Re: [PATCH for-4.1 06/11] RDMA/ocrdma: Prevent allocation of DPP PDs if FW doesnt support it

Re: [PATCH for-4.1 06/11] RDMA/ocrdma: Prevent allocation of DPP PDs if FW doesnt support it

[Doug Ledford]

[-- Attachment #1: Type: text/plain, Size: 1228 bytes --]

On Sat, 2015-05-16 at 13:21 +0530, Selvin Xavier wrote:

> @@ -1468,10 +1471,8 @@ static int ocrdma_mbx_alloc_pd_range(struct ocrdma_dev *dev)
>  
>  	cmd->pd_count = dev->attr.max_pd - dev->attr.max_dpp_pds;
>  	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
> -	if (status)
> -		goto mbx_err;
>  	rsp = (struct ocrdma_alloc_pd_range_rsp *)cmd;
> -	if (rsp->pd_count) {
> +	if (!status && rsp->pd_count) {
>  		dev->pd_mgr->pd_norm_start = rsp->dpp_page_pdid &
>  					OCRDMA_ALLOC_PD_RNG_RSP_START_PDID_MASK;
>  		dev->pd_mgr->max_normal_pd = rsp->pd_count;
> @@ -1486,7 +1487,6 @@ static int ocrdma_mbx_alloc_pd_range(struct ocrdma_dev *dev)
>  	} else {
>  		return -ENOMEM;
>  	}
> -mbx_err:
>  	kfree(cmd);
>  	return status;
>  }

I didn't go into the file to make sure, but this looks like a memory
leak (in fact, it looks like a leak that always existed, specifically if
rsp->pd_count == 0 then the else causes you to leave before the
kfree(cmd) and therefore you leak).  It's now been made worse because it
would now happen both when status != 0 and when pd_count == 0.


-- 
Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
              GPG KeyID: 0E572FDD


[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: [PATCH for-4.1 10/11] RDMA/ocrdma: Prevent returning failures in the resource destroy path

[Doug Ledford]

[-- Attachment #1: Type: text/plain, Size: 3613 bytes --]

On Sat, 2015-05-16 at 13:21 +0530, Selvin Xavier wrote:
> Avoid returning FW failures in the resource destroy verbs.

This patch looks extremely unsafe to someone that doesn't have the
benefit of being able to look at your firmware code.  What firmware
errors are you ignoring?  What state is the card in whenever these
errors are returned?  Is there even a remote possibility that the
firmware being in this state and returning these errors will either A)
cause erroneous attempts to write to user memory or B) interfere with
future attempts to create PDs or other contexts on the card?

> Signed-off-by: Padmanabh Ratnakar <padmanabh.ratnakar-1wcpHE2jlwO1Z/+hSey0Gg@public.gmane.org>
> Signed-off-by: Devesh Sharma <devesh.sharma-1wcpHE2jlwO1Z/+hSey0Gg@public.gmane.org>
> Signed-off-by: Selvin Xavier <selvin.xavier-1wcpHE2jlwO1Z/+hSey0Gg@public.gmane.org>
> ---
>  drivers/infiniband/hw/ocrdma/ocrdma_verbs.c | 17 +++++++----------
>  1 file changed, 7 insertions(+), 10 deletions(-)
> 
> diff --git a/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c b/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c
> index 48ec56f..b1f9d7d 100644
> --- a/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c
> +++ b/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c
> @@ -531,13 +531,12 @@ map_err:
>  
>  int ocrdma_dealloc_ucontext(struct ib_ucontext *ibctx)
>  {
> -	int status = 0;
>  	struct ocrdma_mm *mm, *tmp;
>  	struct ocrdma_ucontext *uctx = get_ocrdma_ucontext(ibctx);
>  	struct ocrdma_dev *dev = get_ocrdma_dev(ibctx->device);
>  	struct pci_dev *pdev = dev->nic_info.pdev;
>  
> -	status = ocrdma_dealloc_ucontext_pd(uctx);
> +	(void)ocrdma_dealloc_ucontext_pd(uctx);
>  
>  	ocrdma_del_mmap(uctx, uctx->ah_tbl.pa, uctx->ah_tbl.len);
>  	dma_free_coherent(&pdev->dev, uctx->ah_tbl.len, uctx->ah_tbl.va,
> @@ -548,7 +547,7 @@ int ocrdma_dealloc_ucontext(struct ib_ucontext *ibctx)
>  		kfree(mm);
>  	}
>  	kfree(uctx);
> -	return status;
> +	return 0;
>  }
>  
>  int ocrdma_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
> @@ -690,7 +689,6 @@ int ocrdma_dealloc_pd(struct ib_pd *ibpd)
>  	struct ocrdma_pd *pd = get_ocrdma_pd(ibpd);
>  	struct ocrdma_dev *dev = get_ocrdma_dev(ibpd->device);
>  	struct ocrdma_ucontext *uctx = NULL;
> -	int status = 0;
>  	u64 usr_db;
>  
>  	uctx = pd->uctx;
> @@ -704,11 +702,11 @@ int ocrdma_dealloc_pd(struct ib_pd *ibpd)
>  
>  		if (is_ucontext_pd(uctx, pd)) {
>  			ocrdma_release_ucontext_pd(uctx);
> -			return status;
> +			return 0;
>  		}
>  	}
> -	status = _ocrdma_dealloc_pd(dev, pd);
> -	return status;
> +	(void)_ocrdma_dealloc_pd(dev, pd);
> +	return 0;
>  }
>  
>  static int ocrdma_alloc_lkey(struct ocrdma_dev *dev, struct ocrdma_mr *mr,
> @@ -1905,13 +1903,12 @@ int ocrdma_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *srq_attr)
>  
>  int ocrdma_destroy_srq(struct ib_srq *ibsrq)
>  {
> -	int status;
>  	struct ocrdma_srq *srq;
>  	struct ocrdma_dev *dev = get_ocrdma_dev(ibsrq->device);
>  
>  	srq = get_ocrdma_srq(ibsrq);
>  
> -	status = ocrdma_mbx_destroy_srq(dev, srq);
> +	(void)ocrdma_mbx_destroy_srq(dev, srq);
>  
>  	if (srq->pd->uctx)
>  		ocrdma_del_mmap(srq->pd->uctx, (u64) srq->rq.pa,
> @@ -1920,7 +1917,7 @@ int ocrdma_destroy_srq(struct ib_srq *ibsrq)
>  	kfree(srq->idx_bit_fields);
>  	kfree(srq->rqe_wr_id_tbl);
>  	kfree(srq);
> -	return status;
> +	return 0;
>  }
>  
>  /* unprivileged verbs and their support functions. */


-- 
Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
              GPG KeyID: 0E572FDD


[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: [PATCH for-4.1 09/11] RDMA/ocrdma: Add CQ usage statistics counters

[Doug Ledford]

[-- Attachment #1: Type: text/plain, Size: 4702 bytes --]

On Sat, 2015-05-16 at 13:21 +0530, Selvin Xavier wrote:
> From: Meghana Cheripady <meghana.cheripady-1wcpHE2jlwO1Z/+hSey0Gg@public.gmane.org>
> 
> Add counters for tracking the CQ usage by applications.

This isn't a bug fix, it's an enhancement.  We are already in the
4.1-rc3+ phase, enhancements aren't really appropriate (even if they
seem minor).  Please drop this one from the patchset.

