aboutsummaryrefslogtreecommitdiff
path: root/drivers/infiniband/hw/qib
diff options
context:
space:
mode:
Diffstat (limited to 'drivers/infiniband/hw/qib')
-rw-r--r--drivers/infiniband/hw/qib/qib.h12
-rw-r--r--drivers/infiniband/hw/qib/qib_diag.c52
-rw-r--r--drivers/infiniband/hw/qib/qib_dma.c21
-rw-r--r--drivers/infiniband/hw/qib/qib_file_ops.c5
-rw-r--r--drivers/infiniband/hw/qib/qib_fs.c8
-rw-r--r--drivers/infiniband/hw/qib/qib_iba6120.c11
-rw-r--r--drivers/infiniband/hw/qib/qib_iba7220.c12
-rw-r--r--drivers/infiniband/hw/qib/qib_iba7322.c53
-rw-r--r--drivers/infiniband/hw/qib/qib_init.c104
-rw-r--r--drivers/infiniband/hw/qib/qib_mad.c46
-rw-r--r--drivers/infiniband/hw/qib/qib_mad.h14
-rw-r--r--drivers/infiniband/hw/qib/qib_mr.c14
-rw-r--r--drivers/infiniband/hw/qib/qib_pcie.c181
-rw-r--r--drivers/infiniband/hw/qib/qib_qp.c5
-rw-r--r--drivers/infiniband/hw/qib/qib_rc.c2
-rw-r--r--drivers/infiniband/hw/qib/qib_ruc.c1
-rw-r--r--drivers/infiniband/hw/qib/qib_ud.c15
-rw-r--r--drivers/infiniband/hw/qib/qib_user_sdma.c142
-rw-r--r--drivers/infiniband/hw/qib/qib_verbs.c8
-rw-r--r--drivers/infiniband/hw/qib/qib_verbs.h26
20 files changed, 436 insertions, 296 deletions
diff --git a/drivers/infiniband/hw/qib/qib.h b/drivers/infiniband/hw/qib/qib.h
index 1946101419a..c00ae093b6f 100644
--- a/drivers/infiniband/hw/qib/qib.h
+++ b/drivers/infiniband/hw/qib/qib.h
@@ -868,8 +868,10 @@ struct qib_devdata {
/* last buffer for user use */
u32 lastctxt_piobuf;
- /* saturating counter of (non-port-specific) device interrupts */
- u32 int_counter;
+ /* reset value */
+ u64 z_int_counter;
+ /* percpu intcounter */
+ u64 __percpu *int_counter;
/* pio bufs allocated per ctxt */
u32 pbufsctxt;
@@ -1184,7 +1186,7 @@ int qib_setup_eagerbufs(struct qib_ctxtdata *);
void qib_set_ctxtcnt(struct qib_devdata *);
int qib_create_ctxts(struct qib_devdata *dd);
struct qib_ctxtdata *qib_create_ctxtdata(struct qib_pportdata *, u32, int);
-void qib_init_pportdata(struct qib_pportdata *, struct qib_devdata *, u8, u8);
+int qib_init_pportdata(struct qib_pportdata *, struct qib_devdata *, u8, u8);
void qib_free_ctxtdata(struct qib_devdata *, struct qib_ctxtdata *);
u32 qib_kreceive(struct qib_ctxtdata *, u32 *, u32 *);
@@ -1449,6 +1451,10 @@ void qib_nomsi(struct qib_devdata *);
void qib_nomsix(struct qib_devdata *);
void qib_pcie_getcmd(struct qib_devdata *, u16 *, u8 *, u8 *);
void qib_pcie_reenable(struct qib_devdata *, u16, u8, u8);
+/* interrupts for device */
+u64 qib_int_counter(struct qib_devdata *);
+/* interrupt for all devices */
+u64 qib_sps_ints(void);
/*
* dma_addr wrappers - all 0's invalid for hw
diff --git a/drivers/infiniband/hw/qib/qib_diag.c b/drivers/infiniband/hw/qib/qib_diag.c
index 1686fd4bda8..5dfda4c5cc9 100644
--- a/drivers/infiniband/hw/qib/qib_diag.c
+++ b/drivers/infiniband/hw/qib/qib_diag.c
@@ -546,7 +546,7 @@ static ssize_t qib_diagpkt_write(struct file *fp,
size_t count, loff_t *off)
{
u32 __iomem *piobuf;
- u32 plen, clen, pbufn;
+ u32 plen, pbufn, maxlen_reserve;
struct qib_diag_xpkt dp;
u32 *tmpbuf = NULL;
struct qib_devdata *dd;
@@ -590,15 +590,20 @@ static ssize_t qib_diagpkt_write(struct file *fp,
}
ppd = &dd->pport[dp.port - 1];
- /* need total length before first word written */
- /* +1 word is for the qword padding */
- plen = sizeof(u32) + dp.len;
- clen = dp.len >> 2;
-
- if ((plen + 4) > ppd->ibmaxlen) {
+ /*
+ * need total length before first word written, plus 2 Dwords. One Dword
+ * is for padding so we get the full user data when not aligned on
+ * a word boundary. The other Dword is to make sure we have room for the
+ * ICRC which gets tacked on later.
