Merge tag 'rdma-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland...
authorLinus Torvalds <torvalds@linux-foundation.org>
Wed, 3 Oct 2012 00:20:40 +0000 (17:20 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Wed, 3 Oct 2012 00:20:40 +0000 (17:20 -0700)
Pull infiniband updates from Roland Dreier:
 "First batch of InfiniBand/RDMA changes for the 3.7 merge window:
   - mlx4 IB support for SR-IOV
   - A couple of SRP initiator fixes
   - Batch of nes hardware driver fixes
   - Fix for long-standing use-after-free crash in IPoIB
   - Other miscellaneous fixes"

This merge also removes a new use of __cancel_delayed_work(), and
replaces it with the regular cancel_delayed_work() that is now irq-safe
thanks to the workqueue updates.

That said, I suspect the sequence in question should probably use
"mod_delayed_work()".  I just did the minimal "don't use deprecated
functions" fixup, though.

* tag 'rdma-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland/infiniband: (45 commits)
  IB/qib: Fix local access validation for user MRs
  mlx4_core: Disable SENSE_PORT for multifunction devices
  mlx4_core: Clean up enabling of SENSE_PORT for older (ConnectX-1/-2) HCAs
  mlx4_core: Stash PCI ID driver_data in mlx4_priv structure
  IB/srp: Avoid having aborted requests hang
  IB/srp: Fix use-after-free in srp_reset_req()
  IB/qib: Add a qib driver version
  RDMA/nes: Fix compilation error when nes_debug is enabled
  RDMA/nes: Print hardware resource type
  RDMA/nes: Fix for crash when TX checksum offload is off
  RDMA/nes: Cosmetic changes
  RDMA/nes: Fix for incorrect MSS when TSO is on
  RDMA/nes: Fix incorrect resolving of the loopback MAC address
  mlx4_core: Fix crash on uninitialized priv->cmd.slave_sem
  mlx4_core: Trivial cleanups to driver log messages
  mlx4_core: Trivial readability fix: "0X30" -> "0x30"
  IB/mlx4: Create paravirt contexts for VFs when master IB driver initializes
  mlx4: Modify proxy/tunnel QP mechanism so that guests do no calculations
  mlx4: Paravirtualize Node Guids for slaves
  mlx4: Activate SR-IOV mode for IB
  ...

1  2 
drivers/infiniband/hw/cxgb4/qp.c
drivers/infiniband/hw/mlx4/alias_GUID.c
drivers/infiniband/hw/nes/nes_hw.c
drivers/infiniband/hw/nes/nes_nic.c
drivers/infiniband/ulp/ipoib/ipoib_main.c
drivers/net/ethernet/mellanox/mlx4/main.c

Simple merge
index 0000000,0fcd5cd..d2fb38d
mode 000000,100644..100644
--- /dev/null
@@@ -1,0 -1,688 +1,688 @@@
 -              __cancel_delayed_work(&dev->sriov.alias_guid.
+ /*
+  * Copyright (c) 2012 Mellanox Technologies. All rights reserved.
+  *
+  * This software is available to you under a choice of one of two
+  * licenses.  You may choose to be licensed under the terms of the GNU
+  * General Public License (GPL) Version 2, available from the file
+  * COPYING in the main directory of this source tree, or the
+  * OpenIB.org BSD license below:
+  *
+  *     Redistribution and use in source and binary forms, with or
+  *     without modification, are permitted provided that the following
+  *     conditions are met:
+  *
+  *      - Redistributions of source code must retain the above
+  *        copyright notice, this list of conditions and the following
+  *        disclaimer.
+  *
+  *      - Redistributions in binary form must reproduce the above
+  *        copyright notice, this list of conditions and the following
+  *        disclaimer in the documentation and/or other materials
+  *        provided with the distribution.
+  *
+  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+  * SOFTWARE.
+  */
+  /***********************************************************/
+ /*This file support the handling of the Alias GUID feature. */
+ /***********************************************************/
+ #include <rdma/ib_mad.h>
+ #include <rdma/ib_smi.h>
+ #include <rdma/ib_cache.h>
+ #include <rdma/ib_sa.h>
+ #include <rdma/ib_pack.h>
+ #include <linux/mlx4/cmd.h>
+ #include <linux/module.h>
+ #include <linux/init.h>
+ #include <linux/errno.h>
+ #include <rdma/ib_user_verbs.h>
+ #include <linux/delay.h>
+ #include "mlx4_ib.h"
+ /*
+ The driver keeps the current state of all guids, as they are in the HW.
