26 files changed, 333 insertions, 231 deletions

diff --git a/drivers/infiniband/core/sysfs.c b/drivers/infiniband/core/sysfs.c
index 15121cb5a1f6..21f9282c1b25 100644
--- a/drivers/infiniband/core/sysfs.c
+++ b/drivers/infiniband/core/sysfs.c
@@ -336,7 +336,7 @@ static ssize_t show_pma_counter(struct ib_port *p, struct port_attribute *attr,
         switch (width) {
         case 4:
                 ret = sprintf(buf, "%u\n", (out_mad->data[40 + offset / 8] >>
-                                            (offset % 4)) & 0xf);
+                                            (4 - (offset % 8))) & 0xf);
                 break;
         case 8:
                 ret = sprintf(buf, "%u\n", out_mad->data[40 + offset / 8]);
diff --git a/drivers/infiniband/hw/ipath/ipath_debug.h b/drivers/infiniband/hw/ipath/ipath_debug.h
index 593e28969c69..46762387f5f8 100644
--- a/drivers/infiniband/hw/ipath/ipath_debug.h
+++ b/drivers/infiniband/hw/ipath/ipath_debug.h
@@ -60,11 +60,11 @@
 #define __IPATH_KERNEL_SEND 0x2000      /* use kernel mode send */
 #define __IPATH_EPKTDBG     0x4000      /* print ethernet packet data */
 #define __IPATH_SMADBG      0x8000      /* sma packet debug */
-#define __IPATH_IPATHDBG    0x10000     /* Ethernet (IPATH) general debug on */
-#define __IPATH_IPATHWARN   0x20000     /* Ethernet (IPATH) warnings on */
-#define __IPATH_IPATHERR    0x40000     /* Ethernet (IPATH) errors on */
-#define __IPATH_IPATHPD     0x80000     /* Ethernet (IPATH) packet dump on */
-#define __IPATH_IPATHTABLE  0x100000    /* Ethernet (IPATH) table dump on */
+#define __IPATH_IPATHDBG    0x10000     /* Ethernet (IPATH) gen debug */
+#define __IPATH_IPATHWARN   0x20000     /* Ethernet (IPATH) warnings */
+#define __IPATH_IPATHERR    0x40000     /* Ethernet (IPATH) errors */
+#define __IPATH_IPATHPD     0x80000     /* Ethernet (IPATH) packet dump */
+#define __IPATH_IPATHTABLE  0x100000    /* Ethernet (IPATH) table dump */
 
 #else                           /* _IPATH_DEBUGGING */
 
@@ -79,11 +79,12 @@
 #define __IPATH_TRSAMPLE  0x0   /* generate trace buffer sample entries */
 #define __IPATH_VERBDBG   0x0   /* very verbose debug */
 #define __IPATH_PKTDBG    0x0   /* print packet data */
-#define __IPATH_PROCDBG   0x0   /* print process startup (init)/exit messages */
+#define __IPATH_PROCDBG   0x0   /* process startup (init)/exit messages */
 /* print mmap/nopage stuff, not using VDBG any more */
 #define __IPATH_MMDBG     0x0
 #define __IPATH_EPKTDBG   0x0   /* print ethernet packet data */
-#define __IPATH_SMADBG    0x0   /* print process startup (init)/exit messages */#define __IPATH_IPATHDBG  0x0   /* Ethernet (IPATH) table dump on */
+#define __IPATH_SMADBG    0x0   /* process startup (init)/exit messages */
+#define __IPATH_IPATHDBG  0x0   /* Ethernet (IPATH) table dump on */
 #define __IPATH_IPATHWARN 0x0   /* Ethernet (IPATH) warnings on   */
 #define __IPATH_IPATHERR  0x0   /* Ethernet (IPATH) errors on   */
 #define __IPATH_IPATHPD   0x0   /* Ethernet (IPATH) packet dump on   */
diff --git a/drivers/infiniband/hw/ipath/ipath_diag.c b/drivers/infiniband/hw/ipath/ipath_diag.c
index 7d3fb6996b41..28ddceb260e8 100644
--- a/drivers/infiniband/hw/ipath/ipath_diag.c
+++ b/drivers/infiniband/hw/ipath/ipath_diag.c
@@ -277,13 +277,14 @@ static int ipath_diag_open(struct inode *in, struct file *fp)
 
 bail:
         spin_unlock_irqrestore(&ipath_devs_lock, flags);
-        mutex_unlock(&ipath_mutex);
 
         /* Only expose a way to reset the device if we
            make it into diag mode. */
         if (ret == 0)
                 ipath_expose_reset(&dd->pcidev->dev);
 
+        mutex_unlock(&ipath_mutex);
+
         return ret;
 }
 
diff --git a/drivers/infiniband/hw/ipath/ipath_driver.c b/drivers/infiniband/hw/ipath/ipath_driver.c
index e7617c3982ea..398add4d4cb1 100644
--- a/drivers/infiniband/hw/ipath/ipath_driver.c
+++ b/drivers/infiniband/hw/ipath/ipath_driver.c
@@ -418,9 +418,19 @@ static int __devinit ipath_init_one(struct pci_dev *pdev,
 
         ret = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
         if (ret) {
-                dev_info(&pdev->dev, "pci_set_dma_mask unit %u "
-                         "fails: %d\n", dd->ipath_unit, ret);
-                goto bail_regions;
+                /*
+                 * if the 64 bit setup fails, try 32 bit.  Some systems
+                 * do not setup 64 bit maps on systems with 2GB or less
+                 * memory installed.
+                 */
+                ret = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
+                if (ret) {
+                        dev_info(&pdev->dev, "pci_set_dma_mask unit %u "
+                                 "fails: %d\n", dd->ipath_unit, ret);
+                        goto bail_regions;
+                }
+                else
+                        ipath_dbg("No 64bit DMA mask, used 32 bit mask\n");
         }
 
         pci_set_master(pdev);
@@ -1949,7 +1959,7 @@ int ipath_reset_device(int unit)
         }
 
         if (dd->ipath_pd)
-                for (i = 1; i < dd->ipath_portcnt; i++) {
+                for (i = 1; i < dd->ipath_cfgports; i++) {
                         if (dd->ipath_pd[i] && dd->ipath_pd[i]->port_cnt) {
                                 ipath_dbg("unit %u port %d is in use "
                                           "(PID %u cmd %s), can't reset\n",
diff --git a/drivers/infiniband/hw/ipath/ipath_init_chip.c b/drivers/infiniband/hw/ipath/ipath_init_chip.c
index 2823ff9c0c62..16f640e1c16e 100644
--- a/drivers/infiniband/hw/ipath/ipath_init_chip.c
+++ b/drivers/infiniband/hw/ipath/ipath_init_chip.c
@@ -53,13 +53,19 @@ MODULE_PARM_DESC(cfgports, "Set max number of ports to use");
 
 /*
  * Number of buffers reserved for driver (layered drivers and SMA
- * send).  Reserved at end of buffer list.
+ * send).  Reserved at end of buffer list.   Initialized based on
+ * number of PIO buffers if not set via module interface.
+ * The problem with this is that it's global, but we'll use different
+ * numbers for different chip types.  So the default value is not
+ * very useful.  I've redefined it for the 1.3 release so that it's
+ * zero unless set by the user to something else, in which case we
+ * try to respect it.
  */
-static ushort ipath_kpiobufs = 32;
+static ushort ipath_kpiobufs;
 
 static int ipath_set_kpiobufs(const char *val, struct kernel_param *kp);
 
-module_param_call(kpiobufs, ipath_set_kpiobufs, param_get_uint,
+module_param_call(kpiobufs, ipath_set_kpiobufs, param_get_ushort,
                   &ipath_kpiobufs, S_IWUSR | S_IRUGO);
 MODULE_PARM_DESC(kpiobufs, "Set number of PIO buffers for driver");
 
@@ -531,8 +537,11 @@ static int init_housekeeping(struct ipath_devdata *dd,
          * Don't clear ipath_flags as 8bit mode was set before
          * entering this func. However, we do set the linkstate to
          * unknown, so we can watch for a transition.
+         * PRESENT is set because we want register reads to work,
+         * and the kernel infrastructure saw it in config space;
+         * We clear it if we have failures.
          */
-        dd->ipath_flags |= IPATH_LINKUNK;
+        dd->ipath_flags |= IPATH_LINKUNK | IPATH_PRESENT;
         dd->ipath_flags &= ~(IPATH_LINKACTIVE | IPATH_LINKARMED |
                              IPATH_LINKDOWN | IPATH_LINKINIT);
 