+ */
+ maxlen_reserve = 2 * sizeof(u32);
+ if (dp.len > ppd->ibmaxlen - maxlen_reserve) {
ret = -EINVAL;
- goto bail; /* before writing pbc */
+ goto bail;
}
+
+ plen = sizeof(u32) + dp.len;
+
tmpbuf = vmalloc(plen);
if (!tmpbuf) {
qib_devinfo(dd->pcidev,
@@ -638,11 +643,11 @@ static ssize_t qib_diagpkt_write(struct file *fp,
*/
if (dd->flags & QIB_PIO_FLUSH_WC) {
qib_flush_wc();
- qib_pio_copy(piobuf + 2, tmpbuf, clen - 1);
+ qib_pio_copy(piobuf + 2, tmpbuf, plen - 1);
qib_flush_wc();
- __raw_writel(tmpbuf[clen - 1], piobuf + clen + 1);
+ __raw_writel(tmpbuf[plen - 1], piobuf + plen + 1);
} else
- qib_pio_copy(piobuf + 2, tmpbuf, clen);
+ qib_pio_copy(piobuf + 2, tmpbuf, plen);
if (dd->flags & QIB_USE_SPCL_TRIG) {
u32 spcl_off = (pbufn >= dd->piobcnt2k) ? 2047 : 1023;
@@ -689,28 +694,23 @@ int qib_register_observer(struct qib_devdata *dd,
const struct diag_observer *op)
{
struct diag_observer_list_elt *olp;
- int ret = -EINVAL;
+ unsigned long flags;
if (!dd || !op)
- goto bail;
- ret = -ENOMEM;
+ return -EINVAL;
olp = vmalloc(sizeof *olp);
if (!olp) {
pr_err("vmalloc for observer failed\n");
- goto bail;
+ return -ENOMEM;
}
- if (olp) {
- unsigned long flags;
- spin_lock_irqsave(&dd->qib_diag_trans_lock, flags);
- olp->op = op;
- olp->next = dd->diag_observer_list;
- dd->diag_observer_list = olp;
- spin_unlock_irqrestore(&dd->qib_diag_trans_lock, flags);
- ret = 0;
- }
-bail:
- return ret;
+ spin_lock_irqsave(&dd->qib_diag_trans_lock, flags);
+ olp->op = op;
+ olp->next = dd->diag_observer_list;
+ dd->diag_observer_list = olp;
+ spin_unlock_irqrestore(&dd->qib_diag_trans_lock, flags);
+
+ return 0;
}
/* Remove all registered observers when device is closed */
diff --git a/drivers/infiniband/hw/qib/qib_dma.c b/drivers/infiniband/hw/qib/qib_dma.c
index 2920bb39a65..59fe092b4b0 100644
--- a/drivers/infiniband/hw/qib/qib_dma.c
+++ b/drivers/infiniband/hw/qib/qib_dma.c
@@ -108,6 +108,10 @@ static int qib_map_sg(struct ib_device *dev, struct scatterlist *sgl,
ret = 0;
break;
}
+ sg->dma_address = addr + sg->offset;
+#ifdef CONFIG_NEED_SG_DMA_LENGTH
+ sg->dma_length = sg->length;
+#endif
}
return ret;
}
@@ -119,21 +123,6 @@ static void qib_unmap_sg(struct ib_device *dev,
BUG_ON(!valid_dma_direction(direction));
}
-static u64 qib_sg_dma_address(struct ib_device *dev, struct scatterlist *sg)
-{
- u64 addr = (u64) page_address(sg_page(sg));
-
- if (addr)
- addr += sg->offset;
- return addr;
-}
-
-static unsigned int qib_sg_dma_len(struct ib_device *dev,
- struct scatterlist *sg)
-{
- return sg->length;
-}
-
static void qib_sync_single_for_cpu(struct ib_device *dev, u64 addr,
size_t size, enum dma_data_direction dir)
{
@@ -173,8 +162,6 @@ struct ib_dma_mapping_ops qib_dma_mapping_ops = {
.unmap_page = qib_dma_unmap_page,
.map_sg = qib_map_sg,
.unmap_sg = qib_unmap_sg,
- .dma_address = qib_sg_dma_address,
- .dma_len = qib_sg_dma_len,
.sync_single_for_cpu = qib_sync_single_for_cpu,
.sync_single_for_device = qib_sync_single_for_device,
.alloc_coherent = qib_dma_alloc_coherent,
diff --git a/drivers/infiniband/hw/qib/qib_file_ops.c b/drivers/infiniband/hw/qib/qib_file_ops.c
index 275f247f9fc..b15e34eeef6 100644
--- a/drivers/infiniband/hw/qib/qib_file_ops.c
+++ b/drivers/infiniband/hw/qib/qib_file_ops.c
@@ -1459,7 +1459,7 @@ static int get_a_ctxt(struct file *fp, const struct qib_user_info *uinfo,
cused++;
else
cfree++;
- if (pusable && cfree && cused < inuse) {
+ if (cfree && cused < inuse) {
udd = dd;
inuse = cused;
}
@@ -1578,7 +1578,7 @@ static int do_qib_user_sdma_queue_create(struct file *fp)
struct qib_ctxtdata *rcd = fd->rcd;
struct qib_devdata *dd = rcd->dd;
- if (dd->flags & QIB_HAS_SEND_DMA)
+ if (dd->flags & QIB_HAS_SEND_DMA) {
fd->pq = qib_user_sdma_queue_create(&dd->pcidev->dev,
dd->unit,
@@ -1586,6 +1586,7 @@ static int do_qib_user_sdma_queue_create(struct file *fp)
fd->subctxt);
if (!fd->pq)
return -ENOMEM;
+ }
return 0;
}
diff --git a/drivers/infiniband/hw/qib/qib_fs.c b/drivers/infiniband/hw/qib/qib_fs.c
index f247fc6e618..cab610ccd50 100644
--- a/drivers/infiniband/hw/qib/qib_fs.c
+++ b/drivers/infiniband/hw/qib/qib_fs.c
@@ -105,6 +105,7 @@ static int create_file(const char *name, umode_t mode,
static ssize_t driver_stats_read(struct file *file, char __user *buf,
size_t count, loff_t *ppos)
{
+ qib_stats.sps_ints = qib_sps_ints();
return simple_read_from_buffer(buf, count, ppos, &qib_stats,
sizeof qib_stats);
}
@@ -456,13 +457,13 @@ static int remove_file(struct dentry *parent, char *name)
spin_lock(&tmp->d_lock);
if (!(d_unhashed(tmp) && tmp->d_inode)) {
- dget_dlock(tmp);
__d_drop(tmp);
spin_unlock(&tmp->d_lock);
simple_unlink(parent->d_inode, tmp);
} else {
spin_unlock(&tmp->d_lock);
}
+ dput(tmp);
ret = 0;
bail:
@@ -491,6 +492,7 @@ static int remove_device_files(struct super_block *sb,
goto bail;
}
+ mutex_lock(&dir->d_inode->i_mutex);
remove_file(dir, "counters");
remove_file(dir, "counter_names");
remove_file(dir, "portcounter_names");
@@ -505,8 +507,10 @@ static int remove_device_files(struct super_block *sb,
}
}
remove_file(dir, "flash");
- d_delete(dir);
+ mutex_unlock(&dir->d_inode->i_mutex);
ret = simple_rmdir(root->d_inode, dir);
+ d_delete(dir);
+ dput(dir);
bail:
mutex_unlock(&root->d_inode->i_mutex);
diff --git a/drivers/infiniband/hw/qib/qib_iba6120.c b/drivers/infiniband/hw/qib/qib_iba6120.c
index 84e593d6007..d68266ac761 100644
--- a/drivers/infiniband/hw/qib/qib_iba6120.c
+++ b/drivers/infiniband/hw/qib/qib_iba6120.c
@@ -1634,9 +1634,7 @@ static irqreturn_t qib_6120intr(int irq, void *data)
goto bail;
}
- qib_stats.sps_ints++;
- if (dd->int_counter != (u32) -1)
- dd->int_counter++;
+ this_cpu_inc(*dd->int_counter);
if (unlikely(istat & (~QLOGIC_IB_I_BITSEXTANT |
QLOGIC_IB_I_GPIO | QLOGIC_IB_I_ERROR)))
@@ -1808,7 +1806,8 @@ static int qib_6120_setup_reset(struct qib_devdata *dd)
* isn't set.
