IB/ipath: simplify layering code

A lot of ipath layer code was only called in one place. Now that the
ipath_core and ib_ipath drivers are merged, it's more sensible to simply
inline the simple stuff that the layer code was doing.
Signed-off-by: Bryan O'Sullivan <bryan.osullivan@qlogic.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>

drivers/infiniband/hw/ipath/ipath_diag.c

@@ -45,7 +45,6 @@
 #include <asm/uaccess.h>
 
 #include "ipath_kernel.h"
-#include "ipath_layer.h"
 #include "ipath_common.h"
 
 int ipath_diag_inuse;

drivers/infiniband/hw/ipath/ipath_driver.c

@@ -39,7 +39,6 @@
 #include <linux/vmalloc.h>
 
 #include "ipath_kernel.h"
-#include "ipath_layer.h"
 #include "ipath_verbs.h"
 #include "ipath_common.h"
 
@@ -508,7 +507,6 @@ static int __devinit ipath_init_one(struct pci_dev *pdev,
 	ipathfs_add_device(dd);
 	ipath_user_add(dd);
 	ipath_diag_add(dd);
-	ipath_layer_add(dd);
 	ipath_register_ib_device(dd);
 
 	goto bail;
@@ -539,7 +537,6 @@ static void __devexit ipath_remove_one(struct pci_dev *pdev)
 
 	dd = pci_get_drvdata(pdev);
 	ipath_unregister_ib_device(dd->verbs_dev);
-	ipath_layer_remove(dd);
 	ipath_diag_remove(dd);
 	ipath_user_remove(dd);
 	ipathfs_remove_device(dd);
@@ -614,11 +611,12 @@ void ipath_disarm_piobufs(struct ipath_devdata *dd, unsigned first,
  *
  * wait up to msecs milliseconds for IB link state change to occur for
  * now, take the easy polling route.  Currently used only by
- * ipath_layer_set_linkstate.  Returns 0 if state reached, otherwise
+ * ipath_set_linkstate.  Returns 0 if state reached, otherwise
  * -ETIMEDOUT state can have multiple states set, for any of several
  * transitions.
  */
-int ipath_wait_linkstate(struct ipath_devdata *dd, u32 state, int msecs)
+static int ipath_wait_linkstate(struct ipath_devdata *dd, u32 state,
+				int msecs)
 {
 	dd->ipath_sma_state_wanted = state;
 	wait_event_interruptible_timeout(ipath_sma_state_wait,
@@ -814,58 +812,6 @@ bail:
 	return skb;
 }
 
-/**
- * ipath_rcv_layer - receive a packet for the layered (ethernet) driver
- * @dd: the infinipath device
- * @etail: the sk_buff number
- * @tlen: the total packet length
- * @hdr: the ethernet header
- *
- * Separate routine for better overall optimization
- */
-static void ipath_rcv_layer(struct ipath_devdata *dd, u32 etail,
-			    u32 tlen, struct ether_header *hdr)
-{
-	u32 elen;
-	u8 pad, *bthbytes;
-	struct sk_buff *skb, *nskb;
-
-	if (dd->ipath_port0_skbs &&
-			hdr->sub_opcode == IPATH_ITH4X_OPCODE_ENCAP) {
-		/*
-		 * Allocate a new sk_buff to replace the one we give
-		 * to the network stack.
-		 */
-		nskb = ipath_alloc_skb(dd, GFP_ATOMIC);
-		if (!nskb) {
-			/* count OK packets that we drop */
-			ipath_stats.sps_krdrops++;
-			return;
-		}
-
-		bthbytes = (u8 *) hdr->bth;
-		pad = (bthbytes[1] >> 4) & 3;
-		/* +CRC32 */
-		elen = tlen - (sizeof(*hdr) + pad + sizeof(u32));
-
-		skb = dd->ipath_port0_skbs[etail];
-		dd->ipath_port0_skbs[etail] = nskb;
-		skb_put(skb, elen);
-
-		dd->ipath_f_put_tid(dd, etail + (u64 __iomem *)
-				    ((char __iomem *) dd->ipath_kregbase
-				     + dd->ipath_rcvegrbase), 0,
-				    virt_to_phys(nskb->data));
-
-		__ipath_layer_rcv(dd, hdr, skb);
-
-		/* another ether packet received */
-		ipath_stats.sps_ether_rpkts++;
-	}
-	else if (hdr->sub_opcode == IPATH_ITH4X_OPCODE_LID_ARP)
-		__ipath_layer_rcv_lid(dd, hdr);
-}
-
 static void ipath_rcv_hdrerr(struct ipath_devdata *dd,
 			     u32 eflags,
 			     u32 l,
@@ -979,22 +925,17 @@ reloop:
 		if (unlikely(eflags))
 			ipath_rcv_hdrerr(dd, eflags, l, etail, rc);
 		else if (etype == RCVHQ_RCV_TYPE_NON_KD) {
-				ipath_ib_rcv(dd->verbs_dev, rc + 1, ebuf,
-					     tlen);
-				if (dd->ipath_lli_counter)
-					dd->ipath_lli_counter--;
-
-		} else if (etype == RCVHQ_RCV_TYPE_EAGER) {
-			if (qp == IPATH_KD_QP &&
-			    bthbytes[0] == ipath_layer_rcv_opcode &&
-			    ebuf)
-				ipath_rcv_layer(dd, etail, tlen,
-						(struct ether_header *)hdr);
-			else
-				ipath_cdbg(PKT, "typ %x, opcode %x (eager, "
-					   "qp=%x), len %x; ignored\n",
-					   etype, bthbytes[0], qp, tlen);
+			ipath_ib_rcv(dd->verbs_dev, rc + 1, ebuf, tlen);
+			if (dd->ipath_lli_counter)
+				dd->ipath_lli_counter--;
+			ipath_cdbg(PKT, "typ %x, opcode %x (eager, "
+				   "qp=%x), len %x; ignored\n",
+				   etype, bthbytes[0], qp, tlen);
 		}
+		else if (etype == RCVHQ_RCV_TYPE_EAGER)
+			ipath_cdbg(PKT, "typ %x, opcode %x (eager, "
+				   "qp=%x), len %x; ignored\n",
+				   etype, bthbytes[0], qp, tlen);
 		else if (etype == RCVHQ_RCV_TYPE_EXPECTED)
 			ipath_dbg("Bug: Expected TID, opcode %x; ignored\n",
 				  be32_to_cpu(hdr->bth[0]) & 0xff);
@@ -1320,13 +1261,6 @@ rescan:
 		goto bail;
 	}
 