@@ -560,6 +569,7 @@ static int init_housekeeping(struct ipath_devdata *dd,
             || (dd->ipath_uregbase & 0xffffffff) == 0xffffffff) {
                 ipath_dev_err(dd, "Register read failures from chip, "
                               "giving up initialization\n");
+                dd->ipath_flags &= ~IPATH_PRESENT;
                 ret = -ENODEV;
                 goto done;
         }
@@ -682,16 +692,14 @@ int ipath_init_chip(struct ipath_devdata *dd, int reinit)
          */
         dd->ipath_pioavregs = ALIGN(val, sizeof(u64) * BITS_PER_BYTE / 2)
                 / (sizeof(u64) * BITS_PER_BYTE / 2);
-        if (!ipath_kpiobufs)    /* have to have at least 1, for SMA */
-                kpiobufs = ipath_kpiobufs = 1;
-        else if ((dd->ipath_piobcnt2k + dd->ipath_piobcnt4k) <
-                 (dd->ipath_cfgports * IPATH_MIN_USER_PORT_BUFCNT)) {
-                dev_info(&dd->pcidev->dev, "Too few PIO buffers (%u) "
-                         "for %u ports to have %u each!\n",
-                         dd->ipath_piobcnt2k + dd->ipath_piobcnt4k,
-                         dd->ipath_cfgports, IPATH_MIN_USER_PORT_BUFCNT);
-                kpiobufs = 1;   /* reserve just the minimum for SMA/ether */
-        } else
+        if (ipath_kpiobufs == 0) {
+                /* not set by user, or set explictly to default  */
+                if ((dd->ipath_piobcnt2k + dd->ipath_piobcnt4k) > 128)
+                        kpiobufs = 32;
+                else
+                        kpiobufs = 16;
+        }
+        else
                 kpiobufs = ipath_kpiobufs;
 
         if (kpiobufs >
diff --git a/drivers/infiniband/hw/ipath/ipath_intr.c b/drivers/infiniband/hw/ipath/ipath_intr.c
index 0bcb428041f3..3e72a1fe3d73 100644
--- a/drivers/infiniband/hw/ipath/ipath_intr.c
+++ b/drivers/infiniband/hw/ipath/ipath_intr.c
@@ -665,14 +665,14 @@ static void handle_layer_pioavail(struct ipath_devdata *dd)
 
         ret = __ipath_layer_intr(dd, IPATH_LAYER_INT_SEND_CONTINUE);
         if (ret > 0)
-                goto clear;
+                goto set;
 
         ret = __ipath_verbs_piobufavail(dd);
         if (ret > 0)
-                goto clear;
+                goto set;
 
         return;
-clear:
+set:
         set_bit(IPATH_S_PIOINTBUFAVAIL, &dd->ipath_sendctrl);
         ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl,
                          dd->ipath_sendctrl);
@@ -719,11 +719,24 @@ static void handle_rcv(struct ipath_devdata *dd, u32 istat)
 irqreturn_t ipath_intr(int irq, void *data, struct pt_regs *regs)
 {
         struct ipath_devdata *dd = data;
-        u32 istat = ipath_read_kreg32(dd, dd->ipath_kregs->kr_intstatus);
+        u32 istat;
         ipath_err_t estat = 0;
         static unsigned unexpected = 0;
         irqreturn_t ret;
 
+        if(!(dd->ipath_flags & IPATH_PRESENT)) {
+                /* this is mostly so we don't try to touch the chip while
+                 * it is being reset */
+                /*
+                 * This return value is perhaps odd, but we do not want the
+                 * interrupt core code to remove our interrupt handler
+                 * because we don't appear to be handling an interrupt
+                 * during a chip reset.
+                 */
+                return IRQ_HANDLED;
+        }
+
+        istat = ipath_read_kreg32(dd, dd->ipath_kregs->kr_intstatus);
         if (unlikely(!istat)) {
                 ipath_stats.sps_nullintr++;
                 ret = IRQ_NONE; /* not our interrupt, or already handled */
diff --git a/drivers/infiniband/hw/ipath/ipath_kernel.h b/drivers/infiniband/hw/ipath/ipath_kernel.h
index 0ce5f19c9d62..e6507f8115bc 100644
--- a/drivers/infiniband/hw/ipath/ipath_kernel.h
+++ b/drivers/infiniband/hw/ipath/ipath_kernel.h
@@ -731,7 +731,7 @@ u64 ipath_read_kreg64_port(const struct ipath_devdata *, ipath_kreg,
 static inline u32 ipath_read_ureg32(const struct ipath_devdata *dd,
                                     ipath_ureg regno, int port)
 {
-        if (!dd->ipath_kregbase)
+        if (!dd->ipath_kregbase || !(dd->ipath_flags & IPATH_PRESENT))
                 return 0;
 
         return readl(regno + (u64 __iomem *)
@@ -762,7 +762,7 @@ static inline void ipath_write_ureg(const struct ipath_devdata *dd,
 static inline u32 ipath_read_kreg32(const struct ipath_devdata *dd,
                                     ipath_kreg regno)
 {
-        if (!dd->ipath_kregbase)
+        if (!dd->ipath_kregbase || !(dd->ipath_flags & IPATH_PRESENT))
                 return -1;
         return readl((u32 __iomem *) & dd->ipath_kregbase[regno]);
 }
@@ -770,7 +770,7 @@ static inline u32 ipath_read_kreg32(const struct ipath_devdata *dd,
 static inline u64 ipath_read_kreg64(const struct ipath_devdata *dd,
                                     ipath_kreg regno)
 {
-        if (!dd->ipath_kregbase)
+        if (!dd->ipath_kregbase || !(dd->ipath_flags & IPATH_PRESENT))
                 return -1;
 
         return readq(&dd->ipath_kregbase[regno]);
@@ -786,7 +786,7 @@ static inline void ipath_write_kreg(const struct ipath_devdata *dd,
 static inline u64 ipath_read_creg(const struct ipath_devdata *dd,
                                   ipath_sreg regno)
 {
-        if (!dd->ipath_kregbase)
+        if (!dd->ipath_kregbase || !(dd->ipath_flags & IPATH_PRESENT))
                 return 0;
 
         return readq(regno + (u64 __iomem *)
@@ -797,7 +797,7 @@ static inline u64 ipath_read_creg(const struct ipath_devdata *dd,
 static inline u32 ipath_read_creg32(const struct ipath_devdata *dd,
                                          ipath_sreg regno)
 {
-        if (!dd->ipath_kregbase)
+        if (!dd->ipath_kregbase || !(dd->ipath_flags & IPATH_PRESENT))
                 return 0;
         return readl(regno + (u64 __iomem *)
                      (dd->ipath_cregbase +
diff --git a/drivers/infiniband/hw/ipath/ipath_layer.c b/drivers/infiniband/hw/ipath/ipath_layer.c
index 69ed1100701a..9cb5258ffed9 100644
--- a/drivers/infiniband/hw/ipath/ipath_layer.c
+++ b/drivers/infiniband/hw/ipath/ipath_layer.c
@@ -46,13 +46,15 @@
 /* Acquire before ipath_devs_lock. */
 static DEFINE_MUTEX(ipath_layer_mutex);
 
+static int ipath_verbs_registered;
+
 u16 ipath_layer_rcv_opcode;
+
 static int (*layer_intr)(void *, u32);
 static int (*layer_rcv)(void *, void *, struct sk_buff *);
 static int (*layer_rcv_lid)(void *, void *);
 static int (*verbs_piobufavail)(void *);
 static void (*verbs_rcv)(void *, void *, void *, u32);
-static int ipath_verbs_registered;
 
 static void *(*layer_add_one)(int, struct ipath_devdata *);
 static void (*layer_remove_one)(void *);
@@ -586,6 +588,8 @@ void ipath_verbs_unregister(void)
         verbs_rcv = NULL;
         verbs_timer_cb = NULL;
 