> 
> Signed-off-by: Meghana Cheripady <meghana.cheripady-1wcpHE2jlwO1Z/+hSey0Gg@public.gmane.org>
> Signed-off-by: Selvin Xavier <selvin.xavier-1wcpHE2jlwO1Z/+hSey0Gg@public.gmane.org>
> ---
>  drivers/infiniband/hw/ocrdma/ocrdma.h       |  2 ++
>  drivers/infiniband/hw/ocrdma/ocrdma_hw.c    |  8 ++++++++
>  drivers/infiniband/hw/ocrdma/ocrdma_stats.c | 15 +++++++++++++++
>  drivers/infiniband/hw/ocrdma/ocrdma_verbs.c |  2 ++
>  4 files changed, 27 insertions(+)
> 
> diff --git a/drivers/infiniband/hw/ocrdma/ocrdma.h b/drivers/infiniband/hw/ocrdma/ocrdma.h
> index ee9e335..8782e1f 100644
> --- a/drivers/infiniband/hw/ocrdma/ocrdma.h
> +++ b/drivers/infiniband/hw/ocrdma/ocrdma.h
> @@ -290,6 +290,8 @@ struct ocrdma_dev {
>  	struct dentry *dir;
>  	atomic_t async_err_stats[OCRDMA_MAX_ASYNC_ERRORS];
>  	atomic_t cqe_err_stats[OCRDMA_MAX_CQE_ERR];
> +	atomic_t act_cq_stats;
> +	atomic_t th_cq_stats;
>  	struct ocrdma_pd_resource_mgr *pd_mgr;
>  };
>  
> diff --git a/drivers/infiniband/hw/ocrdma/ocrdma_hw.c b/drivers/infiniband/hw/ocrdma/ocrdma_hw.c
> index 319675c..596d8c9 100644
> --- a/drivers/infiniband/hw/ocrdma/ocrdma_hw.c
> +++ b/drivers/infiniband/hw/ocrdma/ocrdma_hw.c
> @@ -1720,6 +1720,7 @@ int ocrdma_mbx_create_cq(struct ocrdma_dev *dev, struct ocrdma_cq *cq,
>  	struct ocrdma_create_cq *cmd;
>  	struct ocrdma_create_cq_rsp *rsp;
>  	u32 hw_pages, cqe_size, page_size, cqe_count;
> +	int act_cq_cnt, th_cq_cnt;
>  
>  	if (entries > dev->attr.max_cqe) {
>  		pr_err("%s(%d) max_cqe=0x%x, requester_cqe=0x%x\n",
> @@ -1808,6 +1809,13 @@ int ocrdma_mbx_create_cq(struct ocrdma_dev *dev, struct ocrdma_cq *cq,
>  
>  	rsp = (struct ocrdma_create_cq_rsp *)cmd;
>  	cq->id = (u16) (rsp->rsp.cq_id & OCRDMA_CREATE_CQ_RSP_CQ_ID_MASK);
> +
> +	atomic_inc(&dev->act_cq_stats);
> +	act_cq_cnt = atomic_read(&dev->act_cq_stats);
> +	th_cq_cnt = atomic_read(&dev->th_cq_stats);
> +	if (th_cq_cnt < act_cq_cnt)
> +		atomic_set(&dev->th_cq_stats, act_cq_cnt);
> +
>  	kfree(cmd);
>  	return 0;
>  mbx_err:
> diff --git a/drivers/infiniband/hw/ocrdma/ocrdma_stats.c b/drivers/infiniband/hw/ocrdma/ocrdma_stats.c
> index 48d7ef5..47fef14 100644
> --- a/drivers/infiniband/hw/ocrdma/ocrdma_stats.c
> +++ b/drivers/infiniband/hw/ocrdma/ocrdma_stats.c
> @@ -146,6 +146,8 @@ static char *ocrdma_resource_stats(struct ocrdma_dev *dev)
>  				(u64)rsrc_stats->phy_mr);
>  	pcur += ocrdma_add_stat(stats, pcur, "active_mw",
>  				(u64)rsrc_stats->mw);
> +	pcur += ocrdma_add_stat(stats, pcur, "active_cq_usage_stats",
> +				(u64)atomic_read(&dev->act_cq_stats));
>  
>  	/* Print the threshold stats */
>  	rsrc_stats = &rdma_stats->th_rsrc_stats;
> @@ -200,6 +202,8 @@ static char *ocrdma_resource_stats(struct ocrdma_dev *dev)
>  				(u64)rsrc_stats->phy_mr);
>  	pcur += ocrdma_add_stat(stats, pcur, "threshold_mw",
>  				(u64)rsrc_stats->mw);
> +	pcur += ocrdma_add_stat(stats, pcur, "threshold_cq_usage_stats",
> +				(u64)atomic_read(&dev->th_cq_stats));
>  	return stats;
>  }
>  
> @@ -745,6 +749,16 @@ static const struct file_operations ocrdma_dbg_ops = {
>  	.write = ocrdma_dbgfs_ops_write,
>  };
>  
> +static void ocrdma_init_cq_stats(struct ocrdma_dev *dev)
> +{
> +	int i, cq_cnt = 0;
> +
> +	for (i = 0; i < dev->eq_cnt; i++)
> +		cq_cnt += dev->eq_tbl[i].cq_cnt;
> +	atomic_set(&dev->act_cq_stats, cq_cnt);
> +	atomic_set(&dev->th_cq_stats, cq_cnt);
> +}
> +
>  void ocrdma_add_port_stats(struct ocrdma_dev *dev)
>  {
>  	if (!ocrdma_dbgfs_dir)
> @@ -828,6 +842,7 @@ void ocrdma_add_port_stats(struct ocrdma_dev *dev)
>  		goto err;
>  
>  	mutex_init(&dev->stats_lock);
> +	ocrdma_init_cq_stats(dev);
>  
>  	return;
>  err:
> diff --git a/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c b/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c
> index 9dcb660..48ec56f 100644
> --- a/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c
> +++ b/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c
> @@ -1119,6 +1119,8 @@ int ocrdma_destroy_cq(struct ib_cq *ibcq)
>  				dev->nic_info.db_page_size);
>  	}
>  
> +	atomic_dec(&dev->act_cq_stats);
> +
>  	kfree(cq);
>  	return 0;
>  }


-- 
Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
              GPG KeyID: 0E572FDD


[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

RE: [PATCH for-4.1 06/11] RDMA/ocrdma: Prevent allocation of DPP PDs if FW doesnt support it

[Selvin Xavier]

Thanks Doug for the review.  We will fix this memory leak in V2.

Thanks,
Selvin Xavier

> -----Original Message-----
> From: Doug Ledford [mailto:dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org]
> Sent: Friday, May 15, 2015 9:13 PM
> To: Selvin Xavier
> Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
> Subject: Re: [PATCH for-4.1 06/11] RDMA/ocrdma: Prevent allocation of DPP
> PDs if FW doesnt support it
>
> On Sat, 2015-05-16 at 13:21 +0530, Selvin Xavier wrote:
>
> > @@ -1468,10 +1471,8 @@ static int ocrdma_mbx_alloc_pd_range(struct
> > ocrdma_dev *dev)
> >
> >  	cmd->pd_count = dev->attr.max_pd - dev->attr.max_dpp_pds;
> >  	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
> > -	if (status)
> > -		goto mbx_err;
> >  	rsp = (struct ocrdma_alloc_pd_range_rsp *)cmd;
> > -	if (rsp->pd_count) {
> > +	if (!status && rsp->pd_count) {
> >  		dev->pd_mgr->pd_norm_start = rsp->dpp_page_pdid &
> >
> 	OCRDMA_ALLOC_PD_RNG_RSP_START_PDID_MASK;
> >  		dev->pd_mgr->max_normal_pd = rsp->pd_count; @@ -
> 1486,7 +1487,6 @@
> > static int ocrdma_mbx_alloc_pd_range(struct ocrdma_dev *dev)
> >  	} else {
> >  		return -ENOMEM;
> >  	}
> > -mbx_err:
> >  	kfree(cmd);
> >  	return status;
> >  }
>
> I didn't go into the file to make sure, but this looks like a memory leak
> (in fact,
> it looks like a leak that always existed, specifically if
> rsp->pd_count == 0 then the else causes you to leave before the
> kfree(cmd) and therefore you leak).  It's now been made worse because it
> would now happen both when status != 0 and when pd_count == 0.
>
>
> --
> Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
>               GPG KeyID: 0E572FDD
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