+ Whenever we receive an smp mad GUIDInfo record, the data will be cached.
+ */
+ struct mlx4_alias_guid_work_context {
+       u8 port;
+       struct mlx4_ib_dev     *dev ;
+       struct ib_sa_query     *sa_query;
+       struct completion       done;
+       int                     query_id;
+       struct list_head        list;
+       int                     block_num;
+ };
+ struct mlx4_next_alias_guid_work {
+       u8 port;
+       u8 block_num;
+       struct mlx4_sriov_alias_guid_info_rec_det rec_det;
+ };
+ void mlx4_ib_update_cache_on_guid_change(struct mlx4_ib_dev *dev, int block_num,
+                                        u8 port_num, u8 *p_data)
+ {
+       int i;
+       u64 guid_indexes;
+       int slave_id;
+       int port_index = port_num - 1;
+       if (!mlx4_is_master(dev->dev))
+               return;
+       guid_indexes = be64_to_cpu((__force __be64) dev->sriov.alias_guid.
+                                  ports_guid[port_num - 1].
+                                  all_rec_per_port[block_num].guid_indexes);
+       pr_debug("port: %d, guid_indexes: 0x%llx\n", port_num, guid_indexes);
+       for (i = 0; i < NUM_ALIAS_GUID_IN_REC; i++) {
+               /* The location of the specific index starts from bit number 4
+                * until bit num 11 */
+               if (test_bit(i + 4, (unsigned long *)&guid_indexes)) {
+                       slave_id = (block_num * NUM_ALIAS_GUID_IN_REC) + i ;
+                       if (slave_id >= dev->dev->num_slaves) {
+                               pr_debug("The last slave: %d\n", slave_id);
+                               return;
+                       }
+                       /* cache the guid: */
+                       memcpy(&dev->sriov.demux[port_index].guid_cache[slave_id],
+                              &p_data[i * GUID_REC_SIZE],
+                              GUID_REC_SIZE);
+               } else
+                       pr_debug("Guid number: %d in block: %d"
+                                " was not updated\n", i, block_num);
+       }
+ }
+ static __be64 get_cached_alias_guid(struct mlx4_ib_dev *dev, int port, int index)
+ {
+       if (index >= NUM_ALIAS_GUID_PER_PORT) {
+               pr_err("%s: ERROR: asked for index:%d\n", __func__, index);
+               return  (__force __be64) ((u64) 0xFFFFFFFFFFFFFFFFUL);
+       }
+       return *(__be64 *)&dev->sriov.demux[port - 1].guid_cache[index];
+ }
+ ib_sa_comp_mask mlx4_ib_get_aguid_comp_mask_from_ix(int index)
+ {
+       return IB_SA_COMP_MASK(4 + index);
+ }
+ /*
+  * Whenever new GUID is set/unset (guid table change) create event and
+  * notify the relevant slave (master also should be notified).
+  * If the GUID value is not as we have in the cache the slave will not be
+  * updated; in this case it waits for the smp_snoop or the port management
+  * event to call the function and to update the slave.
+  * block_number - the index of the block (16 blocks available)
+  * port_number - 1 or 2
+  */
+ void mlx4_ib_notify_slaves_on_guid_change(struct mlx4_ib_dev *dev,
+                                         int block_num, u8 port_num,
+                                         u8 *p_data)
+ {
+       int i;
+       u64 guid_indexes;
+       int slave_id;
+       enum slave_port_state new_state;
+       enum slave_port_state prev_state;
+       __be64 tmp_cur_ag, form_cache_ag;
+       enum slave_port_gen_event gen_event;
+       if (!mlx4_is_master(dev->dev))
+               return;
+       guid_indexes = be64_to_cpu((__force __be64) dev->sriov.alias_guid.
+                                  ports_guid[port_num - 1].
