[PATCH 0/24] various fixes for QIB driver

[PATCH 0/24] various fixes for QIB driver

[Mike Marciniszyn]

The following patches (ordered first to last) are for various bug fixes found
during internal testing.

IB/qib: remove IB latency turnoff
IB/qib: add receive header queue size module parameters
IB/qib: add support for the new QME7362 card
IB/qib: add thresholds to VendorPortCounters PMA operation
IB/qib: generate completion callback on errors
IB/qib: set port physical state even if other fields are invalid
IB/qib: UD send with immediate Rx completion has wrong size
IB/qib: Handle transitions from ACTIVE_DEFERRED to ACTIVE better
IB/qib: Multi Florida HCA Hosts panic on reboot
IB/qib: fix context allocation with multiple HCAs
IB/qib: clear WAIT_SEND flags when setting QP to error state
IB/qib: New SERDES init routine and improvements to SI quality
IB/qib: Reset packet list after freeing
IB/qib: Add a few new SERDES tunings
IB/qib: Avoid duplicate writes to the rcv head register
IB/qib: interrupt mitigation fix
IB/qib: change rcv queue/qpn selection
IB/qib: adding fix missing from earlier patch
IB/qib: change qpn increment
IB/qib: RDMA lkey/rkey validation is inefficient for large MRs
IB/qib: Issue pre-emptive NAKs on eager buffer overflow
IB/qib: Un-necessary delayed completions on RC connection
IB/qib: Add a delay to allow SERDES to initialize
IB/qib: Improve SERDES tunning on QMH boards
--
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 01/24] IB/qib: remove IB latency turnoff

[Mike Marciniszyn]

This is required for hardware testing.

Signed-off-by: Mike Marciniszyn <mike.marciniszyn-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/qib/qib_iba7322.c |    2 --
 1 files changed, 0 insertions(+), 2 deletions(-)

diff --git a/drivers/infiniband/hw/qib/qib_iba7322.c b/drivers/infiniband/hw/qib/qib_iba7322.c
index 584d443..9031cd8 100644
--- a/drivers/infiniband/hw/qib/qib_iba7322.c
+++ b/drivers/infiniband/hw/qib/qib_iba7322.c
@@ -7271,8 +7271,6 @@ static int serdes_7322_init(struct qib_pportdata *ppd)
 	ibsd_wr_allchans(ppd, 20, (4 << 13), BMASK(15, 13)); /* SDR */
 
 	data = qib_read_kreg_port(ppd, krp_serdesctrl);
-	/* Turn off IB latency mode */
-	data &= ~SYM_MASK(IBSerdesCtrl_0, IB_LAT_MODE);
 	qib_write_kreg_port(ppd, krp_serdesctrl, data |
 		SYM_MASK(IBSerdesCtrl_0, RXLOSEN));
 

--
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 02/24] IB/qib: add receive header queue size module parameters

[Mike Marciniszyn]

The receive header queue sizes need to modified to support
performance tuning.  Three module parameters are added to
support this.

Signed-off-by: Mike Marciniszyn <mike.marciniszyn-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/qib/qib_iba7322.c |   28 ++++++++++++++++++++++++----
 1 files changed, 24 insertions(+), 4 deletions(-)

diff --git a/drivers/infiniband/hw/qib/qib_iba7322.c b/drivers/infiniband/hw/qib/qib_iba7322.c
index 9031cd8..b35676f 100644
--- a/drivers/infiniband/hw/qib/qib_iba7322.c
+++ b/drivers/infiniband/hw/qib/qib_iba7322.c
@@ -111,6 +111,21 @@ static ushort qib_singleport;
 module_param_named(singleport, qib_singleport, ushort, S_IRUGO);
 MODULE_PARM_DESC(singleport, "Use only IB port 1; more per-port buffer space");
 
+/*
+ * Receive header queue sizes
+ */
+static unsigned qib_rcvhdrcnt;
+module_param_named(rcvhdrcnt, qib_rcvhdrcnt, uint, S_IRUGO);
+MODULE_PARM_DESC(rcvhdrcnt, "receive header count");
+
+static unsigned qib_rcvhdrsize;
+module_param_named(rcvhdrsize, qib_rcvhdrsize, uint, S_IRUGO);
+MODULE_PARM_DESC(rcvhdrsize, "receive header size in 32-bit words");
+
+static unsigned qib_rcvhdrentsize;
+module_param_named(rcvhdrentsize, qib_rcvhdrentsize, uint, S_IRUGO);
+MODULE_PARM_DESC(rcvhdrentsize, "receive header entry size in 32-bit words");
+
 #define MAX_ATTEN_LEN 64 /* plenty for any real system */
 /* for read back, default index is ~5m copper cable */
 static char txselect_list[MAX_ATTEN_LEN] = "10";
@@ -3530,8 +3545,11 @@ static void qib_7322_config_ctxts(struct qib_devdata *dd)
 
 	/* kr_rcvegrcnt changes based on the number of contexts enabled */
 	dd->cspec->rcvegrcnt = qib_read_kreg32(dd, kr_rcvegrcnt);
-	dd->rcvhdrcnt = max(dd->cspec->rcvegrcnt,
-				dd->num_pports > 1 ? 1024U : 2048U);
+	if (qib_rcvhdrcnt)
+		dd->rcvhdrcnt = max(dd->cspec->rcvegrcnt, qib_rcvhdrcnt);
+	else
+		dd->rcvhdrcnt = max(dd->cspec->rcvegrcnt,
+				    dd->num_pports > 1 ? 1024U : 2048U);
 }
 
 static int qib_7322_get_ib_cfg(struct qib_pportdata *ppd, int which)
@@ -6097,8 +6115,10 @@ static int qib_init_7322_variables(struct qib_devdata *dd)
 		ppd++;
 	}
 
-	dd->rcvhdrentsize = QIB_RCVHDR_ENTSIZE;
-	dd->rcvhdrsize = QIB_DFLT_RCVHDRSIZE;
+	dd->rcvhdrentsize = qib_rcvhdrentsize ?
+		qib_rcvhdrentsize : QIB_RCVHDR_ENTSIZE;
+	dd->rcvhdrsize = qib_rcvhdrsize ?
+		qib_rcvhdrsize : QIB_DFLT_RCVHDRSIZE;
 	dd->rhf_offset = dd->rcvhdrentsize - sizeof(u64) / sizeof(u32);
 
 	/* we always allocate at least 2048 bytes for eager buffers */

--
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 03/24] IB/qib: add support for the new QME7362 card

[Mike Marciniszyn]

This patch adds support to recognize another board variation
named QME7362.

Signed-off-by: Mike Marciniszyn <mike.marciniszyn-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/qib/qib_iba7322.c |    4 ++++
 1 files changed, 4 insertions(+), 0 deletions(-)

diff --git a/drivers/infiniband/hw/qib/qib_iba7322.c b/drivers/infiniband/hw/qib/qib_iba7322.c
index b35676f..40f4a23 100644
--- a/drivers/infiniband/hw/qib/qib_iba7322.c
+++ b/drivers/infiniband/hw/qib/qib_iba7322.c
@@ -3172,6 +3172,10 @@ static unsigned qib_7322_boardname(struct qib_devdata *dd)
 	case BOARD_QME7342:
 		n = "InfiniPath_QME7342";
 		break;
+	case 8:
+		n = "InfiniPath_QME7362";
+		dd->flags |= QIB_HAS_QSFP;
+		break;
 	case 15:
 		n = "InfiniPath_QLE7342_TEST";
 		dd->flags |= QIB_HAS_QSFP;

--
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 04/24] IB/qib: add thresholds to VendorPortCounters PMA operation

[Mike Marciniszyn]

This adds thresholds to be specified in the VendorPortCounters PMA
operation.

Signed-off-by: Mike Marciniszyn <mike.marciniszyn-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/qib/qib_mad.c |   87 +++++++++++++++++++++++------------
 1 files changed, 57 insertions(+), 30 deletions(-)

diff --git a/drivers/infiniband/hw/qib/qib_mad.c b/drivers/infiniband/hw/qib/qib_mad.c
index 94b0d1f..9794020 100644
--- a/drivers/infiniband/hw/qib/qib_mad.c
+++ b/drivers/infiniband/hw/qib/qib_mad.c
@@ -1135,10 +1135,10 @@ static int pma_get_classportinfo(struct ib_perf *pmp,
 	p->class_version = 1;
 	p->cap_mask = IB_PMA_CLASS_CAP_EXT_WIDTH;
 	/*
-	 * Set the most significant bit of CM2 to indicate support for
+	 * Set the most significant two bits of CM2 to indicate support for
 	 * congestion statistics
 	 */
-	p->reserved[0] = dd->psxmitwait_supported << 7;
+	p->reserved[0] = (dd->psxmitwait_supported * 3) << 6;
 	/*
 	 * Expected response time is 4.096 usec. * 2^18 == 1.073741824 sec.
 	 */
@@ -1709,8 +1709,14 @@ static int pma_set_portcounters(struct ib_perf *pmp,
 }
 
 static int pma_set_portcounters_cong(struct ib_perf *pmp,
-				     struct ib_device *ibdev, u8 port)
+				     struct ib_device *ibdev, u8 port,
+				     struct ib_perf *pmp_in)
 {
+	/* Congestion PMA packets start at offset 24 not 64 */
+	struct ib_pma_portcounters_cong *p_out =
+			(struct ib_pma_portcounters_cong *)pmp->reserved;
+	struct ib_pma_portcounters_cong *p_in =
+			(struct ib_pma_portcounters_cong *)pmp_in->reserved;
 	struct qib_ibport *ibp = to_iport(ibdev, port);
 	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
 	struct qib_devdata *dd = dd_from_ppd(ppd);
@@ -1724,36 +1730,56 @@ static int pma_set_portcounters_cong(struct ib_perf *pmp,
 	ret = pma_get_portcounters_cong(pmp, ibdev, port);
 
 	if (counter_select & IB_PMA_SEL_CONG_XMIT) {
-		spin_lock_irqsave(&ppd->ibport_data.lock, flags);
-		ppd->cong_stats.counter = 0;
-		dd->f_set_cntr_sample(ppd, QIB_CONG_TIMER_PSINTERVAL,
-				      0x0);
-		spin_unlock_irqrestore(&ppd->ibport_data.lock, flags);
+		if (p_out->port_xmit_wait > p_in->port_xmit_wait) {
+			spin_lock_irqsave(&ppd->ibport_data.lock, flags);
+			ppd->cong_stats.counter = 0;
+			dd->f_set_cntr_sample(ppd, QIB_CONG_TIMER_PSINTERVAL,
+					      0x0);
+			spin_unlock_irqrestore(&ppd->ibport_data.lock, flags);
+		}
 	}
 	if (counter_select & IB_PMA_SEL_CONG_PORT_DATA) {
-		ibp->z_port_xmit_data = cntrs.port_xmit_data;
-		ibp->z_port_rcv_data = cntrs.port_rcv_data;
-		ibp->z_port_xmit_packets = cntrs.port_xmit_packets;
-		ibp->z_port_rcv_packets = cntrs.port_rcv_packets;
+		if (p_out->port_xmit_data > p_in->port_xmit_data)
+			ibp->z_port_xmit_data = cntrs.port_xmit_data;
+		if (p_out->port_rcv_data > p_in->port_rcv_data)
+			ibp->z_port_rcv_data = cntrs.port_rcv_data;
+		if (p_out->port_xmit_packets > p_in->port_xmit_packets)
+			ibp->z_port_xmit_packets = cntrs.port_xmit_packets;
+		if (p_out->port_rcv_packets > p_in->port_rcv_packets)
+			ibp->z_port_rcv_packets = cntrs.port_rcv_packets;
 	}
 	if (counter_select & IB_PMA_SEL_CONG_ALL) {
-		ibp->z_symbol_error_counter =
-			cntrs.symbol_error_counter;
-		ibp->z_link_error_recovery_counter =
-			cntrs.link_error_recovery_counter;
-		ibp->z_link_downed_counter =
-			cntrs.link_downed_counter;
-		ibp->z_port_rcv_errors = cntrs.port_rcv_errors;
-		ibp->z_port_rcv_remphys_errors =
-			cntrs.port_rcv_remphys_errors;
-		ibp->z_port_xmit_discards =
-			cntrs.port_xmit_discards;
-		ibp->z_local_link_integrity_errors =
-			cntrs.local_link_integrity_errors;
-		ibp->z_excessive_buffer_overrun_errors =
-			cntrs.excessive_buffer_overrun_errors;
-		ibp->n_vl15_dropped = 0;
-		ibp->z_vl15_dropped = cntrs.vl15_dropped;
+		if (p_out->symbol_error_counter > p_in->symbol_error_counter)
+			ibp->z_symbol_error_counter =
+				cntrs.symbol_error_counter;
+		if (p_out->link_error_recovery_counter >
+		    p_in->link_error_recovery_counter)
+			ibp->z_link_error_recovery_counter =
+				cntrs.link_error_recovery_counter;
+		if (p_out->link_downed_counter > p_in->link_downed_counter)
+			ibp->z_link_downed_counter =
+				cntrs.link_downed_counter;
+		if (p_out->port_rcv_errors > p_in->port_rcv_errors)
+			ibp->z_port_rcv_errors = cntrs.port_rcv_errors;
+		if (p_out->port_rcv_remphys_errors >
+		    p_in->port_rcv_remphys_errors)
+			ibp->z_port_rcv_remphys_errors =
+				cntrs.port_rcv_remphys_errors;
+		if (p_out->port_xmit_discards > p_in->port_xmit_discards)
+			ibp->z_port_xmit_discards =
+				cntrs.port_xmit_discards;
+		if ((p_out->lli_ebor_errors & 0xf0) >
+		    (p_in->lli_ebor_errors & 0xf0))
+			ibp->z_local_link_integrity_errors =
+				cntrs.local_link_integrity_errors;
+		if ((p_out->lli_ebor_errors & 0x0f) >
+		    (p_in->lli_ebor_errors & 0x0f))
+			ibp->z_excessive_buffer_overrun_errors =
+				cntrs.excessive_buffer_overrun_errors;
+		if (p_out->vl15_dropped > p_in->vl15_dropped) {
+			ibp->n_vl15_dropped = 0;
+			ibp->z_vl15_dropped = cntrs.vl15_dropped;
+		}
 	}
 
 	return ret;
@@ -2004,7 +2030,8 @@ static int process_perf(struct ib_device *ibdev, u8 port,
 			ret = pma_set_portcounters_ext(pmp, ibdev, port);
 			goto bail;
 		case IB_PMA_PORT_COUNTERS_CONG:
-			ret = pma_set_portcounters_cong(pmp, ibdev, port);
+			ret = pma_set_portcounters_cong(pmp, ibdev, port,
+						(struct ib_perf *)in_mad);
 			goto bail;
 		default:
 			pmp->status |= IB_SMP_UNSUP_METH_ATTR;

--
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 04/24] IB/qib: add thresholds to VendorPortCounters PMA operation

[Hal Rosenstock]

On 12/28/2010 10:27 AM, Mike Marciniszyn wrote:
> This adds thresholds to be specified in the VendorPortCounters PMA
> operation.
>
> Signed-off-by: Mike Marciniszyn<mike.marciniszyn-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
> ---
>   drivers/infiniband/hw/qib/qib_mad.c |   87 +++++++++++++++++++++++------------
>   1 files changed, 57 insertions(+), 30 deletions(-)
>
> diff --git a/drivers/infiniband/hw/qib/qib_mad.c b/drivers/infiniband/hw/qib/qib_mad.c
> index 94b0d1f..9794020 100644
> --- a/drivers/infiniband/hw/qib/qib_mad.c
> +++ b/drivers/infiniband/hw/qib/qib_mad.c
> @@ -1135,10 +1135,10 @@ static int pma_get_classportinfo(struct ib_perf *pmp,
>   	p->class_version = 1;
>   	p->cap_mask = IB_PMA_CLASS_CAP_EXT_WIDTH;
>   	/*
> -	 * Set the most significant bit of CM2 to indicate support for
> +	 * Set the most significant two bits of CM2 to indicate support for
>   	 * congestion statistics
>   	 */

Where in the IBA spec do these bits come from ? What are they ?