-	if (updated)
-		/*
-		 * ran out of bufs, now some (at least this one we just
-		 * got) are now available, so tell the layered driver.
-		 */
-		__ipath_layer_intr(dd, IPATH_LAYER_INT_SEND_CONTINUE);
-
 	/*
 	 * set next starting place.  Since it's just an optimization,
 	 * it doesn't matter who wins on this, so no locking
@@ -1503,7 +1437,7 @@ int ipath_waitfor_mdio_cmdready(struct ipath_devdata *dd)
 	return ret;
 }
 
-void ipath_set_ib_lstate(struct ipath_devdata *dd, int which)
+static void ipath_set_ib_lstate(struct ipath_devdata *dd, int which)
 {
 	static const char *what[4] = {
 		[0] = "DOWN",
@@ -1537,6 +1471,180 @@ void ipath_set_ib_lstate(struct ipath_devdata *dd, int which)
 			 dd->ipath_ibcctrl | which);
 }
 
+int ipath_set_linkstate(struct ipath_devdata *dd, u8 newstate)
+{
+	u32 lstate;
+	int ret;
+
+	switch (newstate) {
+	case IPATH_IB_LINKDOWN:
+		ipath_set_ib_lstate(dd, INFINIPATH_IBCC_LINKINITCMD_POLL <<
+				    INFINIPATH_IBCC_LINKINITCMD_SHIFT);
+		/* don't wait */
+		ret = 0;
+		goto bail;
+
+	case IPATH_IB_LINKDOWN_SLEEP:
+		ipath_set_ib_lstate(dd, INFINIPATH_IBCC_LINKINITCMD_SLEEP <<
+				    INFINIPATH_IBCC_LINKINITCMD_SHIFT);
+		/* don't wait */
+		ret = 0;
+		goto bail;
+
+	case IPATH_IB_LINKDOWN_DISABLE:
+		ipath_set_ib_lstate(dd,
+				    INFINIPATH_IBCC_LINKINITCMD_DISABLE <<
+				    INFINIPATH_IBCC_LINKINITCMD_SHIFT);
+		/* don't wait */
+		ret = 0;
+		goto bail;
+
+	case IPATH_IB_LINKINIT:
+		if (dd->ipath_flags & IPATH_LINKINIT) {
+			ret = 0;
+			goto bail;
+		}
+		ipath_set_ib_lstate(dd, INFINIPATH_IBCC_LINKCMD_INIT <<
+				    INFINIPATH_IBCC_LINKCMD_SHIFT);
+		lstate = IPATH_LINKINIT;
+		break;
+
+	case IPATH_IB_LINKARM:
+		if (dd->ipath_flags & IPATH_LINKARMED) {
+			ret = 0;
+			goto bail;
+		}
+		if (!(dd->ipath_flags &
+		      (IPATH_LINKINIT | IPATH_LINKACTIVE))) {
+			ret = -EINVAL;
+			goto bail;
+		}
+		ipath_set_ib_lstate(dd, INFINIPATH_IBCC_LINKCMD_ARMED <<
+				    INFINIPATH_IBCC_LINKCMD_SHIFT);
+		/*
+		 * Since the port can transition to ACTIVE by receiving
+		 * a non VL 15 packet, wait for either state.
+		 */
+		lstate = IPATH_LINKARMED | IPATH_LINKACTIVE;
+		break;
+
+	case IPATH_IB_LINKACTIVE:
+		if (dd->ipath_flags & IPATH_LINKACTIVE) {
+			ret = 0;
+			goto bail;
+		}
+		if (!(dd->ipath_flags & IPATH_LINKARMED)) {
+			ret = -EINVAL;
+			goto bail;
+		}
+		ipath_set_ib_lstate(dd, INFINIPATH_IBCC_LINKCMD_ACTIVE <<
+				    INFINIPATH_IBCC_LINKCMD_SHIFT);
+		lstate = IPATH_LINKACTIVE;
+		break;
+
+	default:
+		ipath_dbg("Invalid linkstate 0x%x requested\n", newstate);
+		ret = -EINVAL;
+		goto bail;
+	}
+	ret = ipath_wait_linkstate(dd, lstate, 2000);
+
+bail:
+	return ret;
+}
+
+/**
+ * ipath_set_mtu - set the MTU
+ * @dd: the infinipath device
+ * @arg: the new MTU
+ *
+ * we can handle "any" incoming size, the issue here is whether we
+ * need to restrict our outgoing size.   For now, we don't do any
+ * sanity checking on this, and we don't deal with what happens to
+ * programs that are already running when the size changes.
+ * NOTE: changing the MTU will usually cause the IBC to go back to
+ * link initialize (IPATH_IBSTATE_INIT) state...
+ */
+int ipath_set_mtu(struct ipath_devdata *dd, u16 arg)
+{
+	u32 piosize;
+	int changed = 0;
+	int ret;
+
+	/*
+	 * mtu is IB data payload max.  It's the largest power of 2 less
+	 * than piosize (or even larger, since it only really controls the
+	 * largest we can receive; we can send the max of the mtu and
+	 * piosize).  We check that it's one of the valid IB sizes.
+	 */
+	if (arg != 256 && arg != 512 && arg != 1024 && arg != 2048 &&
+	    arg != 4096) {
+		ipath_dbg("Trying to set invalid mtu %u, failing\n", arg);
+		ret = -EINVAL;
+		goto bail;
+	}
+	if (dd->ipath_ibmtu == arg) {
+		ret = 0;        /* same as current */
+		goto bail;
+	}
+
+	piosize = dd->ipath_ibmaxlen;
+	dd->ipath_ibmtu = 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) {
+			dd->ipath_ibmaxlen = piosize;
+			changed = 1;
+		}
+	} else if ((arg + IPATH_PIO_MAXIBHDR) != dd->ipath_ibmaxlen) {
+		piosize = arg + IPATH_PIO_MAXIBHDR;
+		ipath_cdbg(VERBOSE, "ibmaxlen was 0x%x, setting to 0x%x "
+			   "(mtu 0x%x)\n", dd->ipath_ibmaxlen, piosize,
+			   arg);
+		dd->ipath_ibmaxlen = piosize;
+		changed = 1;
+	}
+
+	if (changed) {
+		/*
+		 * set the IBC maxpktlength to the size of our pio
+		 * buffers in words
+		 */
+		u64 ibc = dd->ipath_ibcctrl;
+		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->ipath_ibcctrl = ibc;
+		ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl,
+				 dd->ipath_ibcctrl);
+		dd->ipath_f_tidtemplate(dd);
+	}
+
+	ret = 0;
+
+bail:
+	return ret;
+}
+
+int ipath_set_lid(struct ipath_devdata *dd, u32 arg, u8 lmc)
+{
+	dd->ipath_lid = arg;
+	dd->ipath_lmc = lmc;
+
+	return 0;
+}
+
 /**
  * ipath_read_kreg64_port - read a device's per-port 64-bit kernel register
  * @dd: the infinipath device
@@ -1640,13 +1748,6 @@ void ipath_shutdown_device(struct ipath_devdata *dd)
 	ipath_set_ib_lstate(dd, INFINIPATH_IBCC_LINKINITCMD_DISABLE <<
 			    INFINIPATH_IBCC_LINKINITCMD_SHIFT);
 