RE: [PATCH for-4.1 09/11] RDMA/ocrdma: Add CQ usage statistics counters

[Selvin Xavier]

> -----Original Message-----
> From: Doug Ledford [mailto:dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org]
> Sent: Friday, May 15, 2015 9:28 PM
> To: Selvin Xavier
> Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
> Subject: Re: [PATCH for-4.1 09/11] RDMA/ocrdma: Add CQ usage statistics
> counters
>
> On Sat, 2015-05-16 at 13:21 +0530, Selvin Xavier wrote:
> > From: Meghana Cheripady <meghana.cheripady-1wcpHE2jlwO1Z/+hSey0Gg@public.gmane.org>
> >
> > Add counters for tracking the CQ usage by applications.
>
> This isn't a bug fix, it's an enhancement.  We are already in the 4.1-rc3+
> phase, enhancements aren't really appropriate (even if they seem minor).
> Please drop this one from the patchset.
>
Sure.. I will drop this patch from this patch set.
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

RE: [PATCH for-4.1 10/11] RDMA/ocrdma: Prevent returning failures in the resource destroy path

[Selvin Xavier]

> -----Original Message-----
> From: Doug Ledford [mailto:dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org]
> Sent: Friday, May 15, 2015 9:26 PM
> To: Selvin Xavier
> Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
> Subject: Re: [PATCH for-4.1 10/11] RDMA/ocrdma: Prevent returning failures
> in the resource destroy path
>
> On Sat, 2015-05-16 at 13:21 +0530, Selvin Xavier wrote:
> > Avoid returning FW failures in the resource destroy verbs.
>
> This patch looks extremely unsafe to someone that doesn't have the benefit
> of being able to look at your firmware code.  What firmware errors are you
> ignoring?  What state is the card in whenever these errors are returned?
> Is
> there even a remote possibility that the firmware being in this state and
> returning these errors will either A) cause erroneous attempts to write to
> user memory or B) interfere with future attempts to create PDs or other
> contexts on the card?
>


One of the main reasons for these destroy commands to fail is when the
adapter is in  an unrecoverable error, which is handled in this patch.  Case
A will not be applicable in case of UE since adapter is down.   But Case A
is applicable if any rogue application passes wrong parameters like PDID for
destroy and the failure is a genuine error from FW .   I will rework on the
patch to handle these conditions also.

Case B is handled in both conditions. If the adapter is in UE, successive
create requests also will fail and failure is returned to the caller. If
adapter is in proper state,  successive create requests return success with
new resource IDs.

Thanks,
Selvin Xavier
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH for-4.1 00/11] ocrdma bug fixes

[Selvin Xavier]

This patch series includes fixes for few issues found in 
testing and added few info messages for debugging.
Also, added some debug statistics counters and
updated the driver version.

Please apply.

Devesh Sharma (3):
  RDMA/ocrdma: Report EQ full fatal error
  RDMA/ocrdma: Fix QP state transition in destroy_qp
  RDMA/ocrdma: Use VID 0 if PFC is enabled and vlan is not configured

Meghana Cheripady (1):
  RDMA/ocrdma: Add CQ usage statistics counters

Mitesh Ahuja (3):
  RDMA/ocrdma: Fix the request length for RDMA_QUERY_QP mailbox command
    to FW.
  RDMA/ocrdma: Prevent allocation of DPP PDs if FW doesnt support it
  RDMA/ocrdma: Fix dmac resolution for link local address

Naga Irrinki (1):
  RDMA/ocrdma: Fail connection for MTU lesser than 512

Selvin Xavier (3):
  RDMA/ocrdma: Fix EQ destroy failure during driver unload
  RDMA/ocrdma: Prevent returning failures in the resource destroy path
  RDMA/ocrdma: Update ocrdma version number

 drivers/infiniband/hw/ocrdma/ocrdma.h       |  6 +-
 drivers/infiniband/hw/ocrdma/ocrdma_ah.c    | 12 +++-
 drivers/infiniband/hw/ocrdma/ocrdma_hw.c    | 86 +++++++++++++++++++----------
 drivers/infiniband/hw/ocrdma/ocrdma_sli.h   |  9 +++
 drivers/infiniband/hw/ocrdma/ocrdma_stats.c | 15 +++++
 drivers/infiniband/hw/ocrdma/ocrdma_verbs.c | 31 ++++++-----
 6 files changed, 111 insertions(+), 48 deletions(-)

-- 
2.2.0

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH for-4.1 01/11] RDMA/ocrdma: Fix EQ destroy failure during driver unload

[Selvin Xavier]

Changing the destroy sequence of mailbox queue and event queues.
FW expects mailbox queue to be destroyed before desroying the EQs.

Signed-off-by: Selvin Xavier <selvin.xavier-1wcpHE2jlwO1Z/+hSey0Gg@public.gmane.org>
Signed-off-by: Devesh Sharma <devesh.sharma-1wcpHE2jlwO1Z/+hSey0Gg@public.gmane.org>
---
 drivers/infiniband/hw/ocrdma/ocrdma_hw.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/drivers/infiniband/hw/ocrdma/ocrdma_hw.c b/drivers/infiniband/hw/ocrdma/ocrdma_hw.c
index 0c9e959..3a5ea5af 100644
--- a/drivers/infiniband/hw/ocrdma/ocrdma_hw.c
+++ b/drivers/infiniband/hw/ocrdma/ocrdma_hw.c
@@ -3147,9 +3147,9 @@ void ocrdma_cleanup_hw(struct ocrdma_dev *dev)
 	ocrdma_free_pd_pool(dev);
 	ocrdma_mbx_delete_ah_tbl(dev);
 
-	/* cleanup the eqs */
-	ocrdma_destroy_eqs(dev);
-
 	/* cleanup the control path */
 	ocrdma_destroy_mq(dev);
+
+	/* cleanup the eqs */
+	ocrdma_destroy_eqs(dev);
 }
-- 
2.2.0

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH for-4.1 02/11] RDMA/ocrdma: Report EQ full fatal error

[Selvin Xavier]

From: Devesh Sharma <devesh.sharma-1wcpHE2jlwO1Z/+hSey0Gg@public.gmane.org>

Detect when Event Queue (EQ) becomes full and print a warning message.

Signed-off-by: Devesh Sharma <devesh.sharma-1wcpHE2jlwO1Z/+hSey0Gg@public.gmane.org>
Signed-off-by: Selvin Xavier <selvin.xavier-1wcpHE2jlwO1Z/+hSey0Gg@public.gmane.org>
---
 drivers/infiniband/hw/ocrdma/ocrdma_hw.c  | 6 ++++++
 drivers/infiniband/hw/ocrdma/ocrdma_sli.h | 7 +++++++
 2 files changed, 13 insertions(+)

diff --git a/drivers/infiniband/hw/ocrdma/ocrdma_hw.c b/drivers/infiniband/hw/ocrdma/ocrdma_hw.c
index 3a5ea5af..65759ac 100644
--- a/drivers/infiniband/hw/ocrdma/ocrdma_hw.c
+++ b/drivers/infiniband/hw/ocrdma/ocrdma_hw.c
@@ -933,12 +933,18 @@ static irqreturn_t ocrdma_irq_handler(int irq, void *handle)
 	struct ocrdma_eqe eqe;
 	struct ocrdma_eqe *ptr;
 	u16 cq_id;
+	u8 mcode;
 	int budget = eq->cq_cnt;
 
