The size of mlx4_ib_device became too large to be allocated as whole contigous
block of memory. Currently it takes about 55K. On architecture with 4K page it
means 3rd order.
This patch series makes an attempt to split mlx4_ib_device into several parts
and allocate them with less expensive kvzalloc
Jan Dakinevich (4):
IB/core: introduce ->release() callback
IB/mlx4: move iboe field aside from mlx4_ib_dev
IB/mlx4: move pkeys field aside from mlx4_ib_dev
IB/mlx4: move sriov field aside from mlx4_ib_dev
drivers/infiniband/core/device.c | 2 +
drivers/infiniband/hw/mlx4/alias_GUID.c | 192 ++++++++++++++++----------------
drivers/infiniband/hw/mlx4/cm.c | 32 +++---
drivers/infiniband/hw/mlx4/mad.c | 98 ++++++++--------
drivers/infiniband/hw/mlx4/main.c | 93 ++++++++++------
drivers/infiniband/hw/mlx4/mcg.c | 4 +-
drivers/infiniband/hw/mlx4/mlx4_ib.h | 8 +-
drivers/infiniband/hw/mlx4/qp.c | 8 +-
drivers/infiniband/hw/mlx4/sysfs.c | 40 +++----
include/rdma/ib_verbs.h | 2 +
10 files changed, 256 insertions(+), 223 deletions(-)
--
2.1.4
IB infrastructure shares common device instance constructor with
reference counting, and it uses kzalloc() to allocate memory
for device specific instance with incapsulated ib_device field as one
contigous memory block.
The issue is that the device specific instances tend to be too large
and require high page order memory allocation. Unfortunately, kzalloc()
in ib_alloc_device() can not be replaced with kvzalloc() since it would
require a lot of review in all IB driver to prove correctness of the
replacement.
The driver can allocate some heavy partes of their instance for itself
and keep pointers for them in own instance. For this it is important
that the alocated parts have the same life time as ib_device, thus
their deallocation should be based on the same reference counting.
Let suppose:
struct foo_ib_device {
struct ib_device device;
void *part;
...
};
To properly free memory from .foo_ib_part the driver should provide
function for ->release() callback:
void foo_ib_release(struct ib_device *device)
{
struct foo_ib_device *foo = container_of(device, struct foo_ib_device,
device);
kvfree(foo->part);
}
...and initialiaze this callback immediately after foo_ib_device
instance allocation.
struct foo_ib_device *foo;
foo = ib_alloc_device(sizeof(struct foo_ib_device));
foo->device.release = foo_ib_release;
/* allocate parts */
foo->part = kvmalloc(65536, GFP_KERNEL);
Signed-off-by: Jan Dakinevich <[email protected]>
---
drivers/infiniband/core/device.c | 2 ++
include/rdma/ib_verbs.h | 2 ++
2 files changed, 4 insertions(+)
diff --git a/drivers/infiniband/core/device.c b/drivers/infiniband/core/device.c
index db3b627..a8c8b0d 100644
--- a/drivers/infiniband/core/device.c
+++ b/drivers/infiniband/core/device.c
@@ -215,6 +215,8 @@ static void ib_device_release(struct device *device)
ib_cache_release_one(dev);
kfree(dev->port_immutable);
}
+ if (dev->release)
+ dev->release(dev);
kfree(dev);
}
diff --git a/include/rdma/ib_verbs.h b/include/rdma/ib_verbs.h
index e950c2a..fb582bb 100644
--- a/include/rdma/ib_verbs.h
+++ b/include/rdma/ib_verbs.h
@@ -2271,6 +2271,8 @@ struct ib_device {
struct iw_cm_verbs *iwcm;
+ void (*release)(struct ib_device *device);
+
/**
* alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
* driver initialized data. The struct is kfree()'ed by the sysfs
--
2.1.4
This is the 2nd patch of 3 of the work for decreasing size
of mlx4_ib_dev.
The field takes about 36K and could be safely allocated with kvzalloc.
Signed-off-by: Jan Dakinevich <[email protected]>
---
drivers/infiniband/hw/mlx4/mad.c | 18 +++++++++---------
drivers/infiniband/hw/mlx4/main.c | 11 ++++++++---
drivers/infiniband/hw/mlx4/mlx4_ib.h | 2 +-
drivers/infiniband/hw/mlx4/sysfs.c | 30 +++++++++++++++---------------
4 files changed, 33 insertions(+), 28 deletions(-)
diff --git a/drivers/infiniband/hw/mlx4/mad.c b/drivers/infiniband/hw/mlx4/mad.c
index e5466d7..3eceb46 100644
--- a/drivers/infiniband/hw/mlx4/mad.c
+++ b/drivers/infiniband/hw/mlx4/mad.c
@@ -268,9 +268,9 @@ static void smp_snoop(struct ib_device *ibdev, u8 port_num, const struct ib_mad
pr_debug("PKEY[%d] = x%x\n",
i + bn*32, be16_to_cpu(base[i]));
if (be16_to_cpu(base[i]) !=
- dev->pkeys.phys_pkey_cache[port_num - 1][i + bn*32]) {
+ dev->pkeys->phys_pkey_cache[port_num - 1][i + bn*32]) {
pkey_change_bitmap |= (1 << i);
- dev->pkeys.phys_pkey_cache[port_num - 1][i + bn*32] =
+ dev->pkeys->phys_pkey_cache[port_num - 1][i + bn*32] =
be16_to_cpu(base[i]);
}
}
@@ -348,7 +348,7 @@ static void __propagate_pkey_ev(struct mlx4_ib_dev *dev, int port_num,
continue;
for (ix = 0;
ix < dev->dev->caps.pkey_table_len[port_num]; ix++) {
- if (dev->pkeys.virt2phys_pkey[slave][port_num - 1]
+ if (dev->pkeys->virt2phys_pkey[slave][port_num - 1]
[ix] == i + 32 * block) {
err = mlx4_gen_pkey_eqe(dev->dev, slave, port_num);
pr_debug("propagate_pkey_ev: slave %d,"
@@ -455,10 +455,10 @@ static int find_slave_port_pkey_ix(struct mlx4_ib_dev *dev, int slave,
unassigned_pkey_ix = dev->dev->phys_caps.pkey_phys_table_len[port] - 1;
for (i = 0; i < dev->dev->caps.pkey_table_len[port]; i++) {
- if (dev->pkeys.virt2phys_pkey[slave][port - 1][i] == unassigned_pkey_ix)
+ if (dev->pkeys->virt2phys_pkey[slave][port - 1][i] == unassigned_pkey_ix)
continue;
- pkey_ix = dev->pkeys.virt2phys_pkey[slave][port - 1][i];
+ pkey_ix = dev->pkeys->virt2phys_pkey[slave][port - 1][i];
ret = ib_get_cached_pkey(&dev->ib_dev, port, pkey_ix, &slot_pkey);
if (ret)
@@ -546,7 +546,7 @@ int mlx4_ib_send_to_slave(struct mlx4_ib_dev *dev, int slave, u8 port,
return -EINVAL;
tun_pkey_ix = pkey_ix;
} else
- tun_pkey_ix = dev->pkeys.virt2phys_pkey[slave][port - 1][0];
+ tun_pkey_ix = dev->pkeys->virt2phys_pkey[slave][port - 1][0];
dqpn = dev->dev->phys_caps.base_proxy_sqpn + 8 * slave + port + (dest_qpt * 2) - 1;
@@ -1382,11 +1382,11 @@ int mlx4_ib_send_to_wire(struct mlx4_ib_dev *dev, int slave, u8 port,
if (dest_qpt == IB_QPT_SMI) {
src_qpnum = 0;
sqp = &sqp_ctx->qp[0];
- wire_pkey_ix = dev->pkeys.virt2phys_pkey[slave][port - 1][0];
+ wire_pkey_ix = dev->pkeys->virt2phys_pkey[slave][port - 1][0];
} else {
src_qpnum = 1;
sqp = &sqp_ctx->qp[1];
- wire_pkey_ix = dev->pkeys.virt2phys_pkey[slave][port - 1][pkey_index];
+ wire_pkey_ix = dev->pkeys->virt2phys_pkey[slave][port - 1][pkey_index];
}
send_qp = sqp->qp;
@@ -1840,7 +1840,7 @@ static int create_pv_sqp(struct mlx4_ib_demux_pv_ctx *ctx,
&attr.pkey_index);
if (ret || !create_tun)
attr.pkey_index =
- to_mdev(ctx->ib_dev)->pkeys.virt2phys_pkey[ctx->slave][ctx->port - 1][0];
+ to_mdev(ctx->ib_dev)->pkeys->virt2phys_pkey[ctx->slave][ctx->port - 1][0];
attr.qkey = IB_QP1_QKEY;
attr.port_num = ctx->port;
ret = ib_modify_qp(tun_qp->qp, &attr, qp_attr_mask_INIT);
diff --git a/drivers/infiniband/hw/mlx4/main.c b/drivers/infiniband/hw/mlx4/main.c
index 1e3bb67..8ba0103 100644
--- a/drivers/infiniband/hw/mlx4/main.c
+++ b/drivers/infiniband/hw/mlx4/main.c
@@ -2477,12 +2477,12 @@ static void init_pkeys(struct mlx4_ib_dev *ibdev)
for (i = 0;
i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
++i) {
- ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
+ ibdev->pkeys->virt2phys_pkey[slave][port - 1][i] =
/* master has the identity virt2phys pkey mapping */
(slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
- ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
+ ibdev->pkeys->virt2phys_pkey[slave][port - 1][i]);
}
}
}
@@ -2491,7 +2491,7 @@ static void init_pkeys(struct mlx4_ib_dev *ibdev)
for (i = 0;
i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
++i)
- ibdev->pkeys.phys_pkey_cache[port-1][i] =
+ ibdev->pkeys->phys_pkey_cache[port-1][i] =
(i) ? 0 : 0xFFFF;
}
}
@@ -2595,6 +2595,7 @@ static void mlx4_ib_release(struct ib_device *device)
ib_dev);
kvfree(ibdev->iboe);
+ kvfree(ibdev->pkeys);
}
static void *mlx4_ib_add(struct mlx4_dev *dev)
