xeon-phi: update for the xeon phi patches

In ibp cm server change list modification patterns to be atomic. Before adding and removing
from a list would go something like:

1)      Find item in list while holding reader lock
2)      …
3)      Search again for found item while holding writer lock, remove it

Multiple threads could end up holding pointers to items in the list and by the time they get to
deleting it, it may or may not exist in list.

Also: the ibp cm event handler for cm events should always return 0, if it returns non-zero ib_cm
will attempt to remove the cm_id, which is not correct (and causes ib_cm worker thread to dump
stack and hang).

Finally: trivial change an ibscif printk log level to KERN_INFO vs ALERT.

Signed-off-by: Patrick McCormick <patrick.m.mccormick@intel.com>
Signed-off-by: Vladimir Sokolovsky <vlad@mellanox.com>

diff --git a/tech-preview/xeon-phi/0003-add-the-ibp-client-and-server-drivers.patch b/tech-preview/xeon-phi/0003-add-the-ibp-client-and-server-drivers.patch
index 66e736e1f9afbafe55cd3a2046ed13657bfa9e79..88ae06e576ed2280820ec2b91d1b6a70f19f9a1a 100644 (file)
--- a/tech-preview/xeon-phi/0003-add-the-ibp-client-and-server-drivers.patch
+++ b/tech-preview/xeon-phi/0003-add-the-ibp-client-and-server-drivers.patch
@@ -1,4 +1,4 @@
-From 9f6bb49327fb7179785b10a845cf5070494bd988 Mon Sep 17 00:00:00 2001
+From 3da8ce605b1942b5a7705f10cbadd2afe3b5b810 Mon Sep 17 00:00:00 2001
 From: Jerrie Coffman <jerrie.l.coffman@intel.com>
 Date: Fri, 20 May 2016 16:17:43 -0700
 Subject: [PATCH 3/7] add the ibp client and server drivers
@@ -8,6 +8,7 @@ infiniband.  It includes the base ib proxy driver, the mlx4
 and mlx5 client drivers, and sa and cm proxy drivers.
 