+                                  all_rec_per_port[block_num].guid_indexes);
+       pr_debug("port: %d, guid_indexes: 0x%llx\n", port_num, guid_indexes);
+       /*calculate the slaves and notify them*/
+       for (i = 0; i < NUM_ALIAS_GUID_IN_REC; i++) {
+               /* the location of the specific index runs from bits 4..11 */
+               if (!(test_bit(i + 4, (unsigned long *)&guid_indexes)))
+                       continue;
+               slave_id = (block_num * NUM_ALIAS_GUID_IN_REC) + i ;
+               if (slave_id >= dev->dev->num_slaves)
+                       return;
+               tmp_cur_ag = *(__be64 *)&p_data[i * GUID_REC_SIZE];
+               form_cache_ag = get_cached_alias_guid(dev, port_num,
+                                       (NUM_ALIAS_GUID_IN_REC * block_num) + i);
+               /*
+                * Check if guid is not the same as in the cache,
+                * If it is different, wait for the snoop_smp or the port mgmt
+                * change event to update the slave on its port state change
+                */
+               if (tmp_cur_ag != form_cache_ag)
+                       continue;
+               mlx4_gen_guid_change_eqe(dev->dev, slave_id, port_num);
+               /*2 cases: Valid GUID, and Invalid Guid*/
+               if (tmp_cur_ag != MLX4_NOT_SET_GUID) { /*valid GUID*/
+                       prev_state = mlx4_get_slave_port_state(dev->dev, slave_id, port_num);
+                       new_state = set_and_calc_slave_port_state(dev->dev, slave_id, port_num,
+                                                                 MLX4_PORT_STATE_IB_PORT_STATE_EVENT_GID_VALID,
+                                                                 &gen_event);
+                       pr_debug("slave: %d, port: %d prev_port_state: %d,"
+                                " new_port_state: %d, gen_event: %d\n",
+                                slave_id, port_num, prev_state, new_state, gen_event);
+                       if (gen_event == SLAVE_PORT_GEN_EVENT_UP) {
+                               pr_debug("sending PORT_UP event to slave: %d, port: %d\n",
+                                        slave_id, port_num);
+                               mlx4_gen_port_state_change_eqe(dev->dev, slave_id,
+                                                              port_num, MLX4_PORT_CHANGE_SUBTYPE_ACTIVE);
+                       }
+               } else { /* request to invalidate GUID */
+                       set_and_calc_slave_port_state(dev->dev, slave_id, port_num,
+                                                     MLX4_PORT_STATE_IB_EVENT_GID_INVALID,
+                                                     &gen_event);
+                       pr_debug("sending PORT DOWN event to slave: %d, port: %d\n",
+                                slave_id, port_num);
+                       mlx4_gen_port_state_change_eqe(dev->dev, slave_id, port_num,
+                                                      MLX4_PORT_CHANGE_SUBTYPE_DOWN);
+               }
+       }
+ }
+ static void aliasguid_query_handler(int status,
+                                   struct ib_sa_guidinfo_rec *guid_rec,
+                                   void *context)
+ {
+       struct mlx4_ib_dev *dev;
+       struct mlx4_alias_guid_work_context *cb_ctx = context;
+       u8 port_index ;
+       int i;
+       struct mlx4_sriov_alias_guid_info_rec_det *rec;
+       unsigned long flags, flags1;
+       if (!context)
+               return;
+       dev = cb_ctx->dev;
+       port_index = cb_ctx->port - 1;
+       rec = &dev->sriov.alias_guid.ports_guid[port_index].
+               all_rec_per_port[cb_ctx->block_num];
+       if (status) {
+               rec->status = MLX4_GUID_INFO_STATUS_IDLE;
+               pr_debug("(port: %d) failed: status = %d\n",
+                        cb_ctx->port, status);
+               goto out;
+       }
+       if (guid_rec->block_num != cb_ctx->block_num) {
+               pr_err("block num mismatch: %d != %d\n",
+                      cb_ctx->block_num, guid_rec->block_num);
+               goto out;
+       }
+       pr_debug("lid/port: %d/%d, block_num: %d\n",
+                be16_to_cpu(guid_rec->lid), cb_ctx->port,
+                guid_rec->block_num);
+       rec = &dev->sriov.alias_guid.ports_guid[port_index].
+               all_rec_per_port[guid_rec->block_num];
+       rec->status = MLX4_GUID_INFO_STATUS_SET;
+       rec->method = MLX4_GUID_INFO_RECORD_SET;
+       for (i = 0 ; i < NUM_ALIAS_GUID_IN_REC; i++) {
+               __be64 tmp_cur_ag;
+               tmp_cur_ag = *(__be64 *)&guid_rec->guid_info_list[i * GUID_REC_SIZE];
+               /* check if the SM didn't assign one of the records.
+                * if it didn't, if it was not sysadmin request:
+                * ask the SM to give a new GUID, (instead of the driver request).
