            pr_dbg("QP1\n");
             rc = save_mad_recv_buffer(backend_dev, sge, num_sge, ctx);
             if (rc) {
                 complete_work(IBV_WC_GENERAL_ERR, rc, ctx);
@@ -585,22 +559,17 @@ void rdma_backend_post_recv(RdmaBackendDev *backend_dev,
         return;
     }
 
-    pr_dbg("num_sge=%d\n", num_sge);
-
     bctx = g_malloc0(sizeof(*bctx));
     bctx->up_ctx = ctx;
-    bctx->is_tx_req = 0;
 
     rc = rdma_rm_alloc_cqe_ctx(rdma_dev_res, &bctx_id, bctx);
     if (unlikely(rc)) {
-        pr_dbg("Failed to allocate cqe_ctx\n");
         complete_work(IBV_WC_GENERAL_ERR, VENDOR_ERR_NOMEM, ctx);
         goto out_free_bctx;
     }
 
     rc = build_host_sge_array(rdma_dev_res, new_sge, sge, num_sge);
     if (rc) {
-        pr_dbg("Error: Failed to build host SGE array\n");
         complete_work(IBV_WC_GENERAL_ERR, rc, ctx);
         goto out_dealloc_cqe_ctx;
     }
@@ -609,10 +578,9 @@ void rdma_backend_post_recv(RdmaBackendDev *backend_dev,
     wr.sg_list = new_sge;
     wr.wr_id = bctx_id;
     rc = ibv_post_recv(qp->ibqp, &wr, &bad_wr);
-    pr_dbg("ibv_post_recv=%d\n", rc);
     if (rc) {
-        pr_dbg("Fail (%d, %d) to post recv WQE to qpn %d\n", rc, errno,
-                qp->ibqp->qp_num);
+        rdma_error_report("ibv_post_recv fail, qpn=0x%x, rc=%d, errno=%d",
+                          qp->ibqp->qp_num, rc, errno);
         complete_work(IBV_WC_GENERAL_ERR, VENDOR_ERR_FAIL_BACKEND, ctx);
         goto out_dealloc_cqe_ctx;
     }
@@ -630,7 +598,12 @@ int rdma_backend_create_pd(RdmaBackendDev *backend_dev, RdmaBackendPD *pd)
 {
     pd->ibpd = ibv_alloc_pd(backend_dev->context);
 
-    return pd->ibpd ? 0 : -EIO;
+    if (!pd->ibpd) {
+        rdma_error_report("ibv_alloc_pd fail, errno=%d", errno);
+        return -EIO;
+    }
+
+    return 0;
 }
 
 void rdma_backend_destroy_pd(RdmaBackendPD *pd)
@@ -643,16 +616,15 @@ void rdma_backend_destroy_pd(RdmaBackendPD *pd)
 int rdma_backend_create_mr(RdmaBackendMR *mr, RdmaBackendPD *pd, void *addr,
                            size_t length, int access)
 {
-    pr_dbg("addr=0x%p\n", addr);
-    pr_dbg("len=%zu\n", length);
     mr->ibmr = ibv_reg_mr(pd->ibpd, addr, length, access);
-    if (mr->ibmr) {
-        pr_dbg("lkey=0x%x\n", mr->ibmr->lkey);
-        pr_dbg("rkey=0x%x\n", mr->ibmr->rkey);
-        mr->ibpd = pd->ibpd;
+    if (!mr->ibmr) {
+        rdma_error_report("ibv_reg_mr fail, errno=%d", errno);
+        return -EIO;
     }
 
-    return mr->ibmr ? 0 : -EIO;
+    mr->ibpd = pd->ibpd;
+
+    return 0;
 }
 
 void rdma_backend_destroy_mr(RdmaBackendMR *mr)
@@ -667,21 +639,21 @@ int rdma_backend_create_cq(RdmaBackendDev *backend_dev, RdmaBackendCQ *cq,
 {
     int rc;
 
-    pr_dbg("cqe=%d\n", cqe);
-
-    pr_dbg("dev->channel=%p\n", backend_dev->channel);
     cq->ibcq = ibv_create_cq(backend_dev->context, cqe + 1, NULL,
                              backend_dev->channel, 0);
+    if (!cq->ibcq) {
+        rdma_error_report("ibv_create_cq fail, errno=%d", errno);
+        return -EIO;
+    }
 
-    if (cq->ibcq) {
-        rc = ibv_req_notify_cq(cq->ibcq, 0);
-        if (rc) {
-            pr_dbg("Error %d from ibv_req_notify_cq\n", rc);
-        }
-        cq->backend_dev = backend_dev;
+    rc = ibv_req_notify_cq(cq->ibcq, 0);
+    if (rc) {
+        rdma_warn_report("ibv_req_notify_cq fail, rc=%d, errno=%d", rc, errno);
     }
 
-    return cq->ibcq ? 0 : -EIO;
+    cq->backend_dev = backend_dev;
+
+    return 0;
 }
 
 void rdma_backend_destroy_cq(RdmaBackendCQ *cq)
@@ -700,7 +672,6 @@ int rdma_backend_create_qp(RdmaBackendQP *qp, uint8_t qp_type,
     struct ibv_qp_init_attr attr = {0};
 
     qp->ibqp = 0;
-    pr_dbg("qp_type=%d\n", qp_type);
 
     switch (qp_type) {
     case IBV_QPT_GSI:
@@ -713,7 +684,7 @@ int rdma_backend_create_qp(RdmaBackendQP *qp, uint8_t qp_type,
         break;
 
     default:
-        pr_dbg("Unsupported QP type %d\n", qp_type);
+        rdma_error_report("Unsupported QP type %d", qp_type);
         return -EIO;
     }
 
@@ -725,14 +696,9 @@ int rdma_backend_create_qp(RdmaBackendQP *qp, uint8_t qp_type,
     attr.cap.max_send_sge = max_send_sge;
     attr.cap.max_recv_sge = max_recv_sge;
 
-    pr_dbg("max_send_wr=%d\n", max_send_wr);
-    pr_dbg("max_recv_wr=%d\n", max_recv_wr);
-    pr_dbg("max_send_sge=%d\n", max_send_sge);
-    pr_dbg("max_recv_sge=%d\n", max_recv_sge);
-
     qp->ibqp = ibv_create_qp(pd->ibpd, &attr);
-    if (likely(!qp->ibqp)) {
-        pr_dbg("Error from ibv_create_qp\n");
+    if (!qp->ibqp) {
+        rdma_error_report("ibv_create_qp fail, errno=%d", errno);
         return -EIO;
     }
 
@@ -740,8 +706,6 @@ int rdma_backend_create_qp(RdmaBackendQP *qp, uint8_t qp_type,
 
     /* TODO: Query QP to get max_inline_data and save it to be used in send */
 
-    pr_dbg("qpn=0x%x\n", qp->ibqp->qp_num);
-
     return 0;
 }
 
@@ -751,9 +715,6 @@ int rdma_backend_qp_state_init(RdmaBackendDev *backend_dev, RdmaBackendQP *qp,
     struct ibv_qp_attr attr = {0};
     int rc, attr_mask;
 
-    pr_dbg("qpn=0x%x\n", qp->ibqp->qp_num);
-    pr_dbg("sport_num=%d\n", backend_dev->port_num);
-
     attr_mask = IBV_QP_STATE | IBV_QP_PKEY_INDEX | IBV_QP_PORT;
     attr.qp_state        = IBV_QPS_INIT;
     attr.pkey_index      = 0;
@@ -762,21 +723,23 @@ int rdma_backend_qp_state_init(RdmaBackendDev *backend_dev, RdmaBackendQP *qp,
     switch (qp_type) {
     case IBV_QPT_RC:
         attr_mask |= IBV_QP_ACCESS_FLAGS;
+        trace_rdma_backend_rc_qp_state_init(qp->ibqp->qp_num);
         break;
 
     case IBV_QPT_UD:
         attr.qkey = qkey;
         attr_mask |= IBV_QP_QKEY;
+        trace_rdma_backend_ud_qp_state_init(qp->ibqp->qp_num, qkey);
         break;
 
     default:
-        pr_dbg("Unsupported QP type %d\n", qp_type);
+        rdma_error_report("Unsupported QP type %d", qp_type);
         return -EIO;
     }
 
     rc = ibv_modify_qp(qp->ibqp, &attr, attr_mask);
     if (rc) {
-        pr_dbg("Error %d from ibv_modify_qp\n", rc);
+        rdma_error_report("ibv_modify_qp fail, rc=%d, errno=%d", rc, errno);
         return -EIO;
     }
 
@@ -802,14 +765,6 @@ int rdma_backend_qp_state_rtr(RdmaBackendDev *backend_dev, RdmaBackendQP *qp,
 
     switch (qp_type) {
     case IBV_QPT_RC:
-        pr_dbg("dgid=0x%" PRIx64 ",%" PRIx64 "\n",
-               be64_to_cpu(ibv_gid.global.subnet_prefix),
-               be64_to_cpu(ibv_gid.global.interface_id));
-        pr_dbg("dqpn=0x%x\n", dqpn);
-        pr_dbg("sgid_idx=%d\n", qp->sgid_idx);
-        pr_dbg("sport_num=%d\n", backend_dev->port_num);
-        pr_dbg("rq_psn=0x%x\n", rq_psn);
-
         attr.path_mtu               = IBV_MTU_1024;
         attr.dest_qp_num            = dqpn;
         attr.max_dest_rd_atomic     = 1;
@@ -824,20 +779,28 @@ int rdma_backend_qp_state_rtr(RdmaBackendDev *backend_dev, RdmaBackendQP *qp,
         attr_mask |= IBV_QP_AV | IBV_QP_PATH_MTU | IBV_QP_DEST_QPN |
                      IBV_QP_RQ_PSN | IBV_QP_MAX_DEST_RD_ATOMIC |
                      IBV_QP_MIN_RNR_TIMER;
+
+        trace_rdma_backend_rc_qp_state_rtr(qp->ibqp->qp_num,
+                                           be64_to_cpu(ibv_gid.global.
+                                                       subnet_prefix),
+                                           be64_to_cpu(ibv_gid.global.
+                                                       interface_id),
+                                           qp->sgid_idx, dqpn, rq_psn);
         break;
 
     case IBV_QPT_UD:
-        pr_dbg("qkey=0x%x\n", qkey);
         if (use_qkey) {
             attr.qkey = qkey;
             attr_mask |= IBV_QP_QKEY;
         }
+        trace_rdma_backend_ud_qp_state_rtr(qp->ibqp->qp_num, use_qkey ? qkey :
+                                           0);
         break;
     }
 
     rc = ibv_modify_qp(qp->ibqp, &attr, attr_mask);
     if (rc) {
-        pr_dbg("Error %d from ibv_modify_qp\n", rc);
+        rdma_error_report("ibv_modify_qp fail, rc=%d, errno=%d", rc, errno);
         return -EIO;
     }
 
@@ -850,9 +813,6 @@ int rdma_backend_qp_state_rts(RdmaBackendQP *qp, uint8_t qp_type,
     struct ibv_qp_attr attr = {0};
     int rc, attr_mask;
 
-    pr_dbg("qpn=0x%x\n", qp->ibqp->qp_num);
-    pr_dbg("sq_psn=0x%x\n", sq_psn);
-
     attr.qp_state = IBV_QPS_RTS;
     attr.sq_psn = sq_psn;
     attr_mask = IBV_QP_STATE | IBV_QP_SQ_PSN;
@@ -866,20 +826,22 @@ int rdma_backend_qp_state_rts(RdmaBackendQP *qp, uint8_t qp_type,
 
         attr_mask |= IBV_QP_TIMEOUT | IBV_QP_RETRY_CNT | IBV_QP_RNR_RETRY |
                      IBV_QP_MAX_QP_RD_ATOMIC;
+        trace_rdma_backend_rc_qp_state_rts(qp->ibqp->qp_num, sq_psn);
         break;
 
     case IBV_QPT_UD:
         if (use_qkey) {
-            pr_dbg("qkey=0x%x\n", qkey);
             attr.qkey = qkey;
             attr_mask |= IBV_QP_QKEY;
         }
+        trace_rdma_backend_ud_qp_state_rts(qp->ibqp->qp_num, sq_psn,
+                                           use_qkey ? qkey : 0);
         break;
     }
 
     rc = ibv_modify_qp(qp->ibqp, &attr, attr_mask);
     if (rc) {
-        pr_dbg("Error %d from ibv_modify_qp\n", rc);
+        rdma_error_report("ibv_modify_qp fail, rc=%d, errno=%d", rc, errno);
         return -EIO;
     }
 
@@ -890,7 +852,6 @@ int rdma_backend_query_qp(RdmaBackendQP *qp, struct ibv_qp_attr *attr,
                           int attr_mask, struct ibv_qp_init_attr *init_attr)
 {
     if (!qp->ibqp) {
-        pr_dbg("QP1\n");
         attr->qp_state = IBV_QPS_RTS;
         return 0;
     }
@@ -906,20 +867,23 @@ void rdma_backend_destroy_qp(RdmaBackendQP *qp)
 }
 
 #define CHK_ATTR(req, dev, member, fmt) ({ \
-    pr_dbg("%s="fmt","fmt"\n", #member, dev.member, req->member); \
+    trace_rdma_check_dev_attr(#member, dev.member, req->member); \
     if (req->member > dev.member) { \
-        warn_report("%s = "fmt" is higher than host device capability "fmt, \
-                    #member, req->member, dev.member); \
+        rdma_warn_report("%s = "fmt" is higher than host device capability "fmt, \
+                         #member, req->member, dev.member); \
         req->member = dev.member; \
     } \
-    pr_dbg("%s="fmt"\n", #member, req->member); })
+})
 
 static int init_device_caps(RdmaBackendDev *backend_dev,
                             struct ibv_device_attr *dev_attr)
 {
     struct ibv_device_attr bk_dev_attr;
+    int rc;
 
-    if (ibv_query_device(backend_dev->context, &bk_dev_attr)) {
+    rc = ibv_query_device(backend_dev->context, &bk_dev_attr);
+    if (rc) {
+        rdma_error_report("ibv_query_device fail, rc=%d, errno=%d", rc, errno);
         return -EIO;
     }
 
@@ -928,9 +892,7 @@ static int init_device_caps(RdmaBackendDev *backend_dev,
     CHK_ATTR(dev_attr, bk_dev_attr, max_mr_size, "%" PRId64);
     CHK_ATTR(dev_attr, bk_dev_attr, max_qp, "%d");
     CHK_ATTR(dev_attr, bk_dev_attr, max_sge, "%d");
-    CHK_ATTR(dev_attr, bk_dev_attr, max_qp_wr, "%d");
     CHK_ATTR(dev_attr, bk_dev_attr, max_cq, "%d");
-    CHK_ATTR(dev_attr, bk_dev_attr, max_cqe, "%d");
     CHK_ATTR(dev_attr, bk_dev_attr, max_mr, "%d");
     CHK_ATTR(dev_attr, bk_dev_attr, max_pd, "%d");
     CHK_ATTR(dev_attr, bk_dev_attr, max_qp_rd_atom, "%d");
@@ -946,10 +908,6 @@ static inline void build_mad_hdr(struct ibv_grh *grh, union ibv_gid *sgid,
     grh->paylen = htons(paylen);
     grh->sgid = *sgid;
     grh->dgid = *my_gid;
-
-    pr_dbg("paylen=%d (net=0x%x)\n", paylen, grh->paylen);
-    pr_dbg("dgid=0x%llx\n", my_gid->global.interface_id);
-    pr_dbg("sgid=0x%llx\n", sgid->global.interface_id);
 }
 
 static void process_incoming_mad_req(RdmaBackendDev *backend_dev,
@@ -960,21 +918,13 @@ static void process_incoming_mad_req(RdmaBackendDev *backend_dev,
     BackendCtx *bctx;
     char *mad;
 
-    pr_dbg("umad_len=%d\n", msg->umad_len);
-
-#ifdef PVRDMA_DEBUG
-    struct umad_hdr *hdr = (struct umad_hdr *)&msg->umad.mad;
-    pr_dbg("bv %x cls %x cv %x mtd %x st %d tid %" PRIx64 " at %x atm %x\n",
-           hdr->base_version, hdr->mgmt_class, hdr->class_version,
-           hdr->method, hdr->status, be64toh(hdr->tid),
-           hdr->attr_id, hdr->attr_mod);
-#endif
+    trace_mad_message("recv", msg->umad.mad, msg->umad_len);
 
     qemu_mutex_lock(&backend_dev->recv_mads_list.lock);
     o_ctx_id = qlist_pop(backend_dev->recv_mads_list.list);
     qemu_mutex_unlock(&backend_dev->recv_mads_list.lock);
     if (!o_ctx_id) {
-        pr_dbg("No more free MADs buffers, waiting for a while\n");
+        rdma_warn_report("No more free MADs buffers, waiting for a while");
         sleep(THR_POLL_TO);
         return;
     }
@@ -982,12 +932,10 @@ static void process_incoming_mad_req(RdmaBackendDev *backend_dev,
     cqe_ctx_id = qnum_get_uint(qobject_to(QNum, o_ctx_id));
     bctx = rdma_rm_get_cqe_ctx(backend_dev->rdma_dev_res, cqe_ctx_id);
     if (unlikely(!bctx)) {
-        pr_dbg("Error: Fail to find ctx for %ld\n", cqe_ctx_id);
+        rdma_error_report("No matching ctx for req %ld", cqe_ctx_id);
         return;
     }
 
-    pr_dbg("id %ld, bctx %p, ctx %p\n", cqe_ctx_id, bctx, bctx->up_ctx);
-
     mad = rdma_pci_dma_map(backend_dev->dev, bctx->sge.addr,
                            bctx->sge.length);
     if (!mad || bctx->sge.length < msg->umad_len + MAD_HDR_SIZE) {
@@ -995,7 +943,6 @@ static void process_incoming_mad_req(RdmaBackendDev *backend_dev,
                       bctx->up_ctx);
     } else {
         struct ibv_wc wc = {0};
-        pr_dbg_buf("mad", msg->umad.mad, msg->umad_len);
         memset(mad, 0, bctx->sge.length);
         build_mad_hdr((struct ibv_grh *)mad,
                       (union ibv_gid *)&msg->umad.hdr.addr.gid, &msg->hdr.sgid,
@@ -1025,13 +972,11 @@ static void rdmacm_mux_read(void *opaque, const uint8_t *buf, int size)
     RdmaBackendDev *backend_dev = (RdmaBackendDev *)opaque;
     RdmaCmMuxMsg *msg = (RdmaCmMuxMsg *)buf;
 
-    pr_dbg("Got %d bytes\n", size);
-    pr_dbg("msg_type=%d\n", msg->hdr.msg_type);
-    pr_dbg("op_code=%d\n", msg->hdr.op_code);
+    trace_rdmacm_mux("read", msg->hdr.msg_type, msg->hdr.op_code);
 
     if (msg->hdr.msg_type != RDMACM_MUX_MSG_TYPE_REQ &&
         msg->hdr.op_code != RDMACM_MUX_OP_CODE_MAD) {
-            pr_dbg("Error: Not a MAD request, skipping\n");
+            rdma_error_report("Error: Not a MAD request, skipping");
             return;
     }
     process_incoming_mad_req(backend_dev, msg);
@@ -1045,7 +990,7 @@ static int mad_init(RdmaBackendDev *backend_dev, CharBackend *mad_chr_be)
 
     ret = qemu_chr_fe_backend_connected(backend_dev->rdmacm_mux.chr_be);
     if (!ret) {
-        pr_dbg("Missing chardev for MAD multiplexer\n");
+        rdma_error_report("Missing chardev for MAD multiplexer");
         return -EIO;
     }
 
@@ -1063,7 +1008,6 @@ static int mad_init(RdmaBackendDev *backend_dev, CharBackend *mad_chr_be)
 
 static void mad_fini(RdmaBackendDev *backend_dev)
 {
-    pr_dbg("Stopping MAD\n");
     disable_rdmacm_mux_async(backend_dev);
     qemu_chr_fe_disconnect(backend_dev->rdmacm_mux.chr_be);
     if (backend_dev->recv_mads_list.list) {
@@ -1079,17 +1023,15 @@ int rdma_backend_get_gid_index(RdmaBackendDev *backend_dev,
     int ret;
     int i = 0;
 
-    pr_dbg("0x%llx, 0x%llx\n",
-           (long long unsigned int)be64_to_cpu(gid->global.subnet_prefix),
-           (long long unsigned int)be64_to_cpu(gid->global.interface_id));
-
     do {
         ret = ibv_query_gid(backend_dev->context, backend_dev->port_num, i,
                             &sgid);
         i++;
     } while (!ret && (memcmp(&sgid, gid, sizeof(*gid))));
 
-    pr_dbg("gid_index=%d\n", i - 1);
+    trace_rdma_backend_get_gid_index(be64_to_cpu(gid->global.subnet_prefix),
+                                     be64_to_cpu(gid->global.interface_id),
+                                     i - 1);
 
     return ret ? ret : i - 1;
 }
@@ -1100,16 +1042,15 @@ int rdma_backend_add_gid(RdmaBackendDev *backend_dev, const char *ifname,
     RdmaCmMuxMsg msg = {};
     int ret;
 
-    pr_dbg("0x%llx, 0x%llx\n",
-           (long long unsigned int)be64_to_cpu(gid->global.subnet_prefix),
-           (long long unsigned int)be64_to_cpu(gid->global.interface_id));
+    trace_rdma_backend_gid_change("add", be64_to_cpu(gid->global.subnet_prefix),
+                                  be64_to_cpu(gid->global.interface_id));
 
     msg.hdr.op_code = RDMACM_MUX_OP_CODE_REG;
     memcpy(msg.hdr.sgid.raw, gid->raw, sizeof(msg.hdr.sgid));
 
-    ret = exec_rdmacm_mux_req(backend_dev, &msg);
+    ret = rdmacm_mux_send(backend_dev, &msg);
     if (ret) {
-        pr_dbg("Fail to register GID to rdma_umadmux (%d)\n", ret);
+        rdma_error_report("Failed to register GID to rdma_umadmux (%d)", ret);
         return -EIO;
     }
 
@@ -1126,16 +1067,16 @@ int rdma_backend_del_gid(RdmaBackendDev *backend_dev, const char *ifname,
     RdmaCmMuxMsg msg = {};
     int ret;
 
-    pr_dbg("0x%llx, 0x%llx\n",
-           (long long unsigned int)be64_to_cpu(gid->global.subnet_prefix),
-           (long long unsigned int)be64_to_cpu(gid->global.interface_id));
+    trace_rdma_backend_gid_change("del", be64_to_cpu(gid->global.subnet_prefix),
+                                  be64_to_cpu(gid->global.interface_id));
 
     msg.hdr.op_code = RDMACM_MUX_OP_CODE_UNREG;
     memcpy(msg.hdr.sgid.raw, gid->raw, sizeof(msg.hdr.sgid));
 
-    ret = exec_rdmacm_mux_req(backend_dev, &msg);
+    ret = rdmacm_mux_send(backend_dev, &msg);
     if (ret) {
-        pr_dbg("Fail to unregister GID from rdma_umadmux (%d)\n", ret);
+        rdma_error_report("Failed to unregister GID from rdma_umadmux (%d)",
+                          ret);
         return -EIO;
     }
 
@@ -1196,9 +1137,7 @@ int rdma_backend_init(RdmaBackendDev *backend_dev, PCIDevice *pdev,
         backend_dev->ib_dev = *dev_list;
     }
 
-    pr_dbg("Using backend device %s, port %d\n",
-           ibv_get_device_name(backend_dev->ib_dev), backend_dev->port_num);
-    pr_dbg("uverb device %s\n", backend_dev->ib_dev->dev_name);
+    rdma_info_report("uverb device %s", backend_dev->ib_dev->dev_name);
 
     backend_dev->context = ibv_open_device(backend_dev->ib_dev);
     if (!backend_dev->context) {
@@ -1213,7 +1152,6 @@ int rdma_backend_init(RdmaBackendDev *backend_dev, PCIDevice *pdev,
         ret = -EIO;
         goto out_close_device;
     }
-    pr_dbg("dev->backend_dev.channel=%p\n", backend_dev->channel);
 
     ret = init_device_caps(backend_dev, dev_attr);
     if (ret) {
@@ -1225,7 +1163,7 @@ int rdma_backend_init(RdmaBackendDev *backend_dev, PCIDevice *pdev,
 
     ret = mad_init(backend_dev, mad_chr_be);
     if (ret) {
-        error_setg(errp, "Fail to initialize mad");
+        error_setg(errp, "Failed to initialize mad");
         ret = -EIO;
         goto out_destroy_comm_channel;
     }
@@ -1253,13 +1191,11 @@ out:
 
 void rdma_backend_start(RdmaBackendDev *backend_dev)
 {
-    pr_dbg("Starting rdma_backend\n");
     start_comp_thread(backend_dev);
 }
 
 void rdma_backend_stop(RdmaBackendDev *backend_dev)
 {
-    pr_dbg("Stopping rdma_backend\n");
     stop_backend_thread(&backend_dev->comp_thread);
 }
 
diff --git a/hw/rdma/rdma_rm.c b/hw/rdma/rdma_rm.c
index 268ff633a4..5dab4a2189 100644
--- a/hw/rdma/rdma_rm.c
+++ b/hw/rdma/rdma_rm.c
@@ -17,6 +17,7 @@
 #include "qapi/error.h"
 #include "cpu.h"
 
+#include "trace.h"
 #include "rdma_utils.h"
 #include "rdma_backend.h"
 #include "rdma_rm.h"
@@ -49,25 +50,26 @@ static inline void res_tbl_free(RdmaRmResTbl *tbl)
     g_free(tbl->bitmap);
 }
 
-static inline void *res_tbl_get(RdmaRmResTbl *tbl, uint32_t handle)
+static inline void *rdma_res_tbl_get(RdmaRmResTbl *tbl, uint32_t handle)
 {
-    pr_dbg("%s, handle=%d\n", tbl->name, handle);
+    trace_rdma_res_tbl_get(tbl->name, handle);
 
     if ((handle < tbl->tbl_sz) && (test_bit(handle, tbl->bitmap))) {
         return tbl->tbl + handle * tbl->res_sz;
     } else {
-        pr_dbg("Invalid handle %d\n", handle);
+        rdma_error_report("Table %s, invalid handle %d", tbl->name, handle);
         return NULL;
     }
 }
 
-static inline void *res_tbl_alloc(RdmaRmResTbl *tbl, uint32_t *handle)
+static inline void *rdma_res_tbl_alloc(RdmaRmResTbl *tbl, uint32_t *handle)
 {
     qemu_mutex_lock(&tbl->lock);
 
     *handle = find_first_zero_bit(tbl->bitmap, tbl->tbl_sz);
     if (*handle > tbl->tbl_sz) {
-        pr_dbg("Failed to alloc, bitmap is full\n");
+        rdma_error_report("Table %s, failed to allocate, bitmap is full",
+                          tbl->name);
         qemu_mutex_unlock(&tbl->lock);
         return NULL;
     }
@@ -78,14 +80,14 @@ static inline void *res_tbl_alloc(RdmaRmResTbl *tbl, uint32_t *handle)
 
     memset(tbl->tbl + *handle * tbl->res_sz, 0, tbl->res_sz);
 
-    pr_dbg("%s, handle=%d\n", tbl->name, *handle);
+    trace_rdma_res_tbl_alloc(tbl->name, *handle);
 
     return tbl->tbl + *handle * tbl->res_sz;
 }
 
-static inline void res_tbl_dealloc(RdmaRmResTbl *tbl, uint32_t handle)
+static inline void rdma_res_tbl_dealloc(RdmaRmResTbl *tbl, uint32_t handle)
 {
-    pr_dbg("%s, handle=%d\n", tbl->name, handle);
+    trace_rdma_res_tbl_dealloc(tbl->name, handle);
 
     qemu_mutex_lock(&tbl->lock);
 
@@ -102,7 +104,7 @@ int rdma_rm_alloc_pd(RdmaDeviceResources *dev_res, RdmaBackendDev *backend_dev,
     RdmaRmPD *pd;
     int ret = -ENOMEM;
 
-    pd = res_tbl_alloc(&dev_res->pd_tbl, pd_handle);
+    pd = rdma_res_tbl_alloc(&dev_res->pd_tbl, pd_handle);
     if (!pd) {
         goto out;
     }
@@ -118,7 +120,7 @@ int rdma_rm_alloc_pd(RdmaDeviceResources *dev_res, RdmaBackendDev *backend_dev,
     return 0;
 
 out_tbl_dealloc:
-    res_tbl_dealloc(&dev_res->pd_tbl, *pd_handle);
+    rdma_res_tbl_dealloc(&dev_res->pd_tbl, *pd_handle);
 
 out:
     return ret;
@@ -126,7 +128,7 @@ out:
 
 RdmaRmPD *rdma_rm_get_pd(RdmaDeviceResources *dev_res, uint32_t pd_handle)
 {
-    return res_tbl_get(&dev_res->pd_tbl, pd_handle);
+    return rdma_res_tbl_get(&dev_res->pd_tbl, pd_handle);
 }
 
 void rdma_rm_dealloc_pd(RdmaDeviceResources *dev_res, uint32_t pd_handle)
@@ -135,14 +137,14 @@ void rdma_rm_dealloc_pd(RdmaDeviceResources *dev_res, uint32_t pd_handle)
 
     if (pd) {
         rdma_backend_destroy_pd(&pd->backend_pd);
-        res_tbl_dealloc(&dev_res->pd_tbl, pd_handle);
+        rdma_res_tbl_dealloc(&dev_res->pd_tbl, pd_handle);
     }
 }
 
 int rdma_rm_alloc_mr(RdmaDeviceResources *dev_res, uint32_t pd_handle,
-                     uint64_t guest_start, size_t guest_length, void *host_virt,
-                     int access_flags, uint32_t *mr_handle, uint32_t *lkey,
-                     uint32_t *rkey)
+                     uint64_t guest_start, uint64_t guest_length,
+                     void *host_virt, int access_flags, uint32_t *mr_handle,
+                     uint32_t *lkey, uint32_t *rkey)
 {
     RdmaRmMR *mr;
     int ret = 0;
@@ -150,20 +152,15 @@ int rdma_rm_alloc_mr(RdmaDeviceResources *dev_res, uint32_t pd_handle,
 
     pd = rdma_rm_get_pd(dev_res, pd_handle);
     if (!pd) {
-        pr_dbg("Invalid PD\n");
         return -EINVAL;
     }
 
-    mr = res_tbl_alloc(&dev_res->mr_tbl, mr_handle);
+    mr = rdma_res_tbl_alloc(&dev_res->mr_tbl, mr_handle);
     if (!mr) {
-        pr_dbg("Failed to allocate obj in table\n");
         return -ENOMEM;
     }
-    pr_dbg("mr_handle=%d\n", *mr_handle);
-
-    pr_dbg("host_virt=0x%p\n", host_virt);
-    pr_dbg("guest_start=0x%" PRIx64 "\n", guest_start);
-    pr_dbg("length=%zu\n", guest_length);
+    trace_rdma_rm_alloc_mr(*mr_handle, host_virt, guest_start, guest_length,
+                           access_flags);
 
     if (host_virt) {
         mr->virt = host_virt;
@@ -174,7 +171,6 @@ int rdma_rm_alloc_mr(RdmaDeviceResources *dev_res, uint32_t pd_handle,
         ret = rdma_backend_create_mr(&mr->backend_mr, &pd->backend_pd, mr->virt,
                                      mr->length, access_flags);
         if (ret) {
-            pr_dbg("Fail in rdma_backend_create_mr, err=%d\n", ret);
             ret = -EIO;
             goto out_dealloc_mr;
         }
@@ -189,14 +185,14 @@ int rdma_rm_alloc_mr(RdmaDeviceResources *dev_res, uint32_t pd_handle,
     return 0;
 
 out_dealloc_mr:
-    res_tbl_dealloc(&dev_res->mr_tbl, *mr_handle);
+    rdma_res_tbl_dealloc(&dev_res->mr_tbl, *mr_handle);
 
     return ret;
 }
 
 RdmaRmMR *rdma_rm_get_mr(RdmaDeviceResources *dev_res, uint32_t mr_handle)
 {
-    return res_tbl_get(&dev_res->mr_tbl, mr_handle);
+    return rdma_res_tbl_get(&dev_res->mr_tbl, mr_handle);
 }
 
 void rdma_rm_dealloc_mr(RdmaDeviceResources *dev_res, uint32_t mr_handle)
@@ -205,12 +201,12 @@ void rdma_rm_dealloc_mr(RdmaDeviceResources *dev_res, uint32_t mr_handle)
 
     if (mr) {
         rdma_backend_destroy_mr(&mr->backend_mr);
-        pr_dbg("start=0x%" PRIx64 "\n", mr->start);
+        trace_rdma_rm_dealloc_mr(mr_handle, mr->start);
         if (mr->start) {
             mr->virt -= (mr->start & (TARGET_PAGE_SIZE - 1));
             munmap(mr->virt, mr->length);
         }
-        res_tbl_dealloc(&dev_res->mr_tbl, mr_handle);
+        rdma_res_tbl_dealloc(&dev_res->mr_tbl, mr_handle);
     }
 }
 
@@ -222,12 +218,13 @@ int rdma_rm_alloc_uc(RdmaDeviceResources *dev_res, uint32_t pfn,
     /* TODO: Need to make sure pfn is between bar start address and
      * bsd+RDMA_BAR2_UAR_SIZE
     if (pfn > RDMA_BAR2_UAR_SIZE) {
-        pr_err("pfn out of range (%d > %d)\n", pfn, RDMA_BAR2_UAR_SIZE);
+        rdma_error_report("pfn out of range (%d > %d)", pfn,
+                          RDMA_BAR2_UAR_SIZE);
         return -ENOMEM;
     }
     */
 
-    uc = res_tbl_alloc(&dev_res->uc_tbl, uc_handle);
+    uc = rdma_res_tbl_alloc(&dev_res->uc_tbl, uc_handle);
     if (!uc) {
         return -ENOMEM;
     }
@@ -237,7 +234,7 @@ int rdma_rm_alloc_uc(RdmaDeviceResources *dev_res, uint32_t pfn,
 
 RdmaRmUC *rdma_rm_get_uc(RdmaDeviceResources *dev_res, uint32_t uc_handle)
 {
-    return res_tbl_get(&dev_res->uc_tbl, uc_handle);
+    return rdma_res_tbl_get(&dev_res->uc_tbl, uc_handle);
 }
 
 void rdma_rm_dealloc_uc(RdmaDeviceResources *dev_res, uint32_t uc_handle)
@@ -245,13 +242,13 @@ void rdma_rm_dealloc_uc(RdmaDeviceResources *dev_res, uint32_t uc_handle)
     RdmaRmUC *uc = rdma_rm_get_uc(dev_res, uc_handle);
 
     if (uc) {
-        res_tbl_dealloc(&dev_res->uc_tbl, uc_handle);
+        rdma_res_tbl_dealloc(&dev_res->uc_tbl, uc_handle);
     }
 }
 
 RdmaRmCQ *rdma_rm_get_cq(RdmaDeviceResources *dev_res, uint32_t cq_handle)
 {
-    return res_tbl_get(&dev_res->cq_tbl, cq_handle);
+    return rdma_res_tbl_get(&dev_res->cq_tbl, cq_handle);
 }
 
 int rdma_rm_alloc_cq(RdmaDeviceResources *dev_res, RdmaBackendDev *backend_dev,
@@ -260,7 +257,7 @@ int rdma_rm_alloc_cq(RdmaDeviceResources *dev_res, RdmaBackendDev *backend_dev,
     int rc;
     RdmaRmCQ *cq;
 
-    cq = res_tbl_alloc(&dev_res->cq_tbl, cq_handle);
+    cq = rdma_res_tbl_alloc(&dev_res->cq_tbl, cq_handle);
     if (!cq) {
         return -ENOMEM;
     }
@@ -287,8 +284,6 @@ void rdma_rm_req_notify_cq(RdmaDeviceResources *dev_res, uint32_t cq_handle,
 {
     RdmaRmCQ *cq;
 
-    pr_dbg("cq_handle=%d, notify=0x%x\n", cq_handle, notify);
-
     cq = rdma_rm_get_cq(dev_res, cq_handle);
     if (!cq) {
         return;
@@ -297,8 +292,6 @@ void rdma_rm_req_notify_cq(RdmaDeviceResources *dev_res, uint32_t cq_handle,
     if (cq->notify != CNT_SET) {
         cq->notify = notify ? CNT_ARM : CNT_CLEAR;
     }
-
-    pr_dbg("notify=%d\n", cq->notify);
 }
 
 void rdma_rm_dealloc_cq(RdmaDeviceResources *dev_res, uint32_t cq_handle)
@@ -312,7 +305,7 @@ void rdma_rm_dealloc_cq(RdmaDeviceResources *dev_res, uint32_t cq_handle)
 
     rdma_backend_destroy_cq(&cq->backend_cq);
 
-    res_tbl_dealloc(&dev_res->cq_tbl, cq_handle);
+    rdma_res_tbl_dealloc(&dev_res->cq_tbl, cq_handle);
 }
 
 RdmaRmQP *rdma_rm_get_qp(RdmaDeviceResources *dev_res, uint32_t qpn)
@@ -323,6 +316,10 @@ RdmaRmQP *rdma_rm_get_qp(RdmaDeviceResources *dev_res, uint32_t qpn)
 
     g_bytes_unref(key);
 
+    if (!qp) {
+        rdma_error_report("Invalid QP handle %d", qpn);
+    }
+
     return qp;
 }
 
@@ -338,11 +335,8 @@ int rdma_rm_alloc_qp(RdmaDeviceResources *dev_res, uint32_t pd_handle,
     RdmaRmPD *pd;
     uint32_t rm_qpn;
 
-    pr_dbg("qp_type=%d\n", qp_type);
-
     pd = rdma_rm_get_pd(dev_res, pd_handle);
     if (!pd) {
-        pr_err("Invalid pd handle (%d)\n", pd_handle);
         return -EINVAL;
     }
 
@@ -350,8 +344,8 @@ int rdma_rm_alloc_qp(RdmaDeviceResources *dev_res, uint32_t pd_handle,
     rcq = rdma_rm_get_cq(dev_res, recv_cq_handle);
 
     if (!scq || !rcq) {
-        pr_err("Invalid send_cqn or recv_cqn (%d, %d)\n",
-               send_cq_handle, recv_cq_handle);
+        rdma_error_report("Invalid send_cqn or recv_cqn (%d, %d)",
+                          send_cq_handle, recv_cq_handle);
         return -EINVAL;
     }
 
@@ -360,11 +354,10 @@ int rdma_rm_alloc_qp(RdmaDeviceResources *dev_res, uint32_t pd_handle,
         rcq->notify = CNT_SET;
     }
 
-    qp = res_tbl_alloc(&dev_res->qp_tbl, &rm_qpn);
+    qp = rdma_res_tbl_alloc(&dev_res->qp_tbl, &rm_qpn);
     if (!qp) {
         return -ENOMEM;
     }
-    pr_dbg("rm_qpn=%d\n", rm_qpn);
 
     qp->qpn = rm_qpn;
     qp->qp_state = IBV_QPS_RESET;
@@ -382,13 +375,13 @@ int rdma_rm_alloc_qp(RdmaDeviceResources *dev_res, uint32_t pd_handle,
     }
 
     *qpn = rdma_backend_qpn(&qp->backend_qp);
-    pr_dbg("rm_qpn=%d, backend_qpn=0x%x\n", rm_qpn, *qpn);
+    trace_rdma_rm_alloc_qp(rm_qpn, *qpn, qp_type);
     g_hash_table_insert(dev_res->qp_hash, g_bytes_new(qpn, sizeof(*qpn)), qp);
 
     return 0;
 
 out_dealloc_qp:
-    res_tbl_dealloc(&dev_res->qp_tbl, qp->qpn);
+    rdma_res_tbl_dealloc(&dev_res->qp_tbl, qp->qpn);
 
     return rc;
 }
@@ -402,28 +395,22 @@ int rdma_rm_modify_qp(RdmaDeviceResources *dev_res, RdmaBackendDev *backend_dev,
     RdmaRmQP *qp;
     int ret;
 
-    pr_dbg("qpn=0x%x\n", qp_handle);
-    pr_dbg("qkey=0x%x\n", qkey);
-
     qp = rdma_rm_get_qp(dev_res, qp_handle);
     if (!qp) {
         return -EINVAL;
     }
 
-    pr_dbg("qp_type=%d\n", qp->qp_type);
-    pr_dbg("attr_mask=0x%x\n", attr_mask);
-
     if (qp->qp_type == IBV_QPT_SMI) {
-        pr_dbg("QP0 unsupported\n");
+        rdma_error_report("Got QP0 request");
         return -EPERM;
     } else if (qp->qp_type == IBV_QPT_GSI) {
-        pr_dbg("QP1\n");
         return 0;
     }
 
+    trace_rdma_rm_modify_qp(qp_handle, attr_mask, qp_state, sgid_idx);
+
     if (attr_mask & IBV_QP_STATE) {
         qp->qp_state = qp_state;
-        pr_dbg("qp_state=%d\n", qp->qp_state);
 
         if (qp->qp_state == IBV_QPS_INIT) {
             ret = rdma_backend_qp_state_init(backend_dev, &qp->backend_qp,
@@ -435,11 +422,11 @@ int rdma_rm_modify_qp(RdmaDeviceResources *dev_res, RdmaBackendDev *backend_dev,
 
         if (qp->qp_state == IBV_QPS_RTR) {
             /* Get backend gid index */
-            pr_dbg("Guest sgid_idx=%d\n", sgid_idx);
             sgid_idx = rdma_rm_get_backend_gid_index(dev_res, backend_dev,
                                                      sgid_idx);
             if (sgid_idx <= 0) { /* TODO check also less than bk.max_sgid */
-                pr_dbg("Fail to get bk sgid_idx for sgid_idx %d\n", sgid_idx);
+                rdma_error_report("Failed to get bk sgid_idx for sgid_idx %d",
+                                  sgid_idx);
                 return -EIO;
             }
 
@@ -471,15 +458,11 @@ int rdma_rm_query_qp(RdmaDeviceResources *dev_res, RdmaBackendDev *backend_dev,
 {
     RdmaRmQP *qp;
 
-    pr_dbg("qpn=0x%x\n", qp_handle);
-
     qp = rdma_rm_get_qp(dev_res, qp_handle);
     if (!qp) {
         return -EINVAL;
     }
 
-    pr_dbg("qp_type=%d\n", qp->qp_type);
-
     return rdma_backend_query_qp(&qp->backend_qp, attr, attr_mask, init_attr);
 }
 
@@ -499,20 +482,18 @@ void rdma_rm_dealloc_qp(RdmaDeviceResources *dev_res, uint32_t qp_handle)
 
     rdma_backend_destroy_qp(&qp->backend_qp);
 
-    res_tbl_dealloc(&dev_res->qp_tbl, qp->qpn);
+    rdma_res_tbl_dealloc(&dev_res->qp_tbl, qp->qpn);
 }
 
 void *rdma_rm_get_cqe_ctx(RdmaDeviceResources *dev_res, uint32_t cqe_ctx_id)
 {
     void **cqe_ctx;
 
-    cqe_ctx = res_tbl_get(&dev_res->cqe_ctx_tbl, cqe_ctx_id);
+    cqe_ctx = rdma_res_tbl_get(&dev_res->cqe_ctx_tbl, cqe_ctx_id);
     if (!cqe_ctx) {
         return NULL;
     }
 
-    pr_dbg("ctx=%p\n", *cqe_ctx);
-
     return *cqe_ctx;
 }
 
@@ -521,12 +502,11 @@ int rdma_rm_alloc_cqe_ctx(RdmaDeviceResources *dev_res, uint32_t *cqe_ctx_id,
 {
     void **cqe_ctx;
 
-    cqe_ctx = res_tbl_alloc(&dev_res->cqe_ctx_tbl, cqe_ctx_id);
+    cqe_ctx = rdma_res_tbl_alloc(&dev_res->cqe_ctx_tbl, cqe_ctx_id);
     if (!cqe_ctx) {
         return -ENOMEM;
     }
 
-    pr_dbg("ctx=%p\n", ctx);
     *cqe_ctx = ctx;
 
     return 0;
@@ -534,7 +514,7 @@ int rdma_rm_alloc_cqe_ctx(RdmaDeviceResources *dev_res, uint32_t *cqe_ctx_id,
 
 void rdma_rm_dealloc_cqe_ctx(RdmaDeviceResources *dev_res, uint32_t cqe_ctx_id)
 {
-    res_tbl_dealloc(&dev_res->cqe_ctx_tbl, cqe_ctx_id);
+    rdma_res_tbl_dealloc(&dev_res->cqe_ctx_tbl, cqe_ctx_id);
 }
 
 int rdma_rm_add_gid(RdmaDeviceResources *dev_res, RdmaBackendDev *backend_dev,
@@ -544,7 +524,6 @@ int rdma_rm_add_gid(RdmaDeviceResources *dev_res, RdmaBackendDev *backend_dev,
 
     rc = rdma_backend_add_gid(backend_dev, ifname, gid);
     if (rc) {
-        pr_dbg("Fail to add gid\n");
         return -EINVAL;
     }
 