+        ipath_verbs_registered = 0;
+
         mutex_unlock(&ipath_layer_mutex);
 }
 
diff --git a/drivers/infiniband/hw/ipath/ipath_pe800.c b/drivers/infiniband/hw/ipath/ipath_pe800.c
index e1dc4f757062..6318067ab5ec 100644
--- a/drivers/infiniband/hw/ipath/ipath_pe800.c
+++ b/drivers/infiniband/hw/ipath/ipath_pe800.c
@@ -972,6 +972,8 @@ static int ipath_setup_pe_reset(struct ipath_devdata *dd)
         /* Use ERROR so it shows up in logs, etc. */
         ipath_dev_err(dd, "Resetting PE-800 unit %u\n",
                       dd->ipath_unit);
+        /* keep chip from being accessed in a few places */
+        dd->ipath_flags &= ~(IPATH_INITTED|IPATH_PRESENT);
         val = dd->ipath_control | INFINIPATH_C_RESET;
         ipath_write_kreg(dd, dd->ipath_kregs->kr_control, val);
         mb();
@@ -997,6 +999,8 @@ static int ipath_setup_pe_reset(struct ipath_devdata *dd)
                 if ((r = pci_enable_device(dd->pcidev)))
                         ipath_dev_err(dd, "pci_enable_device failed after "
                                       "reset: %d\n", r);
+                /* whether it worked or not, mark as present, again */
+                dd->ipath_flags |= IPATH_PRESENT;
                 val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_revision);
                 if (val == dd->ipath_revision) {
                         ipath_cdbg(VERBOSE, "Got matching revision "
diff --git a/drivers/infiniband/hw/ipath/ipath_registers.h b/drivers/infiniband/hw/ipath/ipath_registers.h
index 1e59750c5f63..402126eb79c9 100644
--- a/drivers/infiniband/hw/ipath/ipath_registers.h
+++ b/drivers/infiniband/hw/ipath/ipath_registers.h
@@ -34,8 +34,9 @@
 #define _IPATH_REGISTERS_H
 
 /*
- * This file should only be included by kernel source, and by the diags.
- * It defines the registers, and their contents, for the InfiniPath HT-400 chip
+ * This file should only be included by kernel source, and by the diags.  It
+ * defines the registers, and their contents, for the InfiniPath HT-400
+ * chip.
  */
 
 /*
@@ -156,8 +157,10 @@
 #define INFINIPATH_IBCC_FLOWCTRLWATERMARK_SHIFT 8
 #define INFINIPATH_IBCC_LINKINITCMD_MASK 0x3ULL
 #define INFINIPATH_IBCC_LINKINITCMD_DISABLE 1
-#define INFINIPATH_IBCC_LINKINITCMD_POLL 2      /* cycle through TS1/TS2 till OK */
-#define INFINIPATH_IBCC_LINKINITCMD_SLEEP 3     /* wait for TS1, then go on */
+/* cycle through TS1/TS2 till OK */
+#define INFINIPATH_IBCC_LINKINITCMD_POLL 2
+/* wait for TS1, then go on */
+#define INFINIPATH_IBCC_LINKINITCMD_SLEEP 3
 #define INFINIPATH_IBCC_LINKINITCMD_SHIFT 16
 #define INFINIPATH_IBCC_LINKCMD_MASK 0x3ULL
 #define INFINIPATH_IBCC_LINKCMD_INIT 1  /* move to 0x11 */
@@ -182,7 +185,8 @@
 #define INFINIPATH_IBCS_LINKSTATE_SHIFT 4
 #define INFINIPATH_IBCS_TXREADY       0x40000000
 #define INFINIPATH_IBCS_TXCREDITOK    0x80000000
-/* link training states (shift by INFINIPATH_IBCS_LINKTRAININGSTATE_SHIFT) */
+/* link training states (shift by
+   INFINIPATH_IBCS_LINKTRAININGSTATE_SHIFT) */
 #define INFINIPATH_IBCS_LT_STATE_DISABLED       0x00
 #define INFINIPATH_IBCS_LT_STATE_LINKUP         0x01
 #define INFINIPATH_IBCS_LT_STATE_POLLACTIVE     0x02
@@ -267,10 +271,12 @@
 /* kr_serdesconfig0 bits */
 #define INFINIPATH_SERDC0_RESET_MASK  0xfULL    /* overal reset bits */
 #define INFINIPATH_SERDC0_RESET_PLL   0x10000000ULL     /* pll reset */
-#define INFINIPATH_SERDC0_TXIDLE      0xF000ULL /* tx idle enables (per lane) */
-#define INFINIPATH_SERDC0_RXDETECT_EN 0xF0000ULL        /* rx detect enables (per lane) */
-#define INFINIPATH_SERDC0_L1PWR_DN       0xF0ULL        /* L1 Power down; use with RXDETECT,
-                                                           Otherwise not used on IB side */
+/* tx idle enables (per lane) */
+#define INFINIPATH_SERDC0_TXIDLE      0xF000ULL
+/* rx detect enables (per lane) */
+#define INFINIPATH_SERDC0_RXDETECT_EN 0xF0000ULL
+/* L1 Power down; use with RXDETECT, Otherwise not used on IB side */
+#define INFINIPATH_SERDC0_L1PWR_DN       0xF0ULL
 
 /* kr_xgxsconfig bits */
 #define INFINIPATH_XGXS_RESET          0x7ULL
@@ -390,12 +396,13 @@ struct ipath_kregs {
         ipath_kreg kr_txintmemsize;
         ipath_kreg kr_xgxsconfig;
         ipath_kreg kr_ibpllcfg;
-        /* use these two (and the following N ports) only with ipath_k*_kreg64_port();
-         * not *kreg64() */
+        /* use these two (and the following N ports) only with
+         * ipath_k*_kreg64_port(); not *kreg64() */
         ipath_kreg kr_rcvhdraddr;
         ipath_kreg kr_rcvhdrtailaddr;
 
-        /* remaining registers are not present on all types of infinipath chips  */
+        /* remaining registers are not present on all types of infinipath
+           chips  */
         ipath_kreg kr_rcvpktledcnt;
         ipath_kreg kr_pcierbuftestreg0;
         ipath_kreg kr_pcierbuftestreg1;
diff --git a/drivers/infiniband/hw/ipath/ipath_ruc.c b/drivers/infiniband/hw/ipath/ipath_ruc.c
index f232e77b78ee..eb81424b3c5b 100644
--- a/drivers/infiniband/hw/ipath/ipath_ruc.c
+++ b/drivers/infiniband/hw/ipath/ipath_ruc.c
@@ -531,19 +531,12 @@ int ipath_post_rc_send(struct ipath_qp *qp, struct ib_send_wr *wr)
         }
         wqe->wr.num_sge = j;
         qp->s_head = next;
-        /*
-         * Wake up the send tasklet if the QP is not waiting
-         * for an RNR timeout.
-         */
-        next = qp->s_rnr_timeout;
         spin_unlock_irqrestore(&qp->s_lock, flags);
 
-        if (next == 0) {
-                if (qp->ibqp.qp_type == IB_QPT_UC)
-                        ipath_do_uc_send((unsigned long) qp);
-                else
-                        ipath_do_rc_send((unsigned long) qp);
-        }
+        if (qp->ibqp.qp_type == IB_QPT_UC)
+                ipath_do_uc_send((unsigned long) qp);
+        else
+                ipath_do_rc_send((unsigned long) qp);
 
         ret = 0;
 
diff --git a/drivers/infiniband/hw/ipath/ipath_sysfs.c b/drivers/infiniband/hw/ipath/ipath_sysfs.c
index 32acd8048b49..f323791cc495 100644
--- a/drivers/infiniband/hw/ipath/ipath_sysfs.c
+++ b/drivers/infiniband/hw/ipath/ipath_sysfs.c
@@ -711,10 +711,22 @@ static struct attribute_group dev_attr_group = {
  * enters diag mode.  A device reset is quite likely to crash the
  * machine entirely, so we don't want to normally make it
  * available.
+ *
+ * Called with ipath_mutex held.
  */
 int ipath_expose_reset(struct device *dev)
 {
-        return device_create_file(dev, &dev_attr_reset);
+        static int exposed;
+        int ret;
+
+        if (!exposed) {
+                ret = device_create_file(dev, &dev_attr_reset);
+                exposed = 1;
+        }
+        else
+                ret = 0;
+
+        return ret;
 }
 