 	do {
 		ptr = ocrdma_get_eqe(eq);
 		eqe = *ptr;
 		ocrdma_le32_to_cpu(&eqe, sizeof(eqe));
+		mcode = (eqe.id_valid & OCRDMA_EQE_MAJOR_CODE_MASK)
+				>> OCRDMA_EQE_MAJOR_CODE_SHIFT;
+		if (mcode == OCRDMA_MAJOR_CODE_SENTINAL)
+			pr_err("EQ full on eqid = 0x%x, eqe = 0x%x\n",
+			       eq->q.id, eqe.id_valid);
 		if ((eqe.id_valid & OCRDMA_EQE_VALID_MASK) == 0)
 			break;
 
diff --git a/drivers/infiniband/hw/ocrdma/ocrdma_sli.h b/drivers/infiniband/hw/ocrdma/ocrdma_sli.h
index 243c87c..baf9b8a 100644
--- a/drivers/infiniband/hw/ocrdma/ocrdma_sli.h
+++ b/drivers/infiniband/hw/ocrdma/ocrdma_sli.h
@@ -1624,12 +1624,19 @@ struct ocrdma_delete_ah_tbl_rsp {
 enum {
 	OCRDMA_EQE_VALID_SHIFT		= 0,
 	OCRDMA_EQE_VALID_MASK		= BIT(0),
+	OCRDMA_EQE_MAJOR_CODE_MASK      = 0x0E,
+	OCRDMA_EQE_MAJOR_CODE_SHIFT     = 0x01,
 	OCRDMA_EQE_FOR_CQE_MASK		= 0xFFFE,
 	OCRDMA_EQE_RESOURCE_ID_SHIFT	= 16,
 	OCRDMA_EQE_RESOURCE_ID_MASK	= 0xFFFF <<
 				OCRDMA_EQE_RESOURCE_ID_SHIFT,
 };
 
+enum major_code {
+	OCRDMA_MAJOR_CODE_COMPLETION    = 0x00,
+	OCRDMA_MAJOR_CODE_SENTINAL      = 0x01
+};
+
 struct ocrdma_eqe {
 	u32 id_valid;
 };
-- 
2.2.0

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH for-4.1 03/11] RDMA/ocrdma: Fix QP state transition in destroy_qp

[Selvin Xavier]

From: Devesh Sharma <devesh.sharma-1wcpHE2jlwO1Z/+hSey0Gg@public.gmane.org>

Don't move QP to error state, if QP is in reset state during QP
destroy operation.

Signed-off-by: Devesh Sharma <devesh.sharma-1wcpHE2jlwO1Z/+hSey0Gg@public.gmane.org>
Signed-off-by: Selvin Xavier <selvin.xavier-1wcpHE2jlwO1Z/+hSey0Gg@public.gmane.org>
---
 drivers/infiniband/hw/ocrdma/ocrdma_verbs.c | 10 ++++++----
 1 file changed, 6 insertions(+), 4 deletions(-)

diff --git a/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c b/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c
index 8771755..06e8ab7 100644
--- a/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c
+++ b/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c
@@ -1721,18 +1721,20 @@ int ocrdma_destroy_qp(struct ib_qp *ibqp)
 	struct ocrdma_qp *qp;
 	struct ocrdma_dev *dev;
 	struct ib_qp_attr attrs;
-	int attr_mask = IB_QP_STATE;
+	int attr_mask;
 	unsigned long flags;
 
 	qp = get_ocrdma_qp(ibqp);
 	dev = get_ocrdma_dev(ibqp->device);
 
-	attrs.qp_state = IB_QPS_ERR;
 	pd = qp->pd;
 
 	/* change the QP state to ERROR */
-	_ocrdma_modify_qp(ibqp, &attrs, attr_mask);
-
+	if (qp->state != OCRDMA_QPS_RST) {
+		attrs.qp_state = IB_QPS_ERR;
+		attr_mask = IB_QP_STATE;
+		_ocrdma_modify_qp(ibqp, &attrs, attr_mask);
+	}
 	/* ensure that CQEs for newly created QP (whose id may be same with
 	 * one which just getting destroyed are same), dont get
 	 * discarded until the old CQEs are discarded.
-- 
2.2.0

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH for-4.1 04/11] RDMA/ocrdma: Use VID 0 if PFC is enabled and vlan is not configured

[Selvin Xavier]

From: Devesh Sharma <devesh.sharma-1wcpHE2jlwO1Z/+hSey0Gg@public.gmane.org>

If the adapter ports are in PFC mode and VLAN is not configured,
use vlan tag 0 for RoCE traffic. Also, log an advisory message
in system logs.

Signed-off-by: Padmanabh Ratnakar <padmanabh.ratnakar-1wcpHE2jlwO1Z/+hSey0Gg@public.gmane.org>
Signed-off-by: Devesh Sharma <devesh.sharma-1wcpHE2jlwO1Z/+hSey0Gg@public.gmane.org>
Signed-off-by: Selvin Xavier <selvin.xavier-1wcpHE2jlwO1Z/+hSey0Gg@public.gmane.org>
---
 drivers/infiniband/hw/ocrdma/ocrdma_ah.c |  8 +++++++-
 drivers/infiniband/hw/ocrdma/ocrdma_hw.c | 12 +++++++++++-
 2 files changed, 18 insertions(+), 2 deletions(-)

diff --git a/drivers/infiniband/hw/ocrdma/ocrdma_ah.c b/drivers/infiniband/hw/ocrdma/ocrdma_ah.c
index d812904..2a5993b 100644
--- a/drivers/infiniband/hw/ocrdma/ocrdma_ah.c
+++ b/drivers/infiniband/hw/ocrdma/ocrdma_ah.c
@@ -56,7 +56,13 @@ static inline int set_av_attr(struct ocrdma_dev *dev, struct ocrdma_ah *ah,
 	vlan_tag = attr->vlan_id;
 	if (!vlan_tag || (vlan_tag > 0xFFF))
 		vlan_tag = dev->pvid;
-	if (vlan_tag && (vlan_tag < 0x1000)) {
+	if (vlan_tag || dev->pfc_state) {
+		if (!vlan_tag) {
+			pr_err("ocrdma%d:Using VLAN with PFC is recommended\n",
+				dev->id);
+			pr_err("ocrdma%d:Using VLAN 0 for this connection\n",
+				dev->id);
+		}
 		eth.eth_type = cpu_to_be16(0x8100);
 		eth.roce_eth_type = cpu_to_be16(OCRDMA_ROCE_ETH_TYPE);
 		vlan_tag |= (dev->sl & 0x07) << OCRDMA_VID_PCP_SHIFT;
diff --git a/drivers/infiniband/hw/ocrdma/ocrdma_hw.c b/drivers/infiniband/hw/ocrdma/ocrdma_hw.c
index 65759ac..da688d7 100644
--- a/drivers/infiniband/hw/ocrdma/ocrdma_hw.c
+++ b/drivers/infiniband/hw/ocrdma/ocrdma_hw.c
@@ -2434,7 +2434,7 @@ static int ocrdma_set_av_params(struct ocrdma_qp *qp,
 	int status;
 	struct ib_ah_attr *ah_attr = &attrs->ah_attr;
 	union ib_gid sgid, zgid;
-	u32 vlan_id;
+	u32 vlan_id = 0xFFFF;
 	u8 mac_addr[6];
 	struct ocrdma_dev *dev = get_ocrdma_dev(qp->ibqp.device);
 
@@ -2474,12 +2474,22 @@ static int ocrdma_set_av_params(struct ocrdma_qp *qp,
 	cmd->params.vlan_dmac_b4_to_b5 = mac_addr[4] | (mac_addr[5] << 8);
 	if (attr_mask & IB_QP_VID) {
 		vlan_id = attrs->vlan_id;
+	} else if (dev->pfc_state) {
+		vlan_id = 0;
+		pr_err("ocrdma%d:Using VLAN with PFC is recommended\n",
+			dev->id);
+		pr_err("ocrdma%d:Using VLAN 0 for this connection\n",
+			dev->id);
+	}
+
+	if (vlan_id < 0x1000) {
 		cmd->params.vlan_dmac_b4_to_b5 |=
 		    vlan_id << OCRDMA_QP_PARAMS_VLAN_SHIFT;
 		cmd->flags |= OCRDMA_QP_PARA_VLAN_EN_VALID;
 		cmd->params.rnt_rc_sl_fl |=
 			(dev->sl & 0x07) << OCRDMA_QP_PARAMS_SL_SHIFT;
 	}
+
 	return 0;
 }
 
-- 
2.2.0

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH for-4.1 05/11] RDMA/ocrdma: Fix the request length for RDMA_QUERY_QP mailbox command to FW.