@@ -2636,6 +2637,10 @@ static void *mlx4_ib_add(struct mlx4_dev *dev)
ibdev->iboe->parent = ibdev;
iboe = ibdev->iboe;
+ ibdev->pkeys = kvzalloc(sizeof(struct pkey_mgt), GFP_KERNEL);
+ if (!ibdev->pkeys)
+ goto err_dealloc;
+
if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
goto err_dealloc;
diff --git a/drivers/infiniband/hw/mlx4/mlx4_ib.h b/drivers/infiniband/hw/mlx4/mlx4_ib.h
index 2996c61..2b5a9b2 100644
--- a/drivers/infiniband/hw/mlx4/mlx4_ib.h
+++ b/drivers/infiniband/hw/mlx4/mlx4_ib.h
@@ -608,7 +608,7 @@ struct mlx4_ib_dev {
struct kobject *ports_parent;
struct kobject *dev_ports_parent[MLX4_MFUNC_MAX];
struct mlx4_ib_iov_port iov_ports[MLX4_MAX_PORTS];
- struct pkey_mgt pkeys;
+ struct pkey_mgt *pkeys;
unsigned long *ib_uc_qpns_bitmap;
int steer_qpn_count;
int steer_qpn_base;
diff --git a/drivers/infiniband/hw/mlx4/sysfs.c b/drivers/infiniband/hw/mlx4/sysfs.c
index e219093..a5b4592a 100644
--- a/drivers/infiniband/hw/mlx4/sysfs.c
+++ b/drivers/infiniband/hw/mlx4/sysfs.c
@@ -447,12 +447,12 @@ static ssize_t show_port_pkey(struct mlx4_port *p, struct port_attribute *attr,
container_of(attr, struct port_table_attribute, attr);
ssize_t ret = -ENODEV;
- if (p->dev->pkeys.virt2phys_pkey[p->slave][p->port_num - 1][tab_attr->index] >=
+ if (p->dev->pkeys->virt2phys_pkey[p->slave][p->port_num - 1][tab_attr->index] >=
(p->dev->dev->caps.pkey_table_len[p->port_num]))
ret = sprintf(buf, "none\n");
else
ret = sprintf(buf, "%d\n",
- p->dev->pkeys.virt2phys_pkey[p->slave]
+ p->dev->pkeys->virt2phys_pkey[p->slave]
[p->port_num - 1][tab_attr->index]);
return ret;
}
@@ -476,8 +476,8 @@ static ssize_t store_port_pkey(struct mlx4_port *p, struct port_attribute *attr,
idx < 0)
return -EINVAL;
- p->dev->pkeys.virt2phys_pkey[p->slave][p->port_num - 1]
- [tab_attr->index] = idx;
+ p->dev->pkeys->virt2phys_pkey[p->slave][p->port_num - 1]
+ [tab_attr->index] = idx;
mlx4_sync_pkey_table(p->dev->dev, p->slave, p->port_num,
tab_attr->index, idx);
err = mlx4_gen_pkey_eqe(p->dev->dev, p->slave, p->port_num);
@@ -687,7 +687,7 @@ static int add_port(struct mlx4_ib_dev *dev, int port_num, int slave)
if (ret)
goto err_free_gid;
- list_add_tail(&p->kobj.entry, &dev->pkeys.pkey_port_list[slave]);
+ list_add_tail(&p->kobj.entry, &dev->pkeys->pkey_port_list[slave]);
return 0;
err_free_gid:
@@ -716,19 +716,19 @@ static int register_one_pkey_tree(struct mlx4_ib_dev *dev, int slave)
get_name(dev, name, slave, sizeof name);
- dev->pkeys.device_parent[slave] =
+ dev->pkeys->device_parent[slave] =
kobject_create_and_add(name, kobject_get(dev->iov_parent));
- if (!dev->pkeys.device_parent[slave]) {
+ if (!dev->pkeys->device_parent[slave]) {
err = -ENOMEM;
goto fail_dev;
}
- INIT_LIST_HEAD(&dev->pkeys.pkey_port_list[slave]);
+ INIT_LIST_HEAD(&dev->pkeys->pkey_port_list[slave]);
dev->dev_ports_parent[slave] =
kobject_create_and_add("ports",
- kobject_get(dev->pkeys.device_parent[slave]));
+ kobject_get(dev->pkeys->device_parent[slave]));
if (!dev->dev_ports_parent[slave]) {
err = -ENOMEM;
@@ -748,7 +748,7 @@ static int register_one_pkey_tree(struct mlx4_ib_dev *dev, int slave)
err_add:
list_for_each_entry_safe(p, t,
- &dev->pkeys.pkey_port_list[slave],
+ &dev->pkeys->pkey_port_list[slave],
entry) {
list_del(&p->entry);
mport = container_of(p, struct mlx4_port, kobj);
@@ -760,9 +760,9 @@ static int register_one_pkey_tree(struct mlx4_ib_dev *dev, int slave)
kobject_put(dev->dev_ports_parent[slave]);
err_ports:
- kobject_put(dev->pkeys.device_parent[slave]);
+ kobject_put(dev->pkeys->device_parent[slave]);
/* extra put for the device_parent create_and_add */
- kobject_put(dev->pkeys.device_parent[slave]);
+ kobject_put(dev->pkeys->device_parent[slave]);
fail_dev:
kobject_put(dev->iov_parent);
@@ -793,7 +793,7 @@ static void unregister_pkey_tree(struct mlx4_ib_dev *device)
for (slave = device->dev->persist->num_vfs; slave >= 0; --slave) {
list_for_each_entry_safe(p, t,
- &device->pkeys.pkey_port_list[slave],
+ &device->pkeys->pkey_port_list[slave],
entry) {
list_del(&p->entry);
port = container_of(p, struct mlx4_port, kobj);
@@ -804,8 +804,8 @@ static void unregister_pkey_tree(struct mlx4_ib_dev *device)
kobject_put(device->dev_ports_parent[slave]);
}
kobject_put(device->dev_ports_parent[slave]);
- kobject_put(device->pkeys.device_parent[slave]);
- kobject_put(device->pkeys.device_parent[slave]);
+ kobject_put(device->pkeys->device_parent[slave]);
+ kobject_put(device->pkeys->device_parent[slave]);
kobject_put(device->iov_parent);
}
}
--
2.1.4
This is the 3rd patch of 3 of the work for decreasing size
of mlx4_ib_dev.
The field takes about 6K and could be safely allocated with kvzalloc.
Signed-off-by: Jan Dakinevich <[email protected]>
---
drivers/infiniband/hw/mlx4/alias_GUID.c | 192 ++++++++++++++++----------------
drivers/infiniband/hw/mlx4/cm.c | 32 +++---
drivers/infiniband/hw/mlx4/mad.c | 80 ++++++-------
drivers/infiniband/hw/mlx4/main.c | 17 ++-
drivers/infiniband/hw/mlx4/mcg.c | 4 +-
drivers/infiniband/hw/mlx4/mlx4_ib.h | 3 +-
drivers/infiniband/hw/mlx4/qp.c | 4 +-
drivers/infiniband/hw/mlx4/sysfs.c | 10 +-
8 files changed, 175 insertions(+), 167 deletions(-)
diff --git a/drivers/infiniband/hw/mlx4/alias_GUID.c b/drivers/infiniband/hw/mlx4/alias_GUID.c
index 155b4df..b5f794d 100644
--- a/drivers/infiniband/hw/mlx4/alias_GUID.c
+++ b/drivers/infiniband/hw/mlx4/alias_GUID.c
@@ -83,7 +83,7 @@ void mlx4_ib_update_cache_on_guid_change(struct mlx4_ib_dev *dev, int block_num,
if (!mlx4_is_master(dev->dev))
return;
- guid_indexes = be64_to_cpu((__force __be64) dev->sriov.alias_guid.
+ 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);
@@ -99,7 +99,7 @@ void mlx4_ib_update_cache_on_guid_change(struct mlx4_ib_dev *dev, int block_num,
}
/* cache the guid: */
- memcpy(&dev->sriov.demux[port_index].guid_cache[slave_id],
+ memcpy(&dev->sriov->demux[port_index].guid_cache[slave_id],
&p_data[i * GUID_REC_SIZE],
GUID_REC_SIZE);
} else
@@ -114,7 +114,7 @@ static __be64 get_cached_alias_guid(struct mlx4_ib_dev *dev, int port, int index
pr_err("%s: ERROR: asked for index:%d\n", __func__, index);
return (__force __be64) -1;
}
- return *(__be64 *)&dev->sriov.demux[port - 1].guid_cache[index];
+ return *(__be64 *)&dev->sriov->demux[port - 1].guid_cache[index];
}
@@ -133,12 +133,12 @@ void mlx4_ib_slave_alias_guid_event(struct mlx4_ib_dev *dev, int slave,
unsigned long flags;
int do_work = 0;
- spin_lock_irqsave(&dev->sriov.alias_guid.ag_work_lock, flags);
- if (dev->sriov.alias_guid.ports_guid[port_index].state_flags &
+ spin_lock_irqsave(&dev->sriov->alias_guid.ag_work_lock, flags);
+ if (dev->sriov->alias_guid.ports_guid[port_index].state_flags &
GUID_STATE_NEED_PORT_INIT)
goto unlock;
if (!slave_init) {
- curr_guid = *(__be64 *)&dev->sriov.
+ curr_guid = *(__be64 *)&dev->sriov->
alias_guid.ports_guid[port_index].
all_rec_per_port[record_num].
all_recs[GUID_REC_SIZE * index];
@@ -151,24 +151,24 @@ void mlx4_ib_slave_alias_guid_event(struct mlx4_ib_dev *dev, int slave,
if (required_guid == cpu_to_be64(MLX4_GUID_FOR_DELETE_VAL))
goto unlock;
}
- *(__be64 *)&dev->sriov.alias_guid.ports_guid[port_index].
+ *(__be64 *)&dev->sriov->alias_guid.ports_guid[port_index].
all_rec_per_port[record_num].
all_recs[GUID_REC_SIZE * index] = required_guid;
- dev->sriov.alias_guid.ports_guid[port_index].
+ dev->sriov->alias_guid.ports_guid[port_index].
all_rec_per_port[record_num].guid_indexes
|= mlx4_ib_get_aguid_comp_mask_from_ix(index);
- dev->sriov.alias_guid.ports_guid[port_index].
+ dev->sriov->alias_guid.ports_guid[port_index].
all_rec_per_port[record_num].status
= MLX4_GUID_INFO_STATUS_IDLE;
/* set to run immediately */
- dev->sriov.alias_guid.ports_guid[port_index].
+ dev->sriov->alias_guid.ports_guid[port_index].
all_rec_per_port[record_num].time_to_run = 0;
- dev->sriov.alias_guid.ports_guid[port_index].