 int ipath_driver_create_group(struct device_driver *drv)
diff --git a/drivers/infiniband/hw/ipath/ipath_ud.c b/drivers/infiniband/hw/ipath/ipath_ud.c
index 01cfb30ee160..e606daf83210 100644
--- a/drivers/infiniband/hw/ipath/ipath_ud.c
+++ b/drivers/infiniband/hw/ipath/ipath_ud.c
@@ -46,8 +46,10 @@
  * This is called from ipath_post_ud_send() to forward a WQE addressed
  * to the same HCA.
  */
-static void ipath_ud_loopback(struct ipath_qp *sqp, struct ipath_sge_state *ss,
-                              u32 length, struct ib_send_wr *wr, struct ib_wc *wc)
+static void ipath_ud_loopback(struct ipath_qp *sqp,
+                              struct ipath_sge_state *ss,
+                              u32 length, struct ib_send_wr *wr,
+                              struct ib_wc *wc)
 {
         struct ipath_ibdev *dev = to_idev(sqp->ibqp.device);
         struct ipath_qp *qp;
diff --git a/drivers/infiniband/hw/ipath/ipath_verbs.c b/drivers/infiniband/hw/ipath/ipath_verbs.c
index 8d2558a01f35..cb9e387c301f 100644
--- a/drivers/infiniband/hw/ipath/ipath_verbs.c
+++ b/drivers/infiniband/hw/ipath/ipath_verbs.c
@@ -449,7 +449,6 @@ static void ipath_ib_timer(void *arg)
 {
         struct ipath_ibdev *dev = (struct ipath_ibdev *) arg;
         struct ipath_qp *resend = NULL;
-        struct ipath_qp *rnr = NULL;
         struct list_head *last;
         struct ipath_qp *qp;
         unsigned long flags;
@@ -465,32 +464,18 @@ static void ipath_ib_timer(void *arg)
         last = &dev->pending[dev->pending_index];
         while (!list_empty(last)) {
                 qp = list_entry(last->next, struct ipath_qp, timerwait);
-                if (last->next == LIST_POISON1 ||
-                    last->next != &qp->timerwait ||
-                    qp->timerwait.prev != last) {
-                        INIT_LIST_HEAD(last);
-                } else {
-                        list_del(&qp->timerwait);
-                        qp->timerwait.prev = (struct list_head *) resend;
-                        resend = qp;
-                        atomic_inc(&qp->refcount);
-                }
+                list_del(&qp->timerwait);
+                qp->timer_next = resend;
+                resend = qp;
+                atomic_inc(&qp->refcount);
         }
         last = &dev->rnrwait;
         if (!list_empty(last)) {
                 qp = list_entry(last->next, struct ipath_qp, timerwait);
                 if (--qp->s_rnr_timeout == 0) {
                         do {
-                                if (last->next == LIST_POISON1 ||
-                                    last->next != &qp->timerwait ||
-                                    qp->timerwait.prev != last) {
-                                        INIT_LIST_HEAD(last);
-                                        break;
-                                }
                                 list_del(&qp->timerwait);
-                                qp->timerwait.prev =
-                                        (struct list_head *) rnr;
-                                rnr = qp;
+                                tasklet_hi_schedule(&qp->s_task);
                                 if (list_empty(last))
                                         break;
                                 qp = list_entry(last->next, struct ipath_qp,
@@ -530,8 +515,7 @@ static void ipath_ib_timer(void *arg)
         spin_unlock_irqrestore(&dev->pending_lock, flags);
 
         /* XXX What if timer fires again while this is running? */
-        for (qp = resend; qp != NULL;
-             qp = (struct ipath_qp *) qp->timerwait.prev) {
+        for (qp = resend; qp != NULL; qp = qp->timer_next) {
                 struct ib_wc wc;
 
                 spin_lock_irqsave(&qp->s_lock, flags);
@@ -545,9 +529,6 @@ static void ipath_ib_timer(void *arg)
                 if (atomic_dec_and_test(&qp->refcount))
                         wake_up(&qp->wait);
         }
-        for (qp = rnr; qp != NULL;
-             qp = (struct ipath_qp *) qp->timerwait.prev)
-                tasklet_hi_schedule(&qp->s_task);
 }
 
 /**
@@ -556,9 +537,9 @@ static void ipath_ib_timer(void *arg)
  *
  * This is called from ipath_intr() at interrupt level when a PIO buffer is
  * available after ipath_verbs_send() returned an error that no buffers were
- * available.  Return 0 if we consumed all the PIO buffers and we still have
+ * available.  Return 1 if we consumed all the PIO buffers and we still have
  * QPs waiting for buffers (for now, just do a tasklet_hi_schedule and
- * return one).
+ * return zero).
  */
 static int ipath_ib_piobufavail(void *arg)
 {
@@ -579,7 +560,7 @@ static int ipath_ib_piobufavail(void *arg)
         spin_unlock_irqrestore(&dev->pending_lock, flags);
 
 bail:
-        return 1;
+        return 0;
 }
 
 static int ipath_query_device(struct ib_device *ibdev,
@@ -1159,7 +1140,7 @@ static ssize_t show_stats(struct class_device *cdev, char *buf)
 
         len = sprintf(buf,
                       "RC resends  %d\n"
-                      "RC QACKs    %d\n"
+                      "RC no QACK  %d\n"
                       "RC ACKs     %d\n"
                       "RC SEQ NAKs %d\n"
                       "RC RDMA seq %d\n"
diff --git a/drivers/infiniband/hw/ipath/ipath_verbs.h b/drivers/infiniband/hw/ipath/ipath_verbs.h
index fcafbc7c9e71..4f8d59300e9b 100644
--- a/drivers/infiniband/hw/ipath/ipath_verbs.h
+++ b/drivers/infiniband/hw/ipath/ipath_verbs.h
@@ -282,7 +282,8 @@ struct ipath_srq {
  */
 struct ipath_qp {
         struct ib_qp ibqp;
-        struct ipath_qp *next;  /* link list for QPN hash table */
+        struct ipath_qp *next;          /* link list for QPN hash table */
+        struct ipath_qp *timer_next;    /* link list for ipath_ib_timer() */
         struct list_head piowait;       /* link for wait PIO buf */
         struct list_head timerwait;     /* link for waiting for timeouts */
         struct ib_ah_attr remote_ah_attr;
diff --git a/drivers/infiniband/hw/ipath/ips_common.h b/drivers/infiniband/hw/ipath/ips_common.h
index 410a764dfcef..ab7cbbbfd03a 100644
--- a/drivers/infiniband/hw/ipath/ips_common.h
+++ b/drivers/infiniband/hw/ipath/ips_common.h
@@ -95,7 +95,7 @@ struct ether_header {
         __u8 seq_num;
         __le32 len;
         /* MUST be of word size due to PIO write requirements */
-        __u32 csum;
+        __le32 csum;
         __le16 csum_offset;
         __le16 flags;
         __u16 first_2_bytes;
diff --git a/drivers/infiniband/hw/mthca/mthca_cq.c b/drivers/infiniband/hw/mthca/mthca_cq.c
index 312cf90731ea..205854e9c662 100644
--- a/drivers/infiniband/hw/mthca/mthca_cq.c
+++ b/drivers/infiniband/hw/mthca/mthca_cq.c
@@ -238,9 +238,9 @@ void mthca_cq_event(struct mthca_dev *dev, u32 cqn,
         spin_lock(&dev->cq_table.lock);
 
         cq = mthca_array_get(&dev->cq_table.cq, cqn & (dev->limits.num_cqs - 1));
-
         if (cq)
-                atomic_inc(&cq->refcount);
+                ++cq->refcount;
+
         spin_unlock(&dev->cq_table.lock);
 
         if (!cq) {
@@ -254,8 +254,10 @@ void mthca_cq_event(struct mthca_dev *dev, u32 cqn,
         if (cq->ibcq.event_handler)
                 cq->ibcq.event_handler(&event, cq->ibcq.cq_context);
 
-        if (atomic_dec_and_test(&cq->refcount))
+        spin_lock(&dev->cq_table.lock);
+        if (!--cq->refcount)
                 wake_up(&cq->wait);
+        spin_unlock(&dev->cq_table.lock);
 }
 
 static inline int is_recv_cqe(struct mthca_cqe *cqe)
@@ -267,23 +269,13 @@ static inline int is_recv_cqe(struct mthca_cqe *cqe)
                 return !(cqe->is_send & 0x80);
 }
 