> -	p->reserved[0] = dd->psxmitwait_supported<<  7;
> +	p->reserved[0] = (dd->psxmitwait_supported * 3)<<  6;

It looks weird to me to set an IBA reserved field. Should the structure 
be updated to contain the real field name ?

-- Hal

>   	/*
>   	 * Expected response time is 4.096 usec. * 2^18 == 1.073741824 sec.
>   	 */
> @@ -1709,8 +1709,14 @@ static int pma_set_portcounters(struct ib_perf *pmp,
>   }
>
>   static int pma_set_portcounters_cong(struct ib_perf *pmp,
> -				     struct ib_device *ibdev, u8 port)
> +				     struct ib_device *ibdev, u8 port,
> +				     struct ib_perf *pmp_in)
>   {
> +	/* Congestion PMA packets start at offset 24 not 64 */
> +	struct ib_pma_portcounters_cong *p_out =
> +			(struct ib_pma_portcounters_cong *)pmp->reserved;
> +	struct ib_pma_portcounters_cong *p_in =
> +			(struct ib_pma_portcounters_cong *)pmp_in->reserved;
>   	struct qib_ibport *ibp = to_iport(ibdev, port);
>   	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
>   	struct qib_devdata *dd = dd_from_ppd(ppd);
> @@ -1724,36 +1730,56 @@ static int pma_set_portcounters_cong(struct ib_perf *pmp,
>   	ret = pma_get_portcounters_cong(pmp, ibdev, port);
>
>   	if (counter_select&  IB_PMA_SEL_CONG_XMIT) {
> -		spin_lock_irqsave(&ppd->ibport_data.lock, flags);
> -		ppd->cong_stats.counter = 0;
> -		dd->f_set_cntr_sample(ppd, QIB_CONG_TIMER_PSINTERVAL,
> -				      0x0);
> -		spin_unlock_irqrestore(&ppd->ibport_data.lock, flags);
> +		if (p_out->port_xmit_wait>  p_in->port_xmit_wait) {
> +			spin_lock_irqsave(&ppd->ibport_data.lock, flags);
> +			ppd->cong_stats.counter = 0;
> +			dd->f_set_cntr_sample(ppd, QIB_CONG_TIMER_PSINTERVAL,
> +					      0x0);
> +			spin_unlock_irqrestore(&ppd->ibport_data.lock, flags);
> +		}
>   	}
>   	if (counter_select&  IB_PMA_SEL_CONG_PORT_DATA) {
> -		ibp->z_port_xmit_data = cntrs.port_xmit_data;
> -		ibp->z_port_rcv_data = cntrs.port_rcv_data;
> -		ibp->z_port_xmit_packets = cntrs.port_xmit_packets;
> -		ibp->z_port_rcv_packets = cntrs.port_rcv_packets;
> +		if (p_out->port_xmit_data>  p_in->port_xmit_data)
> +			ibp->z_port_xmit_data = cntrs.port_xmit_data;
> +		if (p_out->port_rcv_data>  p_in->port_rcv_data)
> +			ibp->z_port_rcv_data = cntrs.port_rcv_data;
> +		if (p_out->port_xmit_packets>  p_in->port_xmit_packets)
> +			ibp->z_port_xmit_packets = cntrs.port_xmit_packets;
> +		if (p_out->port_rcv_packets>  p_in->port_rcv_packets)
> +			ibp->z_port_rcv_packets = cntrs.port_rcv_packets;
>   	}
>   	if (counter_select&  IB_PMA_SEL_CONG_ALL) {
> -		ibp->z_symbol_error_counter =
> -			cntrs.symbol_error_counter;
> -		ibp->z_link_error_recovery_counter =
> -			cntrs.link_error_recovery_counter;
> -		ibp->z_link_downed_counter =
> -			cntrs.link_downed_counter;
> -		ibp->z_port_rcv_errors = cntrs.port_rcv_errors;
> -		ibp->z_port_rcv_remphys_errors =
> -			cntrs.port_rcv_remphys_errors;
> -		ibp->z_port_xmit_discards =
> -			cntrs.port_xmit_discards;
> -		ibp->z_local_link_integrity_errors =
> -			cntrs.local_link_integrity_errors;
> -		ibp->z_excessive_buffer_overrun_errors =
> -			cntrs.excessive_buffer_overrun_errors;
> -		ibp->n_vl15_dropped = 0;
> -		ibp->z_vl15_dropped = cntrs.vl15_dropped;
> +		if (p_out->symbol_error_counter>  p_in->symbol_error_counter)
> +			ibp->z_symbol_error_counter =
> +				cntrs.symbol_error_counter;
> +		if (p_out->link_error_recovery_counter>
> +		    p_in->link_error_recovery_counter)
> +			ibp->z_link_error_recovery_counter =
> +				cntrs.link_error_recovery_counter;
> +		if (p_out->link_downed_counter>  p_in->link_downed_counter)
> +			ibp->z_link_downed_counter =
> +				cntrs.link_downed_counter;
> +		if (p_out->port_rcv_errors>  p_in->port_rcv_errors)
> +			ibp->z_port_rcv_errors = cntrs.port_rcv_errors;
> +		if (p_out->port_rcv_remphys_errors>
> +		    p_in->port_rcv_remphys_errors)
> +			ibp->z_port_rcv_remphys_errors =
> +				cntrs.port_rcv_remphys_errors;
> +		if (p_out->port_xmit_discards>  p_in->port_xmit_discards)
> +			ibp->z_port_xmit_discards =
> +				cntrs.port_xmit_discards;
> +		if ((p_out->lli_ebor_errors&  0xf0)>
> +		    (p_in->lli_ebor_errors&  0xf0))
> +			ibp->z_local_link_integrity_errors =
> +				cntrs.local_link_integrity_errors;
> +		if ((p_out->lli_ebor_errors&  0x0f)>
> +		    (p_in->lli_ebor_errors&  0x0f))
> +			ibp->z_excessive_buffer_overrun_errors =
> +				cntrs.excessive_buffer_overrun_errors;
> +		if (p_out->vl15_dropped>  p_in->vl15_dropped) {
> +			ibp->n_vl15_dropped = 0;
> +			ibp->z_vl15_dropped = cntrs.vl15_dropped;
> +		}
>   	}
>
>   	return ret;
> @@ -2004,7 +2030,8 @@ static int process_perf(struct ib_device *ibdev, u8 port,
>   			ret = pma_set_portcounters_ext(pmp, ibdev, port);
>   			goto bail;
>   		case IB_PMA_PORT_COUNTERS_CONG:
> -			ret = pma_set_portcounters_cong(pmp, ibdev, port);
> +			ret = pma_set_portcounters_cong(pmp, ibdev, port,
> +						(struct ib_perf *)in_mad);
>   			goto bail;
>   		default:
>   			pmp->status |= IB_SMP_UNSUP_METH_ATTR;
>
> --
> 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
>

--
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 04/24] IB/qib: add thresholds to VendorPortCounters PMA operation

[Roland Dreier]

 > >   	p->class_version = 1;
 > >   	p->cap_mask = IB_PMA_CLASS_CAP_EXT_WIDTH;
 > >   	/*
 > > -	 * Set the most significant bit of CM2 to indicate support for
 > > +	 * Set the most significant two bits of CM2 to indicate support for
 > >   	 * congestion statistics
 > >   	 */
 > 
 > Where in the IBA spec do these bits come from ? What are they ?
 > 
 > > -	p->reserved[0] = dd->psxmitwait_supported<<  7;
 > > +	p->reserved[0] = (dd->psxmitwait_supported * 3)<<  6;
 > 
 > It looks weird to me to set an IBA reserved field. Should the
 > structure be updated to contain the real field name ?

The reserved field is (I think) actually the IBA "CapabilityMask2" bits
in ClassPortInfo; I agree that it would be better to update the
structure to name this correctly.

However, both the old and the new code look very fishy to me: bit 2 in
CapabilityMask is never set, which is the generic bit that says "If 1,
the management class agent implements additional class-specific
capabilities (CapabilityMask2)" so AFAICT, nothing should ever look in
CapabilityMask2 anyway.  

It seems that the original code was trying to set
ClassPortInfo.PortCountersXmitWaitSupported (bit 12), maybe.  I can't
fathom what the new code is doing.  And I can't find anything in the
spec that suggests that CM2 in this PMA ClassPortInfo should ever be used.

Mike, can you clarify this patch?  I'll hold off on applying this for now.

 - R.
--
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 05/24] IB/qib: generate completion callback on errors

[Mike Marciniszyn]

According to IBTA vol. 1, C11-30.1.1, a notification callback is
invoked if the CQ is armed for the next solicited completion event
or an error completion. The error case wasn't being generated
previously.

Signed-off-by: Mike Marciniszyn <mike.marciniszyn-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/qib/qib_cq.c |    3 ++-
 1 files changed, 2 insertions(+), 1 deletions(-)

diff --git a/drivers/infiniband/hw/qib/qib_cq.c b/drivers/infiniband/hw/qib/qib_cq.c
index a86cbf8..5246aa4 100644
--- a/drivers/infiniband/hw/qib/qib_cq.c
+++ b/drivers/infiniband/hw/qib/qib_cq.c
@@ -100,7 +100,8 @@ void qib_cq_enter(struct qib_cq *cq, struct ib_wc *entry, int solicited)
 	wc->head = next;
 
 	if (cq->notify == IB_CQ_NEXT_COMP ||
-	    (cq->notify == IB_CQ_SOLICITED && solicited)) {
+	    (cq->notify == IB_CQ_SOLICITED &&
+	     (solicited || entry->status != IB_WC_SUCCESS))) {
 		cq->notify = IB_CQ_NONE;
 		cq->triggered++;
 		/*

--
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 06/24] IB/qib: set port physical state even if other fields are invalid

[Mike Marciniszyn]

The IBTA vol. 1 release 1.2.1 spec. says:
C14-24.2.1: If PortInfo:Portstate=Down, then a SubnSet(PortInfo)
shall make any changes it specifies to PortInfo:PortPhysicalState;
any other result is vendor-dependent.

The patch changes the error handling so that the reply says there
are invalid fields but still attempts to set fields that are
in range including PortInfo:PortPhysicalState.

Signed-off-by: Mike Marciniszyn <mike.marciniszyn-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/qib/qib_mad.c |   45 +++++++++++++++++++----------------
 1 files changed, 25 insertions(+), 20 deletions(-)

diff --git a/drivers/infiniband/hw/qib/qib_mad.c b/drivers/infiniband/hw/qib/qib_mad.c
index 9794020..113807b 100644
--- a/drivers/infiniband/hw/qib/qib_mad.c
+++ b/drivers/infiniband/hw/qib/qib_mad.c
@@ -668,8 +668,8 @@ static int subn_set_portinfo(struct ib_smp *smp, struct ib_device *ibdev,
 	lid = be16_to_cpu(pip->lid);
 	/* Must be a valid unicast LID address. */
 	if (lid == 0 || lid >= QIB_MULTICAST_LID_BASE)
-		goto err;
-	if (ppd->lid != lid || ppd->lmc != (pip->mkeyprot_resv_lmc & 7)) {
+		smp->status |= IB_SMP_INVALID_FIELD;
+	else if (ppd->lid != lid || ppd->lmc != (pip->mkeyprot_resv_lmc & 7)) {
 		if (ppd->lid != lid)
 			qib_set_uevent_bits(ppd, _QIB_EVENT_LID_CHANGE_BIT);
 		if (ppd->lmc != (pip->mkeyprot_resv_lmc & 7))
@@ -683,8 +683,8 @@ static int subn_set_portinfo(struct ib_smp *smp, struct ib_device *ibdev,
 	msl = pip->neighbormtu_mastersmsl & 0xF;
 	/* Must be a valid unicast LID address. */
 	if (smlid == 0 || smlid >= QIB_MULTICAST_LID_BASE)
-		goto err;
-	if (smlid != ibp->sm_lid || msl != ibp->sm_sl) {
+		smp->status |= IB_SMP_INVALID_FIELD;
+	else if (smlid != ibp->sm_lid || msl != ibp->sm_sl) {
 		spin_lock_irqsave(&ibp->lock, flags);
 		if (ibp->sm_ah) {
 			if (smlid != ibp->sm_lid)
@@ -707,8 +707,9 @@ static int subn_set_portinfo(struct ib_smp *smp, struct ib_device *ibdev,
 		if (lwe == 0xFF)
 			lwe = ppd->link_width_supported;
 		else if (lwe >= 16 || (lwe & ~ppd->link_width_supported))
-			goto err;
-		set_link_width_enabled(ppd, lwe);
+			smp->status |= IB_SMP_INVALID_FIELD;
+		else if (lwe != ppd->link_width_enabled)
+			set_link_width_enabled(ppd, lwe);
 	}
 
 	lse = pip->linkspeedactive_enabled & 0xF;
@@ -721,8 +722,9 @@ static int subn_set_portinfo(struct ib_smp *smp, struct ib_device *ibdev,
 		if (lse == 15)
 			lse = ppd->link_speed_supported;
 		else if (lse >= 8 || (lse & ~ppd->link_speed_supported))
-			goto err;
-		set_link_speed_enabled(ppd, lse);
+			smp->status |= IB_SMP_INVALID_FIELD;
+		else if (lse != ppd->link_speed_enabled)
+			set_link_speed_enabled(ppd, lse);
 	}
 