+ dev->sriov->alias_guid.ports_guid[port_index].
all_rec_per_port[record_num].
guids_retry_schedule[index] = 0;
do_work = 1;
unlock:
- spin_unlock_irqrestore(&dev->sriov.alias_guid.ag_work_lock, flags);
+ spin_unlock_irqrestore(&dev->sriov->alias_guid.ag_work_lock, flags);
if (do_work)
mlx4_ib_init_alias_guid_work(dev, port_index);
@@ -201,9 +201,9 @@ void mlx4_ib_notify_slaves_on_guid_change(struct mlx4_ib_dev *dev,
if (!mlx4_is_master(dev->dev))
return;
- rec = &dev->sriov.alias_guid.ports_guid[port_num - 1].
+ rec = &dev->sriov->alias_guid.ports_guid[port_num - 1].
all_rec_per_port[block_num];
- guid_indexes = be64_to_cpu((__force __be64) dev->sriov.alias_guid.
+ 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);
@@ -233,7 +233,7 @@ void mlx4_ib_notify_slaves_on_guid_change(struct mlx4_ib_dev *dev,
if (tmp_cur_ag != form_cache_ag)
continue;
- spin_lock_irqsave(&dev->sriov.alias_guid.ag_work_lock, flags);
+ spin_lock_irqsave(&dev->sriov->alias_guid.ag_work_lock, flags);
required_value = *(__be64 *)&rec->all_recs[i * GUID_REC_SIZE];
if (required_value == cpu_to_be64(MLX4_GUID_FOR_DELETE_VAL))
@@ -245,12 +245,12 @@ void mlx4_ib_notify_slaves_on_guid_change(struct mlx4_ib_dev *dev,
} else {
/* may notify port down if value is 0 */
if (tmp_cur_ag != MLX4_NOT_SET_GUID) {
- spin_unlock_irqrestore(&dev->sriov.
+ spin_unlock_irqrestore(&dev->sriov->
alias_guid.ag_work_lock, flags);
continue;
}
}
- spin_unlock_irqrestore(&dev->sriov.alias_guid.ag_work_lock,
+ spin_unlock_irqrestore(&dev->sriov->alias_guid.ag_work_lock,
flags);
mlx4_gen_guid_change_eqe(dev->dev, slave_id, port_num);
/*2 cases: Valid GUID, and Invalid Guid*/
@@ -304,7 +304,7 @@ static void aliasguid_query_handler(int status,
dev = cb_ctx->dev;
port_index = cb_ctx->port - 1;
- rec = &dev->sriov.alias_guid.ports_guid[port_index].
+ rec = &dev->sriov->alias_guid.ports_guid[port_index].
all_rec_per_port[cb_ctx->block_num];
if (status) {
@@ -324,10 +324,10 @@ static void aliasguid_query_handler(int status,
be16_to_cpu(guid_rec->lid), cb_ctx->port,
guid_rec->block_num);
- rec = &dev->sriov.alias_guid.ports_guid[port_index].
+ rec = &dev->sriov->alias_guid.ports_guid[port_index].
all_rec_per_port[guid_rec->block_num];
- spin_lock_irqsave(&dev->sriov.alias_guid.ag_work_lock, flags);
+ spin_lock_irqsave(&dev->sriov->alias_guid.ag_work_lock, flags);
for (i = 0 ; i < NUM_ALIAS_GUID_IN_REC; i++) {
__be64 sm_response, required_val;
@@ -421,7 +421,7 @@ static void aliasguid_query_handler(int status,
} else {
rec->status = MLX4_GUID_INFO_STATUS_SET;
}
- spin_unlock_irqrestore(&dev->sriov.alias_guid.ag_work_lock, flags);
+ spin_unlock_irqrestore(&dev->sriov->alias_guid.ag_work_lock, flags);
/*
The func is call here to close the cases when the
sm doesn't send smp, so in the sa response the driver
@@ -431,12 +431,12 @@ static void aliasguid_query_handler(int status,
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) {
+ 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) {
get_low_record_time_index(dev, port_index, &resched_delay_sec);
- queue_delayed_work(dev->sriov.alias_guid.ports_guid[port_index].wq,
- &dev->sriov.alias_guid.ports_guid[port_index].
+ queue_delayed_work(dev->sriov->alias_guid.ports_guid[port_index].wq,
+ &dev->sriov->alias_guid.ports_guid[port_index].
alias_guid_work,
msecs_to_jiffies(resched_delay_sec * 1000));
}
@@ -445,8 +445,8 @@ static void aliasguid_query_handler(int status,
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);
+ 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)
@@ -455,13 +455,13 @@ static void invalidate_guid_record(struct mlx4_ib_dev *dev, u8 port, int index)
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
+ dev->sriov->alias_guid.ports_guid[port - 1].all_rec_per_port[index].status
= MLX4_GUID_INFO_STATUS_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].
+ *(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
@@ -474,11 +474,11 @@ static void invalidate_guid_record(struct mlx4_ib_dev *dev, u8 port, int index)
continue;
comp_mask |= mlx4_ib_get_aguid_comp_mask_from_ix(i);
}
- dev->sriov.alias_guid.ports_guid[port - 1].
+ dev->sriov->alias_guid.ports_guid[port - 1].
all_rec_per_port[index].guid_indexes |= comp_mask;
- if (dev->sriov.alias_guid.ports_guid[port - 1].
+ if (dev->sriov->alias_guid.ports_guid[port - 1].
all_rec_per_port[index].guid_indexes)
- dev->sriov.alias_guid.ports_guid[port - 1].
+ dev->sriov->alias_guid.ports_guid[port - 1].
all_rec_per_port[index].status = MLX4_GUID_INFO_STATUS_IDLE;
}
@@ -497,7 +497,7 @@ static int set_guid_rec(struct ib_device *ibdev,
int index = rec->block_num;
struct mlx4_sriov_alias_guid_info_rec_det *rec_det = &rec->rec_det;
struct list_head *head =
- &dev->sriov.alias_guid.ports_guid[port - 1].cb_list;
+ &dev->sriov->alias_guid.ports_guid[port - 1].cb_list;
memset(&attr, 0, sizeof(attr));
err = __mlx4_ib_query_port(ibdev, port, &attr, 1);
@@ -537,12 +537,12 @@ static int set_guid_rec(struct ib_device *ibdev,
rec_det->guid_indexes;
init_completion(&callback_context->done);
- spin_lock_irqsave(&dev->sriov.alias_guid.ag_work_lock, flags1);
+ 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);
+ 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,
+ ib_sa_guid_info_rec_query(dev->sriov->alias_guid.sa_client,
ibdev, port, &guid_info_rec,
comp_mask, rec->method, 1000,
GFP_KERNEL, aliasguid_query_handler,
@@ -552,10 +552,10 @@ static int set_guid_rec(struct ib_device *ibdev,
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);
+ 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);
+ spin_unlock_irqrestore(&dev->sriov->alias_guid.ag_work_lock, flags1);
resched_delay = 1 * HZ;
err = -EAGAIN;
goto new_schedule;
@@ -564,16 +564,16 @@ static int set_guid_rec(struct ib_device *ibdev,
goto out;
new_schedule:
- spin_lock_irqsave(&dev->sriov.going_down_lock, flags);
- spin_lock_irqsave(&dev->sriov.alias_guid.ag_work_lock, flags1);
+ 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,
+ 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);
+ spin_unlock_irqrestore(&dev->sriov->alias_guid.ag_work_lock, flags1);
+ spin_unlock_irqrestore(&dev->sriov->going_down_lock, flags);
out:
return err;
@@ -593,7 +593,7 @@ static void mlx4_ib_guid_port_init(struct mlx4_ib_dev *dev, int port)
!mlx4_is_slave_active(dev->dev, entry))
continue;
guid = mlx4_get_admin_guid(dev->dev, entry, port);
- *(__be64 *)&dev->sriov.alias_guid.ports_guid[port - 1].
+ *(__be64 *)&dev->sriov->alias_guid.ports_guid[port - 1].
all_rec_per_port[j].all_recs
[GUID_REC_SIZE * k] = guid;
pr_debug("guid was set, entry=%d, val=0x%llx, port=%d\n",
@@ -610,32 +610,32 @@ void mlx4_ib_invalidate_all_guid_record(struct mlx4_ib_dev *dev, int port)
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);
+ spin_lock_irqsave(&dev->sriov->going_down_lock, flags);
+ spin_lock_irqsave(&dev->sriov->alias_guid.ag_work_lock, flags1);
- if (dev->sriov.alias_guid.ports_guid[port - 1].state_flags &
+ if (dev->sriov->alias_guid.ports_guid[port - 1].state_flags &
GUID_STATE_NEED_PORT_INIT) {
mlx4_ib_guid_port_init(dev, port);
- dev->sriov.alias_guid.ports_guid[port - 1].state_flags &=
+ dev->sriov->alias_guid.ports_guid[port - 1].state_flags &=
(~GUID_STATE_NEED_PORT_INIT);
}
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) {
+ 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.
+ 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,
+ 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);
+ spin_unlock_irqrestore(&dev->sriov->alias_guid.ag_work_lock, flags1);
+ spin_unlock_irqrestore(&dev->sriov->going_down_lock, flags);
}
static void set_required_record(struct mlx4_ib_dev *dev, u8 port,
@@ -648,7 +648,7 @@ static void set_required_record(struct mlx4_ib_dev *dev, u8 port,
ib_sa_comp_mask delete_guid_indexes = 0;
ib_sa_comp_mask set_guid_indexes = 0;
struct mlx4_sriov_alias_guid_info_rec_det *rec =
- &dev->sriov.alias_guid.ports_guid[port].
+ &dev->sriov->alias_guid.ports_guid[port].
all_rec_per_port[record_index];
for (i = 0; i < NUM_ALIAS_GUID_IN_REC; i++) {
@@ -697,7 +697,7 @@ static int get_low_record_time_index(struct mlx4_ib_dev *dev, u8 port,
int j;
for (j = 0; j < NUM_ALIAS_GUID_REC_IN_PORT; j++) {
- rec = dev->sriov.alias_guid.ports_guid[port].