-	/*
-	 * we are shutting down, so tell the layered driver.  We don't do
-	 * this on just a link state change, much like ethernet, a cable
-	 * unplug, etc. doesn't change driver state
-	 */
-	ipath_layer_intr(dd, IPATH_LAYER_INT_IF_DOWN);
-
 	/* disable IBC */
 	dd->ipath_control &= ~INFINIPATH_C_LINKENABLE;
 	ipath_write_kreg(dd, dd->ipath_kregs->kr_control,

drivers/infiniband/hw/ipath/ipath_file_ops.c

@@ -39,7 +39,6 @@
 #include <asm/pgtable.h>
 
 #include "ipath_kernel.h"
-#include "ipath_layer.h"
 #include "ipath_common.h"
 
 static int ipath_open(struct inode *, struct file *);

drivers/infiniband/hw/ipath/ipath_intr.c

@@ -34,7 +34,6 @@
 #include <linux/pci.h>
 
 #include "ipath_kernel.h"
-#include "ipath_layer.h"
 #include "ipath_verbs.h"
 #include "ipath_common.h"
 
@@ -290,8 +289,6 @@ static void handle_e_ibstatuschanged(struct ipath_devdata *dd,
 		*dd->ipath_statusp |=
 			IPATH_STATUS_IB_READY | IPATH_STATUS_IB_CONF;
 		dd->ipath_f_setextled(dd, lstate, ltstate);
-
-		__ipath_layer_intr(dd, IPATH_LAYER_INT_IF_UP);
 	} else if ((val & IPATH_IBSTATE_MASK) == IPATH_IBSTATE_INIT) {
 		/*
 		 * set INIT and DOWN.  Down is checked by most of the other
@@ -709,10 +706,6 @@ static void handle_layer_pioavail(struct ipath_devdata *dd)
 {
 	int ret;
 
-	ret = __ipath_layer_intr(dd, IPATH_LAYER_INT_SEND_CONTINUE);
-	if (ret > 0)
-		goto set;
-
 	ret = ipath_ib_piobufavail(dd->verbs_dev);
 	if (ret > 0)
 		goto set;

drivers/infiniband/hw/ipath/ipath_kernel.h

@@ -518,16 +518,6 @@ extern struct list_head ipath_dev_list;
 extern spinlock_t ipath_devs_lock;
 extern struct ipath_devdata *ipath_lookup(int unit);
 
-extern u16 ipath_layer_rcv_opcode;
-extern int __ipath_layer_intr(struct ipath_devdata *, u32);
-extern int ipath_layer_intr(struct ipath_devdata *, u32);
-extern int __ipath_layer_rcv(struct ipath_devdata *, void *,
-			     struct sk_buff *);
-extern int __ipath_layer_rcv_lid(struct ipath_devdata *, void *);
-
-void ipath_layer_add(struct ipath_devdata *);
-void ipath_layer_remove(struct ipath_devdata *);
-
 int ipath_init_chip(struct ipath_devdata *, int);
 int ipath_enable_wc(struct ipath_devdata *dd);
 void ipath_disable_wc(struct ipath_devdata *dd);
@@ -575,12 +565,13 @@ void ipath_free_pddata(struct ipath_devdata *, struct ipath_portdata *);
 
 int ipath_parse_ushort(const char *str, unsigned short *valp);
 