*/
dd->flags &= ~(QIB_INITTED | QIB_PRESENT);
- dd->int_counter = 0; /* so we check interrupts work again */
+ /* so we check interrupts work again */
+ dd->z_int_counter = qib_int_counter(dd);
val = dd->control | QLOGIC_IB_C_RESET;
writeq(val, &dd->kregbase[kr_control]);
mb(); /* prevent compiler re-ordering around actual reset */
@@ -3266,7 +3265,9 @@ static int init_6120_variables(struct qib_devdata *dd)
dd->eep_st_masks[2].errs_to_log = ERR_MASK(ResetNegated);
- qib_init_pportdata(ppd, dd, 0, 1);
+ ret = qib_init_pportdata(ppd, dd, 0, 1);
+ if (ret)
+ goto bail;
ppd->link_width_supported = IB_WIDTH_1X | IB_WIDTH_4X;
ppd->link_speed_supported = QIB_IB_SDR;
ppd->link_width_enabled = IB_WIDTH_4X;
diff --git a/drivers/infiniband/hw/qib/qib_iba7220.c b/drivers/infiniband/hw/qib/qib_iba7220.c
index 454c2e7668f..7dec89fdc12 100644
--- a/drivers/infiniband/hw/qib/qib_iba7220.c
+++ b/drivers/infiniband/hw/qib/qib_iba7220.c
@@ -1962,10 +1962,7 @@ static irqreturn_t qib_7220intr(int irq, void *data)
goto bail;
}
- qib_stats.sps_ints++;
- if (dd->int_counter != (u32) -1)
- dd->int_counter++;
-
+ this_cpu_inc(*dd->int_counter);
if (unlikely(istat & (~QLOGIC_IB_I_BITSEXTANT |
QLOGIC_IB_I_GPIO | QLOGIC_IB_I_ERROR)))
unlikely_7220_intr(dd, istat);
@@ -2120,7 +2117,8 @@ static int qib_setup_7220_reset(struct qib_devdata *dd)
* isn't set.
*/
dd->flags &= ~(QIB_INITTED | QIB_PRESENT);
- dd->int_counter = 0; /* so we check interrupts work again */
+ /* so we check interrupts work again */
+ dd->z_int_counter = qib_int_counter(dd);
val = dd->control | QLOGIC_IB_C_RESET;
writeq(val, &dd->kregbase[kr_control]);
mb(); /* prevent compiler reordering around actual reset */
@@ -4061,7 +4059,9 @@ static int qib_init_7220_variables(struct qib_devdata *dd)
init_waitqueue_head(&cpspec->autoneg_wait);
INIT_DELAYED_WORK(&cpspec->autoneg_work, autoneg_7220_work);
- qib_init_pportdata(ppd, dd, 0, 1);
+ ret = qib_init_pportdata(ppd, dd, 0, 1);
+ if (ret)
+ goto bail;
ppd->link_width_supported = IB_WIDTH_1X | IB_WIDTH_4X;
ppd->link_speed_supported = QIB_IB_SDR | QIB_IB_DDR;
diff --git a/drivers/infiniband/hw/qib/qib_iba7322.c b/drivers/infiniband/hw/qib/qib_iba7322.c
index 016e7429adf..a7eb32517a0 100644
--- a/drivers/infiniband/hw/qib/qib_iba7322.c
+++ b/drivers/infiniband/hw/qib/qib_iba7322.c
@@ -2395,6 +2395,11 @@ static int qib_7322_bringup_serdes(struct qib_pportdata *ppd)
qib_write_kreg_port(ppd, krp_ibcctrl_a, ppd->cpspec->ibcctrl_a);
qib_write_kreg(dd, kr_scratch, 0ULL);
+ /* ensure previous Tx parameters are not still forced */
+ qib_write_kreg_port(ppd, krp_tx_deemph_override,
+ SYM_MASK(IBSD_TX_DEEMPHASIS_OVERRIDE_0,
+ reset_tx_deemphasis_override));
+
if (qib_compat_ddr_negotiate) {
ppd->cpspec->ibdeltainprog = 1;
ppd->cpspec->ibsymsnap = read_7322_creg32_port(ppd,
@@ -3110,9 +3115,7 @@ static irqreturn_t qib_7322intr(int irq, void *data)
goto bail;
}
- qib_stats.sps_ints++;
- if (dd->int_counter != (u32) -1)
- dd->int_counter++;
+ this_cpu_inc(*dd->int_counter);
/* handle "errors" of various kinds first, device ahead of port */
if (unlikely(istat & (~QIB_I_BITSEXTANT | QIB_I_GPIO |
@@ -3181,9 +3184,7 @@ static irqreturn_t qib_7322pintr(int irq, void *data)
*/
return IRQ_HANDLED;
- qib_stats.sps_ints++;
- if (dd->int_counter != (u32) -1)
- dd->int_counter++;
+ this_cpu_inc(*dd->int_counter);
/* Clear the interrupt bit we expect to be set. */
qib_write_kreg(dd, kr_intclear, ((1ULL << QIB_I_RCVAVAIL_LSB) |
@@ -3210,9 +3211,7 @@ static irqreturn_t qib_7322bufavail(int irq, void *data)
*/
return IRQ_HANDLED;
- qib_stats.sps_ints++;
- if (dd->int_counter != (u32) -1)
- dd->int_counter++;
+ this_cpu_inc(*dd->int_counter);
/* Clear the interrupt bit we expect to be set. */
qib_write_kreg(dd, kr_intclear, QIB_I_SPIOBUFAVAIL);
@@ -3243,9 +3242,7 @@ static irqreturn_t sdma_intr(int irq, void *data)
*/
return IRQ_HANDLED;
- qib_stats.sps_ints++;
- if (dd->int_counter != (u32) -1)
- dd->int_counter++;
+ this_cpu_inc(*dd->int_counter);
/* Clear the interrupt bit we expect to be set. */
qib_write_kreg(dd, kr_intclear, ppd->hw_pidx ?