[Selvin Xavier]

From: Mitesh Ahuja <mitesh.ahuja-1wcpHE2jlwO1Z/+hSey0Gg@public.gmane.org>

Fix ocrdma_query_qp to pass correct mailbox request length to FW.

Signed-off-by: Mitesh Ahuja <mitesh.ahuja-1wcpHE2jlwO1Z/+hSey0Gg@public.gmane.org>
Signed-off-by: Selvin Xavier <selvin.xavier-1wcpHE2jlwO1Z/+hSey0Gg@public.gmane.org>
---
 drivers/infiniband/hw/ocrdma/ocrdma_hw.c  | 2 +-
 drivers/infiniband/hw/ocrdma/ocrdma_sli.h | 2 ++
 2 files changed, 3 insertions(+), 1 deletion(-)

diff --git a/drivers/infiniband/hw/ocrdma/ocrdma_hw.c b/drivers/infiniband/hw/ocrdma/ocrdma_hw.c
index da688d7..5636eb6 100644
--- a/drivers/infiniband/hw/ocrdma/ocrdma_hw.c
+++ b/drivers/infiniband/hw/ocrdma/ocrdma_hw.c
@@ -2412,7 +2412,7 @@ int ocrdma_mbx_query_qp(struct ocrdma_dev *dev, struct ocrdma_qp *qp,
 	struct ocrdma_query_qp *cmd;
 	struct ocrdma_query_qp_rsp *rsp;
 
-	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_QUERY_QP, sizeof(*cmd));
+	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_QUERY_QP, sizeof(*rsp));
 	if (!cmd)
 		return status;
 	cmd->qp_id = qp->id;
diff --git a/drivers/infiniband/hw/ocrdma/ocrdma_sli.h b/drivers/infiniband/hw/ocrdma/ocrdma_sli.h
index baf9b8a..02ad0ae 100644
--- a/drivers/infiniband/hw/ocrdma/ocrdma_sli.h
+++ b/drivers/infiniband/hw/ocrdma/ocrdma_sli.h
@@ -1176,6 +1176,8 @@ struct ocrdma_query_qp_rsp {
 	struct ocrdma_mqe_hdr hdr;
 	struct ocrdma_mbx_rsp rsp;
 	struct ocrdma_qp_params params;
+	u32 dpp_credits_cqid;
+	u32 rbq_id;
 };
 
 enum {
-- 
2.2.0

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH for-4.1 06/11] RDMA/ocrdma: Prevent allocation of DPP PDs if FW doesnt support it

[Selvin Xavier]

From: Mitesh Ahuja <mitesh.ahuja-1wcpHE2jlwO1Z/+hSey0Gg@public.gmane.org>

If DPP PDs are not supported by the FW, allocate only normal PDs.

Signed-off-by: Mitesh Ahuja <mitesh.ahuja-1wcpHE2jlwO1Z/+hSey0Gg@public.gmane.org>
Signed-off-by: Devesh Sharma <devesh.sharma-1wcpHE2jlwO1Z/+hSey0Gg@public.gmane.org>
Signed-off-by: Selvin Xavier <selvin.xavier-1wcpHE2jlwO1Z/+hSey0Gg@public.gmane.org>
---
 drivers/infiniband/hw/ocrdma/ocrdma_hw.c    | 48 ++++++++++++++---------------
 drivers/infiniband/hw/ocrdma/ocrdma_verbs.c |  2 +-
 2 files changed, 25 insertions(+), 25 deletions(-)

diff --git a/drivers/infiniband/hw/ocrdma/ocrdma_hw.c b/drivers/infiniband/hw/ocrdma/ocrdma_hw.c
index 5636eb6..4b77f34 100644
--- a/drivers/infiniband/hw/ocrdma/ocrdma_hw.c
+++ b/drivers/infiniband/hw/ocrdma/ocrdma_hw.c
@@ -1440,27 +1440,30 @@ static int ocrdma_mbx_alloc_pd_range(struct ocrdma_dev *dev)
 	struct ocrdma_alloc_pd_range_rsp *rsp;
 
 	/* Pre allocate the DPP PDs */
-	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_ALLOC_PD_RANGE, sizeof(*cmd));
-	if (!cmd)
-		return -ENOMEM;
-	cmd->pd_count = dev->attr.max_dpp_pds;
-	cmd->enable_dpp_rsvd |= OCRDMA_ALLOC_PD_ENABLE_DPP;
-	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
-	if (status)
-		goto mbx_err;
-	rsp = (struct ocrdma_alloc_pd_range_rsp *)cmd;
-
-	if ((rsp->dpp_page_pdid & OCRDMA_ALLOC_PD_RSP_DPP) && rsp->pd_count) {
-		dev->pd_mgr->dpp_page_index = rsp->dpp_page_pdid >>
-				OCRDMA_ALLOC_PD_RSP_DPP_PAGE_SHIFT;
-		dev->pd_mgr->pd_dpp_start = rsp->dpp_page_pdid &
-				OCRDMA_ALLOC_PD_RNG_RSP_START_PDID_MASK;
-		dev->pd_mgr->max_dpp_pd = rsp->pd_count;
-		pd_bitmap_size = BITS_TO_LONGS(rsp->pd_count) * sizeof(long);
-		dev->pd_mgr->pd_dpp_bitmap = kzalloc(pd_bitmap_size,
-						     GFP_KERNEL);
+	if (dev->attr.max_dpp_pds) {
+		cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_ALLOC_PD_RANGE,
+					  sizeof(*cmd));
+		if (!cmd)
+			return -ENOMEM;
+		cmd->pd_count = dev->attr.max_dpp_pds;
+		cmd->enable_dpp_rsvd |= OCRDMA_ALLOC_PD_ENABLE_DPP;
+		status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
+		rsp = (struct ocrdma_alloc_pd_range_rsp *)cmd;
+
+		if (!status && (rsp->dpp_page_pdid & OCRDMA_ALLOC_PD_RSP_DPP) &&
+		    rsp->pd_count) {
+			dev->pd_mgr->dpp_page_index = rsp->dpp_page_pdid >>
+					OCRDMA_ALLOC_PD_RSP_DPP_PAGE_SHIFT;
+			dev->pd_mgr->pd_dpp_start = rsp->dpp_page_pdid &
+					OCRDMA_ALLOC_PD_RNG_RSP_START_PDID_MASK;
+			dev->pd_mgr->max_dpp_pd = rsp->pd_count;
+			pd_bitmap_size =
+				BITS_TO_LONGS(rsp->pd_count) * sizeof(long);
+			dev->pd_mgr->pd_dpp_bitmap = kzalloc(pd_bitmap_size,
+							     GFP_KERNEL);
+		}
+		kfree(cmd);
 	}
-	kfree(cmd);
 