+ rec = dev->sriov->alias_guid.ports_guid[port].
all_rec_per_port[j];
if (rec.status == MLX4_GUID_INFO_STATUS_IDLE &&
rec.guid_indexes) {
@@ -727,7 +727,7 @@ static int get_next_record_to_update(struct mlx4_ib_dev *dev, u8 port,
int record_index;
int ret = 0;
- spin_lock_irqsave(&dev->sriov.alias_guid.ag_work_lock, flags);
+ spin_lock_irqsave(&dev->sriov->alias_guid.ag_work_lock, flags);
record_index = get_low_record_time_index(dev, port, NULL);
if (record_index < 0) {
@@ -737,7 +737,7 @@ static int get_next_record_to_update(struct mlx4_ib_dev *dev, u8 port,
set_required_record(dev, port, rec, record_index);
out:
- spin_unlock_irqrestore(&dev->sriov.alias_guid.ag_work_lock, flags);
+ spin_unlock_irqrestore(&dev->sriov->alias_guid.ag_work_lock, flags);
return ret;
}
@@ -753,7 +753,7 @@ static void alias_guid_work(struct work_struct *work)
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);
+ struct mlx4_ib_dev *dev = ib_sriov->parent;
rec = kzalloc(sizeof *rec, GFP_KERNEL);
if (!rec)
@@ -778,33 +778,33 @@ void mlx4_ib_init_alias_guid_work(struct mlx4_ib_dev *dev, int port)
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) {
+ 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) {
/* If there is pending one should cancel then run, otherwise
* won't run till previous one is ended as same work
* struct is used.
*/
- cancel_delayed_work(&dev->sriov.alias_guid.ports_guid[port].
+ cancel_delayed_work(&dev->sriov->alias_guid.ports_guid[port].
alias_guid_work);
- queue_delayed_work(dev->sriov.alias_guid.ports_guid[port].wq,
- &dev->sriov.alias_guid.ports_guid[port].alias_guid_work, 0);
+ 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);
+ 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_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);
+ 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)) {
@@ -823,11 +823,11 @@ void mlx4_ib_destroy_alias_guid_service(struct mlx4_ib_dev *dev)
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);
+ 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);
+ 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)
@@ -839,14 +839,14 @@ int mlx4_ib_init_alias_guid_service(struct mlx4_ib_dev *dev)
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)
+ 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);
+ ib_sa_register_client(dev->sriov->alias_guid.sa_client);
- spin_lock_init(&dev->sriov.alias_guid.ag_work_lock);
+ 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)) {
@@ -856,18 +856,18 @@ int mlx4_ib_init_alias_guid_service(struct mlx4_ib_dev *dev)
}
for (i = 0 ; i < dev->num_ports; i++) {
- memset(&dev->sriov.alias_guid.ports_guid[i], 0,
+ memset(&dev->sriov->alias_guid.ports_guid[i], 0,
sizeof (struct mlx4_sriov_alias_guid_port_rec_det));
- dev->sriov.alias_guid.ports_guid[i].state_flags |=
+ dev->sriov->alias_guid.ports_guid[i].state_flags |=
GUID_STATE_NEED_PORT_INIT;
for (j = 0; j < NUM_ALIAS_GUID_REC_IN_PORT; j++) {
/* mark each val as it was deleted */
- memset(dev->sriov.alias_guid.ports_guid[i].
+ memset(dev->sriov->alias_guid.ports_guid[i].
all_rec_per_port[j].all_recs, 0xFF,
- sizeof(dev->sriov.alias_guid.ports_guid[i].
+ sizeof(dev->sriov->alias_guid.ports_guid[i].
all_rec_per_port[j].all_recs));
}
- INIT_LIST_HEAD(&dev->sriov.alias_guid.ports_guid[i].cb_list);
+ INIT_LIST_HEAD(&dev->sriov->alias_guid.ports_guid[i].cb_list);
/*prepare the records, set them to be allocated by sm*/
if (mlx4_ib_sm_guid_assign)
for (j = 1; j < NUM_ALIAS_GUID_PER_PORT; j++)
@@ -875,31 +875,31 @@ int mlx4_ib_init_alias_guid_service(struct mlx4_ib_dev *dev)
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;
+ dev->sriov->alias_guid.ports_guid[i].parent = &dev->sriov->alias_guid;
+ dev->sriov->alias_guid.ports_guid[i].port = i;
snprintf(alias_wq_name, sizeof alias_wq_name, "alias_guid%d", i);
- dev->sriov.alias_guid.ports_guid[i].wq =
+ dev->sriov->alias_guid.ports_guid[i].wq =
alloc_ordered_workqueue(alias_wq_name, WQ_MEM_RECLAIM);
- if (!dev->sriov.alias_guid.ports_guid[i].wq) {
+ 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,
+ 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;
+ 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;
+ 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;
}
diff --git a/drivers/infiniband/hw/mlx4/cm.c b/drivers/infiniband/hw/mlx4/cm.c
index fedaf82..3712109 100644
--- a/drivers/infiniband/hw/mlx4/cm.c
+++ b/drivers/infiniband/hw/mlx4/cm.c
@@ -143,7 +143,7 @@ static union ib_gid gid_from_req_msg(struct ib_device *ibdev, struct ib_mad *mad
static struct id_map_entry *
id_map_find_by_sl_id(struct ib_device *ibdev, u32 slave_id, u32 sl_cm_id)
{
- struct rb_root *sl_id_map = &to_mdev(ibdev)->sriov.sl_id_map;
+ struct rb_root *sl_id_map = &to_mdev(ibdev)->sriov->sl_id_map;
struct rb_node *node = sl_id_map->rb_node;
while (node) {
@@ -170,7 +170,7 @@ static void id_map_ent_timeout(struct work_struct *work)
struct id_map_entry *ent = container_of(delay, struct id_map_entry, timeout);
struct id_map_entry *db_ent, *found_ent;
struct mlx4_ib_dev *dev = ent->dev;
- struct mlx4_ib_sriov *sriov = &dev->sriov;
+ struct mlx4_ib_sriov *sriov = dev->sriov;
struct rb_root *sl_id_map = &sriov->sl_id_map;
int pv_id = (int) ent->pv_cm_id;
@@ -191,7 +191,7 @@ static void id_map_ent_timeout(struct work_struct *work)
static void id_map_find_del(struct ib_device *ibdev, int pv_cm_id)
{
- struct mlx4_ib_sriov *sriov = &to_mdev(ibdev)->sriov;
+ struct mlx4_ib_sriov *sriov = to_mdev(ibdev)->sriov;
struct rb_root *sl_id_map = &sriov->sl_id_map;
struct id_map_entry *ent, *found_ent;
@@ -209,7 +209,7 @@ static void id_map_find_del(struct ib_device *ibdev, int pv_cm_id)
static void sl_id_map_add(struct ib_device *ibdev, struct id_map_entry *new)
{
- struct rb_root *sl_id_map = &to_mdev(ibdev)->sriov.sl_id_map;
+ struct rb_root *sl_id_map = &to_mdev(ibdev)->sriov->sl_id_map;
struct rb_node **link = &sl_id_map->rb_node, *parent = NULL;
struct id_map_entry *ent;
int slave_id = new->slave_id;
@@ -244,7 +244,7 @@ id_map_alloc(struct ib_device *ibdev, int slave_id, u32 sl_cm_id)
{
int ret;
struct id_map_entry *ent;
- struct mlx4_ib_sriov *sriov = &to_mdev(ibdev)->sriov;
+ struct mlx4_ib_sriov *sriov = to_mdev(ibdev)->sriov;
ent = kmalloc(sizeof (struct id_map_entry), GFP_KERNEL);
if (!ent)
@@ -257,7 +257,7 @@ id_map_alloc(struct ib_device *ibdev, int slave_id, u32 sl_cm_id)
INIT_DELAYED_WORK(&ent->timeout, id_map_ent_timeout);
idr_preload(GFP_KERNEL);
- spin_lock(&to_mdev(ibdev)->sriov.id_map_lock);
+ spin_lock(&to_mdev(ibdev)->sriov->id_map_lock);
ret = idr_alloc_cyclic(&sriov->pv_id_table, ent, 0, 0, GFP_NOWAIT);
if (ret >= 0) {
@@ -282,7 +282,7 @@ static struct id_map_entry *
id_map_get(struct ib_device *ibdev, int *pv_cm_id, int slave_id, int sl_cm_id)
{
struct id_map_entry *ent;
- struct mlx4_ib_sriov *sriov = &to_mdev(ibdev)->sriov;
+ struct mlx4_ib_sriov *sriov = to_mdev(ibdev)->sriov;
spin_lock(&sriov->id_map_lock);
if (*pv_cm_id == -1) {
@@ -298,7 +298,7 @@ id_map_get(struct ib_device *ibdev, int *pv_cm_id, int slave_id, int sl_cm_id)
static void schedule_delayed(struct ib_device *ibdev, struct id_map_entry *id)
{
- struct mlx4_ib_sriov *sriov = &to_mdev(ibdev)->sriov;
+ struct mlx4_ib_sriov *sriov = to_mdev(ibdev)->sriov;
unsigned long flags;
spin_lock(&sriov->id_map_lock);
@@ -404,17 +404,17 @@ int mlx4_ib_demux_cm_handler(struct ib_device *ibdev, int port, int *slave,
void mlx4_ib_cm_paravirt_init(struct mlx4_ib_dev *dev)
{
- spin_lock_init(&dev->sriov.id_map_lock);
- INIT_LIST_HEAD(&dev->sriov.cm_list);
- dev->sriov.sl_id_map = RB_ROOT;
- idr_init(&dev->sriov.pv_id_table);
+ spin_lock_init(&dev->sriov->id_map_lock);