+                */
+               if (tmp_cur_ag == MLX4_NOT_SET_GUID) {
+                       mlx4_ib_warn(&dev->ib_dev, "%s:Record num %d in "
+                                    "block_num: %d was declined by SM, "
+                                    "ownership by %d (0 = driver, 1=sysAdmin,"
+                                    " 2=None)\n", __func__, i,
+                                    guid_rec->block_num, rec->ownership);
+                       if (rec->ownership == MLX4_GUID_DRIVER_ASSIGN) {
+                               /* if it is driver assign, asks for new GUID from SM*/
+                               *(__be64 *)&rec->all_recs[i * GUID_REC_SIZE] =
+                                       MLX4_NOT_SET_GUID;
+                               /* Mark the record as not assigned, and let it
+                                * be sent again in the next work sched.*/
+                               rec->status = MLX4_GUID_INFO_STATUS_IDLE;
+                               rec->guid_indexes |= mlx4_ib_get_aguid_comp_mask_from_ix(i);
+                       }
+               } else {
+                      /* properly assigned record. */
+                      /* We save the GUID we just got from the SM in the
+                       * admin_guid in order to be persistent, and in the
+                       * request from the sm the process will ask for the same GUID */
+                       if (rec->ownership == MLX4_GUID_SYSADMIN_ASSIGN &&
+                           tmp_cur_ag != *(__be64 *)&rec->all_recs[i * GUID_REC_SIZE]) {
+                               /* the sysadmin assignment failed.*/
+                               mlx4_ib_warn(&dev->ib_dev, "%s: Failed to set"
+                                            " admin guid after SysAdmin "
+                                            "configuration. "
+                                            "Record num %d in block_num:%d "
+                                            "was declined by SM, "
+                                            "new val(0x%llx) was kept\n",
+                                             __func__, i,
+                                            guid_rec->block_num,
+                                            be64_to_cpu(*(__be64 *) &
+                                                        rec->all_recs[i * GUID_REC_SIZE]));
+                       } else {
+                               memcpy(&rec->all_recs[i * GUID_REC_SIZE],
+                                      &guid_rec->guid_info_list[i * GUID_REC_SIZE],
+                                      GUID_REC_SIZE);
+                       }
+               }
+       }
+       /*
+       The func is call here to close the cases when the
+       sm doesn't send smp, so in the sa response the driver
+       notifies the slave.
+       */
+       mlx4_ib_notify_slaves_on_guid_change(dev, guid_rec->block_num,
+                                            cb_ctx->port,
+                                            guid_rec->guid_info_list);
+ out:
+       spin_lock_irqsave(&dev->sriov.going_down_lock, flags);
+       spin_lock_irqsave(&dev->sriov.alias_guid.ag_work_lock, flags1);
+       if (!dev->sriov.is_going_down)
+               queue_delayed_work(dev->sriov.alias_guid.ports_guid[port_index].wq,
+                                  &dev->sriov.alias_guid.ports_guid[port_index].
+                                  alias_guid_work, 0);
+       if (cb_ctx->sa_query) {
+               list_del(&cb_ctx->list);
+               kfree(cb_ctx);
+       } else
+               complete(&cb_ctx->done);
+       spin_unlock_irqrestore(&dev->sriov.alias_guid.ag_work_lock, flags1);
+       spin_unlock_irqrestore(&dev->sriov.going_down_lock, flags);
+ }
+ static void invalidate_guid_record(struct mlx4_ib_dev *dev, u8 port, int index)
+ {
+       int i;
+       u64 cur_admin_val;
+       ib_sa_comp_mask comp_mask = 0;
+       dev->sriov.alias_guid.ports_guid[port - 1].all_rec_per_port[index].status
+               = MLX4_GUID_INFO_STATUS_IDLE;
+       dev->sriov.alias_guid.ports_guid[port - 1].all_rec_per_port[index].method
+               = MLX4_GUID_INFO_RECORD_SET;
+       /* calculate the comp_mask for that record.*/
+       for (i = 0; i < NUM_ALIAS_GUID_IN_REC; i++) {
+               cur_admin_val =
+                       *(u64 *)&dev->sriov.alias_guid.ports_guid[port - 1].
+                       all_rec_per_port[index].all_recs[GUID_REC_SIZE * i];
+               /*
+               check the admin value: if it's for delete (~00LL) or
+               it is the first guid of the first record (hw guid) or
+               the records is not in ownership of the sysadmin and the sm doesn't
+               need to assign GUIDs, then don't put it up for assignment.
