IB/ipath: Enable 4KB MTU

Enable use of 4KB MTU.  Since the driver uses more pinned memory for
receive buffers when the 4KB MTU is enabled, whether or not the fabric
supports that MTU, add a "mtu4096" module parameter that can be used to
limit the MTU to 2KB when it is known that 4KB MTUs can't be used
anyway.

Signed-off-by: Dave Olson <dave.olson@qlogic.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>

8 files changed, 45 insertions, 65 deletions

diff --git a/drivers/infiniband/hw/ipath/ipath_driver.c b/drivers/infiniband/hw/ipath/ipath_driver.c
index 367f2a38744e..7121fe84ff8b 100644
--- a/drivers/infiniband/hw/ipath/ipath_driver.c
+++ b/drivers/infiniband/hw/ipath/ipath_driver.c
@@ -73,6 +73,10 @@ module_param_named(debug, ipath_debug, uint, S_IWUSR | S_IRUGO);
 MODULE_PARM_DESC(debug, "mask for debug prints");
 EXPORT_SYMBOL_GPL(ipath_debug);
 
+unsigned ipath_mtu4096 = 1; /* max 4KB IB mtu by default, if supported */
+module_param_named(mtu4096, ipath_mtu4096, uint, S_IRUGO);
+MODULE_PARM_DESC(mtu4096, "enable MTU of 4096 bytes, if supported");
+
 MODULE_LICENSE("GPL");
 MODULE_AUTHOR("QLogic <support@pathscale.com>");
 MODULE_DESCRIPTION("QLogic InfiniPath driver");
@@ -1800,7 +1804,7 @@ int ipath_set_mtu(struct ipath_devdata *dd, u16 arg)
 	 * piosize).  We check that it's one of the valid IB sizes.
 	 */
 	if (arg != 256 && arg != 512 && arg != 1024 && arg != 2048 &&
-	    arg != 4096) {
+	    (arg != 4096 || !ipath_mtu4096)) {
 		ipath_dbg("Trying to set invalid mtu %u, failing\n", arg);
 		ret = -EINVAL;
 		goto bail;
@@ -1816,6 +1820,8 @@ int ipath_set_mtu(struct ipath_devdata *dd, u16 arg)
 	if (arg >= (piosize - IPATH_PIO_MAXIBHDR)) {
 		/* Only if it's not the initial value (or reset to it) */
 		if (piosize != dd->ipath_init_ibmaxlen) {
+			if (arg > piosize && arg <= dd->ipath_init_ibmaxlen)
+				piosize = dd->ipath_init_ibmaxlen;
 			dd->ipath_ibmaxlen = piosize;
 			changed = 1;
 		}
@@ -1829,24 +1835,17 @@ int ipath_set_mtu(struct ipath_devdata *dd, u16 arg)
 	}
 
 	if (changed) {
+		u64 ibc = dd->ipath_ibcctrl, ibdw;
 		/*
-		 * set the IBC maxpktlength to the size of our pio
-		 * buffers in words
+		 * update our housekeeping variables, and set IBC max
+		 * size, same as init code; max IBC is max we allow in
+		 * buffer, less the qword pbc, plus 1 for ICRC, in dwords
 		 */
-		u64 ibc = dd->ipath_ibcctrl;
+		dd->ipath_ibmaxlen = piosize - 2 * sizeof(u32);
+		ibdw = (dd->ipath_ibmaxlen >> 2) + 1;
 		ibc &= ~(INFINIPATH_IBCC_MAXPKTLEN_MASK <<
-			 INFINIPATH_IBCC_MAXPKTLEN_SHIFT);
-
-		piosize = piosize - 2 * sizeof(u32);    /* ignore pbc */
-		dd->ipath_ibmaxlen = piosize;
-		piosize /= sizeof(u32); /* in words */
-		/*
-		 * for ICRC, which we only send in diag test pkt mode, and
-		 * we don't need to worry about that for mtu
-		 */
-		piosize += 1;
-
-		ibc |= piosize << INFINIPATH_IBCC_MAXPKTLEN_SHIFT;
+			 dd->ibcc_mpl_shift);
+		ibc |= ibdw << dd->ibcc_mpl_shift;
 		dd->ipath_ibcctrl = ibc;
 		ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl,
 				 dd->ipath_ibcctrl);
diff --git a/drivers/infiniband/hw/ipath/ipath_file_ops.c b/drivers/infiniband/hw/ipath/ipath_file_ops.c
index 219b62d32eed..cddf29b9554e 100644
--- a/drivers/infiniband/hw/ipath/ipath_file_ops.c
+++ b/drivers/infiniband/hw/ipath/ipath_file_ops.c
@@ -219,7 +219,12 @@ static int ipath_get_base_info(struct file *fp,
 	kinfo->spi_pioalign = dd->ipath_palign;
 