@@ -565,7 +544,6 @@ int rdma_rm_del_gid(RdmaDeviceResources *dev_res, RdmaBackendDev *backend_dev,
     rc = rdma_backend_del_gid(backend_dev, ifname,
                               &dev_res->port.gid_tbl[gid_idx].gid);
     if (rc) {
-        pr_dbg("Fail to delete gid\n");
         return -EINVAL;
     }
 
@@ -580,7 +558,7 @@ int rdma_rm_get_backend_gid_index(RdmaDeviceResources *dev_res,
                                   RdmaBackendDev *backend_dev, int sgid_idx)
 {
     if (unlikely(sgid_idx < 0 || sgid_idx >= MAX_PORT_GIDS)) {
-        pr_dbg("Got invalid sgid_idx %d\n", sgid_idx);
+        rdma_error_report("Got invalid sgid_idx %d", sgid_idx);
         return -EINVAL;
     }
 
@@ -590,9 +568,6 @@ int rdma_rm_get_backend_gid_index(RdmaDeviceResources *dev_res,
                                    &dev_res->port.gid_tbl[sgid_idx].gid);
     }
 
-    pr_dbg("backend_gid_index=%d\n",
-           dev_res->port.gid_tbl[sgid_idx].backend_gid_index);
-
     return dev_res->port.gid_tbl[sgid_idx].backend_gid_index;
 }
 
diff --git a/hw/rdma/rdma_rm.h b/hw/rdma/rdma_rm.h
index 3c602c04c0..9ec87d2667 100644
--- a/hw/rdma/rdma_rm.h
+++ b/hw/rdma/rdma_rm.h
@@ -31,9 +31,9 @@ RdmaRmPD *rdma_rm_get_pd(RdmaDeviceResources *dev_res, uint32_t pd_handle);
 void rdma_rm_dealloc_pd(RdmaDeviceResources *dev_res, uint32_t pd_handle);
 
 int rdma_rm_alloc_mr(RdmaDeviceResources *dev_res, uint32_t pd_handle,
-                     uint64_t guest_start, size_t guest_length, void *host_virt,
-                     int access_flags, uint32_t *mr_handle, uint32_t *lkey,
-                     uint32_t *rkey);
+                     uint64_t guest_start, uint64_t guest_length,
+                     void *host_virt, int access_flags, uint32_t *mr_handle,
+                     uint32_t *lkey, uint32_t *rkey);
 RdmaRmMR *rdma_rm_get_mr(RdmaDeviceResources *dev_res, uint32_t mr_handle);
 void rdma_rm_dealloc_mr(RdmaDeviceResources *dev_res, uint32_t mr_handle);
 
diff --git a/hw/rdma/rdma_utils.c b/hw/rdma/rdma_utils.c
index 4fbea8cde2..b9f07fcda7 100644
--- a/hw/rdma/rdma_utils.c
+++ b/hw/rdma/rdma_utils.c
@@ -14,26 +14,23 @@
  */
 
 #include "qemu/osdep.h"
+#include "trace.h"
 #include "rdma_utils.h"
 
-#ifdef PVRDMA_DEBUG
-unsigned long pr_dbg_cnt;
-#endif
-
 void *rdma_pci_dma_map(PCIDevice *dev, dma_addr_t addr, dma_addr_t plen)
 {
     void *p;
     hwaddr len = plen;
 
     if (!addr) {
-        pr_dbg("addr is NULL\n");
+        rdma_error_report("addr is NULL");
         return NULL;
     }
 
     p = pci_dma_map(dev, addr, &len, DMA_DIRECTION_TO_DEVICE);
     if (!p) {
-        pr_dbg("Fail in pci_dma_map, addr=0x%" PRIx64 ", len=%" PRId64 "\n",
-               addr, len);
+        rdma_error_report("pci_dma_map fail, addr=0x%"PRIx64", len=%"PRId64,
+                          addr, len);
         return NULL;
     }
 
@@ -42,14 +39,14 @@ void *rdma_pci_dma_map(PCIDevice *dev, dma_addr_t addr, dma_addr_t plen)
         return NULL;
     }
 
-    pr_dbg("0x%" PRIx64 " -> %p (len=% " PRId64 ")\n", addr, p, len);
+    trace_rdma_pci_dma_map(addr, p, len);
 
     return p;
 }
 
 void rdma_pci_dma_unmap(PCIDevice *dev, void *buffer, dma_addr_t len)
 {
-    pr_dbg("%p\n", buffer);
+    trace_rdma_pci_dma_unmap(buffer);
     if (buffer) {
         pci_dma_unmap(dev, buffer, len, DMA_DIRECTION_TO_DEVICE, 0);
     }
diff --git a/hw/rdma/rdma_utils.h b/hw/rdma/rdma_utils.h
index 4490ea0b94..acd148837f 100644
--- a/hw/rdma/rdma_utils.h
+++ b/hw/rdma/rdma_utils.h
@@ -17,48 +17,17 @@
 #ifndef RDMA_UTILS_H
 #define RDMA_UTILS_H
 
+#include "qemu/error-report.h"
 #include "hw/pci/pci.h"
 #include "sysemu/dma.h"
 #include "stdio.h"
 
-#define pr_info(fmt, ...) \
-    fprintf(stdout, "%s: %-20s (%3d): " fmt, "rdma",  __func__, __LINE__,\
-           ## __VA_ARGS__)
-
-#define pr_err(fmt, ...) \
-    fprintf(stderr, "%s: Error at %-20s (%3d): " fmt, "rdma", __func__, \
-        __LINE__, ## __VA_ARGS__)
-
-#ifdef PVRDMA_DEBUG
-extern unsigned long pr_dbg_cnt;
-
-#define init_pr_dbg(void) \
-{ \
-    pr_dbg_cnt = 0; \
-}
-
-#define pr_dbg(fmt, ...) \
-    fprintf(stdout, "%lx %ld: %-20s (%3d): " fmt, pthread_self(), pr_dbg_cnt++, \
-            __func__, __LINE__, ## __VA_ARGS__)
-
-#define pr_dbg_buf(title, buf, len) \
-{ \
-    int i; \
-    char *b = g_malloc0(len * 3 + 1); \
-    char b1[4]; \
-    for (i = 0; i < len; i++) { \
-        sprintf(b1, "%.2X ", buf[i] & 0x000000FF); \
-        strcat(b, b1); \
-    } \
-    pr_dbg("%s (%d): %s\n", title, len, b); \
-    g_free(b); \
-}
-
-#else
-#define init_pr_dbg(void)
-#define pr_dbg(fmt, ...)
-#define pr_dbg_buf(title, buf, len)
-#endif
+#define rdma_error_report(fmt, ...) \
+    error_report("%s: " fmt, "rdma", ## __VA_ARGS__)
+#define rdma_warn_report(fmt, ...) \
+    warn_report("%s: " fmt, "rdma", ## __VA_ARGS__)
+#define rdma_info_report(fmt, ...) \
+    info_report("%s: " fmt, "rdma", ## __VA_ARGS__)
 
 void *rdma_pci_dma_map(PCIDevice *dev, dma_addr_t addr, dma_addr_t plen);
 void rdma_pci_dma_unmap(PCIDevice *dev, void *buffer, dma_addr_t len);
diff --git a/hw/rdma/trace-events b/hw/rdma/trace-events
index c4c202e647..0fad56c882 100644
--- a/hw/rdma/trace-events
+++ b/hw/rdma/trace-events
@@ -1,5 +1,31 @@
 # See docs/tracing.txt for syntax documentation.
 
-#hw/rdma/rdma_backend.c
-create_ah_cache_hit(uint64_t subnet, uint64_t net_id) "subnet = 0x%"PRIx64" net_id = 0x%"PRIx64
-create_ah_cache_miss(uint64_t subnet, uint64_t net_id) "subnet = 0x%"PRIx64" net_id = 0x%"PRIx64
+# hw/rdma/rdma_backend.c
+rdma_check_dev_attr(const char *name, int max_bk, int max_fe) "%s: be=%d, fe=%d"
+rdma_create_ah_cache_hit(uint64_t subnet, uint64_t if_id) "subnet=0x%"PRIx64",if_id=0x%"PRIx64
+rdma_create_ah_cache_miss(uint64_t subnet, uint64_t if_id) "subnet=0x%"PRIx64",if_id=0x%"PRIx64
+rdma_poll_cq(int ne, void *ibcq) "Got %d completion(s) from cq %p"
+rdmacm_mux(const char *title, int msg_type, int op_code) "%s: msg_type=%d, op_code=%d"
+rdmacm_mux_check_op_status(int msg_type, int op_code, int err_code) "resp: msg_type=%d, op_code=%d, err_code=%d"
+rdma_mad_message(const char *title, int len, char *data) "mad %s (%d): %s"
+rdma_backend_rc_qp_state_init(uint32_t qpn) "RC QP 0x%x switch to INIT"
+rdma_backend_ud_qp_state_init(uint32_t qpn, uint32_t qkey) "UD QP 0x%x switch to INIT, qkey=0x%x"
+rdma_backend_rc_qp_state_rtr(uint32_t qpn, uint64_t subnet, uint64_t ifid, uint8_t sgid_idx, uint32_t dqpn, uint32_t rq_psn) "RC QP 0x%x switch to RTR, subnet = 0x%"PRIx64", ifid = 0x%"PRIx64 ", sgid_idx=%d, dqpn=0x%x, rq_psn=0x%x"
+rdma_backend_ud_qp_state_rtr(uint32_t qpn, uint32_t qkey) "UD QP 0x%x switch to RTR, qkey=0x%x"
+rdma_backend_rc_qp_state_rts(uint32_t qpn, uint32_t sq_psn) "RC QP 0x%x switch to RTS, sq_psn=0x%x, "
+rdma_backend_ud_qp_state_rts(uint32_t qpn, uint32_t sq_psn, uint32_t qkey) "UD QP 0x%x switch to RTS, sq_psn=0x%x, qkey=0x%x"
+rdma_backend_get_gid_index(uint64_t subnet, uint64_t ifid, int gid_idx) "subnet=0x%"PRIx64", ifid=0x%"PRIx64 ", gid_idx=%d"
+rdma_backend_gid_change(const char *op, uint64_t subnet, uint64_t ifid) "%s subnet=0x%"PRIx64", ifid=0x%"PRIx64
+
+# hw/rdma/rdma_rm.c
+rdma_res_tbl_get(char *name, uint32_t handle) "tbl %s, handle %d"
+rdma_res_tbl_alloc(char *name, uint32_t handle) "tbl %s, handle %d"
+rdma_res_tbl_dealloc(char *name, uint32_t handle) "tbl %s, handle %d"
+rdma_rm_alloc_mr(uint32_t mr_handle, void *host_virt, uint64_t guest_start, uint64_t guest_length, int access_flags) "mr_handle=%d, host_virt=%p, guest_start=0x%"PRIx64", length=%" PRId64", access_flags=0x%x"
+rdma_rm_dealloc_mr(uint32_t mr_handle, uint64_t guest_start) "mr_handle=%d, guest_start=0x%"PRIx64
+rdma_rm_alloc_qp(uint32_t rm_qpn, uint32_t backend_qpn, uint8_t qp_type) "rm_qpn=%d, backend_qpn=0x%x, qp_type=%d"
+rdma_rm_modify_qp(uint32_t qpn, uint32_t attr_mask, int qp_state, uint8_t sgid_idx) "qpn=0x%x, attr_mask=0x%x, qp_state=%d, sgid_idx=%d"
+
+# hw/rdma/rdma_utils.c
+rdma_pci_dma_map(uint64_t addr, void *vaddr, uint64_t len) "0x%"PRIx64" -> %p (len=%" PRId64")"
+rdma_pci_dma_unmap(void *vaddr) "%p"
diff --git a/hw/rdma/vmw/pvrdma.h b/hw/rdma/vmw/pvrdma.h
index ffae36986e..0879224957 100644
--- a/hw/rdma/vmw/pvrdma.h
+++ b/hw/rdma/vmw/pvrdma.h
@@ -127,6 +127,6 @@ static inline void post_interrupt(PVRDMADev *dev, unsigned vector)
     }
 }
 
-int execute_command(PVRDMADev *dev);
+int pvrdma_exec_cmd(PVRDMADev *dev);
 
 #endif
diff --git a/hw/rdma/vmw/pvrdma_cmd.c b/hw/rdma/vmw/pvrdma_cmd.c
index 89920887bf..21a55e225a 100644
--- a/hw/rdma/vmw/pvrdma_cmd.c
+++ b/hw/rdma/vmw/pvrdma_cmd.c
@@ -14,7 +14,6 @@
  */
 
 #include "qemu/osdep.h"
-#include "qemu/error-report.h"
 #include "cpu.h"
 #include "hw/hw.h"
 #include "hw/pci/pci.h"
@@ -24,6 +23,7 @@
 #include "../rdma_rm.h"
 #include "../rdma_utils.h"
 
+#include "trace.h"
 #include "pvrdma.h"
 #include "standard-headers/rdma/vmw_pvrdma-abi.h"
 
@@ -35,40 +35,38 @@ static void *pvrdma_map_to_pdir(PCIDevice *pdev, uint64_t pdir_dma,
     void *host_virt = NULL, *curr_page;
 
     if (!nchunks) {
-        pr_dbg("nchunks=0\n");
+        rdma_error_report("Got nchunks=0");
         return NULL;
     }
 
     dir = rdma_pci_dma_map(pdev, pdir_dma, TARGET_PAGE_SIZE);
     if (!dir) {
-        error_report("PVRDMA: Failed to map to page directory");
+        rdma_error_report("Failed to map to page directory");
         return NULL;
     }
 
     tbl = rdma_pci_dma_map(pdev, dir[0], TARGET_PAGE_SIZE);
     if (!tbl) {
-        error_report("PVRDMA: Failed to map to page table 0");
+        rdma_error_report("Failed to map to page table 0");
         goto out_unmap_dir;
     }
 
     curr_page = rdma_pci_dma_map(pdev, (dma_addr_t)tbl[0], TARGET_PAGE_SIZE);
     if (!curr_page) {
-        error_report("PVRDMA: Failed to map the first page");
+        rdma_error_report("Failed to map the page 0");
         goto out_unmap_tbl;
     }
 
     host_virt = mremap(curr_page, 0, length, MREMAP_MAYMOVE);
-    pr_dbg("mremap %p -> %p\n", curr_page, host_virt);
     if (host_virt == MAP_FAILED) {
         host_virt = NULL;
-        error_report("PVRDMA: Failed to remap memory for host_virt");
+        rdma_error_report("Failed to remap memory for host_virt");
         goto out_unmap_tbl;
     }
+    trace_pvrdma_map_to_pdir_host_virt(curr_page, host_virt);
 
     rdma_pci_dma_unmap(pdev, curr_page, TARGET_PAGE_SIZE);
 
-    pr_dbg("host_virt=%p\n", host_virt);
-
     dir_idx = 0;
     tbl_idx = 1;
     addr_idx = 1;
@@ -76,28 +74,28 @@ static void *pvrdma_map_to_pdir(PCIDevice *pdev, uint64_t pdir_dma,
         if (tbl_idx == TARGET_PAGE_SIZE / sizeof(uint64_t)) {
             tbl_idx = 0;
             dir_idx++;
-            pr_dbg("Mapping to table %d\n", dir_idx);
             rdma_pci_dma_unmap(pdev, tbl, TARGET_PAGE_SIZE);
             tbl = rdma_pci_dma_map(pdev, dir[dir_idx], TARGET_PAGE_SIZE);
             if (!tbl) {
-                error_report("PVRDMA: Failed to map to page table %d", dir_idx);
+                rdma_error_report("Failed to map to page table %d", dir_idx);
                 goto out_unmap_host_virt;
             }
         }
 
-        pr_dbg("guest_dma[%d]=0x%" PRIx64 "\n", addr_idx, tbl[tbl_idx]);
-
         curr_page = rdma_pci_dma_map(pdev, (dma_addr_t)tbl[tbl_idx],
                                      TARGET_PAGE_SIZE);
         if (!curr_page) {
-            error_report("PVRDMA: Failed to map to page %d, dir %d", tbl_idx,
-                         dir_idx);
+            rdma_error_report("Failed to map to page %d, dir %d", tbl_idx,
+                              dir_idx);
             goto out_unmap_host_virt;
         }
 
         mremap(curr_page, 0, TARGET_PAGE_SIZE, MREMAP_MAYMOVE | MREMAP_FIXED,
                host_virt + TARGET_PAGE_SIZE * addr_idx);
 
+        trace_pvrdma_map_to_pdir_next_page(addr_idx, curr_page, host_virt +
+                                           TARGET_PAGE_SIZE * addr_idx);
+
         rdma_pci_dma_unmap(pdev, curr_page, TARGET_PAGE_SIZE);
 
         addr_idx++;
@@ -127,7 +125,6 @@ static int query_port(PVRDMADev *dev, union pvrdma_cmd_req *req,
     struct pvrdma_cmd_query_port_resp *resp = &rsp->query_port_resp;
     struct pvrdma_port_attr attrs = {0};
 
-    pr_dbg("port=%d\n", cmd->port_num);
     if (cmd->port_num > MAX_PORTS) {
         return -EINVAL;
     }
@@ -159,12 +156,10 @@ static int query_pkey(PVRDMADev *dev, union pvrdma_cmd_req *req,
     struct pvrdma_cmd_query_pkey *cmd = &req->query_pkey;
     struct pvrdma_cmd_query_pkey_resp *resp = &rsp->query_pkey_resp;
 
-    pr_dbg("port=%d\n", cmd->port_num);
     if (cmd->port_num > MAX_PORTS) {
         return -EINVAL;
     }
 
-    pr_dbg("index=%d\n", cmd->index);
     if (cmd->index > MAX_PKEYS) {
         return -EINVAL;
     }
@@ -172,7 +167,6 @@ static int query_pkey(PVRDMADev *dev, union pvrdma_cmd_req *req,
     memset(resp, 0, sizeof(*resp));
 
     resp->pkey = PVRDMA_PKEY;
-    pr_dbg("pkey=0x%x\n", resp->pkey);
 
     return 0;
 }
@@ -184,8 +178,6 @@ static int create_pd(PVRDMADev *dev, union pvrdma_cmd_req *req,
     struct pvrdma_cmd_create_pd_resp *resp = &rsp->create_pd_resp;
     int rc;
 
-    pr_dbg("context=0x%x\n", cmd->ctx_handle ? cmd->ctx_handle : 0);
-
     memset(resp, 0, sizeof(*resp));
     rc = rdma_rm_alloc_pd(&dev->rdma_dev_res, &dev->backend_dev,
                           &resp->pd_handle, cmd->ctx_handle);
@@ -198,8 +190,6 @@ static int destroy_pd(PVRDMADev *dev, union pvrdma_cmd_req *req,
 {
     struct pvrdma_cmd_destroy_pd *cmd = &req->destroy_pd;
 
-    pr_dbg("pd_handle=%d\n", cmd->pd_handle);
-
     rdma_rm_dealloc_pd(&dev->rdma_dev_res, cmd->pd_handle);
 
     return 0;
@@ -216,15 +206,11 @@ static int create_mr(PVRDMADev *dev, union pvrdma_cmd_req *req,
 
     memset(resp, 0, sizeof(*resp));
 
-    pr_dbg("pd_handle=%d\n", cmd->pd_handle);
-    pr_dbg("access_flags=0x%x\n", cmd->access_flags);
-    pr_dbg("flags=0x%x\n", cmd->flags);
-
     if (!(cmd->flags & PVRDMA_MR_FLAG_DMA)) {
         host_virt = pvrdma_map_to_pdir(pci_dev, cmd->pdir_dma, cmd->nchunks,
                                        cmd->length);
         if (!host_virt) {
-            pr_dbg("Failed to map to pdir\n");
+            rdma_error_report("Failed to map to pdir");
             return -EINVAL;
         }
     }
@@ -244,8 +230,6 @@ static int destroy_mr(PVRDMADev *dev, union pvrdma_cmd_req *req,
 {
     struct pvrdma_cmd_destroy_mr *cmd = &req->destroy_mr;
 
-    pr_dbg("mr_handle=%d\n", cmd->mr_handle);
-
     rdma_rm_dealloc_mr(&dev->rdma_dev_res, cmd->mr_handle);
 
     return 0;
@@ -260,20 +244,19 @@ static int create_cq_ring(PCIDevice *pci_dev , PvrdmaRing **ring,
     char ring_name[MAX_RING_NAME_SZ];
 
     if (!nchunks || nchunks > PVRDMA_MAX_FAST_REG_PAGES) {
-        pr_dbg("invalid nchunks: %d\n", nchunks);
+        rdma_error_report("Got invalid nchunks: %d", nchunks);
         return rc;
     }
 
-    pr_dbg("pdir_dma=0x%llx\n", (long long unsigned int)pdir_dma);
     dir = rdma_pci_dma_map(pci_dev, pdir_dma, TARGET_PAGE_SIZE);
     if (!dir) {
-        pr_dbg("Failed to map to CQ page directory\n");
+        rdma_error_report("Failed to map to CQ page directory");
         goto out;
     }
 
     tbl = rdma_pci_dma_map(pci_dev, dir[0], TARGET_PAGE_SIZE);
     if (!tbl) {
-        pr_dbg("Failed to map to CQ page table\n");
+        rdma_error_report("Failed to map to CQ page table");
         goto out;
     }
 
@@ -284,7 +267,7 @@ static int create_cq_ring(PCIDevice *pci_dev , PvrdmaRing **ring,
         rdma_pci_dma_map(pci_dev, tbl[0], TARGET_PAGE_SIZE);
 
     if (!r->ring_state) {
-        pr_dbg("Failed to map to CQ ring state\n");
+        rdma_error_report("Failed to map to CQ ring state");
         goto out_free_ring;
     }
 
@@ -339,8 +322,6 @@ static int create_cq(PVRDMADev *dev, union pvrdma_cmd_req *req,
         return rc;
     }
 
-    pr_dbg("ring=%p\n", ring);
-
     rc = rdma_rm_alloc_cq(&dev->rdma_dev_res, &dev->backend_dev, cmd->cqe,
                           &resp->cq_handle, ring);
     if (rc) {
@@ -359,11 +340,9 @@ static int destroy_cq(PVRDMADev *dev, union pvrdma_cmd_req *req,
     RdmaRmCQ *cq;
     PvrdmaRing *ring;
 
-    pr_dbg("cq_handle=%d\n", cmd->cq_handle);
-
     cq = rdma_rm_get_cq(&dev->rdma_dev_res, cmd->cq_handle);
     if (!cq) {
-        pr_dbg("Invalid CQ handle\n");
+        rdma_error_report("Got invalid CQ handle");
         return -EINVAL;
     }
 
@@ -388,42 +367,33 @@ static int create_qp_rings(PCIDevice *pci_dev, uint64_t pdir_dma,
 
     if (!spages || spages > PVRDMA_MAX_FAST_REG_PAGES
         || !rpages || rpages > PVRDMA_MAX_FAST_REG_PAGES) {
-        pr_dbg("invalid pages: %d, %d\n", spages, rpages);
+        rdma_error_report("Got invalid page count for QP ring: %d, %d", spages,
+                          rpages);
         return rc;
     }
 
-    pr_dbg("pdir_dma=0x%llx\n", (long long unsigned int)pdir_dma);
     dir = rdma_pci_dma_map(pci_dev, pdir_dma, TARGET_PAGE_SIZE);
     if (!dir) {
-        pr_dbg("Failed to map to CQ page directory\n");
+        rdma_error_report("Failed to map to CQ page directory");
         goto out;
     }
 
     tbl = rdma_pci_dma_map(pci_dev, dir[0], TARGET_PAGE_SIZE);
     if (!tbl) {
-        pr_dbg("Failed to map to CQ page table\n");
+        rdma_error_report("Failed to map to CQ page table");
         goto out;
     }
 
     sr = g_malloc(2 * sizeof(*rr));
     rr = &sr[1];
-    pr_dbg("sring=%p\n", sr);
-    pr_dbg("rring=%p\n", rr);
 
     *rings = sr;
 
-    pr_dbg("scqe=%d\n", scqe);
-    pr_dbg("smax_sge=%d\n", smax_sge);
-    pr_dbg("spages=%d\n", spages);
-    pr_dbg("rcqe=%d\n", rcqe);
-    pr_dbg("rmax_sge=%d\n", rmax_sge);
-    pr_dbg("rpages=%d\n", rpages);
-
     /* Create send ring */
     sr->ring_state = (struct pvrdma_ring *)
         rdma_pci_dma_map(pci_dev, tbl[0], TARGET_PAGE_SIZE);
     if (!sr->ring_state) {
-        pr_dbg("Failed to map to CQ ring state\n");
+        rdma_error_report("Failed to map to CQ ring state");
         goto out_free_sr_mem;
     }
 
@@ -468,9 +438,7 @@ out:
 
 static void destroy_qp_rings(PvrdmaRing *ring)
 {
-    pr_dbg("sring=%p\n", &ring[0]);
     pvrdma_ring_free(&ring[0]);
-    pr_dbg("rring=%p\n", &ring[1]);
     pvrdma_ring_free(&ring[1]);
 
     rdma_pci_dma_unmap(ring->dev, ring->ring_state, TARGET_PAGE_SIZE);
@@ -487,9 +455,6 @@ static int create_qp(PVRDMADev *dev, union pvrdma_cmd_req *req,
 
     memset(resp, 0, sizeof(*resp));
 
-    pr_dbg("total_chunks=%d\n", cmd->total_chunks);
-    pr_dbg("send_chunks=%d\n", cmd->send_chunks);
-
     rc = create_qp_rings(PCI_DEVICE(dev), cmd->pdir_dma, &rings,
                          cmd->max_send_wr, cmd->max_send_sge, cmd->send_chunks,
                          cmd->max_recv_wr, cmd->max_recv_sge,
@@ -498,8 +463,6 @@ static int create_qp(PVRDMADev *dev, union pvrdma_cmd_req *req,
         return rc;
     }
 
-    pr_dbg("rings=%p\n", rings);
-
     rc = rdma_rm_alloc_qp(&dev->rdma_dev_res, cmd->pd_handle, cmd->qp_type,
                           cmd->max_send_wr, cmd->max_send_sge,
                           cmd->send_cq_handle, cmd->max_recv_wr,
@@ -525,8 +488,6 @@ static int modify_qp(PVRDMADev *dev, union pvrdma_cmd_req *req,
     struct pvrdma_cmd_modify_qp *cmd = &req->modify_qp;
     int rc;
 
-    pr_dbg("qp_handle=%d\n", cmd->qp_handle);
-
     memset(rsp, 0, sizeof(*rsp));
 
     /* No need to verify sgid_index since it is u8 */
@@ -551,9 +512,6 @@ static int query_qp(PVRDMADev *dev, union pvrdma_cmd_req *req,
     struct ibv_qp_init_attr init_attr;
     int rc;
 
-    pr_dbg("qp_handle=%d\n", cmd->qp_handle);
-    pr_dbg("attr_mask=0x%x\n", cmd->attr_mask);
-
     memset(rsp, 0, sizeof(*rsp));
 
     rc = rdma_rm_query_qp(&dev->rdma_dev_res, &dev->backend_dev, cmd->qp_handle,
@@ -572,7 +530,6 @@ static int destroy_qp(PVRDMADev *dev, union pvrdma_cmd_req *req,
 
     qp = rdma_rm_get_qp(&dev->rdma_dev_res, cmd->qp_handle);
     if (!qp) {
-        pr_dbg("Invalid QP handle\n");
         return -EINVAL;
     }
 
@@ -591,16 +548,10 @@ static int create_bind(PVRDMADev *dev, union pvrdma_cmd_req *req,
     int rc;
     union ibv_gid *gid = (union ibv_gid *)&cmd->new_gid;
 
-    pr_dbg("index=%d\n", cmd->index);
-
     if (cmd->index >= MAX_PORT_GIDS) {
         return -EINVAL;
     }
 
-    pr_dbg("gid[%d]=0x%llx,0x%llx\n", cmd->index,
-           (long long unsigned int)be64_to_cpu(gid->global.subnet_prefix),
-           (long long unsigned int)be64_to_cpu(gid->global.interface_id));
-
     rc = rdma_rm_add_gid(&dev->rdma_dev_res, &dev->backend_dev,
                          dev->backend_eth_device_name, gid, cmd->index);
 
@@ -614,8 +565,6 @@ static int destroy_bind(PVRDMADev *dev, union pvrdma_cmd_req *req,
 
     struct pvrdma_cmd_destroy_bind *cmd = &req->destroy_bind;
 
-    pr_dbg("index=%d\n", cmd->index);
-
     if (cmd->index >= MAX_PORT_GIDS) {
         return -EINVAL;
     }
@@ -633,8 +582,6 @@ static int create_uc(PVRDMADev *dev, union pvrdma_cmd_req *req,
     struct pvrdma_cmd_create_uc_resp *resp = &rsp->create_uc_resp;
     int rc;
 
-    pr_dbg("pfn=%d\n", cmd->pfn);
-
     memset(resp, 0, sizeof(*resp));
     rc = rdma_rm_alloc_uc(&dev->rdma_dev_res, cmd->pfn, &resp->ctx_handle);
 
@@ -646,8 +593,6 @@ static int destroy_uc(PVRDMADev *dev, union pvrdma_cmd_req *req,
 {
     struct pvrdma_cmd_destroy_uc *cmd = &req->destroy_uc;
 
-    pr_dbg("ctx_handle=%d\n", cmd->ctx_handle);
-
     rdma_rm_dealloc_uc(&dev->rdma_dev_res, cmd->ctx_handle);
 
     return 0;
@@ -680,22 +625,21 @@ static struct cmd_handler cmd_handlers[] = {
     {PVRDMA_CMD_DESTROY_BIND, PVRDMA_CMD_DESTROY_BIND_RESP_NOOP, destroy_bind},
 };
 
-int execute_command(PVRDMADev *dev)
+int pvrdma_exec_cmd(PVRDMADev *dev)
 {
     int err = 0xFFFF;
     DSRInfo *dsr_info;
 
     dsr_info = &dev->dsr_info;
 
-    pr_dbg("cmd=%d\n", dsr_info->req->hdr.cmd);
     if (dsr_info->req->hdr.cmd >= sizeof(cmd_handlers) /
                       sizeof(struct cmd_handler)) {
-        pr_dbg("Unsupported command\n");
+        rdma_error_report("Unsupported command");
         goto out;
     }
 
     if (!cmd_handlers[dsr_info->req->hdr.cmd].exec) {
-        pr_dbg("Unsupported command (not implemented yet)\n");
+        rdma_error_report("Unsupported command (not implemented yet)");
         goto out;
     }
 
@@ -704,7 +648,8 @@ int execute_command(PVRDMADev *dev)
     dsr_info->rsp->hdr.response = dsr_info->req->hdr.response;
     dsr_info->rsp->hdr.ack = cmd_handlers[dsr_info->req->hdr.cmd].ack;
     dsr_info->rsp->hdr.err = err < 0 ? -err : 0;
-    pr_dbg("rsp->hdr.err=%d\n", dsr_info->rsp->hdr.err);
+
+    trace_pvrdma_exec_cmd(dsr_info->req->hdr.cmd, dsr_info->rsp->hdr.err);
 
 out:
     set_reg_val(dev, PVRDMA_REG_ERR, err);
diff --git a/hw/rdma/vmw/pvrdma_dev_ring.c b/hw/rdma/vmw/pvrdma_dev_ring.c
index e8e5b502f6..d7bc7f5ccc 100644
--- a/hw/rdma/vmw/pvrdma_dev_ring.c
+++ b/hw/rdma/vmw/pvrdma_dev_ring.c
@@ -17,6 +17,8 @@
 #include "hw/pci/pci.h"
 #include "cpu.h"
 
+#include "trace.h"
+
 #include "../rdma_utils.h"
 #include "standard-headers/drivers/infiniband/hw/vmw_pvrdma/pvrdma_ring.h"
 #include "pvrdma_dev_ring.h"
@@ -30,13 +32,10 @@ int pvrdma_ring_init(PvrdmaRing *ring, const char *name, PCIDevice *dev,
 
     strncpy(ring->name, name, MAX_RING_NAME_SZ);
     ring->name[MAX_RING_NAME_SZ - 1] = 0;
-    pr_dbg("Initializing %s ring\n", ring->name);
     ring->dev = dev;
     ring->ring_state = ring_state;
     ring->max_elems = max_elems;
     ring->elem_sz = elem_sz;
-    pr_dbg("ring->elem_sz=%zu\n", ring->elem_sz);
-    pr_dbg("npages=%d\n", npages);
     /* TODO: Give a moment to think if we want to redo driver settings
     atomic_set(&ring->ring_state->prod_tail, 0);
     atomic_set(&ring->ring_state->cons_head, 0);
@@ -46,14 +45,14 @@ int pvrdma_ring_init(PvrdmaRing *ring, const char *name, PCIDevice *dev,
 
     for (i = 0; i < npages; i++) {
         if (!tbl[i]) {
-            pr_err("npages=%ld but tbl[%d] is NULL\n", (long)npages, i);
+            rdma_error_report("npages=%d but tbl[%d] is NULL", npages, i);
             continue;
         }
 
         ring->pages[i] = rdma_pci_dma_map(dev, tbl[i], TARGET_PAGE_SIZE);
         if (!ring->pages[i]) {
             rc = -ENOMEM;
-            pr_dbg("Failed to map to page %d\n", i);
+            rdma_error_report("Failed to map to page %d in ring %s", i, name);
             goto out_free;
         }
         memset(ring->pages[i], 0, TARGET_PAGE_SIZE);
@@ -78,7 +77,7 @@ void *pvrdma_ring_next_elem_read(PvrdmaRing *ring)
 
     e = pvrdma_idx_ring_has_data(ring->ring_state, ring->max_elems, &idx);
     if (e <= 0) {
-        pr_dbg("No more data in ring\n");
+        trace_pvrdma_ring_next_elem_read_no_data(ring->name);
         return NULL;
     }
 
@@ -89,11 +88,6 @@ void *pvrdma_ring_next_elem_read(PvrdmaRing *ring)
 void pvrdma_ring_read_inc(PvrdmaRing *ring)
 {
     pvrdma_idx_ring_inc(&ring->ring_state->cons_head, ring->max_elems);
-    /*
-    pr_dbg("%s: t=%d, h=%d, m=%ld\n", ring->name,
-           ring->ring_state->prod_tail, ring->ring_state->cons_head,
-           ring->max_elems);
-    */
 }
 
 void *pvrdma_ring_next_elem_write(PvrdmaRing *ring)
@@ -103,13 +97,13 @@ void *pvrdma_ring_next_elem_write(PvrdmaRing *ring)
 
     idx = pvrdma_idx_ring_has_space(ring->ring_state, ring->max_elems, &tail);
     if (idx <= 0) {
-        pr_dbg("CQ is full\n");
+        rdma_error_report("CQ is full");
         return NULL;
     }
 
     idx = pvrdma_idx(&ring->ring_state->prod_tail, ring->max_elems);
     if (idx < 0 || tail != idx) {
-        pr_dbg("invalid idx\n");
+        rdma_error_report("Invalid idx %d", idx);
         return NULL;
     }
 
@@ -120,11 +114,6 @@ void *pvrdma_ring_next_elem_write(PvrdmaRing *ring)
 void pvrdma_ring_write_inc(PvrdmaRing *ring)
 {
     pvrdma_idx_ring_inc(&ring->ring_state->prod_tail, ring->max_elems);
-    /*
-    pr_dbg("%s: t=%d, h=%d, m=%ld\n", ring->name,
-           ring->ring_state->prod_tail, ring->ring_state->cons_head,
-           ring->max_elems);
-    */
 }
 
 void pvrdma_ring_free(PvrdmaRing *ring)
@@ -137,7 +126,6 @@ void pvrdma_ring_free(PvrdmaRing *ring)
         return;
     }
 
-    pr_dbg("ring->npages=%d\n", ring->npages);
     while (ring->npages--) {
         rdma_pci_dma_unmap(ring->dev, ring->pages[ring->npages],
                            TARGET_PAGE_SIZE);
diff --git a/hw/rdma/vmw/pvrdma_main.c b/hw/rdma/vmw/pvrdma_main.c
index d2bdb5ba8c..b6061f4b6e 100644
--- a/hw/rdma/vmw/pvrdma_main.c
+++ b/hw/rdma/vmw/pvrdma_main.c
@@ -69,25 +69,22 @@ static int init_dev_ring(PvrdmaRing *ring, struct pvrdma_ring **ring_state,
     uint64_t *dir, *tbl;
     int rc = 0;
 
-    pr_dbg("Initializing device ring %s\n", name);
-    pr_dbg("pdir_dma=0x%llx\n", (long long unsigned int)dir_addr);
-    pr_dbg("num_pages=%d\n", num_pages);
     dir = rdma_pci_dma_map(pci_dev, dir_addr, TARGET_PAGE_SIZE);
     if (!dir) {
-        pr_err("Failed to map to page directory\n");
+        rdma_error_report("Failed to map to page directory (ring %s)", name);
         rc = -ENOMEM;
         goto out;
     }
     tbl = rdma_pci_dma_map(pci_dev, dir[0], TARGET_PAGE_SIZE);
     if (!tbl) {
-        pr_err("Failed to map to page table\n");
+        rdma_error_report("Failed to map to page table (ring %s)", name);
         rc = -ENOMEM;
         goto out_free_dir;
     }
 
     *ring_state = rdma_pci_dma_map(pci_dev, tbl[0], TARGET_PAGE_SIZE);
     if (!*ring_state) {
-        pr_err("Failed to map to ring state\n");
+        rdma_error_report("Failed to map to ring state (ring %s)", name);
         rc = -ENOMEM;
         goto out_free_tbl;
     }
@@ -100,7 +97,6 @@ static int init_dev_ring(PvrdmaRing *ring, struct pvrdma_ring **ring_state,
                           sizeof(struct pvrdma_cqne),
                           (dma_addr_t *)&tbl[1], (dma_addr_t)num_pages - 1);
     if (rc) {
-        pr_err("Failed to initialize ring\n");
         rc = -ENOMEM;
         goto out_free_ring_state;
     }
@@ -155,11 +151,10 @@ static int load_dsr(PVRDMADev *dev)
     free_dsr(dev);
 
     /* Map to DSR */
-    pr_dbg("dsr_dma=0x%llx\n", (long long unsigned int)dev->dsr_info.dma);
     dev->dsr_info.dsr = rdma_pci_dma_map(pci_dev, dev->dsr_info.dma,
                               sizeof(struct pvrdma_device_shared_region));
     if (!dev->dsr_info.dsr) {
-        pr_err("Failed to map to DSR\n");
+        rdma_error_report("Failed to map to DSR");
         rc = -ENOMEM;
         goto out;
     }
@@ -169,21 +164,19 @@ static int load_dsr(PVRDMADev *dev)
     dsr = dsr_info->dsr;
 
     /* Map to command slot */
-    pr_dbg("cmd_dma=0x%llx\n", (long long unsigned int)dsr->cmd_slot_dma);
     dsr_info->req = rdma_pci_dma_map(pci_dev, dsr->cmd_slot_dma,
                                      sizeof(union pvrdma_cmd_req));
     if (!dsr_info->req) {
-        pr_err("Failed to map to command slot address\n");
+        rdma_error_report("Failed to map to command slot address");
         rc = -ENOMEM;
         goto out_free_dsr;
     }
 
     /* Map to response slot */
-    pr_dbg("rsp_dma=0x%llx\n", (long long unsigned int)dsr->resp_slot_dma);
     dsr_info->rsp = rdma_pci_dma_map(pci_dev, dsr->resp_slot_dma,
                                      sizeof(union pvrdma_cmd_resp));
     if (!dsr_info->rsp) {
-        pr_err("Failed to map to response slot address\n");
+        rdma_error_report("Failed to map to response slot address");
         rc = -ENOMEM;
         goto out_free_req;
     }
@@ -193,7 +186,6 @@ static int load_dsr(PVRDMADev *dev)
                        pci_dev, dsr->cq_ring_pages.pdir_dma,
                        dsr->cq_ring_pages.num_pages);
     if (rc) {
-        pr_err("Failed to map to initialize CQ ring\n");
         rc = -ENOMEM;
         goto out_free_rsp;
     }
@@ -203,7 +195,6 @@ static int load_dsr(PVRDMADev *dev)
                        "dev_async", pci_dev, dsr->async_ring_pages.pdir_dma,
                        dsr->async_ring_pages.num_pages);
     if (rc) {
-        pr_err("Failed to map to initialize event ring\n");
         rc = -ENOMEM;
         goto out_free_rsp;
     }
@@ -230,24 +221,15 @@ static void init_dsr_dev_caps(PVRDMADev *dev)
     struct pvrdma_device_shared_region *dsr;
 
     if (dev->dsr_info.dsr == NULL) {
-        pr_err("Can't initialized DSR\n");
+        rdma_error_report("Can't initialized DSR");
         return;
     }
 
     dsr = dev->dsr_info.dsr;
-
     dsr->caps.fw_ver = PVRDMA_FW_VERSION;
-    pr_dbg("fw_ver=0x%" PRIx64 "\n", dsr->caps.fw_ver);
-
     dsr->caps.mode = PVRDMA_DEVICE_MODE_ROCE;
-    pr_dbg("mode=%d\n", dsr->caps.mode);
-
     dsr->caps.gid_types |= PVRDMA_GID_TYPE_FLAG_ROCE_V1;
-    pr_dbg("gid_types=0x%x\n", dsr->caps.gid_types);
-
     dsr->caps.max_uar = RDMA_BAR2_UAR_SIZE;
-    pr_dbg("max_uar=%d\n", dsr->caps.max_uar);
-
     dsr->caps.max_mr_size = dev->dev_attr.max_mr_size;
     dsr->caps.max_qp = dev->dev_attr.max_qp;
     dsr->caps.max_qp_wr = dev->dev_attr.max_qp_wr;
@@ -257,23 +239,11 @@ static void init_dsr_dev_caps(PVRDMADev *dev)
     dsr->caps.max_mr = dev->dev_attr.max_mr;
     dsr->caps.max_pd = dev->dev_attr.max_pd;
     dsr->caps.max_ah = dev->dev_attr.max_ah;
-
     dsr->caps.gid_tbl_len = MAX_GIDS;
-    pr_dbg("gid_tbl_len=%d\n", dsr->caps.gid_tbl_len);
-
     dsr->caps.sys_image_guid = 0;
-    pr_dbg("sys_image_guid=%" PRIx64 "\n", dsr->caps.sys_image_guid);
-
     dsr->caps.node_guid = dev->node_guid;
-    pr_dbg("node_guid=%" PRIx64 "\n", be64_to_cpu(dsr->caps.node_guid));
-
     dsr->caps.phys_port_cnt = MAX_PORTS;
-    pr_dbg("phys_port_cnt=%d\n", dsr->caps.phys_port_cnt);
-
     dsr->caps.max_pkeys = MAX_PKEYS;
-    pr_dbg("max_pkeys=%d\n", dsr->caps.max_pkeys);
-
-    pr_dbg("Initialized\n");
 }
 
 static void uninit_msix(PCIDevice *pdev, int used_vectors)
@@ -319,9 +289,6 @@ static void pvrdma_fini(PCIDevice *pdev)
 {
     PVRDMADev *dev = PVRDMA_DEV(pdev);
 
-    pr_dbg("Closing device %s %x.%x\n", pdev->name, PCI_SLOT(pdev->devfn),
-           PCI_FUNC(pdev->devfn));
-
     pvrdma_qp_ops_fini();
 
     rdma_rm_fini(&dev->rdma_dev_res, &dev->backend_dev,
@@ -335,8 +302,8 @@ static void pvrdma_fini(PCIDevice *pdev)
         uninit_msix(pdev, RDMA_MAX_INTRS);
     }
 