+               */
+               if (MLX4_GUID_FOR_DELETE_VAL == cur_admin_val ||
+                   (!index && !i) ||
+                   MLX4_GUID_NONE_ASSIGN == dev->sriov.alias_guid.
+                   ports_guid[port - 1].all_rec_per_port[index].ownership)
+                       continue;
+               comp_mask |= mlx4_ib_get_aguid_comp_mask_from_ix(i);
+       }
+       dev->sriov.alias_guid.ports_guid[port - 1].
+               all_rec_per_port[index].guid_indexes = comp_mask;
+ }
+ static int set_guid_rec(struct ib_device *ibdev,
+                       u8 port, int index,
+                       struct mlx4_sriov_alias_guid_info_rec_det *rec_det)
+ {
+       int err;
+       struct mlx4_ib_dev *dev = to_mdev(ibdev);
+       struct ib_sa_guidinfo_rec guid_info_rec;
+       ib_sa_comp_mask comp_mask;
+       struct ib_port_attr attr;
+       struct mlx4_alias_guid_work_context *callback_context;
+       unsigned long resched_delay, flags, flags1;
+       struct list_head *head =
+               &dev->sriov.alias_guid.ports_guid[port - 1].cb_list;
+       err = __mlx4_ib_query_port(ibdev, port, &attr, 1);
+       if (err) {
+               pr_debug("mlx4_ib_query_port failed (err: %d), port: %d\n",
+                        err, port);
+               return err;
+       }
+       /*check the port was configured by the sm, otherwise no need to send */
+       if (attr.state != IB_PORT_ACTIVE) {
+               pr_debug("port %d not active...rescheduling\n", port);
+               resched_delay = 5 * HZ;
+               err = -EAGAIN;
+               goto new_schedule;
+       }
+       callback_context = kmalloc(sizeof *callback_context, GFP_KERNEL);
+       if (!callback_context) {
+               err = -ENOMEM;
+               resched_delay = HZ * 5;
+               goto new_schedule;
+       }
+       callback_context->port = port;
+       callback_context->dev = dev;
+       callback_context->block_num = index;
+       memset(&guid_info_rec, 0, sizeof (struct ib_sa_guidinfo_rec));
+       guid_info_rec.lid = cpu_to_be16(attr.lid);
+       guid_info_rec.block_num = index;
+       memcpy(guid_info_rec.guid_info_list, rec_det->all_recs,
+              GUID_REC_SIZE * NUM_ALIAS_GUID_IN_REC);
+       comp_mask = IB_SA_GUIDINFO_REC_LID | IB_SA_GUIDINFO_REC_BLOCK_NUM |
+               rec_det->guid_indexes;
+       init_completion(&callback_context->done);
+       spin_lock_irqsave(&dev->sriov.alias_guid.ag_work_lock, flags1);
+       list_add_tail(&callback_context->list, head);
+       spin_unlock_irqrestore(&dev->sriov.alias_guid.ag_work_lock, flags1);
+       callback_context->query_id =
+               ib_sa_guid_info_rec_query(dev->sriov.alias_guid.sa_client,
+                                         ibdev, port, &guid_info_rec,
+                                         comp_mask, rec_det->method, 1000,
+                                         GFP_KERNEL, aliasguid_query_handler,
+                                         callback_context,
+                                         &callback_context->sa_query);
+       if (callback_context->query_id < 0) {
+               pr_debug("ib_sa_guid_info_rec_query failed, query_id: "
+                        "%d. will reschedule to the next 1 sec.\n",
+                        callback_context->query_id);
+               spin_lock_irqsave(&dev->sriov.alias_guid.ag_work_lock, flags1);
+               list_del(&callback_context->list);
+               kfree(callback_context);
+               spin_unlock_irqrestore(&dev->sriov.alias_guid.ag_work_lock, flags1);
+               resched_delay = 1 * HZ;
+               err = -EAGAIN;
+               goto new_schedule;
+       }
+       err = 0;
+       goto out;
+ new_schedule:
+       spin_lock_irqsave(&dev->sriov.going_down_lock, flags);
+       spin_lock_irqsave(&dev->sriov.alias_guid.ag_work_lock, flags1);
+       invalidate_guid_record(dev, port, index);
+       if (!dev->sriov.is_going_down) {
+               queue_delayed_work(dev->sriov.alias_guid.ports_guid[port - 1].wq,
+                                  &dev->sriov.alias_guid.ports_guid[port - 1].alias_guid_work,
+                                  resched_delay);
+       }
+       spin_unlock_irqrestore(&dev->sriov.alias_guid.ag_work_lock, flags1);
+       spin_unlock_irqrestore(&dev->sriov.going_down_lock, flags);
+ out:
+       return err;
+ }
+ void mlx4_ib_invalidate_all_guid_record(struct mlx4_ib_dev *dev, int port)
+ {
+       int i;
+       unsigned long flags, flags1;
+       pr_debug("port %d\n", port);
+       spin_lock_irqsave(&dev->sriov.going_down_lock, flags);
+       spin_lock_irqsave(&dev->sriov.alias_guid.ag_work_lock, flags1);
+       for (i = 0; i < NUM_ALIAS_GUID_REC_IN_PORT; i++)
+               invalidate_guid_record(dev, port, i);
+       if (mlx4_is_master(dev->dev) && !dev->sriov.is_going_down) {
+               /*
+               make sure no work waits in the queue, if the work is already
+               queued(not on the timer) the cancel will fail. That is not a problem
+               because we just want the work started.