 	kinfo->spi_qpair = IPATH_KD_QP;
-	kinfo->spi_piosize = dd->ipath_ibmaxlen;
+	/*
+	 * user mode PIO buffers are always 2KB, even when 4KB can
+	 * be received, and sent via the kernel; this is ibmaxlen
+	 * for 2K MTU.
+	 */
+	kinfo->spi_piosize = dd->ipath_piosize2k - 2 * sizeof(u32);
 	kinfo->spi_mtu = dd->ipath_ibmaxlen;	/* maxlen, not ibmtu */
 	kinfo->spi_port = pd->port_port;
 	kinfo->spi_subport = subport_fp(fp);
diff --git a/drivers/infiniband/hw/ipath/ipath_iba6120.c b/drivers/infiniband/hw/ipath/ipath_iba6120.c
index 828066e20ad7..a9fc80409e57 100644
--- a/drivers/infiniband/hw/ipath/ipath_iba6120.c
+++ b/drivers/infiniband/hw/ipath/ipath_iba6120.c
@@ -1441,17 +1441,13 @@ static int ipath_pe_early_init(struct ipath_devdata *dd)
 	dd->ipath_egrtidbase = (u64 __iomem *)
 		((char __iomem *) dd->ipath_kregbase + dd->ipath_rcvegrbase);
 
-	/*
-	 * To truly support a 4KB MTU (for usermode), we need to
-	 * bump this to a larger value.  For now, we use them for
-	 * the kernel only.
-	 */
-	dd->ipath_rcvegrbufsize = 2048;
+	dd->ipath_rcvegrbufsize = ipath_mtu4096 ? 4096 : 2048;
 	/*
 	 * the min() check here is currently a nop, but it may not always
 	 * be, depending on just how we do ipath_rcvegrbufsize
 	 */
-	dd->ipath_ibmaxlen = min(dd->ipath_piosize2k,
+	dd->ipath_ibmaxlen = min(ipath_mtu4096 ? dd->ipath_piosize4k :
+				 dd->ipath_piosize2k,
 				 dd->ipath_rcvegrbufsize +
 				 (dd->ipath_rcvhdrentsize << 2));
 	dd->ipath_init_ibmaxlen = dd->ipath_ibmaxlen;
diff --git a/drivers/infiniband/hw/ipath/ipath_init_chip.c b/drivers/infiniband/hw/ipath/ipath_init_chip.c
index 5428affa9102..f0d7848d9bb1 100644
--- a/drivers/infiniband/hw/ipath/ipath_init_chip.c
+++ b/drivers/infiniband/hw/ipath/ipath_init_chip.c
@@ -155,24 +155,13 @@ static int bringup_link(struct ipath_devdata *dd)
 			 dd->ipath_control);
 
 	/*
-	 * Note that prior to try 14 or 15 of IB, the credit scaling
-	 * wasn't working, because it was swapped for writes with the
-	 * 1 bit default linkstate field
+	 * set initial max size pkt IBC will send, including ICRC; it's the
+	 * PIO buffer size in dwords, less 1; also see ipath_set_mtu()
 	 */
+	val = (dd->ipath_ibmaxlen >> 2) + 1;
+	ibc = val << dd->ibcc_mpl_shift;
 
-	/* ignore pbc and align word */
-	val = dd->ipath_piosize2k - 2 * sizeof(u32);
-	/*
-	 * for ICRC, which we only send in diag test pkt mode, and we
-	 * don't need to worry about that for mtu
-	 */
-	val += 1;
-	/*
-	 * Set the IBC maxpktlength to the size of our pio buffers the
-	 * maxpktlength is in words.  This is *not* the IB data MTU.
-	 */
-	ibc = (val / sizeof(u32)) << INFINIPATH_IBCC_MAXPKTLEN_SHIFT;
-	/* in KB */
+	/* flowcontrolwatermark is in units of KBytes */
 	ibc |= 0x5ULL << INFINIPATH_IBCC_FLOWCTRLWATERMARK_SHIFT;
 	/*
 	 * How often flowctrl sent.  More or less in usecs; balance against
@@ -295,12 +284,9 @@ static int init_chip_first(struct ipath_devdata *dd,
 	val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_sendpiosize);
 	dd->ipath_piosize2k = val & ~0U;
 	dd->ipath_piosize4k = val >> 32;
-	/*
-	 * Note: the chips support a maximum MTU of 4096, but the driver
-	 * hasn't implemented this feature yet, so set the initial value
-	 * to 2048.
-	 */
-	dd->ipath_ibmtu = 2048;
+	if (dd->ipath_piosize4k == 0 && ipath_mtu4096)
+		ipath_mtu4096 = 0; /* 4KB not supported by this chip */
+	dd->ipath_ibmtu = ipath_mtu4096 ? 4096 : 2048;
 	val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_sendpiobufcnt);
 	dd->ipath_piobcnt2k = val & ~0U;
 	dd->ipath_piobcnt4k = val >> 32;
diff --git a/drivers/infiniband/hw/ipath/ipath_kernel.h b/drivers/infiniband/hw/ipath/ipath_kernel.h
index 59dc89516243..70c0a0dd6939 100644
--- a/drivers/infiniband/hw/ipath/ipath_kernel.h
+++ b/drivers/infiniband/hw/ipath/ipath_kernel.h
@@ -1066,6 +1066,7 @@ dma_addr_t ipath_map_single(struct pci_dev *, void *, size_t, int);
 #endif
 
 extern unsigned ipath_debug; /* debugging bit mask */
+extern unsigned ipath_mtu4096;
 