@@ -3272,9 +3269,7 @@ static irqreturn_t sdma_idle_intr(int irq, void *data)
*/
return IRQ_HANDLED;
- qib_stats.sps_ints++;
- if (dd->int_counter != (u32) -1)
- dd->int_counter++;
+ this_cpu_inc(*dd->int_counter);
/* Clear the interrupt bit we expect to be set. */
qib_write_kreg(dd, kr_intclear, ppd->hw_pidx ?
@@ -3301,9 +3296,7 @@ static irqreturn_t sdma_progress_intr(int irq, void *data)
*/
return IRQ_HANDLED;
- qib_stats.sps_ints++;
- if (dd->int_counter != (u32) -1)
- dd->int_counter++;
+ this_cpu_inc(*dd->int_counter);
/* Clear the interrupt bit we expect to be set. */
qib_write_kreg(dd, kr_intclear, ppd->hw_pidx ?
@@ -3331,9 +3324,7 @@ static irqreturn_t sdma_cleanup_intr(int irq, void *data)
*/
return IRQ_HANDLED;
- qib_stats.sps_ints++;
- if (dd->int_counter != (u32) -1)
- dd->int_counter++;
+ this_cpu_inc(*dd->int_counter);
/* Clear the interrupt bit we expect to be set. */
qib_write_kreg(dd, kr_intclear, ppd->hw_pidx ?
@@ -3718,7 +3709,8 @@ static int qib_do_7322_reset(struct qib_devdata *dd)
dd->pport->cpspec->ibsymdelta = 0;
dd->pport->cpspec->iblnkerrdelta = 0;
dd->pport->cpspec->ibmalfdelta = 0;
- dd->int_counter = 0; /* so we check interrupts work again */
+ /* so we check interrupts work again */
+ dd->z_int_counter = qib_int_counter(dd);
/*
* Keep chip from being accessed until we are ready. Use
@@ -6190,21 +6182,20 @@ static int setup_txselect(const char *str, struct kernel_param *kp)
{
struct qib_devdata *dd;
unsigned long val;
- int ret;
-
+ char *n;
if (strlen(str) >= MAX_ATTEN_LEN) {
pr_info("txselect_values string too long\n");
return -ENOSPC;
}
- ret = kstrtoul(str, 0, &val);
- if (ret || val >= (TXDDS_TABLE_SZ + TXDDS_EXTRA_SZ +
+ val = simple_strtoul(str, &n, 0);
+ if (n == str || val >= (TXDDS_TABLE_SZ + TXDDS_EXTRA_SZ +
TXDDS_MFG_SZ)) {
pr_info("txselect_values must start with a number < %d\n",
TXDDS_TABLE_SZ + TXDDS_EXTRA_SZ + TXDDS_MFG_SZ);
- return ret ? ret : -EINVAL;
+ return -EINVAL;
}
-
strcpy(txselect_list, str);
+
list_for_each_entry(dd, &qib_dev_list, list)
if (dd->deviceid == PCI_DEVICE_ID_QLOGIC_IB_7322)
set_no_qsfp_atten(dd, 1);
@@ -6553,7 +6544,11 @@ static int qib_init_7322_variables(struct qib_devdata *dd)
}
dd->num_pports++;
- qib_init_pportdata(ppd, dd, pidx, dd->num_pports);
+ ret = qib_init_pportdata(ppd, dd, pidx, dd->num_pports);
+ if (ret) {
+ dd->num_pports--;
+ goto bail;
+ }
ppd->link_width_supported = IB_WIDTH_1X | IB_WIDTH_4X;
ppd->link_width_enabled = IB_WIDTH_4X;
diff --git a/drivers/infiniband/hw/qib/qib_init.c b/drivers/infiniband/hw/qib/qib_init.c
index 24e802f4ea2..8d3c78ddc90 100644
--- a/drivers/infiniband/hw/qib/qib_init.c
+++ b/drivers/infiniband/hw/qib/qib_init.c
@@ -130,7 +130,6 @@ void qib_set_ctxtcnt(struct qib_devdata *dd)
int qib_create_ctxts(struct qib_devdata *dd)
{
unsigned i;
- int ret;
int local_node_id = pcibus_to_node(dd->pcidev->bus);
if (local_node_id < 0)
@@ -145,8 +144,7 @@ int qib_create_ctxts(struct qib_devdata *dd)
if (!dd->rcd) {
qib_dev_err(dd,
"Unable to allocate ctxtdata array, failing\n");
- ret = -ENOMEM;
- goto done;
+ return -ENOMEM;
}
/* create (one or more) kctxt */
@@ -163,15 +161,14 @@ int qib_create_ctxts(struct qib_devdata *dd)
if (!rcd) {
qib_dev_err(dd,
"Unable to allocate ctxtdata for Kernel ctxt, failing\n");
- ret = -ENOMEM;
- goto done;
+ kfree(dd->rcd);
+ dd->rcd = NULL;
+ return -ENOMEM;
}
rcd->pkeys[0] = QIB_DEFAULT_P_KEY;
rcd->seq_cnt = 1;
}
- ret = 0;
-done:
- return ret;
+ return 0;
}
/*
@@ -233,7 +230,7 @@ struct qib_ctxtdata *qib_create_ctxtdata(struct qib_pportdata *ppd, u32 ctxt,
/*
* Common code for initializing the physical port structure.