 	/* Set link down default state. */
@@ -738,7 +740,7 @@ static int subn_set_portinfo(struct ib_smp *smp, struct ib_device *ibdev,
 					IB_LINKINITCMD_POLL);
 		break;
 	default:
-		goto err;
+		smp->status |= IB_SMP_INVALID_FIELD;
 	}
 
 	ibp->mkeyprot = pip->mkeyprot_resv_lmc >> 6;
@@ -748,15 +750,17 @@ static int subn_set_portinfo(struct ib_smp *smp, struct ib_device *ibdev,
 
 	mtu = ib_mtu_enum_to_int((pip->neighbormtu_mastersmsl >> 4) & 0xF);
 	if (mtu == -1)
-		goto err;
-	qib_set_mtu(ppd, mtu);
+		smp->status |= IB_SMP_INVALID_FIELD;
+	else
+		qib_set_mtu(ppd, mtu);
 
 	/* Set operational VLs */
 	vls = (pip->operationalvl_pei_peo_fpi_fpo >> 4) & 0xF;
 	if (vls) {
 		if (vls > ppd->vls_supported)
-			goto err;
-		(void) dd->f_set_ib_cfg(ppd, QIB_IB_CFG_OP_VLS, vls);
+			smp->status |= IB_SMP_INVALID_FIELD;
+		else
+			(void) dd->f_set_ib_cfg(ppd, QIB_IB_CFG_OP_VLS, vls);
 	}
 
 	if (pip->mkey_violations == 0)
@@ -770,10 +774,10 @@ static int subn_set_portinfo(struct ib_smp *smp, struct ib_device *ibdev,
 
 	ore = pip->localphyerrors_overrunerrors;
 	if (set_phyerrthreshold(ppd, (ore >> 4) & 0xF))
-		goto err;
+		smp->status |= IB_SMP_INVALID_FIELD;
 
 	if (set_overrunthreshold(ppd, (ore & 0xF)))
-		goto err;
+		smp->status |= IB_SMP_INVALID_FIELD;
 
 	ibp->subnet_timeout = pip->clientrereg_resv_subnetto & 0x1F;
 
@@ -792,7 +796,7 @@ static int subn_set_portinfo(struct ib_smp *smp, struct ib_device *ibdev,
 	state = pip->linkspeed_portstate & 0xF;
 	lstate = (pip->portphysstate_linkdown >> 4) & 0xF;
 	if (lstate && !(state == IB_PORT_DOWN || state == IB_PORT_NOP))
-		goto err;
+		smp->status |= IB_SMP_INVALID_FIELD;
 
 	/*
 	 * Only state changes of DOWN, ARM, and ACTIVE are valid
@@ -812,8 +816,10 @@ static int subn_set_portinfo(struct ib_smp *smp, struct ib_device *ibdev,
 			lstate = QIB_IB_LINKDOWN;
 		else if (lstate == 3)
 			lstate = QIB_IB_LINKDOWN_DISABLE;
-		else
-			goto err;
+		else {
+			smp->status |= IB_SMP_INVALID_FIELD;
+			break;
+		}
 		spin_lock_irqsave(&ppd->lflags_lock, flags);
 		ppd->lflags &= ~QIBL_LINKV;
 		spin_unlock_irqrestore(&ppd->lflags_lock, flags);
@@ -835,8 +841,7 @@ static int subn_set_portinfo(struct ib_smp *smp, struct ib_device *ibdev,
 		qib_set_linkstate(ppd, QIB_IB_LINKACTIVE);
 		break;
 	default:
-		/* XXX We have already partially updated our state! */
-		goto err;
+		smp->status |= IB_SMP_INVALID_FIELD;
 	}
 
 	ret = subn_get_portinfo(smp, ibdev, port);

--
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 07/24] IB/qib: UD send with immediate Rx completion has wrong size

[Mike Marciniszyn]

The code to generate receive completion entries for UD send with
immediate contains the wrong payload length. This is because when
the code to compute the payload size was moved, the value of hdrsize
didn't get moved too.  The fix is to update tlen directly.

Signed-off-by: Mike Marciniszyn <mike.marciniszyn-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/qib/qib_ud.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/drivers/infiniband/hw/qib/qib_ud.c b/drivers/infiniband/hw/qib/qib_ud.c
index e1b3da2..a4b945d 100644
--- a/drivers/infiniband/hw/qib/qib_ud.c
+++ b/drivers/infiniband/hw/qib/qib_ud.c
@@ -519,7 +519,7 @@ void qib_ud_rcv(struct qib_ibport *ibp, struct qib_ib_header *hdr,
 	    opcode == IB_OPCODE_UD_SEND_ONLY_WITH_IMMEDIATE) {
 		wc.ex.imm_data = ohdr->u.ud.imm_data;
 		wc.wc_flags = IB_WC_WITH_IMM;
-		hdrsize += sizeof(u32);
+		tlen -= sizeof(u32);
 	} else if (opcode == IB_OPCODE_UD_SEND_ONLY) {
 		wc.ex.imm_data = 0;
 		wc.wc_flags = 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 08/24] IB/qib: Handle transitions from ACTIVE_DEFERRED to ACTIVE better

[Mike Marciniszyn]

When the link transitions from ACTIVE_DEFERRED to ACTIVE, the driver
only sees the ACTIVE state. With this change, it will check whether
the state was already ACTIVE and if so, it will not generated IB
events and will not clear symbol error counts.

Signed-off-by: Mike Marciniszyn <mike.marciniszyn-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/qib/qib_intr.c |    3 ++-
 1 files changed, 2 insertions(+), 1 deletions(-)

diff --git a/drivers/infiniband/hw/qib/qib_intr.c b/drivers/infiniband/hw/qib/qib_intr.c
index 54a4082..a693c56 100644
--- a/drivers/infiniband/hw/qib/qib_intr.c
+++ b/drivers/infiniband/hw/qib/qib_intr.c
@@ -131,7 +131,8 @@ void qib_handle_e_ibstatuschanged(struct qib_pportdata *ppd, u64 ibcs)
 			/* start a 75msec timer to clear symbol errors */
 			mod_timer(&ppd->symerr_clear_timer,
 				  msecs_to_jiffies(75));
-		} else if (ltstate == IB_PHYSPORTSTATE_LINKUP) {
+		} else if (ltstate == IB_PHYSPORTSTATE_LINKUP &&
+			   !(ppd->lflags & QIBL_LINKACTIVE)) {
 			/* active, but not active defered */
 			qib_hol_up(ppd); /* useful only for 6120 now */
 			*ppd->statusp |=

--
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 09/24] IB/qib: Multi Florida HCA Hosts panic on reboot

[Mike Marciniszyn]

Add check when setting configured contexts that the value
does not exceed the number of contexts allocated for the
card. If the value exceeds the already allocated count,
set it to what is already allocated.

Signed-off-by: Mike Marciniszyn <mike.marciniszyn-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/qib/qib_init.c |    6 ++++--
 1 files changed, 4 insertions(+), 2 deletions(-)

diff --git a/drivers/infiniband/hw/qib/qib_init.c b/drivers/infiniband/hw/qib/qib_init.c
index f3b5039..7896afb 100644
--- a/drivers/infiniband/hw/qib/qib_init.c
+++ b/drivers/infiniband/hw/qib/qib_init.c
@@ -92,9 +92,11 @@ unsigned long *qib_cpulist;
 /* set number of contexts we'll actually use */
 void qib_set_ctxtcnt(struct qib_devdata *dd)
 {
-	if (!qib_cfgctxts)
+	if (!qib_cfgctxts) {
 		dd->cfgctxts = dd->first_user_ctxt + num_online_cpus();
-	else if (qib_cfgctxts < dd->num_pports)
+		if (dd->cfgctxts > dd->ctxtcnt)
+			dd->cfgctxts = dd->ctxtcnt;
+	} else if (qib_cfgctxts < dd->num_pports)
 		dd->cfgctxts = dd->ctxtcnt;
 	else if (qib_cfgctxts <= dd->ctxtcnt)
 		dd->cfgctxts = qib_cfgctxts;

--
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 10/24] IB/qib: fix context allocation with multiple HCAs

[Mike Marciniszyn]

The driver was incorrectly choosing HCAs on which to allocate
new user contexts based on overall count of usable ports
regardless whether the usable port was on the currently selected
HCA.

Signed-off-by: Mike Marciniszyn <mike.marciniszyn-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/qib/qib_file_ops.c |   10 +++++-----
 1 files changed, 5 insertions(+), 5 deletions(-)

diff --git a/drivers/infiniband/hw/qib/qib_file_ops.c b/drivers/infiniband/hw/qib/qib_file_ops.c
index 79d9971..75bfad1 100644
--- a/drivers/infiniband/hw/qib/qib_file_ops.c
+++ b/drivers/infiniband/hw/qib/qib_file_ops.c
@@ -1379,17 +1379,17 @@ static int get_a_ctxt(struct file *fp, const struct qib_user_info *uinfo,
 		/* find device (with ACTIVE ports) with fewest ctxts in use */
 		for (ndev = 0; ndev < devmax; ndev++) {
 			struct qib_devdata *dd = qib_lookup(ndev);
-			unsigned cused = 0, cfree = 0;
+			unsigned cused = 0, cfree = 0, pusable = 0;
 			if (!dd)
 				continue;
 			if (port && port <= dd->num_pports &&
 			    usable(dd->pport + port - 1))
-				dusable = 1;
+				pusable = 1;
 			else
 				for (i = 0; i < dd->num_pports; i++)
 					if (usable(dd->pport + i))
-						dusable++;
-			if (!dusable)
+						pusable++;
+			if (!pusable)
 				continue;
 			for (ctxt = dd->first_user_ctxt; ctxt < dd->cfgctxts;
 			     ctxt++)
@@ -1397,7 +1397,7 @@ static int get_a_ctxt(struct file *fp, const struct qib_user_info *uinfo,
 					cused++;
 				else
 					cfree++;
-			if (cfree && cused < inuse) {
+			if (pusable && cfree && cused < inuse) {
 				udd = dd;
 				inuse = cused;
 			}

--
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 11/24] IB/qib: clear WAIT_SEND flags when setting QP to error state

[Mike Marciniszyn]

If these flags are set when the QP is transitioned to the error
state, it will wait until the flags are cleared, which may never
happen if the error transition is due to a link going down.

Signed-off-by: Mike Marciniszyn <mike.marciniszyn-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/qib/qib_qp.c |    4 ++++
 1 files changed, 4 insertions(+), 0 deletions(-)

diff --git a/drivers/infiniband/hw/qib/qib_qp.c b/drivers/infiniband/hw/qib/qib_qp.c
index 6c39851..32dacd4 100644
--- a/drivers/infiniband/hw/qib/qib_qp.c
+++ b/drivers/infiniband/hw/qib/qib_qp.c
@@ -468,6 +468,10 @@ int qib_error_qp(struct qib_qp *qp, enum ib_wc_status err)
 		qp->s_flags &= ~(QIB_S_TIMER | QIB_S_WAIT_RNR);
 		del_timer(&qp->s_timer);
 	}
+
+	if (qp->s_flags & QIB_S_ANY_WAIT_SEND)
+		qp->s_flags &= ~QIB_S_ANY_WAIT_SEND;
+
 	spin_lock(&dev->pending_lock);
 	if (!list_empty(&qp->iowait) && !(qp->s_flags & QIB_S_BUSY)) {
 		qp->s_flags &= ~QIB_S_ANY_WAIT_IO;

--
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 12/24] IB/qib: New SERDES init routine and improvements to SI quality

[Mike Marciniszyn]

Implement new SERDES initialization routine and improvements
to SI - disable LE1 adaptation, disable LOS after link-up, set
better SERDES parameters.

Signed-off-by: Mike Marciniszyn <mike.marciniszyn-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/qib/qib_iba7322.c |  270 +++++++++++++++++++++++++++++--
 1 files changed, 255 insertions(+), 15 deletions(-)

diff --git a/drivers/infiniband/hw/qib/qib_iba7322.c b/drivers/infiniband/hw/qib/qib_iba7322.c
index 40f4a23..d23297a 100644
--- a/drivers/infiniband/hw/qib/qib_iba7322.c
+++ b/drivers/infiniband/hw/qib/qib_iba7322.c
@@ -71,6 +71,9 @@ static void qib_7322_mini_pcs_reset(struct qib_pportdata *);
 
 static u32 ahb_mod(struct qib_devdata *, int, int, int, u32, u32);
 static void ibsd_wr_allchans(struct qib_pportdata *, int, unsigned, unsigned);
+static void serdes_7322_los_enable(struct qib_pportdata *, int);
+static int serdes_7322_init_old(struct qib_pportdata *);
+static int serdes_7322_init_new(struct qib_pportdata *);
 
 #define BMASK(msb, lsb) (((1 << ((msb) + 1 - (lsb))) - 1) << (lsb))
 
@@ -1692,6 +1695,8 @@ static void handle_serdes_issues(struct qib_pportdata *ppd, u64 ibcst)
 	    (ibcst & SYM_MASK(IBCStatusA_0, LinkSpeedQDR))) {
 		force_h1(ppd);
 		ppd->cpspec->qdr_reforce = 1;
+		if (!ppd->dd->cspec->r1)
+			serdes_7322_los_enable(ppd, 0);
 	} else if (ppd->cpspec->qdr_reforce &&
 		(ibcst & SYM_MASK(IBCStatusA_0, LinkSpeedQDR)) &&
 		 (ibclt == IB_7322_LT_STATE_CFGENH ||
@@ -1707,15 +1712,32 @@ static void handle_serdes_issues(struct qib_pportdata *ppd, u64 ibcst)
 	      ibclt <= IB_7322_LT_STATE_SLEEPQUIET)))
 		adj_tx_serdes(ppd);
 
-	if (!ppd->cpspec->qdr_dfe_on && ibclt != IB_7322_LT_STATE_LINKUP &&
-	    ibclt <= IB_7322_LT_STATE_SLEEPQUIET) {
-		ppd->cpspec->qdr_dfe_on = 1;
-		ppd->cpspec->qdr_dfe_time = 0;
-		/* On link down, reenable QDR adaptation */
-		qib_write_kreg_port(ppd, krp_static_adapt_dis(2),
-			ppd->dd->cspec->r1 ?
-				    QDR_STATIC_ADAPT_DOWN_R1 :
-				    QDR_STATIC_ADAPT_DOWN);
+	if (ibclt != IB_7322_LT_STATE_LINKUP) {
+		u8 ltstate = qib_7322_phys_portstate(ibcst);
+		u8 pibclt = (u8)SYM_FIELD(ppd->lastibcstat, IBCStatusA_0,
+					  LinkTrainingState);
+		if (!ppd->dd->cspec->r1 &&
+		    pibclt == IB_7322_LT_STATE_LINKUP &&
+		    ltstate != IB_PHYSPORTSTATE_LINK_ERR_RECOVER &&
+		    ltstate != IB_PHYSPORTSTATE_RECOVERY_RETRAIN &&
+		    ltstate != IB_PHYSPORTSTATE_RECOVERY_WAITRMT &&
+		    ltstate != IB_PHYSPORTSTATE_RECOVERY_IDLE)
+			/* If the link went down (but no into recovery,
+			 * turn LOS back on */
+			serdes_7322_los_enable(ppd, 1);
+		if (!ppd->cpspec->qdr_dfe_on &&
+		    ibclt <= IB_7322_LT_STATE_SLEEPQUIET) {
+			ppd->cpspec->qdr_dfe_on = 1;
+			ppd->cpspec->qdr_dfe_time = 0;
+			/* On link down, reenable QDR adaptation */
+			qib_write_kreg_port(ppd, krp_static_adapt_dis(2),
+					    ppd->dd->cspec->r1 ?
+					    QDR_STATIC_ADAPT_DOWN_R1 :
+					    QDR_STATIC_ADAPT_DOWN);
+			printk(KERN_INFO QIB_DRV_NAME
+				" IB%u:%u re-enabled QDR adaptation "
+				"ibclt %x\n", ppd->dd->unit, ppd->port, ibclt);
+		}
 	}
 }
 