+               */
++              cancel_delayed_work(&dev->sriov.alias_guid.
+                                     ports_guid[port - 1].alias_guid_work);
+               queue_delayed_work(dev->sriov.alias_guid.ports_guid[port - 1].wq,
+                                  &dev->sriov.alias_guid.ports_guid[port - 1].alias_guid_work,
+                                  0);
+       }
+       spin_unlock_irqrestore(&dev->sriov.alias_guid.ag_work_lock, flags1);
+       spin_unlock_irqrestore(&dev->sriov.going_down_lock, flags);
+ }
+ /* The function returns the next record that was
+  * not configured (or failed to be configured) */
+ static int get_next_record_to_update(struct mlx4_ib_dev *dev, u8 port,
+                                    struct mlx4_next_alias_guid_work *rec)
+ {
+       int j;
+       unsigned long flags;
+       for (j = 0; j < NUM_ALIAS_GUID_REC_IN_PORT; j++) {
+               spin_lock_irqsave(&dev->sriov.alias_guid.ag_work_lock, flags);
+               if (dev->sriov.alias_guid.ports_guid[port].all_rec_per_port[j].status ==
+                   MLX4_GUID_INFO_STATUS_IDLE) {
+                       memcpy(&rec->rec_det,
+                              &dev->sriov.alias_guid.ports_guid[port].all_rec_per_port[j],
+                              sizeof (struct mlx4_sriov_alias_guid_info_rec_det));
+                       rec->port = port;
+                       rec->block_num = j;
+                       dev->sriov.alias_guid.ports_guid[port].all_rec_per_port[j].status =
+                               MLX4_GUID_INFO_STATUS_PENDING;
+                       spin_unlock_irqrestore(&dev->sriov.alias_guid.ag_work_lock, flags);
+                       return 0;
+               }
+               spin_unlock_irqrestore(&dev->sriov.alias_guid.ag_work_lock, flags);
+       }
+       return -ENOENT;
+ }
+ static void set_administratively_guid_record(struct mlx4_ib_dev *dev, int port,
+                                            int rec_index,
+                                            struct mlx4_sriov_alias_guid_info_rec_det *rec_det)
+ {
+       dev->sriov.alias_guid.ports_guid[port].all_rec_per_port[rec_index].guid_indexes =
+               rec_det->guid_indexes;
+       memcpy(dev->sriov.alias_guid.ports_guid[port].all_rec_per_port[rec_index].all_recs,
+              rec_det->all_recs, NUM_ALIAS_GUID_IN_REC * GUID_REC_SIZE);
+       dev->sriov.alias_guid.ports_guid[port].all_rec_per_port[rec_index].status =
+               rec_det->status;
+ }
+ static void set_all_slaves_guids(struct mlx4_ib_dev *dev, int port)
+ {
+       int j;
+       struct mlx4_sriov_alias_guid_info_rec_det rec_det ;
+       for (j = 0 ; j < NUM_ALIAS_GUID_REC_IN_PORT ; j++) {
+               memset(rec_det.all_recs, 0, NUM_ALIAS_GUID_IN_REC * GUID_REC_SIZE);
+               rec_det.guid_indexes = (!j ? 0 : IB_SA_GUIDINFO_REC_GID0) |
+                       IB_SA_GUIDINFO_REC_GID1 | IB_SA_GUIDINFO_REC_GID2 |
+                       IB_SA_GUIDINFO_REC_GID3 | IB_SA_GUIDINFO_REC_GID4 |
+                       IB_SA_GUIDINFO_REC_GID5 | IB_SA_GUIDINFO_REC_GID6 |
+                       IB_SA_GUIDINFO_REC_GID7;
+               rec_det.status = MLX4_GUID_INFO_STATUS_IDLE;
+               set_administratively_guid_record(dev, port, j, &rec_det);
+       }
+ }
+ static void alias_guid_work(struct work_struct *work)
+ {
+       struct delayed_work *delay = to_delayed_work(work);
+       int ret = 0;
+       struct mlx4_next_alias_guid_work *rec;
+       struct mlx4_sriov_alias_guid_port_rec_det *sriov_alias_port =