*/
-void qib_init_pportdata(struct qib_pportdata *ppd, struct qib_devdata *dd,
+int qib_init_pportdata(struct qib_pportdata *ppd, struct qib_devdata *dd,
u8 hw_pidx, u8 port)
{
int size;
@@ -243,6 +240,7 @@ void qib_init_pportdata(struct qib_pportdata *ppd, struct qib_devdata *dd,
spin_lock_init(&ppd->sdma_lock);
spin_lock_init(&ppd->lflags_lock);
+ spin_lock_init(&ppd->cc_shadow_lock);
init_waitqueue_head(&ppd->state_wait);
init_timer(&ppd->symerr_clear_timer);
@@ -250,8 +248,10 @@ void qib_init_pportdata(struct qib_pportdata *ppd, struct qib_devdata *dd,
ppd->symerr_clear_timer.data = (unsigned long)ppd;
ppd->qib_wq = NULL;
-
- spin_lock_init(&ppd->cc_shadow_lock);
+ ppd->ibport_data.pmastats =
+ alloc_percpu(struct qib_pma_counters);
+ if (!ppd->ibport_data.pmastats)
+ return -ENOMEM;
if (qib_cc_table_size < IB_CCT_MIN_ENTRIES)
goto bail;
@@ -299,7 +299,7 @@ void qib_init_pportdata(struct qib_pportdata *ppd, struct qib_devdata *dd,
goto bail_3;
}
- return;
+ return 0;
bail_3:
kfree(ppd->ccti_entries_shadow);
@@ -313,7 +313,7 @@ bail_1:
bail:
/* User is intentionally disabling the congestion control agent */
if (!qib_cc_table_size)
- return;
+ return 0;
if (qib_cc_table_size < IB_CCT_MIN_ENTRIES) {
qib_cc_table_size = 0;
@@ -324,7 +324,7 @@ bail:
qib_dev_err(dd, "Congestion Control Agent disabled for port %d\n",
port);
- return;
+ return 0;
}
static int init_pioavailregs(struct qib_devdata *dd)
@@ -525,6 +525,7 @@ static void enable_chip(struct qib_devdata *dd)
static void verify_interrupt(unsigned long opaque)
{
struct qib_devdata *dd = (struct qib_devdata *) opaque;
+ u64 int_counter;
if (!dd)
return; /* being torn down */
@@ -533,7 +534,8 @@ static void verify_interrupt(unsigned long opaque)
* If we don't have a lid or any interrupts, let the user know and
* don't bother checking again.
*/
- if (dd->int_counter == 0) {
+ int_counter = qib_int_counter(dd) - dd->z_int_counter;
+ if (int_counter == 0) {
if (!dd->f_intr_fallback(dd))
dev_err(&dd->pcidev->dev,
"No interrupts detected, not usable.\n");
@@ -633,6 +635,12 @@ wq_error:
return -ENOMEM;
}
+static void qib_free_pportdata(struct qib_pportdata *ppd)
+{
+ free_percpu(ppd->ibport_data.pmastats);
+ ppd->ibport_data.pmastats = NULL;
+}
+
/**
* qib_init - do the actual initialization sequence on the chip
* @dd: the qlogic_ib device
@@ -920,6 +928,7 @@ static void qib_shutdown_device(struct qib_devdata *dd)
destroy_workqueue(ppd->qib_wq);
ppd->qib_wq = NULL;
}
+ qib_free_pportdata(ppd);
}
qib_update_eeprom_log(dd);
@@ -1079,9 +1088,34 @@ void qib_free_devdata(struct qib_devdata *dd)
#ifdef CONFIG_DEBUG_FS
qib_dbg_ibdev_exit(&dd->verbs_dev);
#endif
+ free_percpu(dd->int_counter);
ib_dealloc_device(&dd->verbs_dev.ibdev);
}
+u64 qib_int_counter(struct qib_devdata *dd)
+{
+ int cpu;
+ u64 int_counter = 0;
+
+ for_each_possible_cpu(cpu)
+ int_counter += *per_cpu_ptr(dd->int_counter, cpu);
+ return int_counter;
+}
+
+u64 qib_sps_ints(void)
+{
+ unsigned long flags;
+ struct qib_devdata *dd;
+ u64 sps_ints = 0;
+
+ spin_lock_irqsave(&qib_devs_lock, flags);
+ list_for_each_entry(dd, &qib_dev_list, list) {
+ sps_ints += qib_int_counter(dd);
+ }
+ spin_unlock_irqrestore(&qib_devs_lock, flags);
+ return sps_ints;
+}
+
/*
* Allocate our primary per-unit data structure. Must be done via verbs
* allocator, because the verbs cleanup process both does cleanup and
@@ -1097,14 +1131,10 @@ struct qib_devdata *qib_alloc_devdata(struct pci_dev *pdev, size_t extra)
int ret;
dd = (struct qib_devdata *) ib_alloc_device(sizeof(*dd) + extra);
- if (!dd) {
- dd = ERR_PTR(-ENOMEM);
- goto bail;
- }
+ if (!dd)
+ return ERR_PTR(-ENOMEM);
-#ifdef CONFIG_DEBUG_FS
- qib_dbg_ibdev_init(&dd->verbs_dev);
-#endif
+ INIT_LIST_HEAD(&dd->list);
idr_preload(GFP_KERNEL);
spin_lock_irqsave(&qib_devs_lock, flags);
@@ -1121,11 +1151,13 @@ struct qib_devdata *qib_alloc_devdata(struct pci_dev *pdev, size_t extra)
if (ret < 0) {
qib_early_err(&pdev->dev,
"Could not allocate unit ID: error %d\n", -ret);
-#ifdef CONFIG_DEBUG_FS
- qib_dbg_ibdev_exit(&dd->verbs_dev);
-#endif
- ib_dealloc_device(&dd->verbs_dev.ibdev);
- dd = ERR_PTR(ret);
+ goto bail;
+ }
+ dd->int_counter = alloc_percpu(u64);
+ if (!dd->int_counter) {
+ ret = -ENOMEM;
+ qib_early_err(&pdev->dev,
+ "Could not allocate per-cpu int_counter\n");
goto bail;
}
@@ -1139,9 +1171,15 @@ struct qib_devdata *qib_alloc_devdata(struct pci_dev *pdev, size_t extra)
qib_early_err(&pdev->dev,
"Could not alloc cpulist info, cpu affinity might be wrong\n");
}
-
-bail:
+#ifdef CONFIG_DEBUG_FS
+ qib_dbg_ibdev_init(&dd->verbs_dev);
+#endif
return dd;
+bail:
+ if (!list_empty(&dd->list))
+ list_del_init(&dd->list);
+ ib_dealloc_device(&dd->verbs_dev.ibdev);
+ return ERR_PTR(ret);;
}
/*
@@ -1234,7 +1272,7 @@ static int qib_notify_dca(struct notifier_block *nb, unsigned long event,
* Do all the generic driver unit- and chip-independent memory
* allocation and initialization.