-int ipath_wait_linkstate(struct ipath_devdata *, u32, int);
-void ipath_set_ib_lstate(struct ipath_devdata *, int);
 void ipath_kreceive(struct ipath_devdata *);
 int ipath_setrcvhdrsize(struct ipath_devdata *, unsigned);
 int ipath_reset_device(int);
 void ipath_get_faststats(unsigned long);
+int ipath_set_linkstate(struct ipath_devdata *, u8);
+int ipath_set_mtu(struct ipath_devdata *, u16);
+int ipath_set_lid(struct ipath_devdata *, u32, u8);
 
 /* for use in system calls, where we want to know device type, etc. */
 #define port_fp(fp) ((struct ipath_portdata *) (fp)->private_data)

drivers/infiniband/hw/ipath/ipath_layer.h

@@ -40,73 +40,9 @@
  */
 
 struct sk_buff;
-struct ipath_sge_state;
 struct ipath_devdata;
 struct ether_header;
 
-struct ipath_layer_counters {
-	u64 symbol_error_counter;
-	u64 link_error_recovery_counter;
-	u64 link_downed_counter;
-	u64 port_rcv_errors;
-	u64 port_rcv_remphys_errors;
-	u64 port_xmit_discards;
-	u64 port_xmit_data;
-	u64 port_rcv_data;
-	u64 port_xmit_packets;
-	u64 port_rcv_packets;
-	u32 local_link_integrity_errors;
-	u32 excessive_buffer_overrun_errors;
-};
-
-/*
- * A segment is a linear region of low physical memory.
- * XXX Maybe we should use phys addr here and kmap()/kunmap().
- * Used by the verbs layer.
- */
-struct ipath_seg {
-	void *vaddr;
-	size_t length;
-};
-
-/* The number of ipath_segs that fit in a page. */
-#define IPATH_SEGSZ     (PAGE_SIZE / sizeof (struct ipath_seg))
-
-struct ipath_segarray {
-	struct ipath_seg segs[IPATH_SEGSZ];
-};
-
-struct ipath_mregion {
-	u64 user_base;		/* User's address for this region */
-	u64 iova;		/* IB start address of this region */
-	size_t length;
-	u32 lkey;
-	u32 offset;		/* offset (bytes) to start of region */
-	int access_flags;
-	u32 max_segs;		/* number of ipath_segs in all the arrays */
-	u32 mapsz;		/* size of the map array */
-	struct ipath_segarray *map[0];	/* the segments */
-};
-
-/*
- * These keep track of the copy progress within a memory region.
- * Used by the verbs layer.
- */
-struct ipath_sge {
-	struct ipath_mregion *mr;
-	void *vaddr;		/* current pointer into the segment */
-	u32 sge_length;		/* length of the SGE */
-	u32 length;		/* remaining length of the segment */
-	u16 m;			/* current index: mr->map[m] */
-	u16 n;			/* current index: mr->map[m]->segs[n] */
-};
-
-struct ipath_sge_state {
-	struct ipath_sge *sg_list;	/* next SGE to be used if any */
-	struct ipath_sge sge;	/* progress state for the current SGE */
-	u8 num_sge;
-};
-
 int ipath_layer_register(void *(*l_add)(int, struct ipath_devdata *),
 			 void (*l_remove)(void *),
 			 int (*l_intr)(void *, u32),
@@ -119,49 +55,9 @@ int ipath_layer_open(struct ipath_devdata *, u32 * pktmax);
 u16 ipath_layer_get_lid(struct ipath_devdata *dd);
 int ipath_layer_get_mac(struct ipath_devdata *dd, u8 *);
 u16 ipath_layer_get_bcast(struct ipath_devdata *dd);
-u32 ipath_layer_get_cr_errpkey(struct ipath_devdata *dd);
-int ipath_layer_set_linkstate(struct ipath_devdata *dd, u8 state);
-int ipath_layer_set_mtu(struct ipath_devdata *, u16);
-int ipath_set_lid(struct ipath_devdata *, u32, u8);
 int ipath_layer_send_hdr(struct ipath_devdata *dd,
 			 struct ether_header *hdr);
-int ipath_verbs_send(struct ipath_devdata *dd, u32 hdrwords,
-		     u32 * hdr, u32 len, struct ipath_sge_state *ss);
 int ipath_layer_set_piointbufavail_int(struct ipath_devdata *dd);
-int ipath_layer_get_boardname(struct ipath_devdata *dd, char *name,
-			      size_t namelen);
-int ipath_layer_snapshot_counters(struct ipath_devdata *dd, u64 *swords,
-				  u64 *rwords, u64 *spkts, u64 *rpkts,
-				  u64 *xmit_wait);
-int ipath_layer_get_counters(struct ipath_devdata *dd,
-			     struct ipath_layer_counters *cntrs);
-int ipath_layer_want_buffer(struct ipath_devdata *dd);
-int ipath_layer_set_guid(struct ipath_devdata *, __be64 guid);
-__be64 ipath_layer_get_guid(struct ipath_devdata *);
-u32 ipath_layer_get_majrev(struct ipath_devdata *);
-u32 ipath_layer_get_minrev(struct ipath_devdata *);
-u32 ipath_layer_get_pcirev(struct ipath_devdata *);
-u32 ipath_layer_get_flags(struct ipath_devdata *dd);
-struct device *ipath_layer_get_device(struct ipath_devdata *dd);
-u16 ipath_layer_get_deviceid(struct ipath_devdata *dd);
-u32 ipath_layer_get_vendorid(struct ipath_devdata *);
-u64 ipath_layer_get_lastibcstat(struct ipath_devdata *dd);
-u32 ipath_layer_get_ibmtu(struct ipath_devdata *dd);
-int ipath_layer_enable_timer(struct ipath_devdata *dd);
-int ipath_layer_disable_timer(struct ipath_devdata *dd);
-int ipath_layer_set_verbs_flags(struct ipath_devdata *dd, unsigned flags);
-unsigned ipath_layer_get_npkeys(struct ipath_devdata *dd);
-unsigned ipath_layer_get_pkey(struct ipath_devdata *dd, unsigned index);
-int ipath_layer_get_pkeys(struct ipath_devdata *dd, u16 *pkeys);
-int ipath_layer_set_pkeys(struct ipath_devdata *dd, u16 *pkeys);
-int ipath_layer_get_linkdowndefaultstate(struct ipath_devdata *dd);
-int ipath_layer_set_linkdowndefaultstate(struct ipath_devdata *dd,
-					 int sleep);
-int ipath_layer_get_phyerrthreshold(struct ipath_devdata *dd);
-int ipath_layer_set_phyerrthreshold(struct ipath_devdata *dd, unsigned n);
-int ipath_layer_get_overrunthreshold(struct ipath_devdata *dd);
-int ipath_layer_set_overrunthreshold(struct ipath_devdata *dd, unsigned n);
-u32 ipath_layer_get_rcvhdrentsize(struct ipath_devdata *dd);
 
