Signed-off-by: Joachim Fenkes <[email protected]>
---
drivers/infiniband/hw/ehca/ehca_reqs.c | 6 ++++--
1 files changed, 4 insertions(+), 2 deletions(-)
diff --git a/drivers/infiniband/hw/ehca/ehca_reqs.c b/drivers/infiniband/hw/ehca/ehca_reqs.c
index bbe0436..f093b00 100644
--- a/drivers/infiniband/hw/ehca/ehca_reqs.c
+++ b/drivers/infiniband/hw/ehca/ehca_reqs.c
@@ -421,8 +421,10 @@ int ehca_post_send(struct ib_qp *qp,
int ret = 0;
unsigned long flags;
- if (unlikely(my_qp->state != IB_QPS_RTS)) {
- ehca_err(qp->device, "QP not in RTS state qpn=%x", qp->qp_num);
+ /* Reject WR if QP is in RESET, INIT or RTR state */
+ if (unlikely(my_qp->state < IB_QPS_RTS)) {
+ ehca_err(qp->device, "Invalid QP state qp_state=%d qpn=%x",
+ my_qp->state, qp->qp_num);
return -EINVAL;
}
--
1.5.5
Hi.
I checked the code in the ehca driver and noticed that
post RR to a QP is being accepted in any state (including the RESET state).
thanks
Dotan
Joachim Fenkes wrote:
> Signed-off-by: Joachim Fenkes <[email protected]>
> ---
> drivers/infiniband/hw/ehca/ehca_reqs.c | 6 ++++--
> 1 files changed, 4 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/infiniband/hw/ehca/ehca_reqs.c b/drivers/infiniband/hw/ehca/ehca_reqs.c
> index bbe0436..f093b00 100644
> --- a/drivers/infiniband/hw/ehca/ehca_reqs.c
> +++ b/drivers/infiniband/hw/ehca/ehca_reqs.c
> @@ -421,8 +421,10 @@ int ehca_post_send(struct ib_qp *qp,
> int ret = 0;
> unsigned long flags;
>
> - if (unlikely(my_qp->state != IB_QPS_RTS)) {
> - ehca_err(qp->device, "QP not in RTS state qpn=%x", qp->qp_num);
> + /* Reject WR if QP is in RESET, INIT or RTR state */
> + if (unlikely(my_qp->state < IB_QPS_RTS)) {
> + ehca_err(qp->device, "Invalid QP state qp_state=%d qpn=%x",
> + my_qp->state, qp->qp_num);
> return -EINVAL;
> }
>
>
Signed-off-by: Joachim Fenkes <[email protected]>
---
On Friday 06 June 2008 22:20, Dotan Barak wrote:
> I checked the code in the ehca driver and noticed that
> post RR to a QP is being accepted in any state (including the RESET state).
You're right, this is only consistent -- thanks for pointing it out!
Regards,
Joachim
drivers/infiniband/hw/ehca/ehca_reqs.c | 12 ++++++++++--
1 files changed, 10 insertions(+), 2 deletions(-)
diff --git a/drivers/infiniband/hw/ehca/ehca_reqs.c b/drivers/infiniband/hw/ehca/ehca_reqs.c
index f093b00..ad197f4 100644
--- a/drivers/infiniband/hw/ehca/ehca_reqs.c
+++ b/drivers/infiniband/hw/ehca/ehca_reqs.c
@@ -544,8 +544,16 @@ int ehca_post_recv(struct ib_qp *qp,
struct ib_recv_wr *recv_wr,
struct ib_recv_wr **bad_recv_wr)
{
- return internal_post_recv(container_of(qp, struct ehca_qp, ib_qp),
- qp->device, recv_wr, bad_recv_wr);
+ struct ehca_qp *my_qp = container_of(qp, struct ehca_qp, ib_qp);
+
+ /* Reject WR if QP is in RESET state */
+ if (unlikely(my_qp->state == IB_QPS_RESET)) {
+ ehca_err(qp->device, "Invalid QP state qp_state=%d qpn=%x",
+ my_qp->state, qp->qp_num);
+ return -EINVAL;
+ }
+
+ return internal_post_recv(my_qp, qp->device, recv_wr, bad_recv_wr);
}
int ehca_post_srq_recv(struct ib_srq *srq,
--
1.5.5
This gives ehca an autogenerated modalias and therefore enables automatic loading.