+ INIT_LIST_HEAD(&dev->sriov->cm_list);
+ dev->sriov->sl_id_map = RB_ROOT;
+ idr_init(&dev->sriov->pv_id_table);
}
/* slave = -1 ==> all slaves */
/* TBD -- call paravirt clean for single slave. Need for slave RESET event */
void mlx4_ib_cm_paravirt_clean(struct mlx4_ib_dev *dev, int slave)
{
- struct mlx4_ib_sriov *sriov = &dev->sriov;
+ struct mlx4_ib_sriov *sriov = dev->sriov;
struct rb_root *sl_id_map = &sriov->sl_id_map;
struct list_head lh;
struct rb_node *nd;
@@ -423,7 +423,7 @@ void mlx4_ib_cm_paravirt_clean(struct mlx4_ib_dev *dev, int slave)
/* cancel all delayed work queue entries */
INIT_LIST_HEAD(&lh);
spin_lock(&sriov->id_map_lock);
- list_for_each_entry_safe(map, tmp_map, &dev->sriov.cm_list, list) {
+ list_for_each_entry_safe(map, tmp_map, &dev->sriov->cm_list, list) {
if (slave < 0 || slave == map->slave_id) {
if (map->scheduled_delete)
need_flush |= !cancel_delayed_work(&map->timeout);
@@ -446,7 +446,7 @@ void mlx4_ib_cm_paravirt_clean(struct mlx4_ib_dev *dev, int slave)
rb_erase(&ent->node, sl_id_map);
idr_remove(&sriov->pv_id_table, (int) ent->pv_cm_id);
}
- list_splice_init(&dev->sriov.cm_list, &lh);
+ list_splice_init(&dev->sriov->cm_list, &lh);
} else {
/* first, move nodes belonging to slave to db remove list */
nd = rb_first(sl_id_map);
@@ -464,7 +464,7 @@ void mlx4_ib_cm_paravirt_clean(struct mlx4_ib_dev *dev, int slave)
}
/* add remaining nodes from cm_list */
- list_for_each_entry_safe(map, tmp_map, &dev->sriov.cm_list, list) {
+ list_for_each_entry_safe(map, tmp_map, &dev->sriov->cm_list, list) {
if (slave == map->slave_id)
list_move_tail(&map->list, &lh);
}
diff --git a/drivers/infiniband/hw/mlx4/mad.c b/drivers/infiniband/hw/mlx4/mad.c
index 3eceb46..88fe6cd 100644
--- a/drivers/infiniband/hw/mlx4/mad.c
+++ b/drivers/infiniband/hw/mlx4/mad.c
@@ -281,7 +281,7 @@ static void smp_snoop(struct ib_device *ibdev, u8 port_num, const struct ib_mad
if (pkey_change_bitmap) {
mlx4_ib_dispatch_event(dev, port_num,
IB_EVENT_PKEY_CHANGE);
- if (!dev->sriov.is_going_down)
+ if (!dev->sriov->is_going_down)
__propagate_pkey_ev(dev, port_num, bn,
pkey_change_bitmap);
}
@@ -296,7 +296,7 @@ static void smp_snoop(struct ib_device *ibdev, u8 port_num, const struct ib_mad
IB_EVENT_GID_CHANGE);
/*if master, notify relevant slaves*/
if (mlx4_is_master(dev->dev) &&
- !dev->sriov.is_going_down) {
+ !dev->sriov->is_going_down) {
bn = be32_to_cpu(((struct ib_smp *)mad)->attr_mod);
mlx4_ib_update_cache_on_guid_change(dev, bn, port_num,
(u8 *)(&((struct ib_smp *)mad)->data));
@@ -435,7 +435,7 @@ int mlx4_ib_find_real_gid(struct ib_device *ibdev, u8 port, __be64 guid)
int i;
for (i = 0; i < dev->dev->caps.sqp_demux; i++) {
- if (dev->sriov.demux[port - 1].guid_cache[i] == guid)
+ if (dev->sriov->demux[port - 1].guid_cache[i] == guid)
return i;
}
return -1;
@@ -523,7 +523,7 @@ int mlx4_ib_send_to_slave(struct mlx4_ib_dev *dev, int slave, u8 port,
if (dest_qpt > IB_QPT_GSI)
return -EINVAL;
- tun_ctx = dev->sriov.demux[port-1].tun[slave];
+ tun_ctx = dev->sriov->demux[port-1].tun[slave];
/* check if proxy qp created */
if (!tun_ctx || tun_ctx->state != DEMUX_PV_STATE_ACTIVE)
@@ -736,7 +736,7 @@ static int mlx4_ib_demux_mad(struct ib_device *ibdev, u8 port,
if (grh->dgid.global.interface_id ==
cpu_to_be64(IB_SA_WELL_KNOWN_GUID) &&
grh->dgid.global.subnet_prefix == cpu_to_be64(
- atomic64_read(&dev->sriov.demux[port - 1].subnet_prefix))) {
+ atomic64_read(&dev->sriov->demux[port - 1].subnet_prefix))) {
slave = 0;
} else {
slave = mlx4_ib_find_real_gid(ibdev, port,
@@ -1085,7 +1085,7 @@ static void handle_lid_change_event(struct mlx4_ib_dev *dev, u8 port_num)
{
mlx4_ib_dispatch_event(dev, port_num, IB_EVENT_LID_CHANGE);
- if (mlx4_is_master(dev->dev) && !dev->sriov.is_going_down)
+ if (mlx4_is_master(dev->dev) && !dev->sriov->is_going_down)
mlx4_gen_slaves_port_mgt_ev(dev->dev, port_num,
MLX4_EQ_PORT_INFO_LID_CHANGE_MASK);
}
@@ -1096,8 +1096,8 @@ static void handle_client_rereg_event(struct mlx4_ib_dev *dev, u8 port_num)
if (mlx4_is_master(dev->dev)) {
mlx4_ib_invalidate_all_guid_record(dev, port_num);
- if (!dev->sriov.is_going_down) {
- mlx4_ib_mcg_port_cleanup(&dev->sriov.demux[port_num - 1], 0);
+ if (!dev->sriov->is_going_down) {
+ mlx4_ib_mcg_port_cleanup(&dev->sriov->demux[port_num - 1], 0);
mlx4_gen_slaves_port_mgt_ev(dev->dev, port_num,
MLX4_EQ_PORT_INFO_CLIENT_REREG_MASK);
}
@@ -1223,9 +1223,9 @@ void handle_port_mgmt_change_event(struct work_struct *work)
} else {
pr_debug("Changing QP1 subnet prefix for port %d. old=0x%llx. new=0x%llx\n",
port,
- (u64)atomic64_read(&dev->sriov.demux[port - 1].subnet_prefix),
+ (u64)atomic64_read(&dev->sriov->demux[port - 1].subnet_prefix),
be64_to_cpu(gid.global.subnet_prefix));
- atomic64_set(&dev->sriov.demux[port - 1].subnet_prefix,
+ atomic64_set(&dev->sriov->demux[port - 1].subnet_prefix,
be64_to_cpu(gid.global.subnet_prefix));
}
}
@@ -1242,7 +1242,7 @@ void handle_port_mgmt_change_event(struct work_struct *work)
case MLX4_DEV_PMC_SUBTYPE_PKEY_TABLE:
mlx4_ib_dispatch_event(dev, port, IB_EVENT_PKEY_CHANGE);
- if (mlx4_is_master(dev->dev) && !dev->sriov.is_going_down)
+ if (mlx4_is_master(dev->dev) && !dev->sriov->is_going_down)
propagate_pkey_ev(dev, port, eqe);
break;
case MLX4_DEV_PMC_SUBTYPE_GUID_INFO:
@@ -1250,7 +1250,7 @@ void handle_port_mgmt_change_event(struct work_struct *work)
if (!mlx4_is_master(dev->dev))
mlx4_ib_dispatch_event(dev, port, IB_EVENT_GID_CHANGE);
/*if master, notify relevant slaves*/
- else if (!dev->sriov.is_going_down) {
+ else if (!dev->sriov->is_going_down) {
tbl_block = GET_BLK_PTR_FROM_EQE(eqe);
change_bitmap = GET_MASK_FROM_EQE(eqe);
handle_slaves_guid_change(dev, port, tbl_block, change_bitmap);
@@ -1299,10 +1299,10 @@ static void mlx4_ib_tunnel_comp_handler(struct ib_cq *cq, void *arg)
unsigned long flags;
struct mlx4_ib_demux_pv_ctx *ctx = cq->cq_context;
struct mlx4_ib_dev *dev = to_mdev(ctx->ib_dev);
- spin_lock_irqsave(&dev->sriov.going_down_lock, flags);
- if (!dev->sriov.is_going_down && ctx->state == DEMUX_PV_STATE_ACTIVE)
+ spin_lock_irqsave(&dev->sriov->going_down_lock, flags);
+ if (!dev->sriov->is_going_down && ctx->state == DEMUX_PV_STATE_ACTIVE)
queue_work(ctx->wq, &ctx->work);
- spin_unlock_irqrestore(&dev->sriov.going_down_lock, flags);
+ spin_unlock_irqrestore(&dev->sriov->going_down_lock, flags);
}
static int mlx4_ib_post_pv_qp_buf(struct mlx4_ib_demux_pv_ctx *ctx,
@@ -1373,7 +1373,7 @@ int mlx4_ib_send_to_wire(struct mlx4_ib_dev *dev, int slave, u8 port,
u16 wire_pkey_ix;
int src_qpnum;
- sqp_ctx = dev->sriov.sqps[port-1];
+ sqp_ctx = dev->sriov->sqps[port-1];
/* check if proxy qp created */
if (!sqp_ctx || sqp_ctx->state != DEMUX_PV_STATE_ACTIVE)
@@ -1960,9 +1960,9 @@ static int alloc_pv_object(struct mlx4_ib_dev *dev, int slave, int port,
static void free_pv_object(struct mlx4_ib_dev *dev, int slave, int port)
{
- if (dev->sriov.demux[port - 1].tun[slave]) {
- kfree(dev->sriov.demux[port - 1].tun[slave]);
- dev->sriov.demux[port - 1].tun[slave] = NULL;
+ if (dev->sriov->demux[port - 1].tun[slave]) {
+ kfree(dev->sriov->demux[port - 1].tun[slave]);
+ dev->sriov->demux[port - 1].tun[slave] = NULL;
}
}
@@ -2036,7 +2036,7 @@ static int create_pv_resources(struct ib_device *ibdev, int slave, int port,
else
INIT_WORK(&ctx->work, mlx4_ib_sqp_comp_worker);
- ctx->wq = to_mdev(ibdev)->sriov.demux[port - 1].wq;
+ ctx->wq = to_mdev(ibdev)->sriov->demux[port - 1].wq;
ret = ib_req_notify_cq(ctx->cq, IB_CQ_NEXT_COMP);
if (ret) {
@@ -2107,25 +2107,25 @@ static int mlx4_ib_tunnels_update(struct mlx4_ib_dev *dev, int slave,
int ret = 0;
if (!do_init) {