 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_ALLOC_PD_RANGE, sizeof(*cmd));
 	if (!cmd)
@@ -1468,10 +1471,8 @@ static int ocrdma_mbx_alloc_pd_range(struct ocrdma_dev *dev)
 
 	cmd->pd_count = dev->attr.max_pd - dev->attr.max_dpp_pds;
 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
-	if (status)
-		goto mbx_err;
 	rsp = (struct ocrdma_alloc_pd_range_rsp *)cmd;
-	if (rsp->pd_count) {
+	if (!status && rsp->pd_count) {
 		dev->pd_mgr->pd_norm_start = rsp->dpp_page_pdid &
 					OCRDMA_ALLOC_PD_RNG_RSP_START_PDID_MASK;
 		dev->pd_mgr->max_normal_pd = rsp->pd_count;
@@ -1486,7 +1487,6 @@ static int ocrdma_mbx_alloc_pd_range(struct ocrdma_dev *dev)
 	} else {
 		return -ENOMEM;
 	}
-mbx_err:
 	kfree(cmd);
 	return status;
 }
diff --git a/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c b/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c
index 06e8ab7..9dcb660 100644
--- a/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c
+++ b/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c
@@ -365,7 +365,7 @@ static struct ocrdma_pd *_ocrdma_alloc_pd(struct ocrdma_dev *dev,
 	if (!pd)
 		return ERR_PTR(-ENOMEM);
 
-	if (udata && uctx) {
+	if (udata && uctx && dev->attr.max_dpp_pds) {
 		pd->dpp_enabled =
 			ocrdma_get_asic_type(dev) == OCRDMA_ASIC_GEN_SKH_R;
 		pd->num_dpp_qp =
-- 
2.2.0

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH for-4.1 07/11] RDMA/ocrdma: Fix dmac resolution for link local address

[Selvin Xavier]

From: Mitesh Ahuja <mitesh.ahuja-1wcpHE2jlwO1Z/+hSey0Gg@public.gmane.org>

rdma_addr_find_dmac_by_grh fails to resolve dmac for link local address.
Use rdma_get_ll_mac to resolve the link local address.

Signed-off-by: Padmanabh Ratnakar <padmanabh.ratnakar-1wcpHE2jlwO1Z/+hSey0Gg@public.gmane.org>
Signed-off-by: Mitesh Ahuja <mitesh.ahuja-1wcpHE2jlwO1Z/+hSey0Gg@public.gmane.org>
Signed-off-by: Devesh Sharma <devesh.sharma-1wcpHE2jlwO1Z/+hSey0Gg@public.gmane.org>
Signed-off-by: Selvin Xavier <selvin.xavier-1wcpHE2jlwO1Z/+hSey0Gg@public.gmane.org>
---
 drivers/infiniband/hw/ocrdma/ocrdma.h    | 2 ++
 drivers/infiniband/hw/ocrdma/ocrdma_ah.c | 4 +++-
 2 files changed, 5 insertions(+), 1 deletion(-)

diff --git a/drivers/infiniband/hw/ocrdma/ocrdma.h b/drivers/infiniband/hw/ocrdma/ocrdma.h
index c9780d9..ee9e335 100644
--- a/drivers/infiniband/hw/ocrdma/ocrdma.h
+++ b/drivers/infiniband/hw/ocrdma/ocrdma.h
@@ -515,6 +515,8 @@ static inline int ocrdma_resolve_dmac(struct ocrdma_dev *dev,
 	memcpy(&in6, ah_attr->grh.dgid.raw, sizeof(in6));
 	if (rdma_is_multicast_addr(&in6))
 		rdma_get_mcast_mac(&in6, mac_addr);
+	else if (rdma_link_local_addr(&in6))
+		rdma_get_ll_mac(&in6, mac_addr);
 	else
 		memcpy(mac_addr, ah_attr->dmac, ETH_ALEN);
 	return 0;
diff --git a/drivers/infiniband/hw/ocrdma/ocrdma_ah.c b/drivers/infiniband/hw/ocrdma/ocrdma_ah.c
index 2a5993b..f5a5ea836 100644
--- a/drivers/infiniband/hw/ocrdma/ocrdma_ah.c
+++ b/drivers/infiniband/hw/ocrdma/ocrdma_ah.c
@@ -127,7 +127,9 @@ struct ib_ah *ocrdma_create_ah(struct ib_pd *ibpd, struct ib_ah_attr *attr)
 		goto av_conf_err;
 	}
 
-	if (pd->uctx) {
+	if ((pd->uctx) &&
+	    (!rdma_is_multicast_addr((struct in6_addr *)attr->grh.dgid.raw)) &&
+	    (!rdma_link_local_addr((struct in6_addr *)attr->grh.dgid.raw))) {
 		status = rdma_addr_find_dmac_by_grh(&sgid, &attr->grh.dgid,
                                         attr->dmac, &attr->vlan_id);
 		if (status) {
-- 
2.2.0

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH for-4.1 08/11] RDMA/ocrdma: Fail connection for MTU lesser than 512

[Selvin Xavier]

From: Naga Irrinki <naga.irrinki-1wcpHE2jlwO1Z/+hSey0Gg@public.gmane.org>

HW currently restricts the IB MTU range between 512 and 4096.
Fail connection for MTUs lesser than 512.

Signed-off-by: Naga Irrinki <naga.irrinki-1wcpHE2jlwO1Z/+hSey0Gg@public.gmane.org>
Signed-off-by: Selvin Xavier <selvin.xavier-1wcpHE2jlwO1Z/+hSey0Gg@public.gmane.org>
---
 drivers/infiniband/hw/ocrdma/ocrdma_hw.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/drivers/infiniband/hw/ocrdma/ocrdma_hw.c b/drivers/infiniband/hw/ocrdma/ocrdma_hw.c
index 4b77f34..319675c 100644
--- a/drivers/infiniband/hw/ocrdma/ocrdma_hw.c
+++ b/drivers/infiniband/hw/ocrdma/ocrdma_hw.c
@@ -2535,8 +2535,10 @@ static int ocrdma_set_qp_params(struct ocrdma_qp *qp,
 		cmd->flags |= OCRDMA_QP_PARA_DST_QPN_VALID;
 	}
 	if (attr_mask & IB_QP_PATH_MTU) {
-		if (attrs->path_mtu < IB_MTU_256 ||
+		if (attrs->path_mtu < IB_MTU_512 ||
 		    attrs->path_mtu > IB_MTU_4096) {
+			pr_err("ocrdma%d: IB MTU %d is not supported\n",
+			       dev->id, ib_mtu_enum_to_int(attrs->path_mtu));
 			status = -EINVAL;
 			goto pmtu_err;
 		}
-- 
2.2.0

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH for-4.1 09/11] RDMA/ocrdma: Add CQ usage statistics counters

[Selvin Xavier]

From: Meghana Cheripady <meghana.cheripady-1wcpHE2jlwO1Z/+hSey0Gg@public.gmane.org>

Add counters for tracking the CQ usage by applications.