Signed-off-by: Joachim Fenkes <[email protected]>
---
drivers/infiniband/hw/ehca/ehca_main.c | 1 +
1 files changed, 1 insertions(+), 0 deletions(-)
diff --git a/drivers/infiniband/hw/ehca/ehca_main.c b/drivers/infiniband/hw/ehca/ehca_main.c
index 482103e..598844d 100644
--- a/drivers/infiniband/hw/ehca/ehca_main.c
+++ b/drivers/infiniband/hw/ehca/ehca_main.c
@@ -923,6 +923,7 @@ static struct of_device_id ehca_device_table[] =
},
{},
};
+MODULE_DEVICE_TABLE(of, ehca_device_table);
static struct of_platform_driver ehca_driver = {
.name = "ehca",
--
1.5.5
A previous fix introduced a regression where port activation events were
dropped unconditionally if port autodetection was not enabled. Fixed.
Signed-off-by: Joachim Fenkes <[email protected]>
---
Roland -- this patch is made against your for-linus branch. Please review
and apply if you think it's okay. Hope it's not too late for the next kernel.
Joachim
drivers/infiniband/hw/ehca/ehca_irq.c | 45 +++++++++++++++++++-------------
1 files changed, 27 insertions(+), 18 deletions(-)
diff --git a/drivers/infiniband/hw/ehca/ehca_irq.c b/drivers/infiniband/hw/ehca/ehca_irq.c
index 9e43459..757035e 100644
--- a/drivers/infiniband/hw/ehca/ehca_irq.c
+++ b/drivers/infiniband/hw/ehca/ehca_irq.c
@@ -359,34 +359,43 @@ static void notify_port_conf_change(struct ehca_shca *shca, int port_num)
*old_attr = new_attr;
}
+/* replay modify_qp for sqps -- return 0 if all is well, 1 if AQP1 destroyed */
+static int replay_modify_qp(struct ehca_sport *sport)
+{
+ int aqp1_destroyed;
+ unsigned long flags;
+
+ spin_lock_irqsave(&sport->mod_sqp_lock, flags);
+
+ aqp1_destroyed = !sport->ibqp_sqp[IB_QPT_GSI];
+
+ if (sport->ibqp_sqp[IB_QPT_SMI])
+ ehca_recover_sqp(sport->ibqp_sqp[IB_QPT_SMI]);
+ if (!aqp1_destroyed)
+ ehca_recover_sqp(sport->ibqp_sqp[IB_QPT_GSI]);
+
+ spin_unlock_irqrestore(&sport->mod_sqp_lock, flags);
+
+ return aqp1_destroyed;
+}
+
static void parse_ec(struct ehca_shca *shca, u64 eqe)
{
u8 ec = EHCA_BMASK_GET(NEQE_EVENT_CODE, eqe);
u8 port = EHCA_BMASK_GET(NEQE_PORT_NUMBER, eqe);
u8 spec_event;
struct ehca_sport *sport = &shca->sport[port - 1];
- unsigned long flags;
switch (ec) {
case 0x30: /* port availability change */
if (EHCA_BMASK_GET(NEQE_PORT_AVAILABILITY, eqe)) {
- /* only for autodetect mode important */
- if (ehca_nr_ports >= 0)
- break;
-
- int suppress_event;
- /* replay modify_qp for sqps */
- spin_lock_irqsave(&sport->mod_sqp_lock, flags);
- suppress_event = !sport->ibqp_sqp[IB_QPT_GSI];
- if (sport->ibqp_sqp[IB_QPT_SMI])
- ehca_recover_sqp(sport->ibqp_sqp[IB_QPT_SMI]);
- if (!suppress_event)
- ehca_recover_sqp(sport->ibqp_sqp[IB_QPT_GSI]);
- spin_unlock_irqrestore(&sport->mod_sqp_lock, flags);
-
- /* AQP1 was destroyed, ignore this event */
- if (suppress_event)
- break;
+ /* only replay modify_qp calls in autodetect mode;
+ * if AQP1 was destroyed, the port is already down
+ * again and we can drop the event.