-    pr_dbg("Device %s %x.%x is down\n", pdev->name, PCI_SLOT(pdev->devfn),
-           PCI_FUNC(pdev->devfn));
+    rdma_info_report("Device %s %x.%x is down", pdev->name,
+                     PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
 }
 
 static void pvrdma_stop(PVRDMADev *dev)
@@ -353,32 +320,26 @@ static void activate_device(PVRDMADev *dev)
 {
     pvrdma_start(dev);
     set_reg_val(dev, PVRDMA_REG_ERR, 0);
-    pr_dbg("Device activated\n");
 }
 
 static int unquiesce_device(PVRDMADev *dev)
 {
-    pr_dbg("Device unquiesced\n");
     return 0;
 }
 
 static void reset_device(PVRDMADev *dev)
 {
     pvrdma_stop(dev);
-
-    pr_dbg("Device reset complete\n");
 }
 
-static uint64_t regs_read(void *opaque, hwaddr addr, unsigned size)
+static uint64_t pvrdma_regs_read(void *opaque, hwaddr addr, unsigned size)
 {
     PVRDMADev *dev = opaque;
     uint32_t val;
 
-    /* pr_dbg("addr=0x%lx, size=%d\n", addr, size); */
-
     if (get_reg_val(dev, addr, &val)) {
-        pr_dbg("Error trying to read REG value from address 0x%x\n",
-               (uint32_t)addr);
+        rdma_error_report("Failed to read REG value from address 0x%x",
+                          (uint32_t)addr);
         return -EINVAL;
     }
 
@@ -387,25 +348,24 @@ static uint64_t regs_read(void *opaque, hwaddr addr, unsigned size)
     return val;
 }
 
-static void regs_write(void *opaque, hwaddr addr, uint64_t val, unsigned size)
+static void pvrdma_regs_write(void *opaque, hwaddr addr, uint64_t val,
+                              unsigned size)
 {
     PVRDMADev *dev = opaque;
 
-    /* pr_dbg("addr=0x%lx, val=0x%x, size=%d\n", addr, (uint32_t)val, size); */
-
     if (set_reg_val(dev, addr, val)) {
-        pr_err("Fail to set REG value, addr=0x%" PRIx64 ", val=0x%" PRIx64 "\n",
-               addr, val);
+        rdma_error_report("Failed to set REG value, addr=0x%"PRIx64 ", val=0x%"PRIx64,
+                          addr, val);
         return;
     }
 
-    trace_pvrdma_regs_write(addr, val);
-
     switch (addr) {
     case PVRDMA_REG_DSRLOW:
+        trace_pvrdma_regs_write(addr, val, "DSRLOW", "");
         dev->dsr_info.dma = val;
         break;
     case PVRDMA_REG_DSRHIGH:
+        trace_pvrdma_regs_write(addr, val, "DSRHIGH", "");
         dev->dsr_info.dma |= val << 32;
         load_dsr(dev);
         init_dsr_dev_caps(dev);
@@ -413,23 +373,27 @@ static void regs_write(void *opaque, hwaddr addr, uint64_t val, unsigned size)
     case PVRDMA_REG_CTL:
         switch (val) {
         case PVRDMA_DEVICE_CTL_ACTIVATE:
+            trace_pvrdma_regs_write(addr, val, "CTL", "ACTIVATE");
             activate_device(dev);
             break;
         case PVRDMA_DEVICE_CTL_UNQUIESCE:
+            trace_pvrdma_regs_write(addr, val, "CTL", "UNQUIESCE");
             unquiesce_device(dev);
             break;
         case PVRDMA_DEVICE_CTL_RESET:
+            trace_pvrdma_regs_write(addr, val, "CTL", "URESET");
             reset_device(dev);
             break;
         }
         break;
     case PVRDMA_REG_IMR:
-        pr_dbg("Interrupt mask=0x%" PRIx64 "\n", val);
+        trace_pvrdma_regs_write(addr, val, "INTR_MASK", "");
         dev->interrupt_mask = val;
         break;
     case PVRDMA_REG_REQUEST:
         if (val == 0) {
-            execute_command(dev);
+            trace_pvrdma_regs_write(addr, val, "REQUEST", "");
+            pvrdma_exec_cmd(dev);
         }
         break;
     default:
@@ -438,8 +402,8 @@ static void regs_write(void *opaque, hwaddr addr, uint64_t val, unsigned size)
 }
 
 static const MemoryRegionOps regs_ops = {
-    .read = regs_read,
-    .write = regs_write,
+    .read = pvrdma_regs_read,
+    .write = pvrdma_regs_write,
     .endianness = DEVICE_LITTLE_ENDIAN,
     .impl = {
         .min_access_size = sizeof(uint32_t),
@@ -447,54 +411,58 @@ static const MemoryRegionOps regs_ops = {
     },
 };
 
-static uint64_t uar_read(void *opaque, hwaddr addr, unsigned size)
+static uint64_t pvrdma_uar_read(void *opaque, hwaddr addr, unsigned size)
 {
     return 0xffffffff;
 }
 
-static void uar_write(void *opaque, hwaddr addr, uint64_t val, unsigned size)
+static void pvrdma_uar_write(void *opaque, hwaddr addr, uint64_t val,
+                             unsigned size)
 {
     PVRDMADev *dev = opaque;
 
-    /* pr_dbg("addr=0x%lx, val=0x%x, size=%d\n", addr, (uint32_t)val, size); */
-
     switch (addr & 0xFFF) { /* Mask with 0xFFF as each UC gets page */
     case PVRDMA_UAR_QP_OFFSET:
-        pr_dbg("UAR QP command, addr=0x%" PRIx64 ", val=0x%" PRIx64 "\n",
-               (uint64_t)addr, val);
         if (val & PVRDMA_UAR_QP_SEND) {
+            trace_pvrdma_uar_write(addr, val, "QP", "SEND",
+                                   val & PVRDMA_UAR_HANDLE_MASK, 0);
             pvrdma_qp_send(dev, val & PVRDMA_UAR_HANDLE_MASK);
         }
         if (val & PVRDMA_UAR_QP_RECV) {
+            trace_pvrdma_uar_write(addr, val, "QP", "RECV",
+                                   val & PVRDMA_UAR_HANDLE_MASK, 0);
             pvrdma_qp_recv(dev, val & PVRDMA_UAR_HANDLE_MASK);
         }
         break;
     case PVRDMA_UAR_CQ_OFFSET:
-        /* pr_dbg("UAR CQ cmd, addr=0x%x, val=0x%lx\n", (uint32_t)addr, val); */
         if (val & PVRDMA_UAR_CQ_ARM) {
+            trace_pvrdma_uar_write(addr, val, "CQ", "ARM",
+                                   val & PVRDMA_UAR_HANDLE_MASK,
+                                   !!(val & PVRDMA_UAR_CQ_ARM_SOL));
             rdma_rm_req_notify_cq(&dev->rdma_dev_res,
                                   val & PVRDMA_UAR_HANDLE_MASK,
                                   !!(val & PVRDMA_UAR_CQ_ARM_SOL));
         }
         if (val & PVRDMA_UAR_CQ_ARM_SOL) {
-            pr_dbg("UAR_CQ_ARM_SOL (%" PRIx64 ")\n",
-                   val & PVRDMA_UAR_HANDLE_MASK);
+            trace_pvrdma_uar_write(addr, val, "CQ", "ARMSOL - not supported", 0,
+                                   0);
         }
         if (val & PVRDMA_UAR_CQ_POLL) {
-            pr_dbg("UAR_CQ_POLL (%" PRIx64 ")\n", val & PVRDMA_UAR_HANDLE_MASK);
+            trace_pvrdma_uar_write(addr, val, "CQ", "POLL",
+                                   val & PVRDMA_UAR_HANDLE_MASK, 0);
             pvrdma_cq_poll(&dev->rdma_dev_res, val & PVRDMA_UAR_HANDLE_MASK);
         }
         break;
     default:
-        pr_err("Unsupported command, addr=0x%" PRIx64 ", val=0x%" PRIx64 "\n",
-               addr, val);
+        rdma_error_report("Unsupported command, addr=0x%"PRIx64", val=0x%"PRIx64,
+                          addr, val);
         break;
     }
 }
 
 static const MemoryRegionOps uar_ops = {
-    .read = uar_read,
-    .write = uar_write,
+    .read = pvrdma_uar_read,
+    .write = pvrdma_uar_write,
     .endianness = DEVICE_LITTLE_ENDIAN,
     .impl = {
         .min_access_size = sizeof(uint32_t),
@@ -551,11 +519,9 @@ static void init_dev_caps(PVRDMADev *dev)
                               (wr_sz + sizeof(struct pvrdma_sge) *
                               dev->dev_attr.max_sge) - TARGET_PAGE_SIZE;
                               /* First page is ring state  ^^^^ */
-    pr_dbg("max_qp_wr=%d\n", dev->dev_attr.max_qp_wr);
 
     dev->dev_attr.max_cqe = pg_tbl_bytes / sizeof(struct pvrdma_cqe) -
                             TARGET_PAGE_SIZE; /* First page is ring state */
-    pr_dbg("max_cqe=%d\n", dev->dev_attr.max_cqe);
 }
 
 static int pvrdma_check_ram_shared(Object *obj, void *opaque)
@@ -585,10 +551,8 @@ static void pvrdma_realize(PCIDevice *pdev, Error **errp)
     bool ram_shared = false;
     PCIDevice *func0;
 
-    init_pr_dbg();
-
-    pr_dbg("Initializing device %s %x.%x\n", pdev->name,
-           PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
+    rdma_info_report("Initializing device %s %x.%x", pdev->name,
+                     PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
 
     if (TARGET_PAGE_SIZE != getpagesize()) {
         error_setg(errp, "Target page size must be the same as host page size");
@@ -598,8 +562,6 @@ static void pvrdma_realize(PCIDevice *pdev, Error **errp)
     func0 = pci_get_function_0(pdev);
     /* Break if not vmxnet3 device in slot 0 */
     if (strcmp(object_get_typename(&func0->qdev.parent_obj), TYPE_VMXNET3)) {
-        pr_dbg("func0 type is %s\n",
-               object_get_typename(&func0->qdev.parent_obj));
         error_setg(errp, "Device on %x.0 must be %s", PCI_SLOT(pdev->devfn),
                    TYPE_VMXNET3);
         return;
@@ -656,7 +618,7 @@ static void pvrdma_realize(PCIDevice *pdev, Error **errp)
 out:
     if (rc) {
         pvrdma_fini(pdev);
-        error_append_hint(errp, "Device fail to load\n");
+        error_append_hint(errp, "Device failed to load\n");
     }
 }
 
diff --git a/hw/rdma/vmw/pvrdma_qp_ops.c b/hw/rdma/vmw/pvrdma_qp_ops.c
index ce5a60e184..16db726dac 100644
--- a/hw/rdma/vmw/pvrdma_qp_ops.c
+++ b/hw/rdma/vmw/pvrdma_qp_ops.c
@@ -19,6 +19,8 @@
 #include "../rdma_rm.h"
 #include "../rdma_backend.h"
 
+#include "trace.h"
+
 #include "pvrdma.h"
 #include "standard-headers/rdma/vmw_pvrdma-abi.h"
 #include "pvrdma_qp_ops.h"
@@ -55,18 +57,14 @@ static int pvrdma_post_cqe(PVRDMADev *dev, uint32_t cq_handle,
     RdmaRmCQ *cq = rdma_rm_get_cq(&dev->rdma_dev_res, cq_handle);
 
     if (unlikely(!cq)) {
-        pr_dbg("Invalid cqn %d\n", cq_handle);
         return -EINVAL;
     }
 
     ring = (PvrdmaRing *)cq->opaque;
-    pr_dbg("ring=%p\n", ring);
 
     /* Step #1: Put CQE on CQ ring */
-    pr_dbg("Writing CQE\n");
     cqe1 = pvrdma_ring_next_elem_write(ring);
     if (unlikely(!cqe1)) {
-        pr_dbg("No CQEs in ring\n");
         return -EINVAL;
     }
 
@@ -80,19 +78,13 @@ static int pvrdma_post_cqe(PVRDMADev *dev, uint32_t cq_handle,
     cqe1->wc_flags = wc->wc_flags;
     cqe1->vendor_err = wc->vendor_err;
 
-    pr_dbg("wr_id=%" PRIx64 "\n", cqe1->wr_id);
-    pr_dbg("qp=0x%lx\n", cqe1->qp);
-    pr_dbg("opcode=%d\n", cqe1->opcode);
-    pr_dbg("status=%d\n", cqe1->status);
-    pr_dbg("byte_len=%d\n", cqe1->byte_len);
-    pr_dbg("src_qp=%d\n", cqe1->src_qp);
-    pr_dbg("wc_flags=%d\n", cqe1->wc_flags);
-    pr_dbg("vendor_err=%d\n", cqe1->vendor_err);
+    trace_pvrdma_post_cqe(cq_handle, cq->notify, cqe1->wr_id, cqe1->qp,
+                          cqe1->opcode, cqe1->status, cqe1->byte_len,
+                          cqe1->src_qp, cqe1->wc_flags, cqe1->vendor_err);
 
     pvrdma_ring_write_inc(ring);
 
     /* Step #2: Put CQ number on dsr completion ring */
-    pr_dbg("Writing CQNE\n");
     cqne = pvrdma_ring_next_elem_write(&dev->dsr_info.cq);
     if (unlikely(!cqne)) {
         return -EINVAL;
@@ -101,7 +93,6 @@ static int pvrdma_post_cqe(PVRDMADev *dev, uint32_t cq_handle,
     cqne->info = cq_handle;
     pvrdma_ring_write_inc(&dev->dsr_info.cq);
 
-    pr_dbg("cq->notify=%d\n", cq->notify);
     if (cq->notify != CNT_CLEAR) {
         if (cq->notify == CNT_ARM) {
             cq->notify = CNT_CLEAR;
@@ -151,23 +142,17 @@ void pvrdma_qp_send(PVRDMADev *dev, uint32_t qp_handle)
     int sgid_idx;
     union ibv_gid *sgid;
 
-    pr_dbg("qp_handle=0x%x\n", qp_handle);
-
     qp = rdma_rm_get_qp(&dev->rdma_dev_res, qp_handle);
     if (unlikely(!qp)) {
-        pr_dbg("Invalid qpn\n");
         return;
     }
 
     ring = (PvrdmaRing *)qp->opaque;
-    pr_dbg("sring=%p\n", ring);
 
     wqe = (struct PvrdmaSqWqe *)pvrdma_ring_next_elem_read(ring);
     while (wqe) {
         CompHandlerCtx *comp_ctx;
 
-        pr_dbg("wr_id=%" PRIx64 "\n", wqe->hdr.wr_id);
-
         /* Prepare CQE */
         comp_ctx = g_malloc(sizeof(CompHandlerCtx));
         comp_ctx->dev = dev;
@@ -178,26 +163,25 @@ void pvrdma_qp_send(PVRDMADev *dev, uint32_t qp_handle)
 
         sgid = rdma_rm_get_gid(&dev->rdma_dev_res, wqe->hdr.wr.ud.av.gid_index);
         if (!sgid) {
-            pr_dbg("Fail to get gid for idx %d\n", wqe->hdr.wr.ud.av.gid_index);
+            rdma_error_report("Failed to get gid for idx %d",
+                              wqe->hdr.wr.ud.av.gid_index);
             complete_with_error(VENDOR_ERR_INV_GID_IDX, comp_ctx);
             continue;
         }
-        pr_dbg("sgid_id=%d, sgid=0x%llx\n", wqe->hdr.wr.ud.av.gid_index,
-               sgid->global.interface_id);
 
         sgid_idx = rdma_rm_get_backend_gid_index(&dev->rdma_dev_res,
                                                  &dev->backend_dev,
                                                  wqe->hdr.wr.ud.av.gid_index);
         if (sgid_idx <= 0) {
-            pr_dbg("Fail to get bk sgid_idx for sgid_idx %d\n",
-                   wqe->hdr.wr.ud.av.gid_index);
+            rdma_error_report("Failed to get bk sgid_idx for sgid_idx %d",
+                              wqe->hdr.wr.ud.av.gid_index);
             complete_with_error(VENDOR_ERR_INV_GID_IDX, comp_ctx);
             continue;
         }
 
         if (wqe->hdr.num_sge > dev->dev_attr.max_sge) {
-            pr_dbg("Invalid num_sge=%d (max %d)\n", wqe->hdr.num_sge,
-                   dev->dev_attr.max_sge);
+            rdma_error_report("Invalid num_sge=%d (max %d)", wqe->hdr.num_sge,
+                              dev->dev_attr.max_sge);
             complete_with_error(VENDOR_ERR_INV_NUM_SGE, comp_ctx);
             continue;
         }
@@ -221,23 +205,17 @@ void pvrdma_qp_recv(PVRDMADev *dev, uint32_t qp_handle)
     PvrdmaRqWqe *wqe;
     PvrdmaRing *ring;
 
-    pr_dbg("qp_handle=0x%x\n", qp_handle);
-
     qp = rdma_rm_get_qp(&dev->rdma_dev_res, qp_handle);
     if (unlikely(!qp)) {
-        pr_dbg("Invalid qpn\n");
         return;
     }
 
     ring = &((PvrdmaRing *)qp->opaque)[1];
-    pr_dbg("rring=%p\n", ring);
 
     wqe = (struct PvrdmaRqWqe *)pvrdma_ring_next_elem_read(ring);
     while (wqe) {
         CompHandlerCtx *comp_ctx;
 
-        pr_dbg("wr_id=%" PRIx64 "\n", wqe->hdr.wr_id);
-
         /* Prepare CQE */
         comp_ctx = g_malloc(sizeof(CompHandlerCtx));
         comp_ctx->dev = dev;
@@ -247,8 +225,8 @@ void pvrdma_qp_recv(PVRDMADev *dev, uint32_t qp_handle)
         comp_ctx->cqe.opcode = IBV_WC_RECV;
 
         if (wqe->hdr.num_sge > dev->dev_attr.max_sge) {
-            pr_dbg("Invalid num_sge=%d (max %d)\n", wqe->hdr.num_sge,
-                   dev->dev_attr.max_sge);
+            rdma_error_report("Invalid num_sge=%d (max %d)", wqe->hdr.num_sge,
+                              dev->dev_attr.max_sge);
             complete_with_error(VENDOR_ERR_INV_NUM_SGE, comp_ctx);
             continue;
         }
@@ -270,7 +248,6 @@ void pvrdma_cq_poll(RdmaDeviceResources *dev_res, uint32_t cq_handle)
 
     cq = rdma_rm_get_cq(dev_res, cq_handle);
     if (!cq) {
-        pr_dbg("Invalid CQ# %d\n", cq_handle);
         return;
     }
 
diff --git a/hw/rdma/vmw/trace-events b/hw/rdma/vmw/trace-events
index b3f9e2b19f..0122266ad7 100644
--- a/hw/rdma/vmw/trace-events
+++ b/hw/rdma/vmw/trace-events
@@ -1,5 +1,17 @@
 # See docs/tracing.txt for syntax documentation.
 
 # hw/rdma/vmw/pvrdma_main.c
-pvrdma_regs_read(uint64_t addr, uint64_t val) "regs[0x%"PRIx64"] = 0x%"PRIx64
-pvrdma_regs_write(uint64_t addr, uint64_t val) "regs[0x%"PRIx64"] = 0x%"PRIx64
+pvrdma_regs_read(uint64_t addr, uint64_t val) "pvrdma.regs[0x%"PRIx64"]=0x%"PRIx64
+pvrdma_regs_write(uint64_t addr, uint64_t val, const char *reg_name, const char *val_name) "pvrdma.regs[0x%"PRIx64"]=0x%"PRIx64" (%s %s)"
+pvrdma_uar_write(uint64_t addr, uint64_t val, const char *reg_name, const char *val_name, int val1, int val2) "uar[0x%"PRIx64"]=0x%"PRIx64" (cls=%s, op=%s, obj=%d, val=%d)"
+
+# hw/rdma/vmw/pvrdma_cmd.c
+pvrdma_map_to_pdir_host_virt(void *vfirst, void *vremaped) "mremap %p -> %p"
+pvrdma_map_to_pdir_next_page(int page_idx, void *vnext, void *vremaped) "mremap [%d] %p -> %p"
+pvrdma_exec_cmd(int cmd, int err) "cmd=%d, err=%d"
+
+# hw/rdma/vmw/pvrdma_dev_ring.c
+pvrdma_ring_next_elem_read_no_data(char *ring_name) "pvrdma_ring %s is empty"
+
+# hw/rdma/vmw/pvrdma_qp_ops.c
+pvrdma_post_cqe(uint32_t cq_handle, int notify, uint64_t wr_id, uint64_t qpn, uint32_t op_code, uint32_t status, uint32_t byte_len, uint32_t src_qp, uint32_t wc_flags, uint32_t vendor_err) "cq_handle=%d, notify=%d, wr_id=0x%"PRIx64", qpn=0x%"PRIx64", opcode=%d, status=%d, byte_len=%d, src_qp=%d, wc_flags=%d, vendor_err=%d"
-- 
2.17.2

[Qemu-devel] [PATCH v4 2/9] hw/rdma: Introduce protected qlist

[Yuval Shaia]

To make code more readable move handling of protected list to a
rdma_utils

Signed-off-by: Yuval Shaia <yuval.shaia@oracle.com>
Reviewed-by: Marcel Apfelbaum <marcel.apfelbaum@gmail.com>
---
 hw/rdma/rdma_backend.c      | 20 +++++--------------
 hw/rdma/rdma_backend_defs.h |  8 ++------
 hw/rdma/rdma_utils.c        | 39 +++++++++++++++++++++++++++++++++++++
 hw/rdma/rdma_utils.h        |  9 +++++++++
 4 files changed, 55 insertions(+), 21 deletions(-)

diff --git a/hw/rdma/rdma_backend.c b/hw/rdma/rdma_backend.c
index 24bac00a20..0ed14751be 100644
--- a/hw/rdma/rdma_backend.c
+++ b/hw/rdma/rdma_backend.c
@@ -527,9 +527,7 @@ static unsigned int save_mad_recv_buffer(RdmaBackendDev *backend_dev,
     bctx->up_ctx = ctx;
     bctx->sge = *sge;
 
-    qemu_mutex_lock(&backend_dev->recv_mads_list.lock);
-    qlist_append_int(backend_dev->recv_mads_list.list, bctx_id);
-    qemu_mutex_unlock(&backend_dev->recv_mads_list.lock);
+    rdma_protected_qlist_append_int64(&backend_dev->recv_mads_list, bctx_id);
 
     return 0;
 }
@@ -913,23 +911,19 @@ static inline void build_mad_hdr(struct ibv_grh *grh, union ibv_gid *sgid,
 static void process_incoming_mad_req(RdmaBackendDev *backend_dev,
                                      RdmaCmMuxMsg *msg)
 {
-    QObject *o_ctx_id;
     unsigned long cqe_ctx_id;
     BackendCtx *bctx;
     char *mad;
 
     trace_mad_message("recv", msg->umad.mad, msg->umad_len);
 
-    qemu_mutex_lock(&backend_dev->recv_mads_list.lock);
-    o_ctx_id = qlist_pop(backend_dev->recv_mads_list.list);
-    qemu_mutex_unlock(&backend_dev->recv_mads_list.lock);
-    if (!o_ctx_id) {
+    cqe_ctx_id = rdma_protected_qlist_pop_int64(&backend_dev->recv_mads_list);
+    if (cqe_ctx_id == -ENOENT) {
         rdma_warn_report("No more free MADs buffers, waiting for a while");
         sleep(THR_POLL_TO);
         return;
     }
 
-    cqe_ctx_id = qnum_get_uint(qobject_to(QNum, o_ctx_id));
     bctx = rdma_rm_get_cqe_ctx(backend_dev->rdma_dev_res, cqe_ctx_id);
     if (unlikely(!bctx)) {
         rdma_error_report("No matching ctx for req %ld", cqe_ctx_id);
@@ -994,8 +988,7 @@ static int mad_init(RdmaBackendDev *backend_dev, CharBackend *mad_chr_be)
         return -EIO;
     }
 
-    qemu_mutex_init(&backend_dev->recv_mads_list.lock);
-    backend_dev->recv_mads_list.list = qlist_new();
+    rdma_protected_qlist_init(&backend_dev->recv_mads_list);
 
     enable_rdmacm_mux_async(backend_dev);
 
@@ -1010,10 +1003,7 @@ static void mad_fini(RdmaBackendDev *backend_dev)
 {
     disable_rdmacm_mux_async(backend_dev);
     qemu_chr_fe_disconnect(backend_dev->rdmacm_mux.chr_be);
-    if (backend_dev->recv_mads_list.list) {
-        qlist_destroy_obj(QOBJECT(backend_dev->recv_mads_list.list));
-        qemu_mutex_destroy(&backend_dev->recv_mads_list.lock);
-    }
+    rdma_protected_qlist_destroy(&backend_dev->recv_mads_list);
 }
 
 int rdma_backend_get_gid_index(RdmaBackendDev *backend_dev,
diff --git a/hw/rdma/rdma_backend_defs.h b/hw/rdma/rdma_backend_defs.h
index 15ae8b970e..a8c15b09ab 100644
--- a/hw/rdma/rdma_backend_defs.h
+++ b/hw/rdma/rdma_backend_defs.h
@@ -20,6 +20,7 @@
 #include "chardev/char-fe.h"
 #include <infiniband/verbs.h>
 #include "contrib/rdmacm-mux/rdmacm-mux.h"
+#include "rdma_utils.h"
 
 typedef struct RdmaDeviceResources RdmaDeviceResources;
 
@@ -30,11 +31,6 @@ typedef struct RdmaBackendThread {
     bool is_running; /* Set by the thread to report its status */
 } RdmaBackendThread;
 
-typedef struct RecvMadList {
-    QemuMutex lock;
-    QList *list;
-} RecvMadList;
-
 typedef struct RdmaCmMux {
     CharBackend *chr_be;
     int can_receive;
@@ -48,7 +44,7 @@ typedef struct RdmaBackendDev {
     struct ibv_context *context;
     struct ibv_comp_channel *channel;
     uint8_t port_num;
-    RecvMadList recv_mads_list;
+    RdmaProtectedQList recv_mads_list;
     RdmaCmMux rdmacm_mux;
 } RdmaBackendDev;
 
diff --git a/hw/rdma/rdma_utils.c b/hw/rdma/rdma_utils.c
index b9f07fcda7..0a8abe572d 100644
--- a/hw/rdma/rdma_utils.c
+++ b/hw/rdma/rdma_utils.c
@@ -14,6 +14,8 @@
  */
 
 #include "qemu/osdep.h"
+#include "qapi/qmp/qlist.h"
+#include "qapi/qmp/qnum.h"
 #include "trace.h"
 #include "rdma_utils.h"
 
@@ -51,3 +53,40 @@ void rdma_pci_dma_unmap(PCIDevice *dev, void *buffer, dma_addr_t len)
         pci_dma_unmap(dev, buffer, len, DMA_DIRECTION_TO_DEVICE, 0);
     }
 }
+
+void rdma_protected_qlist_init(RdmaProtectedQList *list)
+{
+    qemu_mutex_init(&list->lock);
+    list->list = qlist_new();
+}
+
+void rdma_protected_qlist_destroy(RdmaProtectedQList *list)
+{
+    if (list->list) {
+        qlist_destroy_obj(QOBJECT(list->list));
+        qemu_mutex_destroy(&list->lock);
+        list->list = NULL;
+    }
+}
+
+void rdma_protected_qlist_append_int64(RdmaProtectedQList *list, int64_t value)
+{
+    qemu_mutex_lock(&list->lock);
+    qlist_append_int(list->list, value);
+    qemu_mutex_unlock(&list->lock);
+}
+
+int64_t rdma_protected_qlist_pop_int64(RdmaProtectedQList *list)
+{
+    QObject *obj;
+
+    qemu_mutex_lock(&list->lock);
+    obj = qlist_pop(list->list);
+    qemu_mutex_unlock(&list->lock);
+
+    if (!obj) {
+        return -ENOENT;
+    }
+
+    return qnum_get_uint(qobject_to(QNum, obj));
+}
diff --git a/hw/rdma/rdma_utils.h b/hw/rdma/rdma_utils.h
index acd148837f..a8bf1d4fec 100644
--- a/hw/rdma/rdma_utils.h
+++ b/hw/rdma/rdma_utils.h
@@ -29,8 +29,17 @@
 #define rdma_info_report(fmt, ...) \
     info_report("%s: " fmt, "rdma", ## __VA_ARGS__)
 
+typedef struct RdmaProtectedQList {
+    QemuMutex lock;
+    QList *list;
+} RdmaProtectedQList;
+
 void *rdma_pci_dma_map(PCIDevice *dev, dma_addr_t addr, dma_addr_t plen);
 void rdma_pci_dma_unmap(PCIDevice *dev, void *buffer, dma_addr_t len);
+void rdma_protected_qlist_init(RdmaProtectedQList *list);
+void rdma_protected_qlist_destroy(RdmaProtectedQList *list);
+void rdma_protected_qlist_append_int64(RdmaProtectedQList *list, int64_t value);
+int64_t rdma_protected_qlist_pop_int64(RdmaProtectedQList *list);
 
 static inline void addrconf_addr_eui48(uint8_t *eui, const char *addr)
 {
-- 
2.17.2

[Qemu-devel] [PATCH v4 3/9] hw/rdma: Protect against concurrent execution of poll_cq

[Yuval Shaia]

The function rdma_poll_cq is called from two contexts - completion
handler thread which sense new completion on backend channel and
explicitly as result of guest issuing poll_cq command.

Add lock to protect against concurrent executions.

Signed-off-by: Yuval Shaia <yuval.shaia@oracle.com>
Reviewed-by: Marcel Apfelbaum <marcel.apfelbaum@gmail.com>
---
 hw/rdma/rdma_backend.c | 2 ++
 hw/rdma/rdma_rm.c      | 4 ++++
 hw/rdma/rdma_rm_defs.h | 1 +
 3 files changed, 7 insertions(+)

diff --git a/hw/rdma/rdma_backend.c b/hw/rdma/rdma_backend.c
index 0ed14751be..9679b842d1 100644
--- a/hw/rdma/rdma_backend.c
+++ b/hw/rdma/rdma_backend.c
@@ -70,6 +70,7 @@ static void rdma_poll_cq(RdmaDeviceResources *rdma_dev_res, struct ibv_cq *ibcq)
     BackendCtx *bctx;
     struct ibv_wc wc[2];
 
+    qemu_mutex_lock(&rdma_dev_res->lock);
     do {
         ne = ibv_poll_cq(ibcq, ARRAY_SIZE(wc), wc);
 
@@ -89,6 +90,7 @@ static void rdma_poll_cq(RdmaDeviceResources *rdma_dev_res, struct ibv_cq *ibcq)
             g_free(bctx);
         }
     } while (ne > 0);
+    qemu_mutex_unlock(&rdma_dev_res->lock);
 
     if (ne < 0) {
         rdma_error_report("ibv_poll_cq fail, rc=%d, errno=%d", ne, errno);
diff --git a/hw/rdma/rdma_rm.c b/hw/rdma/rdma_rm.c
index 5dab4a2189..14580ca379 100644
--- a/hw/rdma/rdma_rm.c
+++ b/hw/rdma/rdma_rm.c
@@ -618,12 +618,16 @@ int rdma_rm_init(RdmaDeviceResources *dev_res, struct ibv_device_attr *dev_attr,
 
     init_ports(dev_res);
 
+    qemu_mutex_init(&dev_res->lock);
+
     return 0;
 }
 
 void rdma_rm_fini(RdmaDeviceResources *dev_res, RdmaBackendDev *backend_dev,
                   const char *ifname)
 {
+    qemu_mutex_destroy(&dev_res->lock);
+
     fini_ports(dev_res, backend_dev, ifname);
 
     res_tbl_free(&dev_res->uc_tbl);
diff --git a/hw/rdma/rdma_rm_defs.h b/hw/rdma/rdma_rm_defs.h
index 0ba61d1838..f0ee1f3072 100644
--- a/hw/rdma/rdma_rm_defs.h
+++ b/hw/rdma/rdma_rm_defs.h
@@ -105,6 +105,7 @@ typedef struct RdmaDeviceResources {
     RdmaRmResTbl cq_tbl;
     RdmaRmResTbl cqe_ctx_tbl;
     GHashTable *qp_hash; /* Keeps mapping between real and emulated */
+    QemuMutex lock;
 } RdmaDeviceResources;
 
 #endif
-- 
2.17.2

[Qemu-devel] [PATCH v4 4/9] {hmp, hw/pvrdma}: Expose device internals via monitor interface

[Yuval Shaia]

Allow interrogating device internals through HMP interface.
The exposed indicators can be used for troubleshooting by developers or
sysadmin.
There is no need to expose these attributes to a management system (e.x.
libvirt) because (1) most of them are not "device-management' related
info and (2) there is no guarantee the interface is stable.

Signed-off-by: Yuval Shaia <yuval.shaia@oracle.com>
---
 hmp-commands-info.hx       | 14 ++++++++
 hmp.c                      | 27 +++++++++++++++
 hmp.h                      |  1 +
 hw/rdma/Makefile.objs      |  2 +-
 hw/rdma/rdma_backend.c     | 70 +++++++++++++++++++++++++++++---------
 hw/rdma/rdma_hmp.c         | 30 ++++++++++++++++
 hw/rdma/rdma_rm.c          | 60 ++++++++++++++++++++++++++++++++
 hw/rdma/rdma_rm.h          |  1 +
 hw/rdma/rdma_rm_defs.h     | 27 ++++++++++++++-
 hw/rdma/vmw/pvrdma.h       |  5 +++
 hw/rdma/vmw/pvrdma_main.c  | 21 ++++++++++++
 include/hw/rdma/rdma_hmp.h | 40 ++++++++++++++++++++++
 12 files changed, 279 insertions(+), 19 deletions(-)
 create mode 100644 hw/rdma/rdma_hmp.c
 create mode 100644 include/hw/rdma/rdma_hmp.h

diff --git a/hmp-commands-info.hx b/hmp-commands-info.hx
index cbee8b944d..c59444c461 100644
--- a/hmp-commands-info.hx
+++ b/hmp-commands-info.hx
@@ -202,6 +202,20 @@ STEXI
 @item info pic
 @findex info pic
 Show PIC state.
+ETEXI
+
+    {
+        .name       = "rdma",
+        .args_type  = "",
+        .params     = "",
+        .help       = "show RDMA state",
+        .cmd        = hmp_info_rdma,
+    },
+
+STEXI
+@item info rdma
+@findex info rdma
+Show RDMA state.
 ETEXI
 
     {
diff --git a/hmp.c b/hmp.c
index 1e006eeb49..68511b2441 100644
--- a/hmp.c
+++ b/hmp.c
@@ -51,6 +51,7 @@
 #include "qemu/error-report.h"
 #include "exec/ramlist.h"
 #include "hw/intc/intc.h"
+#include "hw/rdma/rdma_hmp.h"
 #include "migration/snapshot.h"
 #include "migration/misc.h"
 
@@ -968,6 +969,32 @@ void hmp_info_pic(Monitor *mon, const QDict *qdict)
                                    hmp_info_pic_foreach, mon);
 }
 
+static int hmp_info_rdma_foreach(Object *obj, void *opaque)
+{
+    RdmaStatsProvider *rdma;
+    RdmaStatsProviderClass *k;
+    Monitor *mon = opaque;
+
+    if (object_dynamic_cast(obj, TYPE_RDMA_STATS_PROVIDER)) {
+        rdma = RDMA_STATS_PROVIDER(obj);
+        k = RDMA_STATS_PROVIDER_GET_CLASS(obj);
+        if (k->print_statistics) {
+            k->print_statistics(mon, rdma);
+        } else {
+            monitor_printf(mon, "RDMA statistics not available for %s.\n",
+                           object_get_typename(obj));
+        }
+    }
+
+    return 0;
+}
+
+void hmp_info_rdma(Monitor *mon, const QDict *qdict)
+{
+    object_child_foreach_recursive(object_get_root(),
+                                   hmp_info_rdma_foreach, mon);
+}
+
 void hmp_info_pci(Monitor *mon, const QDict *qdict)
 {
     PciInfoList *info_list, *info;
diff --git a/hmp.h b/hmp.h
index 5f1addcca2..666949afc3 100644
--- a/hmp.h
+++ b/hmp.h
@@ -36,6 +36,7 @@ void hmp_info_spice(Monitor *mon, const QDict *qdict);
 void hmp_info_balloon(Monitor *mon, const QDict *qdict);
 void hmp_info_irq(Monitor *mon, const QDict *qdict);
 void hmp_info_pic(Monitor *mon, const QDict *qdict);
+void hmp_info_rdma(Monitor *mon, const QDict *qdict);
 void hmp_info_pci(Monitor *mon, const QDict *qdict);
 void hmp_info_block_jobs(Monitor *mon, const QDict *qdict);
 void hmp_info_tpm(Monitor *mon, const QDict *qdict);
diff --git a/hw/rdma/Makefile.objs b/hw/rdma/Makefile.objs
index bd36cbf51c..dd59faff51 100644
--- a/hw/rdma/Makefile.objs
+++ b/hw/rdma/Makefile.objs
@@ -1,5 +1,5 @@
 ifeq ($(CONFIG_PVRDMA),y)
-obj-$(CONFIG_PCI) += rdma_utils.o rdma_backend.o rdma_rm.o
+obj-$(CONFIG_PCI) += rdma_utils.o rdma_backend.o rdma_rm.o rdma_hmp.o
 obj-$(CONFIG_PCI) += vmw/pvrdma_dev_ring.o vmw/pvrdma_cmd.o \
                      vmw/pvrdma_qp_ops.o vmw/pvrdma_main.o
 endif
diff --git a/hw/rdma/rdma_backend.c b/hw/rdma/rdma_backend.c
index 9679b842d1..bc2fefcf93 100644
--- a/hw/rdma/rdma_backend.c
+++ b/hw/rdma/rdma_backend.c
@@ -64,9 +64,9 @@ static inline void complete_work(enum ibv_wc_status status, uint32_t vendor_err,
     comp_handler(ctx, &wc);
 }
 
-static void rdma_poll_cq(RdmaDeviceResources *rdma_dev_res, struct ibv_cq *ibcq)
+static int rdma_poll_cq(RdmaDeviceResources *rdma_dev_res, struct ibv_cq *ibcq)
 {
-    int i, ne;
+    int i, ne, total_ne = 0;
     BackendCtx *bctx;
     struct ibv_wc wc[2];
 
@@ -89,12 +89,18 @@ static void rdma_poll_cq(RdmaDeviceResources *rdma_dev_res, struct ibv_cq *ibcq)
             rdma_rm_dealloc_cqe_ctx(rdma_dev_res, wc[i].wr_id);
             g_free(bctx);
         }
+        total_ne += ne;
     } while (ne > 0);
+    atomic_sub(&rdma_dev_res->stats.missing_cqe, total_ne);
     qemu_mutex_unlock(&rdma_dev_res->lock);
 
     if (ne < 0) {
         rdma_error_report("ibv_poll_cq fail, rc=%d, errno=%d", ne, errno);
     }
+
+    rdma_dev_res->stats.completions += total_ne;
+
+    return total_ne;
 }
 
 static void *comp_handler_thread(void *arg)
@@ -122,6 +128,9 @@ static void *comp_handler_thread(void *arg)
     while (backend_dev->comp_thread.run) {
         do {
             rc = qemu_poll_ns(pfds, 1, THR_POLL_TO * (int64_t)SCALE_MS);
+            if (!rc) {
+                backend_dev->rdma_dev_res->stats.poll_cq_ppoll_to++;
+            }
         } while (!rc && backend_dev->comp_thread.run);
 
         if (backend_dev->comp_thread.run) {
@@ -138,6 +147,7 @@ static void *comp_handler_thread(void *arg)
                                   errno);
             }
 
+            backend_dev->rdma_dev_res->stats.poll_cq_from_bk++;
             rdma_poll_cq(backend_dev->rdma_dev_res, ev_cq);
 
             ibv_ack_cq_events(ev_cq, 1);
@@ -271,7 +281,13 @@ int rdma_backend_query_port(RdmaBackendDev *backend_dev,
 
 void rdma_backend_poll_cq(RdmaDeviceResources *rdma_dev_res, RdmaBackendCQ *cq)
 {
-    rdma_poll_cq(rdma_dev_res, cq->ibcq);
+    int polled;
+
+    rdma_dev_res->stats.poll_cq_from_guest++;
+    polled = rdma_poll_cq(rdma_dev_res, cq->ibcq);
+    if (!polled) {
+        rdma_dev_res->stats.poll_cq_from_guest_empty++;
+    }
 }
 
 static GHashTable *ah_hash;
@@ -333,7 +349,7 @@ static void ah_cache_init(void)
 
 static int build_host_sge_array(RdmaDeviceResources *rdma_dev_res,
                                 struct ibv_sge *dsge, struct ibv_sge *ssge,
-                                uint8_t num_sge)
+                                uint8_t num_sge, uint64_t *total_length)
 {
     RdmaRmMR *mr;
     int ssge_idx;
@@ -349,6 +365,8 @@ static int build_host_sge_array(RdmaDeviceResources *rdma_dev_res,
         dsge->length = ssge[ssge_idx].length;
         dsge->lkey = rdma_backend_mr_lkey(&mr->backend_mr);
 
+        *total_length += dsge->length;
+
         dsge++;
     }
 
@@ -445,8 +463,10 @@ void rdma_backend_post_send(RdmaBackendDev *backend_dev,
             rc = mad_send(backend_dev, sgid_idx, sgid, sge, num_sge);
             if (rc) {
                 complete_work(IBV_WC_GENERAL_ERR, VENDOR_ERR_MAD_SEND, ctx);
+                backend_dev->rdma_dev_res->stats.mad_tx_err++;
             } else {
                 complete_work(IBV_WC_SUCCESS, 0, ctx);
+                backend_dev->rdma_dev_res->stats.mad_tx++;
             }
         }
         return;
@@ -458,20 +478,21 @@ void rdma_backend_post_send(RdmaBackendDev *backend_dev,
     rc = rdma_rm_alloc_cqe_ctx(backend_dev->rdma_dev_res, &bctx_id, bctx);
     if (unlikely(rc)) {
         complete_work(IBV_WC_GENERAL_ERR, VENDOR_ERR_NOMEM, ctx);
-        goto out_free_bctx;
+        goto err_free_bctx;
     }
 
-    rc = build_host_sge_array(backend_dev->rdma_dev_res, new_sge, sge, num_sge);
+    rc = build_host_sge_array(backend_dev->rdma_dev_res, new_sge, sge, num_sge,
+                              &backend_dev->rdma_dev_res->stats.tx_len);
     if (rc) {
         complete_work(IBV_WC_GENERAL_ERR, rc, ctx);
-        goto out_dealloc_cqe_ctx;
+        goto err_dealloc_cqe_ctx;
     }
 
     if (qp_type == IBV_QPT_UD) {
         wr.wr.ud.ah = create_ah(backend_dev, qp->ibpd, sgid_idx, dgid);
         if (!wr.wr.ud.ah) {
             complete_work(IBV_WC_GENERAL_ERR, VENDOR_ERR_FAIL_BACKEND, ctx);
-            goto out_dealloc_cqe_ctx;
+            goto err_dealloc_cqe_ctx;
         }
         wr.wr.ud.remote_qpn = dqpn;
         wr.wr.ud.remote_qkey = dqkey;
@@ -488,15 +509,19 @@ void rdma_backend_post_send(RdmaBackendDev *backend_dev,
         rdma_error_report("ibv_post_send fail, qpn=0x%x, rc=%d, errno=%d",
                           qp->ibqp->qp_num, rc, errno);
         complete_work(IBV_WC_GENERAL_ERR, VENDOR_ERR_FAIL_BACKEND, ctx);
-        goto out_dealloc_cqe_ctx;
+        goto err_dealloc_cqe_ctx;
     }
 
+    atomic_inc(&backend_dev->rdma_dev_res->stats.missing_cqe);
+    backend_dev->rdma_dev_res->stats.tx++;
+
     return;
 
-out_dealloc_cqe_ctx:
+err_dealloc_cqe_ctx:
+    backend_dev->rdma_dev_res->stats.tx_err++;
     rdma_rm_dealloc_cqe_ctx(backend_dev->rdma_dev_res, bctx_id);
 
-out_free_bctx:
+err_free_bctx:
     g_free(bctx);
 }
 
@@ -554,6 +579,9 @@ void rdma_backend_post_recv(RdmaBackendDev *backend_dev,
             rc = save_mad_recv_buffer(backend_dev, sge, num_sge, ctx);
             if (rc) {
                 complete_work(IBV_WC_GENERAL_ERR, rc, ctx);
+                rdma_dev_res->stats.mad_rx_bufs_err++;
+            } else {
+                rdma_dev_res->stats.mad_rx_bufs++;
             }
         }
         return;
@@ -565,13 +593,14 @@ void rdma_backend_post_recv(RdmaBackendDev *backend_dev,
     rc = rdma_rm_alloc_cqe_ctx(rdma_dev_res, &bctx_id, bctx);
     if (unlikely(rc)) {
         complete_work(IBV_WC_GENERAL_ERR, VENDOR_ERR_NOMEM, ctx);
-        goto out_free_bctx;
+        goto err_free_bctx;
     }
 