 Signed-off-by: Jerrie Coffman <jerrie.l.coffman@intel.com>
+Signed-off-by: Patrick McCormick <patrick.m.mccorick@intel.com>
 ---
  drivers/infiniband/Kconfig                  |    2 +
  drivers/infiniband/Makefile                 |    2 +
@@ -21,12 +22,12 @@ Signed-off-by: Jerrie Coffman <jerrie.l.coffman@intel.com>
  drivers/infiniband/ibp/cm/cm_client_msg.c   |  785 ++++++++
  drivers/infiniband/ibp/cm/cm_ibp_abi.h      |  405 ++++
  drivers/infiniband/ibp/cm/cm_proxy.c        |  752 +++++++
- drivers/infiniband/ibp/cm/cm_server_msg.c   | 1044 ++++++++++
+ drivers/infiniband/ibp/cm/cm_server_msg.c   | 1065 ++++++++++
  drivers/infiniband/ibp/cm/common.h          |   55 +
  drivers/infiniband/ibp/cm/ibp-abi.h         |   94 +
  drivers/infiniband/ibp/cm/ibp_exports.h     |   50 +
  drivers/infiniband/ibp/cm/server.c          |  221 +++
- drivers/infiniband/ibp/cm/server.h          |  129 ++
+ drivers/infiniband/ibp/cm/server.h          |  128 ++
  drivers/infiniband/ibp/cm/server_msg.c      |  176 ++
  drivers/infiniband/ibp/compat.h             |  101 +
  drivers/infiniband/ibp/drv/Makefile         |   10 +
@@ -71,7 +72,7 @@ Signed-off-by: Jerrie Coffman <jerrie.l.coffman@intel.com>
  drivers/infiniband/ibp/sa/server.c          |  218 +++
  drivers/infiniband/ibp/sa/server.h          |  173 ++
  drivers/infiniband/ibp/sa/server_msg.c      |  185 ++
- 62 files changed, 22315 insertions(+)
+ 62 files changed, 22335 insertions(+)
  create mode 100644 drivers/infiniband/ibp/Kconfig
  create mode 100644 drivers/infiniband/ibp/Makefile
  create mode 100644 drivers/infiniband/ibp/cm/Makefile
@@ -2770,10 +2771,10 @@ index 0000000..f08608e
 +EXPORT_SYMBOL(ib_cm_init_qp_attr);
 diff --git a/drivers/infiniband/ibp/cm/cm_server_msg.c b/drivers/infiniband/ibp/cm/cm_server_msg.c
 new file mode 100644
-index 0000000..884be47
+index 0000000..02b3fb3
 --- /dev/null
 +++ b/drivers/infiniband/ibp/cm/cm_server_msg.c
-@@ -0,0 +1,1044 @@
+@@ -0,0 +1,1065 @@
 +/*
 + * Copyright (c) 2011-2013 Intel Corporation. All rights reserved.
 + *
@@ -2866,19 +2867,9 @@ index 0000000..884be47
 +      a->preference                   = b->preference;
 +}
 +
-+void cleanup_cm_entry_list(void)
-+{
-+      struct cm_entry *entry, *next;
-+
-+      down_write(&list_rwsem);
-+      list_for_each_entry_safe(entry, next, &cm_entry_list, list)
-+              kfree(entry);
-+      up_write(&list_rwsem);
-+}
-+
 +static struct cm_entry *find_cm_entry(struct ib_cm_id *cm_id)
 +{
-+      struct cm_entry *entry;
++      struct cm_entry *entry;
 +
 +      down_read(&list_rwsem);
 +
@@ -2888,39 +2879,72 @@ index 0000000..884be47
 +
 +      print_err("Could not find cm id %p\n", cm_id);
 +      entry = NULL;
++
 +out:
 +      up_read(&list_rwsem);
 +
 +      return entry;
 +}
 +
++static struct cm_entry *remove_cm_entry(struct ib_cm_id *cm_id)
++{
++      struct cm_entry                         *entry, *found = NULL;
++
++      down_write(&list_rwsem);
++
++      list_for_each_entry(entry, &cm_entry_list, list)
++              if (entry->cm_id == cm_id) {
++                      found = entry;
++                      break;
++              }
++
++      if (!found) {
++              print_err("could not find cm id %p\n", cm_id);
++      } else {
++              list_del(&found->list);
++      }
++
++      up_write(&list_rwsem);
++
++      return found;
++}
++
 +/* find the entry id for the listen cm id so we can add the new cm id
 + * that is being accepted to the list so it can be found on future events
 + */
 +static struct cm_entry *find_cm_entry_and_add(struct ib_cm_id *listen_id,
 +                                            struct ib_cm_id *cm_id)
 +{
-+      struct cm_entry *listen_entry, *entry;
++      struct cm_entry                         *entry;
++      struct cm_entry                         *listen_entry = NULL;
++
++      down_write(&list_rwsem);
++
++      list_for_each_entry(entry, &cm_entry_list, list)
++              if (entry->cm_id == listen_id) {
++                      listen_entry = entry;
++                      break;
++              }
 +
-+      listen_entry = find_cm_entry(listen_id);
 +      if (!listen_entry) {
 +              print_err("Could not find listen id %p\n", listen_id);
-+              return NULL;
++              goto out;
 +      }
 +
 +      entry = kzalloc(sizeof(struct cm_entry), GFP_KERNEL);
 +      if (!entry) {
 +              print_err("kzalloc failed\n");
-+              return NULL;
++              listen_entry = NULL;
++              goto out;
 +      }
 +
 +      entry->client = listen_entry->client;
 +      entry->cm_id = cm_id;
 +
-+      down_write(&list_rwsem);
 +      list_add(&entry->list, &cm_entry_list);
-+      up_write(&list_rwsem);
 +
++out:
++      up_write(&list_rwsem);
 +      return listen_entry;
 +}
 +
@@ -3063,7 +3087,7 @@ index 0000000..884be47
 +                           data_length + info_length, GFP_KERNEL);
 +      if (!event_work) {
 +              print_err("kzalloc failed\n");
-+              return -ENOMEM;
++              return 0;
 +      }
 +
 +      if (ib_cm_event->event == IB_CM_REQ_RECEIVED) {
@@ -3078,8 +3102,9 @@ index 0000000..884be47
 +              entry = find_cm_entry(cm_id);
 +
 +      if (!entry) {
++              print_err("event (%d) on non-existent id: %p\n", ib_cm_event->event, cm_id);
 +              kfree(event_work);
-+              return -EINVAL;
++              return 0;
 +      }
 +
 +      client = entry->client;
@@ -3280,16 +3305,13 @@ index 0000000..884be47
 +      msg     = (struct ibp_response_msg *) client->tx_buf;
 +      len     = sizeof(*msg);
 +
-+      entry = find_cm_entry(cm_id);
-+      if (!entry)
++      entry = remove_cm_entry(cm_id);
++      if (!entry) {
++              ret = -EINVAL;
 +              goto send_resp;
-+
-+      down_write(&list_rwsem);
-+      list_del(&entry->list);
-+      up_write(&list_rwsem);
++      }
 +
 +      kfree(entry);
-+
 +      ib_destroy_cm_id(cm_id);
 +
 +send_resp:
@@ -4264,10 +4286,10 @@ index 0000000..08fe284
 +module_exit(ibp_cm_server_exit);
 diff --git a/drivers/infiniband/ibp/cm/server.h b/drivers/infiniband/ibp/cm/server.h
 new file mode 100644
-index 0000000..cd71a90
+index 0000000..362e4e4a
 --- /dev/null
 +++ b/drivers/infiniband/ibp/cm/server.h
-@@ -0,0 +1,129 @@
+@@ -0,0 +1,128 @@
 +/*
 + * Copyright (c) 2011-2013 Intel Corporation. All rights reserved.
 + *
@@ -4372,7 +4394,6 @@ index 0000000..cd71a90
 +      } while (0)
 +
 +int ibp_process_recvs(void *p);
-+void cleanup_cm_entry_list(void);
 +
 +int ibp_cmd_create_cm_id(struct ibp_client *client, struct ibp_msg_header *hdr);
 +int ibp_cmd_destroy_cm_id(struct ibp_client *client,
@@ -22828,5 +22849,5 @@ index 0000000..2d396d8
 +      return ret;
 +}
 -- 
-2.7.0
+1.8.3.1
 