-void mthca_cq_clean(struct mthca_dev *dev, u32 cqn, u32 qpn,
+void mthca_cq_clean(struct mthca_dev *dev, struct mthca_cq *cq, u32 qpn,
                     struct mthca_srq *srq)
 {
-        struct mthca_cq *cq;
         struct mthca_cqe *cqe;
         u32 prod_index;
         int nfreed = 0;
 
-        spin_lock_irq(&dev->cq_table.lock);
-        cq = mthca_array_get(&dev->cq_table.cq, cqn & (dev->limits.num_cqs - 1));
-        if (cq)
-                atomic_inc(&cq->refcount);
-        spin_unlock_irq(&dev->cq_table.lock);
-
-        if (!cq)
-                return;
-
         spin_lock_irq(&cq->lock);
 
         /*
@@ -301,7 +293,7 @@ void mthca_cq_clean(struct mthca_dev *dev, u32 cqn, u32 qpn,
 
         if (0)
                 mthca_dbg(dev, "Cleaning QPN %06x from CQN %06x; ci %d, pi %d\n",
-                          qpn, cqn, cq->cons_index, prod_index);
+                          qpn, cq->cqn, cq->cons_index, prod_index);
 
         /*
          * Now sweep backwards through the CQ, removing CQ entries
@@ -325,8 +317,6 @@ void mthca_cq_clean(struct mthca_dev *dev, u32 cqn, u32 qpn,
         }
 
         spin_unlock_irq(&cq->lock);
-        if (atomic_dec_and_test(&cq->refcount))
-                wake_up(&cq->wait);
 }
 
 void mthca_cq_resize_copy_cqes(struct mthca_cq *cq)
@@ -821,7 +811,7 @@ int mthca_init_cq(struct mthca_dev *dev, int nent,
         }
 
         spin_lock_init(&cq->lock);
-        atomic_set(&cq->refcount, 1);
+        cq->refcount = 1;
         init_waitqueue_head(&cq->wait);
 
         memset(cq_context, 0, sizeof *cq_context);
@@ -896,6 +886,17 @@ err_out:
         return err;
 }
 
+static inline int get_cq_refcount(struct mthca_dev *dev, struct mthca_cq *cq)
+{
+        int c;
+
+        spin_lock_irq(&dev->cq_table.lock);
+        c = cq->refcount;
+        spin_unlock_irq(&dev->cq_table.lock);
+
+        return c;
+}
+
 void mthca_free_cq(struct mthca_dev *dev,
                    struct mthca_cq *cq)
 {
@@ -929,6 +930,7 @@ void mthca_free_cq(struct mthca_dev *dev,
         spin_lock_irq(&dev->cq_table.lock);
         mthca_array_clear(&dev->cq_table.cq,
                           cq->cqn & (dev->limits.num_cqs - 1));
+        --cq->refcount;
         spin_unlock_irq(&dev->cq_table.lock);
 
         if (dev->mthca_flags & MTHCA_FLAG_MSI_X)
@@ -936,8 +938,7 @@ void mthca_free_cq(struct mthca_dev *dev,
         else
                 synchronize_irq(dev->pdev->irq);
 
-        atomic_dec(&cq->refcount);
-        wait_event(cq->wait, !atomic_read(&cq->refcount));
+        wait_event(cq->wait, !get_cq_refcount(dev, cq));
 
         if (cq->is_kernel) {
                 mthca_free_cq_buf(dev, &cq->buf, cq->ibcq.cqe);
diff --git a/drivers/infiniband/hw/mthca/mthca_dev.h b/drivers/infiniband/hw/mthca/mthca_dev.h
index 4c1dcb4c1822..f8160b8de090 100644
--- a/drivers/infiniband/hw/mthca/mthca_dev.h
+++ b/drivers/infiniband/hw/mthca/mthca_dev.h
@@ -496,7 +496,7 @@ void mthca_free_cq(struct mthca_dev *dev,
 void mthca_cq_completion(struct mthca_dev *dev, u32 cqn);
 void mthca_cq_event(struct mthca_dev *dev, u32 cqn,
                     enum ib_event_type event_type);
-void mthca_cq_clean(struct mthca_dev *dev, u32 cqn, u32 qpn,
+void mthca_cq_clean(struct mthca_dev *dev, struct mthca_cq *cq, u32 qpn,
                     struct mthca_srq *srq);
 void mthca_cq_resize_copy_cqes(struct mthca_cq *cq);
 int mthca_alloc_cq_buf(struct mthca_dev *dev, struct mthca_cq_buf *buf, int nent);
diff --git a/drivers/infiniband/hw/mthca/mthca_mr.c b/drivers/infiniband/hw/mthca/mthca_mr.c
index 25e1c1db9a40..a486dec1707e 100644
--- a/drivers/infiniband/hw/mthca/mthca_mr.c
+++ b/drivers/infiniband/hw/mthca/mthca_mr.c
@@ -761,6 +761,7 @@ void mthca_arbel_fmr_unmap(struct mthca_dev *dev, struct mthca_fmr *fmr)
 
 int __devinit mthca_init_mr_table(struct mthca_dev *dev)
 {
+        unsigned long addr;
         int err, i;
 
         err = mthca_alloc_init(&dev->mr_table.mpt_alloc,
@@ -796,9 +797,12 @@ int __devinit mthca_init_mr_table(struct mthca_dev *dev)
                         goto err_fmr_mpt;
                 }
 
+                addr = pci_resource_start(dev->pdev, 4) +
+                        ((pci_resource_len(dev->pdev, 4) - 1) &
+                         dev->mr_table.mpt_base);
+
                 dev->mr_table.tavor_fmr.mpt_base =
-                        ioremap(dev->mr_table.mpt_base,
-                                (1 << i) * sizeof (struct mthca_mpt_entry));
+                        ioremap(addr, (1 << i) * sizeof(struct mthca_mpt_entry));
 
                 if (!dev->mr_table.tavor_fmr.mpt_base) {
                         mthca_warn(dev, "MPT ioremap for FMR failed.\n");
@@ -806,9 +810,12 @@ int __devinit mthca_init_mr_table(struct mthca_dev *dev)
                         goto err_fmr_mpt;
                 }
 
+                addr = pci_resource_start(dev->pdev, 4) +
+                        ((pci_resource_len(dev->pdev, 4) - 1) &
+                         dev->mr_table.mtt_base);
+
                 dev->mr_table.tavor_fmr.mtt_base =
-                        ioremap(dev->mr_table.mtt_base,
-                                (1 << i) * MTHCA_MTT_SEG_SIZE);
+                        ioremap(addr, (1 << i) * MTHCA_MTT_SEG_SIZE);
                 if (!dev->mr_table.tavor_fmr.mtt_base) {
                         mthca_warn(dev, "MTT ioremap for FMR failed.\n");
                         err = -ENOMEM;
diff --git a/drivers/infiniband/hw/mthca/mthca_provider.c b/drivers/infiniband/hw/mthca/mthca_provider.c
index 565a24b1756f..a2eae8a30167 100644
--- a/drivers/infiniband/hw/mthca/mthca_provider.c
+++ b/drivers/infiniband/hw/mthca/mthca_provider.c
@@ -306,7 +306,7 @@ static int mthca_query_gid(struct ib_device *ibdev, u8 port,
                 goto out;
         }
 