+ */
+ if (ehca_nr_ports < 0)
+ if (replay_modify_qp(sport))
+ break;
sport->port_state = IB_PORT_ACTIVE;
dispatch_port_event(shca, port, IB_EVENT_PORT_ACTIVE,
--
1.5.5
> A previous fix introduced a regression where port activation events were
> dropped unconditionally if port autodetection was not enabled. Fixed.
Is this a fix to "IB/ehca: Remove reference to special QP in case of
port activation failure"? Because if so I can roll it into that patch,
since Linus hasn't pulled it yet.
- R.
From: Michael Ellerman <[email protected]>
shca_list_lock is taken from softirq context in ehca_poll_eqs, so we need to
lock IRQ safe elsewhere.
Signed-off-by: Michael Ellerman <[email protected]>
Acked-by: Joachim Fenkes <[email protected]>
---
drivers/infiniband/hw/ehca/ehca_main.c | 17 ++++++++++-------
1 files changed, 10 insertions(+), 7 deletions(-)
diff --git a/drivers/infiniband/hw/ehca/ehca_main.c b/drivers/infiniband/hw/ehca/ehca_main.c
index bb02a86..021c454 100644
--- a/drivers/infiniband/hw/ehca/ehca_main.c
+++ b/drivers/infiniband/hw/ehca/ehca_main.c
@@ -717,6 +717,7 @@ static int __devinit ehca_probe(struct of_device *dev,
const u64 *handle;
struct ib_pd *ibpd;
int ret, i, eq_size;
+ u64 flags;
handle = of_get_property(dev->node, "ibm,hca-handle", NULL);
if (!handle) {
@@ -830,9 +831,9 @@ static int __devinit ehca_probe(struct of_device *dev,
ehca_err(&shca->ib_device,
"Cannot create device attributes ret=%d", ret);
- spin_lock(&shca_list_lock);
+ spin_lock_irqsave(&shca_list_lock, flags);
list_add(&shca->shca_list, &shca_list);
- spin_unlock(&shca_list_lock);
+ spin_unlock_irqrestore(&shca_list_lock, flags);
return 0;
@@ -878,6 +879,7 @@ probe1:
static int __devexit ehca_remove(struct of_device *dev)
{
struct ehca_shca *shca = dev->dev.driver_data;
+ u64 flags;
int ret;
sysfs_remove_group(&dev->dev.kobj, &ehca_dev_attr_grp);
@@ -915,9 +917,9 @@ static int __devexit ehca_remove(struct of_device *dev)
ib_dealloc_device(&shca->ib_device);
- spin_lock(&shca_list_lock);
+ spin_lock_irqsave(&shca_list_lock, flags);
list_del(&shca->shca_list);
- spin_unlock(&shca_list_lock);
+ spin_unlock_irqrestore(&shca_list_lock, flags);
return ret;
}
@@ -975,6 +977,7 @@ static int ehca_mem_notifier(struct notifier_block *nb,
unsigned long action, void *data)
{
static unsigned long ehca_dmem_warn_time;
+ unsigned long flags;
switch (action) {
case MEM_CANCEL_OFFLINE:
@@ -985,12 +988,12 @@ static int ehca_mem_notifier(struct notifier_block *nb,
case MEM_GOING_ONLINE:
case MEM_GOING_OFFLINE:
/* only ok if no hca is attached to the lpar */
- spin_lock(&shca_list_lock);
+ spin_lock_irqsave(&shca_list_lock, flags);
if (list_empty(&shca_list)) {
- spin_unlock(&shca_list_lock);
+ spin_unlock_irqrestore(&shca_list_lock, flags);
return NOTIFY_OK;
} else {
- spin_unlock(&shca_list_lock);
+ spin_unlock_irqrestore(&shca_list_lock, flags);
if (printk_timed_ratelimit(&ehca_dmem_warn_time,
30 * 1000))
ehca_gen_err("DMEM operations are not allowed"
--
1.5.5
On Fri, 2008-11-21 at 16:37 +0100, Joachim Fenkes wrote:
> + u64 flags;
> - spin_lock(&shca_list_lock);
> + spin_lock_irqsave(&shca_list_lock, flags);
That's wrong and I think will give a warning on all machines where
u64 != unsigned long. Might not particularly matter in this case.