-    rc = build_host_sge_array(rdma_dev_res, new_sge, sge, num_sge);
+    rc = build_host_sge_array(rdma_dev_res, new_sge, sge, num_sge,
+                              &backend_dev->rdma_dev_res->stats.rx_bufs_len);
     if (rc) {
         complete_work(IBV_WC_GENERAL_ERR, rc, ctx);
-        goto out_dealloc_cqe_ctx;
+        goto err_dealloc_cqe_ctx;
     }
 
     wr.num_sge = num_sge;
@@ -582,15 +611,19 @@ void rdma_backend_post_recv(RdmaBackendDev *backend_dev,
         rdma_error_report("ibv_post_recv fail, qpn=0x%x, rc=%d, errno=%d",
                           qp->ibqp->qp_num, rc, errno);
         complete_work(IBV_WC_GENERAL_ERR, VENDOR_ERR_FAIL_BACKEND, ctx);
-        goto out_dealloc_cqe_ctx;
+        goto err_dealloc_cqe_ctx;
     }
 
+    atomic_inc(&backend_dev->rdma_dev_res->stats.missing_cqe);
+    rdma_dev_res->stats.rx_bufs++;
+
     return;
 
-out_dealloc_cqe_ctx:
+err_dealloc_cqe_ctx:
+    backend_dev->rdma_dev_res->stats.rx_bufs_err++;
     rdma_rm_dealloc_cqe_ctx(rdma_dev_res, bctx_id);
 
-out_free_bctx:
+err_free_bctx:
     g_free(bctx);
 }
 
@@ -929,12 +962,14 @@ static void process_incoming_mad_req(RdmaBackendDev *backend_dev,
     bctx = rdma_rm_get_cqe_ctx(backend_dev->rdma_dev_res, cqe_ctx_id);
     if (unlikely(!bctx)) {
         rdma_error_report("No matching ctx for req %ld", cqe_ctx_id);
+        backend_dev->rdma_dev_res->stats.mad_rx_err++;
         return;
     }
 
     mad = rdma_pci_dma_map(backend_dev->dev, bctx->sge.addr,
                            bctx->sge.length);
     if (!mad || bctx->sge.length < msg->umad_len + MAD_HDR_SIZE) {
+        backend_dev->rdma_dev_res->stats.mad_rx_err++;
         complete_work(IBV_WC_GENERAL_ERR, VENDOR_ERR_INV_MAD_BUFF,
                       bctx->up_ctx);
     } else {
@@ -949,6 +984,7 @@ static void process_incoming_mad_req(RdmaBackendDev *backend_dev,
         wc.byte_len = msg->umad_len;
         wc.status = IBV_WC_SUCCESS;
         wc.wc_flags = IBV_WC_GRH;
+        backend_dev->rdma_dev_res->stats.mad_rx++;
         comp_handler(bctx->up_ctx, &wc);
     }
 
diff --git a/hw/rdma/rdma_hmp.c b/hw/rdma/rdma_hmp.c
new file mode 100644
index 0000000000..c5814473c5
--- /dev/null
+++ b/hw/rdma/rdma_hmp.c
@@ -0,0 +1,30 @@
+/*
+ * RDMA device: Human Monitor interface
+ *
+ * Copyright (C) 2018 Oracle
+ * Copyright (C) 2018 Red Hat Inc
+ *
+ * Authors:
+ *     Yuval Shaia <yuval.shaia@oracle.com>
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
+ * See the COPYING file in the top-level directory.
+ *
+ */
+
+#include "qemu/osdep.h"
+#include "hw/rdma/rdma_hmp.h"
+#include "qemu/module.h"
+
+static const TypeInfo rdma_hmp_info = {
+    .name = TYPE_RDMA_STATS_PROVIDER,
+    .parent = TYPE_INTERFACE,
+    .class_size = sizeof(RdmaStatsProviderClass),
+};
+
+static void rdma_hmp_register_types(void)
+{
+    type_register_static(&rdma_hmp_info);
+}
+
+type_init(rdma_hmp_register_types)
diff --git a/hw/rdma/rdma_rm.c b/hw/rdma/rdma_rm.c
index 14580ca379..e019de1a14 100644
--- a/hw/rdma/rdma_rm.c
+++ b/hw/rdma/rdma_rm.c
@@ -16,6 +16,7 @@
 #include "qemu/osdep.h"
 #include "qapi/error.h"
 #include "cpu.h"
+#include "monitor/monitor.h"
 
 #include "trace.h"
 #include "rdma_utils.h"
@@ -26,6 +27,58 @@
 #define PG_DIR_SZ { TARGET_PAGE_SIZE / sizeof(__u64) }
 #define PG_TBL_SZ { TARGET_PAGE_SIZE / sizeof(__u64) }
 
+void rdma_dump_device_counters(Monitor *mon, RdmaDeviceResources *dev_res)
+{
+    monitor_printf(mon, "\ttx               : %" PRId64 "\n",
+                   dev_res->stats.tx);
+    monitor_printf(mon, "\ttx_len           : %" PRId64 "\n",
+                   dev_res->stats.tx_len);
+    monitor_printf(mon, "\ttx_err           : %" PRId64 "\n",
+                   dev_res->stats.tx_err);
+    monitor_printf(mon, "\trx_bufs          : %" PRId64 "\n",
+                   dev_res->stats.rx_bufs);
+    monitor_printf(mon, "\trx_bufs_len      : %" PRId64 "\n",
+                   dev_res->stats.rx_bufs_len);
+    monitor_printf(mon, "\trx_bufs_err      : %" PRId64 "\n",
+                   dev_res->stats.rx_bufs_err);
+    monitor_printf(mon, "\tcomps            : %" PRId64 "\n",
+                   dev_res->stats.completions);
+    monitor_printf(mon, "\tmissing_comps    : %" PRId32 "\n",
+                   dev_res->stats.missing_cqe);
+    monitor_printf(mon, "\tpoll_cq (bk)     : %" PRId64 "\n",
+                   dev_res->stats.poll_cq_from_bk);
+    monitor_printf(mon, "\tpoll_cq_ppoll_to : %" PRId64 "\n",
+                   dev_res->stats.poll_cq_ppoll_to);
+    monitor_printf(mon, "\tpoll_cq (fe)     : %" PRId64 "\n",
+                   dev_res->stats.poll_cq_from_guest);
+    monitor_printf(mon, "\tpoll_cq_empty    : %" PRId64 "\n",
+                   dev_res->stats.poll_cq_from_guest_empty);
+    monitor_printf(mon, "\tmad_tx           : %" PRId64 "\n",
+                   dev_res->stats.mad_tx);
+    monitor_printf(mon, "\tmad_tx_err       : %" PRId64 "\n",
+                   dev_res->stats.mad_tx_err);
+    monitor_printf(mon, "\tmad_rx           : %" PRId64 "\n",
+                   dev_res->stats.mad_rx);
+    monitor_printf(mon, "\tmad_rx_err       : %" PRId64 "\n",
+                   dev_res->stats.mad_rx_err);
+    monitor_printf(mon, "\tmad_rx_bufs      : %" PRId64 "\n",
+                   dev_res->stats.mad_rx_bufs);
+    monitor_printf(mon, "\tmad_rx_bufs_err  : %" PRId64 "\n",
+                   dev_res->stats.mad_rx_bufs_err);
+    monitor_printf(mon, "\tPDs              : %" PRId32 "\n",
+                   dev_res->pd_tbl.used);
+    monitor_printf(mon, "\tMRs              : %" PRId32 "\n",
+                   dev_res->mr_tbl.used);
+    monitor_printf(mon, "\tUCs              : %" PRId32 "\n",
+                   dev_res->uc_tbl.used);
+    monitor_printf(mon, "\tQPs              : %" PRId32 "\n",
+                   dev_res->qp_tbl.used);
+    monitor_printf(mon, "\tCQs              : %" PRId32 "\n",
+                   dev_res->cq_tbl.used);
+    monitor_printf(mon, "\tCEQ_CTXs         : %" PRId32 "\n",
+                   dev_res->cqe_ctx_tbl.used);
+}
+
 static inline void res_tbl_init(const char *name, RdmaRmResTbl *tbl,
                                 uint32_t tbl_sz, uint32_t res_sz)
 {
@@ -37,6 +90,7 @@ static inline void res_tbl_init(const char *name, RdmaRmResTbl *tbl,
     tbl->bitmap = bitmap_new(tbl_sz);
     tbl->tbl_sz = tbl_sz;
     tbl->res_sz = res_sz;
+    tbl->used = 0;
     qemu_mutex_init(&tbl->lock);
 }
 
@@ -76,6 +130,8 @@ static inline void *rdma_res_tbl_alloc(RdmaRmResTbl *tbl, uint32_t *handle)
 
     set_bit(*handle, tbl->bitmap);
 
+    tbl->used++;
+
     qemu_mutex_unlock(&tbl->lock);
 
     memset(tbl->tbl + *handle * tbl->res_sz, 0, tbl->res_sz);
@@ -93,6 +149,7 @@ static inline void rdma_res_tbl_dealloc(RdmaRmResTbl *tbl, uint32_t handle)
 
     if (handle < tbl->tbl_sz) {
         clear_bit(handle, tbl->bitmap);
+        tbl->used--;
     }
 
     qemu_mutex_unlock(&tbl->lock);
@@ -620,6 +677,9 @@ int rdma_rm_init(RdmaDeviceResources *dev_res, struct ibv_device_attr *dev_attr,
 
     qemu_mutex_init(&dev_res->lock);
 
+    memset(&dev_res->stats, 0, sizeof(dev_res->stats));
+    atomic_set(&dev_res->stats.missing_cqe, 0);
+
     return 0;
 }
 
diff --git a/hw/rdma/rdma_rm.h b/hw/rdma/rdma_rm.h
index 9ec87d2667..a527d306f6 100644
--- a/hw/rdma/rdma_rm.h
+++ b/hw/rdma/rdma_rm.h
@@ -20,6 +20,7 @@
 #include "rdma_backend_defs.h"
 #include "rdma_rm_defs.h"
 
+void rdma_dump_device_counters(Monitor *mon, RdmaDeviceResources *dev_res);
 int rdma_rm_init(RdmaDeviceResources *dev_res, struct ibv_device_attr *dev_attr,
                  Error **errp);
 void rdma_rm_fini(RdmaDeviceResources *dev_res, RdmaBackendDev *backend_dev,
diff --git a/hw/rdma/rdma_rm_defs.h b/hw/rdma/rdma_rm_defs.h
index f0ee1f3072..4b8d704cfe 100644
--- a/hw/rdma/rdma_rm_defs.h
+++ b/hw/rdma/rdma_rm_defs.h
@@ -34,7 +34,9 @@
 #define MAX_QP_INIT_RD_ATOM   16
 #define MAX_AH                64
 
-#define MAX_RM_TBL_NAME 16
+#define MAX_RM_TBL_NAME             16
+#define MAX_CONSEQ_EMPTY_POLL_CQ    4096 /* considered as error above this */
+
 typedef struct RdmaRmResTbl {
     char name[MAX_RM_TBL_NAME];
     QemuMutex lock;
@@ -42,6 +44,7 @@ typedef struct RdmaRmResTbl {
     size_t tbl_sz;
     size_t res_sz;
     void *tbl;
+    uint32_t used; /* number of used entries in the table */
 } RdmaRmResTbl;
 
 typedef struct RdmaRmPD {
@@ -96,6 +99,27 @@ typedef struct RdmaRmPort {
     enum ibv_port_state state;
 } RdmaRmPort;
 
+typedef struct RdmaRmStats {
+    uint64_t tx;
+    uint64_t tx_len;
+    uint64_t tx_err;
+    uint64_t rx_bufs;
+    uint64_t rx_bufs_len;
+    uint64_t rx_bufs_err;
+    uint64_t completions;
+    uint64_t mad_tx;
+    uint64_t mad_tx_err;
+    uint64_t mad_rx;
+    uint64_t mad_rx_err;
+    uint64_t mad_rx_bufs;
+    uint64_t mad_rx_bufs_err;
+    uint64_t poll_cq_from_bk;
+    uint64_t poll_cq_from_guest;
+    uint64_t poll_cq_from_guest_empty;
+    uint64_t poll_cq_ppoll_to;
+    uint32_t missing_cqe;
+} RdmaRmStats;
+
 typedef struct RdmaDeviceResources {
     RdmaRmPort port;
     RdmaRmResTbl pd_tbl;
@@ -106,6 +130,7 @@ typedef struct RdmaDeviceResources {
     RdmaRmResTbl cqe_ctx_tbl;
     GHashTable *qp_hash; /* Keeps mapping between real and emulated */
     QemuMutex lock;
+    RdmaRmStats stats;
 } RdmaDeviceResources;
 
 #endif
diff --git a/hw/rdma/vmw/pvrdma.h b/hw/rdma/vmw/pvrdma.h
index 0879224957..167706ec2c 100644
--- a/hw/rdma/vmw/pvrdma.h
+++ b/hw/rdma/vmw/pvrdma.h
@@ -70,6 +70,10 @@ typedef struct DSRInfo {
     PvrdmaRing cq;
 } DSRInfo;
 
+typedef struct PVRDMADevStats {
+    uint64_t commands;
+} PVRDMADevStats;
+
 typedef struct PVRDMADev {
     PCIDevice parent_obj;
     MemoryRegion msix;
@@ -89,6 +93,7 @@ typedef struct PVRDMADev {
     CharBackend mad_chr;
     VMXNET3State *func0;
     Notifier shutdown_notifier;
+    PVRDMADevStats stats;
 } PVRDMADev;
 #define PVRDMA_DEV(dev) OBJECT_CHECK(PVRDMADev, (dev), PVRDMA_HW_NAME)
 
diff --git a/hw/rdma/vmw/pvrdma_main.c b/hw/rdma/vmw/pvrdma_main.c
index b6061f4b6e..659331ac93 100644
--- a/hw/rdma/vmw/pvrdma_main.c
+++ b/hw/rdma/vmw/pvrdma_main.c
@@ -25,6 +25,8 @@
 #include "cpu.h"
 #include "trace.h"
 #include "sysemu/sysemu.h"
+#include "monitor/monitor.h"
+#include "hw/rdma/rdma_hmp.h"
 
 #include "../rdma_rm.h"
 #include "../rdma_backend.h"
@@ -55,6 +57,18 @@ static Property pvrdma_dev_properties[] = {
     DEFINE_PROP_END_OF_LIST(),
 };
 
+static void pvrdma_print_statistics(Monitor *mon, RdmaStatsProvider *obj)
+{
+    PVRDMADev *dev = PVRDMA_DEV(obj);
+    PCIDevice *pdev = PCI_DEVICE(dev);
+
+    monitor_printf(mon, "%s, %x.%x\n", pdev->name, PCI_SLOT(pdev->devfn),
+                   PCI_FUNC(pdev->devfn));
+    monitor_printf(mon, "\tcommands         : %" PRId64 "\n",
+                   dev->stats.commands);
+    rdma_dump_device_counters(mon, &dev->rdma_dev_res);
+}
+
 static void free_dev_ring(PCIDevice *pci_dev, PvrdmaRing *ring,
                           void *ring_state)
 {
@@ -394,6 +408,7 @@ static void pvrdma_regs_write(void *opaque, hwaddr addr, uint64_t val,
         if (val == 0) {
             trace_pvrdma_regs_write(addr, val, "REQUEST", "");
             pvrdma_exec_cmd(dev);
+            dev->stats.commands++;
         }
         break;
     default:
@@ -612,6 +627,8 @@ static void pvrdma_realize(PCIDevice *pdev, Error **errp)
         goto out;
     }
 
+    memset(&dev->stats, 0, sizeof(dev->stats));
+
     dev->shutdown_notifier.notify = pvrdma_shutdown_notifier;
     qemu_register_shutdown_notifier(&dev->shutdown_notifier);
 
@@ -631,6 +648,7 @@ static void pvrdma_class_init(ObjectClass *klass, void *data)
 {
     DeviceClass *dc = DEVICE_CLASS(klass);
     PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
+    RdmaStatsProviderClass *ir = RDMA_STATS_PROVIDER_CLASS(klass);
 
     k->realize = pvrdma_realize;
     k->exit = pvrdma_exit;
@@ -642,6 +660,8 @@ static void pvrdma_class_init(ObjectClass *klass, void *data)
     dc->desc = "RDMA Device";
     dc->props = pvrdma_dev_properties;
     set_bit(DEVICE_CATEGORY_NETWORK, dc->categories);
+
+    ir->print_statistics = pvrdma_print_statistics;
 }
 
 static const TypeInfo pvrdma_info = {
@@ -651,6 +671,7 @@ static const TypeInfo pvrdma_info = {
     .class_init = pvrdma_class_init,
     .interfaces = (InterfaceInfo[]) {
         { INTERFACE_CONVENTIONAL_PCI_DEVICE },
+        { TYPE_RDMA_STATS_PROVIDER },
         { }
     }
 };
diff --git a/include/hw/rdma/rdma_hmp.h b/include/hw/rdma/rdma_hmp.h
new file mode 100644
index 0000000000..dd23f2bc84
--- /dev/null
+++ b/include/hw/rdma/rdma_hmp.h
@@ -0,0 +1,40 @@
+/*
+ * RDMA device: Human Monitor interface
+ *
+ * Copyright (C) 2019 Oracle
+ * Copyright (C) 2019 Red Hat Inc
+ *
+ * Authors:
+ *     Yuval Shaia <yuval.shaia@oracle.com>
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
+ * See the COPYING file in the top-level directory.
+ *
+ */
+
+#ifndef RDMA_HMP_H
+#define RDMA_HMP_H
+
+#include "qom/object.h"
+
+#define TYPE_RDMA_STATS_PROVIDER "rdma"
+
+#define RDMA_STATS_PROVIDER_CLASS(klass) \
+    OBJECT_CLASS_CHECK(RdmaStatsProviderClass, (klass), \
+                       TYPE_RDMA_STATS_PROVIDER)
+#define RDMA_STATS_PROVIDER_GET_CLASS(obj) \
+    OBJECT_GET_CLASS(RdmaStatsProviderClass, (obj), \
+                     TYPE_RDMA_STATS_PROVIDER)
+#define RDMA_STATS_PROVIDER(obj) \
+    INTERFACE_CHECK(RdmaStatsProvider, (obj), \
+                    TYPE_RDMA_STATS_PROVIDER)
+
+typedef struct RdmaStatsProvider RdmaStatsProvider;
+
+typedef struct RdmaStatsProviderClass {
+    InterfaceClass parent;
+
+    void (*print_statistics)(Monitor *mon, RdmaStatsProvider *obj);
+} RdmaStatsProviderClass;
+
+#endif
-- 
2.17.2

Re: [Qemu-devel] [PATCH v4 4/9] {hmp, hw/pvrdma}: Expose device internals via monitor interface

[Yuval Shaia]

On Sun, Mar 03, 2019 at 10:33:40PM +0200, Yuval Shaia wrote:
> Allow interrogating device internals through HMP interface.
> The exposed indicators can be used for troubleshooting by developers or
> sysadmin.
> There is no need to expose these attributes to a management system (e.x.
> libvirt) because (1) most of them are not "device-management' related
> info and (2) there is no guarantee the interface is stable.
> 
> Signed-off-by: Yuval Shaia <yuval.shaia@oracle.com>
> ---
>  hmp-commands-info.hx       | 14 ++++++++
>  hmp.c                      | 27 +++++++++++++++
>  hmp.h                      |  1 +
>  hw/rdma/Makefile.objs      |  2 +-
>  hw/rdma/rdma_backend.c     | 70 +++++++++++++++++++++++++++++---------
>  hw/rdma/rdma_hmp.c         | 30 ++++++++++++++++
>  hw/rdma/rdma_rm.c          | 60 ++++++++++++++++++++++++++++++++
>  hw/rdma/rdma_rm.h          |  1 +
>  hw/rdma/rdma_rm_defs.h     | 27 ++++++++++++++-
>  hw/rdma/vmw/pvrdma.h       |  5 +++
>  hw/rdma/vmw/pvrdma_main.c  | 21 ++++++++++++
>  include/hw/rdma/rdma_hmp.h | 40 ++++++++++++++++++++++
>  12 files changed, 279 insertions(+), 19 deletions(-)
>  create mode 100644 hw/rdma/rdma_hmp.c
>  create mode 100644 include/hw/rdma/rdma_hmp.h
> 
> diff --git a/hmp-commands-info.hx b/hmp-commands-info.hx
> index cbee8b944d..c59444c461 100644
> --- a/hmp-commands-info.hx
> +++ b/hmp-commands-info.hx
> @@ -202,6 +202,20 @@ STEXI
>  @item info pic
>  @findex info pic
>  Show PIC state.
> +ETEXI
> +
> +    {
> +        .name       = "rdma",
> +        .args_type  = "",
> +        .params     = "",
> +        .help       = "show RDMA state",
> +        .cmd        = hmp_info_rdma,
> +    },
> +
> +STEXI
> +@item info rdma
> +@findex info rdma
> +Show RDMA state.
>  ETEXI
>  
>      {
> diff --git a/hmp.c b/hmp.c
> index 1e006eeb49..68511b2441 100644
> --- a/hmp.c
> +++ b/hmp.c
> @@ -51,6 +51,7 @@
>  #include "qemu/error-report.h"
>  #include "exec/ramlist.h"
>  #include "hw/intc/intc.h"
> +#include "hw/rdma/rdma_hmp.h"
>  #include "migration/snapshot.h"
>  #include "migration/misc.h"
>  
> @@ -968,6 +969,32 @@ void hmp_info_pic(Monitor *mon, const QDict *qdict)
>                                     hmp_info_pic_foreach, mon);
>  }
>  
> +static int hmp_info_rdma_foreach(Object *obj, void *opaque)
> +{
> +    RdmaStatsProvider *rdma;
> +    RdmaStatsProviderClass *k;
> +    Monitor *mon = opaque;
> +
> +    if (object_dynamic_cast(obj, TYPE_RDMA_STATS_PROVIDER)) {
> +        rdma = RDMA_STATS_PROVIDER(obj);
> +        k = RDMA_STATS_PROVIDER_GET_CLASS(obj);
> +        if (k->print_statistics) {
> +            k->print_statistics(mon, rdma);
> +        } else {
> +            monitor_printf(mon, "RDMA statistics not available for %s.\n",
> +                           object_get_typename(obj));
> +        }
> +    }
> +
> +    return 0;
> +}
> +
> +void hmp_info_rdma(Monitor *mon, const QDict *qdict)
> +{
> +    object_child_foreach_recursive(object_get_root(),
> +                                   hmp_info_rdma_foreach, mon);
> +}
> +

Hi Markus and David,
Is this implementation is acceptable by you?

Thanks,
Yuval


>  void hmp_info_pci(Monitor *mon, const QDict *qdict)
>  {
>      PciInfoList *info_list, *info;
> diff --git a/hmp.h b/hmp.h
> index 5f1addcca2..666949afc3 100644
> --- a/hmp.h
> +++ b/hmp.h
> @@ -36,6 +36,7 @@ void hmp_info_spice(Monitor *mon, const QDict *qdict);
>  void hmp_info_balloon(Monitor *mon, const QDict *qdict);
>  void hmp_info_irq(Monitor *mon, const QDict *qdict);
>  void hmp_info_pic(Monitor *mon, const QDict *qdict);
> +void hmp_info_rdma(Monitor *mon, const QDict *qdict);
>  void hmp_info_pci(Monitor *mon, const QDict *qdict);
>  void hmp_info_block_jobs(Monitor *mon, const QDict *qdict);
>  void hmp_info_tpm(Monitor *mon, const QDict *qdict);
> diff --git a/hw/rdma/Makefile.objs b/hw/rdma/Makefile.objs
> index bd36cbf51c..dd59faff51 100644
> --- a/hw/rdma/Makefile.objs
> +++ b/hw/rdma/Makefile.objs
> @@ -1,5 +1,5 @@
>  ifeq ($(CONFIG_PVRDMA),y)
> -obj-$(CONFIG_PCI) += rdma_utils.o rdma_backend.o rdma_rm.o
> +obj-$(CONFIG_PCI) += rdma_utils.o rdma_backend.o rdma_rm.o rdma_hmp.o
>  obj-$(CONFIG_PCI) += vmw/pvrdma_dev_ring.o vmw/pvrdma_cmd.o \
>                       vmw/pvrdma_qp_ops.o vmw/pvrdma_main.o
>  endif
> diff --git a/hw/rdma/rdma_backend.c b/hw/rdma/rdma_backend.c
> index 9679b842d1..bc2fefcf93 100644
> --- a/hw/rdma/rdma_backend.c
> +++ b/hw/rdma/rdma_backend.c
> @@ -64,9 +64,9 @@ static inline void complete_work(enum ibv_wc_status status, uint32_t vendor_err,
>      comp_handler(ctx, &wc);
>  }
>  
> -static void rdma_poll_cq(RdmaDeviceResources *rdma_dev_res, struct ibv_cq *ibcq)
> +static int rdma_poll_cq(RdmaDeviceResources *rdma_dev_res, struct ibv_cq *ibcq)
>  {
> -    int i, ne;
> +    int i, ne, total_ne = 0;
>      BackendCtx *bctx;
>      struct ibv_wc wc[2];
>  
> @@ -89,12 +89,18 @@ static void rdma_poll_cq(RdmaDeviceResources *rdma_dev_res, struct ibv_cq *ibcq)
>              rdma_rm_dealloc_cqe_ctx(rdma_dev_res, wc[i].wr_id);
>              g_free(bctx);
>          }
> +        total_ne += ne;
>      } while (ne > 0);
> +    atomic_sub(&rdma_dev_res->stats.missing_cqe, total_ne);
>      qemu_mutex_unlock(&rdma_dev_res->lock);
>  
>      if (ne < 0) {
>          rdma_error_report("ibv_poll_cq fail, rc=%d, errno=%d", ne, errno);
>      }
> +
> +    rdma_dev_res->stats.completions += total_ne;
> +
> +    return total_ne;
>  }
>  
>  static void *comp_handler_thread(void *arg)
> @@ -122,6 +128,9 @@ static void *comp_handler_thread(void *arg)
>      while (backend_dev->comp_thread.run) {
>          do {
>              rc = qemu_poll_ns(pfds, 1, THR_POLL_TO * (int64_t)SCALE_MS);
> +            if (!rc) {
> +                backend_dev->rdma_dev_res->stats.poll_cq_ppoll_to++;
> +            }
>          } while (!rc && backend_dev->comp_thread.run);
>  
>          if (backend_dev->comp_thread.run) {
> @@ -138,6 +147,7 @@ static void *comp_handler_thread(void *arg)
>                                    errno);
>              }
>  
> +            backend_dev->rdma_dev_res->stats.poll_cq_from_bk++;
>              rdma_poll_cq(backend_dev->rdma_dev_res, ev_cq);
>  
>              ibv_ack_cq_events(ev_cq, 1);
> @@ -271,7 +281,13 @@ int rdma_backend_query_port(RdmaBackendDev *backend_dev,
>  
>  void rdma_backend_poll_cq(RdmaDeviceResources *rdma_dev_res, RdmaBackendCQ *cq)
>  {
> -    rdma_poll_cq(rdma_dev_res, cq->ibcq);
> +    int polled;
> +
> +    rdma_dev_res->stats.poll_cq_from_guest++;
> +    polled = rdma_poll_cq(rdma_dev_res, cq->ibcq);
> +    if (!polled) {
> +        rdma_dev_res->stats.poll_cq_from_guest_empty++;
> +    }
>  }
>  
>  static GHashTable *ah_hash;
> @@ -333,7 +349,7 @@ static void ah_cache_init(void)
>  
>  static int build_host_sge_array(RdmaDeviceResources *rdma_dev_res,
>                                  struct ibv_sge *dsge, struct ibv_sge *ssge,
> -                                uint8_t num_sge)
> +                                uint8_t num_sge, uint64_t *total_length)
>  {
>      RdmaRmMR *mr;
>      int ssge_idx;
> @@ -349,6 +365,8 @@ static int build_host_sge_array(RdmaDeviceResources *rdma_dev_res,
>          dsge->length = ssge[ssge_idx].length;
>          dsge->lkey = rdma_backend_mr_lkey(&mr->backend_mr);
>  
> +        *total_length += dsge->length;
> +
>          dsge++;
>      }
>  
> @@ -445,8 +463,10 @@ void rdma_backend_post_send(RdmaBackendDev *backend_dev,
>              rc = mad_send(backend_dev, sgid_idx, sgid, sge, num_sge);
>              if (rc) {
>                  complete_work(IBV_WC_GENERAL_ERR, VENDOR_ERR_MAD_SEND, ctx);
> +                backend_dev->rdma_dev_res->stats.mad_tx_err++;
>              } else {
>                  complete_work(IBV_WC_SUCCESS, 0, ctx);
> +                backend_dev->rdma_dev_res->stats.mad_tx++;
>              }
>          }
>          return;
> @@ -458,20 +478,21 @@ void rdma_backend_post_send(RdmaBackendDev *backend_dev,
>      rc = rdma_rm_alloc_cqe_ctx(backend_dev->rdma_dev_res, &bctx_id, bctx);
>      if (unlikely(rc)) {
>          complete_work(IBV_WC_GENERAL_ERR, VENDOR_ERR_NOMEM, ctx);
> -        goto out_free_bctx;
> +        goto err_free_bctx;
>      }
>  
> -    rc = build_host_sge_array(backend_dev->rdma_dev_res, new_sge, sge, num_sge);
> +    rc = build_host_sge_array(backend_dev->rdma_dev_res, new_sge, sge, num_sge,
> +                              &backend_dev->rdma_dev_res->stats.tx_len);
>      if (rc) {
>          complete_work(IBV_WC_GENERAL_ERR, rc, ctx);
> -        goto out_dealloc_cqe_ctx;
> +        goto err_dealloc_cqe_ctx;
>      }
>  
>      if (qp_type == IBV_QPT_UD) {
>          wr.wr.ud.ah = create_ah(backend_dev, qp->ibpd, sgid_idx, dgid);
>          if (!wr.wr.ud.ah) {
>              complete_work(IBV_WC_GENERAL_ERR, VENDOR_ERR_FAIL_BACKEND, ctx);
> -            goto out_dealloc_cqe_ctx;
> +            goto err_dealloc_cqe_ctx;
>          }
>          wr.wr.ud.remote_qpn = dqpn;
>          wr.wr.ud.remote_qkey = dqkey;
> @@ -488,15 +509,19 @@ void rdma_backend_post_send(RdmaBackendDev *backend_dev,
>          rdma_error_report("ibv_post_send fail, qpn=0x%x, rc=%d, errno=%d",
>                            qp->ibqp->qp_num, rc, errno);
>          complete_work(IBV_WC_GENERAL_ERR, VENDOR_ERR_FAIL_BACKEND, ctx);
> -        goto out_dealloc_cqe_ctx;
> +        goto err_dealloc_cqe_ctx;
>      }
>  
> +    atomic_inc(&backend_dev->rdma_dev_res->stats.missing_cqe);
> +    backend_dev->rdma_dev_res->stats.tx++;
> +
>      return;
>  
> -out_dealloc_cqe_ctx:
> +err_dealloc_cqe_ctx:
> +    backend_dev->rdma_dev_res->stats.tx_err++;
>      rdma_rm_dealloc_cqe_ctx(backend_dev->rdma_dev_res, bctx_id);
>  
> -out_free_bctx:
> +err_free_bctx:
>      g_free(bctx);
>  }
>  
> @@ -554,6 +579,9 @@ void rdma_backend_post_recv(RdmaBackendDev *backend_dev,
>              rc = save_mad_recv_buffer(backend_dev, sge, num_sge, ctx);
>              if (rc) {
>                  complete_work(IBV_WC_GENERAL_ERR, rc, ctx);
> +                rdma_dev_res->stats.mad_rx_bufs_err++;
> +            } else {
> +                rdma_dev_res->stats.mad_rx_bufs++;
>              }
>          }
>          return;
> @@ -565,13 +593,14 @@ void rdma_backend_post_recv(RdmaBackendDev *backend_dev,
>      rc = rdma_rm_alloc_cqe_ctx(rdma_dev_res, &bctx_id, bctx);
>      if (unlikely(rc)) {
>          complete_work(IBV_WC_GENERAL_ERR, VENDOR_ERR_NOMEM, ctx);
> -        goto out_free_bctx;
> +        goto err_free_bctx;
>      }
>  
> -    rc = build_host_sge_array(rdma_dev_res, new_sge, sge, num_sge);
> +    rc = build_host_sge_array(rdma_dev_res, new_sge, sge, num_sge,
> +                              &backend_dev->rdma_dev_res->stats.rx_bufs_len);
>      if (rc) {
>          complete_work(IBV_WC_GENERAL_ERR, rc, ctx);
> -        goto out_dealloc_cqe_ctx;
> +        goto err_dealloc_cqe_ctx;
>      }
>  
>      wr.num_sge = num_sge;
> @@ -582,15 +611,19 @@ void rdma_backend_post_recv(RdmaBackendDev *backend_dev,
>          rdma_error_report("ibv_post_recv fail, qpn=0x%x, rc=%d, errno=%d",
>                            qp->ibqp->qp_num, rc, errno);
>          complete_work(IBV_WC_GENERAL_ERR, VENDOR_ERR_FAIL_BACKEND, ctx);
> -        goto out_dealloc_cqe_ctx;
> +        goto err_dealloc_cqe_ctx;
>      }
>  
> +    atomic_inc(&backend_dev->rdma_dev_res->stats.missing_cqe);
> +    rdma_dev_res->stats.rx_bufs++;
> +
>      return;
>  
> -out_dealloc_cqe_ctx:
> +err_dealloc_cqe_ctx:
> +    backend_dev->rdma_dev_res->stats.rx_bufs_err++;
>      rdma_rm_dealloc_cqe_ctx(rdma_dev_res, bctx_id);
>  
> -out_free_bctx:
> +err_free_bctx:
>      g_free(bctx);
>  }
>  
> @@ -929,12 +962,14 @@ static void process_incoming_mad_req(RdmaBackendDev *backend_dev,
>      bctx = rdma_rm_get_cqe_ctx(backend_dev->rdma_dev_res, cqe_ctx_id);
>      if (unlikely(!bctx)) {
>          rdma_error_report("No matching ctx for req %ld", cqe_ctx_id);
> +        backend_dev->rdma_dev_res->stats.mad_rx_err++;
>          return;
>      }
>  
>      mad = rdma_pci_dma_map(backend_dev->dev, bctx->sge.addr,
>                             bctx->sge.length);
>      if (!mad || bctx->sge.length < msg->umad_len + MAD_HDR_SIZE) {
> +        backend_dev->rdma_dev_res->stats.mad_rx_err++;
>          complete_work(IBV_WC_GENERAL_ERR, VENDOR_ERR_INV_MAD_BUFF,
>                        bctx->up_ctx);
>      } else {
> @@ -949,6 +984,7 @@ static void process_incoming_mad_req(RdmaBackendDev *backend_dev,
>          wc.byte_len = msg->umad_len;
>          wc.status = IBV_WC_SUCCESS;
>          wc.wc_flags = IBV_WC_GRH;
> +        backend_dev->rdma_dev_res->stats.mad_rx++;
>          comp_handler(bctx->up_ctx, &wc);
>      }
>  
> diff --git a/hw/rdma/rdma_hmp.c b/hw/rdma/rdma_hmp.c
> new file mode 100644
> index 0000000000..c5814473c5
> --- /dev/null
> +++ b/hw/rdma/rdma_hmp.c
> @@ -0,0 +1,30 @@
> +/*
> + * RDMA device: Human Monitor interface
> + *
> + * Copyright (C) 2018 Oracle
> + * Copyright (C) 2018 Red Hat Inc
> + *
> + * Authors:
> + *     Yuval Shaia <yuval.shaia@oracle.com>
> + *
> + * This work is licensed under the terms of the GNU GPL, version 2 or later.
> + * See the COPYING file in the top-level directory.
> + *
> + */
> +
> +#include "qemu/osdep.h"
> +#include "hw/rdma/rdma_hmp.h"
> +#include "qemu/module.h"
> +
> +static const TypeInfo rdma_hmp_info = {
> +    .name = TYPE_RDMA_STATS_PROVIDER,
> +    .parent = TYPE_INTERFACE,
> +    .class_size = sizeof(RdmaStatsProviderClass),
> +};
> +
> +static void rdma_hmp_register_types(void)
> +{
> +    type_register_static(&rdma_hmp_info);
> +}
> +
> +type_init(rdma_hmp_register_types)
> diff --git a/hw/rdma/rdma_rm.c b/hw/rdma/rdma_rm.c
> index 14580ca379..e019de1a14 100644
> --- a/hw/rdma/rdma_rm.c
> +++ b/hw/rdma/rdma_rm.c
> @@ -16,6 +16,7 @@
>  #include "qemu/osdep.h"
>  #include "qapi/error.h"
>  #include "cpu.h"
> +#include "monitor/monitor.h"
>  
>  #include "trace.h"
>  #include "rdma_utils.h"
> @@ -26,6 +27,58 @@
>  #define PG_DIR_SZ { TARGET_PAGE_SIZE / sizeof(__u64) }
>  #define PG_TBL_SZ { TARGET_PAGE_SIZE / sizeof(__u64) }
>  
> +void rdma_dump_device_counters(Monitor *mon, RdmaDeviceResources *dev_res)
> +{
> +    monitor_printf(mon, "\ttx               : %" PRId64 "\n",
> +                   dev_res->stats.tx);
> +    monitor_printf(mon, "\ttx_len           : %" PRId64 "\n",
> +                   dev_res->stats.tx_len);
> +    monitor_printf(mon, "\ttx_err           : %" PRId64 "\n",
> +                   dev_res->stats.tx_err);
> +    monitor_printf(mon, "\trx_bufs          : %" PRId64 "\n",
> +                   dev_res->stats.rx_bufs);
> +    monitor_printf(mon, "\trx_bufs_len      : %" PRId64 "\n",
> +                   dev_res->stats.rx_bufs_len);
> +    monitor_printf(mon, "\trx_bufs_err      : %" PRId64 "\n",
> +                   dev_res->stats.rx_bufs_err);
> +    monitor_printf(mon, "\tcomps            : %" PRId64 "\n",
> +                   dev_res->stats.completions);
> +    monitor_printf(mon, "\tmissing_comps    : %" PRId32 "\n",
> +                   dev_res->stats.missing_cqe);
> +    monitor_printf(mon, "\tpoll_cq (bk)     : %" PRId64 "\n",
> +                   dev_res->stats.poll_cq_from_bk);
> +    monitor_printf(mon, "\tpoll_cq_ppoll_to : %" PRId64 "\n",
> +                   dev_res->stats.poll_cq_ppoll_to);
> +    monitor_printf(mon, "\tpoll_cq (fe)     : %" PRId64 "\n",
> +                   dev_res->stats.poll_cq_from_guest);
> +    monitor_printf(mon, "\tpoll_cq_empty    : %" PRId64 "\n",
> +                   dev_res->stats.poll_cq_from_guest_empty);
> +    monitor_printf(mon, "\tmad_tx           : %" PRId64 "\n",
> +                   dev_res->stats.mad_tx);
> +    monitor_printf(mon, "\tmad_tx_err       : %" PRId64 "\n",
> +                   dev_res->stats.mad_tx_err);
> +    monitor_printf(mon, "\tmad_rx           : %" PRId64 "\n",
> +                   dev_res->stats.mad_rx);
> +    monitor_printf(mon, "\tmad_rx_err       : %" PRId64 "\n",
> +                   dev_res->stats.mad_rx_err);
> +    monitor_printf(mon, "\tmad_rx_bufs      : %" PRId64 "\n",
> +                   dev_res->stats.mad_rx_bufs);
> +    monitor_printf(mon, "\tmad_rx_bufs_err  : %" PRId64 "\n",
> +                   dev_res->stats.mad_rx_bufs_err);
> +    monitor_printf(mon, "\tPDs              : %" PRId32 "\n",
> +                   dev_res->pd_tbl.used);
> +    monitor_printf(mon, "\tMRs              : %" PRId32 "\n",
> +                   dev_res->mr_tbl.used);
> +    monitor_printf(mon, "\tUCs              : %" PRId32 "\n",
> +                   dev_res->uc_tbl.used);
> +    monitor_printf(mon, "\tQPs              : %" PRId32 "\n",
> +                   dev_res->qp_tbl.used);
> +    monitor_printf(mon, "\tCQs              : %" PRId32 "\n",
> +                   dev_res->cq_tbl.used);
> +    monitor_printf(mon, "\tCEQ_CTXs         : %" PRId32 "\n",
> +                   dev_res->cqe_ctx_tbl.used);
> +}
> +
>  static inline void res_tbl_init(const char *name, RdmaRmResTbl *tbl,
>                                  uint32_t tbl_sz, uint32_t res_sz)
>  {
> @@ -37,6 +90,7 @@ static inline void res_tbl_init(const char *name, RdmaRmResTbl *tbl,
>      tbl->bitmap = bitmap_new(tbl_sz);
>      tbl->tbl_sz = tbl_sz;
>      tbl->res_sz = res_sz;
> +    tbl->used = 0;
>      qemu_mutex_init(&tbl->lock);
>  }
>  
> @@ -76,6 +130,8 @@ static inline void *rdma_res_tbl_alloc(RdmaRmResTbl *tbl, uint32_t *handle)
>  
>      set_bit(*handle, tbl->bitmap);
>  
> +    tbl->used++;
> +
>      qemu_mutex_unlock(&tbl->lock);
>  
>      memset(tbl->tbl + *handle * tbl->res_sz, 0, tbl->res_sz);
> @@ -93,6 +149,7 @@ static inline void rdma_res_tbl_dealloc(RdmaRmResTbl *tbl, uint32_t handle)
>  
>      if (handle < tbl->tbl_sz) {
>          clear_bit(handle, tbl->bitmap);
> +        tbl->used--;
>      }
>  
>      qemu_mutex_unlock(&tbl->lock);
> @@ -620,6 +677,9 @@ int rdma_rm_init(RdmaDeviceResources *dev_res, struct ibv_device_attr *dev_attr,
>  
>      qemu_mutex_init(&dev_res->lock);
>  
> +    memset(&dev_res->stats, 0, sizeof(dev_res->stats));
> +    atomic_set(&dev_res->stats.missing_cqe, 0);
> +
>      return 0;
>  }
>  
> diff --git a/hw/rdma/rdma_rm.h b/hw/rdma/rdma_rm.h
> index 9ec87d2667..a527d306f6 100644
> --- a/hw/rdma/rdma_rm.h
> +++ b/hw/rdma/rdma_rm.h
> @@ -20,6 +20,7 @@
>  #include "rdma_backend_defs.h"
>  #include "rdma_rm_defs.h"
>  
> +void rdma_dump_device_counters(Monitor *mon, RdmaDeviceResources *dev_res);
>  int rdma_rm_init(RdmaDeviceResources *dev_res, struct ibv_device_attr *dev_attr,
>                   Error **errp);
>  void rdma_rm_fini(RdmaDeviceResources *dev_res, RdmaBackendDev *backend_dev,
> diff --git a/hw/rdma/rdma_rm_defs.h b/hw/rdma/rdma_rm_defs.h
> index f0ee1f3072..4b8d704cfe 100644
> --- a/hw/rdma/rdma_rm_defs.h
> +++ b/hw/rdma/rdma_rm_defs.h
> @@ -34,7 +34,9 @@
>  #define MAX_QP_INIT_RD_ATOM   16
>  #define MAX_AH                64
>  
> -#define MAX_RM_TBL_NAME 16
> +#define MAX_RM_TBL_NAME             16
> +#define MAX_CONSEQ_EMPTY_POLL_CQ    4096 /* considered as error above this */
> +
>  typedef struct RdmaRmResTbl {
>      char name[MAX_RM_TBL_NAME];
>      QemuMutex lock;
> @@ -42,6 +44,7 @@ typedef struct RdmaRmResTbl {
>      size_t tbl_sz;
>      size_t res_sz;
>      void *tbl;
> +    uint32_t used; /* number of used entries in the table */
>  } RdmaRmResTbl;
>  
>  typedef struct RdmaRmPD {
> @@ -96,6 +99,27 @@ typedef struct RdmaRmPort {
>      enum ibv_port_state state;
>  } RdmaRmPort;
>  
> +typedef struct RdmaRmStats {
> +    uint64_t tx;
> +    uint64_t tx_len;
> +    uint64_t tx_err;
> +    uint64_t rx_bufs;
> +    uint64_t rx_bufs_len;
> +    uint64_t rx_bufs_err;
> +    uint64_t completions;
> +    uint64_t mad_tx;
> +    uint64_t mad_tx_err;
> +    uint64_t mad_rx;
> +    uint64_t mad_rx_err;
> +    uint64_t mad_rx_bufs;
> +    uint64_t mad_rx_bufs_err;
> +    uint64_t poll_cq_from_bk;
> +    uint64_t poll_cq_from_guest;
> +    uint64_t poll_cq_from_guest_empty;
> +    uint64_t poll_cq_ppoll_to;
> +    uint32_t missing_cqe;
> +} RdmaRmStats;
> +
>  typedef struct RdmaDeviceResources {
>      RdmaRmPort port;
>      RdmaRmResTbl pd_tbl;
> @@ -106,6 +130,7 @@ typedef struct RdmaDeviceResources {
>      RdmaRmResTbl cqe_ctx_tbl;
>      GHashTable *qp_hash; /* Keeps mapping between real and emulated */
>      QemuMutex lock;
> +    RdmaRmStats stats;
>  } RdmaDeviceResources;
>  
>  #endif
> diff --git a/hw/rdma/vmw/pvrdma.h b/hw/rdma/vmw/pvrdma.h
> index 0879224957..167706ec2c 100644
> --- a/hw/rdma/vmw/pvrdma.h
> +++ b/hw/rdma/vmw/pvrdma.h
> @@ -70,6 +70,10 @@ typedef struct DSRInfo {
>      PvrdmaRing cq;
>  } DSRInfo;
>  
> +typedef struct PVRDMADevStats {
> +    uint64_t commands;
> +} PVRDMADevStats;
> +
>  typedef struct PVRDMADev {
>      PCIDevice parent_obj;
>      MemoryRegion msix;
> @@ -89,6 +93,7 @@ typedef struct PVRDMADev {
>      CharBackend mad_chr;
>      VMXNET3State *func0;
>      Notifier shutdown_notifier;
> +    PVRDMADevStats stats;
>  } PVRDMADev;
>  #define PVRDMA_DEV(dev) OBJECT_CHECK(PVRDMADev, (dev), PVRDMA_HW_NAME)
>  
> diff --git a/hw/rdma/vmw/pvrdma_main.c b/hw/rdma/vmw/pvrdma_main.c
> index b6061f4b6e..659331ac93 100644
> --- a/hw/rdma/vmw/pvrdma_main.c
> +++ b/hw/rdma/vmw/pvrdma_main.c
> @@ -25,6 +25,8 @@
>  #include "cpu.h"
>  #include "trace.h"
>  #include "sysemu/sysemu.h"
> +#include "monitor/monitor.h"
> +#include "hw/rdma/rdma_hmp.h"
>  
>  #include "../rdma_rm.h"
>  #include "../rdma_backend.h"
> @@ -55,6 +57,18 @@ static Property pvrdma_dev_properties[] = {
>      DEFINE_PROP_END_OF_LIST(),
>  };
>  
> +static void pvrdma_print_statistics(Monitor *mon, RdmaStatsProvider *obj)
> +{
> +    PVRDMADev *dev = PVRDMA_DEV(obj);
> +    PCIDevice *pdev = PCI_DEVICE(dev);
> +
> +    monitor_printf(mon, "%s, %x.%x\n", pdev->name, PCI_SLOT(pdev->devfn),
> +                   PCI_FUNC(pdev->devfn));
> +    monitor_printf(mon, "\tcommands         : %" PRId64 "\n",
> +                   dev->stats.commands);
> +    rdma_dump_device_counters(mon, &dev->rdma_dev_res);
> +}
> +
>  static void free_dev_ring(PCIDevice *pci_dev, PvrdmaRing *ring,
>                            void *ring_state)
>  {
> @@ -394,6 +408,7 @@ static void pvrdma_regs_write(void *opaque, hwaddr addr, uint64_t val,
>          if (val == 0) {
>              trace_pvrdma_regs_write(addr, val, "REQUEST", "");
>              pvrdma_exec_cmd(dev);
> +            dev->stats.commands++;
>          }
>          break;
>      default:
> @@ -612,6 +627,8 @@ static void pvrdma_realize(PCIDevice *pdev, Error **errp)
>          goto out;
>      }
>  
> +    memset(&dev->stats, 0, sizeof(dev->stats));
> +
>      dev->shutdown_notifier.notify = pvrdma_shutdown_notifier;
>      qemu_register_shutdown_notifier(&dev->shutdown_notifier);
>  
> @@ -631,6 +648,7 @@ static void pvrdma_class_init(ObjectClass *klass, void *data)
>  {
>      DeviceClass *dc = DEVICE_CLASS(klass);
>      PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
> +    RdmaStatsProviderClass *ir = RDMA_STATS_PROVIDER_CLASS(klass);
>  
>      k->realize = pvrdma_realize;
>      k->exit = pvrdma_exit;
> @@ -642,6 +660,8 @@ static void pvrdma_class_init(ObjectClass *klass, void *data)
>      dc->desc = "RDMA Device";
>      dc->props = pvrdma_dev_properties;
>      set_bit(DEVICE_CATEGORY_NETWORK, dc->categories);
> +
> +    ir->print_statistics = pvrdma_print_statistics;
>  }
>  
>  static const TypeInfo pvrdma_info = {
> @@ -651,6 +671,7 @@ static const TypeInfo pvrdma_info = {
>      .class_init = pvrdma_class_init,
>      .interfaces = (InterfaceInfo[]) {
>          { INTERFACE_CONVENTIONAL_PCI_DEVICE },
> +        { TYPE_RDMA_STATS_PROVIDER },
>          { }
>      }
>  };
> diff --git a/include/hw/rdma/rdma_hmp.h b/include/hw/rdma/rdma_hmp.h
> new file mode 100644
> index 0000000000..dd23f2bc84
> --- /dev/null
> +++ b/include/hw/rdma/rdma_hmp.h
> @@ -0,0 +1,40 @@
> +/*
> + * RDMA device: Human Monitor interface
> + *
> + * Copyright (C) 2019 Oracle
> + * Copyright (C) 2019 Red Hat Inc
> + *
> + * Authors:
> + *     Yuval Shaia <yuval.shaia@oracle.com>
> + *
> + * This work is licensed under the terms of the GNU GPL, version 2 or later.
> + * See the COPYING file in the top-level directory.
> + *
> + */
> +
> +#ifndef RDMA_HMP_H
> +#define RDMA_HMP_H
> +
> +#include "qom/object.h"
> +
> +#define TYPE_RDMA_STATS_PROVIDER "rdma"
> +
> +#define RDMA_STATS_PROVIDER_CLASS(klass) \
> +    OBJECT_CLASS_CHECK(RdmaStatsProviderClass, (klass), \
> +                       TYPE_RDMA_STATS_PROVIDER)
> +#define RDMA_STATS_PROVIDER_GET_CLASS(obj) \
> +    OBJECT_GET_CLASS(RdmaStatsProviderClass, (obj), \
> +                     TYPE_RDMA_STATS_PROVIDER)
> +#define RDMA_STATS_PROVIDER(obj) \
> +    INTERFACE_CHECK(RdmaStatsProvider, (obj), \
> +                    TYPE_RDMA_STATS_PROVIDER)
> +
> +typedef struct RdmaStatsProvider RdmaStatsProvider;
> +
> +typedef struct RdmaStatsProviderClass {
> +    InterfaceClass parent;
> +
> +    void (*print_statistics)(Monitor *mon, RdmaStatsProvider *obj);
> +} RdmaStatsProviderClass;
> +
> +#endif
> -- 
> 2.17.2
> 
> 