- clean_vf_mcast(&dev->sriov.demux[port - 1], slave);
+ clean_vf_mcast(&dev->sriov->demux[port - 1], slave);
/* for master, destroy real sqp resources */
if (slave == mlx4_master_func_num(dev->dev))
destroy_pv_resources(dev, slave, port,
- dev->sriov.sqps[port - 1], 1);
+ dev->sriov->sqps[port - 1], 1);
/* destroy the tunnel qp resources */
destroy_pv_resources(dev, slave, port,
- dev->sriov.demux[port - 1].tun[slave], 1);
+ dev->sriov->demux[port - 1].tun[slave], 1);
return 0;
}
/* create the tunnel qp resources */
ret = create_pv_resources(&dev->ib_dev, slave, port, 1,
- dev->sriov.demux[port - 1].tun[slave]);
+ dev->sriov->demux[port - 1].tun[slave]);
/* for master, create the real sqp resources */
if (!ret && slave == mlx4_master_func_num(dev->dev))
ret = create_pv_resources(&dev->ib_dev, slave, port, 0,
- dev->sriov.sqps[port - 1]);
+ dev->sriov->sqps[port - 1]);
return ret;
}
@@ -2276,8 +2276,8 @@ int mlx4_ib_init_sriov(struct mlx4_ib_dev *dev)
if (!mlx4_is_mfunc(dev->dev))
return 0;
- dev->sriov.is_going_down = 0;
- spin_lock_init(&dev->sriov.going_down_lock);
+ dev->sriov->is_going_down = 0;
+ spin_lock_init(&dev->sriov->going_down_lock);
mlx4_ib_cm_paravirt_init(dev);
mlx4_ib_warn(&dev->ib_dev, "multi-function enabled\n");
@@ -2312,14 +2312,14 @@ int mlx4_ib_init_sriov(struct mlx4_ib_dev *dev)
err = __mlx4_ib_query_gid(&dev->ib_dev, i + 1, 0, &gid, 1);
if (err)
goto demux_err;
- dev->sriov.demux[i].guid_cache[0] = gid.global.interface_id;
- atomic64_set(&dev->sriov.demux[i].subnet_prefix,
+ dev->sriov->demux[i].guid_cache[0] = gid.global.interface_id;
+ atomic64_set(&dev->sriov->demux[i].subnet_prefix,
be64_to_cpu(gid.global.subnet_prefix));
err = alloc_pv_object(dev, mlx4_master_func_num(dev->dev), i + 1,
- &dev->sriov.sqps[i]);
+ &dev->sriov->sqps[i]);
if (err)
goto demux_err;
- err = mlx4_ib_alloc_demux_ctx(dev, &dev->sriov.demux[i], i + 1);
+ err = mlx4_ib_alloc_demux_ctx(dev, &dev->sriov->demux[i], i + 1);
if (err)
goto free_pv;
}
@@ -2331,7 +2331,7 @@ int mlx4_ib_init_sriov(struct mlx4_ib_dev *dev)
demux_err:
while (--i >= 0) {
free_pv_object(dev, mlx4_master_func_num(dev->dev), i + 1);
- mlx4_ib_free_demux_ctx(&dev->sriov.demux[i]);
+ mlx4_ib_free_demux_ctx(&dev->sriov->demux[i]);
}
mlx4_ib_device_unregister_sysfs(dev);
@@ -2352,16 +2352,16 @@ void mlx4_ib_close_sriov(struct mlx4_ib_dev *dev)
if (!mlx4_is_mfunc(dev->dev))
return;
- spin_lock_irqsave(&dev->sriov.going_down_lock, flags);
- dev->sriov.is_going_down = 1;
- spin_unlock_irqrestore(&dev->sriov.going_down_lock, flags);
+ spin_lock_irqsave(&dev->sriov->going_down_lock, flags);
+ dev->sriov->is_going_down = 1;
+ spin_unlock_irqrestore(&dev->sriov->going_down_lock, flags);
if (mlx4_is_master(dev->dev)) {
for (i = 0; i < dev->num_ports; i++) {
- flush_workqueue(dev->sriov.demux[i].ud_wq);
- mlx4_ib_free_sqp_ctx(dev->sriov.sqps[i]);
- kfree(dev->sriov.sqps[i]);
- dev->sriov.sqps[i] = NULL;
- mlx4_ib_free_demux_ctx(&dev->sriov.demux[i]);
+ flush_workqueue(dev->sriov->demux[i].ud_wq);
+ mlx4_ib_free_sqp_ctx(dev->sriov->sqps[i]);
+ kfree(dev->sriov->sqps[i]);
+ dev->sriov->sqps[i] = NULL;
+ mlx4_ib_free_demux_ctx(&dev->sriov->demux[i]);
}
mlx4_ib_cm_paravirt_clean(dev, -1);
diff --git a/drivers/infiniband/hw/mlx4/main.c b/drivers/infiniband/hw/mlx4/main.c
index 8ba0103..6d1483d 100644
--- a/drivers/infiniband/hw/mlx4/main.c
+++ b/drivers/infiniband/hw/mlx4/main.c
@@ -2596,6 +2596,7 @@ static void mlx4_ib_release(struct ib_device *device)
kvfree(ibdev->iboe);
kvfree(ibdev->pkeys);
+ kvfree(ibdev->sriov);
}
static void *mlx4_ib_add(struct mlx4_dev *dev)
@@ -2641,6 +2642,12 @@ static void *mlx4_ib_add(struct mlx4_dev *dev)
if (!ibdev->pkeys)
goto err_dealloc;
+ ibdev->sriov = kvzalloc(sizeof(struct mlx4_ib_sriov), GFP_KERNEL);
+ if (!ibdev->sriov)
+ goto err_dealloc;
+
+ ibdev->sriov->parent = ibdev;
+
if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
goto err_dealloc;
@@ -3152,13 +3159,13 @@ static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
dm[i]->dev = ibdev;
}
/* initialize or tear down tunnel QPs for the slave */
- spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
- if (!ibdev->sriov.is_going_down) {
+ spin_lock_irqsave(&ibdev->sriov->going_down_lock, flags);
+ if (!ibdev->sriov->is_going_down) {
for (i = 0; i < ports; i++)
- queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
- spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
+ queue_work(ibdev->sriov->demux[i].ud_wq, &dm[i]->work);
+ spin_unlock_irqrestore(&ibdev->sriov->going_down_lock, flags);
} else {
- spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
+ spin_unlock_irqrestore(&ibdev->sriov->going_down_lock, flags);
for (i = 0; i < ports; i++)
kfree(dm[i]);
}
diff --git a/drivers/infiniband/hw/mlx4/mcg.c b/drivers/infiniband/hw/mlx4/mcg.c
index 81ffc00..6415326 100644
--- a/drivers/infiniband/hw/mlx4/mcg.c
+++ b/drivers/infiniband/hw/mlx4/mcg.c
@@ -884,7 +884,7 @@ int mlx4_ib_mcg_demux_handler(struct ib_device *ibdev, int port, int slave,
{
struct mlx4_ib_dev *dev = to_mdev(ibdev);
struct ib_sa_mcmember_data *rec = (struct ib_sa_mcmember_data *)mad->data;
- struct mlx4_ib_demux_ctx *ctx = &dev->sriov.demux[port - 1];
+ struct mlx4_ib_demux_ctx *ctx = &dev->sriov->demux[port - 1];
struct mcast_group *group;
switch (mad->mad_hdr.method) {
@@ -933,7 +933,7 @@ int mlx4_ib_mcg_multiplex_handler(struct ib_device *ibdev, int port,
{
struct mlx4_ib_dev *dev = to_mdev(ibdev);
struct ib_sa_mcmember_data *rec = (struct ib_sa_mcmember_data *)sa_mad->data;
- struct mlx4_ib_demux_ctx *ctx = &dev->sriov.demux[port - 1];
+ struct mlx4_ib_demux_ctx *ctx = &dev->sriov->demux[port - 1];
struct mcast_group *group;
struct mcast_req *req;
int may_create = 0;
diff --git a/drivers/infiniband/hw/mlx4/mlx4_ib.h b/drivers/infiniband/hw/mlx4/mlx4_ib.h
index 2b5a9b2..dfe3a5c 100644
--- a/drivers/infiniband/hw/mlx4/mlx4_ib.h
+++ b/drivers/infiniband/hw/mlx4/mlx4_ib.h
@@ -501,6 +501,7 @@ struct mlx4_ib_sriov {
spinlock_t id_map_lock;
struct rb_root sl_id_map;
struct idr pv_id_table;
+ struct mlx4_ib_dev *parent;
};
struct gid_cache_context {
@@ -597,7 +598,7 @@ struct mlx4_ib_dev {
struct ib_ah *sm_ah[MLX4_MAX_PORTS];
spinlock_t sm_lock;
atomic64_t sl2vl[MLX4_MAX_PORTS];
- struct mlx4_ib_sriov sriov;
+ struct mlx4_ib_sriov *sriov;
struct mutex cap_mask_mutex;
bool ib_active;
diff --git a/drivers/infiniband/hw/mlx4/qp.c b/drivers/infiniband/hw/mlx4/qp.c
index 853ef6f..f001e2f 100644
--- a/drivers/infiniband/hw/mlx4/qp.c
+++ b/drivers/infiniband/hw/mlx4/qp.c
@@ -3031,11 +3031,11 @@ static int build_mlx_header(struct mlx4_ib_sqp *sqp, const struct ib_ud_wr *wr,
* we must use our own cache
*/
sqp->ud_header.grh.source_gid.global.subnet_prefix =
- cpu_to_be64(atomic64_read(&(to_mdev(ib_dev)->sriov.
+ cpu_to_be64(atomic64_read(&(to_mdev(ib_dev)->sriov->
demux[sqp->qp.port - 1].
subnet_prefix)));
sqp->ud_header.grh.source_gid.global.interface_id =
- to_mdev(ib_dev)->sriov.demux[sqp->qp.port - 1].
+ to_mdev(ib_dev)->sriov->demux[sqp->qp.port - 1].
guid_cache[ah->av.ib.gid_index];
} else {
sqp->ud_header.grh.source_gid =
diff --git a/drivers/infiniband/hw/mlx4/sysfs.c b/drivers/infiniband/hw/mlx4/sysfs.c
index a5b4592a..15f7c38 100644
--- a/drivers/infiniband/hw/mlx4/sysfs.c
+++ b/drivers/infiniband/hw/mlx4/sysfs.c
@@ -84,25 +84,25 @@ static ssize_t store_admin_alias_guid(struct device *dev,
pr_err("GUID 0 block 0 is RO\n");
return count;
}
- spin_lock_irqsave(&mdev->sriov.alias_guid.ag_work_lock, flags);
+ spin_lock_irqsave(&mdev->sriov->alias_guid.ag_work_lock, flags);
sscanf(buf, "%llx", &sysadmin_ag_val);
- *(__be64 *)&mdev->sriov.alias_guid.ports_guid[port->num - 1].