 /* ipath_ether interrupt values */
 #define IPATH_LAYER_INT_IF_UP 0x2

drivers/infiniband/hw/ipath/ipath_mr.c

@@ -36,6 +36,18 @@
 
 #include "ipath_verbs.h"
 
+/* Fast memory region */
+struct ipath_fmr {
+	struct ib_fmr ibfmr;
+	u8 page_shift;
+	struct ipath_mregion mr;        /* must be last */
+};
+
+static inline struct ipath_fmr *to_ifmr(struct ib_fmr *ibfmr)
+{
+	return container_of(ibfmr, struct ipath_fmr, ibfmr);
+}
+
 /**
  * ipath_get_dma_mr - get a DMA memory region
  * @pd: protection domain for this memory region

drivers/infiniband/hw/ipath/ipath_qp.c

@@ -461,7 +461,7 @@ int ipath_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 			goto inval;
 
 	if (attr_mask & IB_QP_PKEY_INDEX)
-		if (attr->pkey_index >= ipath_layer_get_npkeys(dev->dd))
+		if (attr->pkey_index >= ipath_get_npkeys(dev->dd))
 			goto inval;
 
 	if (attr_mask & IB_QP_MIN_RNR_TIMER)
@@ -644,6 +644,33 @@ __be32 ipath_compute_aeth(struct ipath_qp *qp)
 	return cpu_to_be32(aeth);
 }
 
+/**
+ * set_verbs_flags - set the verbs layer flags
+ * @dd: the infinipath device
+ * @flags: the flags to set
+ */
+static int set_verbs_flags(struct ipath_devdata *dd, unsigned flags)
+{
+	struct ipath_devdata *ss;
+	unsigned long lflags;
+
+	spin_lock_irqsave(&ipath_devs_lock, lflags);
+
+	list_for_each_entry(ss, &ipath_dev_list, ipath_list) {
+		if (!(ss->ipath_flags & IPATH_INITTED))
+			continue;
+		if ((flags & IPATH_VERBS_KERNEL_SMA) &&
+		    !(*ss->ipath_statusp & IPATH_STATUS_SMA))
+			*ss->ipath_statusp |= IPATH_STATUS_OIB_SMA;
+		else
+			*ss->ipath_statusp &= ~IPATH_STATUS_OIB_SMA;
+	}
+
+	spin_unlock_irqrestore(&ipath_devs_lock, lflags);
+
+	return 0;
+}
+
 /**
  * ipath_create_qp - create a queue pair for a device
  * @ibpd: the protection domain who's device we create the queue pair for
@@ -760,8 +787,7 @@ struct ib_qp *ipath_create_qp(struct ib_pd *ibpd,
 
 		/* Tell the core driver that the kernel SMA is present. */
 		if (init_attr->qp_type == IB_QPT_SMI)
-			ipath_layer_set_verbs_flags(dev->dd,
-						    IPATH_VERBS_KERNEL_SMA);
+			set_verbs_flags(dev->dd, IPATH_VERBS_KERNEL_SMA);
 		break;
 
 	default:
@@ -838,7 +864,7 @@ int ipath_destroy_qp(struct ib_qp *ibqp)
 
 	/* Tell the core driver that the kernel SMA is gone. */
 	if (qp->ibqp.qp_type == IB_QPT_SMI)
-		ipath_layer_set_verbs_flags(dev->dd, 0);
+		set_verbs_flags(dev->dd, 0);
 
 	spin_lock_irqsave(&qp->s_lock, flags);
 	qp->state = IB_QPS_ERR;

drivers/infiniband/hw/ipath/ipath_rc.c

@@ -32,7 +32,7 @@
  */
 
 #include "ipath_verbs.h"
-#include "ipath_common.h"
+#include "ipath_kernel.h"
 