*/
-static int __init qlogic_ib_init(void)
+static int __init qib_ib_init(void)
{
int ret;
@@ -1278,12 +1316,12 @@ bail:
return ret;
}
-module_init(qlogic_ib_init);
+module_init(qib_ib_init);
/*
* Do the non-unit driver cleanup, memory free, etc. at unload.
*/
-static void __exit qlogic_ib_cleanup(void)
+static void __exit qib_ib_cleanup(void)
{
int ret;
@@ -1308,7 +1346,7 @@ static void __exit qlogic_ib_cleanup(void)
qib_dev_cleanup();
}
-module_exit(qlogic_ib_cleanup);
+module_exit(qib_ib_cleanup);
/* this can only be called after a successful initialization */
static void cleanup_device_data(struct qib_devdata *dd)
diff --git a/drivers/infiniband/hw/qib/qib_mad.c b/drivers/infiniband/hw/qib/qib_mad.c
index ccb119143d2..22c720e5740 100644
--- a/drivers/infiniband/hw/qib/qib_mad.c
+++ b/drivers/infiniband/hw/qib/qib_mad.c
@@ -1028,7 +1028,7 @@ static int set_pkeys(struct qib_devdata *dd, u8 port, u16 *pkeys)
event.event = IB_EVENT_PKEY_CHANGE;
event.device = &dd->verbs_dev.ibdev;
- event.element.port_num = 1;
+ event.element.port_num = port;
ib_dispatch_event(&event);
}
return 0;
@@ -1634,6 +1634,23 @@ static int pma_get_portcounters_cong(struct ib_pma_mad *pmp,
return reply((struct ib_smp *)pmp);
}
+static void qib_snapshot_pmacounters(
+ struct qib_ibport *ibp,
+ struct qib_pma_counters *pmacounters)
+{
+ struct qib_pma_counters *p;
+ int cpu;
+
+ memset(pmacounters, 0, sizeof(*pmacounters));
+ for_each_possible_cpu(cpu) {
+ p = per_cpu_ptr(ibp->pmastats, cpu);
+ pmacounters->n_unicast_xmit += p->n_unicast_xmit;
+ pmacounters->n_unicast_rcv += p->n_unicast_rcv;
+ pmacounters->n_multicast_xmit += p->n_multicast_xmit;
+ pmacounters->n_multicast_rcv += p->n_multicast_rcv;
+ }
+}
+
static int pma_get_portcounters_ext(struct ib_pma_mad *pmp,
struct ib_device *ibdev, u8 port)
{
@@ -1642,6 +1659,7 @@ static int pma_get_portcounters_ext(struct ib_pma_mad *pmp,
struct qib_ibport *ibp = to_iport(ibdev, port);
struct qib_pportdata *ppd = ppd_from_ibp(ibp);
u64 swords, rwords, spkts, rpkts, xwait;
+ struct qib_pma_counters pma;
u8 port_select = p->port_select;
memset(pmp->data, 0, sizeof(pmp->data));
@@ -1664,10 +1682,17 @@ static int pma_get_portcounters_ext(struct ib_pma_mad *pmp,
p->port_rcv_data = cpu_to_be64(rwords);
p->port_xmit_packets = cpu_to_be64(spkts);
p->port_rcv_packets = cpu_to_be64(rpkts);
- p->port_unicast_xmit_packets = cpu_to_be64(ibp->n_unicast_xmit);
- p->port_unicast_rcv_packets = cpu_to_be64(ibp->n_unicast_rcv);
- p->port_multicast_xmit_packets = cpu_to_be64(ibp->n_multicast_xmit);
- p->port_multicast_rcv_packets = cpu_to_be64(ibp->n_multicast_rcv);
+
+ qib_snapshot_pmacounters(ibp, &pma);
+
+ p->port_unicast_xmit_packets = cpu_to_be64(pma.n_unicast_xmit
+ - ibp->z_unicast_xmit);
+ p->port_unicast_rcv_packets = cpu_to_be64(pma.n_unicast_rcv
+ - ibp->z_unicast_rcv);
+ p->port_multicast_xmit_packets = cpu_to_be64(pma.n_multicast_xmit
+ - ibp->z_multicast_xmit);
+ p->port_multicast_rcv_packets = cpu_to_be64(pma.n_multicast_rcv
+ - ibp->z_multicast_rcv);
bail:
return reply((struct ib_smp *) pmp);
@@ -1795,6 +1820,7 @@ static int pma_set_portcounters_ext(struct ib_pma_mad *pmp,
struct qib_ibport *ibp = to_iport(ibdev, port);
struct qib_pportdata *ppd = ppd_from_ibp(ibp);
u64 swords, rwords, spkts, rpkts, xwait;
+ struct qib_pma_counters pma;
qib_snapshot_counters(ppd, &swords, &rwords, &spkts, &rpkts, &xwait);
@@ -1810,17 +1836,19 @@ static int pma_set_portcounters_ext(struct ib_pma_mad *pmp,
if (p->counter_select & IB_PMA_SELX_PORT_RCV_PACKETS)
ibp->z_port_rcv_packets = rpkts;
+ qib_snapshot_pmacounters(ibp, &pma);
+
if (p->counter_select & IB_PMA_SELX_PORT_UNI_XMIT_PACKETS)
- ibp->n_unicast_xmit = 0;
+ ibp->z_unicast_xmit = pma.n_unicast_xmit;
if (p->counter_select & IB_PMA_SELX_PORT_UNI_RCV_PACKETS)
- ibp->n_unicast_rcv = 0;
+ ibp->z_unicast_rcv = pma.n_unicast_rcv;
if (p->counter_select & IB_PMA_SELX_PORT_MULTI_XMIT_PACKETS)