+               container_of(delay, struct mlx4_sriov_alias_guid_port_rec_det,
+                            alias_guid_work);
+       struct mlx4_sriov_alias_guid *sriov_alias_guid = sriov_alias_port->parent;
+       struct mlx4_ib_sriov *ib_sriov = container_of(sriov_alias_guid,
+                                               struct mlx4_ib_sriov,
+                                               alias_guid);
+       struct mlx4_ib_dev *dev = container_of(ib_sriov, struct mlx4_ib_dev, sriov);
+       rec = kzalloc(sizeof *rec, GFP_KERNEL);
+       if (!rec) {
+               pr_err("alias_guid_work: No Memory\n");
+               return;
+       }
+       pr_debug("starting [port: %d]...\n", sriov_alias_port->port + 1);
+       ret = get_next_record_to_update(dev, sriov_alias_port->port, rec);
+       if (ret) {
+               pr_debug("No more records to update.\n");
+               goto out;
+       }
+       set_guid_rec(&dev->ib_dev, rec->port + 1, rec->block_num,
+                    &rec->rec_det);
+ out:
+       kfree(rec);
+ }
+ void mlx4_ib_init_alias_guid_work(struct mlx4_ib_dev *dev, int port)
+ {
+       unsigned long flags, flags1;
+       if (!mlx4_is_master(dev->dev))
+               return;
+       spin_lock_irqsave(&dev->sriov.going_down_lock, flags);
+       spin_lock_irqsave(&dev->sriov.alias_guid.ag_work_lock, flags1);
+       if (!dev->sriov.is_going_down) {
+               queue_delayed_work(dev->sriov.alias_guid.ports_guid[port].wq,
+                          &dev->sriov.alias_guid.ports_guid[port].alias_guid_work, 0);
+       }
+       spin_unlock_irqrestore(&dev->sriov.alias_guid.ag_work_lock, flags1);
+       spin_unlock_irqrestore(&dev->sriov.going_down_lock, flags);
+ }
+ void mlx4_ib_destroy_alias_guid_service(struct mlx4_ib_dev *dev)
+ {
+       int i;
+       struct mlx4_ib_sriov *sriov = &dev->sriov;
+       struct mlx4_alias_guid_work_context *cb_ctx;
+       struct mlx4_sriov_alias_guid_port_rec_det *det;
+       struct ib_sa_query *sa_query;
+       unsigned long flags;
+       for (i = 0 ; i < dev->num_ports; i++) {
+               cancel_delayed_work(&dev->sriov.alias_guid.ports_guid[i].alias_guid_work);
+               det = &sriov->alias_guid.ports_guid[i];
+               spin_lock_irqsave(&sriov->alias_guid.ag_work_lock, flags);
+               while (!list_empty(&det->cb_list)) {
+                       cb_ctx = list_entry(det->cb_list.next,
+                                           struct mlx4_alias_guid_work_context,
+                                           list);
+                       sa_query = cb_ctx->sa_query;
+                       cb_ctx->sa_query = NULL;
+                       list_del(&cb_ctx->list);
+                       spin_unlock_irqrestore(&sriov->alias_guid.ag_work_lock, flags);
+                       ib_sa_cancel_query(cb_ctx->query_id, sa_query);
+                       wait_for_completion(&cb_ctx->done);
+                       kfree(cb_ctx);
+                       spin_lock_irqsave(&sriov->alias_guid.ag_work_lock, flags);
+               }
+               spin_unlock_irqrestore(&sriov->alias_guid.ag_work_lock, flags);
+       }
+       for (i = 0 ; i < dev->num_ports; i++) {
+               flush_workqueue(dev->sriov.alias_guid.ports_guid[i].wq);
+               destroy_workqueue(dev->sriov.alias_guid.ports_guid[i].wq);
+       }
+       ib_sa_unregister_client(dev->sriov.alias_guid.sa_client);
+       kfree(dev->sriov.alias_guid.sa_client);
+ }
+ int mlx4_ib_init_alias_guid_service(struct mlx4_ib_dev *dev)
+ {
+       char alias_wq_name[15];