Re: [Qemu-devel] [PATCH v4 4/9] {hmp, hw/pvrdma}: Expose device internals via monitor interface

[Dr. David Alan Gilbert]

* Yuval Shaia (yuval.shaia@oracle.com) wrote:
> On Sun, Mar 03, 2019 at 10:33:40PM +0200, Yuval Shaia wrote:
> > Allow interrogating device internals through HMP interface.
> > The exposed indicators can be used for troubleshooting by developers or
> > sysadmin.
> > There is no need to expose these attributes to a management system (e.x.
> > libvirt) because (1) most of them are not "device-management' related
> > info and (2) there is no guarantee the interface is stable.
> > 
> > Signed-off-by: Yuval Shaia <yuval.shaia@oracle.com>
> > ---
> >  hmp-commands-info.hx       | 14 ++++++++
> >  hmp.c                      | 27 +++++++++++++++
> >  hmp.h                      |  1 +
> >  hw/rdma/Makefile.objs      |  2 +-
> >  hw/rdma/rdma_backend.c     | 70 +++++++++++++++++++++++++++++---------
> >  hw/rdma/rdma_hmp.c         | 30 ++++++++++++++++
> >  hw/rdma/rdma_rm.c          | 60 ++++++++++++++++++++++++++++++++
> >  hw/rdma/rdma_rm.h          |  1 +
> >  hw/rdma/rdma_rm_defs.h     | 27 ++++++++++++++-
> >  hw/rdma/vmw/pvrdma.h       |  5 +++
> >  hw/rdma/vmw/pvrdma_main.c  | 21 ++++++++++++
> >  include/hw/rdma/rdma_hmp.h | 40 ++++++++++++++++++++++
> >  12 files changed, 279 insertions(+), 19 deletions(-)
> >  create mode 100644 hw/rdma/rdma_hmp.c
> >  create mode 100644 include/hw/rdma/rdma_hmp.h
> > 
> > diff --git a/hmp-commands-info.hx b/hmp-commands-info.hx
> > index cbee8b944d..c59444c461 100644
> > --- a/hmp-commands-info.hx
> > +++ b/hmp-commands-info.hx
> > @@ -202,6 +202,20 @@ STEXI
> >  @item info pic
> >  @findex info pic
> >  Show PIC state.
> > +ETEXI
> > +
> > +    {
> > +        .name       = "rdma",
> > +        .args_type  = "",
> > +        .params     = "",
> > +        .help       = "show RDMA state",
> > +        .cmd        = hmp_info_rdma,
> > +    },
> > +
> > +STEXI
> > +@item info rdma
> > +@findex info rdma
> > +Show RDMA state.
> >  ETEXI
> >  
> >      {
> > diff --git a/hmp.c b/hmp.c
> > index 1e006eeb49..68511b2441 100644
> > --- a/hmp.c
> > +++ b/hmp.c
> > @@ -51,6 +51,7 @@
> >  #include "qemu/error-report.h"
> >  #include "exec/ramlist.h"
> >  #include "hw/intc/intc.h"
> > +#include "hw/rdma/rdma_hmp.h"
> >  #include "migration/snapshot.h"
> >  #include "migration/misc.h"
> >  
> > @@ -968,6 +969,32 @@ void hmp_info_pic(Monitor *mon, const QDict *qdict)
> >                                     hmp_info_pic_foreach, mon);
> >  }
> >  
> > +static int hmp_info_rdma_foreach(Object *obj, void *opaque)
> > +{
> > +    RdmaStatsProvider *rdma;
> > +    RdmaStatsProviderClass *k;
> > +    Monitor *mon = opaque;
> > +
> > +    if (object_dynamic_cast(obj, TYPE_RDMA_STATS_PROVIDER)) {
> > +        rdma = RDMA_STATS_PROVIDER(obj);
> > +        k = RDMA_STATS_PROVIDER_GET_CLASS(obj);
> > +        if (k->print_statistics) {
> > +            k->print_statistics(mon, rdma);
> > +        } else {
> > +            monitor_printf(mon, "RDMA statistics not available for %s.\n",
> > +                           object_get_typename(obj));
> > +        }
> > +    }
> > +
> > +    return 0;
> > +}
> > +
> > +void hmp_info_rdma(Monitor *mon, const QDict *qdict)
> > +{
> > +    object_child_foreach_recursive(object_get_root(),
> > +                                   hmp_info_rdma_foreach, mon);
> > +}
> > +
> 
> Hi Markus and David,
> Is this implementation is acceptable by you?

I think from the HMP side I'm fine, so:

Acked-by: Dr. David Alan Gilbert <dgilbert@redhat.com>

I would say that there's really two parts to this patch, all the places
you change your code to gather the stats, and the separate pieces that
wire it into HMP;  if you had to repost for another reason I'd
split the patch like that.

Dave

> Thanks,
> Yuval
> 
> 
> >  void hmp_info_pci(Monitor *mon, const QDict *qdict)
> >  {
> >      PciInfoList *info_list, *info;
> > diff --git a/hmp.h b/hmp.h
> > index 5f1addcca2..666949afc3 100644
> > --- a/hmp.h
> > +++ b/hmp.h
> > @@ -36,6 +36,7 @@ void hmp_info_spice(Monitor *mon, const QDict *qdict);
> >  void hmp_info_balloon(Monitor *mon, const QDict *qdict);
> >  void hmp_info_irq(Monitor *mon, const QDict *qdict);
> >  void hmp_info_pic(Monitor *mon, const QDict *qdict);
> > +void hmp_info_rdma(Monitor *mon, const QDict *qdict);
> >  void hmp_info_pci(Monitor *mon, const QDict *qdict);
> >  void hmp_info_block_jobs(Monitor *mon, const QDict *qdict);
> >  void hmp_info_tpm(Monitor *mon, const QDict *qdict);
> > diff --git a/hw/rdma/Makefile.objs b/hw/rdma/Makefile.objs
> > index bd36cbf51c..dd59faff51 100644
> > --- a/hw/rdma/Makefile.objs
> > +++ b/hw/rdma/Makefile.objs
> > @@ -1,5 +1,5 @@
> >  ifeq ($(CONFIG_PVRDMA),y)
> > -obj-$(CONFIG_PCI) += rdma_utils.o rdma_backend.o rdma_rm.o
> > +obj-$(CONFIG_PCI) += rdma_utils.o rdma_backend.o rdma_rm.o rdma_hmp.o
> >  obj-$(CONFIG_PCI) += vmw/pvrdma_dev_ring.o vmw/pvrdma_cmd.o \
> >                       vmw/pvrdma_qp_ops.o vmw/pvrdma_main.o
> >  endif
> > diff --git a/hw/rdma/rdma_backend.c b/hw/rdma/rdma_backend.c
> > index 9679b842d1..bc2fefcf93 100644
> > --- a/hw/rdma/rdma_backend.c
> > +++ b/hw/rdma/rdma_backend.c
> > @@ -64,9 +64,9 @@ static inline void complete_work(enum ibv_wc_status status, uint32_t vendor_err,
> >      comp_handler(ctx, &wc);
> >  }
> >  
> > -static void rdma_poll_cq(RdmaDeviceResources *rdma_dev_res, struct ibv_cq *ibcq)
> > +static int rdma_poll_cq(RdmaDeviceResources *rdma_dev_res, struct ibv_cq *ibcq)
> >  {
> > -    int i, ne;
> > +    int i, ne, total_ne = 0;
> >      BackendCtx *bctx;
> >      struct ibv_wc wc[2];
> >  
> > @@ -89,12 +89,18 @@ static void rdma_poll_cq(RdmaDeviceResources *rdma_dev_res, struct ibv_cq *ibcq)
> >              rdma_rm_dealloc_cqe_ctx(rdma_dev_res, wc[i].wr_id);
> >              g_free(bctx);
> >          }
> > +        total_ne += ne;
> >      } while (ne > 0);
> > +    atomic_sub(&rdma_dev_res->stats.missing_cqe, total_ne);
> >      qemu_mutex_unlock(&rdma_dev_res->lock);
> >  
> >      if (ne < 0) {
> >          rdma_error_report("ibv_poll_cq fail, rc=%d, errno=%d", ne, errno);
> >      }
> > +
> > +    rdma_dev_res->stats.completions += total_ne;
> > +
> > +    return total_ne;
> >  }
> >  
> >  static void *comp_handler_thread(void *arg)
> > @@ -122,6 +128,9 @@ static void *comp_handler_thread(void *arg)
> >      while (backend_dev->comp_thread.run) {
> >          do {
> >              rc = qemu_poll_ns(pfds, 1, THR_POLL_TO * (int64_t)SCALE_MS);
> > +            if (!rc) {
> > +                backend_dev->rdma_dev_res->stats.poll_cq_ppoll_to++;
> > +            }
> >          } while (!rc && backend_dev->comp_thread.run);
> >  
> >          if (backend_dev->comp_thread.run) {
> > @@ -138,6 +147,7 @@ static void *comp_handler_thread(void *arg)
> >                                    errno);
> >              }
> >  
> > +            backend_dev->rdma_dev_res->stats.poll_cq_from_bk++;
> >              rdma_poll_cq(backend_dev->rdma_dev_res, ev_cq);
> >  
> >              ibv_ack_cq_events(ev_cq, 1);
> > @@ -271,7 +281,13 @@ int rdma_backend_query_port(RdmaBackendDev *backend_dev,
> >  
> >  void rdma_backend_poll_cq(RdmaDeviceResources *rdma_dev_res, RdmaBackendCQ *cq)
> >  {
> > -    rdma_poll_cq(rdma_dev_res, cq->ibcq);
> > +    int polled;
> > +
> > +    rdma_dev_res->stats.poll_cq_from_guest++;
> > +    polled = rdma_poll_cq(rdma_dev_res, cq->ibcq);
> > +    if (!polled) {
> > +        rdma_dev_res->stats.poll_cq_from_guest_empty++;
> > +    }
> >  }
> >  
> >  static GHashTable *ah_hash;
> > @@ -333,7 +349,7 @@ static void ah_cache_init(void)
> >  
> >  static int build_host_sge_array(RdmaDeviceResources *rdma_dev_res,
> >                                  struct ibv_sge *dsge, struct ibv_sge *ssge,
> > -                                uint8_t num_sge)
> > +                                uint8_t num_sge, uint64_t *total_length)
> >  {
> >      RdmaRmMR *mr;
> >      int ssge_idx;
> > @@ -349,6 +365,8 @@ static int build_host_sge_array(RdmaDeviceResources *rdma_dev_res,
> >          dsge->length = ssge[ssge_idx].length;
> >          dsge->lkey = rdma_backend_mr_lkey(&mr->backend_mr);
> >  
> > +        *total_length += dsge->length;
> > +
> >          dsge++;
> >      }
> >  
> > @@ -445,8 +463,10 @@ void rdma_backend_post_send(RdmaBackendDev *backend_dev,
> >              rc = mad_send(backend_dev, sgid_idx, sgid, sge, num_sge);
> >              if (rc) {
> >                  complete_work(IBV_WC_GENERAL_ERR, VENDOR_ERR_MAD_SEND, ctx);
> > +                backend_dev->rdma_dev_res->stats.mad_tx_err++;
> >              } else {
> >                  complete_work(IBV_WC_SUCCESS, 0, ctx);
> > +                backend_dev->rdma_dev_res->stats.mad_tx++;
> >              }
> >          }
> >          return;
> > @@ -458,20 +478,21 @@ void rdma_backend_post_send(RdmaBackendDev *backend_dev,
> >      rc = rdma_rm_alloc_cqe_ctx(backend_dev->rdma_dev_res, &bctx_id, bctx);
> >      if (unlikely(rc)) {
> >          complete_work(IBV_WC_GENERAL_ERR, VENDOR_ERR_NOMEM, ctx);
> > -        goto out_free_bctx;
> > +        goto err_free_bctx;
> >      }
> >  
> > -    rc = build_host_sge_array(backend_dev->rdma_dev_res, new_sge, sge, num_sge);
> > +    rc = build_host_sge_array(backend_dev->rdma_dev_res, new_sge, sge, num_sge,
> > +                              &backend_dev->rdma_dev_res->stats.tx_len);
> >      if (rc) {
> >          complete_work(IBV_WC_GENERAL_ERR, rc, ctx);
> > -        goto out_dealloc_cqe_ctx;
> > +        goto err_dealloc_cqe_ctx;
> >      }
> >  
> >      if (qp_type == IBV_QPT_UD) {
> >          wr.wr.ud.ah = create_ah(backend_dev, qp->ibpd, sgid_idx, dgid);
> >          if (!wr.wr.ud.ah) {
> >              complete_work(IBV_WC_GENERAL_ERR, VENDOR_ERR_FAIL_BACKEND, ctx);
> > -            goto out_dealloc_cqe_ctx;
> > +            goto err_dealloc_cqe_ctx;
> >          }
> >          wr.wr.ud.remote_qpn = dqpn;
> >          wr.wr.ud.remote_qkey = dqkey;
> > @@ -488,15 +509,19 @@ void rdma_backend_post_send(RdmaBackendDev *backend_dev,
> >          rdma_error_report("ibv_post_send fail, qpn=0x%x, rc=%d, errno=%d",
> >                            qp->ibqp->qp_num, rc, errno);
> >          complete_work(IBV_WC_GENERAL_ERR, VENDOR_ERR_FAIL_BACKEND, ctx);
> > -        goto out_dealloc_cqe_ctx;
> > +        goto err_dealloc_cqe_ctx;
> >      }
> >  
> > +    atomic_inc(&backend_dev->rdma_dev_res->stats.missing_cqe);
> > +    backend_dev->rdma_dev_res->stats.tx++;
> > +
> >      return;
> >  
> > -out_dealloc_cqe_ctx:
> > +err_dealloc_cqe_ctx:
> > +    backend_dev->rdma_dev_res->stats.tx_err++;
> >      rdma_rm_dealloc_cqe_ctx(backend_dev->rdma_dev_res, bctx_id);
> >  
> > -out_free_bctx:
> > +err_free_bctx:
> >      g_free(bctx);
> >  }
> >  
> > @@ -554,6 +579,9 @@ void rdma_backend_post_recv(RdmaBackendDev *backend_dev,
> >              rc = save_mad_recv_buffer(backend_dev, sge, num_sge, ctx);
> >              if (rc) {
> >                  complete_work(IBV_WC_GENERAL_ERR, rc, ctx);
> > +                rdma_dev_res->stats.mad_rx_bufs_err++;
> > +            } else {
> > +                rdma_dev_res->stats.mad_rx_bufs++;
> >              }
> >          }
> >          return;
> > @@ -565,13 +593,14 @@ void rdma_backend_post_recv(RdmaBackendDev *backend_dev,
> >      rc = rdma_rm_alloc_cqe_ctx(rdma_dev_res, &bctx_id, bctx);
> >      if (unlikely(rc)) {
> >          complete_work(IBV_WC_GENERAL_ERR, VENDOR_ERR_NOMEM, ctx);
> > -        goto out_free_bctx;
> > +        goto err_free_bctx;
> >      }
> >  
> > -    rc = build_host_sge_array(rdma_dev_res, new_sge, sge, num_sge);
> > +    rc = build_host_sge_array(rdma_dev_res, new_sge, sge, num_sge,
> > +                              &backend_dev->rdma_dev_res->stats.rx_bufs_len);
> >      if (rc) {
> >          complete_work(IBV_WC_GENERAL_ERR, rc, ctx);
> > -        goto out_dealloc_cqe_ctx;
> > +        goto err_dealloc_cqe_ctx;
> >      }
> >  
> >      wr.num_sge = num_sge;
> > @@ -582,15 +611,19 @@ void rdma_backend_post_recv(RdmaBackendDev *backend_dev,
> >          rdma_error_report("ibv_post_recv fail, qpn=0x%x, rc=%d, errno=%d",
> >                            qp->ibqp->qp_num, rc, errno);
> >          complete_work(IBV_WC_GENERAL_ERR, VENDOR_ERR_FAIL_BACKEND, ctx);
> > -        goto out_dealloc_cqe_ctx;
> > +        goto err_dealloc_cqe_ctx;
> >      }
> >  
> > +    atomic_inc(&backend_dev->rdma_dev_res->stats.missing_cqe);
> > +    rdma_dev_res->stats.rx_bufs++;
> > +
> >      return;
> >  
> > -out_dealloc_cqe_ctx:
> > +err_dealloc_cqe_ctx:
> > +    backend_dev->rdma_dev_res->stats.rx_bufs_err++;
> >      rdma_rm_dealloc_cqe_ctx(rdma_dev_res, bctx_id);
> >  
> > -out_free_bctx:
> > +err_free_bctx:
> >      g_free(bctx);
> >  }
> >  
> > @@ -929,12 +962,14 @@ static void process_incoming_mad_req(RdmaBackendDev *backend_dev,
> >      bctx = rdma_rm_get_cqe_ctx(backend_dev->rdma_dev_res, cqe_ctx_id);
> >      if (unlikely(!bctx)) {
> >          rdma_error_report("No matching ctx for req %ld", cqe_ctx_id);
> > +        backend_dev->rdma_dev_res->stats.mad_rx_err++;
> >          return;
> >      }
> >  
> >      mad = rdma_pci_dma_map(backend_dev->dev, bctx->sge.addr,
> >                             bctx->sge.length);
> >      if (!mad || bctx->sge.length < msg->umad_len + MAD_HDR_SIZE) {
> > +        backend_dev->rdma_dev_res->stats.mad_rx_err++;
> >          complete_work(IBV_WC_GENERAL_ERR, VENDOR_ERR_INV_MAD_BUFF,
> >                        bctx->up_ctx);
> >      } else {
> > @@ -949,6 +984,7 @@ static void process_incoming_mad_req(RdmaBackendDev *backend_dev,
> >          wc.byte_len = msg->umad_len;
> >          wc.status = IBV_WC_SUCCESS;
> >          wc.wc_flags = IBV_WC_GRH;
> > +        backend_dev->rdma_dev_res->stats.mad_rx++;
> >          comp_handler(bctx->up_ctx, &wc);
> >      }
> >  
> > diff --git a/hw/rdma/rdma_hmp.c b/hw/rdma/rdma_hmp.c
> > new file mode 100644
> > index 0000000000..c5814473c5
> > --- /dev/null
> > +++ b/hw/rdma/rdma_hmp.c
> > @@ -0,0 +1,30 @@
> > +/*
> > + * RDMA device: Human Monitor interface
> > + *
> > + * Copyright (C) 2018 Oracle
> > + * Copyright (C) 2018 Red Hat Inc
> > + *
> > + * Authors:
> > + *     Yuval Shaia <yuval.shaia@oracle.com>
> > + *
> > + * This work is licensed under the terms of the GNU GPL, version 2 or later.
> > + * See the COPYING file in the top-level directory.
> > + *
> > + */
> > +
> > +#include "qemu/osdep.h"
> > +#include "hw/rdma/rdma_hmp.h"
> > +#include "qemu/module.h"
> > +
> > +static const TypeInfo rdma_hmp_info = {
> > +    .name = TYPE_RDMA_STATS_PROVIDER,
> > +    .parent = TYPE_INTERFACE,
> > +    .class_size = sizeof(RdmaStatsProviderClass),
> > +};
> > +
> > +static void rdma_hmp_register_types(void)
> > +{
> > +    type_register_static(&rdma_hmp_info);
> > +}
> > +
> > +type_init(rdma_hmp_register_types)
> > diff --git a/hw/rdma/rdma_rm.c b/hw/rdma/rdma_rm.c
> > index 14580ca379..e019de1a14 100644
> > --- a/hw/rdma/rdma_rm.c
> > +++ b/hw/rdma/rdma_rm.c
> > @@ -16,6 +16,7 @@
> >  #include "qemu/osdep.h"
> >  #include "qapi/error.h"
> >  #include "cpu.h"
> > +#include "monitor/monitor.h"
> >  
> >  #include "trace.h"
> >  #include "rdma_utils.h"
> > @@ -26,6 +27,58 @@
> >  #define PG_DIR_SZ { TARGET_PAGE_SIZE / sizeof(__u64) }
> >  #define PG_TBL_SZ { TARGET_PAGE_SIZE / sizeof(__u64) }
> >  
> > +void rdma_dump_device_counters(Monitor *mon, RdmaDeviceResources *dev_res)
> > +{
> > +    monitor_printf(mon, "\ttx               : %" PRId64 "\n",
> > +                   dev_res->stats.tx);
> > +    monitor_printf(mon, "\ttx_len           : %" PRId64 "\n",
> > +                   dev_res->stats.tx_len);
> > +    monitor_printf(mon, "\ttx_err           : %" PRId64 "\n",
> > +                   dev_res->stats.tx_err);
> > +    monitor_printf(mon, "\trx_bufs          : %" PRId64 "\n",
> > +                   dev_res->stats.rx_bufs);
> > +    monitor_printf(mon, "\trx_bufs_len      : %" PRId64 "\n",
> > +                   dev_res->stats.rx_bufs_len);
> > +    monitor_printf(mon, "\trx_bufs_err      : %" PRId64 "\n",
> > +                   dev_res->stats.rx_bufs_err);
> > +    monitor_printf(mon, "\tcomps            : %" PRId64 "\n",
> > +                   dev_res->stats.completions);
> > +    monitor_printf(mon, "\tmissing_comps    : %" PRId32 "\n",
> > +                   dev_res->stats.missing_cqe);
> > +    monitor_printf(mon, "\tpoll_cq (bk)     : %" PRId64 "\n",
> > +                   dev_res->stats.poll_cq_from_bk);
> > +    monitor_printf(mon, "\tpoll_cq_ppoll_to : %" PRId64 "\n",
> > +                   dev_res->stats.poll_cq_ppoll_to);
> > +    monitor_printf(mon, "\tpoll_cq (fe)     : %" PRId64 "\n",
> > +                   dev_res->stats.poll_cq_from_guest);
> > +    monitor_printf(mon, "\tpoll_cq_empty    : %" PRId64 "\n",
> > +                   dev_res->stats.poll_cq_from_guest_empty);
> > +    monitor_printf(mon, "\tmad_tx           : %" PRId64 "\n",
> > +                   dev_res->stats.mad_tx);
> > +    monitor_printf(mon, "\tmad_tx_err       : %" PRId64 "\n",
> > +                   dev_res->stats.mad_tx_err);
> > +    monitor_printf(mon, "\tmad_rx           : %" PRId64 "\n",
> > +                   dev_res->stats.mad_rx);
> > +    monitor_printf(mon, "\tmad_rx_err       : %" PRId64 "\n",
> > +                   dev_res->stats.mad_rx_err);
> > +    monitor_printf(mon, "\tmad_rx_bufs      : %" PRId64 "\n",
> > +                   dev_res->stats.mad_rx_bufs);
> > +    monitor_printf(mon, "\tmad_rx_bufs_err  : %" PRId64 "\n",
> > +                   dev_res->stats.mad_rx_bufs_err);
> > +    monitor_printf(mon, "\tPDs              : %" PRId32 "\n",
> > +                   dev_res->pd_tbl.used);
> > +    monitor_printf(mon, "\tMRs              : %" PRId32 "\n",
> > +                   dev_res->mr_tbl.used);
> > +    monitor_printf(mon, "\tUCs              : %" PRId32 "\n",
> > +                   dev_res->uc_tbl.used);
> > +    monitor_printf(mon, "\tQPs              : %" PRId32 "\n",
> > +                   dev_res->qp_tbl.used);
> > +    monitor_printf(mon, "\tCQs              : %" PRId32 "\n",
> > +                   dev_res->cq_tbl.used);
> > +    monitor_printf(mon, "\tCEQ_CTXs         : %" PRId32 "\n",
> > +                   dev_res->cqe_ctx_tbl.used);
> > +}
> > +
> >  static inline void res_tbl_init(const char *name, RdmaRmResTbl *tbl,
> >                                  uint32_t tbl_sz, uint32_t res_sz)
> >  {
> > @@ -37,6 +90,7 @@ static inline void res_tbl_init(const char *name, RdmaRmResTbl *tbl,
> >      tbl->bitmap = bitmap_new(tbl_sz);
> >      tbl->tbl_sz = tbl_sz;
> >      tbl->res_sz = res_sz;
> > +    tbl->used = 0;
> >      qemu_mutex_init(&tbl->lock);
> >  }
> >  
> > @@ -76,6 +130,8 @@ static inline void *rdma_res_tbl_alloc(RdmaRmResTbl *tbl, uint32_t *handle)
> >  
> >      set_bit(*handle, tbl->bitmap);
> >  
> > +    tbl->used++;
> > +
> >      qemu_mutex_unlock(&tbl->lock);
> >  
> >      memset(tbl->tbl + *handle * tbl->res_sz, 0, tbl->res_sz);
> > @@ -93,6 +149,7 @@ static inline void rdma_res_tbl_dealloc(RdmaRmResTbl *tbl, uint32_t handle)
> >  
> >      if (handle < tbl->tbl_sz) {
> >          clear_bit(handle, tbl->bitmap);
> > +        tbl->used--;
> >      }
> >  
> >      qemu_mutex_unlock(&tbl->lock);
> > @@ -620,6 +677,9 @@ int rdma_rm_init(RdmaDeviceResources *dev_res, struct ibv_device_attr *dev_attr,
> >  
> >      qemu_mutex_init(&dev_res->lock);
> >  
> > +    memset(&dev_res->stats, 0, sizeof(dev_res->stats));
> > +    atomic_set(&dev_res->stats.missing_cqe, 0);
> > +
> >      return 0;
> >  }
> >  
> > diff --git a/hw/rdma/rdma_rm.h b/hw/rdma/rdma_rm.h
> > index 9ec87d2667..a527d306f6 100644
> > --- a/hw/rdma/rdma_rm.h
> > +++ b/hw/rdma/rdma_rm.h
> > @@ -20,6 +20,7 @@
> >  #include "rdma_backend_defs.h"
> >  #include "rdma_rm_defs.h"
> >  
> > +void rdma_dump_device_counters(Monitor *mon, RdmaDeviceResources *dev_res);
> >  int rdma_rm_init(RdmaDeviceResources *dev_res, struct ibv_device_attr *dev_attr,
> >                   Error **errp);
> >  void rdma_rm_fini(RdmaDeviceResources *dev_res, RdmaBackendDev *backend_dev,
> > diff --git a/hw/rdma/rdma_rm_defs.h b/hw/rdma/rdma_rm_defs.h
> > index f0ee1f3072..4b8d704cfe 100644
> > --- a/hw/rdma/rdma_rm_defs.h
> > +++ b/hw/rdma/rdma_rm_defs.h
> > @@ -34,7 +34,9 @@
> >  #define MAX_QP_INIT_RD_ATOM   16
> >  #define MAX_AH                64
> >  
> > -#define MAX_RM_TBL_NAME 16
> > +#define MAX_RM_TBL_NAME             16
> > +#define MAX_CONSEQ_EMPTY_POLL_CQ    4096 /* considered as error above this */
> > +
> >  typedef struct RdmaRmResTbl {
> >      char name[MAX_RM_TBL_NAME];
> >      QemuMutex lock;
> > @@ -42,6 +44,7 @@ typedef struct RdmaRmResTbl {
> >      size_t tbl_sz;
> >      size_t res_sz;
> >      void *tbl;
> > +    uint32_t used; /* number of used entries in the table */
> >  } RdmaRmResTbl;
> >  
> >  typedef struct RdmaRmPD {
> > @@ -96,6 +99,27 @@ typedef struct RdmaRmPort {
> >      enum ibv_port_state state;
> >  } RdmaRmPort;
> >  
> > +typedef struct RdmaRmStats {
> > +    uint64_t tx;
> > +    uint64_t tx_len;
> > +    uint64_t tx_err;
> > +    uint64_t rx_bufs;
> > +    uint64_t rx_bufs_len;
> > +    uint64_t rx_bufs_err;
> > +    uint64_t completions;
> > +    uint64_t mad_tx;
> > +    uint64_t mad_tx_err;
> > +    uint64_t mad_rx;
> > +    uint64_t mad_rx_err;
> > +    uint64_t mad_rx_bufs;
> > +    uint64_t mad_rx_bufs_err;
> > +    uint64_t poll_cq_from_bk;
> > +    uint64_t poll_cq_from_guest;
> > +    uint64_t poll_cq_from_guest_empty;
> > +    uint64_t poll_cq_ppoll_to;
> > +    uint32_t missing_cqe;
> > +} RdmaRmStats;
> > +
> >  typedef struct RdmaDeviceResources {
> >      RdmaRmPort port;
> >      RdmaRmResTbl pd_tbl;
> > @@ -106,6 +130,7 @@ typedef struct RdmaDeviceResources {
> >      RdmaRmResTbl cqe_ctx_tbl;
> >      GHashTable *qp_hash; /* Keeps mapping between real and emulated */
> >      QemuMutex lock;
> > +    RdmaRmStats stats;
> >  } RdmaDeviceResources;
> >  
> >  #endif
> > diff --git a/hw/rdma/vmw/pvrdma.h b/hw/rdma/vmw/pvrdma.h
> > index 0879224957..167706ec2c 100644
> > --- a/hw/rdma/vmw/pvrdma.h
> > +++ b/hw/rdma/vmw/pvrdma.h
> > @@ -70,6 +70,10 @@ typedef struct DSRInfo {
> >      PvrdmaRing cq;
> >  } DSRInfo;
> >  
> > +typedef struct PVRDMADevStats {
> > +    uint64_t commands;
> > +} PVRDMADevStats;
> > +
> >  typedef struct PVRDMADev {
> >      PCIDevice parent_obj;
> >      MemoryRegion msix;
> > @@ -89,6 +93,7 @@ typedef struct PVRDMADev {
> >      CharBackend mad_chr;
> >      VMXNET3State *func0;
> >      Notifier shutdown_notifier;
> > +    PVRDMADevStats stats;
> >  } PVRDMADev;
> >  #define PVRDMA_DEV(dev) OBJECT_CHECK(PVRDMADev, (dev), PVRDMA_HW_NAME)
> >  
> > diff --git a/hw/rdma/vmw/pvrdma_main.c b/hw/rdma/vmw/pvrdma_main.c
> > index b6061f4b6e..659331ac93 100644
> > --- a/hw/rdma/vmw/pvrdma_main.c
> > +++ b/hw/rdma/vmw/pvrdma_main.c
> > @@ -25,6 +25,8 @@
> >  #include "cpu.h"
> >  #include "trace.h"
> >  #include "sysemu/sysemu.h"
> > +#include "monitor/monitor.h"
> > +#include "hw/rdma/rdma_hmp.h"
> >  
> >  #include "../rdma_rm.h"
> >  #include "../rdma_backend.h"
> > @@ -55,6 +57,18 @@ static Property pvrdma_dev_properties[] = {
> >      DEFINE_PROP_END_OF_LIST(),
> >  };
> >  
> > +static void pvrdma_print_statistics(Monitor *mon, RdmaStatsProvider *obj)
> > +{
> > +    PVRDMADev *dev = PVRDMA_DEV(obj);
> > +    PCIDevice *pdev = PCI_DEVICE(dev);
> > +
> > +    monitor_printf(mon, "%s, %x.%x\n", pdev->name, PCI_SLOT(pdev->devfn),
> > +                   PCI_FUNC(pdev->devfn));
> > +    monitor_printf(mon, "\tcommands         : %" PRId64 "\n",
> > +                   dev->stats.commands);
> > +    rdma_dump_device_counters(mon, &dev->rdma_dev_res);
> > +}
> > +
> >  static void free_dev_ring(PCIDevice *pci_dev, PvrdmaRing *ring,
> >                            void *ring_state)
> >  {
> > @@ -394,6 +408,7 @@ static void pvrdma_regs_write(void *opaque, hwaddr addr, uint64_t val,
> >          if (val == 0) {
> >              trace_pvrdma_regs_write(addr, val, "REQUEST", "");
> >              pvrdma_exec_cmd(dev);
> > +            dev->stats.commands++;
> >          }
> >          break;
> >      default:
> > @@ -612,6 +627,8 @@ static void pvrdma_realize(PCIDevice *pdev, Error **errp)
> >          goto out;
> >      }
> >  
> > +    memset(&dev->stats, 0, sizeof(dev->stats));
> > +
> >      dev->shutdown_notifier.notify = pvrdma_shutdown_notifier;
> >      qemu_register_shutdown_notifier(&dev->shutdown_notifier);
> >  
> > @@ -631,6 +648,7 @@ static void pvrdma_class_init(ObjectClass *klass, void *data)
> >  {
> >      DeviceClass *dc = DEVICE_CLASS(klass);
> >      PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
> > +    RdmaStatsProviderClass *ir = RDMA_STATS_PROVIDER_CLASS(klass);
> >  
> >      k->realize = pvrdma_realize;
> >      k->exit = pvrdma_exit;
> > @@ -642,6 +660,8 @@ static void pvrdma_class_init(ObjectClass *klass, void *data)
> >      dc->desc = "RDMA Device";
> >      dc->props = pvrdma_dev_properties;
> >      set_bit(DEVICE_CATEGORY_NETWORK, dc->categories);
> > +
> > +    ir->print_statistics = pvrdma_print_statistics;
> >  }
> >  
> >  static const TypeInfo pvrdma_info = {
> > @@ -651,6 +671,7 @@ static const TypeInfo pvrdma_info = {
> >      .class_init = pvrdma_class_init,
> >      .interfaces = (InterfaceInfo[]) {
> >          { INTERFACE_CONVENTIONAL_PCI_DEVICE },
> > +        { TYPE_RDMA_STATS_PROVIDER },
> >          { }
> >      }
> >  };
> > diff --git a/include/hw/rdma/rdma_hmp.h b/include/hw/rdma/rdma_hmp.h
> > new file mode 100644
> > index 0000000000..dd23f2bc84
> > --- /dev/null
> > +++ b/include/hw/rdma/rdma_hmp.h
> > @@ -0,0 +1,40 @@
> > +/*
> > + * RDMA device: Human Monitor interface
> > + *
> > + * Copyright (C) 2019 Oracle
> > + * Copyright (C) 2019 Red Hat Inc
> > + *
> > + * Authors:
> > + *     Yuval Shaia <yuval.shaia@oracle.com>
> > + *
> > + * This work is licensed under the terms of the GNU GPL, version 2 or later.
> > + * See the COPYING file in the top-level directory.
> > + *
> > + */
> > +
> > +#ifndef RDMA_HMP_H
> > +#define RDMA_HMP_H
> > +
> > +#include "qom/object.h"
> > +
> > +#define TYPE_RDMA_STATS_PROVIDER "rdma"
> > +
> > +#define RDMA_STATS_PROVIDER_CLASS(klass) \
> > +    OBJECT_CLASS_CHECK(RdmaStatsProviderClass, (klass), \
> > +                       TYPE_RDMA_STATS_PROVIDER)
> > +#define RDMA_STATS_PROVIDER_GET_CLASS(obj) \
> > +    OBJECT_GET_CLASS(RdmaStatsProviderClass, (obj), \
> > +                     TYPE_RDMA_STATS_PROVIDER)
> > +#define RDMA_STATS_PROVIDER(obj) \
> > +    INTERFACE_CHECK(RdmaStatsProvider, (obj), \
> > +                    TYPE_RDMA_STATS_PROVIDER)
> > +
> > +typedef struct RdmaStatsProvider RdmaStatsProvider;
> > +
> > +typedef struct RdmaStatsProviderClass {
> > +    InterfaceClass parent;
> > +
> > +    void (*print_statistics)(Monitor *mon, RdmaStatsProvider *obj);
> > +} RdmaStatsProviderClass;
> > +
> > +#endif
> > -- 
> > 2.17.2
> > 
> > 
--
Dr. David Alan Gilbert / dgilbert@redhat.com / Manchester, UK

Re: [Qemu-devel] [PATCH v4 4/9] {hmp, hw/pvrdma}: Expose device internals via monitor interface

[Yuval Shaia]

On Wed, Mar 06, 2019 at 12:20:49PM +0000, Dr. David Alan Gilbert wrote:
> * Yuval Shaia (yuval.shaia@oracle.com) wrote:
> > On Sun, Mar 03, 2019 at 10:33:40PM +0200, Yuval Shaia wrote:
> > > Allow interrogating device internals through HMP interface.
> > > The exposed indicators can be used for troubleshooting by developers or
> > > sysadmin.
> > > There is no need to expose these attributes to a management system (e.x.
> > > libvirt) because (1) most of them are not "device-management' related
> > > info and (2) there is no guarantee the interface is stable.
> > > 
> > > Signed-off-by: Yuval Shaia <yuval.shaia@oracle.com>
> > > ---
> > >  hmp-commands-info.hx       | 14 ++++++++
> > >  hmp.c                      | 27 +++++++++++++++
> > >  hmp.h                      |  1 +
> > >  hw/rdma/Makefile.objs      |  2 +-
> > >  hw/rdma/rdma_backend.c     | 70 +++++++++++++++++++++++++++++---------
> > >  hw/rdma/rdma_hmp.c         | 30 ++++++++++++++++
> > >  hw/rdma/rdma_rm.c          | 60 ++++++++++++++++++++++++++++++++
> > >  hw/rdma/rdma_rm.h          |  1 +
> > >  hw/rdma/rdma_rm_defs.h     | 27 ++++++++++++++-
> > >  hw/rdma/vmw/pvrdma.h       |  5 +++
> > >  hw/rdma/vmw/pvrdma_main.c  | 21 ++++++++++++
> > >  include/hw/rdma/rdma_hmp.h | 40 ++++++++++++++++++++++
> > >  12 files changed, 279 insertions(+), 19 deletions(-)
> > >  create mode 100644 hw/rdma/rdma_hmp.c
> > >  create mode 100644 include/hw/rdma/rdma_hmp.h
> > > 
> > > diff --git a/hmp-commands-info.hx b/hmp-commands-info.hx
> > > index cbee8b944d..c59444c461 100644
> > > --- a/hmp-commands-info.hx
> > > +++ b/hmp-commands-info.hx
> > > @@ -202,6 +202,20 @@ STEXI
> > >  @item info pic
> > >  @findex info pic
> > >  Show PIC state.
> > > +ETEXI
> > > +
> > > +    {
> > > +        .name       = "rdma",
> > > +        .args_type  = "",
> > > +        .params     = "",
> > > +        .help       = "show RDMA state",
> > > +        .cmd        = hmp_info_rdma,
> > > +    },
> > > +
> > > +STEXI
> > > +@item info rdma
> > > +@findex info rdma
> > > +Show RDMA state.
> > >  ETEXI
> > >  
> > >      {
> > > diff --git a/hmp.c b/hmp.c
> > > index 1e006eeb49..68511b2441 100644
> > > --- a/hmp.c
> > > +++ b/hmp.c
> > > @@ -51,6 +51,7 @@
> > >  #include "qemu/error-report.h"
> > >  #include "exec/ramlist.h"
> > >  #include "hw/intc/intc.h"
> > > +#include "hw/rdma/rdma_hmp.h"
> > >  #include "migration/snapshot.h"
> > >  #include "migration/misc.h"
> > >  
> > > @@ -968,6 +969,32 @@ void hmp_info_pic(Monitor *mon, const QDict *qdict)
> > >                                     hmp_info_pic_foreach, mon);
> > >  }
> > >  
> > > +static int hmp_info_rdma_foreach(Object *obj, void *opaque)
> > > +{
> > > +    RdmaStatsProvider *rdma;
> > > +    RdmaStatsProviderClass *k;
> > > +    Monitor *mon = opaque;
> > > +
> > > +    if (object_dynamic_cast(obj, TYPE_RDMA_STATS_PROVIDER)) {
> > > +        rdma = RDMA_STATS_PROVIDER(obj);
> > > +        k = RDMA_STATS_PROVIDER_GET_CLASS(obj);
> > > +        if (k->print_statistics) {
> > > +            k->print_statistics(mon, rdma);
> > > +        } else {
> > > +            monitor_printf(mon, "RDMA statistics not available for %s.\n",
> > > +                           object_get_typename(obj));
> > > +        }
> > > +    }
> > > +
> > > +    return 0;
> > > +}
> > > +
> > > +void hmp_info_rdma(Monitor *mon, const QDict *qdict)
> > > +{
> > > +    object_child_foreach_recursive(object_get_root(),
> > > +                                   hmp_info_rdma_foreach, mon);
> > > +}
> > > +
> > 
> > Hi Markus and David,
> > Is this implementation is acceptable by you?
> 
> I think from the HMP side I'm fine, so:
> 
> Acked-by: Dr. David Alan Gilbert <dgilbert@redhat.com>

Thanks for that.