@@ -5544,7 +5566,7 @@ static void qsfp_7322_event(struct work_struct *work)
 		u64 now = get_jiffies_64();
 		if (time_after64(now, pwrup))
 			break;
-		msleep(1);
+		msleep(20);
 	}
 	ret = qib_refresh_qsfp_cache(ppd, &qd->cache);
 	/*
@@ -6519,7 +6541,7 @@ static void qib_7322_txchk_change(struct qib_devdata *dd, u32 start,
 		/* make sure we see an updated copy next time around */
 		sendctrl_7322_mod(dd->pport, QIB_SENDCTRL_AVAIL_BLIP);
 		sleeps++;
-		msleep(1);
+		msleep(20);
 	}
 
 	switch (which) {
@@ -7234,9 +7256,30 @@ static void ibsd_wr_allchans(struct qib_pportdata *ppd, int addr, unsigned data,
 	}
 }
 
+static void serdes_7322_los_enable(struct qib_pportdata *ppd, int enable)
+{
+	u64 data = qib_read_kreg_port(ppd, krp_serdesctrl);
+	printk(KERN_INFO QIB_DRV_NAME " Turning LOS %s for port %d\n",
+		 (enable ? "on" : "off"), ppd->port);
+	if (enable)
+		data |= SYM_MASK(IBSerdesCtrl_0, RXLOSEN);
+	else
+		data &= ~SYM_MASK(IBSerdesCtrl_0, RXLOSEN);
+	qib_write_kreg_port(ppd, krp_serdesctrl, data);
+}
+
 static int serdes_7322_init(struct qib_pportdata *ppd)
 {
-	u64 data;
+	int ret = 0;
+	if (ppd->dd->cspec->r1)
+		ret = serdes_7322_init_old(ppd);
+	else
+		ret = serdes_7322_init_new(ppd);
+	return ret;
+}
+
+static int serdes_7322_init_old(struct qib_pportdata *ppd)
+{
 	u32 le_val;
 
 	/*
@@ -7294,9 +7337,7 @@ static int serdes_7322_init(struct qib_pportdata *ppd)
 	ibsd_wr_allchans(ppd, 20, (2 << 10), BMASK(12, 10)); /* DDR */
 	ibsd_wr_allchans(ppd, 20, (4 << 13), BMASK(15, 13)); /* SDR */
 
-	data = qib_read_kreg_port(ppd, krp_serdesctrl);
-	qib_write_kreg_port(ppd, krp_serdesctrl, data |
-		SYM_MASK(IBSerdesCtrl_0, RXLOSEN));
+	serdes_7322_los_enable(ppd, 1);
 
 	/* rxbistena; set 0 to avoid effects of it switch later */
 	ibsd_wr_allchans(ppd, 9, 0 << 15, 1 << 15);
@@ -7336,6 +7377,205 @@ static int serdes_7322_init(struct qib_pportdata *ppd)
 	return 0;
 }
 
+static int serdes_7322_init_new(struct qib_pportdata *ppd)
+{
+	u64 tstart;
+	u32 le_val, rxcaldone;
+	int chan, chan_done = (1 << SERDES_CHANS) - 1;
+
+	/*
+	 * Initialize the Tx DDS tables.  Also done every QSFP event,
+	 * for adapters with QSFP
+	 */
+	init_txdds_table(ppd, 0);
+
+	/* Clear cmode-override, may be set from older driver */
+	ahb_mod(ppd->dd, IBSD(ppd->hw_pidx), 5, 10, 0 << 14, 1 << 14);
+
+	/* ensure no tx overrides from earlier driver loads */
+	qib_write_kreg_port(ppd, krp_tx_deemph_override,
+		SYM_MASK(IBSD_TX_DEEMPHASIS_OVERRIDE_0,
+		reset_tx_deemphasis_override));
+
+	/* START OF LSI SUGGESTED SERDES BRINGUP */
+	/* Reset - Calibration Setup */
+	/*       Stop DFE adaptaion */
+	ibsd_wr_allchans(ppd, 1, 0, BMASK(9, 1));
+	/*       Disable LE1 */
+	ibsd_wr_allchans(ppd, 13, 0, BMASK(5, 5));
+	/*       Disable autoadapt for LE1 */
+	ibsd_wr_allchans(ppd, 1, 0, BMASK(15, 15));
+	/*       Disable LE2 */
+	ibsd_wr_allchans(ppd, 13, 0, BMASK(6, 6));
+	/*       Disable VGA */
+	ibsd_wr_allchans(ppd, 5, 0, BMASK(0, 0));
+	/*       Disable AFE Offset Cancel */
+	ibsd_wr_allchans(ppd, 12, 0, BMASK(12, 12));
+	/*       Disable Timing Loop */
+	ibsd_wr_allchans(ppd, 2, 0, BMASK(3, 3));
+	/*       Disable Frequency Loop */
+	ibsd_wr_allchans(ppd, 2, 0, BMASK(4, 4));
+	/*       Disable Baseline Wander Correction */
+	ibsd_wr_allchans(ppd, 13, 0, BMASK(13, 13));
+	/*       Disable RX Calibration */
+	ibsd_wr_allchans(ppd, 4, 0, BMASK(10, 10));
+	/*       Disable RX Offset Calibration */
+	ibsd_wr_allchans(ppd, 12, 0, BMASK(4, 4));
+	/*       Select BB CDR */
+	ibsd_wr_allchans(ppd, 2, (1 << 15), BMASK(15, 15));
+	/*       CDR Step Size */
+	ibsd_wr_allchans(ppd, 5, 0, BMASK(9, 8));
+	/*       Enable phase Calibration */
+	ibsd_wr_allchans(ppd, 12, (1 << 5), BMASK(5, 5));
+	/*       DFE Bandwidth [2:14-12] */
+	ibsd_wr_allchans(ppd, 2, (4 << 12), BMASK(14, 12));
+	/*       DFE Config (4 taps only) */
+	ibsd_wr_allchans(ppd, 16, 0, BMASK(1, 0));
+	/*       Gain Loop Bandwidth */
+	if (!ppd->dd->cspec->r1) {
+		ibsd_wr_allchans(ppd, 12, 1 << 12, BMASK(12, 12));
+		ibsd_wr_allchans(ppd, 12, 2 << 8, BMASK(11, 8));
+	} else {
+		ibsd_wr_allchans(ppd, 19, (3 << 11), BMASK(13, 11));
+	}
+	/*       Baseline Wander Correction Gain [13:4-0] (leave as default) */
+	/*       Baseline Wander Correction Gain [3:7-5] (leave as default) */
+	/*       Data Rate Select [5:7-6] (leave as default) */
+	/*       RX Parralel Word Width [3:10-8] (leave as default) */
+
+	/* RX REST */
+	/*       Single- or Multi-channel reset */
+	/*       RX Analog reset */
+	/*       RX Digital reset */
+	ibsd_wr_allchans(ppd, 0, 0, BMASK(15, 13));
+	msleep(20);
+	/*       RX Analog reset */
+	ibsd_wr_allchans(ppd, 0, (1 << 14), BMASK(14, 14));
+	msleep(20);
+	/*       RX Digital reset */
+	ibsd_wr_allchans(ppd, 0, (1 << 13), BMASK(13, 13));
+	msleep(20);
+
+	/* setup LoS params; these are subsystem, so chan == 5 */
+	/* LoS filter threshold_count on, ch 0-3, set to 8 */
+	ahb_mod(ppd->dd, IBSD(ppd->hw_pidx), 5, 5, 8 << 11, BMASK(14, 11));
+	ahb_mod(ppd->dd, IBSD(ppd->hw_pidx), 5, 7, 8 << 4, BMASK(7, 4));
+	ahb_mod(ppd->dd, IBSD(ppd->hw_pidx), 5, 8, 8 << 11, BMASK(14, 11));
+	ahb_mod(ppd->dd, IBSD(ppd->hw_pidx), 5, 10, 8 << 4, BMASK(7, 4));
+
+	/* LoS filter threshold_count off, ch 0-3, set to 4 */
+	ahb_mod(ppd->dd, IBSD(ppd->hw_pidx), 5, 6, 4 << 0, BMASK(3, 0));
+	ahb_mod(ppd->dd, IBSD(ppd->hw_pidx), 5, 7, 4 << 8, BMASK(11, 8));
+	ahb_mod(ppd->dd, IBSD(ppd->hw_pidx), 5, 9, 4 << 0, BMASK(3, 0));
+	ahb_mod(ppd->dd, IBSD(ppd->hw_pidx), 5, 10, 4 << 8, BMASK(11, 8));
+
+	/* LoS filter select enabled */
+	ahb_mod(ppd->dd, IBSD(ppd->hw_pidx), 5, 9, 1 << 15, 1 << 15);
+
+	/* LoS target data:  SDR=4, DDR=2, QDR=1 */
+	ibsd_wr_allchans(ppd, 14, (1 << 3), BMASK(5, 3)); /* QDR */
+	ibsd_wr_allchans(ppd, 20, (2 << 10), BMASK(12, 10)); /* DDR */
+	ibsd_wr_allchans(ppd, 20, (4 << 13), BMASK(15, 13)); /* SDR */
+
+	/* Turn on LOS on initial SERDES init */
+	serdes_7322_los_enable(ppd, 1);
+	/* FLoop LOS gate: PPM filter  enabled */
+	ibsd_wr_allchans(ppd, 38, 0 << 10, 1 << 10);
+
+	/* RX LATCH CALIBRATION */
+	/*       Enable Eyefinder Phase Calibration latch */
+	ibsd_wr_allchans(ppd, 15, 1, BMASK(0, 0));
+	/*       Enable RX Offset Calibration latch */
+	ibsd_wr_allchans(ppd, 12, (1 << 4), BMASK(4, 4));
+	msleep(20);
+	/*       Start Calibration */
+	ibsd_wr_allchans(ppd, 4, (1 << 10), BMASK(10, 10));
+	tstart = get_jiffies_64();
+	while (chan_done &&
+	       !time_after64(tstart, tstart + msecs_to_jiffies(500))) {
+		msleep(20);
+		for (chan = 0; chan < SERDES_CHANS; ++chan) {
+			rxcaldone = ahb_mod(ppd->dd, IBSD(ppd->hw_pidx),
+					    (chan + (chan >> 1)),
+					    25, 0, 0);
+			if ((~rxcaldone & (u32)BMASK(9, 9)) == 0 &&
+			    (~chan_done & (1 << chan)) == 0)
+				chan_done &= ~(1 << chan);
+		}
+	}
+	if (chan_done) {
+		printk(KERN_INFO QIB_DRV_NAME
+			 " Serdes %d calibration not done after .5 sec: 0x%x\n",
+			 IBSD(ppd->hw_pidx), chan_done);
+	} else {
+		for (chan = 0; chan < SERDES_CHANS; ++chan) {
+			rxcaldone = ahb_mod(ppd->dd, IBSD(ppd->hw_pidx),
+					    (chan + (chan >> 1)),
+					    25, 0, 0);
+			if ((~rxcaldone & (u32)BMASK(10, 10)) == 0)
+				printk(KERN_INFO QIB_DRV_NAME
+					 " Serdes %d chan %d calibration "
+					 "failed\n", IBSD(ppd->hw_pidx), chan);
+		}
+	}
+
+	/*       Turn off Calibration */
+	ibsd_wr_allchans(ppd, 4, 0, BMASK(10, 10));
+	msleep(20);
+
+	/* BRING RX UP */
+	/*       Set LE2 value (May be overridden in qsfp_7322_event) */
+	le_val = IS_QME(ppd->dd) ? LE2_QME : LE2_DEFAULT;
+	ibsd_wr_allchans(ppd, 13, (le_val << 7), BMASK(9, 7));
+	/*       Set LE2 Loop bandwidth */
+	ibsd_wr_allchans(ppd, 3, (7 << 5), BMASK(7, 5));
+	/*       Enable LE2 */
+	ibsd_wr_allchans(ppd, 13, (1 << 6), BMASK(6, 6));
+	msleep(20);
+	/*       Enable H0 only */
+	ibsd_wr_allchans(ppd, 1, 1, BMASK(9, 1));
+	/* gain hi stop 32 (22) (6:1) lo stop 7 (10:7) target 22 (13) (15:11) */
+	le_val = (ppd->dd->cspec->r1 || IS_QME(ppd->dd)) ? 0xb6c0 : 0x6bac;
+	ibsd_wr_allchans(ppd, 21, le_val, 0xfffe);
+	/*       Enable VGA */
+	ibsd_wr_allchans(ppd, 5, 0, BMASK(0, 0));
+	msleep(20);
+	/*       Set Frequency Loop Bandwidth */
+	ibsd_wr_allchans(ppd, 2, (7 << 5), BMASK(8, 5));
+	/*       Enable Frequency Loop */
+	ibsd_wr_allchans(ppd, 2, (1 << 4), BMASK(4, 4));
+	/*       Set Timing Loop Bandwidth */
+	ibsd_wr_allchans(ppd, 2, 0, BMASK(11, 9));
+	/*       Enable Timing Loop */
+	ibsd_wr_allchans(ppd, 2, (1 << 3), BMASK(3, 3));
+	msleep(50);
+	/*       Enable DFE
+	 *       Set receive adaptation mode.  SDR and DDR adaptation are
+	 *       always on, and QDR is initially enabled; later disabled.
+	 */
+	qib_write_kreg_port(ppd, krp_static_adapt_dis(0), 0ULL);
+	qib_write_kreg_port(ppd, krp_static_adapt_dis(1), 0ULL);
+	qib_write_kreg_port(ppd, krp_static_adapt_dis(2),
+			    ppd->dd->cspec->r1 ?
+			    QDR_STATIC_ADAPT_DOWN_R1 : QDR_STATIC_ADAPT_DOWN);
+	ppd->cpspec->qdr_dfe_on = 1;
+	/*       Disable LE1  */
+	ibsd_wr_allchans(ppd, 13, (0 << 5), (1 << 5));
+	/*       Disable auto adapt for LE1 */
+	ibsd_wr_allchans(ppd, 1, (0 << 15), BMASK(15, 15));
+	msleep(20);
+	/*       Enable AFE Offset Cancel */
+	ibsd_wr_allchans(ppd, 12, (1 << 12), BMASK(12, 12));
+	/*       Enable Baseline Wander Correction */
+	ibsd_wr_allchans(ppd, 12, (1 << 13), BMASK(13, 13));
+	/* Termination: rxtermctrl_r2d addr 11 bits [12:11] = 1 */
+	ibsd_wr_allchans(ppd, 11, (1 << 11), BMASK(12, 11));
+	/* VGA output common mode */
+	ibsd_wr_allchans(ppd, 12, (3 << 2), BMASK(3, 2));
+
+	return 0;
+}
+
 /* start adjust QMH serdes parameters */
 
 static void set_man_code(struct qib_pportdata *ppd, int chan, int code)

--
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 13/24] IB/qib: Reset packet list after freeing

[Mike Marciniszyn]

Reset the list pointers after freeing the SDMA packet
list. This is done to any potential double-free cases.