 /* cut down ridiculously long IB macro names */
 #define OP(x) IB_OPCODE_RC_##x
@@ -540,7 +540,7 @@ static void send_rc_ack(struct ipath_qp *qp)
 		lrh0 = IPATH_LRH_GRH;
 	}
 	/* read pkey_index w/o lock (its atomic) */
-	bth0 = ipath_layer_get_pkey(dev->dd, qp->s_pkey_index);
+	bth0 = ipath_get_pkey(dev->dd, qp->s_pkey_index);
 	if (qp->r_nak_state)
 		ohdr->u.aeth = cpu_to_be32((qp->r_msn & IPATH_MSN_MASK) |
 					    (qp->r_nak_state <<
@@ -557,7 +557,7 @@ static void send_rc_ack(struct ipath_qp *qp)
 	hdr.lrh[0] = cpu_to_be16(lrh0);
 	hdr.lrh[1] = cpu_to_be16(qp->remote_ah_attr.dlid);
 	hdr.lrh[2] = cpu_to_be16(hwords + SIZE_OF_CRC);
-	hdr.lrh[3] = cpu_to_be16(ipath_layer_get_lid(dev->dd));
+	hdr.lrh[3] = cpu_to_be16(dev->dd->ipath_lid);
 	ohdr->bth[0] = cpu_to_be32(bth0);
 	ohdr->bth[1] = cpu_to_be32(qp->remote_qpn);
 	ohdr->bth[2] = cpu_to_be32(qp->r_ack_psn & IPATH_PSN_MASK);
@@ -1323,8 +1323,7 @@ void ipath_rc_rcv(struct ipath_ibdev *dev, struct ipath_ib_header *hdr,
 		 * the eager header buffer size to 56 bytes so the last 4
 		 * bytes of the BTH header (PSN) is in the data buffer.
 		 */
-		header_in_data =
-			ipath_layer_get_rcvhdrentsize(dev->dd) == 16;
+		header_in_data = dev->dd->ipath_rcvhdrentsize == 16;
 		if (header_in_data) {
 			psn = be32_to_cpu(((__be32 *) data)[0]);
 			data += sizeof(__be32);

drivers/infiniband/hw/ipath/ipath_ruc.c

@@ -470,6 +470,15 @@ done:
 		wake_up(&qp->wait);
 }
 
+static int want_buffer(struct ipath_devdata *dd)
+{
+	set_bit(IPATH_S_PIOINTBUFAVAIL, &dd->ipath_sendctrl);
+	ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl,
+			 dd->ipath_sendctrl);
+
+	return 0;
+}
+
 /**
  * ipath_no_bufs_available - tell the layer driver we need buffers
  * @qp: the QP that caused the problem
@@ -486,7 +495,7 @@ void ipath_no_bufs_available(struct ipath_qp *qp, struct ipath_ibdev *dev)
 		list_add_tail(&qp->piowait, &dev->piowait);
 	spin_unlock_irqrestore(&dev->pending_lock, flags);
 	/*
-	 * Note that as soon as ipath_layer_want_buffer() is called and
+	 * Note that as soon as want_buffer() is called and
 	 * possibly before it returns, ipath_ib_piobufavail()
 	 * could be called.  If we are still in the tasklet function,
 	 * tasklet_hi_schedule() will not call us until the next time
@@ -496,7 +505,7 @@ void ipath_no_bufs_available(struct ipath_qp *qp, struct ipath_ibdev *dev)
 	 */
 	clear_bit(IPATH_S_BUSY, &qp->s_flags);
 	tasklet_unlock(&qp->s_task);
-	ipath_layer_want_buffer(dev->dd);
+	want_buffer(dev->dd);
 	dev->n_piowait++;
 }
 
@@ -611,7 +620,7 @@ u32 ipath_make_grh(struct ipath_ibdev *dev, struct ib_grh *hdr,
 	hdr->hop_limit = grh->hop_limit;
 	/* The SGID is 32-bit aligned. */
 	hdr->sgid.global.subnet_prefix = dev->gid_prefix;
-	hdr->sgid.global.interface_id = ipath_layer_get_guid(dev->dd);
+	hdr->sgid.global.interface_id = dev->dd->ipath_guid;
 	hdr->dgid = grh->dgid;
 
 	/* GRH header size in 32-bit words. */
@@ -643,8 +652,7 @@ void ipath_do_ruc_send(unsigned long data)
 	if (test_and_set_bit(IPATH_S_BUSY, &qp->s_flags))
 		goto bail;
 
-	if (unlikely(qp->remote_ah_attr.dlid ==
-		     ipath_layer_get_lid(dev->dd))) {
+	if (unlikely(qp->remote_ah_attr.dlid == dev->dd->ipath_lid)) {
 		ipath_ruc_loopback(qp);
 		goto clear;
 	}
@@ -711,8 +719,8 @@ again:
 	qp->s_hdr.lrh[1] = cpu_to_be16(qp->remote_ah_attr.dlid);
 	qp->s_hdr.lrh[2] = cpu_to_be16(qp->s_hdrwords + nwords +
 				       SIZE_OF_CRC);
-	qp->s_hdr.lrh[3] = cpu_to_be16(ipath_layer_get_lid(dev->dd));
-	bth0 |= ipath_layer_get_pkey(dev->dd, qp->s_pkey_index);
+	qp->s_hdr.lrh[3] = cpu_to_be16(dev->dd->ipath_lid);
+	bth0 |= ipath_get_pkey(dev->dd, qp->s_pkey_index);
 	bth0 |= extra_bytes << 20;
 	ohdr->bth[0] = cpu_to_be32(bth0);
 	ohdr->bth[1] = cpu_to_be32(qp->remote_qpn);

drivers/infiniband/hw/ipath/ipath_sysfs.c

@@ -35,7 +35,6 @@
 #include <linux/pci.h>
 
 #include "ipath_kernel.h"
-#include "ipath_layer.h"
 #include "ipath_common.h"
 
 /**
@@ -227,7 +226,6 @@ static ssize_t store_mlid(struct device *dev,
 	unit = dd->ipath_unit;
 
 	dd->ipath_mlid = mlid;
-	ipath_layer_intr(dd, IPATH_LAYER_INT_BCAST);
 
 	goto bail;
 invalid:
@@ -467,7 +465,7 @@ static ssize_t store_link_state(struct device *dev,
 	if (ret < 0)
 		goto invalid;
 