+       int ret = 0;
+       int i, j, k;
+       union ib_gid gid;
+       if (!mlx4_is_master(dev->dev))
+               return 0;
+       dev->sriov.alias_guid.sa_client =
+               kzalloc(sizeof *dev->sriov.alias_guid.sa_client, GFP_KERNEL);
+       if (!dev->sriov.alias_guid.sa_client)
+               return -ENOMEM;
+       ib_sa_register_client(dev->sriov.alias_guid.sa_client);
+       spin_lock_init(&dev->sriov.alias_guid.ag_work_lock);
+       for (i = 1; i <= dev->num_ports; ++i) {
+               if (dev->ib_dev.query_gid(&dev->ib_dev , i, 0, &gid)) {
+                       ret = -EFAULT;
+                       goto err_unregister;
+               }
+       }
+       for (i = 0 ; i < dev->num_ports; i++) {
+               memset(&dev->sriov.alias_guid.ports_guid[i], 0,
+                      sizeof (struct mlx4_sriov_alias_guid_port_rec_det));
+               /*Check if the SM doesn't need to assign the GUIDs*/
+               for (j = 0; j < NUM_ALIAS_GUID_REC_IN_PORT; j++) {
+                       if (mlx4_ib_sm_guid_assign) {
+                               dev->sriov.alias_guid.ports_guid[i].
+                                       all_rec_per_port[j].
+                                       ownership = MLX4_GUID_DRIVER_ASSIGN;
+                               continue;
+                       }
+                       dev->sriov.alias_guid.ports_guid[i].all_rec_per_port[j].
+                                       ownership = MLX4_GUID_NONE_ASSIGN;
+                       /*mark each val as it was deleted,
+                         till the sysAdmin will give it valid val*/
+                       for (k = 0; k < NUM_ALIAS_GUID_IN_REC; k++) {
+                               *(__be64 *)&dev->sriov.alias_guid.ports_guid[i].
+                                       all_rec_per_port[j].all_recs[GUID_REC_SIZE * k] =
+                                               cpu_to_be64(MLX4_GUID_FOR_DELETE_VAL);
+                       }
+               }
+               INIT_LIST_HEAD(&dev->sriov.alias_guid.ports_guid[i].cb_list);
+               /*prepare the records, set them to be allocated by sm*/
+               for (j = 0 ; j < NUM_ALIAS_GUID_REC_IN_PORT; j++)
+                       invalidate_guid_record(dev, i + 1, j);
+               dev->sriov.alias_guid.ports_guid[i].parent = &dev->sriov.alias_guid;
+               dev->sriov.alias_guid.ports_guid[i].port  = i;
+               if (mlx4_ib_sm_guid_assign)
+                       set_all_slaves_guids(dev, i);
+               snprintf(alias_wq_name, sizeof alias_wq_name, "alias_guid%d", i);
+               dev->sriov.alias_guid.ports_guid[i].wq =
+                       create_singlethread_workqueue(alias_wq_name);
+               if (!dev->sriov.alias_guid.ports_guid[i].wq) {
+                       ret = -ENOMEM;
+                       goto err_thread;
+               }
+               INIT_DELAYED_WORK(&dev->sriov.alias_guid.ports_guid[i].alias_guid_work,
+                         alias_guid_work);
+       }
+       return 0;
+ err_thread:
+       for (--i; i >= 0; i--) {
+               destroy_workqueue(dev->sriov.alias_guid.ports_guid[i].wq);
+               dev->sriov.alias_guid.ports_guid[i].wq = NULL;
+       }
+ err_unregister:
+       ib_sa_unregister_client(dev->sriov.alias_guid.sa_client);
+       kfree(dev->sriov.alias_guid.sa_client);
+       dev->sriov.alias_guid.sa_client = NULL;
+       pr_err("init_alias_guid_service: Failed. (ret:%d)\n", ret);
+       return ret;
+ }
Simple merge
Simple merge