Hope that some one can review the 'counters' part of the patch.

> 
> I would say that there's really two parts to this patch, all the places
> you change your code to gather the stats, and the separate pieces that
> wire it into HMP;  if you had to repost for another reason I'd
> split the patch like that.

Actually that how it was in v0 but then Markus (i think) comes up with the
idea to squash them as the *only* consumer of these counters is HMP.

But in general, i do agree to this but also can understand the other side.

> 
> Dave
> 
> > Thanks,
> > Yuval
> > 
> > 
> > >  void hmp_info_pci(Monitor *mon, const QDict *qdict)
> > >  {
> > >      PciInfoList *info_list, *info;
> > > diff --git a/hmp.h b/hmp.h
> > > index 5f1addcca2..666949afc3 100644
> > > --- a/hmp.h
> > > +++ b/hmp.h
> > > @@ -36,6 +36,7 @@ void hmp_info_spice(Monitor *mon, const QDict *qdict);
> > >  void hmp_info_balloon(Monitor *mon, const QDict *qdict);
> > >  void hmp_info_irq(Monitor *mon, const QDict *qdict);
> > >  void hmp_info_pic(Monitor *mon, const QDict *qdict);
> > > +void hmp_info_rdma(Monitor *mon, const QDict *qdict);
> > >  void hmp_info_pci(Monitor *mon, const QDict *qdict);
> > >  void hmp_info_block_jobs(Monitor *mon, const QDict *qdict);
> > >  void hmp_info_tpm(Monitor *mon, const QDict *qdict);
> > > diff --git a/hw/rdma/Makefile.objs b/hw/rdma/Makefile.objs
> > > index bd36cbf51c..dd59faff51 100644
> > > --- a/hw/rdma/Makefile.objs
> > > +++ b/hw/rdma/Makefile.objs
> > > @@ -1,5 +1,5 @@
> > >  ifeq ($(CONFIG_PVRDMA),y)
> > > -obj-$(CONFIG_PCI) += rdma_utils.o rdma_backend.o rdma_rm.o
> > > +obj-$(CONFIG_PCI) += rdma_utils.o rdma_backend.o rdma_rm.o rdma_hmp.o
> > >  obj-$(CONFIG_PCI) += vmw/pvrdma_dev_ring.o vmw/pvrdma_cmd.o \
> > >                       vmw/pvrdma_qp_ops.o vmw/pvrdma_main.o
> > >  endif
> > > diff --git a/hw/rdma/rdma_backend.c b/hw/rdma/rdma_backend.c
> > > index 9679b842d1..bc2fefcf93 100644
> > > --- a/hw/rdma/rdma_backend.c
> > > +++ b/hw/rdma/rdma_backend.c
> > > @@ -64,9 +64,9 @@ static inline void complete_work(enum ibv_wc_status status, uint32_t vendor_err,
> > >      comp_handler(ctx, &wc);
> > >  }
> > >  
> > > -static void rdma_poll_cq(RdmaDeviceResources *rdma_dev_res, struct ibv_cq *ibcq)
> > > +static int rdma_poll_cq(RdmaDeviceResources *rdma_dev_res, struct ibv_cq *ibcq)
> > >  {
> > > -    int i, ne;
> > > +    int i, ne, total_ne = 0;
> > >      BackendCtx *bctx;
> > >      struct ibv_wc wc[2];
> > >  
> > > @@ -89,12 +89,18 @@ static void rdma_poll_cq(RdmaDeviceResources *rdma_dev_res, struct ibv_cq *ibcq)
> > >              rdma_rm_dealloc_cqe_ctx(rdma_dev_res, wc[i].wr_id);
> > >              g_free(bctx);
> > >          }
> > > +        total_ne += ne;
> > >      } while (ne > 0);
> > > +    atomic_sub(&rdma_dev_res->stats.missing_cqe, total_ne);
> > >      qemu_mutex_unlock(&rdma_dev_res->lock);
> > >  
> > >      if (ne < 0) {
> > >          rdma_error_report("ibv_poll_cq fail, rc=%d, errno=%d", ne, errno);
> > >      }
> > > +
> > > +    rdma_dev_res->stats.completions += total_ne;
> > > +
> > > +    return total_ne;
> > >  }
> > >  
> > >  static void *comp_handler_thread(void *arg)
> > > @@ -122,6 +128,9 @@ static void *comp_handler_thread(void *arg)
> > >      while (backend_dev->comp_thread.run) {
> > >          do {
> > >              rc = qemu_poll_ns(pfds, 1, THR_POLL_TO * (int64_t)SCALE_MS);
> > > +            if (!rc) {
> > > +                backend_dev->rdma_dev_res->stats.poll_cq_ppoll_to++;
> > > +            }
> > >          } while (!rc && backend_dev->comp_thread.run);
> > >  
> > >          if (backend_dev->comp_thread.run) {
> > > @@ -138,6 +147,7 @@ static void *comp_handler_thread(void *arg)
> > >                                    errno);
> > >              }
> > >  
> > > +            backend_dev->rdma_dev_res->stats.poll_cq_from_bk++;
> > >              rdma_poll_cq(backend_dev->rdma_dev_res, ev_cq);
> > >  
> > >              ibv_ack_cq_events(ev_cq, 1);
> > > @@ -271,7 +281,13 @@ int rdma_backend_query_port(RdmaBackendDev *backend_dev,
> > >  
> > >  void rdma_backend_poll_cq(RdmaDeviceResources *rdma_dev_res, RdmaBackendCQ *cq)
> > >  {
> > > -    rdma_poll_cq(rdma_dev_res, cq->ibcq);
> > > +    int polled;
> > > +
> > > +    rdma_dev_res->stats.poll_cq_from_guest++;
> > > +    polled = rdma_poll_cq(rdma_dev_res, cq->ibcq);
> > > +    if (!polled) {
> > > +        rdma_dev_res->stats.poll_cq_from_guest_empty++;
> > > +    }
> > >  }
> > >  
> > >  static GHashTable *ah_hash;
> > > @@ -333,7 +349,7 @@ static void ah_cache_init(void)
> > >  
> > >  static int build_host_sge_array(RdmaDeviceResources *rdma_dev_res,
> > >                                  struct ibv_sge *dsge, struct ibv_sge *ssge,
> > > -                                uint8_t num_sge)
> > > +                                uint8_t num_sge, uint64_t *total_length)
> > >  {
> > >      RdmaRmMR *mr;
> > >      int ssge_idx;
> > > @@ -349,6 +365,8 @@ static int build_host_sge_array(RdmaDeviceResources *rdma_dev_res,
> > >          dsge->length = ssge[ssge_idx].length;
> > >          dsge->lkey = rdma_backend_mr_lkey(&mr->backend_mr);
> > >  
> > > +        *total_length += dsge->length;
> > > +
> > >          dsge++;
> > >      }
> > >  
> > > @@ -445,8 +463,10 @@ void rdma_backend_post_send(RdmaBackendDev *backend_dev,
> > >              rc = mad_send(backend_dev, sgid_idx, sgid, sge, num_sge);
> > >              if (rc) {
> > >                  complete_work(IBV_WC_GENERAL_ERR, VENDOR_ERR_MAD_SEND, ctx);
> > > +                backend_dev->rdma_dev_res->stats.mad_tx_err++;
> > >              } else {
> > >                  complete_work(IBV_WC_SUCCESS, 0, ctx);
> > > +                backend_dev->rdma_dev_res->stats.mad_tx++;
> > >              }
> > >          }
> > >          return;
> > > @@ -458,20 +478,21 @@ void rdma_backend_post_send(RdmaBackendDev *backend_dev,
> > >      rc = rdma_rm_alloc_cqe_ctx(backend_dev->rdma_dev_res, &bctx_id, bctx);
> > >      if (unlikely(rc)) {
> > >          complete_work(IBV_WC_GENERAL_ERR, VENDOR_ERR_NOMEM, ctx);
> > > -        goto out_free_bctx;
> > > +        goto err_free_bctx;
> > >      }
> > >  
> > > -    rc = build_host_sge_array(backend_dev->rdma_dev_res, new_sge, sge, num_sge);
> > > +    rc = build_host_sge_array(backend_dev->rdma_dev_res, new_sge, sge, num_sge,
> > > +                              &backend_dev->rdma_dev_res->stats.tx_len);
> > >      if (rc) {
> > >          complete_work(IBV_WC_GENERAL_ERR, rc, ctx);
> > > -        goto out_dealloc_cqe_ctx;
> > > +        goto err_dealloc_cqe_ctx;
> > >      }
> > >  
> > >      if (qp_type == IBV_QPT_UD) {
> > >          wr.wr.ud.ah = create_ah(backend_dev, qp->ibpd, sgid_idx, dgid);
> > >          if (!wr.wr.ud.ah) {
> > >              complete_work(IBV_WC_GENERAL_ERR, VENDOR_ERR_FAIL_BACKEND, ctx);
> > > -            goto out_dealloc_cqe_ctx;
> > > +            goto err_dealloc_cqe_ctx;
> > >          }
> > >          wr.wr.ud.remote_qpn = dqpn;
> > >          wr.wr.ud.remote_qkey = dqkey;
> > > @@ -488,15 +509,19 @@ void rdma_backend_post_send(RdmaBackendDev *backend_dev,
> > >          rdma_error_report("ibv_post_send fail, qpn=0x%x, rc=%d, errno=%d",
> > >                            qp->ibqp->qp_num, rc, errno);
> > >          complete_work(IBV_WC_GENERAL_ERR, VENDOR_ERR_FAIL_BACKEND, ctx);
> > > -        goto out_dealloc_cqe_ctx;
> > > +        goto err_dealloc_cqe_ctx;
> > >      }
> > >  
> > > +    atomic_inc(&backend_dev->rdma_dev_res->stats.missing_cqe);
> > > +    backend_dev->rdma_dev_res->stats.tx++;
> > > +
> > >      return;
> > >  
> > > -out_dealloc_cqe_ctx:
> > > +err_dealloc_cqe_ctx:
> > > +    backend_dev->rdma_dev_res->stats.tx_err++;
> > >      rdma_rm_dealloc_cqe_ctx(backend_dev->rdma_dev_res, bctx_id);
> > >  
> > > -out_free_bctx:
> > > +err_free_bctx:
> > >      g_free(bctx);
> > >  }
> > >  
> > > @@ -554,6 +579,9 @@ void rdma_backend_post_recv(RdmaBackendDev *backend_dev,
> > >              rc = save_mad_recv_buffer(backend_dev, sge, num_sge, ctx);
> > >              if (rc) {
> > >                  complete_work(IBV_WC_GENERAL_ERR, rc, ctx);
> > > +                rdma_dev_res->stats.mad_rx_bufs_err++;
> > > +            } else {
> > > +                rdma_dev_res->stats.mad_rx_bufs++;
> > >              }
> > >          }
> > >          return;
> > > @@ -565,13 +593,14 @@ void rdma_backend_post_recv(RdmaBackendDev *backend_dev,
> > >      rc = rdma_rm_alloc_cqe_ctx(rdma_dev_res, &bctx_id, bctx);
> > >      if (unlikely(rc)) {
> > >          complete_work(IBV_WC_GENERAL_ERR, VENDOR_ERR_NOMEM, ctx);
> > > -        goto out_free_bctx;
> > > +        goto err_free_bctx;
> > >      }
> > >  
> > > -    rc = build_host_sge_array(rdma_dev_res, new_sge, sge, num_sge);
> > > +    rc = build_host_sge_array(rdma_dev_res, new_sge, sge, num_sge,
> > > +                              &backend_dev->rdma_dev_res->stats.rx_bufs_len);
> > >      if (rc) {
> > >          complete_work(IBV_WC_GENERAL_ERR, rc, ctx);
> > > -        goto out_dealloc_cqe_ctx;
> > > +        goto err_dealloc_cqe_ctx;
> > >      }
> > >  
> > >      wr.num_sge = num_sge;
> > > @@ -582,15 +611,19 @@ void rdma_backend_post_recv(RdmaBackendDev *backend_dev,
> > >          rdma_error_report("ibv_post_recv fail, qpn=0x%x, rc=%d, errno=%d",
> > >                            qp->ibqp->qp_num, rc, errno);
> > >          complete_work(IBV_WC_GENERAL_ERR, VENDOR_ERR_FAIL_BACKEND, ctx);
> > > -        goto out_dealloc_cqe_ctx;
> > > +        goto err_dealloc_cqe_ctx;
> > >      }
> > >  
> > > +    atomic_inc(&backend_dev->rdma_dev_res->stats.missing_cqe);
> > > +    rdma_dev_res->stats.rx_bufs++;
> > > +
> > >      return;
> > >  
> > > -out_dealloc_cqe_ctx:
> > > +err_dealloc_cqe_ctx:
> > > +    backend_dev->rdma_dev_res->stats.rx_bufs_err++;
> > >      rdma_rm_dealloc_cqe_ctx(rdma_dev_res, bctx_id);
> > >  
> > > -out_free_bctx:
> > > +err_free_bctx:
> > >      g_free(bctx);
> > >  }
> > >  
> > > @@ -929,12 +962,14 @@ static void process_incoming_mad_req(RdmaBackendDev *backend_dev,
> > >      bctx = rdma_rm_get_cqe_ctx(backend_dev->rdma_dev_res, cqe_ctx_id);
> > >      if (unlikely(!bctx)) {
> > >          rdma_error_report("No matching ctx for req %ld", cqe_ctx_id);
> > > +        backend_dev->rdma_dev_res->stats.mad_rx_err++;
> > >          return;
> > >      }
> > >  
> > >      mad = rdma_pci_dma_map(backend_dev->dev, bctx->sge.addr,
> > >                             bctx->sge.length);
> > >      if (!mad || bctx->sge.length < msg->umad_len + MAD_HDR_SIZE) {
> > > +        backend_dev->rdma_dev_res->stats.mad_rx_err++;
> > >          complete_work(IBV_WC_GENERAL_ERR, VENDOR_ERR_INV_MAD_BUFF,
> > >                        bctx->up_ctx);
> > >      } else {
> > > @@ -949,6 +984,7 @@ static void process_incoming_mad_req(RdmaBackendDev *backend_dev,
> > >          wc.byte_len = msg->umad_len;
> > >          wc.status = IBV_WC_SUCCESS;
> > >          wc.wc_flags = IBV_WC_GRH;
> > > +        backend_dev->rdma_dev_res->stats.mad_rx++;
> > >          comp_handler(bctx->up_ctx, &wc);
> > >      }
> > >  
> > > diff --git a/hw/rdma/rdma_hmp.c b/hw/rdma/rdma_hmp.c
> > > new file mode 100644
> > > index 0000000000..c5814473c5
> > > --- /dev/null
> > > +++ b/hw/rdma/rdma_hmp.c
> > > @@ -0,0 +1,30 @@
> > > +/*
> > > + * RDMA device: Human Monitor interface
> > > + *
> > > + * Copyright (C) 2018 Oracle
> > > + * Copyright (C) 2018 Red Hat Inc
> > > + *
> > > + * Authors:
> > > + *     Yuval Shaia <yuval.shaia@oracle.com>
> > > + *
> > > + * This work is licensed under the terms of the GNU GPL, version 2 or later.
> > > + * See the COPYING file in the top-level directory.
> > > + *
> > > + */
> > > +
> > > +#include "qemu/osdep.h"
> > > +#include "hw/rdma/rdma_hmp.h"
> > > +#include "qemu/module.h"
> > > +
> > > +static const TypeInfo rdma_hmp_info = {
> > > +    .name = TYPE_RDMA_STATS_PROVIDER,
> > > +    .parent = TYPE_INTERFACE,
> > > +    .class_size = sizeof(RdmaStatsProviderClass),
> > > +};
> > > +
> > > +static void rdma_hmp_register_types(void)
> > > +{
> > > +    type_register_static(&rdma_hmp_info);
> > > +}
> > > +
> > > +type_init(rdma_hmp_register_types)
> > > diff --git a/hw/rdma/rdma_rm.c b/hw/rdma/rdma_rm.c
> > > index 14580ca379..e019de1a14 100644
> > > --- a/hw/rdma/rdma_rm.c
> > > +++ b/hw/rdma/rdma_rm.c
> > > @@ -16,6 +16,7 @@
> > >  #include "qemu/osdep.h"
> > >  #include "qapi/error.h"
> > >  #include "cpu.h"
> > > +#include "monitor/monitor.h"
> > >  
> > >  #include "trace.h"
> > >  #include "rdma_utils.h"
> > > @@ -26,6 +27,58 @@
> > >  #define PG_DIR_SZ { TARGET_PAGE_SIZE / sizeof(__u64) }
> > >  #define PG_TBL_SZ { TARGET_PAGE_SIZE / sizeof(__u64) }
> > >  
> > > +void rdma_dump_device_counters(Monitor *mon, RdmaDeviceResources *dev_res)
> > > +{
> > > +    monitor_printf(mon, "\ttx               : %" PRId64 "\n",
> > > +                   dev_res->stats.tx);
> > > +    monitor_printf(mon, "\ttx_len           : %" PRId64 "\n",
> > > +                   dev_res->stats.tx_len);
> > > +    monitor_printf(mon, "\ttx_err           : %" PRId64 "\n",
> > > +                   dev_res->stats.tx_err);
> > > +    monitor_printf(mon, "\trx_bufs          : %" PRId64 "\n",
> > > +                   dev_res->stats.rx_bufs);
> > > +    monitor_printf(mon, "\trx_bufs_len      : %" PRId64 "\n",
> > > +                   dev_res->stats.rx_bufs_len);
> > > +    monitor_printf(mon, "\trx_bufs_err      : %" PRId64 "\n",
> > > +                   dev_res->stats.rx_bufs_err);
> > > +    monitor_printf(mon, "\tcomps            : %" PRId64 "\n",
> > > +                   dev_res->stats.completions);
> > > +    monitor_printf(mon, "\tmissing_comps    : %" PRId32 "\n",
> > > +                   dev_res->stats.missing_cqe);
> > > +    monitor_printf(mon, "\tpoll_cq (bk)     : %" PRId64 "\n",
> > > +                   dev_res->stats.poll_cq_from_bk);
> > > +    monitor_printf(mon, "\tpoll_cq_ppoll_to : %" PRId64 "\n",
> > > +                   dev_res->stats.poll_cq_ppoll_to);
> > > +    monitor_printf(mon, "\tpoll_cq (fe)     : %" PRId64 "\n",
> > > +                   dev_res->stats.poll_cq_from_guest);
> > > +    monitor_printf(mon, "\tpoll_cq_empty    : %" PRId64 "\n",
> > > +                   dev_res->stats.poll_cq_from_guest_empty);
> > > +    monitor_printf(mon, "\tmad_tx           : %" PRId64 "\n",
> > > +                   dev_res->stats.mad_tx);
> > > +    monitor_printf(mon, "\tmad_tx_err       : %" PRId64 "\n",
> > > +                   dev_res->stats.mad_tx_err);
> > > +    monitor_printf(mon, "\tmad_rx           : %" PRId64 "\n",
> > > +                   dev_res->stats.mad_rx);
> > > +    monitor_printf(mon, "\tmad_rx_err       : %" PRId64 "\n",
> > > +                   dev_res->stats.mad_rx_err);
> > > +    monitor_printf(mon, "\tmad_rx_bufs      : %" PRId64 "\n",
> > > +                   dev_res->stats.mad_rx_bufs);
> > > +    monitor_printf(mon, "\tmad_rx_bufs_err  : %" PRId64 "\n",
> > > +                   dev_res->stats.mad_rx_bufs_err);
> > > +    monitor_printf(mon, "\tPDs              : %" PRId32 "\n",
> > > +                   dev_res->pd_tbl.used);
> > > +    monitor_printf(mon, "\tMRs              : %" PRId32 "\n",
> > > +                   dev_res->mr_tbl.used);
> > > +    monitor_printf(mon, "\tUCs              : %" PRId32 "\n",
> > > +                   dev_res->uc_tbl.used);
> > > +    monitor_printf(mon, "\tQPs              : %" PRId32 "\n",
> > > +                   dev_res->qp_tbl.used);
> > > +    monitor_printf(mon, "\tCQs              : %" PRId32 "\n",
> > > +                   dev_res->cq_tbl.used);
> > > +    monitor_printf(mon, "\tCEQ_CTXs         : %" PRId32 "\n",
> > > +                   dev_res->cqe_ctx_tbl.used);
> > > +}
> > > +
> > >  static inline void res_tbl_init(const char *name, RdmaRmResTbl *tbl,
> > >                                  uint32_t tbl_sz, uint32_t res_sz)
> > >  {
> > > @@ -37,6 +90,7 @@ static inline void res_tbl_init(const char *name, RdmaRmResTbl *tbl,
> > >      tbl->bitmap = bitmap_new(tbl_sz);
> > >      tbl->tbl_sz = tbl_sz;
> > >      tbl->res_sz = res_sz;
> > > +    tbl->used = 0;
> > >      qemu_mutex_init(&tbl->lock);
> > >  }
> > >  
> > > @@ -76,6 +130,8 @@ static inline void *rdma_res_tbl_alloc(RdmaRmResTbl *tbl, uint32_t *handle)
> > >  
> > >      set_bit(*handle, tbl->bitmap);
> > >  
> > > +    tbl->used++;
> > > +
> > >      qemu_mutex_unlock(&tbl->lock);
> > >  
> > >      memset(tbl->tbl + *handle * tbl->res_sz, 0, tbl->res_sz);
> > > @@ -93,6 +149,7 @@ static inline void rdma_res_tbl_dealloc(RdmaRmResTbl *tbl, uint32_t handle)
> > >  
> > >      if (handle < tbl->tbl_sz) {
> > >          clear_bit(handle, tbl->bitmap);
> > > +        tbl->used--;
> > >      }
> > >  
> > >      qemu_mutex_unlock(&tbl->lock);
> > > @@ -620,6 +677,9 @@ int rdma_rm_init(RdmaDeviceResources *dev_res, struct ibv_device_attr *dev_attr,
> > >  
> > >      qemu_mutex_init(&dev_res->lock);
> > >  
> > > +    memset(&dev_res->stats, 0, sizeof(dev_res->stats));
> > > +    atomic_set(&dev_res->stats.missing_cqe, 0);
> > > +
> > >      return 0;
> > >  }
> > >  
> > > diff --git a/hw/rdma/rdma_rm.h b/hw/rdma/rdma_rm.h
> > > index 9ec87d2667..a527d306f6 100644
> > > --- a/hw/rdma/rdma_rm.h
> > > +++ b/hw/rdma/rdma_rm.h
> > > @@ -20,6 +20,7 @@
> > >  #include "rdma_backend_defs.h"
> > >  #include "rdma_rm_defs.h"
> > >  
> > > +void rdma_dump_device_counters(Monitor *mon, RdmaDeviceResources *dev_res);
> > >  int rdma_rm_init(RdmaDeviceResources *dev_res, struct ibv_device_attr *dev_attr,
> > >                   Error **errp);
> > >  void rdma_rm_fini(RdmaDeviceResources *dev_res, RdmaBackendDev *backend_dev,
> > > diff --git a/hw/rdma/rdma_rm_defs.h b/hw/rdma/rdma_rm_defs.h
> > > index f0ee1f3072..4b8d704cfe 100644
> > > --- a/hw/rdma/rdma_rm_defs.h
> > > +++ b/hw/rdma/rdma_rm_defs.h
> > > @@ -34,7 +34,9 @@
> > >  #define MAX_QP_INIT_RD_ATOM   16
> > >  #define MAX_AH                64
> > >  
> > > -#define MAX_RM_TBL_NAME 16
> > > +#define MAX_RM_TBL_NAME             16
> > > +#define MAX_CONSEQ_EMPTY_POLL_CQ    4096 /* considered as error above this */
> > > +
> > >  typedef struct RdmaRmResTbl {
> > >      char name[MAX_RM_TBL_NAME];
> > >      QemuMutex lock;
> > > @@ -42,6 +44,7 @@ typedef struct RdmaRmResTbl {
> > >      size_t tbl_sz;
> > >      size_t res_sz;
> > >      void *tbl;
> > > +    uint32_t used; /* number of used entries in the table */
> > >  } RdmaRmResTbl;
> > >  
> > >  typedef struct RdmaRmPD {
> > > @@ -96,6 +99,27 @@ typedef struct RdmaRmPort {
> > >      enum ibv_port_state state;
> > >  } RdmaRmPort;
> > >  
> > > +typedef struct RdmaRmStats {
> > > +    uint64_t tx;
> > > +    uint64_t tx_len;
> > > +    uint64_t tx_err;
> > > +    uint64_t rx_bufs;
> > > +    uint64_t rx_bufs_len;
> > > +    uint64_t rx_bufs_err;
> > > +    uint64_t completions;
> > > +    uint64_t mad_tx;
> > > +    uint64_t mad_tx_err;
> > > +    uint64_t mad_rx;
> > > +    uint64_t mad_rx_err;
> > > +    uint64_t mad_rx_bufs;
> > > +    uint64_t mad_rx_bufs_err;
> > > +    uint64_t poll_cq_from_bk;
> > > +    uint64_t poll_cq_from_guest;
> > > +    uint64_t poll_cq_from_guest_empty;
> > > +    uint64_t poll_cq_ppoll_to;
> > > +    uint32_t missing_cqe;
> > > +} RdmaRmStats;
> > > +
> > >  typedef struct RdmaDeviceResources {
> > >      RdmaRmPort port;
> > >      RdmaRmResTbl pd_tbl;
> > > @@ -106,6 +130,7 @@ typedef struct RdmaDeviceResources {
> > >      RdmaRmResTbl cqe_ctx_tbl;
> > >      GHashTable *qp_hash; /* Keeps mapping between real and emulated */
> > >      QemuMutex lock;
> > > +    RdmaRmStats stats;
> > >  } RdmaDeviceResources;
> > >  
> > >  #endif
> > > diff --git a/hw/rdma/vmw/pvrdma.h b/hw/rdma/vmw/pvrdma.h
> > > index 0879224957..167706ec2c 100644
> > > --- a/hw/rdma/vmw/pvrdma.h
> > > +++ b/hw/rdma/vmw/pvrdma.h
> > > @@ -70,6 +70,10 @@ typedef struct DSRInfo {
> > >      PvrdmaRing cq;
> > >  } DSRInfo;
> > >  
> > > +typedef struct PVRDMADevStats {
> > > +    uint64_t commands;
> > > +} PVRDMADevStats;
> > > +
> > >  typedef struct PVRDMADev {
> > >      PCIDevice parent_obj;
> > >      MemoryRegion msix;
> > > @@ -89,6 +93,7 @@ typedef struct PVRDMADev {
> > >      CharBackend mad_chr;
> > >      VMXNET3State *func0;
> > >      Notifier shutdown_notifier;
> > > +    PVRDMADevStats stats;
> > >  } PVRDMADev;
> > >  #define PVRDMA_DEV(dev) OBJECT_CHECK(PVRDMADev, (dev), PVRDMA_HW_NAME)
> > >  
> > > diff --git a/hw/rdma/vmw/pvrdma_main.c b/hw/rdma/vmw/pvrdma_main.c
> > > index b6061f4b6e..659331ac93 100644
> > > --- a/hw/rdma/vmw/pvrdma_main.c
> > > +++ b/hw/rdma/vmw/pvrdma_main.c
> > > @@ -25,6 +25,8 @@
> > >  #include "cpu.h"
> > >  #include "trace.h"
> > >  #include "sysemu/sysemu.h"
> > > +#include "monitor/monitor.h"
> > > +#include "hw/rdma/rdma_hmp.h"
> > >  
> > >  #include "../rdma_rm.h"
> > >  #include "../rdma_backend.h"
> > > @@ -55,6 +57,18 @@ static Property pvrdma_dev_properties[] = {
> > >      DEFINE_PROP_END_OF_LIST(),
> > >  };
> > >  
> > > +static void pvrdma_print_statistics(Monitor *mon, RdmaStatsProvider *obj)
> > > +{
> > > +    PVRDMADev *dev = PVRDMA_DEV(obj);
> > > +    PCIDevice *pdev = PCI_DEVICE(dev);
> > > +
> > > +    monitor_printf(mon, "%s, %x.%x\n", pdev->name, PCI_SLOT(pdev->devfn),
> > > +                   PCI_FUNC(pdev->devfn));
> > > +    monitor_printf(mon, "\tcommands         : %" PRId64 "\n",
> > > +                   dev->stats.commands);
> > > +    rdma_dump_device_counters(mon, &dev->rdma_dev_res);
> > > +}
> > > +
> > >  static void free_dev_ring(PCIDevice *pci_dev, PvrdmaRing *ring,
> > >                            void *ring_state)
> > >  {
> > > @@ -394,6 +408,7 @@ static void pvrdma_regs_write(void *opaque, hwaddr addr, uint64_t val,
> > >          if (val == 0) {
> > >              trace_pvrdma_regs_write(addr, val, "REQUEST", "");
> > >              pvrdma_exec_cmd(dev);
> > > +            dev->stats.commands++;
> > >          }
> > >          break;
> > >      default:
> > > @@ -612,6 +627,8 @@ static void pvrdma_realize(PCIDevice *pdev, Error **errp)
> > >          goto out;
> > >      }
> > >  
> > > +    memset(&dev->stats, 0, sizeof(dev->stats));
> > > +
> > >      dev->shutdown_notifier.notify = pvrdma_shutdown_notifier;
> > >      qemu_register_shutdown_notifier(&dev->shutdown_notifier);
> > >  
> > > @@ -631,6 +648,7 @@ static void pvrdma_class_init(ObjectClass *klass, void *data)
> > >  {
> > >      DeviceClass *dc = DEVICE_CLASS(klass);
> > >      PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
> > > +    RdmaStatsProviderClass *ir = RDMA_STATS_PROVIDER_CLASS(klass);
> > >  
> > >      k->realize = pvrdma_realize;
> > >      k->exit = pvrdma_exit;
> > > @@ -642,6 +660,8 @@ static void pvrdma_class_init(ObjectClass *klass, void *data)
> > >      dc->desc = "RDMA Device";
> > >      dc->props = pvrdma_dev_properties;
> > >      set_bit(DEVICE_CATEGORY_NETWORK, dc->categories);
> > > +
> > > +    ir->print_statistics = pvrdma_print_statistics;
> > >  }
> > >  
> > >  static const TypeInfo pvrdma_info = {
> > > @@ -651,6 +671,7 @@ static const TypeInfo pvrdma_info = {
> > >      .class_init = pvrdma_class_init,
> > >      .interfaces = (InterfaceInfo[]) {
> > >          { INTERFACE_CONVENTIONAL_PCI_DEVICE },
> > > +        { TYPE_RDMA_STATS_PROVIDER },
> > >          { }
> > >      }
> > >  };
> > > diff --git a/include/hw/rdma/rdma_hmp.h b/include/hw/rdma/rdma_hmp.h
> > > new file mode 100644
> > > index 0000000000..dd23f2bc84
> > > --- /dev/null
> > > +++ b/include/hw/rdma/rdma_hmp.h
> > > @@ -0,0 +1,40 @@
> > > +/*
> > > + * RDMA device: Human Monitor interface
> > > + *
> > > + * Copyright (C) 2019 Oracle
> > > + * Copyright (C) 2019 Red Hat Inc
> > > + *
> > > + * Authors:
> > > + *     Yuval Shaia <yuval.shaia@oracle.com>
> > > + *
> > > + * This work is licensed under the terms of the GNU GPL, version 2 or later.
> > > + * See the COPYING file in the top-level directory.
> > > + *
> > > + */
> > > +
> > > +#ifndef RDMA_HMP_H
> > > +#define RDMA_HMP_H
> > > +
> > > +#include "qom/object.h"
> > > +
> > > +#define TYPE_RDMA_STATS_PROVIDER "rdma"
> > > +
> > > +#define RDMA_STATS_PROVIDER_CLASS(klass) \
> > > +    OBJECT_CLASS_CHECK(RdmaStatsProviderClass, (klass), \
> > > +                       TYPE_RDMA_STATS_PROVIDER)
> > > +#define RDMA_STATS_PROVIDER_GET_CLASS(obj) \
> > > +    OBJECT_GET_CLASS(RdmaStatsProviderClass, (obj), \
> > > +                     TYPE_RDMA_STATS_PROVIDER)
> > > +#define RDMA_STATS_PROVIDER(obj) \
> > > +    INTERFACE_CHECK(RdmaStatsProvider, (obj), \
> > > +                    TYPE_RDMA_STATS_PROVIDER)
> > > +
> > > +typedef struct RdmaStatsProvider RdmaStatsProvider;
> > > +
> > > +typedef struct RdmaStatsProviderClass {
> > > +    InterfaceClass parent;
> > > +
> > > +    void (*print_statistics)(Monitor *mon, RdmaStatsProvider *obj);
> > > +} RdmaStatsProviderClass;
> > > +
> > > +#endif
> > > -- 
> > > 2.17.2
> > > 
> > > 
> --
> Dr. David Alan Gilbert / dgilbert@redhat.com / Manchester, UK

Re: [Qemu-devel] [PATCH v4 4/9] {hmp, hw/pvrdma}: Expose device internals via monitor interface

[Marcel Apfelbaum]