- ibp->n_multicast_xmit = 0;
+ ibp->z_multicast_xmit = pma.n_multicast_xmit;
if (p->counter_select & IB_PMA_SELX_PORT_MULTI_RCV_PACKETS)
- ibp->n_multicast_rcv = 0;
+ ibp->z_multicast_rcv = pma.n_multicast_rcv;
return pma_get_portcounters_ext(pmp, ibdev, port);
}
diff --git a/drivers/infiniband/hw/qib/qib_mad.h b/drivers/infiniband/hw/qib/qib_mad.h
index 28874f8606f..941d4d50d8e 100644
--- a/drivers/infiniband/hw/qib/qib_mad.h
+++ b/drivers/infiniband/hw/qib/qib_mad.h
@@ -54,7 +54,7 @@ struct ib_node_info {
__be32 revision;
u8 local_port_num;
u8 vendor_id[3];
-} __attribute__ ((packed));
+} __packed;
struct ib_mad_notice_attr {
u8 generic_type;
@@ -73,7 +73,7 @@ struct ib_mad_notice_attr {
__be16 reserved;
__be16 lid; /* where violation happened */
u8 port_num; /* where violation happened */
- } __attribute__ ((packed)) ntc_129_131;
+ } __packed ntc_129_131;
struct {
__be16 reserved;
@@ -83,14 +83,14 @@ struct ib_mad_notice_attr {
__be32 new_cap_mask; /* new capability mask */
u8 reserved3;
u8 change_flags; /* low 3 bits only */
- } __attribute__ ((packed)) ntc_144;
+ } __packed ntc_144;
struct {
__be16 reserved;
__be16 lid; /* lid where sys guid changed */
__be16 reserved2;
__be64 new_sys_guid;
- } __attribute__ ((packed)) ntc_145;
+ } __packed ntc_145;
struct {
__be16 reserved;
@@ -104,7 +104,7 @@ struct ib_mad_notice_attr {
u8 reserved3;
u8 dr_trunc_hop;
u8 dr_rtn_path[30];
- } __attribute__ ((packed)) ntc_256;
+ } __packed ntc_256;
struct {
__be16 reserved;
@@ -115,7 +115,7 @@ struct ib_mad_notice_attr {
__be32 qp2; /* high 8 bits reserved */
union ib_gid gid1;
union ib_gid gid2;
- } __attribute__ ((packed)) ntc_257_258;
+ } __packed ntc_257_258;
} details;
};
@@ -209,7 +209,7 @@ struct ib_pma_portcounters_cong {
__be64 port_rcv_packets;
__be64 port_xmit_wait;
__be64 port_adr_events;
-} __attribute__ ((packed));
+} __packed;
#define IB_PMA_CONG_HW_CONTROL_TIMER 0x00
#define IB_PMA_CONG_HW_CONTROL_SAMPLE 0x01
diff --git a/drivers/infiniband/hw/qib/qib_mr.c b/drivers/infiniband/hw/qib/qib_mr.c
index e6687ded821..9bbb55347cc 100644
--- a/drivers/infiniband/hw/qib/qib_mr.c
+++ b/drivers/infiniband/hw/qib/qib_mr.c
@@ -232,8 +232,8 @@ struct ib_mr *qib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
{
struct qib_mr *mr;
struct ib_umem *umem;
- struct ib_umem_chunk *chunk;
- int n, m, i;
+ struct scatterlist *sg;
+ int n, m, entry;
struct ib_mr *ret;
if (length == 0) {
@@ -246,9 +246,7 @@ struct ib_mr *qib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
if (IS_ERR(umem))
return (void *) umem;
- n = 0;
- list_for_each_entry(chunk, &umem->chunk_list, list)
- n += chunk->nents;
+ n = umem->nmap;
mr = alloc_mr(n, pd);
if (IS_ERR(mr)) {
@@ -268,11 +266,10 @@ struct ib_mr *qib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
mr->mr.page_shift = ilog2(umem->page_size);
m = 0;
n = 0;
- list_for_each_entry(chunk, &umem->chunk_list, list) {
- for (i = 0; i < chunk->nents; i++) {
+ for_each_sg(umem->sg_head.sgl, sg, umem->nmap, entry) {
void *vaddr;
- vaddr = page_address(sg_page(&chunk->page_list[i]));
+ vaddr = page_address(sg_page(sg));
if (!vaddr) {
ret = ERR_PTR(-EINVAL);
goto bail;
@@ -284,7 +281,6 @@ struct ib_mr *qib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
m++;
n = 0;
}
- }
}
ret = &mr->ibmr;
diff --git a/drivers/infiniband/hw/qib/qib_pcie.c b/drivers/infiniband/hw/qib/qib_pcie.c
index 3f14009fb66..61a0046efb7 100644
--- a/drivers/infiniband/hw/qib/qib_pcie.c
+++ b/drivers/infiniband/hw/qib/qib_pcie.c
@@ -51,8 +51,8 @@
* file calls, even though this violates some
* expectations of harmlessness.
*/
-static int qib_tune_pcie_caps(struct qib_devdata *);
-static int qib_tune_pcie_coalesce(struct qib_devdata *);
+static void qib_tune_pcie_caps(struct qib_devdata *);
+static void qib_tune_pcie_coalesce(struct qib_devdata *);
/*
* Do all the common PCIe setup and initialization.