diff --git a/tech-preview/xeon-phi/0006-Add-ibscif-driver.patch b/tech-preview/xeon-phi/0006-Add-ibscif-driver.patch
index 96fea44ac2a74735294de715c0de8360c04dcc52..2e1b98dce6e0629273d770a18ded584c5563eade 100644 (file)
--- a/tech-preview/xeon-phi/0006-Add-ibscif-driver.patch
+++ b/tech-preview/xeon-phi/0006-Add-ibscif-driver.patch
@@ -1,4 +1,4 @@
-From a5a092dd4b1ec62e32643dcb2f51153aa51ce491 Mon Sep 17 00:00:00 2001
+From 0cb3f19158193f0dbf649b0b68c95b42498c02eb Mon Sep 17 00:00:00 2001
 From: Jerrie Coffman <jerrie.l.coffman@intel.com>
 Date: Thu, 1 Sep 2016 14:39:08 -0700
 Subject: [PATCH 6/7] Add ibscif driver
@@ -6943,7 +6943,7 @@ index 0000000..3ce5763
 +#endif /* IBSCIF_PROTOCOL_H */
 diff --git a/drivers/infiniband/hw/scif/ibscif_provider.c b/drivers/infiniband/hw/scif/ibscif_provider.c
 new file mode 100644
-index 0000000..9954532
+index 0000000..f17404b
 --- /dev/null
 +++ b/drivers/infiniband/hw/scif/ibscif_provider.c
 @@ -0,0 +1,424 @@
@@ -7279,7 +7279,7 @@ index 0000000..9954532
 +
 +      node_cnt = ret;
 +      dev->node_id = my_node_id;
-+      printk(KERN_ALERT PFX "%s: my node_id is %d\n", __func__, dev->node_id);
++      printk(KERN_INFO PFX "%s: my node_id is %d\n", __func__, dev->node_id);
 +
 +      ret = scif_bind(dev->listen_ep, SCIF_OFED_PORT_0);
 +      if (ret < 0) {