-        memcpy(gid->raw + 8, out_mad->data + (index % 8) * 16, 8);
+        memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
 
  out:
         kfree(in_mad);
diff --git a/drivers/infiniband/hw/mthca/mthca_provider.h b/drivers/infiniband/hw/mthca/mthca_provider.h
index 6676a786d690..179a8f610d0f 100644
--- a/drivers/infiniband/hw/mthca/mthca_provider.h
+++ b/drivers/infiniband/hw/mthca/mthca_provider.h
@@ -139,11 +139,12 @@ struct mthca_ah {
  * a qp may be locked, with the send cq locked first.  No other
  * nesting should be done.
  *
- * Each struct mthca_cq/qp also has an atomic_t ref count.  The
- * pointer from the cq/qp_table to the struct counts as one reference.
- * This reference also is good for access through the consumer API, so
- * modifying the CQ/QP etc doesn't need to take another reference.
- * Access because of a completion being polled does need a reference.
+ * Each struct mthca_cq/qp also has an ref count, protected by the
+ * corresponding table lock.  The pointer from the cq/qp_table to the
+ * struct counts as one reference.  This reference also is good for
+ * access through the consumer API, so modifying the CQ/QP etc doesn't
+ * need to take another reference.  Access to a QP because of a
+ * completion being polled does not need a reference either.
  *
  * Finally, each struct mthca_cq/qp has a wait_queue_head_t for the
  * destroy function to sleep on.
@@ -159,8 +160,9 @@ struct mthca_ah {
  * - decrement ref count; if zero, wake up waiters
  *
  * To destroy a CQ/QP, we can do the following:
- * - lock cq/qp_table, remove pointer, unlock cq/qp_table lock
- * - decrement ref count
+ * - lock cq/qp_table
+ * - remove pointer and decrement ref count
+ * - unlock cq/qp_table lock
  * - wait_event until ref count is zero
  *
  * It is the consumer's responsibilty to make sure that no QP
@@ -197,7 +199,7 @@ struct mthca_cq_resize {
 struct mthca_cq {
         struct ib_cq            ibcq;
         spinlock_t              lock;
-        atomic_t                refcount;
+        int                     refcount;
         int                     cqn;
         u32                     cons_index;
         struct mthca_cq_buf     buf;
@@ -217,7 +219,7 @@ struct mthca_cq {
 struct mthca_srq {
         struct ib_srq           ibsrq;
         spinlock_t              lock;
-        atomic_t                refcount;
+        int                     refcount;
         int                     srqn;
         int                     max;
         int                     max_gs;
@@ -254,7 +256,7 @@ struct mthca_wq {
 
 struct mthca_qp {
         struct ib_qp           ibqp;
-        atomic_t               refcount;
+        int                    refcount;
         u32                    qpn;
         int                    is_direct;
         u8                     port; /* for SQP and memfree use only */
diff --git a/drivers/infiniband/hw/mthca/mthca_qp.c b/drivers/infiniband/hw/mthca/mthca_qp.c
index f37b0e367323..19765f6f8d58 100644
--- a/drivers/infiniband/hw/mthca/mthca_qp.c
+++ b/drivers/infiniband/hw/mthca/mthca_qp.c
@@ -240,7 +240,7 @@ void mthca_qp_event(struct mthca_dev *dev, u32 qpn,
         spin_lock(&dev->qp_table.lock);
         qp = mthca_array_get(&dev->qp_table.qp, qpn & (dev->limits.num_qps - 1));
         if (qp)
-                atomic_inc(&qp->refcount);
+                ++qp->refcount;
         spin_unlock(&dev->qp_table.lock);
 
         if (!qp) {
@@ -257,8 +257,10 @@ void mthca_qp_event(struct mthca_dev *dev, u32 qpn,
         if (qp->ibqp.event_handler)
                 qp->ibqp.event_handler(&event, qp->ibqp.qp_context);
 
-        if (atomic_dec_and_test(&qp->refcount))
+        spin_lock(&dev->qp_table.lock);
+        if (!--qp->refcount)
                 wake_up(&qp->wait);
+        spin_unlock(&dev->qp_table.lock);
 }
 
 static int to_mthca_state(enum ib_qp_state ib_state)
@@ -833,10 +835,10 @@ int mthca_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, int attr_mask)
          * entries and reinitialize the QP.
          */
         if (new_state == IB_QPS_RESET && !qp->ibqp.uobject) {
-                mthca_cq_clean(dev, to_mcq(qp->ibqp.send_cq)->cqn, qp->qpn,
+                mthca_cq_clean(dev, to_mcq(qp->ibqp.send_cq), qp->qpn,
                                qp->ibqp.srq ? to_msrq(qp->ibqp.srq) : NULL);
                 if (qp->ibqp.send_cq != qp->ibqp.recv_cq)
-                        mthca_cq_clean(dev, to_mcq(qp->ibqp.recv_cq)->cqn, qp->qpn,
+                        mthca_cq_clean(dev, to_mcq(qp->ibqp.recv_cq), qp->qpn,
                                        qp->ibqp.srq ? to_msrq(qp->ibqp.srq) : NULL);
 
                 mthca_wq_init(&qp->sq);
@@ -1096,7 +1098,7 @@ static int mthca_alloc_qp_common(struct mthca_dev *dev,
         int ret;
         int i;
 
-        atomic_set(&qp->refcount, 1);
+        qp->refcount = 1;
         init_waitqueue_head(&qp->wait);
         qp->state        = IB_QPS_RESET;
         qp->atomic_rd_en = 0;
@@ -1318,6 +1320,17 @@ int mthca_alloc_sqp(struct mthca_dev *dev,
         return err;
 }
 
+static inline int get_qp_refcount(struct mthca_dev *dev, struct mthca_qp *qp)
+{
+        int c;
+
+        spin_lock_irq(&dev->qp_table.lock);
+        c = qp->refcount;
+        spin_unlock_irq(&dev->qp_table.lock);
+
+        return c;
+}
+
 void mthca_free_qp(struct mthca_dev *dev,
                    struct mthca_qp *qp)
 {
@@ -1339,14 +1352,14 @@ void mthca_free_qp(struct mthca_dev *dev,
         spin_lock(&dev->qp_table.lock);
         mthca_array_clear(&dev->qp_table.qp,
                           qp->qpn & (dev->limits.num_qps - 1));
+        --qp->refcount;
         spin_unlock(&dev->qp_table.lock);
 
         if (send_cq != recv_cq)
                 spin_unlock(&recv_cq->lock);
         spin_unlock_irq(&send_cq->lock);
 
-        atomic_dec(&qp->refcount);
-        wait_event(qp->wait, !atomic_read(&qp->refcount));
+        wait_event(qp->wait, !get_qp_refcount(dev, qp));
 
         if (qp->state != IB_QPS_RESET)
                 mthca_MODIFY_QP(dev, qp->state, IB_QPS_RESET, qp->qpn, 0,
@@ -1358,10 +1371,10 @@ void mthca_free_qp(struct mthca_dev *dev,
          * unref the mem-free tables and free the QPN in our table.
          */
         if (!qp->ibqp.uobject) {
-                mthca_cq_clean(dev, to_mcq(qp->ibqp.send_cq)->cqn, qp->qpn,
+                mthca_cq_clean(dev, to_mcq(qp->ibqp.send_cq), qp->qpn,
                                qp->ibqp.srq ? to_msrq(qp->ibqp.srq) : NULL);
                 if (qp->ibqp.send_cq != qp->ibqp.recv_cq)
-                        mthca_cq_clean(dev, to_mcq(qp->ibqp.recv_cq)->cqn, qp->qpn,
+                        mthca_cq_clean(dev, to_mcq(qp->ibqp.recv_cq), qp->qpn,
                                        qp->ibqp.srq ? to_msrq(qp->ibqp.srq) : NULL);
 
                 mthca_free_memfree(dev, qp);
diff --git a/drivers/infiniband/hw/mthca/mthca_srq.c b/drivers/infiniband/hw/mthca/mthca_srq.c
index adcaf85355ae..1ea433291fa7 100644
--- a/drivers/infiniband/hw/mthca/mthca_srq.c
+++ b/drivers/infiniband/hw/mthca/mthca_srq.c
@@ -241,7 +241,7 @@ int mthca_alloc_srq(struct mthca_dev *dev, struct mthca_pd *pd,
                 goto err_out_mailbox;
 
         spin_lock_init(&srq->lock);
-        atomic_set(&srq->refcount, 1);
+        srq->refcount = 1;
         init_waitqueue_head(&srq->wait);
 
         if (mthca_is_memfree(dev))
@@ -308,6 +308,17 @@ err_out:
         return err;
 }
 
+static inline int get_srq_refcount(struct mthca_dev *dev, struct mthca_srq *srq)
+{
+        int c;
+
+        spin_lock_irq(&dev->srq_table.lock);
+        c = srq->refcount;
+        spin_unlock_irq(&dev->srq_table.lock);
+
+        return c;
+}
+
 void mthca_free_srq(struct mthca_dev *dev, struct mthca_srq *srq)
 {
         struct mthca_mailbox *mailbox;
@@ -329,10 +340,10 @@ void mthca_free_srq(struct mthca_dev *dev, struct mthca_srq *srq)
         spin_lock_irq(&dev->srq_table.lock);
         mthca_array_clear(&dev->srq_table.srq,
                           srq->srqn & (dev->limits.num_srqs - 1));
+        --srq->refcount;
         spin_unlock_irq(&dev->srq_table.lock);
 