Hi Yuval,

On 3/3/19 10:33 PM, Yuval Shaia wrote:
> Allow interrogating device internals through HMP interface.
> The exposed indicators can be used for troubleshooting by developers or
> sysadmin.
> There is no need to expose these attributes to a management system (e.x.
> libvirt) because (1) most of them are not "device-management' related
> info and (2) there is no guarantee the interface is stable.
>
> Signed-off-by: Yuval Shaia <yuval.shaia@oracle.com>
> ---
>   hmp-commands-info.hx       | 14 ++++++++
>   hmp.c                      | 27 +++++++++++++++
>   hmp.h                      |  1 +
>   hw/rdma/Makefile.objs      |  2 +-
>   hw/rdma/rdma_backend.c     | 70 +++++++++++++++++++++++++++++---------
>   hw/rdma/rdma_hmp.c         | 30 ++++++++++++++++
>   hw/rdma/rdma_rm.c          | 60 ++++++++++++++++++++++++++++++++
>   hw/rdma/rdma_rm.h          |  1 +
>   hw/rdma/rdma_rm_defs.h     | 27 ++++++++++++++-
>   hw/rdma/vmw/pvrdma.h       |  5 +++
>   hw/rdma/vmw/pvrdma_main.c  | 21 ++++++++++++
>   include/hw/rdma/rdma_hmp.h | 40 ++++++++++++++++++++++
>   12 files changed, 279 insertions(+), 19 deletions(-)
>   create mode 100644 hw/rdma/rdma_hmp.c
>   create mode 100644 include/hw/rdma/rdma_hmp.h
>
> diff --git a/hmp-commands-info.hx b/hmp-commands-info.hx
> index cbee8b944d..c59444c461 100644
> --- a/hmp-commands-info.hx
> +++ b/hmp-commands-info.hx
> @@ -202,6 +202,20 @@ STEXI
>   @item info pic
>   @findex info pic
>   Show PIC state.
> +ETEXI
> +
> +    {
> +        .name       = "rdma",
> +        .args_type  = "",
> +        .params     = "",
> +        .help       = "show RDMA state",
> +        .cmd        = hmp_info_rdma,
> +    },
> +
> +STEXI
> +@item info rdma
> +@findex info rdma
> +Show RDMA state.
>   ETEXI
>   
>       {
> diff --git a/hmp.c b/hmp.c
> index 1e006eeb49..68511b2441 100644
> --- a/hmp.c
> +++ b/hmp.c
> @@ -51,6 +51,7 @@
>   #include "qemu/error-report.h"
>   #include "exec/ramlist.h"
>   #include "hw/intc/intc.h"
> +#include "hw/rdma/rdma_hmp.h"
>   #include "migration/snapshot.h"
>   #include "migration/misc.h"
>   
> @@ -968,6 +969,32 @@ void hmp_info_pic(Monitor *mon, const QDict *qdict)
>                                      hmp_info_pic_foreach, mon);
>   }
>   
> +static int hmp_info_rdma_foreach(Object *obj, void *opaque)
> +{
> +    RdmaStatsProvider *rdma;
> +    RdmaStatsProviderClass *k;
> +    Monitor *mon = opaque;
> +
> +    if (object_dynamic_cast(obj, TYPE_RDMA_STATS_PROVIDER)) {
> +        rdma = RDMA_STATS_PROVIDER(obj);
> +        k = RDMA_STATS_PROVIDER_GET_CLASS(obj);
> +        if (k->print_statistics) {
> +            k->print_statistics(mon, rdma);
> +        } else {
> +            monitor_printf(mon, "RDMA statistics not available for %s.\n",
> +                           object_get_typename(obj));
> +        }
> +    }
> +
> +    return 0;
> +}
> +
> +void hmp_info_rdma(Monitor *mon, const QDict *qdict)
> +{
> +    object_child_foreach_recursive(object_get_root(),
> +                                   hmp_info_rdma_foreach, mon);
> +}
> +
>   void hmp_info_pci(Monitor *mon, const QDict *qdict)
>   {
>       PciInfoList *info_list, *info;
> diff --git a/hmp.h b/hmp.h
> index 5f1addcca2..666949afc3 100644
> --- a/hmp.h
> +++ b/hmp.h
> @@ -36,6 +36,7 @@ void hmp_info_spice(Monitor *mon, const QDict *qdict);
>   void hmp_info_balloon(Monitor *mon, const QDict *qdict);
>   void hmp_info_irq(Monitor *mon, const QDict *qdict);
>   void hmp_info_pic(Monitor *mon, const QDict *qdict);
> +void hmp_info_rdma(Monitor *mon, const QDict *qdict);
>   void hmp_info_pci(Monitor *mon, const QDict *qdict);
>   void hmp_info_block_jobs(Monitor *mon, const QDict *qdict);
>   void hmp_info_tpm(Monitor *mon, const QDict *qdict);
> diff --git a/hw/rdma/Makefile.objs b/hw/rdma/Makefile.objs
> index bd36cbf51c..dd59faff51 100644
> --- a/hw/rdma/Makefile.objs
> +++ b/hw/rdma/Makefile.objs
> @@ -1,5 +1,5 @@
>   ifeq ($(CONFIG_PVRDMA),y)
> -obj-$(CONFIG_PCI) += rdma_utils.o rdma_backend.o rdma_rm.o
> +obj-$(CONFIG_PCI) += rdma_utils.o rdma_backend.o rdma_rm.o rdma_hmp.o
>   obj-$(CONFIG_PCI) += vmw/pvrdma_dev_ring.o vmw/pvrdma_cmd.o \
>                        vmw/pvrdma_qp_ops.o vmw/pvrdma_main.o
>   endif
> diff --git a/hw/rdma/rdma_backend.c b/hw/rdma/rdma_backend.c
> index 9679b842d1..bc2fefcf93 100644
> --- a/hw/rdma/rdma_backend.c
> +++ b/hw/rdma/rdma_backend.c
> @@ -64,9 +64,9 @@ static inline void complete_work(enum ibv_wc_status status, uint32_t vendor_err,
>       comp_handler(ctx, &wc);
>   }
>   
> -static void rdma_poll_cq(RdmaDeviceResources *rdma_dev_res, struct ibv_cq *ibcq)
> +static int rdma_poll_cq(RdmaDeviceResources *rdma_dev_res, struct ibv_cq *ibcq)
>   {
> -    int i, ne;
> +    int i, ne, total_ne = 0;
>       BackendCtx *bctx;
>       struct ibv_wc wc[2];
>   
> @@ -89,12 +89,18 @@ static void rdma_poll_cq(RdmaDeviceResources *rdma_dev_res, struct ibv_cq *ibcq)
>               rdma_rm_dealloc_cqe_ctx(rdma_dev_res, wc[i].wr_id);
>               g_free(bctx);
>           }
> +        total_ne += ne;
>       } while (ne > 0);
> +    atomic_sub(&rdma_dev_res->stats.missing_cqe, total_ne);
>       qemu_mutex_unlock(&rdma_dev_res->lock);
>   
>       if (ne < 0) {
>           rdma_error_report("ibv_poll_cq fail, rc=%d, errno=%d", ne, errno);
>       }
> +
> +    rdma_dev_res->stats.completions += total_ne;
> +
> +    return total_ne;
>   }
>   
>   static void *comp_handler_thread(void *arg)
> @@ -122,6 +128,9 @@ static void *comp_handler_thread(void *arg)
>       while (backend_dev->comp_thread.run) {
>           do {
>               rc = qemu_poll_ns(pfds, 1, THR_POLL_TO * (int64_t)SCALE_MS);
> +            if (!rc) {
> +                backend_dev->rdma_dev_res->stats.poll_cq_ppoll_to++;
> +            }
>           } while (!rc && backend_dev->comp_thread.run);
>   
>           if (backend_dev->comp_thread.run) {
> @@ -138,6 +147,7 @@ static void *comp_handler_thread(void *arg)
>                                     errno);
>               }
>   
> +            backend_dev->rdma_dev_res->stats.poll_cq_from_bk++;
>               rdma_poll_cq(backend_dev->rdma_dev_res, ev_cq);
>   
>               ibv_ack_cq_events(ev_cq, 1);
> @@ -271,7 +281,13 @@ int rdma_backend_query_port(RdmaBackendDev *backend_dev,
>   
>   void rdma_backend_poll_cq(RdmaDeviceResources *rdma_dev_res, RdmaBackendCQ *cq)
>   {
> -    rdma_poll_cq(rdma_dev_res, cq->ibcq);
> +    int polled;
> +
> +    rdma_dev_res->stats.poll_cq_from_guest++;
> +    polled = rdma_poll_cq(rdma_dev_res, cq->ibcq);
> +    if (!polled) {
> +        rdma_dev_res->stats.poll_cq_from_guest_empty++;
> +    }
>   }
>   
>   static GHashTable *ah_hash;
> @@ -333,7 +349,7 @@ static void ah_cache_init(void)
>   
>   static int build_host_sge_array(RdmaDeviceResources *rdma_dev_res,
>                                   struct ibv_sge *dsge, struct ibv_sge *ssge,
> -                                uint8_t num_sge)
> +                                uint8_t num_sge, uint64_t *total_length)
>   {
>       RdmaRmMR *mr;
>       int ssge_idx;
> @@ -349,6 +365,8 @@ static int build_host_sge_array(RdmaDeviceResources *rdma_dev_res,
>           dsge->length = ssge[ssge_idx].length;
>           dsge->lkey = rdma_backend_mr_lkey(&mr->backend_mr);
>   
> +        *total_length += dsge->length;
> +
>           dsge++;
>       }
>   
> @@ -445,8 +463,10 @@ void rdma_backend_post_send(RdmaBackendDev *backend_dev,
>               rc = mad_send(backend_dev, sgid_idx, sgid, sge, num_sge);
>               if (rc) {
>                   complete_work(IBV_WC_GENERAL_ERR, VENDOR_ERR_MAD_SEND, ctx);
> +                backend_dev->rdma_dev_res->stats.mad_tx_err++;
>               } else {
>                   complete_work(IBV_WC_SUCCESS, 0, ctx);
> +                backend_dev->rdma_dev_res->stats.mad_tx++;
>               }
>           }
>           return;
> @@ -458,20 +478,21 @@ void rdma_backend_post_send(RdmaBackendDev *backend_dev,
>       rc = rdma_rm_alloc_cqe_ctx(backend_dev->rdma_dev_res, &bctx_id, bctx);
>       if (unlikely(rc)) {
>           complete_work(IBV_WC_GENERAL_ERR, VENDOR_ERR_NOMEM, ctx);
> -        goto out_free_bctx;
> +        goto err_free_bctx;
>       }
>   
> -    rc = build_host_sge_array(backend_dev->rdma_dev_res, new_sge, sge, num_sge);
> +    rc = build_host_sge_array(backend_dev->rdma_dev_res, new_sge, sge, num_sge,
> +                              &backend_dev->rdma_dev_res->stats.tx_len);
>       if (rc) {
>           complete_work(IBV_WC_GENERAL_ERR, rc, ctx);
> -        goto out_dealloc_cqe_ctx;
> +        goto err_dealloc_cqe_ctx;
>       }
>   
>       if (qp_type == IBV_QPT_UD) {
>           wr.wr.ud.ah = create_ah(backend_dev, qp->ibpd, sgid_idx, dgid);
>           if (!wr.wr.ud.ah) {
>               complete_work(IBV_WC_GENERAL_ERR, VENDOR_ERR_FAIL_BACKEND, ctx);
> -            goto out_dealloc_cqe_ctx;
> +            goto err_dealloc_cqe_ctx;
>           }
>           wr.wr.ud.remote_qpn = dqpn;
>           wr.wr.ud.remote_qkey = dqkey;
> @@ -488,15 +509,19 @@ void rdma_backend_post_send(RdmaBackendDev *backend_dev,
>           rdma_error_report("ibv_post_send fail, qpn=0x%x, rc=%d, errno=%d",
>                             qp->ibqp->qp_num, rc, errno);
>           complete_work(IBV_WC_GENERAL_ERR, VENDOR_ERR_FAIL_BACKEND, ctx);
> -        goto out_dealloc_cqe_ctx;
> +        goto err_dealloc_cqe_ctx;
>       }
>   
> +    atomic_inc(&backend_dev->rdma_dev_res->stats.missing_cqe);
> +    backend_dev->rdma_dev_res->stats.tx++;
> +
>       return;
>   
> -out_dealloc_cqe_ctx:
> +err_dealloc_cqe_ctx:
> +    backend_dev->rdma_dev_res->stats.tx_err++;
>       rdma_rm_dealloc_cqe_ctx(backend_dev->rdma_dev_res, bctx_id);
>   
> -out_free_bctx:
> +err_free_bctx:
>       g_free(bctx);
>   }
>   
> @@ -554,6 +579,9 @@ void rdma_backend_post_recv(RdmaBackendDev *backend_dev,
>               rc = save_mad_recv_buffer(backend_dev, sge, num_sge, ctx);
>               if (rc) {
>                   complete_work(IBV_WC_GENERAL_ERR, rc, ctx);
> +                rdma_dev_res->stats.mad_rx_bufs_err++;
> +            } else {
> +                rdma_dev_res->stats.mad_rx_bufs++;
>               }
>           }
>           return;
> @@ -565,13 +593,14 @@ void rdma_backend_post_recv(RdmaBackendDev *backend_dev,
>       rc = rdma_rm_alloc_cqe_ctx(rdma_dev_res, &bctx_id, bctx);
>       if (unlikely(rc)) {
>           complete_work(IBV_WC_GENERAL_ERR, VENDOR_ERR_NOMEM, ctx);
> -        goto out_free_bctx;
> +        goto err_free_bctx;
>       }
>   
> -    rc = build_host_sge_array(rdma_dev_res, new_sge, sge, num_sge);
> +    rc = build_host_sge_array(rdma_dev_res, new_sge, sge, num_sge,
> +                              &backend_dev->rdma_dev_res->stats.rx_bufs_len);
>       if (rc) {
>           complete_work(IBV_WC_GENERAL_ERR, rc, ctx);
> -        goto out_dealloc_cqe_ctx;
> +        goto err_dealloc_cqe_ctx;
>       }
>   
>       wr.num_sge = num_sge;
> @@ -582,15 +611,19 @@ void rdma_backend_post_recv(RdmaBackendDev *backend_dev,
>           rdma_error_report("ibv_post_recv fail, qpn=0x%x, rc=%d, errno=%d",
>                             qp->ibqp->qp_num, rc, errno);
>           complete_work(IBV_WC_GENERAL_ERR, VENDOR_ERR_FAIL_BACKEND, ctx);
> -        goto out_dealloc_cqe_ctx;
> +        goto err_dealloc_cqe_ctx;
>       }
>   
> +    atomic_inc(&backend_dev->rdma_dev_res->stats.missing_cqe);
> +    rdma_dev_res->stats.rx_bufs++;
> +
>       return;
>   
> -out_dealloc_cqe_ctx:
> +err_dealloc_cqe_ctx:
> +    backend_dev->rdma_dev_res->stats.rx_bufs_err++;
>       rdma_rm_dealloc_cqe_ctx(rdma_dev_res, bctx_id);
>   
> -out_free_bctx:
> +err_free_bctx:
>       g_free(bctx);
>   }
>   
> @@ -929,12 +962,14 @@ static void process_incoming_mad_req(RdmaBackendDev *backend_dev,
>       bctx = rdma_rm_get_cqe_ctx(backend_dev->rdma_dev_res, cqe_ctx_id);
>       if (unlikely(!bctx)) {
>           rdma_error_report("No matching ctx for req %ld", cqe_ctx_id);
> +        backend_dev->rdma_dev_res->stats.mad_rx_err++;
>           return;
>       }
>   
>       mad = rdma_pci_dma_map(backend_dev->dev, bctx->sge.addr,
>                              bctx->sge.length);
>       if (!mad || bctx->sge.length < msg->umad_len + MAD_HDR_SIZE) {
> +        backend_dev->rdma_dev_res->stats.mad_rx_err++;
>           complete_work(IBV_WC_GENERAL_ERR, VENDOR_ERR_INV_MAD_BUFF,
>                         bctx->up_ctx);
>       } else {
> @@ -949,6 +984,7 @@ static void process_incoming_mad_req(RdmaBackendDev *backend_dev,
>           wc.byte_len = msg->umad_len;
>           wc.status = IBV_WC_SUCCESS;
>           wc.wc_flags = IBV_WC_GRH;
> +        backend_dev->rdma_dev_res->stats.mad_rx++;
>           comp_handler(bctx->up_ctx, &wc);
>       }
>   
> diff --git a/hw/rdma/rdma_hmp.c b/hw/rdma/rdma_hmp.c
> new file mode 100644
> index 0000000000..c5814473c5
> --- /dev/null
> +++ b/hw/rdma/rdma_hmp.c
> @@ -0,0 +1,30 @@
> +/*
> + * RDMA device: Human Monitor interface
> + *
> + * Copyright (C) 2018 Oracle
> + * Copyright (C) 2018 Red Hat Inc
> + *
> + * Authors:
> + *     Yuval Shaia <yuval.shaia@oracle.com>
> + *
> + * This work is licensed under the terms of the GNU GPL, version 2 or later.
> + * See the COPYING file in the top-level directory.
> + *
> + */
> +
> +#include "qemu/osdep.h"
> +#include "hw/rdma/rdma_hmp.h"
> +#include "qemu/module.h"
> +
> +static const TypeInfo rdma_hmp_info = {
> +    .name = TYPE_RDMA_STATS_PROVIDER,
> +    .parent = TYPE_INTERFACE,
> +    .class_size = sizeof(RdmaStatsProviderClass),
> +};
> +
> +static void rdma_hmp_register_types(void)
> +{
> +    type_register_static(&rdma_hmp_info);
> +}
> +
> +type_init(rdma_hmp_register_types)
> diff --git a/hw/rdma/rdma_rm.c b/hw/rdma/rdma_rm.c
> index 14580ca379..e019de1a14 100644
> --- a/hw/rdma/rdma_rm.c
> +++ b/hw/rdma/rdma_rm.c
> @@ -16,6 +16,7 @@
>   #include "qemu/osdep.h"
>   #include "qapi/error.h"
>   #include "cpu.h"
> +#include "monitor/monitor.h"
>   
>   #include "trace.h"
>   #include "rdma_utils.h"
> @@ -26,6 +27,58 @@
>   #define PG_DIR_SZ { TARGET_PAGE_SIZE / sizeof(__u64) }
>   #define PG_TBL_SZ { TARGET_PAGE_SIZE / sizeof(__u64) }
>   
> +void rdma_dump_device_counters(Monitor *mon, RdmaDeviceResources *dev_res)
> +{
> +    monitor_printf(mon, "\ttx               : %" PRId64 "\n",
> +                   dev_res->stats.tx);
> +    monitor_printf(mon, "\ttx_len           : %" PRId64 "\n",
> +                   dev_res->stats.tx_len);
> +    monitor_printf(mon, "\ttx_err           : %" PRId64 "\n",
> +                   dev_res->stats.tx_err);
> +    monitor_printf(mon, "\trx_bufs          : %" PRId64 "\n",
> +                   dev_res->stats.rx_bufs);
> +    monitor_printf(mon, "\trx_bufs_len      : %" PRId64 "\n",
> +                   dev_res->stats.rx_bufs_len);
> +    monitor_printf(mon, "\trx_bufs_err      : %" PRId64 "\n",
> +                   dev_res->stats.rx_bufs_err);
> +    monitor_printf(mon, "\tcomps            : %" PRId64 "\n",
> +                   dev_res->stats.completions);
> +    monitor_printf(mon, "\tmissing_comps    : %" PRId32 "\n",
> +                   dev_res->stats.missing_cqe);
> +    monitor_printf(mon, "\tpoll_cq (bk)     : %" PRId64 "\n",
> +                   dev_res->stats.poll_cq_from_bk);
> +    monitor_printf(mon, "\tpoll_cq_ppoll_to : %" PRId64 "\n",
> +                   dev_res->stats.poll_cq_ppoll_to);
> +    monitor_printf(mon, "\tpoll_cq (fe)     : %" PRId64 "\n",
> +                   dev_res->stats.poll_cq_from_guest);
> +    monitor_printf(mon, "\tpoll_cq_empty    : %" PRId64 "\n",
> +                   dev_res->stats.poll_cq_from_guest_empty);
> +    monitor_printf(mon, "\tmad_tx           : %" PRId64 "\n",
> +                   dev_res->stats.mad_tx);
> +    monitor_printf(mon, "\tmad_tx_err       : %" PRId64 "\n",
> +                   dev_res->stats.mad_tx_err);
> +    monitor_printf(mon, "\tmad_rx           : %" PRId64 "\n",
> +                   dev_res->stats.mad_rx);
> +    monitor_printf(mon, "\tmad_rx_err       : %" PRId64 "\n",
> +                   dev_res->stats.mad_rx_err);
> +    monitor_printf(mon, "\tmad_rx_bufs      : %" PRId64 "\n",
> +                   dev_res->stats.mad_rx_bufs);
> +    monitor_printf(mon, "\tmad_rx_bufs_err  : %" PRId64 "\n",
> +                   dev_res->stats.mad_rx_bufs_err);
> +    monitor_printf(mon, "\tPDs              : %" PRId32 "\n",
> +                   dev_res->pd_tbl.used);
> +    monitor_printf(mon, "\tMRs              : %" PRId32 "\n",
> +                   dev_res->mr_tbl.used);
> +    monitor_printf(mon, "\tUCs              : %" PRId32 "\n",
> +                   dev_res->uc_tbl.used);
> +    monitor_printf(mon, "\tQPs              : %" PRId32 "\n",
> +                   dev_res->qp_tbl.used);
> +    monitor_printf(mon, "\tCQs              : %" PRId32 "\n",
> +                   dev_res->cq_tbl.used);
> +    monitor_printf(mon, "\tCEQ_CTXs         : %" PRId32 "\n",
> +                   dev_res->cqe_ctx_tbl.used);
> +}
> +
>   static inline void res_tbl_init(const char *name, RdmaRmResTbl *tbl,
>                                   uint32_t tbl_sz, uint32_t res_sz)
>   {
> @@ -37,6 +90,7 @@ static inline void res_tbl_init(const char *name, RdmaRmResTbl *tbl,
>       tbl->bitmap = bitmap_new(tbl_sz);
>       tbl->tbl_sz = tbl_sz;
>       tbl->res_sz = res_sz;
> +    tbl->used = 0;
>       qemu_mutex_init(&tbl->lock);
>   }
>   
> @@ -76,6 +130,8 @@ static inline void *rdma_res_tbl_alloc(RdmaRmResTbl *tbl, uint32_t *handle)
>   
>       set_bit(*handle, tbl->bitmap);
>   
> +    tbl->used++;
> +
>       qemu_mutex_unlock(&tbl->lock);
>   
>       memset(tbl->tbl + *handle * tbl->res_sz, 0, tbl->res_sz);
> @@ -93,6 +149,7 @@ static inline void rdma_res_tbl_dealloc(RdmaRmResTbl *tbl, uint32_t handle)
>   
>       if (handle < tbl->tbl_sz) {
>           clear_bit(handle, tbl->bitmap);
> +        tbl->used--;
>       }
>   
>       qemu_mutex_unlock(&tbl->lock);
> @@ -620,6 +677,9 @@ int rdma_rm_init(RdmaDeviceResources *dev_res, struct ibv_device_attr *dev_attr,
>   
>       qemu_mutex_init(&dev_res->lock);
>   
> +    memset(&dev_res->stats, 0, sizeof(dev_res->stats));
> +    atomic_set(&dev_res->stats.missing_cqe, 0);
> +
>       return 0;
>   }
>   
> diff --git a/hw/rdma/rdma_rm.h b/hw/rdma/rdma_rm.h
> index 9ec87d2667..a527d306f6 100644
> --- a/hw/rdma/rdma_rm.h
> +++ b/hw/rdma/rdma_rm.h
> @@ -20,6 +20,7 @@
>   #include "rdma_backend_defs.h"
>   #include "rdma_rm_defs.h"
>   
> +void rdma_dump_device_counters(Monitor *mon, RdmaDeviceResources *dev_res);
>   int rdma_rm_init(RdmaDeviceResources *dev_res, struct ibv_device_attr *dev_attr,
>                    Error **errp);
>   void rdma_rm_fini(RdmaDeviceResources *dev_res, RdmaBackendDev *backend_dev,
> diff --git a/hw/rdma/rdma_rm_defs.h b/hw/rdma/rdma_rm_defs.h
> index f0ee1f3072..4b8d704cfe 100644
> --- a/hw/rdma/rdma_rm_defs.h
> +++ b/hw/rdma/rdma_rm_defs.h
> @@ -34,7 +34,9 @@
>   #define MAX_QP_INIT_RD_ATOM   16
>   #define MAX_AH                64
>   
> -#define MAX_RM_TBL_NAME 16
> +#define MAX_RM_TBL_NAME             16
> +#define MAX_CONSEQ_EMPTY_POLL_CQ    4096 /* considered as error above this */
> +
>   typedef struct RdmaRmResTbl {
>       char name[MAX_RM_TBL_NAME];
>       QemuMutex lock;
> @@ -42,6 +44,7 @@ typedef struct RdmaRmResTbl {
>       size_t tbl_sz;
>       size_t res_sz;
>       void *tbl;
> +    uint32_t used; /* number of used entries in the table */
>   } RdmaRmResTbl;
>   
>   typedef struct RdmaRmPD {
> @@ -96,6 +99,27 @@ typedef struct RdmaRmPort {
>       enum ibv_port_state state;
>   } RdmaRmPort;
>   
> +typedef struct RdmaRmStats {
> +    uint64_t tx;
> +    uint64_t tx_len;
> +    uint64_t tx_err;
> +    uint64_t rx_bufs;
> +    uint64_t rx_bufs_len;
> +    uint64_t rx_bufs_err;
> +    uint64_t completions;
> +    uint64_t mad_tx;
> +    uint64_t mad_tx_err;
> +    uint64_t mad_rx;
> +    uint64_t mad_rx_err;
> +    uint64_t mad_rx_bufs;
> +    uint64_t mad_rx_bufs_err;
> +    uint64_t poll_cq_from_bk;
> +    uint64_t poll_cq_from_guest;
> +    uint64_t poll_cq_from_guest_empty;
> +    uint64_t poll_cq_ppoll_to;
> +    uint32_t missing_cqe;
> +} RdmaRmStats;
> +
>   typedef struct RdmaDeviceResources {
>       RdmaRmPort port;
>       RdmaRmResTbl pd_tbl;
> @@ -106,6 +130,7 @@ typedef struct RdmaDeviceResources {
>       RdmaRmResTbl cqe_ctx_tbl;
>       GHashTable *qp_hash; /* Keeps mapping between real and emulated */
>       QemuMutex lock;
> +    RdmaRmStats stats;
>   } RdmaDeviceResources;
>   
>   #endif
> diff --git a/hw/rdma/vmw/pvrdma.h b/hw/rdma/vmw/pvrdma.h
> index 0879224957..167706ec2c 100644
> --- a/hw/rdma/vmw/pvrdma.h
> +++ b/hw/rdma/vmw/pvrdma.h
> @@ -70,6 +70,10 @@ typedef struct DSRInfo {
>       PvrdmaRing cq;
>   } DSRInfo;
>   
> +typedef struct PVRDMADevStats {
> +    uint64_t commands;
> +} PVRDMADevStats;
> +
>   typedef struct PVRDMADev {
>       PCIDevice parent_obj;
>       MemoryRegion msix;
> @@ -89,6 +93,7 @@ typedef struct PVRDMADev {
>       CharBackend mad_chr;
>       VMXNET3State *func0;
>       Notifier shutdown_notifier;
> +    PVRDMADevStats stats;
>   } PVRDMADev;
>   #define PVRDMA_DEV(dev) OBJECT_CHECK(PVRDMADev, (dev), PVRDMA_HW_NAME)
>   
> diff --git a/hw/rdma/vmw/pvrdma_main.c b/hw/rdma/vmw/pvrdma_main.c
> index b6061f4b6e..659331ac93 100644
> --- a/hw/rdma/vmw/pvrdma_main.c
> +++ b/hw/rdma/vmw/pvrdma_main.c
> @@ -25,6 +25,8 @@
>   #include "cpu.h"
>   #include "trace.h"
>   #include "sysemu/sysemu.h"
> +#include "monitor/monitor.h"
> +#include "hw/rdma/rdma_hmp.h"
>   
>   #include "../rdma_rm.h"
>   #include "../rdma_backend.h"
> @@ -55,6 +57,18 @@ static Property pvrdma_dev_properties[] = {
>       DEFINE_PROP_END_OF_LIST(),
>   };
>   
> +static void pvrdma_print_statistics(Monitor *mon, RdmaStatsProvider *obj)
> +{
> +    PVRDMADev *dev = PVRDMA_DEV(obj);
> +    PCIDevice *pdev = PCI_DEVICE(dev);
> +
> +    monitor_printf(mon, "%s, %x.%x\n", pdev->name, PCI_SLOT(pdev->devfn),
> +                   PCI_FUNC(pdev->devfn));
> +    monitor_printf(mon, "\tcommands         : %" PRId64 "\n",
> +                   dev->stats.commands);
> +    rdma_dump_device_counters(mon, &dev->rdma_dev_res);
> +}
> +
>   static void free_dev_ring(PCIDevice *pci_dev, PvrdmaRing *ring,
>                             void *ring_state)
>   {
> @@ -394,6 +408,7 @@ static void pvrdma_regs_write(void *opaque, hwaddr addr, uint64_t val,
>           if (val == 0) {
>               trace_pvrdma_regs_write(addr, val, "REQUEST", "");
>               pvrdma_exec_cmd(dev);
> +            dev->stats.commands++;
>           }
>           break;
>       default:
> @@ -612,6 +627,8 @@ static void pvrdma_realize(PCIDevice *pdev, Error **errp)
>           goto out;
>       }
>   
> +    memset(&dev->stats, 0, sizeof(dev->stats));
> +
>       dev->shutdown_notifier.notify = pvrdma_shutdown_notifier;
>       qemu_register_shutdown_notifier(&dev->shutdown_notifier);
>   
> @@ -631,6 +648,7 @@ static void pvrdma_class_init(ObjectClass *klass, void *data)
>   {
>       DeviceClass *dc = DEVICE_CLASS(klass);
>       PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
> +    RdmaStatsProviderClass *ir = RDMA_STATS_PROVIDER_CLASS(klass);
>   
>       k->realize = pvrdma_realize;
>       k->exit = pvrdma_exit;
> @@ -642,6 +660,8 @@ static void pvrdma_class_init(ObjectClass *klass, void *data)
>       dc->desc = "RDMA Device";
>       dc->props = pvrdma_dev_properties;
>       set_bit(DEVICE_CATEGORY_NETWORK, dc->categories);
> +
> +    ir->print_statistics = pvrdma_print_statistics;
>   }
>   
>   static const TypeInfo pvrdma_info = {
> @@ -651,6 +671,7 @@ static const TypeInfo pvrdma_info = {
>       .class_init = pvrdma_class_init,
>       .interfaces = (InterfaceInfo[]) {
>           { INTERFACE_CONVENTIONAL_PCI_DEVICE },
> +        { TYPE_RDMA_STATS_PROVIDER },
>           { }
>       }
>   };
> diff --git a/include/hw/rdma/rdma_hmp.h b/include/hw/rdma/rdma_hmp.h
> new file mode 100644
> index 0000000000..dd23f2bc84
> --- /dev/null
> +++ b/include/hw/rdma/rdma_hmp.h
> @@ -0,0 +1,40 @@
> +/*
> + * RDMA device: Human Monitor interface
> + *
> + * Copyright (C) 2019 Oracle
> + * Copyright (C) 2019 Red Hat Inc
> + *
> + * Authors:
> + *     Yuval Shaia <yuval.shaia@oracle.com>
> + *
> + * This work is licensed under the terms of the GNU GPL, version 2 or later.
> + * See the COPYING file in the top-level directory.
> + *
> + */
> +
> +#ifndef RDMA_HMP_H
> +#define RDMA_HMP_H
> +
> +#include "qom/object.h"
> +
> +#define TYPE_RDMA_STATS_PROVIDER "rdma"
> +
> +#define RDMA_STATS_PROVIDER_CLASS(klass) \
> +    OBJECT_CLASS_CHECK(RdmaStatsProviderClass, (klass), \
> +                       TYPE_RDMA_STATS_PROVIDER)
> +#define RDMA_STATS_PROVIDER_GET_CLASS(obj) \
> +    OBJECT_GET_CLASS(RdmaStatsProviderClass, (obj), \
> +                     TYPE_RDMA_STATS_PROVIDER)
> +#define RDMA_STATS_PROVIDER(obj) \
> +    INTERFACE_CHECK(RdmaStatsProvider, (obj), \
> +                    TYPE_RDMA_STATS_PROVIDER)
> +
> +typedef struct RdmaStatsProvider RdmaStatsProvider;
> +
> +typedef struct RdmaStatsProviderClass {
> +    InterfaceClass parent;
> +
> +    void (*print_statistics)(Monitor *mon, RdmaStatsProvider *obj);
> +} RdmaStatsProviderClass;
> +
> +#endif


RDMAStats implementation looks ok to me.

Reviewed-by: Marcel Apfelbaum<marcel.apfelbaum@gmail.com>

Thanks,
Marcel

Re: [Qemu-devel] [PATCH v4 4/9] {hmp, hw/pvrdma}: Expose device internals via monitor interface

[Markus Armbruster]

Yuval Shaia <yuval.shaia@oracle.com> writes:

> Allow interrogating device internals through HMP interface.
> The exposed indicators can be used for troubleshooting by developers or
> sysadmin.
> There is no need to expose these attributes to a management system (e.x.
> libvirt) because (1) most of them are not "device-management' related
> info and (2) there is no guarantee the interface is stable.
>
> Signed-off-by: Yuval Shaia <yuval.shaia@oracle.com>
> ---
>  hmp-commands-info.hx       | 14 ++++++++
>  hmp.c                      | 27 +++++++++++++++
>  hmp.h                      |  1 +
>  hw/rdma/Makefile.objs      |  2 +-
>  hw/rdma/rdma_backend.c     | 70 +++++++++++++++++++++++++++++---------
>  hw/rdma/rdma_hmp.c         | 30 ++++++++++++++++
>  hw/rdma/rdma_rm.c          | 60 ++++++++++++++++++++++++++++++++
>  hw/rdma/rdma_rm.h          |  1 +
>  hw/rdma/rdma_rm_defs.h     | 27 ++++++++++++++-
>  hw/rdma/vmw/pvrdma.h       |  5 +++
>  hw/rdma/vmw/pvrdma_main.c  | 21 ++++++++++++
>  include/hw/rdma/rdma_hmp.h | 40 ++++++++++++++++++++++
>  12 files changed, 279 insertions(+), 19 deletions(-)
>  create mode 100644 hw/rdma/rdma_hmp.c
>  create mode 100644 include/hw/rdma/rdma_hmp.h
>
> diff --git a/hmp-commands-info.hx b/hmp-commands-info.hx
> index cbee8b944d..c59444c461 100644
> --- a/hmp-commands-info.hx
> +++ b/hmp-commands-info.hx
> @@ -202,6 +202,20 @@ STEXI
>  @item info pic
>  @findex info pic
>  Show PIC state.
> +ETEXI
> +
> +    {
> +        .name       = "rdma",
> +        .args_type  = "",
> +        .params     = "",
> +        .help       = "show RDMA state",
> +        .cmd        = hmp_info_rdma,
> +    },
> +
> +STEXI
> +@item info rdma
> +@findex info rdma
> +Show RDMA state.
>  ETEXI
>  
>      {
> diff --git a/hmp.c b/hmp.c
> index 1e006eeb49..68511b2441 100644
> --- a/hmp.c
> +++ b/hmp.c
> @@ -51,6 +51,7 @@
>  #include "qemu/error-report.h"
>  #include "exec/ramlist.h"
>  #include "hw/intc/intc.h"
> +#include "hw/rdma/rdma_hmp.h"
>  #include "migration/snapshot.h"
>  #include "migration/misc.h"
>  
> @@ -968,6 +969,32 @@ void hmp_info_pic(Monitor *mon, const QDict *qdict)
>                                     hmp_info_pic_foreach, mon);
>  }
>  
> +static int hmp_info_rdma_foreach(Object *obj, void *opaque)
> +{
> +    RdmaStatsProvider *rdma;
> +    RdmaStatsProviderClass *k;
> +    Monitor *mon = opaque;
> +
> +    if (object_dynamic_cast(obj, TYPE_RDMA_STATS_PROVIDER)) {
> +        rdma = RDMA_STATS_PROVIDER(obj);
> +        k = RDMA_STATS_PROVIDER_GET_CLASS(obj);
> +        if (k->print_statistics) {
> +            k->print_statistics(mon, rdma);
> +        } else {
> +            monitor_printf(mon, "RDMA statistics not available for %s.\n",
> +                           object_get_typename(obj));
> +        }
> +    }
> +
> +    return 0;
> +}
> +
> +void hmp_info_rdma(Monitor *mon, const QDict *qdict)
> +{
> +    object_child_foreach_recursive(object_get_root(),
> +                                   hmp_info_rdma_foreach, mon);
> +}
> +
>  void hmp_info_pci(Monitor *mon, const QDict *qdict)
>  {
>      PciInfoList *info_list, *info;
> diff --git a/hmp.h b/hmp.h
> index 5f1addcca2..666949afc3 100644
> --- a/hmp.h
> +++ b/hmp.h
> @@ -36,6 +36,7 @@ void hmp_info_spice(Monitor *mon, const QDict *qdict);
>  void hmp_info_balloon(Monitor *mon, const QDict *qdict);
>  void hmp_info_irq(Monitor *mon, const QDict *qdict);
>  void hmp_info_pic(Monitor *mon, const QDict *qdict);
> +void hmp_info_rdma(Monitor *mon, const QDict *qdict);
>  void hmp_info_pci(Monitor *mon, const QDict *qdict);
>  void hmp_info_block_jobs(Monitor *mon, const QDict *qdict);
>  void hmp_info_tpm(Monitor *mon, const QDict *qdict);

HMP stuff looks good to me.

> diff --git a/hw/rdma/Makefile.objs b/hw/rdma/Makefile.objs
> index bd36cbf51c..dd59faff51 100644
> --- a/hw/rdma/Makefile.objs
> +++ b/hw/rdma/Makefile.objs
> @@ -1,5 +1,5 @@
>  ifeq ($(CONFIG_PVRDMA),y)
> -obj-$(CONFIG_PCI) += rdma_utils.o rdma_backend.o rdma_rm.o
> +obj-$(CONFIG_PCI) += rdma_utils.o rdma_backend.o rdma_rm.o rdma_hmp.o
>  obj-$(CONFIG_PCI) += vmw/pvrdma_dev_ring.o vmw/pvrdma_cmd.o \
>                       vmw/pvrdma_qp_ops.o vmw/pvrdma_main.o
>  endif
> diff --git a/hw/rdma/rdma_backend.c b/hw/rdma/rdma_backend.c
> index 9679b842d1..bc2fefcf93 100644
> --- a/hw/rdma/rdma_backend.c
> +++ b/hw/rdma/rdma_backend.c

The changes to this file look like they're for gathering stats.  The
patch would be much simpler to review had you kept the stats gathering
separate.  I can only review the HMP / QOM part, and to do that, I have
to identify what to ignore.

[...]
> diff --git a/hw/rdma/rdma_hmp.c b/hw/rdma/rdma_hmp.c
> new file mode 100644
> index 0000000000..c5814473c5
> --- /dev/null
> +++ b/hw/rdma/rdma_hmp.c
> @@ -0,0 +1,30 @@
> +/*
> + * RDMA device: Human Monitor interface

The file name and this comment are a bit akward.  Yes, you create
TYPE_RDMA_STATS_PROVIDER for use in HMP info rdma, but there's
absolutely nothing HMP-related in this file.  Same for rdma_hmp.h below.

Call them rdma_stats.c and rdma_stats.h?

> + *
> + * Copyright (C) 2018 Oracle
> + * Copyright (C) 2018 Red Hat Inc
> + *
> + * Authors:
> + *     Yuval Shaia <yuval.shaia@oracle.com>
> + *
> + * This work is licensed under the terms of the GNU GPL, version 2 or later.
> + * See the COPYING file in the top-level directory.
> + *
> + */
> +
> +#include "qemu/osdep.h"
> +#include "hw/rdma/rdma_hmp.h"
> +#include "qemu/module.h"
> +
> +static const TypeInfo rdma_hmp_info = {
> +    .name = TYPE_RDMA_STATS_PROVIDER,
> +    .parent = TYPE_INTERFACE,
> +    .class_size = sizeof(RdmaStatsProviderClass),
> +};
> +
> +static void rdma_hmp_register_types(void)
> +{
> +    type_register_static(&rdma_hmp_info);
> +}
> +
> +type_init(rdma_hmp_register_types)

Also rename _hmp_ to _stats_.

> diff --git a/hw/rdma/rdma_rm.c b/hw/rdma/rdma_rm.c
> index 14580ca379..e019de1a14 100644
> --- a/hw/rdma/rdma_rm.c
> +++ b/hw/rdma/rdma_rm.c
> @@ -16,6 +16,7 @@
>  #include "qemu/osdep.h"
>  #include "qapi/error.h"
>  #include "cpu.h"
> +#include "monitor/monitor.h"
>  
>  #include "trace.h"
>  #include "rdma_utils.h"
> @@ -26,6 +27,58 @@
>  #define PG_DIR_SZ { TARGET_PAGE_SIZE / sizeof(__u64) }
>  #define PG_TBL_SZ { TARGET_PAGE_SIZE / sizeof(__u64) }
>  
> +void rdma_dump_device_counters(Monitor *mon, RdmaDeviceResources *dev_res)
> +{
> +    monitor_printf(mon, "\ttx               : %" PRId64 "\n",
> +                   dev_res->stats.tx);
> +    monitor_printf(mon, "\ttx_len           : %" PRId64 "\n",
> +                   dev_res->stats.tx_len);
> +    monitor_printf(mon, "\ttx_err           : %" PRId64 "\n",
> +                   dev_res->stats.tx_err);
> +    monitor_printf(mon, "\trx_bufs          : %" PRId64 "\n",
> +                   dev_res->stats.rx_bufs);
> +    monitor_printf(mon, "\trx_bufs_len      : %" PRId64 "\n",
> +                   dev_res->stats.rx_bufs_len);
> +    monitor_printf(mon, "\trx_bufs_err      : %" PRId64 "\n",
> +                   dev_res->stats.rx_bufs_err);
> +    monitor_printf(mon, "\tcomps            : %" PRId64 "\n",
> +                   dev_res->stats.completions);
> +    monitor_printf(mon, "\tmissing_comps    : %" PRId32 "\n",
> +                   dev_res->stats.missing_cqe);
> +    monitor_printf(mon, "\tpoll_cq (bk)     : %" PRId64 "\n",
> +                   dev_res->stats.poll_cq_from_bk);
> +    monitor_printf(mon, "\tpoll_cq_ppoll_to : %" PRId64 "\n",
> +                   dev_res->stats.poll_cq_ppoll_to);
> +    monitor_printf(mon, "\tpoll_cq (fe)     : %" PRId64 "\n",
> +                   dev_res->stats.poll_cq_from_guest);
> +    monitor_printf(mon, "\tpoll_cq_empty    : %" PRId64 "\n",
> +                   dev_res->stats.poll_cq_from_guest_empty);
> +    monitor_printf(mon, "\tmad_tx           : %" PRId64 "\n",
> +                   dev_res->stats.mad_tx);
> +    monitor_printf(mon, "\tmad_tx_err       : %" PRId64 "\n",
> +                   dev_res->stats.mad_tx_err);
> +    monitor_printf(mon, "\tmad_rx           : %" PRId64 "\n",
> +                   dev_res->stats.mad_rx);
> +    monitor_printf(mon, "\tmad_rx_err       : %" PRId64 "\n",
> +                   dev_res->stats.mad_rx_err);
> +    monitor_printf(mon, "\tmad_rx_bufs      : %" PRId64 "\n",
> +                   dev_res->stats.mad_rx_bufs);
> +    monitor_printf(mon, "\tmad_rx_bufs_err  : %" PRId64 "\n",
> +                   dev_res->stats.mad_rx_bufs_err);
> +    monitor_printf(mon, "\tPDs              : %" PRId32 "\n",
> +                   dev_res->pd_tbl.used);
> +    monitor_printf(mon, "\tMRs              : %" PRId32 "\n",
> +                   dev_res->mr_tbl.used);
> +    monitor_printf(mon, "\tUCs              : %" PRId32 "\n",
> +                   dev_res->uc_tbl.used);
> +    monitor_printf(mon, "\tQPs              : %" PRId32 "\n",
> +                   dev_res->qp_tbl.used);
> +    monitor_printf(mon, "\tCQs              : %" PRId32 "\n",
> +                   dev_res->cq_tbl.used);
> +    monitor_printf(mon, "\tCEQ_CTXs         : %" PRId32 "\n",
> +                   dev_res->cqe_ctx_tbl.used);
> +}
> +
>  static inline void res_tbl_init(const char *name, RdmaRmResTbl *tbl,
>                                  uint32_t tbl_sz, uint32_t res_sz)
>  {
> @@ -37,6 +90,7 @@ static inline void res_tbl_init(const char *name, RdmaRmResTbl *tbl,

More stats gathering, I think.

[...]
> diff --git a/hw/rdma/rdma_rm.h b/hw/rdma/rdma_rm.h
> index 9ec87d2667..a527d306f6 100644
> --- a/hw/rdma/rdma_rm.h
> +++ b/hw/rdma/rdma_rm.h
> @@ -20,6 +20,7 @@
>  #include "rdma_backend_defs.h"
>  #include "rdma_rm_defs.h"
>  
> +void rdma_dump_device_counters(Monitor *mon, RdmaDeviceResources *dev_res);
>  int rdma_rm_init(RdmaDeviceResources *dev_res, struct ibv_device_attr *dev_attr,
>                   Error **errp);
>  void rdma_rm_fini(RdmaDeviceResources *dev_res, RdmaBackendDev *backend_dev,
> diff --git a/hw/rdma/rdma_rm_defs.h b/hw/rdma/rdma_rm_defs.h
> index f0ee1f3072..4b8d704cfe 100644
> --- a/hw/rdma/rdma_rm_defs.h
> +++ b/hw/rdma/rdma_rm_defs.h

Still more stats gathering, I think.

[...]
> diff --git a/hw/rdma/vmw/pvrdma_main.c b/hw/rdma/vmw/pvrdma_main.c
> index b6061f4b6e..659331ac93 100644
> --- a/hw/rdma/vmw/pvrdma_main.c
> +++ b/hw/rdma/vmw/pvrdma_main.c
> @@ -25,6 +25,8 @@
>  #include "cpu.h"
>  #include "trace.h"
>  #include "sysemu/sysemu.h"
> +#include "monitor/monitor.h"
> +#include "hw/rdma/rdma_hmp.h"
>  
>  #include "../rdma_rm.h"
>  #include "../rdma_backend.h"
> @@ -55,6 +57,18 @@ static Property pvrdma_dev_properties[] = {
>      DEFINE_PROP_END_OF_LIST(),
>  };
>  
> +static void pvrdma_print_statistics(Monitor *mon, RdmaStatsProvider *obj)
> +{
> +    PVRDMADev *dev = PVRDMA_DEV(obj);
> +    PCIDevice *pdev = PCI_DEVICE(dev);
> +
> +    monitor_printf(mon, "%s, %x.%x\n", pdev->name, PCI_SLOT(pdev->devfn),
> +                   PCI_FUNC(pdev->devfn));
> +    monitor_printf(mon, "\tcommands         : %" PRId64 "\n",
> +                   dev->stats.commands);
> +    rdma_dump_device_counters(mon, &dev->rdma_dev_res);
> +}
> +
>  static void free_dev_ring(PCIDevice *pci_dev, PvrdmaRing *ring,
>                            void *ring_state)
>  {
> @@ -394,6 +408,7 @@ static void pvrdma_regs_write(void *opaque, hwaddr addr, uint64_t val,

Still more stats gathering, I think.

[...]
> @@ -631,6 +648,7 @@ static void pvrdma_class_init(ObjectClass *klass, void *data)
>  {
>      DeviceClass *dc = DEVICE_CLASS(klass);
>      PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
> +    RdmaStatsProviderClass *ir = RDMA_STATS_PROVIDER_CLASS(klass);
>  
>      k->realize = pvrdma_realize;
>      k->exit = pvrdma_exit;
> @@ -642,6 +660,8 @@ static void pvrdma_class_init(ObjectClass *klass, void *data)
>      dc->desc = "RDMA Device";
>      dc->props = pvrdma_dev_properties;
>      set_bit(DEVICE_CATEGORY_NETWORK, dc->categories);
> +
> +    ir->print_statistics = pvrdma_print_statistics;
>  }
>  
>  static const TypeInfo pvrdma_info = {
> @@ -651,6 +671,7 @@ static const TypeInfo pvrdma_info = {
>      .class_init = pvrdma_class_init,
>      .interfaces = (InterfaceInfo[]) {
>          { INTERFACE_CONVENTIONAL_PCI_DEVICE },
> +        { TYPE_RDMA_STATS_PROVIDER },

We use INTERFACE_ rather than TYPE_ for some, but not all interfaces.
You follow TYPE_INTERRUPT_STATS_PROVIDER precedence.  We suck.  Not your
problem to solve.

>          { }
>      }
>  };
> diff --git a/include/hw/rdma/rdma_hmp.h b/include/hw/rdma/rdma_hmp.h
> new file mode 100644
> index 0000000000..dd23f2bc84
> --- /dev/null
> +++ b/include/hw/rdma/rdma_hmp.h
> @@ -0,0 +1,40 @@
> +/*
> + * RDMA device: Human Monitor interface
> + *
> + * Copyright (C) 2019 Oracle
> + * Copyright (C) 2019 Red Hat Inc
> + *
> + * Authors:
> + *     Yuval Shaia <yuval.shaia@oracle.com>
> + *
> + * This work is licensed under the terms of the GNU GPL, version 2 or later.
> + * See the COPYING file in the top-level directory.
> + *
> + */
> +
> +#ifndef RDMA_HMP_H
> +#define RDMA_HMP_H
> +
> +#include "qom/object.h"
> +
> +#define TYPE_RDMA_STATS_PROVIDER "rdma"
> +
> +#define RDMA_STATS_PROVIDER_CLASS(klass) \
> +    OBJECT_CLASS_CHECK(RdmaStatsProviderClass, (klass), \
> +                       TYPE_RDMA_STATS_PROVIDER)
> +#define RDMA_STATS_PROVIDER_GET_CLASS(obj) \
> +    OBJECT_GET_CLASS(RdmaStatsProviderClass, (obj), \
> +                     TYPE_RDMA_STATS_PROVIDER)
> +#define RDMA_STATS_PROVIDER(obj) \
> +    INTERFACE_CHECK(RdmaStatsProvider, (obj), \
> +                    TYPE_RDMA_STATS_PROVIDER)
> +
> +typedef struct RdmaStatsProvider RdmaStatsProvider;
> +
> +typedef struct RdmaStatsProviderClass {
> +    InterfaceClass parent;
> +
> +    void (*print_statistics)(Monitor *mon, RdmaStatsProvider *obj);
> +} RdmaStatsProviderClass;
> +
> +#endif

I think using QOM to find the RDMA devices came out rather nicely.
Thanks for giving it a try.

Please consider the renames I recommended.