 #define IPATH_MAX_PARITY_ATTEMPTS 10000 /* max times to try recovery */
 
diff --git a/drivers/infiniband/hw/ipath/ipath_mad.c b/drivers/infiniband/hw/ipath/ipath_mad.c
index b34b91d3723a..aca876bae1c4 100644
--- a/drivers/infiniband/hw/ipath/ipath_mad.c
+++ b/drivers/infiniband/hw/ipath/ipath_mad.c
@@ -292,13 +292,9 @@ static int recv_subn_get_portinfo(struct ib_smp *smp,
 	/* pip->vl_arb_high_cap; // only one VL */
 	/* pip->vl_arb_low_cap; // only one VL */
 	/* InitTypeReply = 0 */
-	/*
-	 * Note: the chips support a maximum MTU of 4096, but the driver
-	 * hasn't implemented this feature yet, so set the maximum value
-	 * to 2048.
-	 */
-	pip->inittypereply_mtucap = IB_MTU_2048;
-	// HCAs ignore VLStallCount and HOQLife
+	/* our mtu cap depends on whether 4K MTU enabled or not */
+	pip->inittypereply_mtucap = ipath_mtu4096 ? IB_MTU_4096 : IB_MTU_2048;
+	/* HCAs ignore VLStallCount and HOQLife */
 	/* pip->vlstallcnt_hoqlife; */
 	pip->operationalvl_pei_peo_fpi_fpo = 0x10;	/* OVLs = 1 */
 	pip->mkey_violations = cpu_to_be16(dev->mkey_violations);
@@ -491,6 +487,8 @@ static int recv_subn_set_portinfo(struct ib_smp *smp,
 		mtu = 2048;
 		break;
 	case IB_MTU_4096:
+		if (!ipath_mtu4096)
+			goto err;
 		mtu = 4096;
 		break;
 	default:
diff --git a/drivers/infiniband/hw/ipath/ipath_qp.c b/drivers/infiniband/hw/ipath/ipath_qp.c
index 087ed3166479..5c8094a03643 100644
--- a/drivers/infiniband/hw/ipath/ipath_qp.c
+++ b/drivers/infiniband/hw/ipath/ipath_qp.c
@@ -516,13 +516,13 @@ int ipath_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 			goto inval;
 
 	/*
-	 * Note: the chips support a maximum MTU of 4096, but the driver
-	 * hasn't implemented this feature yet, so don't allow Path MTU
-	 * values greater than 2048.
+	 * don't allow invalid Path MTU values or greater than 2048
+	 * unless we are configured for a 4KB MTU
 	 */
-	if (attr_mask & IB_QP_PATH_MTU)
-		if (attr->path_mtu > IB_MTU_2048)
-			goto inval;
+	if ((attr_mask & IB_QP_PATH_MTU) &&
+		(ib_mtu_enum_to_int(attr->path_mtu) == -1 ||
+		(attr->path_mtu > IB_MTU_2048 && !ipath_mtu4096)))
+		goto inval;
 
 	if (attr_mask & IB_QP_PATH_MIG_STATE)
 		if (attr->path_mig_state != IB_MIG_MIGRATED &&
diff --git a/drivers/infiniband/hw/ipath/ipath_verbs.c b/drivers/infiniband/hw/ipath/ipath_verbs.c
index 32d8f882e56c..012ccb4f9a37 100644
--- a/drivers/infiniband/hw/ipath/ipath_verbs.c
+++ b/drivers/infiniband/hw/ipath/ipath_verbs.c
@@ -1201,12 +1201,7 @@ static int ipath_query_port(struct ib_device *ibdev,
 	props->max_vl_num = 1;		/* VLCap = VL0 */
 	props->init_type_reply = 0;
 
-	/*
-	 * Note: the chip supports a maximum MTU of 4096, but the driver
-	 * hasn't implemented this feature yet, so set the maximum value
-	 * to 2048.
-	 */
-	props->max_mtu = IB_MTU_2048;
+	props->max_mtu = ipath_mtu4096 ? IB_MTU_4096 : IB_MTU_2048;
 	switch (dd->ipath_ibmtu) {
 	case 4096:
 		mtu = IB_MTU_4096;