Also, generally it seems wrong to say "fix lockdep failure" when the
patch really fixes a bug that lockdep happened to find.
johannes
shca_list_lock is taken from softirq context in ehca_poll_eqs, so we need to
lock IRQ safe elsewhere.
Signed-off-by: Michael Ellerman <[email protected]>
Signed-off-by: Joachim Fenkes <[email protected]>
---
On Friday 21 November 2008 17:02, Johannes Berg wrote:
> On Fri, 2008-11-21 at 16:37 +0100, Joachim Fenkes wrote:
>
> > + u64 flags;
>
> > - spin_lock(&shca_list_lock);
> > + spin_lock_irqsave(&shca_list_lock, flags);
>
> That's wrong and I think will give a warning on all machines where
> u64 != unsigned long. Might not particularly matter in this case.
Doesn't matter for a ppc64 only driver, but you're right nonetheless. Thanks.
> Also, generally it seems wrong to say "fix lockdep failure" when the
> patch really fixes a bug that lockdep happened to find.
Whatever -- changed.
Here's the updated patch.
Regards,
Joachim
drivers/infiniband/hw/ehca/ehca_main.c | 17 ++++++++++-------
1 files changed, 10 insertions(+), 7 deletions(-)
diff --git a/drivers/infiniband/hw/ehca/ehca_main.c b/drivers/infiniband/hw/ehca/ehca_main.c
index bb02a86..169aa1a 100644
--- a/drivers/infiniband/hw/ehca/ehca_main.c
+++ b/drivers/infiniband/hw/ehca/ehca_main.c
@@ -717,6 +717,7 @@ static int __devinit ehca_probe(struct of_device *dev,
const u64 *handle;
struct ib_pd *ibpd;
int ret, i, eq_size;
+ unsigned long flags;
handle = of_get_property(dev->node, "ibm,hca-handle", NULL);
if (!handle) {
@@ -830,9 +831,9 @@ static int __devinit ehca_probe(struct of_device *dev,
ehca_err(&shca->ib_device,
"Cannot create device attributes ret=%d", ret);
- spin_lock(&shca_list_lock);
+ spin_lock_irqsave(&shca_list_lock, flags);
list_add(&shca->shca_list, &shca_list);
- spin_unlock(&shca_list_lock);
+ spin_unlock_irqrestore(&shca_list_lock, flags);
return 0;
@@ -878,6 +879,7 @@ probe1:
static int __devexit ehca_remove(struct of_device *dev)
{
struct ehca_shca *shca = dev->dev.driver_data;
+ unsigned long flags;
int ret;
sysfs_remove_group(&dev->dev.kobj, &ehca_dev_attr_grp);
@@ -915,9 +917,9 @@ static int __devexit ehca_remove(struct of_device *dev)
ib_dealloc_device(&shca->ib_device);
- spin_lock(&shca_list_lock);
+ spin_lock_irqsave(&shca_list_lock, flags);
list_del(&shca->shca_list);
- spin_unlock(&shca_list_lock);
+ spin_unlock_irqrestore(&shca_list_lock, flags);
return ret;
}
@@ -975,6 +977,7 @@ static int ehca_mem_notifier(struct notifier_block *nb,
unsigned long action, void *data)
{
static unsigned long ehca_dmem_warn_time;
+ unsigned long flags;
switch (action) {
case MEM_CANCEL_OFFLINE:
@@ -985,12 +988,12 @@ static int ehca_mem_notifier(struct notifier_block *nb,
case MEM_GOING_ONLINE:
case MEM_GOING_OFFLINE:
/* only ok if no hca is attached to the lpar */
- spin_lock(&shca_list_lock);
+ spin_lock_irqsave(&shca_list_lock, flags);
if (list_empty(&shca_list)) {
- spin_unlock(&shca_list_lock);
+ spin_unlock_irqrestore(&shca_list_lock, flags);
return NOTIFY_OK;
} else {
- spin_unlock(&shca_list_lock);
+ spin_unlock_irqrestore(&shca_list_lock, flags);
if (printk_timed_ratelimit(&ehca_dmem_warn_time,
30 * 1000))
ehca_gen_err("DMEM operations are not allowed"
--
1.5.5
Looks good... I'll add this for 2.6.29, since as far as I can tell this
bug has been there approximately forever already.