Signed-off-by: Mike Marciniszyn <mike.marciniszyn-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/qib/qib_user_sdma.c |    1 +
 1 files changed, 1 insertions(+), 0 deletions(-)

diff --git a/drivers/infiniband/hw/qib/qib_user_sdma.c b/drivers/infiniband/hw/qib/qib_user_sdma.c
index 4c19e06..66208bc 100644
--- a/drivers/infiniband/hw/qib/qib_user_sdma.c
+++ b/drivers/infiniband/hw/qib/qib_user_sdma.c
@@ -382,6 +382,7 @@ static void qib_user_sdma_free_pkt_list(struct device *dev,
 
 		kmem_cache_free(pq->pkt_slab, pkt);
 	}
+	INIT_LIST_HEAD(list);
 }
 
 /*

--
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 14/24] IB/qib: Add a few new SERDES tunings

[Mike Marciniszyn]

Add new SERDES tuning to aid manufacturing.

Signed-off-by: Mike Marciniszyn <mike.marciniszyn-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/qib/qib_iba7322.c |   29 ++++++++++++++++++++++++++---
 1 files changed, 26 insertions(+), 3 deletions(-)

diff --git a/drivers/infiniband/hw/qib/qib_iba7322.c b/drivers/infiniband/hw/qib/qib_iba7322.c
index d23297a..a9c8c72 100644
--- a/drivers/infiniband/hw/qib/qib_iba7322.c
+++ b/drivers/infiniband/hw/qib/qib_iba7322.c
@@ -562,6 +562,7 @@ static void write_tx_serdes_param(struct qib_pportdata *, struct txdds_ent *);
 
 #define TXDDS_TABLE_SZ 16 /* number of entries per speed in onchip table */
 #define TXDDS_EXTRA_SZ 13 /* number of extra tx settings entries */
+#define TXDDS_MFG_SZ 2    /* number of mfg tx settings entries */
 #define SERDES_CHANS 4 /* yes, it's obvious, but one less magic number */
 
 #define H1_FORCE_VAL 8
@@ -5623,6 +5624,7 @@ static void set_no_qsfp_atten(struct qib_devdata *dd, int change)
 	u32 pidx, unit, port, deflt, h1;
 	unsigned long val;
 	int any = 0, seth1;
+	int txdds_size;
 
 	str = txselect_list;
 
@@ -5631,6 +5633,10 @@ static void set_no_qsfp_atten(struct qib_devdata *dd, int change)
 	for (pidx = 0; pidx < dd->num_pports; ++pidx)
 		dd->pport[pidx].cpspec->no_eep = deflt;
 
+	txdds_size = TXDDS_TABLE_SZ + TXDDS_EXTRA_SZ;
+	if (IS_QME(dd) || IS_QMH(dd))
+		txdds_size += TXDDS_MFG_SZ;
+
 	while (*nxt && nxt[1]) {
 		str = ++nxt;
 		unit = simple_strtoul(str, &nxt, 0);
@@ -5653,7 +5659,7 @@ static void set_no_qsfp_atten(struct qib_devdata *dd, int change)
 				;
 			continue;
 		}
-		if (val >= TXDDS_TABLE_SZ + TXDDS_EXTRA_SZ)
+		if (val >= txdds_size)
 			continue;
 		seth1 = 0;
 		h1 = 0; /* gcc thinks it might be used uninitted */
@@ -5705,10 +5711,11 @@ static int setup_txselect(const char *str, struct kernel_param *kp)
 		return -ENOSPC;
 	}
 	val = simple_strtoul(str, &n, 0);
-	if (n == str || val >= (TXDDS_TABLE_SZ + TXDDS_EXTRA_SZ)) {
+	if (n == str || val >= (TXDDS_TABLE_SZ + TXDDS_EXTRA_SZ +
+				TXDDS_MFG_SZ)) {
 		printk(KERN_INFO QIB_DRV_NAME
 		       "txselect_values must start with a number < %d\n",
-			TXDDS_TABLE_SZ + TXDDS_EXTRA_SZ);
+			TXDDS_TABLE_SZ + TXDDS_EXTRA_SZ + TXDDS_MFG_SZ);
 		return -EINVAL;
 	}
 	strcpy(txselect_list, str);
@@ -7039,6 +7046,12 @@ static const struct txdds_ent txdds_extra_qdr[TXDDS_EXTRA_SZ] = {
 	{  0, 1,  0, 12 },	/* QMH7342 backplane settings */
 };
 
+static const struct txdds_ent txdds_extra_mfg[TXDDS_MFG_SZ] = {
+	/* amp, pre, main, post */
+	{ 0, 0, 0, 0 },         /* QME7342 mfg settings */
+	{ 0, 0, 0, 6 },         /* QME7342 P2 mfg settings */
+};
+
 static const struct txdds_ent *get_atten_table(const struct txdds_ent *txdds,
 					       unsigned atten)
 {
@@ -7112,6 +7125,16 @@ static void find_best_ent(struct qib_pportdata *ppd,
 		*sdr_dds = &txdds_extra_sdr[idx];
 		*ddr_dds = &txdds_extra_ddr[idx];
 		*qdr_dds = &txdds_extra_qdr[idx];
+	} else if ((IS_QME(ppd->dd) || IS_QMH(ppd->dd)) &&
+		   ppd->cpspec->no_eep < (TXDDS_TABLE_SZ + TXDDS_EXTRA_SZ +
+					  TXDDS_MFG_SZ)) {
+		idx = ppd->cpspec->no_eep - (TXDDS_TABLE_SZ + TXDDS_EXTRA_SZ);
+		printk(KERN_INFO QIB_DRV_NAME
+			" IB%u:%u use idx %u into txdds_mfg\n",
+			ppd->dd->unit, ppd->port, idx);
+		*sdr_dds = &txdds_extra_mfg[idx];
+		*ddr_dds = &txdds_extra_mfg[idx];
+		*qdr_dds = &txdds_extra_mfg[idx];
 	} else {
 		/* this shouldn't happen, it's range checked */
 		*sdr_dds = txdds_sdr + qib_long_atten;

--
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 15/24] IB/qib: Avoid duplicate writes to the rcv head register

[Mike Marciniszyn]

Avoid duplicate writes to the head register as this can lead to lost interrupts
if the context goes full before the second write is done.

Signed-off-by: Mike Marciniszyn <mike.marciniszyn-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/qib/qib_driver.c |   10 +++++++++-
 1 files changed, 9 insertions(+), 1 deletions(-)

diff --git a/drivers/infiniband/hw/qib/qib_driver.c b/drivers/infiniband/hw/qib/qib_driver.c
index 9cd1936..3ed0d5a 100644
--- a/drivers/infiniband/hw/qib/qib_driver.c
+++ b/drivers/infiniband/hw/qib/qib_driver.c
@@ -71,6 +71,11 @@ MODULE_DESCRIPTION("QLogic IB driver");
  */
 #define QIB_PIO_MAXIBHDR 128
 
+/*
+ * QIB_MAX_PKT_RCV is the max # if packets processed per receive interrupt.
+ */
+#define QIB_MAX_PKT_RECV 64
+
 struct qlogic_ib_stats qib_stats;
 
 const char *qib_get_unit_name(int unit)
@@ -335,7 +340,7 @@ u32 qib_kreceive(struct qib_ctxtdata *rcd, u32 *llic, u32 *npkts)
 		smp_rmb();  /* prevent speculative reads of dma'ed hdrq */
 	}
 
-	for (last = 0, i = 1; !last && i <= 64; i += !last) {
+	for (last = 0, i = 1; !last; i += !last) {
 		hdr = dd->f_get_msgheader(dd, rhf_addr);
 		eflags = qib_hdrget_err_flags(rhf_addr);
 		etype = qib_hdrget_rcv_type(rhf_addr);
@@ -384,6 +389,9 @@ move_along:
 		l += rsize;
 		if (l >= maxcnt)
 			l = 0;
+		if (i == QIB_MAX_PKT_RECV)
+			last = 1;
+
 		rhf_addr = (__le32 *) rcd->rcvhdrq + l + dd->rhf_offset;
 		if (dd->flags & QIB_NODMA_RTAIL) {
 			u32 seq = qib_hdrget_seq(rhf_addr);

--
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 16/24] IB/qib: interrupt mitigation fix

[Mike Marciniszyn]

For SusieQ we need to write to the interrupt timer register before updating the
header queue head with interrupt count. This is to ensure that the timer is
enabled properly and a receive available interrupt is delivered else this
interrupt can be lost if the receiver header/eager queues are full before the
timer is enabled.

Signed-off-by: Mike Marciniszyn <mike.marciniszyn-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/qib/qib.h         |    2 +-
 drivers/infiniband/hw/qib/qib_driver.c  |    4 ++--
 drivers/infiniband/hw/qib/qib_iba6120.c |    2 +-
 drivers/infiniband/hw/qib/qib_iba7220.c |    2 +-
 drivers/infiniband/hw/qib/qib_iba7322.c |   10 +++++++---
 5 files changed, 12 insertions(+), 8 deletions(-)

diff --git a/drivers/infiniband/hw/qib/qib.h b/drivers/infiniband/hw/qib/qib.h
index 64c9e7d..73225ee 100644
--- a/drivers/infiniband/hw/qib/qib.h
+++ b/drivers/infiniband/hw/qib/qib.h
@@ -766,7 +766,7 @@ struct qib_devdata {
 	void (*f_sdma_hw_start_up)(struct qib_pportdata *);
 	void (*f_sdma_init_early)(struct qib_pportdata *);
 	void (*f_set_cntr_sample)(struct qib_pportdata *, u32, u32);
-	void (*f_update_usrhead)(struct qib_ctxtdata *, u64, u32, u32);
+	void (*f_update_usrhead)(struct qib_ctxtdata *, u64, u32, u32, u32);
 	u32 (*f_hdrqempty)(struct qib_ctxtdata *);
 	u64 (*f_portcntr)(struct qib_pportdata *, u32);
 	u32 (*f_read_cntrs)(struct qib_devdata *, loff_t, char **,
diff --git a/drivers/infiniband/hw/qib/qib_driver.c b/drivers/infiniband/hw/qib/qib_driver.c
index 3ed0d5a..816a6bd 100644
--- a/drivers/infiniband/hw/qib/qib_driver.c
+++ b/drivers/infiniband/hw/qib/qib_driver.c
@@ -410,7 +410,7 @@ move_along:
 		 */
 		lval = l;
 		if (!last && !(i & 0xf)) {
-			dd->f_update_usrhead(rcd, lval, updegr, etail);
+			dd->f_update_usrhead(rcd, lval, updegr, etail, i);
 			updegr = 0;
 		}
 	}
@@ -452,7 +452,7 @@ bail:
 	 * if no packets were processed.
 	 */
 	lval = (u64)rcd->head | dd->rhdrhead_intr_off;
-	dd->f_update_usrhead(rcd, lval, updegr, etail);
+	dd->f_update_usrhead(rcd, lval, updegr, etail, i);
 	return crcs;
 }
 
diff --git a/drivers/infiniband/hw/qib/qib_iba6120.c b/drivers/infiniband/hw/qib/qib_iba6120.c
index a5e29db..774dea8 100644
--- a/drivers/infiniband/hw/qib/qib_iba6120.c
+++ b/drivers/infiniband/hw/qib/qib_iba6120.c
@@ -2074,7 +2074,7 @@ static void qib_6120_config_ctxts(struct qib_devdata *dd)
 }
 
 static void qib_update_6120_usrhead(struct qib_ctxtdata *rcd, u64 hd,
-				    u32 updegr, u32 egrhd)
+				    u32 updegr, u32 egrhd, u32 npkts)
 {
 	qib_write_ureg(rcd->dd, ur_rcvhdrhead, hd, rcd->ctxt);
 	if (updegr)
diff --git a/drivers/infiniband/hw/qib/qib_iba7220.c b/drivers/infiniband/hw/qib/qib_iba7220.c
index 6fd8d74..df49e8e 100644
--- a/drivers/infiniband/hw/qib/qib_iba7220.c
+++ b/drivers/infiniband/hw/qib/qib_iba7220.c
@@ -2703,7 +2703,7 @@ static int qib_7220_set_loopback(struct qib_pportdata *ppd, const char *what)
 }
 
 static void qib_update_7220_usrhead(struct qib_ctxtdata *rcd, u64 hd,
-				    u32 updegr, u32 egrhd)
+				    u32 updegr, u32 egrhd, u32 npkts)
 {
 	qib_write_ureg(rcd->dd, ur_rcvhdrhead, hd, rcd->ctxt);
 	if (updegr)
diff --git a/drivers/infiniband/hw/qib/qib_iba7322.c b/drivers/infiniband/hw/qib/qib_iba7322.c
index a9c8c72..9bc6d08 100644
--- a/drivers/infiniband/hw/qib/qib_iba7322.c
+++ b/drivers/infiniband/hw/qib/qib_iba7322.c
@@ -2823,7 +2823,6 @@ static irqreturn_t qib_7322intr(int irq, void *data)
 				ctxtrbits &= ~rmask;
 				if (dd->rcd[i]) {
 					qib_kreceive(dd->rcd[i], NULL, &npkts);
-					adjust_rcv_timeout(dd->rcd[i], npkts);
 				}
 			}
 			rmask <<= 1;
@@ -2873,7 +2872,6 @@ static irqreturn_t qib_7322pintr(int irq, void *data)
 		       (1ULL << QIB_I_RCVURG_LSB)) << rcd->ctxt);
 
 	qib_kreceive(rcd, NULL, &npkts);
-	adjust_rcv_timeout(rcd, npkts);
 
 	return IRQ_HANDLED;
 }
@@ -4047,8 +4045,14 @@ static int qib_7322_set_ib_table(struct qib_pportdata *ppd, int which, void *t)
 }
 
 static void qib_update_7322_usrhead(struct qib_ctxtdata *rcd, u64 hd,
-				    u32 updegr, u32 egrhd)
+				    u32 updegr, u32 egrhd, u32 npkts)
 {
+	/*
+	 * Need to write timeout register before updating rcvhdrhead to ensure
+	 * that the timer is enabled on reception of a packet.
+	 */
+	if (hd >> IBA7322_HDRHEAD_PKTINT_SHIFT)
+		adjust_rcv_timeout(rcd, npkts);
 	qib_write_ureg(rcd->dd, ur_rcvhdrhead, hd, rcd->ctxt);
 	qib_write_ureg(rcd->dd, ur_rcvhdrhead, hd, rcd->ctxt);
 	if (updegr)

--
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 17/24] IB/qib: change rcv queue/qpn selection

[Mike Marciniszyn]

The basic idea is that on SusieQ, the difficult part of mapping QPN to context
is handled by the mapping registers so the generic QPN allocation doesn't need
to worry about chip specifics. For Monty and Linda, there is no mapping table
so the qpt->mask (same as dd->qpn_mask), is used to see if the QPN to context
falls within [zero..dd->n_krcv_queues).