Signed-off-by: Meghana Cheripady <meghana.cheripady-1wcpHE2jlwO1Z/+hSey0Gg@public.gmane.org>
Signed-off-by: Selvin Xavier <selvin.xavier-1wcpHE2jlwO1Z/+hSey0Gg@public.gmane.org>
---
 drivers/infiniband/hw/ocrdma/ocrdma.h       |  2 ++
 drivers/infiniband/hw/ocrdma/ocrdma_hw.c    |  8 ++++++++
 drivers/infiniband/hw/ocrdma/ocrdma_stats.c | 15 +++++++++++++++
 drivers/infiniband/hw/ocrdma/ocrdma_verbs.c |  2 ++
 4 files changed, 27 insertions(+)

diff --git a/drivers/infiniband/hw/ocrdma/ocrdma.h b/drivers/infiniband/hw/ocrdma/ocrdma.h
index ee9e335..8782e1f 100644
--- a/drivers/infiniband/hw/ocrdma/ocrdma.h
+++ b/drivers/infiniband/hw/ocrdma/ocrdma.h
@@ -290,6 +290,8 @@ struct ocrdma_dev {
 	struct dentry *dir;
 	atomic_t async_err_stats[OCRDMA_MAX_ASYNC_ERRORS];
 	atomic_t cqe_err_stats[OCRDMA_MAX_CQE_ERR];
+	atomic_t act_cq_stats;
+	atomic_t th_cq_stats;
 	struct ocrdma_pd_resource_mgr *pd_mgr;
 };
 
diff --git a/drivers/infiniband/hw/ocrdma/ocrdma_hw.c b/drivers/infiniband/hw/ocrdma/ocrdma_hw.c
index 319675c..596d8c9 100644
--- a/drivers/infiniband/hw/ocrdma/ocrdma_hw.c
+++ b/drivers/infiniband/hw/ocrdma/ocrdma_hw.c
@@ -1720,6 +1720,7 @@ int ocrdma_mbx_create_cq(struct ocrdma_dev *dev, struct ocrdma_cq *cq,
 	struct ocrdma_create_cq *cmd;
 	struct ocrdma_create_cq_rsp *rsp;
 	u32 hw_pages, cqe_size, page_size, cqe_count;
+	int act_cq_cnt, th_cq_cnt;
 
 	if (entries > dev->attr.max_cqe) {
 		pr_err("%s(%d) max_cqe=0x%x, requester_cqe=0x%x\n",
@@ -1808,6 +1809,13 @@ int ocrdma_mbx_create_cq(struct ocrdma_dev *dev, struct ocrdma_cq *cq,
 
 	rsp = (struct ocrdma_create_cq_rsp *)cmd;
 	cq->id = (u16) (rsp->rsp.cq_id & OCRDMA_CREATE_CQ_RSP_CQ_ID_MASK);
+
+	atomic_inc(&dev->act_cq_stats);
+	act_cq_cnt = atomic_read(&dev->act_cq_stats);
+	th_cq_cnt = atomic_read(&dev->th_cq_stats);
+	if (th_cq_cnt < act_cq_cnt)
+		atomic_set(&dev->th_cq_stats, act_cq_cnt);
+
 	kfree(cmd);
 	return 0;
 mbx_err:
diff --git a/drivers/infiniband/hw/ocrdma/ocrdma_stats.c b/drivers/infiniband/hw/ocrdma/ocrdma_stats.c
index 48d7ef5..47fef14 100644
--- a/drivers/infiniband/hw/ocrdma/ocrdma_stats.c
+++ b/drivers/infiniband/hw/ocrdma/ocrdma_stats.c
@@ -146,6 +146,8 @@ static char *ocrdma_resource_stats(struct ocrdma_dev *dev)
 				(u64)rsrc_stats->phy_mr);
 	pcur += ocrdma_add_stat(stats, pcur, "active_mw",
 				(u64)rsrc_stats->mw);
+	pcur += ocrdma_add_stat(stats, pcur, "active_cq_usage_stats",
+				(u64)atomic_read(&dev->act_cq_stats));
 
 	/* Print the threshold stats */
 	rsrc_stats = &rdma_stats->th_rsrc_stats;
@@ -200,6 +202,8 @@ static char *ocrdma_resource_stats(struct ocrdma_dev *dev)
 				(u64)rsrc_stats->phy_mr);
 	pcur += ocrdma_add_stat(stats, pcur, "threshold_mw",
 				(u64)rsrc_stats->mw);
+	pcur += ocrdma_add_stat(stats, pcur, "threshold_cq_usage_stats",
+				(u64)atomic_read(&dev->th_cq_stats));
 	return stats;
 }
 
@@ -745,6 +749,16 @@ static const struct file_operations ocrdma_dbg_ops = {
 	.write = ocrdma_dbgfs_ops_write,
 };
 
+static void ocrdma_init_cq_stats(struct ocrdma_dev *dev)
+{
+	int i, cq_cnt = 0;
+
+	for (i = 0; i < dev->eq_cnt; i++)
+		cq_cnt += dev->eq_tbl[i].cq_cnt;
+	atomic_set(&dev->act_cq_stats, cq_cnt);
+	atomic_set(&dev->th_cq_stats, cq_cnt);
+}
+
 void ocrdma_add_port_stats(struct ocrdma_dev *dev)
 {
 	if (!ocrdma_dbgfs_dir)
@@ -828,6 +842,7 @@ void ocrdma_add_port_stats(struct ocrdma_dev *dev)
 		goto err;
 
 	mutex_init(&dev->stats_lock);
+	ocrdma_init_cq_stats(dev);
 
 	return;
 err:
diff --git a/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c b/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c
index 9dcb660..48ec56f 100644
--- a/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c
+++ b/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c
@@ -1119,6 +1119,8 @@ int ocrdma_destroy_cq(struct ib_cq *ibcq)
 				dev->nic_info.db_page_size);
 	}
 
+	atomic_dec(&dev->act_cq_stats);
+
 	kfree(cq);
 	return 0;
 }
-- 
2.2.0

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH for-4.1 10/11] RDMA/ocrdma: Prevent returning failures in the resource destroy path

[Selvin Xavier]

Avoid returning FW failures in the resource destroy verbs.

Signed-off-by: Padmanabh Ratnakar <padmanabh.ratnakar-1wcpHE2jlwO1Z/+hSey0Gg@public.gmane.org>
Signed-off-by: Devesh Sharma <devesh.sharma-1wcpHE2jlwO1Z/+hSey0Gg@public.gmane.org>
Signed-off-by: Selvin Xavier <selvin.xavier-1wcpHE2jlwO1Z/+hSey0Gg@public.gmane.org>
---
 drivers/infiniband/hw/ocrdma/ocrdma_verbs.c | 17 +++++++----------
 1 file changed, 7 insertions(+), 10 deletions(-)

diff --git a/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c b/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c
index 48ec56f..b1f9d7d 100644
--- a/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c
+++ b/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c
@@ -531,13 +531,12 @@ map_err:
 
 int ocrdma_dealloc_ucontext(struct ib_ucontext *ibctx)
 {
-	int status = 0;
 	struct ocrdma_mm *mm, *tmp;
 	struct ocrdma_ucontext *uctx = get_ocrdma_ucontext(ibctx);
 	struct ocrdma_dev *dev = get_ocrdma_dev(ibctx->device);
 	struct pci_dev *pdev = dev->nic_info.pdev;
 
-	status = ocrdma_dealloc_ucontext_pd(uctx);
+	(void)ocrdma_dealloc_ucontext_pd(uctx);
 
 	ocrdma_del_mmap(uctx, uctx->ah_tbl.pa, uctx->ah_tbl.len);
 	dma_free_coherent(&pdev->dev, uctx->ah_tbl.len, uctx->ah_tbl.va,
@@ -548,7 +547,7 @@ int ocrdma_dealloc_ucontext(struct ib_ucontext *ibctx)
 		kfree(mm);
 	}
 	kfree(uctx);
-	return status;
+	return 0;
 }
 
 int ocrdma_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
@@ -690,7 +689,6 @@ int ocrdma_dealloc_pd(struct ib_pd *ibpd)
 	struct ocrdma_pd *pd = get_ocrdma_pd(ibpd);
 	struct ocrdma_dev *dev = get_ocrdma_dev(ibpd->device);
 	struct ocrdma_ucontext *uctx = NULL;
-	int status = 0;
 	u64 usr_db;
 
 	uctx = pd->uctx;
@@ -704,11 +702,11 @@ int ocrdma_dealloc_pd(struct ib_pd *ibpd)
 
 		if (is_ucontext_pd(uctx, pd)) {
 			ocrdma_release_ucontext_pd(uctx);
-			return status;
+			return 0;
 		}
 	}
-	status = _ocrdma_dealloc_pd(dev, pd);
-	return status;
+	(void)_ocrdma_dealloc_pd(dev, pd);
+	return 0;
 }
 
 static int ocrdma_alloc_lkey(struct ocrdma_dev *dev, struct ocrdma_mr *mr,
@@ -1905,13 +1903,12 @@ int ocrdma_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *srq_attr)
 
 int ocrdma_destroy_srq(struct ib_srq *ibsrq)
 {
-	int status;
 	struct ocrdma_srq *srq;
 	struct ocrdma_dev *dev = get_ocrdma_dev(ibsrq->device);
 
 	srq = get_ocrdma_srq(ibsrq);
 