-	r = ipath_layer_set_linkstate(dd, state);
+	r = ipath_set_linkstate(dd, state);
 	if (r < 0) {
 		ret = r;
 		goto bail;
@@ -502,7 +500,7 @@ static ssize_t store_mtu(struct device *dev,
 	if (ret < 0)
 		goto invalid;
 
-	r = ipath_layer_set_mtu(dd, mtu);
+	r = ipath_set_mtu(dd, mtu);
 	if (r < 0)
 		ret = r;
 

drivers/infiniband/hw/ipath/ipath_uc.c

@@ -32,7 +32,7 @@
  */
 
 #include "ipath_verbs.h"
-#include "ipath_common.h"
+#include "ipath_kernel.h"
 
 /* cut down ridiculously long IB macro names */
 #define OP(x) IB_OPCODE_UC_##x
@@ -261,8 +261,7 @@ void ipath_uc_rcv(struct ipath_ibdev *dev, struct ipath_ib_header *hdr,
 		 * size to 56 bytes so the last 4 bytes of
 		 * the BTH header (PSN) is in the data buffer.
 		 */
-		header_in_data =
-			ipath_layer_get_rcvhdrentsize(dev->dd) == 16;
+		header_in_data = dev->dd->ipath_rcvhdrentsize == 16;
 		if (header_in_data) {
 			psn = be32_to_cpu(((__be32 *) data)[0]);
 			data += sizeof(__be32);

drivers/infiniband/hw/ipath/ipath_ud.c

@@ -353,7 +353,7 @@ int ipath_post_ud_send(struct ipath_qp *qp, struct ib_send_wr *wr)
 		ss.num_sge++;
 	}
 	/* Check for invalid packet size. */
-	if (len > ipath_layer_get_ibmtu(dev->dd)) {
+	if (len > dev->dd->ipath_ibmtu) {
 		ret = -EINVAL;
 		goto bail;
 	}
@@ -375,7 +375,7 @@ int ipath_post_ud_send(struct ipath_qp *qp, struct ib_send_wr *wr)
 		dev->n_unicast_xmit++;
 		lid = ah_attr->dlid &
 			~((1 << (dev->mkeyprot_resv_lmc & 7)) - 1);
-		if (unlikely(lid == ipath_layer_get_lid(dev->dd))) {
+		if (unlikely(lid == dev->dd->ipath_lid)) {
 			/*
 			 * Pass in an uninitialized ib_wc to save stack
 			 * space.
@@ -404,7 +404,7 @@ int ipath_post_ud_send(struct ipath_qp *qp, struct ib_send_wr *wr)
 		qp->s_hdr.u.l.grh.sgid.global.subnet_prefix =
 			dev->gid_prefix;
 		qp->s_hdr.u.l.grh.sgid.global.interface_id =
-			ipath_layer_get_guid(dev->dd);
+			dev->dd->ipath_guid;
 		qp->s_hdr.u.l.grh.dgid = ah_attr->grh.dgid;
 		/*
 		 * Don't worry about sending to locally attached multicast
@@ -434,7 +434,7 @@ int ipath_post_ud_send(struct ipath_qp *qp, struct ib_send_wr *wr)
 	qp->s_hdr.lrh[0] = cpu_to_be16(lrh0);
 	qp->s_hdr.lrh[1] = cpu_to_be16(ah_attr->dlid);	/* DEST LID */
 	qp->s_hdr.lrh[2] = cpu_to_be16(hwords + nwords + SIZE_OF_CRC);
-	lid = ipath_layer_get_lid(dev->dd);
+	lid = dev->dd->ipath_lid;
 	if (lid) {
 		lid |= ah_attr->src_path_bits &
 			((1 << (dev->mkeyprot_resv_lmc & 7)) - 1);
@@ -445,7 +445,7 @@ int ipath_post_ud_send(struct ipath_qp *qp, struct ib_send_wr *wr)
 		bth0 |= 1 << 23;
 	bth0 |= extra_bytes << 20;
 	bth0 |= qp->ibqp.qp_type == IB_QPT_SMI ? IPATH_DEFAULT_P_KEY :
-		ipath_layer_get_pkey(dev->dd, qp->s_pkey_index);
+		ipath_get_pkey(dev->dd, qp->s_pkey_index);
 	ohdr->bth[0] = cpu_to_be32(bth0);
 	/*
 	 * Use the multicast QP if the destination LID is a multicast LID.
@@ -531,8 +531,7 @@ void ipath_ud_rcv(struct ipath_ibdev *dev, struct ipath_ib_header *hdr,
 		 * the eager header buffer size to 56 bytes so the last 12
 		 * bytes of the IB header is in the data buffer.
 		 */
-		header_in_data =
-			ipath_layer_get_rcvhdrentsize(dev->dd) == 16;
+		header_in_data = dev->dd->ipath_rcvhdrentsize == 16;
 		if (header_in_data) {
 			qkey = be32_to_cpu(((__be32 *) data)[1]);
 			src_qp = be32_to_cpu(((__be32 *) data)[2]);

drivers/infiniband/hw/ipath/ipath_verbs.h

@@ -153,19 +153,6 @@ struct ipath_mcast {
 	int n_attached;
 };
 
-/* Memory region */
-struct ipath_mr {
-	struct ib_mr ibmr;
-	struct ipath_mregion mr;	/* must be last */
-};
-
-/* Fast memory region */
-struct ipath_fmr {
-	struct ib_fmr ibfmr;
-	u8 page_shift;
-	struct ipath_mregion mr;	/* must be last */
-};
-
 /* Protection domain */
 struct ipath_pd {
 	struct ib_pd ibpd;
@@ -216,6 +203,54 @@ struct ipath_cq {
 	struct ipath_mmap_info *ip;
 };
 