+ *(__be64 *)&mdev->sriov->alias_guid.ports_guid[port->num - 1].
all_rec_per_port[record_num].
all_recs[GUID_REC_SIZE * guid_index_in_rec] =
cpu_to_be64(sysadmin_ag_val);
/* Change the state to be pending for update */
- mdev->sriov.alias_guid.ports_guid[port->num - 1].all_rec_per_port[record_num].status
+ mdev->sriov->alias_guid.ports_guid[port->num - 1].all_rec_per_port[record_num].status
= MLX4_GUID_INFO_STATUS_IDLE ;
mlx4_set_admin_guid(mdev->dev, cpu_to_be64(sysadmin_ag_val),
mlx4_ib_iov_dentry->entry_num,
port->num);
/* set the record index */
- mdev->sriov.alias_guid.ports_guid[port->num - 1].all_rec_per_port[record_num].guid_indexes
+ mdev->sriov->alias_guid.ports_guid[port->num - 1].all_rec_per_port[record_num].guid_indexes
|= mlx4_ib_get_aguid_comp_mask_from_ix(guid_index_in_rec);
- spin_unlock_irqrestore(&mdev->sriov.alias_guid.ag_work_lock, flags);
+ spin_unlock_irqrestore(&mdev->sriov->alias_guid.ag_work_lock, flags);
mlx4_ib_init_alias_guid_work(mdev, port->num - 1);
return count;
--
2.1.4
This is the 1st patch of 3 of the work for decreasing size
of mlx4_ib_dev.
The field takes about 8K and could be safely allocated with kvzalloc.
Signed-off-by: Jan Dakinevich <[email protected]>
---
drivers/infiniband/hw/mlx4/main.c | 65 ++++++++++++++++++++++--------------
drivers/infiniband/hw/mlx4/mlx4_ib.h | 3 +-
drivers/infiniband/hw/mlx4/qp.c | 4 +--
3 files changed, 44 insertions(+), 28 deletions(-)
diff --git a/drivers/infiniband/hw/mlx4/main.c b/drivers/infiniband/hw/mlx4/main.c
index 0bbeaaa..1e3bb67 100644
--- a/drivers/infiniband/hw/mlx4/main.c
+++ b/drivers/infiniband/hw/mlx4/main.c
@@ -249,7 +249,7 @@ static int mlx4_ib_update_gids(struct gid_entry *gids,
static int mlx4_ib_add_gid(const struct ib_gid_attr *attr, void **context)
{
struct mlx4_ib_dev *ibdev = to_mdev(attr->device);
- struct mlx4_ib_iboe *iboe = &ibdev->iboe;
+ struct mlx4_ib_iboe *iboe = ibdev->iboe;
struct mlx4_port_gid_table *port_gid_table;
int free = -1, found = -1;
int ret = 0;
@@ -327,7 +327,7 @@ static int mlx4_ib_del_gid(const struct ib_gid_attr *attr, void **context)
{
struct gid_cache_context *ctx = *context;
struct mlx4_ib_dev *ibdev = to_mdev(attr->device);
- struct mlx4_ib_iboe *iboe = &ibdev->iboe;
+ struct mlx4_ib_iboe *iboe = ibdev->iboe;
struct mlx4_port_gid_table *port_gid_table;
int ret = 0;
int hw_update = 0;
@@ -382,7 +382,7 @@ static int mlx4_ib_del_gid(const struct ib_gid_attr *attr, void **context)
int mlx4_ib_gid_index_to_real_index(struct mlx4_ib_dev *ibdev,
const struct ib_gid_attr *attr)
{
- struct mlx4_ib_iboe *iboe = &ibdev->iboe;
+ struct mlx4_ib_iboe *iboe = ibdev->iboe;
struct gid_cache_context *ctx = NULL;
struct mlx4_port_gid_table *port_gid_table;
int real_index = -EINVAL;
@@ -742,7 +742,7 @@ static int eth_link_query_port(struct ib_device *ibdev, u8 port,
{
struct mlx4_ib_dev *mdev = to_mdev(ibdev);
- struct mlx4_ib_iboe *iboe = &mdev->iboe;
+ struct mlx4_ib_iboe *iboe = mdev->iboe;
struct net_device *ndev;
enum ib_mtu tmp;
struct mlx4_cmd_mailbox *mailbox;
@@ -1415,11 +1415,11 @@ int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
if (!mqp->port)
return 0;
- spin_lock_bh(&mdev->iboe.lock);
- ndev = mdev->iboe.netdevs[mqp->port - 1];
+ spin_lock_bh(&mdev->iboe->lock);
+ ndev = mdev->iboe->netdevs[mqp->port - 1];
if (ndev)
dev_hold(ndev);
- spin_unlock_bh(&mdev->iboe.lock);
+ spin_unlock_bh(&mdev->iboe->lock);
if (ndev) {
ret = 1;
@@ -2078,11 +2078,11 @@ static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
mutex_lock(&mqp->mutex);
ge = find_gid_entry(mqp, gid->raw);
if (ge) {
- spin_lock_bh(&mdev->iboe.lock);
- ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
+ spin_lock_bh(&mdev->iboe->lock);
+ ndev = ge->added ? mdev->iboe->netdevs[ge->port - 1] : NULL;
if (ndev)
dev_hold(ndev);
- spin_unlock_bh(&mdev->iboe.lock);
+ spin_unlock_bh(&mdev->iboe->lock);
if (ndev)
dev_put(ndev);
list_del(&ge->list);
@@ -2373,7 +2373,7 @@ static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
new_smac = mlx4_mac_to_u64(dev->dev_addr);
read_unlock(&dev_base_lock);
- atomic64_set(&ibdev->iboe.mac[port - 1], new_smac);
+ atomic64_set(&ibdev->iboe->mac[port - 1], new_smac);
/* no need for update QP1 and mac registration in non-SRIOV */
if (!mlx4_is_mfunc(ibdev->dev))
@@ -2429,7 +2429,7 @@ static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
ASSERT_RTNL();
- iboe = &ibdev->iboe;
+ iboe = ibdev->iboe;
spin_lock_bh(&iboe->lock);
mlx4_foreach_ib_transport_port(port, ibdev->dev) {
@@ -2453,13 +2453,13 @@ static int mlx4_ib_netdev_event(struct notifier_block *this,
unsigned long event, void *ptr)
{
struct net_device *dev = netdev_notifier_info_to_dev(ptr);
- struct mlx4_ib_dev *ibdev;
+ struct mlx4_ib_iboe *iboe;
if (!net_eq(dev_net(dev), &init_net))
return NOTIFY_DONE;
- ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
- mlx4_ib_scan_netdevs(ibdev, dev, event);
+ iboe = container_of(this, struct mlx4_ib_iboe, nb);
+ mlx4_ib_scan_netdevs(iboe->parent, dev, event);
return NOTIFY_DONE;
}
@@ -2589,6 +2589,14 @@ static void get_fw_ver_str(struct ib_device *device, char *str)
(int) dev->dev->caps.fw_ver & 0xffff);
}
+static void mlx4_ib_release(struct ib_device *device)
+{
+ struct mlx4_ib_dev *ibdev = container_of(device, struct mlx4_ib_dev,
+ ib_dev);
+
+ kvfree(ibdev->iboe);
+}
+
static void *mlx4_ib_add(struct mlx4_dev *dev)
{
struct mlx4_ib_dev *ibdev;
@@ -2619,7 +2627,14 @@ static void *mlx4_ib_add(struct mlx4_dev *dev)
return NULL;
}
- iboe = &ibdev->iboe;
+ ibdev->ib_dev.release = mlx4_ib_release;
+
+ ibdev->iboe = kvzalloc(sizeof(struct mlx4_ib_iboe), GFP_KERNEL);
+ if (!ibdev->iboe)
+ goto err_dealloc;
+
+ ibdev->iboe->parent = ibdev;
+ iboe = ibdev->iboe;
if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
goto err_dealloc;
@@ -2948,10 +2963,10 @@ static void *mlx4_ib_add(struct mlx4_dev *dev)
return ibdev;
err_notif:
- if (ibdev->iboe.nb.notifier_call) {
- if (unregister_netdevice_notifier(&ibdev->iboe.nb))
+ if (ibdev->iboe->nb.notifier_call) {
+ if (unregister_netdevice_notifier(&ibdev->iboe->nb))
pr_warn("failure unregistering notifier\n");
- ibdev->iboe.nb.notifier_call = NULL;
+ ibdev->iboe->nb.notifier_call = NULL;
}
flush_workqueue(wq);
@@ -3073,10 +3088,10 @@ static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
mlx4_ib_mad_cleanup(ibdev);
ib_unregister_device(&ibdev->ib_dev);
mlx4_ib_diag_cleanup(ibdev);
- if (ibdev->iboe.nb.notifier_call) {
- if (unregister_netdevice_notifier(&ibdev->iboe.nb))
+ if (ibdev->iboe->nb.notifier_call) {
+ if (unregister_netdevice_notifier(&ibdev->iboe->nb))
pr_warn("failure unregistering notifier\n");
- ibdev->iboe.nb.notifier_call = NULL;
+ ibdev->iboe->nb.notifier_call = NULL;
}
mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
@@ -3218,9 +3233,9 @@ static void handle_bonded_port_state_event(struct work_struct *work)
struct ib_event ibev;
kfree(ew);
- spin_lock_bh(&ibdev->iboe.lock);
+ spin_lock_bh(&ibdev->iboe->lock);
for (i = 0; i < MLX4_MAX_PORTS; ++i) {
- struct net_device *curr_netdev = ibdev->iboe.netdevs[i];
+ struct net_device *curr_netdev = ibdev->iboe->netdevs[i];
enum ib_port_state curr_port_state;
if (!curr_netdev)
@@ -3234,7 +3249,7 @@ static void handle_bonded_port_state_event(struct work_struct *work)
bonded_port_state = (bonded_port_state != IB_PORT_ACTIVE) ?