On Fri, 2008-11-21 at 17:02 +0100, Johannes Berg wrote:
> On Fri, 2008-11-21 at 16:37 +0100, Joachim Fenkes wrote:
>
> > + u64 flags;
>
> > - spin_lock(&shca_list_lock);
> > + spin_lock_irqsave(&shca_list_lock, flags);
>
> That's wrong and I think will give a warning on all machines where
> u64 != unsigned long. Might not particularly matter in this case.
Crud, sorry.
> Also, generally it seems wrong to say "fix lockdep failure" when the
> patch really fixes a bug that lockdep happened to find.
True. I guess it should be "fix locking error found with lockdep", to
make it clear no one has actually hit the bug.
cheers
--
Michael Ellerman
OzLabs, IBM Australia Development Lab
wwweb: http://michael.ellerman.id.au
phone: +61 2 6212 1183 (tie line 70 21183)
We do not inherit the earth from our ancestors,
we borrow it from our children. - S.M.A.R.T Person
The error message printed when the eHCA driver prevents memory hotplug is
misleading -- the user might think that hot-removing the lhca, hotplugging
memory, then hot-adding the lhca again will work, but it doesn't.
Signed-off-by: Joachim Fenkes <[email protected]>
---
drivers/infiniband/hw/ehca/ehca_main.c | 3 +--
1 files changed, 1 insertions(+), 2 deletions(-)
diff --git a/drivers/infiniband/hw/ehca/ehca_main.c b/drivers/infiniband/hw/ehca/ehca_main.c
index bb02a86..bec7e02 100644
--- a/drivers/infiniband/hw/ehca/ehca_main.c
+++ b/drivers/infiniband/hw/ehca/ehca_main.c
@@ -994,8 +994,7 @@ static int ehca_mem_notifier(struct notifier_block *nb,
if (printk_timed_ratelimit(&ehca_dmem_warn_time,
30 * 1000))
ehca_gen_err("DMEM operations are not allowed"
- "as long as an ehca adapter is"
- "attached to the LPAR");
+ "in conjunction with eHCA");
return NOTIFY_BAD;
}
}
--
1.5.5
> The error message printed when the eHCA driver prevents memory hotplug is
> misleading -- the user might think that hot-removing the lhca, hotplugging
> memory, then hot-adding the lhca again will work, but it doesn't.
That's too bad... I applied this patch but out of curiousity, why
doesn't the hot-remove/hot-add work? I would have thought that
re-registering all of memory after the hot-add would do the right thing.
Roland Dreier <[email protected]> wrote on 26.11.2008 00:13:51:
> That's too bad... I applied this patch but out of curiousity, why
> doesn't the hot-remove/hot-add work? I would have thought that
> re-registering all of memory after the hot-add would do the right thing.
That's right, but right now, we simply try to register all of memory from
KERNELBASE to high_memory, which works right until we have memory holes in
the middle; then the hypervisor will reject our page registrations. Same
goes for huge (16GB) pages, by the way. We're working on a solution to
this.
Cheers,
Joachim
Roland Dreier <[email protected]> wrote on 10.11.2008 21:36:23:
> > A previous fix introduced a regression where port activation events
were
> > dropped unconditionally if port autodetection was not enabled. Fixed.
>
> Is this a fix to "IB/ehca: Remove reference to special QP in case of
> port activation failure"? Because if so I can roll it into that patch,
> since Linus hasn't pulled it yet.
Yes, that would be splendid, thank you!
Cheers,
Joachim
All the fields in the control block are nicely right-aligned, so no masking
is necessary.