-        atomic_dec(&srq->refcount);
-        wait_event(srq->wait, !atomic_read(&srq->refcount));
+        wait_event(srq->wait, !get_srq_refcount(dev, srq));
 
         if (!srq->ibsrq.uobject) {
                 mthca_free_srq_buf(dev, srq);
@@ -414,7 +425,7 @@ void mthca_srq_event(struct mthca_dev *dev, u32 srqn,
         spin_lock(&dev->srq_table.lock);
         srq = mthca_array_get(&dev->srq_table.srq, srqn & (dev->limits.num_srqs - 1));
         if (srq)
-                atomic_inc(&srq->refcount);
+                ++srq->refcount;
         spin_unlock(&dev->srq_table.lock);
 
         if (!srq) {
@@ -431,8 +442,10 @@ void mthca_srq_event(struct mthca_dev *dev, u32 srqn,
         srq->ibsrq.event_handler(&event, srq->ibsrq.srq_context);
 
 out:
-        if (atomic_dec_and_test(&srq->refcount))
+        spin_lock(&dev->srq_table.lock);
+        if (!--srq->refcount)
                 wake_up(&srq->wait);
+        spin_unlock(&dev->srq_table.lock);
 }
 
 /*
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_vlan.c b/drivers/infiniband/ulp/ipoib/ipoib_vlan.c
index 4ca175553f9f..f887780e8093 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_vlan.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_vlan.c
@@ -158,10 +158,8 @@ int ipoib_vlan_delete(struct net_device *pdev, unsigned short pkey)
                 if (priv->pkey == pkey) {
                         unregister_netdev(priv->dev);
                         ipoib_dev_cleanup(priv->dev);
-
                         list_del(&priv->list);
-
-                        kfree(priv);
+                        free_netdev(priv->dev);
 
                         ret = 0;
                         break;
diff --git a/drivers/infiniband/ulp/srp/ib_srp.c b/drivers/infiniband/ulp/srp/ib_srp.c
index 5bb55742ada6..c32ce4348e1b 100644
--- a/drivers/infiniband/ulp/srp/ib_srp.c
+++ b/drivers/infiniband/ulp/srp/ib_srp.c
@@ -409,6 +409,34 @@ static int srp_connect_target(struct srp_target_port *target)
         }
 }
 
+static void srp_unmap_data(struct scsi_cmnd *scmnd,
+                           struct srp_target_port *target,
+                           struct srp_request *req)
+{
+        struct scatterlist *scat;
+        int nents;
+
+        if (!scmnd->request_buffer ||
+            (scmnd->sc_data_direction != DMA_TO_DEVICE &&
+             scmnd->sc_data_direction != DMA_FROM_DEVICE))
+                return;
+
+        /*
+         * This handling of non-SG commands can be killed when the
+         * SCSI midlayer no longer generates non-SG commands.
+         */
+        if (likely(scmnd->use_sg)) {
+                nents = scmnd->use_sg;
+                scat  = scmnd->request_buffer;
+        } else {
+                nents = 1;
+                scat  = &req->fake_sg;
+        }
+
+        dma_unmap_sg(target->srp_host->dev->dma_device, scat, nents,
+                     scmnd->sc_data_direction);
+}
+
 static int srp_reconnect_target(struct srp_target_port *target)
 {
         struct ib_cm_id *new_cm_id;
@@ -455,16 +483,16 @@ static int srp_reconnect_target(struct srp_target_port *target)
         list_for_each_entry(req, &target->req_queue, list) {
                 req->scmnd->result = DID_RESET << 16;
                 req->scmnd->scsi_done(req->scmnd);
+                srp_unmap_data(req->scmnd, target, req);
         }
 
         target->rx_head  = 0;
         target->tx_head  = 0;
         target->tx_tail  = 0;
-        target->req_head = 0;
-        for (i = 0; i < SRP_SQ_SIZE - 1; ++i)
-                target->req_ring[i].next = i + 1;
-        target->req_ring[SRP_SQ_SIZE - 1].next = -1;
+        INIT_LIST_HEAD(&target->free_reqs);
         INIT_LIST_HEAD(&target->req_queue);
+        for (i = 0; i < SRP_SQ_SIZE; ++i)
+                list_add_tail(&target->req_ring[i].list, &target->free_reqs);
 
         ret = srp_connect_target(target);
         if (ret)
@@ -589,40 +617,10 @@ static int srp_map_data(struct scsi_cmnd *scmnd, struct srp_target_port *target,
         return len;
 }
 
-static void srp_unmap_data(struct scsi_cmnd *scmnd,
-                           struct srp_target_port *target,
-                           struct srp_request *req)
-{
-        struct scatterlist *scat;
-        int nents;
-
-        if (!scmnd->request_buffer ||
-            (scmnd->sc_data_direction != DMA_TO_DEVICE &&
-             scmnd->sc_data_direction != DMA_FROM_DEVICE))
-                return;
-
-        /*
-         * This handling of non-SG commands can be killed when the
-         * SCSI midlayer no longer generates non-SG commands.
-         */
-        if (likely(scmnd->use_sg)) {
-                nents = scmnd->use_sg;
-                scat  = scmnd->request_buffer;
-        } else {
-                nents = 1;
-                scat  = &req->fake_sg;
-        }
-
-        dma_unmap_sg(target->srp_host->dev->dma_device, scat, nents,
-                     scmnd->sc_data_direction);
-}
-
-static void srp_remove_req(struct srp_target_port *target, struct srp_request *req,
-                           int index)
+static void srp_remove_req(struct srp_target_port *target, struct srp_request *req)
 {
-        list_del(&req->list);
-        req->next = target->req_head;
-        target->req_head = index;
+        srp_unmap_data(req->scmnd, target, req);
+        list_move_tail(&req->list, &target->free_reqs);
 }
 
 static void srp_process_rsp(struct srp_target_port *target, struct srp_rsp *rsp)
@@ -647,7 +645,7 @@ static void srp_process_rsp(struct srp_target_port *target, struct srp_rsp *rsp)
                         req->tsk_status = rsp->data[3];
                 complete(&req->done);
         } else {
-                scmnd         = req->scmnd;
+                scmnd = req->scmnd;
                 if (!scmnd)
                         printk(KERN_ERR "Null scmnd for RSP w/tag %016llx\n",
                                (unsigned long long) rsp->tag);
@@ -665,14 +663,11 @@ static void srp_process_rsp(struct srp_target_port *target, struct srp_rsp *rsp)
                 else if (rsp->flags & (SRP_RSP_FLAG_DIOVER | SRP_RSP_FLAG_DIUNDER))
                         scmnd->resid = be32_to_cpu(rsp->data_in_res_cnt);
 
-                srp_unmap_data(scmnd, target, req);
-
                 if (!req->tsk_mgmt) {
-                        req->scmnd = NULL;
                         scmnd->host_scribble = (void *) -1L;
                         scmnd->scsi_done(scmnd);
 
-                        srp_remove_req(target, req, rsp->tag & ~SRP_TAG_TSK_MGMT);
+                        srp_remove_req(target, req);
                 } else
                         req->cmd_done = 1;
         }
@@ -859,7 +854,6 @@ static int srp_queuecommand(struct scsi_cmnd *scmnd,
         struct srp_request *req;
         struct srp_iu *iu;
         struct srp_cmd *cmd;
-        long req_index;
         int len;
 
         if (target->state == SRP_TARGET_CONNECTING)
@@ -879,22 +873,20 @@ static int srp_queuecommand(struct scsi_cmnd *scmnd,
         dma_sync_single_for_cpu(target->srp_host->dev->dma_device, iu->dma,
                                 SRP_MAX_IU_LEN, DMA_TO_DEVICE);
 