+/*
+ * A segment is a linear region of low physical memory.
+ * XXX Maybe we should use phys addr here and kmap()/kunmap().
+ * Used by the verbs layer.
+ */
+struct ipath_seg {
+	void *vaddr;
+	size_t length;
+};
+
+/* The number of ipath_segs that fit in a page. */
+#define IPATH_SEGSZ     (PAGE_SIZE / sizeof (struct ipath_seg))
+
+struct ipath_segarray {
+	struct ipath_seg segs[IPATH_SEGSZ];
+};
+
+struct ipath_mregion {
+	u64 user_base;		/* User's address for this region */
+	u64 iova;		/* IB start address of this region */
+	size_t length;
+	u32 lkey;
+	u32 offset;		/* offset (bytes) to start of region */
+	int access_flags;
+	u32 max_segs;		/* number of ipath_segs in all the arrays */
+	u32 mapsz;		/* size of the map array */
+	struct ipath_segarray *map[0];	/* the segments */
+};
+
+/*
+ * These keep track of the copy progress within a memory region.
+ * Used by the verbs layer.
+ */
+struct ipath_sge {
+	struct ipath_mregion *mr;
+	void *vaddr;		/* current pointer into the segment */
+	u32 sge_length;		/* length of the SGE */
+	u32 length;		/* remaining length of the segment */
+	u16 m;			/* current index: mr->map[m] */
+	u16 n;			/* current index: mr->map[m]->segs[n] */
+};
+
+/* Memory region */
+struct ipath_mr {
+	struct ib_mr ibmr;
+	struct ipath_mregion mr;	/* must be last */
+};
+
 /*
  * Send work request queue entry.
  * The size of the sg_list is determined when the QP is created and stored
@@ -270,6 +305,12 @@ struct ipath_srq {
 	u32 limit;
 };
 
+struct ipath_sge_state {
+	struct ipath_sge *sg_list;      /* next SGE to be used if any */
+	struct ipath_sge sge;   /* progress state for the current SGE */
+	u8 num_sge;
+};
+
 /*
  * Variables prefixed with s_ are for the requester (sender).
  * Variables prefixed with r_ are for the responder (receiver).
@@ -500,8 +541,19 @@ struct ipath_ibdev {
 	struct ipath_opcode_stats opstats[128];
 };
 
-struct ipath_ucontext {
-	struct ib_ucontext ibucontext;
+struct ipath_verbs_counters {
+	u64 symbol_error_counter;
+	u64 link_error_recovery_counter;
+	u64 link_downed_counter;
+	u64 port_rcv_errors;
+	u64 port_rcv_remphys_errors;
+	u64 port_xmit_discards;
+	u64 port_xmit_data;
+	u64 port_rcv_data;
+	u64 port_xmit_packets;
+	u64 port_rcv_packets;
+	u32 local_link_integrity_errors;
+	u32 excessive_buffer_overrun_errors;
 };
 
 static inline struct ipath_mr *to_imr(struct ib_mr *ibmr)
@@ -509,11 +561,6 @@ static inline struct ipath_mr *to_imr(struct ib_mr *ibmr)
 	return container_of(ibmr, struct ipath_mr, ibmr);
 }
 
-static inline struct ipath_fmr *to_ifmr(struct ib_fmr *ibfmr)
-{
-	return container_of(ibfmr, struct ipath_fmr, ibfmr);
-}
-
 static inline struct ipath_pd *to_ipd(struct ib_pd *ibpd)
 {
 	return container_of(ibpd, struct ipath_pd, ibpd);
@@ -551,12 +598,6 @@ int ipath_process_mad(struct ib_device *ibdev,
 		      struct ib_grh *in_grh,
 		      struct ib_mad *in_mad, struct ib_mad *out_mad);
 
-static inline struct ipath_ucontext *to_iucontext(struct ib_ucontext
-						  *ibucontext)
-{
-	return container_of(ibucontext, struct ipath_ucontext, ibucontext);
-}
-
 /*
  * Compare the lower 24 bits of the two values.
  * Returns an integer <, ==, or > than zero.
@@ -568,6 +609,13 @@ static inline int ipath_cmp24(u32 a, u32 b)
 
 struct ipath_mcast *ipath_mcast_find(union ib_gid *mgid);
 
+int ipath_snapshot_counters(struct ipath_devdata *dd, u64 *swords,
+			    u64 *rwords, u64 *spkts, u64 *rpkts,
+			    u64 *xmit_wait);
+
+int ipath_get_counters(struct ipath_devdata *dd,
+		       struct ipath_verbs_counters *cntrs);
+
 int ipath_multicast_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid);
 
 int ipath_multicast_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid);
@@ -598,6 +646,9 @@ void ipath_sqerror_qp(struct ipath_qp *qp, struct ib_wc *wc);
 
 void ipath_get_credit(struct ipath_qp *qp, u32 aeth);
 
+int ipath_verbs_send(struct ipath_devdata *dd, u32 hdrwords,
+		     u32 *hdr, u32 len, struct ipath_sge_state *ss);
+
 void ipath_cq_enter(struct ipath_cq *cq, struct ib_wc *entry, int sig);
 
 int ipath_rkey_ok(struct ipath_ibdev *dev, struct ipath_sge_state *ss,
@@ -721,6 +772,12 @@ int ipath_ib_piobufavail(struct ipath_ibdev *);
 
 void ipath_ib_timer(struct ipath_ibdev *);
 
+unsigned ipath_get_npkeys(struct ipath_devdata *);
+
+u32 ipath_get_cr_errpkey(struct ipath_devdata *);
+
+unsigned ipath_get_pkey(struct ipath_devdata *, unsigned);
+
 extern const enum ib_wc_opcode ib_ipath_wc_opcode[];
 
 extern const u8 ipath_cvt_physportstate[];