Signed-off-by: Joachim Fenkes <[email protected]>
---
drivers/infiniband/hw/ehca/ehca_classes_pSeries.h | 28 ---------------------
drivers/infiniband/hw/ehca/ehca_qp.c | 18 +++---------
2 files changed, 5 insertions(+), 41 deletions(-)
diff --git a/drivers/infiniband/hw/ehca/ehca_classes_pSeries.h b/drivers/infiniband/hw/ehca/ehca_classes_pSeries.h
index 1798e64..689c357 100644
--- a/drivers/infiniband/hw/ehca/ehca_classes_pSeries.h
+++ b/drivers/infiniband/hw/ehca/ehca_classes_pSeries.h
@@ -165,7 +165,6 @@ struct hcp_modify_qp_control_block {
#define MQPCB_MASK_ALT_P_KEY_IDX EHCA_BMASK_IBM( 7, 7)
#define MQPCB_MASK_RDMA_ATOMIC_CTRL EHCA_BMASK_IBM( 8, 8)
#define MQPCB_MASK_QP_STATE EHCA_BMASK_IBM( 9, 9)
-#define MQPCB_QP_STATE EHCA_BMASK_IBM(24, 31)
#define MQPCB_MASK_RDMA_NR_ATOMIC_RESP_RES EHCA_BMASK_IBM(11, 11)
#define MQPCB_MASK_PATH_MIGRATION_STATE EHCA_BMASK_IBM(12, 12)
#define MQPCB_MASK_RDMA_ATOMIC_OUTST_DEST_QP EHCA_BMASK_IBM(13, 13)
@@ -176,60 +175,33 @@ struct hcp_modify_qp_control_block {
#define MQPCB_MASK_RETRY_COUNT EHCA_BMASK_IBM(18, 18)
#define MQPCB_MASK_TIMEOUT EHCA_BMASK_IBM(19, 19)
#define MQPCB_MASK_PATH_MTU EHCA_BMASK_IBM(20, 20)
-#define MQPCB_PATH_MTU EHCA_BMASK_IBM(24, 31)
#define MQPCB_MASK_MAX_STATIC_RATE EHCA_BMASK_IBM(21, 21)
-#define MQPCB_MAX_STATIC_RATE EHCA_BMASK_IBM(24, 31)
#define MQPCB_MASK_DLID EHCA_BMASK_IBM(22, 22)
-#define MQPCB_DLID EHCA_BMASK_IBM(16, 31)
#define MQPCB_MASK_RNR_RETRY_COUNT EHCA_BMASK_IBM(23, 23)
-#define MQPCB_RNR_RETRY_COUNT EHCA_BMASK_IBM(29, 31)
#define MQPCB_MASK_SOURCE_PATH_BITS EHCA_BMASK_IBM(24, 24)
-#define MQPCB_SOURCE_PATH_BITS EHCA_BMASK_IBM(25, 31)
#define MQPCB_MASK_TRAFFIC_CLASS EHCA_BMASK_IBM(25, 25)
-#define MQPCB_TRAFFIC_CLASS EHCA_BMASK_IBM(24, 31)
#define MQPCB_MASK_HOP_LIMIT EHCA_BMASK_IBM(26, 26)
-#define MQPCB_HOP_LIMIT EHCA_BMASK_IBM(24, 31)
#define MQPCB_MASK_SOURCE_GID_IDX EHCA_BMASK_IBM(27, 27)
-#define MQPCB_SOURCE_GID_IDX EHCA_BMASK_IBM(24, 31)
#define MQPCB_MASK_FLOW_LABEL EHCA_BMASK_IBM(28, 28)
-#define MQPCB_FLOW_LABEL EHCA_BMASK_IBM(12, 31)
#define MQPCB_MASK_DEST_GID EHCA_BMASK_IBM(30, 30)
#define MQPCB_MASK_SERVICE_LEVEL_AL EHCA_BMASK_IBM(31, 31)
-#define MQPCB_SERVICE_LEVEL_AL EHCA_BMASK_IBM(28, 31)
#define MQPCB_MASK_SEND_GRH_FLAG_AL EHCA_BMASK_IBM(32, 32)
-#define MQPCB_SEND_GRH_FLAG_AL EHCA_BMASK_IBM(31, 31)