curr_port_state : IB_PORT_ACTIVE;
}
- spin_unlock_bh(&ibdev->iboe.lock);
+ spin_unlock_bh(&ibdev->iboe->lock);
ibev.device = &ibdev->ib_dev;
ibev.element.port_num = 1;
diff --git a/drivers/infiniband/hw/mlx4/mlx4_ib.h b/drivers/infiniband/hw/mlx4/mlx4_ib.h
index e10dccc..2996c61 100644
--- a/drivers/infiniband/hw/mlx4/mlx4_ib.h
+++ b/drivers/infiniband/hw/mlx4/mlx4_ib.h
@@ -524,6 +524,7 @@ struct mlx4_ib_iboe {
atomic64_t mac[MLX4_MAX_PORTS];
struct notifier_block nb;
struct mlx4_port_gid_table gids[MLX4_MAX_PORTS];
+ struct mlx4_ib_dev *parent;
};
struct pkey_mgt {
@@ -600,7 +601,7 @@ struct mlx4_ib_dev {
struct mutex cap_mask_mutex;
bool ib_active;
- struct mlx4_ib_iboe iboe;
+ struct mlx4_ib_iboe *iboe;
struct mlx4_ib_counters counters_table[MLX4_MAX_PORTS];
int *eq_table;
struct kobject *iov_parent;
diff --git a/drivers/infiniband/hw/mlx4/qp.c b/drivers/infiniband/hw/mlx4/qp.c
index 6dd3cd2..853ef6f 100644
--- a/drivers/infiniband/hw/mlx4/qp.c
+++ b/drivers/infiniband/hw/mlx4/qp.c
@@ -1868,7 +1868,7 @@ static int handle_eth_ud_smac_index(struct mlx4_ib_dev *dev,
u64 u64_mac;
int smac_index;
- u64_mac = atomic64_read(&dev->iboe.mac[qp->port - 1]);
+ u64_mac = atomic64_read(&dev->iboe->mac[qp->port - 1]);
context->pri_path.sched_queue = MLX4_IB_DEFAULT_SCHED_QUEUE | ((qp->port - 1) << 6);
if (!qp->pri.smac && !qp->pri.smac_port) {
@@ -2926,7 +2926,7 @@ static int fill_gid_by_hw_index(struct mlx4_ib_dev *ibdev, u8 port_num,
int index, union ib_gid *gid,
enum ib_gid_type *gid_type)
{
- struct mlx4_ib_iboe *iboe = &ibdev->iboe;
+ struct mlx4_ib_iboe *iboe = ibdev->iboe;
struct mlx4_port_gid_table *port_gid_table;
unsigned long flags;
--
2.1.4
On Tue, Sep 18, 2018 at 04:03:43PM +0300, Jan Dakinevich wrote:
> IB infrastructure shares common device instance constructor with
> reference counting, and it uses kzalloc() to allocate memory
> for device specific instance with incapsulated ib_device field as one
> contigous memory block.
>
> The issue is that the device specific instances tend to be too large
> and require high page order memory allocation. Unfortunately, kzalloc()
> in ib_alloc_device() can not be replaced with kvzalloc() since it would
> require a lot of review in all IB driver to prove correctness of the
> replacement.
>
> The driver can allocate some heavy partes of their instance for itself
> and keep pointers for them in own instance. For this it is important
> that the alocated parts have the same life time as ib_device, thus
> their deallocation should be based on the same reference counting.
>
> Let suppose:
>
> struct foo_ib_device {
> struct ib_device device;
>
> void *part;
>
> ...
> };
>
> To properly free memory from .foo_ib_part the driver should provide
> function for ->release() callback:
>
> void foo_ib_release(struct ib_device *device)
> {
> struct foo_ib_device *foo = container_of(device, struct foo_ib_device,
> device);
>
> kvfree(foo->part);
> }
>
> ...and initialiaze this callback immediately after foo_ib_device
> instance allocation.
>
> struct foo_ib_device *foo;
>
> foo = ib_alloc_device(sizeof(struct foo_ib_device));
>
> foo->device.release = foo_ib_release;
>
> /* allocate parts */
> foo->part = kvmalloc(65536, GFP_KERNEL);
>
> Signed-off-by: Jan Dakinevich <[email protected]>
> drivers/infiniband/core/device.c | 2 ++
> include/rdma/ib_verbs.h | 2 ++
> 2 files changed, 4 insertions(+)
>
> diff --git a/drivers/infiniband/core/device.c b/drivers/infiniband/core/device.c
> index db3b627..a8c8b0d 100644
> +++ b/drivers/infiniband/core/device.c
> @@ -215,6 +215,8 @@ static void ib_device_release(struct device *device)
> ib_cache_release_one(dev);
> kfree(dev->port_immutable);
> }
> + if (dev->release)
> + dev->release(dev);
> kfree(dev);
> }
Nope, the driver module could be unloaded at this point.
The driver should free memory after its call to ib_unregister_device
returns.
Jason
On Tue, Sep 18, 2018 at 04:03:42PM +0300, Jan Dakinevich wrote:
> The size of mlx4_ib_device became too large to be allocated as whole contigous
> block of memory. Currently it takes about 55K. On architecture with 4K page it
> means 3rd order.
>
> This patch series makes an attempt to split mlx4_ib_device into several parts
> and allocate them with less expensive kvzalloc
Why split it up? Any reason not to just allocate the whole thing with
kvzalloc?
Jason
On Tue, Sep 18, 2018 at 08:46:23AM -0600, Jason Gunthorpe wrote:
> On Tue, Sep 18, 2018 at 04:03:42PM +0300, Jan Dakinevich wrote:
> > The size of mlx4_ib_device became too large to be allocated as whole contigous
> > block of memory. Currently it takes about 55K. On architecture with 4K page it
> > means 3rd order.
> >
> > This patch series makes an attempt to split mlx4_ib_device into several parts
> > and allocate them with less expensive kvzalloc
>
> Why split it up? Any reason not to just allocate the whole thing with
> kvzalloc?
And before we are rushing to dissect mlx4_ib driver, can you
explain the rationale behind this change? The mlx4_ib driver
represents high-performance device which needs enough memory
resources to operate. Those devices are limited by number
of PCIs and SRIOV VFs (upto 126) and very rare allocated/deallocated.
I would like to see real rationale behind such change.
Thanks
>
> Jason
On Wed, 19 Sep 2018 00:23:51 +0300
Leon Romanovsky <[email protected]> wrote:
> On Tue, Sep 18, 2018 at 08:46:23AM -0600, Jason Gunthorpe wrote:
> > On Tue, Sep 18, 2018 at 04:03:42PM +0300, Jan Dakinevich wrote:
> > > The size of mlx4_ib_device became too large to be allocated as
> > > whole contigous block of memory. Currently it takes about 55K. On
> > > architecture with 4K page it means 3rd order.
> > >
> > > This patch series makes an attempt to split mlx4_ib_device into
> > > several parts and allocate them with less expensive kvzalloc
> >
> > Why split it up? Any reason not to just allocate the whole thing
> > with kvzalloc?
>
This allocation could be triggered by userspace. It means that at
_arbitrary_ time kernel could be asked for high order allocation.
This case is considered unacceptable for system under significant load,
since kernel would try to satisfy this memory request wasting the
overall performance.
> And before we are rushing to dissect mlx4_ib driver, can you
> explain the rationale behind this change? The mlx4_ib driver
> represents high-performance device which needs enough memory
> resources to operate. Those devices are limited by number
> of PCIs and SRIOV VFs (upto 126) and very rare allocated/deallocated.
>
> I would like to see real rationale behind such change.
>
> Thanks
>
> >
> > Jason
--
Best regards
Jan Dakinevich
On Tue, 18 Sep 2018 08:46:23 -0600
Jason Gunthorpe <[email protected]> wrote:
> On Tue, Sep 18, 2018 at 04:03:42PM +0300, Jan Dakinevich wrote:
> > The size of mlx4_ib_device became too large to be allocated as
> > whole contigous block of memory. Currently it takes about 55K. On
> > architecture with 4K page it means 3rd order.
> >
> > This patch series makes an attempt to split mlx4_ib_device into
> > several parts and allocate them with less expensive kvzalloc
>
> Why split it up? Any reason not to just allocate the whole thing with
> kvzalloc?
>
To allocate whole ib_device with kvmalloc I will need replace kzalloc()
by kvzalloc() in ib_alloc_device() and then review allocation, to make
sure that no one uses this memory for DMA.
Although, I could introduce new ib_*alloc_device allocator for these
needs...
> Jason
--
Best regards
Jan Dakinevich
On Wed, Sep 26, 2018 at 06:43:42PM +0300, Jan Dakinevich wrote:
> On Wed, 19 Sep 2018 00:23:51 +0300
> Leon Romanovsky <[email protected]> wrote:
>
> > On Tue, Sep 18, 2018 at 08:46:23AM -0600, Jason Gunthorpe wrote:
> > > On Tue, Sep 18, 2018 at 04:03:42PM +0300, Jan Dakinevich wrote:
> > > > The size of mlx4_ib_device became too large to be allocated as
> > > > whole contigous block of memory. Currently it takes about 55K. On
> > > > architecture with 4K page it means 3rd order.
> > > >
> > > > This patch series makes an attempt to split mlx4_ib_device into
> > > > several parts and allocate them with less expensive kvzalloc
> > >
> > > Why split it up? Any reason not to just allocate the whole thing
> > > with kvzalloc?
> >
>
> This allocation could be triggered by userspace. It means that at
> _arbitrary_ time kernel could be asked for high order allocation.
>
> This case is considered unacceptable for system under significant load,
> since kernel would try to satisfy this memory request wasting the
> overall performance.
In such case, you won't do very much with mlx4_ib device. It will be
unusable.
>
> > And before we are rushing to dissect mlx4_ib driver, can you
> > explain the rationale behind this change? The mlx4_ib driver
> > represents high-performance device which needs enough memory
> > resources to operate. Those devices are limited by number
> > of PCIs and SRIOV VFs (upto 126) and very rare allocated/deallocated.
> >
> > I would like to see real rationale behind such change.
> >
> > Thanks
> >
> > >
> > > Jason
>
>
>
> --
> Best regards
> Jan Dakinevich
On Wed, Sep 26, 2018 at 06:48:09PM +0300, Jan Dakinevich wrote:
> On Tue, 18 Sep 2018 08:46:23 -0600
> Jason Gunthorpe <[email protected]> wrote:
>
> > On Tue, Sep 18, 2018 at 04:03:42PM +0300, Jan Dakinevich wrote:
> > > The size of mlx4_ib_device became too large to be allocated as
> > > whole contigous block of memory. Currently it takes about 55K. On
> > > architecture with 4K page it means 3rd order.
> > >
> > > This patch series makes an attempt to split mlx4_ib_device into
> > > several parts and allocate them with less expensive kvzalloc
> >
> > Why split it up? Any reason not to just allocate the whole thing with
> > kvzalloc?
> >
>
> To allocate whole ib_device with kvmalloc I will need replace kzalloc()
> by kvzalloc() in ib_alloc_device() and then review allocation, to make
> sure that no one uses this memory for DMA.
I would be shocked if some driver was DMA'ing out of struct ib_device
memory..
Jason