Signed-off-by: Mike Marciniszyn <mike.marciniszyn-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/qib/qib_iba7220.c |    2 +-
 drivers/infiniband/hw/qib/qib_iba7322.c |    8 ++------
 drivers/infiniband/hw/qib/qib_qp.c      |   26 ++++++++++----------------
 drivers/infiniband/hw/qib/qib_verbs.h   |   10 ++--------
 4 files changed, 15 insertions(+), 31 deletions(-)

diff --git a/drivers/infiniband/hw/qib/qib_iba7220.c b/drivers/infiniband/hw/qib/qib_iba7220.c
index df49e8e..127a0d5 100644
--- a/drivers/infiniband/hw/qib/qib_iba7220.c
+++ b/drivers/infiniband/hw/qib/qib_iba7220.c
@@ -2297,7 +2297,7 @@ static void qib_7220_config_ctxts(struct qib_devdata *dd)
 	nchipctxts = qib_read_kreg32(dd, kr_portcnt);
 	dd->cspec->numctxts = nchipctxts;
 	if (qib_n_krcv_queues > 1) {
-		dd->qpn_mask = 0x3f;
+		dd->qpn_mask = 0x3e;
 		dd->first_user_ctxt = qib_n_krcv_queues * dd->num_pports;
 		if (dd->first_user_ctxt > nchipctxts)
 			dd->first_user_ctxt = nchipctxts;
diff --git a/drivers/infiniband/hw/qib/qib_iba7322.c b/drivers/infiniband/hw/qib/qib_iba7322.c
index 9bc6d08..d3b4938 100644
--- a/drivers/infiniband/hw/qib/qib_iba7322.c
+++ b/drivers/infiniband/hw/qib/qib_iba7322.c
@@ -3515,11 +3515,6 @@ static void qib_7322_config_ctxts(struct qib_devdata *dd)
 	nchipctxts = qib_read_kreg32(dd, kr_contextcnt);
 	dd->cspec->numctxts = nchipctxts;
 	if (qib_n_krcv_queues > 1 && dd->num_pports) {
-		/*
-		 * Set the mask for which bits from the QPN are used
-		 * to select a context number.
-		 */
-		dd->qpn_mask = 0x3f;
 		dd->first_user_ctxt = NUM_IB_PORTS +
 			(qib_n_krcv_queues - 1) * dd->num_pports;
 		if (dd->first_user_ctxt > nchipctxts)
@@ -5865,7 +5860,8 @@ static void write_7322_initregs(struct qib_devdata *dd)
 		unsigned n, regno;
 		unsigned long flags;
 
-		if (!dd->qpn_mask || !dd->pport[pidx].link_speed_supported)
+		if (dd->n_krcv_queues < 2 ||
+			!dd->pport[pidx].link_speed_supported)
 			continue;
 
 		ppd = &dd->pport[pidx];
diff --git a/drivers/infiniband/hw/qib/qib_qp.c b/drivers/infiniband/hw/qib/qib_qp.c
index 32dacd4..eaab008 100644
--- a/drivers/infiniband/hw/qib/qib_qp.c
+++ b/drivers/infiniband/hw/qib/qib_qp.c
@@ -48,13 +48,12 @@ static inline unsigned mk_qpn(struct qib_qpn_table *qpt,
 
 static inline unsigned find_next_offset(struct qib_qpn_table *qpt,
 					struct qpn_map *map, unsigned off,
-					unsigned r)
+					unsigned n)
 {
 	if (qpt->mask) {
 		off++;
-		if ((off & qpt->mask) >> 1 != r)
-			off = ((off & qpt->mask) ?
-				(off | qpt->mask) + 1 : off) | (r << 1);
+		if (((off & qpt->mask) >> 1) >= n)
+			off = (off | qpt->mask) + 2;
 	} else
 		off = find_next_zero_bit(map->page, BITS_PER_PAGE, off);
 	return off;
@@ -123,7 +122,6 @@ static int alloc_qpn(struct qib_devdata *dd, struct qib_qpn_table *qpt,
 	u32 i, offset, max_scan, qpn;
 	struct qpn_map *map;
 	u32 ret;
-	int r;
 
 	if (type == IB_QPT_SMI || type == IB_QPT_GSI) {
 		unsigned n;
@@ -139,15 +137,11 @@ static int alloc_qpn(struct qib_devdata *dd, struct qib_qpn_table *qpt,
 		goto bail;
 	}
 
-	r = smp_processor_id();
-	if (r >= dd->n_krcv_queues)
-		r %= dd->n_krcv_queues;
 	qpn = qpt->last + 1;
 	if (qpn >= QPN_MAX)
 		qpn = 2;
-	if (qpt->mask && ((qpn & qpt->mask) >> 1) != r)
-		qpn = ((qpn & qpt->mask) ? (qpn | qpt->mask) + 1 : qpn) |
-			(r << 1);
+	if (qpt->mask && ((qpn & qpt->mask) >> 1) >= dd->n_krcv_queues)
+		qpn = (qpn | qpt->mask) + 2;
 	offset = qpn & BITS_PER_PAGE_MASK;
 	map = &qpt->map[qpn / BITS_PER_PAGE];
 	max_scan = qpt->nmaps - !offset;
@@ -163,7 +157,8 @@ static int alloc_qpn(struct qib_devdata *dd, struct qib_qpn_table *qpt,
 				ret = qpn;
 				goto bail;
 			}
-			offset = find_next_offset(qpt, map, offset, r);
+			offset = find_next_offset(qpt, map, offset,
+				dd->n_krcv_queues);
 			qpn = mk_qpn(qpt, map, offset);
 			/*
 			 * This test differs from alloc_pidmap().
@@ -183,13 +178,13 @@ static int alloc_qpn(struct qib_devdata *dd, struct qib_qpn_table *qpt,
 			if (qpt->nmaps == QPNMAP_ENTRIES)
 				break;
 			map = &qpt->map[qpt->nmaps++];
-			offset = qpt->mask ? (r << 1) : 0;
+			offset = 0;
 		} else if (map < &qpt->map[qpt->nmaps]) {
 			++map;
-			offset = qpt->mask ? (r << 1) : 0;
+			offset = 0;
 		} else {
 			map = &qpt->map[0];
-			offset = qpt->mask ? (r << 1) : 2;
+			offset = 2;
 		}
 		qpn = mk_qpn(qpt, map, offset);
 	}
@@ -1065,7 +1060,6 @@ struct ib_qp *qib_create_qp(struct ib_pd *ibpd,
 		}
 		qp->ibqp.qp_num = err;
 		qp->port_num = init_attr->port_num;
-		qp->processor_id = smp_processor_id();
 		qib_reset_qp(qp, init_attr->qp_type);
 		break;
 
diff --git a/drivers/infiniband/hw/qib/qib_verbs.h b/drivers/infiniband/hw/qib/qib_verbs.h
index bd57c12..a08ceab 100644
--- a/drivers/infiniband/hw/qib/qib_verbs.h
+++ b/drivers/infiniband/hw/qib/qib_verbs.h
@@ -435,7 +435,6 @@ struct qib_qp {
 	spinlock_t r_lock;      /* used for APM */
 	spinlock_t s_lock;
 	atomic_t s_dma_busy;
-	unsigned processor_id;	/* Processor ID QP is bound to */
 	u32 s_flags;
 	u32 s_cur_size;         /* size of send packet in bytes */
 	u32 s_len;              /* total length of s_sge */
@@ -813,13 +812,8 @@ extern struct workqueue_struct *qib_cq_wq;
  */
 static inline void qib_schedule_send(struct qib_qp *qp)
 {
-	if (qib_send_ok(qp)) {
-		if (qp->processor_id == smp_processor_id())
-			queue_work(qib_wq, &qp->s_work);
-		else
-			queue_work_on(qp->processor_id,
-				      qib_wq, &qp->s_work);
-	}
+	if (qib_send_ok(qp))
+		queue_work(qib_wq, &qp->s_work);
 }
 
 static inline int qib_pkey_ok(u16 pkey1, u16 pkey2)

--
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 18/24] IB/qib: adding fix missing from earlier patch

[Mike Marciniszyn]

rcv/error race fix was missing part
of a fix that makes merges impossible.

Signed-off-by: Mike Marciniszyn <mike.marciniszyn-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/qib/qib_ud.c |   55 ++++++++++++++++--------------------
 1 files changed, 24 insertions(+), 31 deletions(-)

diff --git a/drivers/infiniband/hw/qib/qib_ud.c b/drivers/infiniband/hw/qib/qib_ud.c
index a4b945d..4a51fd1 100644
--- a/drivers/infiniband/hw/qib/qib_ud.c
+++ b/drivers/infiniband/hw/qib/qib_ud.c
@@ -445,13 +445,14 @@ void qib_ud_rcv(struct qib_ibport *ibp, struct qib_ib_header *hdr,
 	qkey = be32_to_cpu(ohdr->u.ud.deth[0]);
 	src_qp = be32_to_cpu(ohdr->u.ud.deth[1]) & QIB_QPN_MASK;
 
-	/* Get the number of bytes the message was padded by. */
+	/*
+	 * Get the number of bytes the message was padded by
+	 * and drop incomplete packets.
+	 */
 	pad = (be32_to_cpu(ohdr->bth[0]) >> 20) & 3;
-	if (unlikely(tlen < (hdrsize + pad + 4))) {
-		/* Drop incomplete packets. */
-		ibp->n_pkt_drops++;
-		goto bail;
-	}
+	if (unlikely(tlen < (hdrsize + pad + 4)))
+		goto drop;
+
 	tlen -= hdrsize + pad + 4;
 
 	/*
@@ -460,10 +461,8 @@ void qib_ud_rcv(struct qib_ibport *ibp, struct qib_ib_header *hdr,
 	 */
 	if (qp->ibqp.qp_num) {
 		if (unlikely(hdr->lrh[1] == IB_LID_PERMISSIVE ||
-			     hdr->lrh[3] == IB_LID_PERMISSIVE)) {
-			ibp->n_pkt_drops++;
-			goto bail;
-		}
+			     hdr->lrh[3] == IB_LID_PERMISSIVE))
+			goto drop;
 		if (qp->ibqp.qp_num > 1) {
 			u16 pkey1, pkey2;
 
@@ -476,7 +475,7 @@ void qib_ud_rcv(struct qib_ibport *ibp, struct qib_ib_header *hdr,
 						0xF,
 					      src_qp, qp->ibqp.qp_num,
 					      hdr->lrh[3], hdr->lrh[1]);
-				goto bail;
+				return;
 			}
 		}
 		if (unlikely(qkey != qp->qkey)) {
@@ -484,30 +483,24 @@ void qib_ud_rcv(struct qib_ibport *ibp, struct qib_ib_header *hdr,
 				      (be16_to_cpu(hdr->lrh[0]) >> 4) & 0xF,
 				      src_qp, qp->ibqp.qp_num,
 				      hdr->lrh[3], hdr->lrh[1]);
-			goto bail;
+			return;
 		}
 		/* Drop invalid MAD packets (see 13.5.3.1). */
 		if (unlikely(qp->ibqp.qp_num == 1 &&
 			     (tlen != 256 ||
-			      (be16_to_cpu(hdr->lrh[0]) >> 12) == 15))) {
-			ibp->n_pkt_drops++;
-			goto bail;
-		}
+			      (be16_to_cpu(hdr->lrh[0]) >> 12) == 15)))
+			goto drop;
 	} else {
 		struct ib_smp *smp;
 
 		/* Drop invalid MAD packets (see 13.5.3.1). */
-		if (tlen != 256 || (be16_to_cpu(hdr->lrh[0]) >> 12) != 15) {
-			ibp->n_pkt_drops++;
-			goto bail;
-		}
+		if (tlen != 256 || (be16_to_cpu(hdr->lrh[0]) >> 12) != 15)
+			goto drop;
 		smp = (struct ib_smp *) data;
 		if ((hdr->lrh[1] == IB_LID_PERMISSIVE ||
 		     hdr->lrh[3] == IB_LID_PERMISSIVE) &&
-		    smp->mgmt_class != IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) {
-			ibp->n_pkt_drops++;
-			goto bail;
-		}
+		    smp->mgmt_class != IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE)
+			goto drop;
 	}
 
 	/*
@@ -523,10 +516,8 @@ void qib_ud_rcv(struct qib_ibport *ibp, struct qib_ib_header *hdr,
 	} else if (opcode == IB_OPCODE_UD_SEND_ONLY) {
 		wc.ex.imm_data = 0;
 		wc.wc_flags = 0;
-	} else {
-		ibp->n_pkt_drops++;
-		goto bail;
-	}
+	} else
+		goto drop;
 
 	/*
 	 * A GRH is expected to preceed the data even if not
@@ -556,8 +547,7 @@ void qib_ud_rcv(struct qib_ibport *ibp, struct qib_ib_header *hdr,
 	/* Silently drop packets which are too big. */
 	if (unlikely(wc.byte_len > qp->r_len)) {
 		qp->r_flags |= QIB_R_REUSE_SGE;
-		ibp->n_pkt_drops++;
-		return;
+		goto drop;
 	}
 	if (has_grh) {
 		qib_copy_sge(&qp->r_sge, &hdr->u.l.grh,
@@ -594,5 +584,8 @@ void qib_ud_rcv(struct qib_ibport *ibp, struct qib_ib_header *hdr,
 	qib_cq_enter(to_icq(qp->ibqp.recv_cq), &wc,
 		     (ohdr->bth[0] &
 			cpu_to_be32(IB_BTH_SOLICITED)) != 0);
-bail:;
+	return;
+
+drop:
+	ibp->n_pkt_drops++;
 }

--
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 18/24] IB/qib: adding fix missing from earlier patch

[Roland Dreier]

 > rcv/error race fix was missing part
 > of a fix that makes merges impossible.