-        req_index = target->req_head;
+        req = list_entry(target->free_reqs.next, struct srp_request, list);
 
         scmnd->scsi_done     = done;
         scmnd->result        = 0;
-        scmnd->host_scribble = (void *) req_index;
+        scmnd->host_scribble = (void *) (long) req->index;
 
         cmd = iu->buf;
         memset(cmd, 0, sizeof *cmd);
 
         cmd->opcode = SRP_CMD;
         cmd->lun    = cpu_to_be64((u64) scmnd->device->lun << 48);
-        cmd->tag    = req_index;
+        cmd->tag    = req->index;
         memcpy(cmd->cdb, scmnd->cmnd, scmnd->cmd_len);
 
-        req = &target->req_ring[req_index];
-
         req->scmnd    = scmnd;
         req->cmd      = iu;
         req->cmd_done = 0;
@@ -919,8 +911,7 @@ static int srp_queuecommand(struct scsi_cmnd *scmnd,
                 goto err_unmap;
         }
 
-        target->req_head = req->next;
-        list_add_tail(&req->list, &target->req_queue);
+        list_move_tail(&req->list, &target->req_queue);
 
         return 0;
 
@@ -1143,30 +1134,20 @@ static int srp_cm_handler(struct ib_cm_id *cm_id, struct ib_cm_event *event)
         return 0;
 }
 
-static int srp_send_tsk_mgmt(struct scsi_cmnd *scmnd, u8 func)
+static int srp_send_tsk_mgmt(struct srp_target_port *target,
+                             struct srp_request *req, u8 func)
 {
-        struct srp_target_port *target = host_to_target(scmnd->device->host);
-        struct srp_request *req;
         struct srp_iu *iu;
         struct srp_tsk_mgmt *tsk_mgmt;
-        int req_index;
-        int ret = FAILED;
 
         spin_lock_irq(target->scsi_host->host_lock);
 
         if (target->state == SRP_TARGET_DEAD ||
             target->state == SRP_TARGET_REMOVED) {
-                scmnd->result = DID_BAD_TARGET << 16;
+                req->scmnd->result = DID_BAD_TARGET << 16;
                 goto out;
         }
 
-        if (scmnd->host_scribble == (void *) -1L)
-                goto out;
-
-        req_index = (long) scmnd->host_scribble;
-        printk(KERN_ERR "Abort for req_index %d\n", req_index);
-
-        req = &target->req_ring[req_index];
         init_completion(&req->done);
 
         iu = __srp_get_tx_iu(target);
@@ -1177,10 +1158,10 @@ static int srp_send_tsk_mgmt(struct scsi_cmnd *scmnd, u8 func)
         memset(tsk_mgmt, 0, sizeof *tsk_mgmt);
 
         tsk_mgmt->opcode        = SRP_TSK_MGMT;
-        tsk_mgmt->lun           = cpu_to_be64((u64) scmnd->device->lun << 48);
-        tsk_mgmt->tag           = req_index | SRP_TAG_TSK_MGMT;
+        tsk_mgmt->lun           = cpu_to_be64((u64) req->scmnd->device->lun << 48);
+        tsk_mgmt->tag           = req->index | SRP_TAG_TSK_MGMT;
         tsk_mgmt->tsk_mgmt_func = func;
-        tsk_mgmt->task_tag      = req_index;
+        tsk_mgmt->task_tag      = req->index;
 
         if (__srp_post_send(target, iu, sizeof *tsk_mgmt))
                 goto out;
@@ -1188,37 +1169,85 @@ static int srp_send_tsk_mgmt(struct scsi_cmnd *scmnd, u8 func)
         req->tsk_mgmt = iu;
 
         spin_unlock_irq(target->scsi_host->host_lock);
+
         if (!wait_for_completion_timeout(&req->done,
                                          msecs_to_jiffies(SRP_ABORT_TIMEOUT_MS)))
-                return FAILED;
-        spin_lock_irq(target->scsi_host->host_lock);
+                return -1;
 
-        if (req->cmd_done) {
-                srp_remove_req(target, req, req_index);
-                scmnd->scsi_done(scmnd);
-        } else if (!req->tsk_status) {
-                srp_remove_req(target, req, req_index);
-                scmnd->result = DID_ABORT << 16;
-                ret = SUCCESS;
-        }
+        return 0;
 
 out:
         spin_unlock_irq(target->scsi_host->host_lock);
-        return ret;
+        return -1;
+}
+
+static int srp_find_req(struct srp_target_port *target,
+                        struct scsi_cmnd *scmnd,
+                        struct srp_request **req)
+{
+        if (scmnd->host_scribble == (void *) -1L)
+                return -1;
+
+        *req = &target->req_ring[(long) scmnd->host_scribble];
+
+        return 0;
 }
 
 static int srp_abort(struct scsi_cmnd *scmnd)
 {
+        struct srp_target_port *target = host_to_target(scmnd->device->host);
+        struct srp_request *req;
+        int ret = SUCCESS;
+
         printk(KERN_ERR "SRP abort called\n");
 
-        return srp_send_tsk_mgmt(scmnd, SRP_TSK_ABORT_TASK);
+        if (srp_find_req(target, scmnd, &req))
+                return FAILED;
+        if (srp_send_tsk_mgmt(target, req, SRP_TSK_ABORT_TASK))
+                return FAILED;
+
+        spin_lock_irq(target->scsi_host->host_lock);
+
+        if (req->cmd_done) {
+                srp_remove_req(target, req);
+                scmnd->scsi_done(scmnd);
+        } else if (!req->tsk_status) {
+                srp_remove_req(target, req);
+                scmnd->result = DID_ABORT << 16;
+        } else
+                ret = FAILED;
+
+        spin_unlock_irq(target->scsi_host->host_lock);
+
+        return ret;
 }
 
 static int srp_reset_device(struct scsi_cmnd *scmnd)
 {
+        struct srp_target_port *target = host_to_target(scmnd->device->host);
+        struct srp_request *req, *tmp;
+
         printk(KERN_ERR "SRP reset_device called\n");
 
-        return srp_send_tsk_mgmt(scmnd, SRP_TSK_LUN_RESET);
+        if (srp_find_req(target, scmnd, &req))
+                return FAILED;
+        if (srp_send_tsk_mgmt(target, req, SRP_TSK_LUN_RESET))
+                return FAILED;
+        if (req->tsk_status)
+                return FAILED;
+
+        spin_lock_irq(target->scsi_host->host_lock);
+
+        list_for_each_entry_safe(req, tmp, &target->req_queue, list)
+                if (req->scmnd->device == scmnd->device) {
+                        req->scmnd->result = DID_RESET << 16;
+                        scmnd->scsi_done(scmnd);
+                        srp_remove_req(target, req);
+                }
+
+        spin_unlock_irq(target->scsi_host->host_lock);
+
+        return SUCCESS;
 }
 
 static int srp_reset_host(struct scsi_cmnd *scmnd)
@@ -1518,10 +1547,12 @@ static ssize_t srp_create_target(struct class_device *class_dev,
 
         INIT_WORK(&target->work, srp_reconnect_work, target);
 
-        for (i = 0; i < SRP_SQ_SIZE - 1; ++i)
-                target->req_ring[i].next = i + 1;
-        target->req_ring[SRP_SQ_SIZE - 1].next = -1;
+        INIT_LIST_HEAD(&target->free_reqs);
         INIT_LIST_HEAD(&target->req_queue);
+        for (i = 0; i < SRP_SQ_SIZE; ++i) {
+                target->req_ring[i].index = i;
+                list_add_tail(&target->req_ring[i].list, &target->free_reqs);
+        }
 
         ret = srp_parse_options(buf, target);
         if (ret)
diff --git a/drivers/infiniband/ulp/srp/ib_srp.h b/drivers/infiniband/ulp/srp/ib_srp.h
index bd7f7c3115de..c5cd43aae860 100644
--- a/drivers/infiniband/ulp/srp/ib_srp.h
+++ b/drivers/infiniband/ulp/srp/ib_srp.h
@@ -101,7 +101,7 @@ struct srp_request {
          */
         struct scatterlist      fake_sg;
         struct completion       done;
-        short                   next;
+        short                   index;
         u8                      cmd_done;
         u8                      tsk_status;
 };
@@ -133,7 +133,7 @@ struct srp_target_port {
         unsigned                tx_tail;
         struct srp_iu          *tx_ring[SRP_SQ_SIZE + 1];
 
-        int                     req_head;
+        struct list_head        free_reqs;
         struct list_head        req_queue;
         struct srp_request      req_ring[SRP_SQ_SIZE];