-	status = ocrdma_mbx_destroy_srq(dev, srq);
+	(void)ocrdma_mbx_destroy_srq(dev, srq);
 
 	if (srq->pd->uctx)
 		ocrdma_del_mmap(srq->pd->uctx, (u64) srq->rq.pa,
@@ -1920,7 +1917,7 @@ int ocrdma_destroy_srq(struct ib_srq *ibsrq)
 	kfree(srq->idx_bit_fields);
 	kfree(srq->rqe_wr_id_tbl);
 	kfree(srq);
-	return status;
+	return 0;
 }
 
 /* unprivileged verbs and their support functions. */
-- 
2.2.0

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH for-4.1 11/11] RDMA/ocrdma: Update ocrdma version number

[Selvin Xavier]

Updating the driver version to 10.6.0.0

Signed-off-by: Selvin Xavier <selvin.xavier-1wcpHE2jlwO1Z/+hSey0Gg@public.gmane.org>
---
 drivers/infiniband/hw/ocrdma/ocrdma.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/infiniband/hw/ocrdma/ocrdma.h b/drivers/infiniband/hw/ocrdma/ocrdma.h
index 8782e1f..5ea13f2 100644
--- a/drivers/infiniband/hw/ocrdma/ocrdma.h
+++ b/drivers/infiniband/hw/ocrdma/ocrdma.h
@@ -40,7 +40,7 @@
 #include <be_roce.h>
 #include "ocrdma_sli.h"
 
-#define OCRDMA_ROCE_DRV_VERSION "10.4.205.0u"
+#define OCRDMA_ROCE_DRV_VERSION "10.6.0.0"
 
 #define OCRDMA_ROCE_DRV_DESC "Emulex OneConnect RoCE Driver"
 #define OCRDMA_NODE_DESC "Emulex OneConnect RoCE HCA"
-- 
2.2.0

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

RE: [PATCH for-4.1 10/11] RDMA/ocrdma: Prevent returning failures in the resource destroy path

[Selvin Xavier]

> -----Original Message-----
> From: Selvin Xavier [mailto:selvin.xavier-1wcpHE2jlwO1Z/+hSey0Gg@public.gmane.org]
> Sent: Friday, May 15, 2015 11:47 PM
> To: 'Doug Ledford'
> Cc: 'linux-rdma-u79uwXL29TY76Z2rM5mHXA@public.gmane.org'
> Subject: RE: [PATCH for-4.1 10/11] RDMA/ocrdma: Prevent returning failures
> in the resource destroy path
>
>
> > -----Original Message-----
> > From: Doug Ledford [mailto:dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org]
> > Sent: Friday, May 15, 2015 9:26 PM
> > To: Selvin Xavier
> > Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
> > Subject: Re: [PATCH for-4.1 10/11] RDMA/ocrdma: Prevent returning
> > failures in the resource destroy path
> >
> > On Sat, 2015-05-16 at 13:21 +0530, Selvin Xavier wrote:
> > > Avoid returning FW failures in the resource destroy verbs.
> >
> > This patch looks extremely unsafe to someone that doesn't have the
> > benefit of being able to look at your firmware code.  What firmware
> > errors are you ignoring?  What state is the card in whenever these
> > errors are returned?  Is there even a remote possibility that the
> > firmware being in this state and returning these errors will either A)
> > cause erroneous attempts to write to user memory or B) interfere with
> > future attempts to create PDs or other contexts on the card?
> >
>
>
> One of the main reasons for these destroy commands to fail is when the
> adapter is in  an unrecoverable error, which is handled in this patch.
> Case A
> will not be applicable in case of UE since adapter is down.   But Case A
> is
> applicable if any rogue application passes wrong parameters like PDID for
> destroy and the failure is a genuine error from FW .   I will rework on
> the
> patch to handle these conditions also.
>

> Case B is handled in both conditions. If the adapter is in UE, successive
> create
> requests also will fail and failure is returned to the caller. If adapter
> is in
> proper state,  successive create requests return success with new resource
> IDs.
>

Doug,

The driver is returning success for other destroy routines also and free the
corresponding driver resources.  As part of the changes suggested by you, we
will change all the destroy routines to return error status for all the FW
command failures and avoid freeing the driver resources  in case of failure.
Since this change need more testing, I will drop this  patch from the series
for 4.1 RC.  I will cut a patch for 4.2 after our testing.

Thanks,
Selvin
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH for-4.1 10/11] RDMA/ocrdma: Prevent returning failures in the resource destroy path

[Jason Gunthorpe]

On Mon, May 18, 2015 at 05:16:14PM +0530, Selvin Xavier wrote:

> The driver is returning success for other destroy routines also and
> free the corresponding driver resources.  As part of the changes
> suggested by you, we will change all the destroy routines to return
> error status for all the FW command failures and avoid freeing the
> driver resources in case of failure.  Since this change need more
> testing, I will drop this patch from the series for 4.1 RC.  I will
> cut a patch for 4.2 after our testing.

Didn't we discuss this already? Destroy routines cannot fail.

If the card's firmware fails during destroy then that is an
unrecoverable error and your driver and firmware are no longer
synchronized. Panic the kernel or do some kind of (synchronous)
card-reset recovery.

Don't expect the caller to handle this, and don't leak IB core memory
when it happens.

Jason
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

RE: [PATCH for-4.1 10/11] RDMA/ocrdma: Prevent returning failures in the resource destroy path

[Selvin Xavier]

Hi Jason
 My reply is delayed since  I was on vacation.

> -----Original Message-----
> From: Jason Gunthorpe [mailto:jgunthorpe-ePGOBjL8dl3ta4EC/59zMFaTQe2KTcn/@public.gmane.org]
> Sent: Tuesday, May 19, 2015 11:53 PM
> To: Selvin Xavier
> Cc: Doug Ledford; linux-rdma-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
> Subject: Re: [PATCH for-4.1 10/11] RDMA/ocrdma: Prevent returning
failures
> in the resource destroy path
>
> On Mon, May 18, 2015 at 05:16:14PM +0530, Selvin Xavier wrote:
>
> > The driver is returning success for other destroy routines also and
> > free the corresponding driver resources.  As part of the changes
> > suggested by you, we will change all the destroy routines to return
> > error status for all the FW command failures and avoid freeing the
> > driver resources in case of failure.  Since this change need more
> > testing, I will drop this patch from the series for 4.1 RC.  I will
> > cut a patch for 4.2 after our testing.
>
> Didn't we discuss this already? Destroy routines cannot fail.
>
> If the card's firmware fails during destroy then that is an
unrecoverable error
> and your driver and firmware are no longer synchronized. Panic the
kernel or
> do some kind of (synchronous) card-reset recovery.
>
> Don't expect the caller to handle this, and don't leak IB core memory
when it
> happens.
>
Got it .   Will do the same in all destroy APIs.

> Jason
Thanks,
Selvin
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html