Which fix?  What merges are impossible?  Is it a good thing or a bad
thing for merges to be impossible?

 - R.
--
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 19/24] IB/qib: change qpn increment

[Mike Marciniszyn]

Changing from +1 to +2 allows for better QP
distribution across receive contexts.

Signed-off-by: Mike Marciniszyn <mike.marciniszyn-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/qib/qib_qp.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/drivers/infiniband/hw/qib/qib_qp.c b/drivers/infiniband/hw/qib/qib_qp.c
index eaab008..e16751f 100644
--- a/drivers/infiniband/hw/qib/qib_qp.c
+++ b/drivers/infiniband/hw/qib/qib_qp.c
@@ -137,7 +137,7 @@ static int alloc_qpn(struct qib_devdata *dd, struct qib_qpn_table *qpt,
 		goto bail;
 	}
 
-	qpn = qpt->last + 1;
+	qpn = qpt->last + 2;
 	if (qpn >= QPN_MAX)
 		qpn = 2;
 	if (qpt->mask && ((qpn & qpt->mask) >> 1) >= dd->n_krcv_queues)

--
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 20/24] IB/qib: RDMA lkey/rkey validation is inefficient for large MRs

[Mike Marciniszyn]

The current code loops during rkey/lkey validiation to isolate the mr
for the RDMA, which is expensive when the current operation is
inside a very large memory region.

This fix optimizes rkey/lkey validation routines for user memory regions and
fast memory regions.  The mr entry can be isolated by shifts/mods instead of
looping.  The existing loop is preserved for phys memory regions for now.

Signed-off-by: Mike Marciniszyn <mike.marciniszyn-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/qib/qib_keys.c  |   74 ++++++++++++++++++++++++---------
 drivers/infiniband/hw/qib/qib_mr.c    |    8 ++--
 drivers/infiniband/hw/qib/qib_verbs.h |    1 
 3 files changed, 60 insertions(+), 23 deletions(-)

diff --git a/drivers/infiniband/hw/qib/qib_keys.c b/drivers/infiniband/hw/qib/qib_keys.c
index 4b80eb1..756d160 100644
--- a/drivers/infiniband/hw/qib/qib_keys.c
+++ b/drivers/infiniband/hw/qib/qib_keys.c
@@ -158,31 +158,47 @@ int qib_lkey_ok(struct qib_lkey_table *rkt, struct qib_pd *pd,
 		isge->sge_length = sge->length;
 		isge->m = 0;
 		isge->n = 0;
+		spin_unlock_irqrestore(&rkt->lock, flags);
 		goto ok;
 	}
 	mr = rkt->table[(sge->lkey >> (32 - ib_qib_lkey_table_size))];
 	if (unlikely(mr == NULL || mr->lkey != sge->lkey ||
 		     mr->pd != &pd->ibpd))
 		goto bail;
+	atomic_inc(&mr->refcount);
+	spin_unlock_irqrestore(&rkt->lock, flags);
 
 	off = sge->addr - mr->user_base;
 	if (unlikely(sge->addr < mr->user_base ||
 		     off + sge->length > mr->length ||
 		     (mr->access_flags & acc) != acc))
-		goto bail;
+		return ret;
 
 	off += mr->offset;
-	m = 0;
-	n = 0;
-	while (off >= mr->map[m]->segs[n].length) {
-		off -= mr->map[m]->segs[n].length;
-		n++;
-		if (n >= QIB_SEGSZ) {
-			m++;
-			n = 0;
+	if (mr->page_shift) {
+		/*
+		page sizes are uniform power of 2 so no loop is necessary
+		entries_spanned_by_off is the number of times the loop below
+		would have executed.
+		*/
+		size_t entries_spanned_by_off;
+
+		entries_spanned_by_off = off >> mr->page_shift;
+		off -= (entries_spanned_by_off << mr->page_shift);
+		m = entries_spanned_by_off/QIB_SEGSZ;
+		n = entries_spanned_by_off%QIB_SEGSZ;
+	} else {
+		m = 0;
+		n = 0;
+		while (off >= mr->map[m]->segs[n].length) {
+			off -= mr->map[m]->segs[n].length;
+			n++;
+			if (n >= QIB_SEGSZ) {
+				m++;
+				n = 0;
+			}
 		}
 	}
-	atomic_inc(&mr->refcount);
 	isge->mr = mr;
 	isge->vaddr = mr->map[m]->segs[n].vaddr + off;
 	isge->length = mr->map[m]->segs[n].length - off;
@@ -191,6 +207,7 @@ int qib_lkey_ok(struct qib_lkey_table *rkt, struct qib_pd *pd,
 	isge->n = n;
 ok:
 	ret = 1;
+	return ret;
 bail:
 	spin_unlock_irqrestore(&rkt->lock, flags);
 	return ret;
@@ -237,30 +254,46 @@ int qib_rkey_ok(struct qib_qp *qp, struct qib_sge *sge,
 		sge->sge_length = len;
 		sge->m = 0;
 		sge->n = 0;
+		spin_unlock_irqrestore(&rkt->lock, flags);
 		goto ok;
 	}
 
 	mr = rkt->table[(rkey >> (32 - ib_qib_lkey_table_size))];
 	if (unlikely(mr == NULL || mr->lkey != rkey || qp->ibqp.pd != mr->pd))
 		goto bail;
+	atomic_inc(&mr->refcount);
+	spin_unlock_irqrestore(&rkt->lock, flags);
 
 	off = vaddr - mr->iova;
 	if (unlikely(vaddr < mr->iova || off + len > mr->length ||
 		     (mr->access_flags & acc) == 0))
-		goto bail;
+		return ret;
 
 	off += mr->offset;
-	m = 0;
-	n = 0;
-	while (off >= mr->map[m]->segs[n].length) {
-		off -= mr->map[m]->segs[n].length;
-		n++;
-		if (n >= QIB_SEGSZ) {
-			m++;
-			n = 0;
+	if (mr->page_shift) {
+		/*
+		page sizes are uniform power of 2 so no loop is necessary
+		entries_spanned_by_off is the number of times the loop below
+		would have executed.
+		*/
+		size_t entries_spanned_by_off;
+
+		entries_spanned_by_off = off >> mr->page_shift;
+		off -= (entries_spanned_by_off << mr->page_shift);
+		m = entries_spanned_by_off/QIB_SEGSZ;
+		n = entries_spanned_by_off%QIB_SEGSZ;
+	} else {
+		m = 0;
+		n = 0;
+		while (off >= mr->map[m]->segs[n].length) {
+			off -= mr->map[m]->segs[n].length;
+			n++;
+			if (n >= QIB_SEGSZ) {
+				m++;
+				n = 0;
+			}
 		}
 	}
-	atomic_inc(&mr->refcount);
 	sge->mr = mr;
 	sge->vaddr = mr->map[m]->segs[n].vaddr + off;
 	sge->length = mr->map[m]->segs[n].length - off;
@@ -269,6 +302,7 @@ int qib_rkey_ok(struct qib_qp *qp, struct qib_sge *sge,
 	sge->n = n;
 ok:
 	ret = 1;
+	return ret;
 bail:
 	spin_unlock_irqrestore(&rkt->lock, flags);
 	return ret;
diff --git a/drivers/infiniband/hw/qib/qib_mr.c b/drivers/infiniband/hw/qib/qib_mr.c
index 5f95f0f..08944e2 100644
--- a/drivers/infiniband/hw/qib/qib_mr.c
+++ b/drivers/infiniband/hw/qib/qib_mr.c
@@ -39,7 +39,6 @@
 /* Fast memory region */
 struct qib_fmr {
 	struct ib_fmr ibfmr;
-	u8 page_shift;
 	struct qib_mregion mr;        /* must be last */
 };
 
@@ -107,6 +106,7 @@ static struct qib_mr *alloc_mr(int count, struct qib_lkey_table *lk_table)
 			goto bail;
 	}
 	mr->mr.mapsz = m;
+	mr->mr.page_shift = 0;
 	mr->mr.max_segs = count;
 
 	/*
@@ -231,6 +231,8 @@ struct ib_mr *qib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 	mr->mr.access_flags = mr_access_flags;
 	mr->umem = umem;
 
+	if (is_power_of_2(umem->page_size))
+		mr->mr.page_shift = ilog2(umem->page_size);
 	m = 0;
 	n = 0;
 	list_for_each_entry(chunk, &umem->chunk_list, list) {
@@ -390,7 +392,7 @@ struct ib_fmr *qib_alloc_fmr(struct ib_pd *pd, int mr_access_flags,
 	fmr->mr.offset = 0;
 	fmr->mr.access_flags = mr_access_flags;
 	fmr->mr.max_segs = fmr_attr->max_pages;
-	fmr->page_shift = fmr_attr->page_shift;
+	fmr->mr.page_shift = fmr_attr->page_shift;
 
 	atomic_set(&fmr->mr.refcount, 0);
 	ret = &fmr->ibfmr;
@@ -437,7 +439,7 @@ int qib_map_phys_fmr(struct ib_fmr *ibfmr, u64 *page_list,
 	spin_lock_irqsave(&rkt->lock, flags);
 	fmr->mr.user_base = iova;
 	fmr->mr.iova = iova;
-	ps = 1 << fmr->page_shift;
+	ps = 1 << fmr->mr.page_shift;
 	fmr->mr.length = list_len * ps;
 	m = 0;
 	n = 0;
diff --git a/drivers/infiniband/hw/qib/qib_verbs.h b/drivers/infiniband/hw/qib/qib_verbs.h
index a08ceab..63b22a9 100644
--- a/drivers/infiniband/hw/qib/qib_verbs.h
+++ b/drivers/infiniband/hw/qib/qib_verbs.h
@@ -301,6 +301,7 @@ struct qib_mregion {
 	int access_flags;
 	u32 max_segs;           /* number of qib_segs in all the arrays */
 	u32 mapsz;              /* size of the map array */
+	u8  page_shift;         /* 0 - non unform/non powerof2 sizes */
 	atomic_t refcount;
 	struct qib_segarray *map[0];    /* the segments */
 };

--
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 21/24] IB/qib: Issue pre-emptive NAKs on eager buffer overflow

[Mike Marciniszyn]

Under congestion resulting in eager buffer overflow attempt to send
pre-emptive NAKs if header queue entries with TID errors are generated and
a valid header is present. This prevents long timeouts and flow restarts if
a trailing set of packets are dropped due to eager overflows. Pre-emptive
NAKs are currently only supported for RDMA Writes.

Signed-off-by: Mike Marciniszyn <mike.marciniszyn-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/qib/qib_driver.c |  141 +++++++++++++++++++++++++++++++-
 1 files changed, 137 insertions(+), 4 deletions(-)

diff --git a/drivers/infiniband/hw/qib/qib_driver.c b/drivers/infiniband/hw/qib/qib_driver.c
index 816a6bd..23e584f 100644
--- a/drivers/infiniband/hw/qib/qib_driver.c
+++ b/drivers/infiniband/hw/qib/qib_driver.c
@@ -289,14 +289,147 @@ static inline void *qib_get_egrbuf(const struct qib_ctxtdata *rcd, u32 etail)
  * Returns 1 if error was a CRC, else 0.
  * Needed for some chip's synthesized error counters.
  */
-static u32 qib_rcv_hdrerr(struct qib_pportdata *ppd, u32 ctxt,
-			  u32 eflags, u32 l, u32 etail, __le32 *rhf_addr,
-			  struct qib_message_header *hdr)
+static u32 qib_rcv_hdrerr(struct qib_ctxtdata *rcd, struct qib_pportdata *ppd,
+			  u32 ctxt, u32 eflags, u32 l, u32 etail,
+			  __le32 *rhf_addr, struct qib_message_header *rhdr)
 {
 	u32 ret = 0;
 
 	if (eflags & (QLOGIC_IB_RHF_H_ICRCERR | QLOGIC_IB_RHF_H_VCRCERR))
 		ret = 1;
+	else if (eflags == QLOGIC_IB_RHF_H_TIDERR) {
+		/* For TIDERR and RC QPs premptively schedule a NAK */
+		struct qib_ib_header *hdr = (struct qib_ib_header *) rhdr;
+		struct qib_other_headers *ohdr = NULL;
+		struct qib_ibport *ibp = &ppd->ibport_data;
+		struct qib_qp *qp = NULL;
+		u32 tlen = qib_hdrget_length_in_bytes(rhf_addr);
+		u16 lid  = be16_to_cpu(hdr->lrh[1]);
+		int lnh = be16_to_cpu(hdr->lrh[0]) & 3;
+		u32 qp_num;
+		u32 opcode;
+		u32 psn;
+		int diff;
+		unsigned long flags;
+
+		/* Sanity check packet */
+		if (tlen < 24)
+			goto drop;
+
+		if (lid < QIB_MULTICAST_LID_BASE) {
+			lid &= ~((1 << ppd->lmc) - 1);
+			if (unlikely(lid != ppd->lid))
+				goto drop;
+		}
+
+		/* Check for GRH */
+		if (lnh == QIB_LRH_BTH)
+			ohdr = &hdr->u.oth;
+		else if (lnh == QIB_LRH_GRH) {
+			u32 vtf;
+
+			ohdr = &hdr->u.l.oth;
+			if (hdr->u.l.grh.next_hdr != IB_GRH_NEXT_HDR)
+				goto drop;
+			vtf = be32_to_cpu(hdr->u.l.grh.version_tclass_flow);
+			if ((vtf >> IB_GRH_VERSION_SHIFT) != IB_GRH_VERSION)
+				goto drop;
+		} else
+			goto drop;
+
+		/* Get opcode and PSN from packet */
+		opcode = be32_to_cpu(ohdr->bth[0]);
+		opcode >>= 24;
+		psn = be32_to_cpu(ohdr->bth[2]);
+
+		/* Get the destination QP number. */
+		qp_num = be32_to_cpu(ohdr->bth[1]) & QIB_QPN_MASK;
+		if (qp_num != QIB_MULTICAST_QPN) {
+			int ruc_res;
+			qp = qib_lookup_qpn(ibp, qp_num);
+			if (!qp)
+				goto drop;
+
+			/*
+			 * Handle only RC QPs - for other QP types drop error
+			 * packet.
+			 */
+			spin_lock(&qp->r_lock);
+
+			/* Check for valid receive state. */
+			if (!(ib_qib_state_ops[qp->state] &
+			      QIB_PROCESS_RECV_OK)) {
+				ibp->n_pkt_drops++;
+				goto unlock;
+			}
+
+			switch (qp->ibqp.qp_type) {
+			case IB_QPT_RC:
+				spin_lock_irqsave(&qp->s_lock, flags);
+				ruc_res =
+					qib_ruc_check_hdr(
+						ibp, hdr,
+						lnh == QIB_LRH_GRH,
+						qp,
+						be32_to_cpu(ohdr->bth[0]));
+				if (ruc_res) {
+					spin_unlock_irqrestore(&qp->s_lock,
+							       flags);
+					goto unlock;
+				}
+				spin_unlock_irqrestore(&qp->s_lock, flags);
+
+				/* Only deal with RDMA Writes for now */
+				if (opcode <
+				    IB_OPCODE_RC_RDMA_READ_RESPONSE_FIRST) {
+					diff = qib_cmp24(psn, qp->r_psn);
+					if (!qp->r_nak_state && diff >= 0) {
+						ibp->n_rc_seqnak++;
+						qp->r_nak_state =
+							IB_NAK_PSN_ERROR;
+						/* Use the expected PSN. */
+						qp->r_ack_psn = qp->r_psn;
+						/*
+						 * Wait to send the sequence
+						 * NAK until all packets
+						 * in the receive queue have
+						 * been processed.
+						 * Otherwise, we end up
+						 * propagating congestion.
+						 */
+						if (list_empty(&qp->rspwait)) {
+							qp->r_flags |=
+								QIB_R_RSP_NAK;
+							atomic_inc(
+								&qp->refcount);
+							list_add_tail(
+							 &qp->rspwait,
+							 &rcd->qp_wait_list);
+						}
+					} /* Out of sequence NAK */
+				} /* QP Request NAKs */
+				break;
+			case IB_QPT_SMI:
+			case IB_QPT_GSI:
+			case IB_QPT_UD:
+			case IB_QPT_UC:
+			default:
+				/* For now don't handle any other QP types */
+				break;
+			}
+
+unlock:
+			spin_unlock(&qp->r_lock);
+			/*
+			 * Notify qib_destroy_qp() if it is waiting
+			 * for us to finish.
+			 */
+			if (atomic_dec_and_test(&qp->refcount))
+				wake_up(&qp->wait);
+		} /* Unicast QP */
+	} /* Valid packet with TIDErr */
+
+drop:
 	return ret;
 }
 