#define MQPCB_MASK_RETRY_COUNT_AL EHCA_BMASK_IBM(33, 33)
-#define MQPCB_RETRY_COUNT_AL EHCA_BMASK_IBM(29, 31)
#define MQPCB_MASK_TIMEOUT_AL EHCA_BMASK_IBM(34, 34)
-#define MQPCB_TIMEOUT_AL EHCA_BMASK_IBM(27, 31)
#define MQPCB_MASK_MAX_STATIC_RATE_AL EHCA_BMASK_IBM(35, 35)
-#define MQPCB_MAX_STATIC_RATE_AL EHCA_BMASK_IBM(24, 31)
#define MQPCB_MASK_DLID_AL EHCA_BMASK_IBM(36, 36)
-#define MQPCB_DLID_AL EHCA_BMASK_IBM(16, 31)
#define MQPCB_MASK_RNR_RETRY_COUNT_AL EHCA_BMASK_IBM(37, 37)
-#define MQPCB_RNR_RETRY_COUNT_AL EHCA_BMASK_IBM(29, 31)
#define MQPCB_MASK_SOURCE_PATH_BITS_AL EHCA_BMASK_IBM(38, 38)
-#define MQPCB_SOURCE_PATH_BITS_AL EHCA_BMASK_IBM(25, 31)
#define MQPCB_MASK_TRAFFIC_CLASS_AL EHCA_BMASK_IBM(39, 39)
-#define MQPCB_TRAFFIC_CLASS_AL EHCA_BMASK_IBM(24, 31)
#define MQPCB_MASK_HOP_LIMIT_AL EHCA_BMASK_IBM(40, 40)
-#define MQPCB_HOP_LIMIT_AL EHCA_BMASK_IBM(24, 31)
#define MQPCB_MASK_SOURCE_GID_IDX_AL EHCA_BMASK_IBM(41, 41)
-#define MQPCB_SOURCE_GID_IDX_AL EHCA_BMASK_IBM(24, 31)
#define MQPCB_MASK_FLOW_LABEL_AL EHCA_BMASK_IBM(42, 42)
-#define MQPCB_FLOW_LABEL_AL EHCA_BMASK_IBM(12, 31)
#define MQPCB_MASK_DEST_GID_AL EHCA_BMASK_IBM(44, 44)
#define MQPCB_MASK_MAX_NR_OUTST_SEND_WR EHCA_BMASK_IBM(45, 45)
-#define MQPCB_MAX_NR_OUTST_SEND_WR EHCA_BMASK_IBM(16, 31)
#define MQPCB_MASK_MAX_NR_OUTST_RECV_WR EHCA_BMASK_IBM(46, 46)
-#define MQPCB_MAX_NR_OUTST_RECV_WR EHCA_BMASK_IBM(16, 31)
#define MQPCB_MASK_DISABLE_ETE_CREDIT_CHECK EHCA_BMASK_IBM(47, 47)
-#define MQPCB_DISABLE_ETE_CREDIT_CHECK EHCA_BMASK_IBM(31, 31)
-#define MQPCB_QP_NUMBER EHCA_BMASK_IBM( 8, 31)
#define MQPCB_MASK_QP_ENABLE EHCA_BMASK_IBM(48, 48)
-#define MQPCB_QP_ENABLE EHCA_BMASK_IBM(31, 31)
#define MQPCB_MASK_CURR_SRQ_LIMIT EHCA_BMASK_IBM(49, 49)
-#define MQPCB_CURR_SRQ_LIMIT EHCA_BMASK_IBM(16, 31)
#define MQPCB_MASK_QP_AFF_ASYN_EV_LOG_REG EHCA_BMASK_IBM(50, 50)
#define MQPCB_MASK_SHARED_RQ_HNDL EHCA_BMASK_IBM(51, 51)
diff --git a/drivers/infiniband/hw/ehca/ehca_qp.c b/drivers/infiniband/hw/ehca/ehca_qp.c
index ead4e71..0338f1f 100644
--- a/drivers/infiniband/hw/ehca/ehca_qp.c
+++ b/drivers/infiniband/hw/ehca/ehca_qp.c
@@ -1962,19 +1962,13 @@ int ehca_query_qp(struct ib_qp *qp,
qp_attr->cap.max_inline_data = my_qp->sq_max_inline_data_size;
qp_attr->dest_qp_num = qpcb->dest_qp_nr;
- qp_attr->pkey_index =
- EHCA_BMASK_GET(MQPCB_PRIM_P_KEY_IDX, qpcb->prim_p_key_idx);