Acked-by: Markus Armbruster <armbru@redhat.com>

Re: [Qemu-devel] [PATCH v4 4/9] {hmp, hw/pvrdma}: Expose device internals via monitor interface

[Yuval Shaia]

On Fri, Mar 08, 2019 at 05:37:45PM +0100, Markus Armbruster wrote:
> Yuval Shaia <yuval.shaia@oracle.com> writes:
> 
> > Allow interrogating device internals through HMP interface.
> > The exposed indicators can be used for troubleshooting by developers or
> > sysadmin.
> > There is no need to expose these attributes to a management system (e.x.
> > libvirt) because (1) most of them are not "device-management' related
> > info and (2) there is no guarantee the interface is stable.
> >
> > Signed-off-by: Yuval Shaia <yuval.shaia@oracle.com>
> > ---
> >  hmp-commands-info.hx       | 14 ++++++++
> >  hmp.c                      | 27 +++++++++++++++
> >  hmp.h                      |  1 +
> >  hw/rdma/Makefile.objs      |  2 +-
> >  hw/rdma/rdma_backend.c     | 70 +++++++++++++++++++++++++++++---------
> >  hw/rdma/rdma_hmp.c         | 30 ++++++++++++++++
> >  hw/rdma/rdma_rm.c          | 60 ++++++++++++++++++++++++++++++++
> >  hw/rdma/rdma_rm.h          |  1 +
> >  hw/rdma/rdma_rm_defs.h     | 27 ++++++++++++++-
> >  hw/rdma/vmw/pvrdma.h       |  5 +++
> >  hw/rdma/vmw/pvrdma_main.c  | 21 ++++++++++++
> >  include/hw/rdma/rdma_hmp.h | 40 ++++++++++++++++++++++
> >  12 files changed, 279 insertions(+), 19 deletions(-)
> >  create mode 100644 hw/rdma/rdma_hmp.c
> >  create mode 100644 include/hw/rdma/rdma_hmp.h
> >
> > diff --git a/hmp-commands-info.hx b/hmp-commands-info.hx
> > index cbee8b944d..c59444c461 100644
> > --- a/hmp-commands-info.hx
> > +++ b/hmp-commands-info.hx
> > @@ -202,6 +202,20 @@ STEXI
> >  @item info pic
> >  @findex info pic
> >  Show PIC state.
> > +ETEXI
> > +
> > +    {
> > +        .name       = "rdma",
> > +        .args_type  = "",
> > +        .params     = "",
> > +        .help       = "show RDMA state",
> > +        .cmd        = hmp_info_rdma,
> > +    },
> > +
> > +STEXI
> > +@item info rdma
> > +@findex info rdma
> > +Show RDMA state.
> >  ETEXI
> >  
> >      {
> > diff --git a/hmp.c b/hmp.c
> > index 1e006eeb49..68511b2441 100644
> > --- a/hmp.c
> > +++ b/hmp.c
> > @@ -51,6 +51,7 @@
> >  #include "qemu/error-report.h"
> >  #include "exec/ramlist.h"
> >  #include "hw/intc/intc.h"
> > +#include "hw/rdma/rdma_hmp.h"
> >  #include "migration/snapshot.h"
> >  #include "migration/misc.h"
> >  
> > @@ -968,6 +969,32 @@ void hmp_info_pic(Monitor *mon, const QDict *qdict)
> >                                     hmp_info_pic_foreach, mon);
> >  }
> >  
> > +static int hmp_info_rdma_foreach(Object *obj, void *opaque)
> > +{
> > +    RdmaStatsProvider *rdma;
> > +    RdmaStatsProviderClass *k;
> > +    Monitor *mon = opaque;
> > +
> > +    if (object_dynamic_cast(obj, TYPE_RDMA_STATS_PROVIDER)) {
> > +        rdma = RDMA_STATS_PROVIDER(obj);
> > +        k = RDMA_STATS_PROVIDER_GET_CLASS(obj);
> > +        if (k->print_statistics) {
> > +            k->print_statistics(mon, rdma);
> > +        } else {
> > +            monitor_printf(mon, "RDMA statistics not available for %s.\n",
> > +                           object_get_typename(obj));
> > +        }
> > +    }
> > +
> > +    return 0;
> > +}
> > +
> > +void hmp_info_rdma(Monitor *mon, const QDict *qdict)
> > +{
> > +    object_child_foreach_recursive(object_get_root(),
> > +                                   hmp_info_rdma_foreach, mon);
> > +}
> > +
> >  void hmp_info_pci(Monitor *mon, const QDict *qdict)
> >  {
> >      PciInfoList *info_list, *info;
> > diff --git a/hmp.h b/hmp.h
> > index 5f1addcca2..666949afc3 100644
> > --- a/hmp.h
> > +++ b/hmp.h
> > @@ -36,6 +36,7 @@ void hmp_info_spice(Monitor *mon, const QDict *qdict);
> >  void hmp_info_balloon(Monitor *mon, const QDict *qdict);
> >  void hmp_info_irq(Monitor *mon, const QDict *qdict);
> >  void hmp_info_pic(Monitor *mon, const QDict *qdict);
> > +void hmp_info_rdma(Monitor *mon, const QDict *qdict);
> >  void hmp_info_pci(Monitor *mon, const QDict *qdict);
> >  void hmp_info_block_jobs(Monitor *mon, const QDict *qdict);
> >  void hmp_info_tpm(Monitor *mon, const QDict *qdict);
> 
> HMP stuff looks good to me.
> 
> > diff --git a/hw/rdma/Makefile.objs b/hw/rdma/Makefile.objs
> > index bd36cbf51c..dd59faff51 100644
> > --- a/hw/rdma/Makefile.objs
> > +++ b/hw/rdma/Makefile.objs
> > @@ -1,5 +1,5 @@
> >  ifeq ($(CONFIG_PVRDMA),y)
> > -obj-$(CONFIG_PCI) += rdma_utils.o rdma_backend.o rdma_rm.o
> > +obj-$(CONFIG_PCI) += rdma_utils.o rdma_backend.o rdma_rm.o rdma_hmp.o
> >  obj-$(CONFIG_PCI) += vmw/pvrdma_dev_ring.o vmw/pvrdma_cmd.o \
> >                       vmw/pvrdma_qp_ops.o vmw/pvrdma_main.o
> >  endif
> > diff --git a/hw/rdma/rdma_backend.c b/hw/rdma/rdma_backend.c
> > index 9679b842d1..bc2fefcf93 100644
> > --- a/hw/rdma/rdma_backend.c
> > +++ b/hw/rdma/rdma_backend.c
> 
> The changes to this file look like they're for gathering stats.  The
> patch would be much simpler to review had you kept the stats gathering
> separate.  I can only review the HMP / QOM part, and to do that, I have
> to identify what to ignore.
> 
> [...]
> > diff --git a/hw/rdma/rdma_hmp.c b/hw/rdma/rdma_hmp.c
> > new file mode 100644
> > index 0000000000..c5814473c5
> > --- /dev/null
> > +++ b/hw/rdma/rdma_hmp.c
> > @@ -0,0 +1,30 @@
> > +/*
> > + * RDMA device: Human Monitor interface
> 
> The file name and this comment are a bit akward.  Yes, you create
> TYPE_RDMA_STATS_PROVIDER for use in HMP info rdma, but there's
> absolutely nothing HMP-related in this file.  Same for rdma_hmp.h below.
> 
> Call them rdma_stats.c and rdma_stats.h?
> 
> > + *
> > + * Copyright (C) 2018 Oracle
> > + * Copyright (C) 2018 Red Hat Inc
> > + *
> > + * Authors:
> > + *     Yuval Shaia <yuval.shaia@oracle.com>
> > + *
> > + * This work is licensed under the terms of the GNU GPL, version 2 or later.
> > + * See the COPYING file in the top-level directory.
> > + *
> > + */
> > +
> > +#include "qemu/osdep.h"
> > +#include "hw/rdma/rdma_hmp.h"
> > +#include "qemu/module.h"
> > +
> > +static const TypeInfo rdma_hmp_info = {
> > +    .name = TYPE_RDMA_STATS_PROVIDER,
> > +    .parent = TYPE_INTERFACE,
> > +    .class_size = sizeof(RdmaStatsProviderClass),
> > +};
> > +
> > +static void rdma_hmp_register_types(void)
> > +{
> > +    type_register_static(&rdma_hmp_info);
> > +}
> > +
> > +type_init(rdma_hmp_register_types)
> 
> Also rename _hmp_ to _stats_.
> 
> > diff --git a/hw/rdma/rdma_rm.c b/hw/rdma/rdma_rm.c
> > index 14580ca379..e019de1a14 100644
> > --- a/hw/rdma/rdma_rm.c
> > +++ b/hw/rdma/rdma_rm.c
> > @@ -16,6 +16,7 @@
> >  #include "qemu/osdep.h"
> >  #include "qapi/error.h"
> >  #include "cpu.h"
> > +#include "monitor/monitor.h"
> >  
> >  #include "trace.h"
> >  #include "rdma_utils.h"
> > @@ -26,6 +27,58 @@
> >  #define PG_DIR_SZ { TARGET_PAGE_SIZE / sizeof(__u64) }
> >  #define PG_TBL_SZ { TARGET_PAGE_SIZE / sizeof(__u64) }
> >  
> > +void rdma_dump_device_counters(Monitor *mon, RdmaDeviceResources *dev_res)
> > +{
> > +    monitor_printf(mon, "\ttx               : %" PRId64 "\n",
> > +                   dev_res->stats.tx);
> > +    monitor_printf(mon, "\ttx_len           : %" PRId64 "\n",
> > +                   dev_res->stats.tx_len);
> > +    monitor_printf(mon, "\ttx_err           : %" PRId64 "\n",
> > +                   dev_res->stats.tx_err);
> > +    monitor_printf(mon, "\trx_bufs          : %" PRId64 "\n",
> > +                   dev_res->stats.rx_bufs);
> > +    monitor_printf(mon, "\trx_bufs_len      : %" PRId64 "\n",
> > +                   dev_res->stats.rx_bufs_len);
> > +    monitor_printf(mon, "\trx_bufs_err      : %" PRId64 "\n",
> > +                   dev_res->stats.rx_bufs_err);
> > +    monitor_printf(mon, "\tcomps            : %" PRId64 "\n",
> > +                   dev_res->stats.completions);
> > +    monitor_printf(mon, "\tmissing_comps    : %" PRId32 "\n",
> > +                   dev_res->stats.missing_cqe);
> > +    monitor_printf(mon, "\tpoll_cq (bk)     : %" PRId64 "\n",
> > +                   dev_res->stats.poll_cq_from_bk);
> > +    monitor_printf(mon, "\tpoll_cq_ppoll_to : %" PRId64 "\n",
> > +                   dev_res->stats.poll_cq_ppoll_to);
> > +    monitor_printf(mon, "\tpoll_cq (fe)     : %" PRId64 "\n",
> > +                   dev_res->stats.poll_cq_from_guest);
> > +    monitor_printf(mon, "\tpoll_cq_empty    : %" PRId64 "\n",
> > +                   dev_res->stats.poll_cq_from_guest_empty);
> > +    monitor_printf(mon, "\tmad_tx           : %" PRId64 "\n",
> > +                   dev_res->stats.mad_tx);
> > +    monitor_printf(mon, "\tmad_tx_err       : %" PRId64 "\n",
> > +                   dev_res->stats.mad_tx_err);
> > +    monitor_printf(mon, "\tmad_rx           : %" PRId64 "\n",
> > +                   dev_res->stats.mad_rx);
> > +    monitor_printf(mon, "\tmad_rx_err       : %" PRId64 "\n",
> > +                   dev_res->stats.mad_rx_err);
> > +    monitor_printf(mon, "\tmad_rx_bufs      : %" PRId64 "\n",
> > +                   dev_res->stats.mad_rx_bufs);
> > +    monitor_printf(mon, "\tmad_rx_bufs_err  : %" PRId64 "\n",
> > +                   dev_res->stats.mad_rx_bufs_err);
> > +    monitor_printf(mon, "\tPDs              : %" PRId32 "\n",
> > +                   dev_res->pd_tbl.used);
> > +    monitor_printf(mon, "\tMRs              : %" PRId32 "\n",
> > +                   dev_res->mr_tbl.used);
> > +    monitor_printf(mon, "\tUCs              : %" PRId32 "\n",
> > +                   dev_res->uc_tbl.used);
> > +    monitor_printf(mon, "\tQPs              : %" PRId32 "\n",
> > +                   dev_res->qp_tbl.used);
> > +    monitor_printf(mon, "\tCQs              : %" PRId32 "\n",
> > +                   dev_res->cq_tbl.used);
> > +    monitor_printf(mon, "\tCEQ_CTXs         : %" PRId32 "\n",
> > +                   dev_res->cqe_ctx_tbl.used);
> > +}
> > +
> >  static inline void res_tbl_init(const char *name, RdmaRmResTbl *tbl,
> >                                  uint32_t tbl_sz, uint32_t res_sz)
> >  {
> > @@ -37,6 +90,7 @@ static inline void res_tbl_init(const char *name, RdmaRmResTbl *tbl,
> 
> More stats gathering, I think.
> 
> [...]
> > diff --git a/hw/rdma/rdma_rm.h b/hw/rdma/rdma_rm.h
> > index 9ec87d2667..a527d306f6 100644
> > --- a/hw/rdma/rdma_rm.h
> > +++ b/hw/rdma/rdma_rm.h
> > @@ -20,6 +20,7 @@
> >  #include "rdma_backend_defs.h"
> >  #include "rdma_rm_defs.h"
> >  
> > +void rdma_dump_device_counters(Monitor *mon, RdmaDeviceResources *dev_res);
> >  int rdma_rm_init(RdmaDeviceResources *dev_res, struct ibv_device_attr *dev_attr,
> >                   Error **errp);
> >  void rdma_rm_fini(RdmaDeviceResources *dev_res, RdmaBackendDev *backend_dev,
> > diff --git a/hw/rdma/rdma_rm_defs.h b/hw/rdma/rdma_rm_defs.h
> > index f0ee1f3072..4b8d704cfe 100644
> > --- a/hw/rdma/rdma_rm_defs.h
> > +++ b/hw/rdma/rdma_rm_defs.h
> 
> Still more stats gathering, I think.
> 
> [...]
> > diff --git a/hw/rdma/vmw/pvrdma_main.c b/hw/rdma/vmw/pvrdma_main.c
> > index b6061f4b6e..659331ac93 100644
> > --- a/hw/rdma/vmw/pvrdma_main.c
> > +++ b/hw/rdma/vmw/pvrdma_main.c
> > @@ -25,6 +25,8 @@
> >  #include "cpu.h"
> >  #include "trace.h"
> >  #include "sysemu/sysemu.h"
> > +#include "monitor/monitor.h"
> > +#include "hw/rdma/rdma_hmp.h"
> >  
> >  #include "../rdma_rm.h"
> >  #include "../rdma_backend.h"
> > @@ -55,6 +57,18 @@ static Property pvrdma_dev_properties[] = {
> >      DEFINE_PROP_END_OF_LIST(),
> >  };
> >  
> > +static void pvrdma_print_statistics(Monitor *mon, RdmaStatsProvider *obj)
> > +{
> > +    PVRDMADev *dev = PVRDMA_DEV(obj);
> > +    PCIDevice *pdev = PCI_DEVICE(dev);
> > +
> > +    monitor_printf(mon, "%s, %x.%x\n", pdev->name, PCI_SLOT(pdev->devfn),
> > +                   PCI_FUNC(pdev->devfn));
> > +    monitor_printf(mon, "\tcommands         : %" PRId64 "\n",
> > +                   dev->stats.commands);
> > +    rdma_dump_device_counters(mon, &dev->rdma_dev_res);
> > +}
> > +
> >  static void free_dev_ring(PCIDevice *pci_dev, PvrdmaRing *ring,
> >                            void *ring_state)
> >  {
> > @@ -394,6 +408,7 @@ static void pvrdma_regs_write(void *opaque, hwaddr addr, uint64_t val,
> 
> Still more stats gathering, I think.
> 
> [...]
> > @@ -631,6 +648,7 @@ static void pvrdma_class_init(ObjectClass *klass, void *data)
> >  {
> >      DeviceClass *dc = DEVICE_CLASS(klass);
> >      PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
> > +    RdmaStatsProviderClass *ir = RDMA_STATS_PROVIDER_CLASS(klass);
> >  
> >      k->realize = pvrdma_realize;
> >      k->exit = pvrdma_exit;
> > @@ -642,6 +660,8 @@ static void pvrdma_class_init(ObjectClass *klass, void *data)
> >      dc->desc = "RDMA Device";
> >      dc->props = pvrdma_dev_properties;
> >      set_bit(DEVICE_CATEGORY_NETWORK, dc->categories);
> > +
> > +    ir->print_statistics = pvrdma_print_statistics;
> >  }
> >  
> >  static const TypeInfo pvrdma_info = {
> > @@ -651,6 +671,7 @@ static const TypeInfo pvrdma_info = {
> >      .class_init = pvrdma_class_init,
> >      .interfaces = (InterfaceInfo[]) {
> >          { INTERFACE_CONVENTIONAL_PCI_DEVICE },
> > +        { TYPE_RDMA_STATS_PROVIDER },
> 
> We use INTERFACE_ rather than TYPE_ for some, but not all interfaces.
> You follow TYPE_INTERRUPT_STATS_PROVIDER precedence.  We suck.  Not your
> problem to solve.
> 
> >          { }
> >      }
> >  };
> > diff --git a/include/hw/rdma/rdma_hmp.h b/include/hw/rdma/rdma_hmp.h
> > new file mode 100644
> > index 0000000000..dd23f2bc84
> > --- /dev/null
> > +++ b/include/hw/rdma/rdma_hmp.h
> > @@ -0,0 +1,40 @@
> > +/*
> > + * RDMA device: Human Monitor interface
> > + *
> > + * Copyright (C) 2019 Oracle
> > + * Copyright (C) 2019 Red Hat Inc
> > + *
> > + * Authors:
> > + *     Yuval Shaia <yuval.shaia@oracle.com>
> > + *
> > + * This work is licensed under the terms of the GNU GPL, version 2 or later.
> > + * See the COPYING file in the top-level directory.
> > + *
> > + */
> > +
> > +#ifndef RDMA_HMP_H
> > +#define RDMA_HMP_H
> > +
> > +#include "qom/object.h"
> > +
> > +#define TYPE_RDMA_STATS_PROVIDER "rdma"
> > +
> > +#define RDMA_STATS_PROVIDER_CLASS(klass) \
> > +    OBJECT_CLASS_CHECK(RdmaStatsProviderClass, (klass), \
> > +                       TYPE_RDMA_STATS_PROVIDER)
> > +#define RDMA_STATS_PROVIDER_GET_CLASS(obj) \
> > +    OBJECT_GET_CLASS(RdmaStatsProviderClass, (obj), \
> > +                     TYPE_RDMA_STATS_PROVIDER)
> > +#define RDMA_STATS_PROVIDER(obj) \
> > +    INTERFACE_CHECK(RdmaStatsProvider, (obj), \
> > +                    TYPE_RDMA_STATS_PROVIDER)
> > +
> > +typedef struct RdmaStatsProvider RdmaStatsProvider;
> > +
> > +typedef struct RdmaStatsProviderClass {
> > +    InterfaceClass parent;
> > +
> > +    void (*print_statistics)(Monitor *mon, RdmaStatsProvider *obj);
> > +} RdmaStatsProviderClass;
> > +
> > +#endif
> 
> I think using QOM to find the RDMA devices came out rather nicely.
> Thanks for giving it a try.
> 
> Please consider the renames I recommended.

Will do, it perfectly make sense.

> 
> Acked-by: Markus Armbruster <armbru@redhat.com>

Thanks.

Re: [Qemu-devel] [PATCH v4 4/9] {hmp, hw/pvrdma}: Expose device internals via monitor interface

[Yuval Shaia]

> 
> [...]
> > diff --git a/hw/rdma/rdma_hmp.c b/hw/rdma/rdma_hmp.c
> > new file mode 100644
> > index 0000000000..c5814473c5
> > --- /dev/null
> > +++ b/hw/rdma/rdma_hmp.c
> > @@ -0,0 +1,30 @@
> > +/*
> > + * RDMA device: Human Monitor interface
> 
> The file name and this comment are a bit akward.  Yes, you create
> TYPE_RDMA_STATS_PROVIDER for use in HMP info rdma, but there's
> absolutely nothing HMP-related in this file.  Same for rdma_hmp.h below.
> 
> Call them rdma_stats.c and rdma_stats.h?

Renamed to rdma.h and rdma.c, wo knows what other things might be added in
the future.

> 
> > + *
> > + * Copyright (C) 2018 Oracle
> > + * Copyright (C) 2018 Red Hat Inc
> > + *
> > + * Authors:
> > + *     Yuval Shaia <yuval.shaia@oracle.com>
> > + *
> > + * This work is licensed under the terms of the GNU GPL, version 2 or later.
> > + * See the COPYING file in the top-level directory.
> > + *
> > + */
> > +
> > +#include "qemu/osdep.h"
> > +#include "hw/rdma/rdma_hmp.h"
> > +#include "qemu/module.h"
> > +
> > +static const TypeInfo rdma_hmp_info = {
> > +    .name = TYPE_RDMA_STATS_PROVIDER,
> > +    .parent = TYPE_INTERFACE,
> > +    .class_size = sizeof(RdmaStatsProviderClass),
> > +};
> > +
> > +static void rdma_hmp_register_types(void)
> > +{
> > +    type_register_static(&rdma_hmp_info);
> > +}
> > +
> > +type_init(rdma_hmp_register_types)
> 
> Also rename _hmp_ to _stats_.

Ditto.

> 

[Qemu-devel] [PATCH v4 5/9] hw/rdma: Free all MAD receive buffers when device is closed

[Yuval Shaia]

When device is going down free all saved MAD buffers.

Signed-off-by: Yuval Shaia <yuval.shaia@oracle.com>
Reviewed-by: Marcel Apfelbaum <marcel.apfelbaum@gmail.com>
---
 hw/rdma/rdma_backend.c    | 34 +++++++++++++++++++++++++++++++++-
 hw/rdma/vmw/pvrdma_main.c |  2 ++
 2 files changed, 35 insertions(+), 1 deletion(-)

diff --git a/hw/rdma/rdma_backend.c b/hw/rdma/rdma_backend.c
index bc2fefcf93..a65f5737e4 100644
--- a/hw/rdma/rdma_backend.c
+++ b/hw/rdma/rdma_backend.c
@@ -64,6 +64,33 @@ static inline void complete_work(enum ibv_wc_status status, uint32_t vendor_err,
     comp_handler(ctx, &wc);
 }
 
+static void free_cqe_ctx(gpointer data, gpointer user_data)
+{
+    BackendCtx *bctx;
+    RdmaDeviceResources *rdma_dev_res = user_data;
+    unsigned long cqe_ctx_id = GPOINTER_TO_INT(data);
+
+    bctx = rdma_rm_get_cqe_ctx(rdma_dev_res, cqe_ctx_id);
+    if (bctx) {
+        rdma_rm_dealloc_cqe_ctx(rdma_dev_res, cqe_ctx_id);
+    }
+    g_free(bctx);
+}
+
+static void clean_recv_mads(RdmaBackendDev *backend_dev)
+{
+    unsigned long cqe_ctx_id;
+
+    do {
+        cqe_ctx_id = rdma_protected_qlist_pop_int64(&backend_dev->
+                                                    recv_mads_list);
+        if (cqe_ctx_id != -ENOENT) {
+            free_cqe_ctx(GINT_TO_POINTER(cqe_ctx_id),
+                         backend_dev->rdma_dev_res);
+        }
+    } while (cqe_ctx_id != -ENOENT);
+}
+
 static int rdma_poll_cq(RdmaDeviceResources *rdma_dev_res, struct ibv_cq *ibcq)
 {
     int i, ne, total_ne = 0;
@@ -1037,6 +1064,11 @@ static int mad_init(RdmaBackendDev *backend_dev, CharBackend *mad_chr_be)
     return 0;
 }
 
+static void mad_stop(RdmaBackendDev *backend_dev)
+{
+    clean_recv_mads(backend_dev);
+}
+
 static void mad_fini(RdmaBackendDev *backend_dev)
 {
     disable_rdmacm_mux_async(backend_dev);
@@ -1224,12 +1256,12 @@ void rdma_backend_start(RdmaBackendDev *backend_dev)
 
 void rdma_backend_stop(RdmaBackendDev *backend_dev)
 {
+    mad_stop(backend_dev);
     stop_backend_thread(&backend_dev->comp_thread);
 }
 
 void rdma_backend_fini(RdmaBackendDev *backend_dev)
 {
-    rdma_backend_stop(backend_dev);
     mad_fini(backend_dev);
     g_hash_table_destroy(ah_hash);
     ibv_destroy_comp_channel(backend_dev->channel);
diff --git a/hw/rdma/vmw/pvrdma_main.c b/hw/rdma/vmw/pvrdma_main.c
index 659331ac93..b795f80666 100644
--- a/hw/rdma/vmw/pvrdma_main.c
+++ b/hw/rdma/vmw/pvrdma_main.c
@@ -305,6 +305,8 @@ static void pvrdma_fini(PCIDevice *pdev)
 
     pvrdma_qp_ops_fini();
 
+    rdma_backend_stop(&dev->backend_dev);
+
     rdma_rm_fini(&dev->rdma_dev_res, &dev->backend_dev,
                  dev->backend_eth_device_name);
 
-- 
2.17.2

[Qemu-devel] [PATCH v4 6/9] hw/rdma: Free all receive buffers when QP is destroyed

[Yuval Shaia]

When QP is destroyed the backend QP is destroyed as well. This ensures
we clean all received buffer we posted to it.
However, a contexts of these buffers are still remain in the device.
Fix it by maintaining a list of buffer's context and free them when QP
is destroyed.

Signed-off-by: Yuval Shaia <yuval.shaia@oracle.com>
Reviewed-by: Marcel Apfelbaum <marcel.apfelbaum@gmail.com>
---
 hw/rdma/rdma_backend.c      | 26 ++++++++++++++++++++------
 hw/rdma/rdma_backend.h      |  2 +-
 hw/rdma/rdma_backend_defs.h |  2 +-
 hw/rdma/rdma_rm.c           |  2 +-
 hw/rdma/rdma_utils.c        | 29 +++++++++++++++++++++++++++++
 hw/rdma/rdma_utils.h        | 11 +++++++++++
 6 files changed, 63 insertions(+), 9 deletions(-)

diff --git a/hw/rdma/rdma_backend.c b/hw/rdma/rdma_backend.c
index a65f5737e4..d511ca282b 100644
--- a/hw/rdma/rdma_backend.c
+++ b/hw/rdma/rdma_backend.c
@@ -39,6 +39,7 @@
 typedef struct BackendCtx {
     void *up_ctx;
     struct ibv_sge sge; /* Used to save MAD recv buffer */
+    RdmaBackendQP *backend_qp; /* To maintain recv buffers */
 } BackendCtx;
 
 struct backend_umad {
@@ -73,6 +74,7 @@ static void free_cqe_ctx(gpointer data, gpointer user_data)
     bctx = rdma_rm_get_cqe_ctx(rdma_dev_res, cqe_ctx_id);
     if (bctx) {
         rdma_rm_dealloc_cqe_ctx(rdma_dev_res, cqe_ctx_id);
+        atomic_dec(&rdma_dev_res->stats.missing_cqe);
     }
     g_free(bctx);
 }
@@ -85,13 +87,15 @@ static void clean_recv_mads(RdmaBackendDev *backend_dev)
         cqe_ctx_id = rdma_protected_qlist_pop_int64(&backend_dev->
                                                     recv_mads_list);
         if (cqe_ctx_id != -ENOENT) {
+            atomic_inc(&backend_dev->rdma_dev_res->stats.missing_cqe);
             free_cqe_ctx(GINT_TO_POINTER(cqe_ctx_id),
                          backend_dev->rdma_dev_res);
         }
     } while (cqe_ctx_id != -ENOENT);
 }
 
-static int rdma_poll_cq(RdmaDeviceResources *rdma_dev_res, struct ibv_cq *ibcq)
+static int rdma_poll_cq(RdmaBackendDev *backend_dev,
+                        RdmaDeviceResources *rdma_dev_res, struct ibv_cq *ibcq)
 {
     int i, ne, total_ne = 0;
     BackendCtx *bctx;
@@ -113,6 +117,8 @@ static int rdma_poll_cq(RdmaDeviceResources *rdma_dev_res, struct ibv_cq *ibcq)
 
             comp_handler(bctx->up_ctx, &wc[i]);
 
+            rdma_protected_gslist_remove_int32(&bctx->backend_qp->cqe_ctx_list,
+                                               wc[i].wr_id);
             rdma_rm_dealloc_cqe_ctx(rdma_dev_res, wc[i].wr_id);
             g_free(bctx);
         }
@@ -175,14 +181,12 @@ static void *comp_handler_thread(void *arg)
             }
 
             backend_dev->rdma_dev_res->stats.poll_cq_from_bk++;
-            rdma_poll_cq(backend_dev->rdma_dev_res, ev_cq);
+            rdma_poll_cq(backend_dev, backend_dev->rdma_dev_res, ev_cq);
 
             ibv_ack_cq_events(ev_cq, 1);
         }
     }
 
-    /* TODO: Post cqe for all remaining buffs that were posted */
-
     backend_dev->comp_thread.is_running = false;
 
     qemu_thread_exit(0);
@@ -311,7 +315,7 @@ void rdma_backend_poll_cq(RdmaDeviceResources *rdma_dev_res, RdmaBackendCQ *cq)
     int polled;
 
     rdma_dev_res->stats.poll_cq_from_guest++;
-    polled = rdma_poll_cq(rdma_dev_res, cq->ibcq);
+    polled = rdma_poll_cq(cq->backend_dev, rdma_dev_res, cq->ibcq);
     if (!polled) {
         rdma_dev_res->stats.poll_cq_from_guest_empty++;
     }
@@ -501,6 +505,7 @@ void rdma_backend_post_send(RdmaBackendDev *backend_dev,
 
     bctx = g_malloc0(sizeof(*bctx));
     bctx->up_ctx = ctx;
+    bctx->backend_qp = qp;
 
     rc = rdma_rm_alloc_cqe_ctx(backend_dev->rdma_dev_res, &bctx_id, bctx);
     if (unlikely(rc)) {
@@ -508,6 +513,8 @@ void rdma_backend_post_send(RdmaBackendDev *backend_dev,
         goto err_free_bctx;
     }
 
+    rdma_protected_gslist_append_int32(&qp->cqe_ctx_list, bctx_id);
+
     rc = build_host_sge_array(backend_dev->rdma_dev_res, new_sge, sge, num_sge,
                               &backend_dev->rdma_dev_res->stats.tx_len);
     if (rc) {
@@ -616,6 +623,7 @@ void rdma_backend_post_recv(RdmaBackendDev *backend_dev,
 
     bctx = g_malloc0(sizeof(*bctx));
     bctx->up_ctx = ctx;
+    bctx->backend_qp = qp;
 
     rc = rdma_rm_alloc_cqe_ctx(rdma_dev_res, &bctx_id, bctx);
     if (unlikely(rc)) {
@@ -623,6 +631,8 @@ void rdma_backend_post_recv(RdmaBackendDev *backend_dev,
         goto err_free_bctx;
     }
 
+    rdma_protected_gslist_append_int32(&qp->cqe_ctx_list, bctx_id);
+
     rc = build_host_sge_array(rdma_dev_res, new_sge, sge, num_sge,
                               &backend_dev->rdma_dev_res->stats.rx_bufs_len);
     if (rc) {
@@ -762,6 +772,8 @@ int rdma_backend_create_qp(RdmaBackendQP *qp, uint8_t qp_type,
         return -EIO;
     }
 
+    rdma_protected_gslist_init(&qp->cqe_ctx_list);
+
     qp->ibpd = pd->ibpd;
 
     /* TODO: Query QP to get max_inline_data and save it to be used in send */
@@ -919,11 +931,13 @@ int rdma_backend_query_qp(RdmaBackendQP *qp, struct ibv_qp_attr *attr,
     return ibv_query_qp(qp->ibqp, attr, attr_mask, init_attr);
 }
 
-void rdma_backend_destroy_qp(RdmaBackendQP *qp)
+void rdma_backend_destroy_qp(RdmaBackendQP *qp, RdmaDeviceResources *dev_res)
 {
     if (qp->ibqp) {
         ibv_destroy_qp(qp->ibqp);
     }
+    g_slist_foreach(qp->cqe_ctx_list.list, free_cqe_ctx, dev_res);
+    rdma_protected_gslist_destroy(&qp->cqe_ctx_list);
 }
 
 #define CHK_ATTR(req, dev, member, fmt) ({ \
diff --git a/hw/rdma/rdma_backend.h b/hw/rdma/rdma_backend.h
index 5114c90e67..cb5efa2a3a 100644
--- a/hw/rdma/rdma_backend.h
+++ b/hw/rdma/rdma_backend.h
@@ -102,7 +102,7 @@ int rdma_backend_qp_state_rts(RdmaBackendQP *qp, uint8_t qp_type,
                               uint32_t sq_psn, uint32_t qkey, bool use_qkey);
 int rdma_backend_query_qp(RdmaBackendQP *qp, struct ibv_qp_attr *attr,
                           int attr_mask, struct ibv_qp_init_attr *init_attr);
-void rdma_backend_destroy_qp(RdmaBackendQP *qp);
+void rdma_backend_destroy_qp(RdmaBackendQP *qp, RdmaDeviceResources *dev_res);
 
 void rdma_backend_post_send(RdmaBackendDev *backend_dev,
                             RdmaBackendQP *qp, uint8_t qp_type,
diff --git a/hw/rdma/rdma_backend_defs.h b/hw/rdma/rdma_backend_defs.h
index a8c15b09ab..817153dc8c 100644
--- a/hw/rdma/rdma_backend_defs.h
+++ b/hw/rdma/rdma_backend_defs.h
@@ -26,7 +26,6 @@ typedef struct RdmaDeviceResources RdmaDeviceResources;
 
 typedef struct RdmaBackendThread {
     QemuThread thread;
-    QemuMutex mutex;
     bool run; /* Set by thread manager to let thread know it should exit */
     bool is_running; /* Set by the thread to report its status */
 } RdmaBackendThread;
@@ -66,6 +65,7 @@ typedef struct RdmaBackendQP {
     struct ibv_pd *ibpd;
     struct ibv_qp *ibqp;
     uint8_t sgid_idx;
+    RdmaProtectedGSList cqe_ctx_list;
 } RdmaBackendQP;
 
 #endif
diff --git a/hw/rdma/rdma_rm.c b/hw/rdma/rdma_rm.c
index e019de1a14..4c9fe1591e 100644
--- a/hw/rdma/rdma_rm.c
+++ b/hw/rdma/rdma_rm.c
@@ -537,7 +537,7 @@ void rdma_rm_dealloc_qp(RdmaDeviceResources *dev_res, uint32_t qp_handle)
         return;
     }
 
-    rdma_backend_destroy_qp(&qp->backend_qp);
+    rdma_backend_destroy_qp(&qp->backend_qp, dev_res);
 
     rdma_res_tbl_dealloc(&dev_res->qp_tbl, qp->qpn);
 }
diff --git a/hw/rdma/rdma_utils.c b/hw/rdma/rdma_utils.c
index 0a8abe572d..73f279104c 100644
--- a/hw/rdma/rdma_utils.c
+++ b/hw/rdma/rdma_utils.c
@@ -90,3 +90,32 @@ int64_t rdma_protected_qlist_pop_int64(RdmaProtectedQList *list)
 
     return qnum_get_uint(qobject_to(QNum, obj));
 }
+
+void rdma_protected_gslist_init(RdmaProtectedGSList *list)
+{
+    qemu_mutex_init(&list->lock);
+}
+
+void rdma_protected_gslist_destroy(RdmaProtectedGSList *list)
+{
+    if (list->list) {
+        g_slist_free(list->list);
+        list->list = NULL;
+    }
+}
+
+void rdma_protected_gslist_append_int32(RdmaProtectedGSList *list,
+                                        int32_t value)
+{
+    qemu_mutex_lock(&list->lock);
+    list->list = g_slist_prepend(list->list, GINT_TO_POINTER(value));
+    qemu_mutex_unlock(&list->lock);
+}
+
+void rdma_protected_gslist_remove_int32(RdmaProtectedGSList *list,
+                                        int32_t value)
+{
+    qemu_mutex_lock(&list->lock);
+    list->list = g_slist_remove(list->list, GINT_TO_POINTER(value));
+    qemu_mutex_unlock(&list->lock);
+}
diff --git a/hw/rdma/rdma_utils.h b/hw/rdma/rdma_utils.h
index a8bf1d4fec..2d42249691 100644
--- a/hw/rdma/rdma_utils.h
+++ b/hw/rdma/rdma_utils.h
@@ -34,12 +34,23 @@ typedef struct RdmaProtectedQList {
     QList *list;
 } RdmaProtectedQList;
 
+typedef struct RdmaProtectedGSList {
+    QemuMutex lock;
+    GSList *list;
+} RdmaProtectedGSList;
+
 void *rdma_pci_dma_map(PCIDevice *dev, dma_addr_t addr, dma_addr_t plen);
 void rdma_pci_dma_unmap(PCIDevice *dev, void *buffer, dma_addr_t len);
 void rdma_protected_qlist_init(RdmaProtectedQList *list);
 void rdma_protected_qlist_destroy(RdmaProtectedQList *list);
 void rdma_protected_qlist_append_int64(RdmaProtectedQList *list, int64_t value);
 int64_t rdma_protected_qlist_pop_int64(RdmaProtectedQList *list);
+void rdma_protected_gslist_init(RdmaProtectedGSList *list);
+void rdma_protected_gslist_destroy(RdmaProtectedGSList *list);
+void rdma_protected_gslist_append_int32(RdmaProtectedGSList *list,
+                                        int32_t value);
+void rdma_protected_gslist_remove_int32(RdmaProtectedGSList *list,
+                                        int32_t value);
 
 static inline void addrconf_addr_eui48(uint8_t *eui, const char *addr)
 {
-- 
2.17.2

[Qemu-devel] [PATCH v4 7/9] hw/pvrdma: Delete unneeded function argument

[Yuval Shaia]

The function's argument rdma_dev_res is not needed as it is stored in
the backend_dev object at init.

Signed-off-by: Yuval Shaia <yuval.shaia@oracle.com>
Reviewed-by: Marcel Apfelbaum <marcel.apfelbaum@gmail.com>
---
 hw/rdma/rdma_backend.c      | 13 ++++++-------
 hw/rdma/rdma_backend.h      |  1 -
 hw/rdma/vmw/pvrdma_qp_ops.c |  3 +--
 3 files changed, 7 insertions(+), 10 deletions(-)

diff --git a/hw/rdma/rdma_backend.c b/hw/rdma/rdma_backend.c
index d511ca282b..66185bd487 100644
--- a/hw/rdma/rdma_backend.c
+++ b/hw/rdma/rdma_backend.c
@@ -594,7 +594,6 @@ static unsigned int save_mad_recv_buffer(RdmaBackendDev *backend_dev,
 }
 
 void rdma_backend_post_recv(RdmaBackendDev *backend_dev,
-                            RdmaDeviceResources *rdma_dev_res,
                             RdmaBackendQP *qp, uint8_t qp_type,
                             struct ibv_sge *sge, uint32_t num_sge, void *ctx)
 {
@@ -613,9 +612,9 @@ void rdma_backend_post_recv(RdmaBackendDev *backend_dev,
             rc = save_mad_recv_buffer(backend_dev, sge, num_sge, ctx);
             if (rc) {
                 complete_work(IBV_WC_GENERAL_ERR, rc, ctx);
-                rdma_dev_res->stats.mad_rx_bufs_err++;
+                backend_dev->rdma_dev_res->stats.mad_rx_bufs_err++;
             } else {
-                rdma_dev_res->stats.mad_rx_bufs++;
+                backend_dev->rdma_dev_res->stats.mad_rx_bufs++;
             }
         }
         return;
@@ -625,7 +624,7 @@ void rdma_backend_post_recv(RdmaBackendDev *backend_dev,
     bctx->up_ctx = ctx;
     bctx->backend_qp = qp;
 
-    rc = rdma_rm_alloc_cqe_ctx(rdma_dev_res, &bctx_id, bctx);
+    rc = rdma_rm_alloc_cqe_ctx(backend_dev->rdma_dev_res, &bctx_id, bctx);
     if (unlikely(rc)) {
         complete_work(IBV_WC_GENERAL_ERR, VENDOR_ERR_NOMEM, ctx);
         goto err_free_bctx;
@@ -633,7 +632,7 @@ void rdma_backend_post_recv(RdmaBackendDev *backend_dev,
 
     rdma_protected_gslist_append_int32(&qp->cqe_ctx_list, bctx_id);
 
-    rc = build_host_sge_array(rdma_dev_res, new_sge, sge, num_sge,
+    rc = build_host_sge_array(backend_dev->rdma_dev_res, new_sge, sge, num_sge,
                               &backend_dev->rdma_dev_res->stats.rx_bufs_len);
     if (rc) {
         complete_work(IBV_WC_GENERAL_ERR, rc, ctx);
@@ -652,13 +651,13 @@ void rdma_backend_post_recv(RdmaBackendDev *backend_dev,
     }
 
     atomic_inc(&backend_dev->rdma_dev_res->stats.missing_cqe);
-    rdma_dev_res->stats.rx_bufs++;
+    backend_dev->rdma_dev_res->stats.rx_bufs++;
 
     return;
 
 err_dealloc_cqe_ctx:
     backend_dev->rdma_dev_res->stats.rx_bufs_err++;
-    rdma_rm_dealloc_cqe_ctx(rdma_dev_res, bctx_id);
+    rdma_rm_dealloc_cqe_ctx(backend_dev->rdma_dev_res, bctx_id);
 
 err_free_bctx:
     g_free(bctx);
diff --git a/hw/rdma/rdma_backend.h b/hw/rdma/rdma_backend.h
index cb5efa2a3a..5d507a1c41 100644
--- a/hw/rdma/rdma_backend.h
+++ b/hw/rdma/rdma_backend.h
@@ -111,7 +111,6 @@ void rdma_backend_post_send(RdmaBackendDev *backend_dev,
                             union ibv_gid *dgid, uint32_t dqpn, uint32_t dqkey,
                             void *ctx);
 void rdma_backend_post_recv(RdmaBackendDev *backend_dev,
-                            RdmaDeviceResources *rdma_dev_res,
                             RdmaBackendQP *qp, uint8_t qp_type,
                             struct ibv_sge *sge, uint32_t num_sge, void *ctx);
 
diff --git a/hw/rdma/vmw/pvrdma_qp_ops.c b/hw/rdma/vmw/pvrdma_qp_ops.c
index 16db726dac..508d8fca3c 100644
--- a/hw/rdma/vmw/pvrdma_qp_ops.c
+++ b/hw/rdma/vmw/pvrdma_qp_ops.c
@@ -231,8 +231,7 @@ void pvrdma_qp_recv(PVRDMADev *dev, uint32_t qp_handle)
             continue;
         }
 
-        rdma_backend_post_recv(&dev->backend_dev, &dev->rdma_dev_res,
-                               &qp->backend_qp, qp->qp_type,
+        rdma_backend_post_recv(&dev->backend_dev, &qp->backend_qp, qp->qp_type,
                                (struct ibv_sge *)&wqe->sge[0], wqe->hdr.num_sge,
                                comp_ctx);
 
-- 
2.17.2

[Qemu-devel] [PATCH v4 8/9] hw/pvrdma: Delete pvrdma_exit function

[Yuval Shaia]

This hook is not called and was implemented by mistake.

Signed-off-by: Yuval Shaia <yuval.shaia@oracle.com>
Reviewed-by: Marcel Apfelbaum <marcel.apfelbaum@gmail.com>
---
 hw/rdma/vmw/pvrdma_main.c | 6 ------
 1 file changed, 6 deletions(-)

diff --git a/hw/rdma/vmw/pvrdma_main.c b/hw/rdma/vmw/pvrdma_main.c
index b795f80666..4e4a43eac4 100644
--- a/hw/rdma/vmw/pvrdma_main.c
+++ b/hw/rdma/vmw/pvrdma_main.c
@@ -641,11 +641,6 @@ out:
     }
 }
 
-static void pvrdma_exit(PCIDevice *pdev)
-{
-    pvrdma_fini(pdev);
-}
-
 static void pvrdma_class_init(ObjectClass *klass, void *data)
 {
     DeviceClass *dc = DEVICE_CLASS(klass);
@@ -653,7 +648,6 @@ static void pvrdma_class_init(ObjectClass *klass, void *data)
     RdmaStatsProviderClass *ir = RDMA_STATS_PROVIDER_CLASS(klass);
 
     k->realize = pvrdma_realize;
-    k->exit = pvrdma_exit;
     k->vendor_id = PCI_VENDOR_ID_VMWARE;
     k->device_id = PCI_DEVICE_ID_VMWARE_PVRDMA;
     k->revision = 0x00;
-- 
2.17.2

[Qemu-devel] [PATCH v4 9/9] hw/pvrdma: Unregister from shutdown notifier when device goes down

[Yuval Shaia]

This hook was installed to close the device when VM is going down.
After the device is closed there is no need to be informed on VM
shutdown.

Signed-off-by: Yuval Shaia <yuval.shaia@oracle.com>
Reviewed-by: Marcel Apfelbaum <marcel.apfelbaum@gmail.com>
---
 hw/rdma/vmw/pvrdma_main.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/hw/rdma/vmw/pvrdma_main.c b/hw/rdma/vmw/pvrdma_main.c
index 4e4a43eac4..cb6eefe029 100644
--- a/hw/rdma/vmw/pvrdma_main.c
+++ b/hw/rdma/vmw/pvrdma_main.c
@@ -303,6 +303,8 @@ static void pvrdma_fini(PCIDevice *pdev)
 {
     PVRDMADev *dev = PVRDMA_DEV(pdev);
 
+    notifier_remove(&dev->shutdown_notifier);
+
     pvrdma_qp_ops_fini();
 
     rdma_backend_stop(&dev->backend_dev);
-- 
2.17.2