@@ -376,7 +509,7 @@ u32 qib_kreceive(struct qib_ctxtdata *rcd, u32 *llic, u32 *npkts)
 		 * packets; only qibhdrerr should be set.
 		 */
 		if (unlikely(eflags))
-			crcs += qib_rcv_hdrerr(ppd, rcd->ctxt, eflags, l,
+			crcs += qib_rcv_hdrerr(rcd, ppd, rcd->ctxt, eflags, l,
 					       etail, rhf_addr, hdr);
 		else if (etype == RCVHQ_RCV_TYPE_NON_KD) {
 			qib_ib_rcv(rcd, hdr, ebuf, tlen);

--
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 22/24] IB/qib: Un-necessary delayed completions on RC connection

[Mike Marciniszyn]

Currently on receipt of a response message (ACKs, RDMA Response, Atomic
Responses etc.) if the SDMA completion counter is not advanced the
driver delays the completion of the WQE. In most cases this is overly
pessimistic as the response (ACK) to a previously transmitted send
implies that the send is complete. Ensure that SDMA queue is progressed
appropriately before determining if a send has delayed completions.

Signed-off-by: Mike Marciniszyn <mike.marciniszyn-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/qib/qib_rc.c |   24 ++++++++++++++++++++++++
 1 files changed, 24 insertions(+), 0 deletions(-)

diff --git a/drivers/infiniband/hw/qib/qib_rc.c b/drivers/infiniband/hw/qib/qib_rc.c
index 955fb71..8245237 100644
--- a/drivers/infiniband/hw/qib/qib_rc.c
+++ b/drivers/infiniband/hw/qib/qib_rc.c
@@ -1407,6 +1407,7 @@ static void qib_rc_rcv_resp(struct qib_ibport *ibp,
 			    struct qib_ctxtdata *rcd)
 {
 	struct qib_swqe *wqe;
+	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
 	enum ib_wc_status status;
 	unsigned long flags;
 	int diff;
@@ -1414,6 +1415,29 @@ static void qib_rc_rcv_resp(struct qib_ibport *ibp,
 	u32 aeth;
 	u64 val;
 
+	if (opcode != OP(RDMA_READ_RESPONSE_MIDDLE)) {
+		/*
+		 * If ACK'd PSN on SDMA busy list try to make progress to
+		 * reclaim SDMA credits.
+		 */
+		if ((qib_cmp24(psn, qp->s_sending_psn) >= 0) &&
+		    (qib_cmp24(qp->s_sending_psn, qp->s_sending_hpsn) <= 0)) {
+
+			/*
+			 * If send tasklet not running attempt to progress
+			 * SDMA queue.
+			 */
+			if (!(qp->s_flags & QIB_S_BUSY)) {
+				/* Acquire SDMA Lock */
+				spin_lock_irqsave(&ppd->sdma_lock, flags);
+				/* Invoke sdma make progress */
+				qib_sdma_make_progress(ppd);
+				/* Release SDMA Lock */
+				spin_unlock_irqrestore(&ppd->sdma_lock, flags);
+			}
+		}
+	}
+
 	spin_lock_irqsave(&qp->s_lock, flags);
 
 	/* Ignore invalid responses. */

--
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 23/24] IB/qib: Add a delay to allow SERDES to initialize

[Mike Marciniszyn]

Adding a small delay would allow the SERDES to adapt better,
which would improve the link negotiation.

Signed-off-by: Mike Marciniszyn <mike.marciniszyn-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/qib/qib_iba7322.c |    2 ++
 1 files changed, 2 insertions(+), 0 deletions(-)

diff --git a/drivers/infiniband/hw/qib/qib_iba7322.c b/drivers/infiniband/hw/qib/qib_iba7322.c
index d3b4938..0ab4b72 100644
--- a/drivers/infiniband/hw/qib/qib_iba7322.c
+++ b/drivers/infiniband/hw/qib/qib_iba7322.c
@@ -7596,6 +7596,8 @@ static int serdes_7322_init_new(struct qib_pportdata *ppd)
 	/* VGA output common mode */
 	ibsd_wr_allchans(ppd, 12, (3 << 2), BMASK(3, 2));
 
+	/* Add a delay to allow the SERDES to init */
+	msleep(3000);
 	return 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 24/24] IB/qib: Improve SERDES tunning on QMH boards

[Mike Marciniszyn]

Improve the QMH SERDES tunning on initial driver load by
removing the delay after SERDES init and having the driver
go through a link state change.

Signed-off-by: Mike Marciniszyn <mike.marciniszyn-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/qib/qib_iba7322.c |   28 +++++++++++++++++++++++-----
 1 files changed, 23 insertions(+), 5 deletions(-)

diff --git a/drivers/infiniband/hw/qib/qib_iba7322.c b/drivers/infiniband/hw/qib/qib_iba7322.c
index 0ab4b72..dbbb0e8 100644
--- a/drivers/infiniband/hw/qib/qib_iba7322.c
+++ b/drivers/infiniband/hw/qib/qib_iba7322.c
@@ -623,6 +623,7 @@ struct qib_chippport_specific {
 	u8 ibmalfusesnap;
 	struct qib_qsfp_data qsfp_data;
 	char epmsgbuf[192]; /* for port error interrupt msg buffer */
+	u8 bounced;
 };
 
 static struct {
@@ -1742,6 +1743,8 @@ static void handle_serdes_issues(struct qib_pportdata *ppd, u64 ibcst)
 	}
 }
 
+static int qib_7322_set_ib_cfg(struct qib_pportdata *, int, u32);
+
 /*
  * This is per-pport error handling.
  * will likely get it's own MSIx interrupt (one for each port,
@@ -1878,7 +1881,23 @@ static noinline void handle_7322_p_errors(struct qib_pportdata *ppd)
 		    IB_PHYSPORTSTATE_DISABLED)
 			qib_set_ib_7322_lstate(ppd, 0,
 			       QLOGIC_IB_IBCC_LINKINITCMD_DISABLE);
-		else
+		else {
+			u32 lstate;
+			/*
+			 * We need the current logical link state before
+			 * lflags are set in handle_e_ibstatuschanged.
+			 */
+			lstate = qib_7322_iblink_state(ibcs);
+
+			if (IS_QMH(dd) && !ppd->cpspec->bounced &&
+			    ltstate == IB_PHYSPORTSTATE_LINKUP &&
+			    (lstate >= IB_PORT_INIT &&
+				lstate <= IB_PORT_ACTIVE)) {
+				ppd->cpspec->bounced = 1;
+				qib_7322_set_ib_cfg(ppd, QIB_IB_CFG_LSTATE,
+					IB_LINKCMD_DOWN | IB_LINKINITCMD_POLL);
+			}
+
 			/*
 			 * Since going into a recovery state causes the link
 			 * state to go down and since recovery is transitory,
@@ -1892,6 +1911,7 @@ static noinline void handle_7322_p_errors(struct qib_pportdata *ppd)
 			    ltstate != IB_PHYSPORTSTATE_RECOVERY_WAITRMT &&
 			    ltstate != IB_PHYSPORTSTATE_RECOVERY_IDLE)
 				qib_handle_e_ibstatuschanged(ppd, ibcs);
+		}
 	}
 	if (*msg && iserr)
 		qib_dev_porterr(dd, ppd->port, "%s error\n", msg);
@@ -7282,8 +7302,8 @@ static void ibsd_wr_allchans(struct qib_pportdata *ppd, int addr, unsigned data,
 static void serdes_7322_los_enable(struct qib_pportdata *ppd, int enable)
 {
 	u64 data = qib_read_kreg_port(ppd, krp_serdesctrl);
-	printk(KERN_INFO QIB_DRV_NAME " Turning LOS %s for port %d\n",
-		 (enable ? "on" : "off"), ppd->port);
+	printk(KERN_INFO QIB_DRV_NAME " IB%u:%u Turning LOS %s\n",
+		ppd->dd->unit, ppd->port, (enable ? "on" : "off"));
 	if (enable)
 		data |= SYM_MASK(IBSerdesCtrl_0, RXLOSEN);
 	else
@@ -7596,8 +7616,6 @@ static int serdes_7322_init_new(struct qib_pportdata *ppd)
 	/* VGA output common mode */
 	ibsd_wr_allchans(ppd, 12, (3 << 2), BMASK(3, 2));
 
-	/* Add a delay to allow the SERDES to init */
-	msleep(3000);
 	return 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 24/24] IB/qib: Improve SERDES tunning on QMH boards

[Roland Dreier]

 > @@ -7596,8 +7616,6 @@ static int serdes_7322_init_new(struct qib_pportdata *ppd)
 >  	/* VGA output common mode */
 >  	ibsd_wr_allchans(ppd, 12, (3 << 2), BMASK(3, 2));
 >  
 > -	/* Add a delay to allow the SERDES to init */
 > -	msleep(3000);
 >  	return 0;
 >  }

Am I confused or is this chunk completely reverting patch 23?  Is there
any reason you want that patch in the kernel history?

 - R.
--
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 24/24] IB/qib: Improve SERDES tunning on QMH boards

[Mike Marciniszyn]

I noticed that, but was concerned with conflicts, which there are none in this case.

23 and 24 can combined to get rid of the hunk below.   I have no issue with losing the history for 23.

Mike

-----Original Message-----
From: Roland Dreier [mailto:rdreier-FYB4Gu1CFyUAvxtiuMwx3w@public.gmane.org] 
Sent: Tuesday, January 04, 2011 1:55 PM
To: Mike Marciniszyn
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
Subject: Re: [PATCH 24/24] IB/qib: Improve SERDES tunning on QMH boards

 > @@ -7596,8 +7616,6 @@ static int serdes_7322_init_new(struct qib_pportdata *ppd)
 >  	/* VGA output common mode */
 >  	ibsd_wr_allchans(ppd, 12, (3 << 2), BMASK(3, 2));
 >  
 > -	/* Add a delay to allow the SERDES to init */
 > -	msleep(3000);
 >  	return 0;
 >  }

Am I confused or is this chunk completely reverting patch 23?  Is there
any reason you want that patch in the kernel history?

 - R.

--
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 0/24] various fixes for QIB driver

[Roland Dreier]

 > The following patches (ordered first to last) are for various bug fixes found
 > during internal testing.

OK, but in the future please send patches in smaller sets (say, no more
than 10 at a time).  It's hard to review and apply big patch bombs.

Also when I apply all this stuff, you will be listed as the author of
all the patches.  If there's someone else you want to give credit too,
you can put a line like

From: X Y Z <x.y.z-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>

as the first line of the body of your email.  (Since you are sending the
emails, your signed-off-by should be the last line before the "---")

 - R.
--
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 0/24] various fixes for QIB driver

[Mike Marciniszyn]

Understood on the volume.   We hadn't had any fixes for a bit, there was a considerable backlog.

Mike

-----Original Message-----
From: Roland Dreier [mailto:rdreier-FYB4Gu1CFyUAvxtiuMwx3w@public.gmane.org] 
Sent: Tuesday, January 04, 2011 1:59 PM
To: Mike Marciniszyn
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
Subject: Re: [PATCH 0/24] various fixes for QIB driver

 > The following patches (ordered first to last) are for various bug fixes found
 > during internal testing.

OK, but in the future please send patches in smaller sets (say, no more
than 10 at a time).  It's hard to review and apply big patch bombs.

Also when I apply all this stuff, you will be listed as the author of
all the patches.  If there's someone else you want to give credit too,
you can put a line like

From: X Y Z <x.y.z-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>

as the first line of the body of your email.  (Since you are sending the
emails, your signed-off-by should be the last line before the "---")

 - R.

--
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 0/24] various fixes for QIB driver

[Roland Dreier]

OK, I applied the whole list except for

IB/qib: add thresholds to VendorPortCounters PMA operation

and I collapsed

IB/qib: Add a delay to allow SERDES to initialize
IB/qib: Improve SERDES tunning on QMH boards

into one patch.
--
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 0/24] various fixes for QIB driver

[Mike Marciniszyn]

Thanks!

I'm working on a v2 patch for the lone holdout.

Mike

-----Original Message-----
From: Roland Dreier [mailto:rdreier-FYB4Gu1CFyUAvxtiuMwx3w@public.gmane.org] 
Sent: Wednesday, January 05, 2011 1:08 PM
To: Mike Marciniszyn
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
Subject: Re: [PATCH 0/24] various fixes for QIB driver

OK, I applied the whole list except for

IB/qib: add thresholds to VendorPortCounters PMA operation

and I collapsed

IB/qib: Add a delay to allow SERDES to initialize
IB/qib: Improve SERDES tunning on QMH boards

into one patch.

--
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