-
- qp_attr->port_num =
- EHCA_BMASK_GET(MQPCB_PRIM_PHYS_PORT, qpcb->prim_phys_port);
-
+ qp_attr->pkey_index = qpcb->prim_p_key_idx;
+ qp_attr->port_num = qpcb->prim_phys_port;
qp_attr->timeout = qpcb->timeout;
qp_attr->retry_cnt = qpcb->retry_count;
qp_attr->rnr_retry = qpcb->rnr_retry_count;
- qp_attr->alt_pkey_index =
- EHCA_BMASK_GET(MQPCB_PRIM_P_KEY_IDX, qpcb->alt_p_key_idx);
-
+ qp_attr->alt_pkey_index = qpcb->alt_p_key_idx;
qp_attr->alt_port_num = qpcb->alt_phys_port;
qp_attr->alt_timeout = qpcb->timeout_al;
@@ -2061,8 +2055,7 @@ int ehca_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr,
update_mask |=
EHCA_BMASK_SET(MQPCB_MASK_CURR_SRQ_LIMIT, 1)
| EHCA_BMASK_SET(MQPCB_MASK_QP_AFF_ASYN_EV_LOG_REG, 1);
- mqpcb->curr_srq_limit =
- EHCA_BMASK_SET(MQPCB_CURR_SRQ_LIMIT, attr->srq_limit);
+ mqpcb->curr_srq_limit = attr->srq_limit;
mqpcb->qp_aff_asyn_ev_log_reg =
EHCA_BMASK_SET(QPX_AAELOG_RESET_SRQ_LIMIT, 1);
}
@@ -2125,8 +2118,7 @@ int ehca_query_srq(struct ib_srq *srq, struct ib_srq_attr *srq_attr)
srq_attr->max_wr = qpcb->max_nr_outst_recv_wr - 1;
srq_attr->max_sge = 3;
- srq_attr->srq_limit = EHCA_BMASK_GET(
- MQPCB_CURR_SRQ_LIMIT, qpcb->curr_srq_limit);
+ srq_attr->srq_limit = qpcb->curr_srq_limit;
if (ehca_debug_level >= 2)
ehca_dmp(qpcb, 4*70, "qp_num=%x", my_qp->real_qp_num);
--
1.5.5
looks fine, applied for 2.6.31
Was reporting CQE flags in the wrong bit positions, causing consumers to
miss incoming immediate data.
Signed-off-by: Joachim Fenkes <[email protected]>
---
Please review and queue for 2.6.32 if you think it's okay. Thanks!
Joachim
drivers/infiniband/hw/ehca/ehca_reqs.c | 6 +++++-
1 files changed, 5 insertions(+), 1 deletions(-)
diff --git a/drivers/infiniband/hw/ehca/ehca_reqs.c b/drivers/infiniband/hw/ehca/ehca_reqs.c
index 5a3d96f..8fd88cd 100644
--- a/drivers/infiniband/hw/ehca/ehca_reqs.c
+++ b/drivers/infiniband/hw/ehca/ehca_reqs.c
@@ -786,7 +786,11 @@ repoll:
wc->slid = cqe->rlid;
wc->dlid_path_bits = cqe->dlid;
wc->src_qp = cqe->remote_qp_number;
- wc->wc_flags = cqe->w_completion_flags;
+ /*
+ * HW has "Immed data present" and "GRH present" in bits 6 and 5.
+ * SW defines those in bits 1 and 0, so we can just shift and mask.
+ */
+ wc->wc_flags = (cqe->w_completion_flags >> 5) & 3;
wc->ex.imm_data = cpu_to_be32(cqe->immediate_data);
wc->sl = cqe->service_level;
--
1.6.0.4