@@ -197,46 +197,47 @@ static void qib_msix_setup(struct qib_devdata *dd, int pos, u32 *msixcnt,
struct qib_msix_entry *qib_msix_entry)
{
int ret;
- u32 tabsize = 0;
- u16 msix_flags;
+ int nvec = *msixcnt;
struct msix_entry *msix_entry;
int i;
+ ret = pci_msix_vec_count(dd->pcidev);
+ if (ret < 0)
+ goto do_intx;
+
+ nvec = min(nvec, ret);
+
/* We can't pass qib_msix_entry array to qib_msix_setup
* so use a dummy msix_entry array and copy the allocated
* irq back to the qib_msix_entry array. */
- msix_entry = kmalloc(*msixcnt * sizeof(*msix_entry), GFP_KERNEL);
- if (!msix_entry) {
- ret = -ENOMEM;
+ msix_entry = kmalloc(nvec * sizeof(*msix_entry), GFP_KERNEL);
+ if (!msix_entry)
goto do_intx;
- }
- for (i = 0; i < *msixcnt; i++)
+
+ for (i = 0; i < nvec; i++)
msix_entry[i] = qib_msix_entry[i].msix;
- pci_read_config_word(dd->pcidev, pos + PCI_MSIX_FLAGS, &msix_flags);
- tabsize = 1 + (msix_flags & PCI_MSIX_FLAGS_QSIZE);
- if (tabsize > *msixcnt)
- tabsize = *msixcnt;
- ret = pci_enable_msix(dd->pcidev, msix_entry, tabsize);
- if (ret > 0) {
- tabsize = ret;
- ret = pci_enable_msix(dd->pcidev, msix_entry, tabsize);
- }
-do_intx:
- if (ret) {
- qib_dev_err(dd,
- "pci_enable_msix %d vectors failed: %d, falling back to INTx\n",
- tabsize, ret);
- tabsize = 0;
- }
- for (i = 0; i < tabsize; i++)
+ ret = pci_enable_msix_range(dd->pcidev, msix_entry, 1, nvec);
+ if (ret < 0)
+ goto free_msix_entry;
+ else
+ nvec = ret;
+
+ for (i = 0; i < nvec; i++)
qib_msix_entry[i].msix = msix_entry[i];
+
kfree(msix_entry);
- *msixcnt = tabsize;
+ *msixcnt = nvec;
+ return;
- if (ret)
- qib_enable_intx(dd->pcidev);
+free_msix_entry:
+ kfree(msix_entry);
+do_intx:
+ qib_dev_err(dd, "pci_enable_msix_range %d vectors failed: %d, "
+ "falling back to INTx\n", nvec, ret);
+ *msixcnt = 0;
+ qib_enable_intx(dd->pcidev);
}
/**
@@ -476,30 +477,6 @@ void qib_pcie_reenable(struct qib_devdata *dd, u16 cmd, u8 iline, u8 cline)
"pci_enable_device failed after reset: %d\n", r);
}
-/* code to adjust PCIe capabilities. */
-
-static int fld2val(int wd, int mask)
-{
- int lsbmask;
-
- if (!mask)
- return 0;
- wd &= mask;
- lsbmask = mask ^ (mask & (mask - 1));
- wd /= lsbmask;
- return wd;
-}
-
-static int val2fld(int wd, int mask)
-{
- int lsbmask;
-
- if (!mask)
- return 0;
- lsbmask = mask ^ (mask & (mask - 1));
- wd *= lsbmask;
- return wd;
-}
static int qib_pcie_coalesce;
module_param_named(pcie_coalesce, qib_pcie_coalesce, int, S_IRUGO);
@@ -511,7 +488,7 @@ MODULE_PARM_DESC(pcie_coalesce, "tune PCIe colescing on some Intel chipsets");
* of these chipsets, with some BIOS settings, and enabling it on those
* systems may result in the system crashing, and/or data corruption.
*/
-static int qib_tune_pcie_coalesce(struct qib_devdata *dd)
+static void qib_tune_pcie_coalesce(struct qib_devdata *dd)
{
int r;
struct pci_dev *parent;
@@ -519,18 +496,18 @@ static int qib_tune_pcie_coalesce(struct qib_devdata *dd)
u32 mask, bits, val;
if (!qib_pcie_coalesce)
- return 0;
+ return;
/* Find out supported and configured values for parent (root) */
parent = dd->pcidev->bus->self;
if (parent->bus->parent) {
qib_devinfo(dd->pcidev, "Parent not root\n");
- return 1;
+ return;
}
if (!pci_is_pcie(parent))
- return 1;
+ return;
if (parent->vendor != 0x8086)
- return 1;
+ return;
/*
* - bit 12: Max_rdcmp_Imt_EN: need to set to 1
@@ -563,13 +540,12 @@ static int qib_tune_pcie_coalesce(struct qib_devdata *dd)
mask = (3U << 24) | (7U << 10);
} else {
/* not one of the chipsets that we know about */
- return 1;
+ return;
}
pci_read_config_dword(parent, 0x48, &val);
val &= ~mask;
val |= bits;
r = pci_write_config_dword(parent, 0x48, val);
- return 0;
}
/*
@@ -580,55 +556,44 @@ static int qib_pcie_caps;
module_param_named(pcie_caps, qib_pcie_caps, int, S_IRUGO);
MODULE_PARM_DESC(pcie_caps, "Max PCIe tuning: Payload (0..3), ReadReq (4..7)");
-static int qib_tune_pcie_caps(struct qib_devdata *dd)
+static void qib_tune_pcie_caps(struct qib_devdata *dd)
{
- int ret = 1; /* Assume the worst */
struct pci_dev *parent;
- u16 pcaps, pctl, ecaps, ectl;
- int rc_sup, ep_sup;
- int rc_cur, ep_cur;
+ u16 rc_mpss, rc_mps, ep_mpss, ep_mps;
+ u16 rc_mrrs, ep_mrrs, max_mrrs;
/* Find out supported and configured values for parent (root) */
parent = dd->pcidev->bus->self;
- if (parent->bus->parent) {
+ if (!pci_is_root_bus(parent->bus)) {
qib_devinfo(dd->pcidev, "Parent not root\n");
- goto bail;
+ return;
}
if (!pci_is_pcie(parent) || !pci_is_pcie(dd->pcidev))
- goto bail;
- pcie_capability_read_word(parent, PCI_EXP_DEVCAP, &pcaps);
- pcie_capability_read_word(parent, PCI_EXP_DEVCTL, &pctl);
+ return;
+
+ rc_mpss = parent->pcie_mpss;
+ rc_mps = ffs(pcie_get_mps(parent)) - 8;
/* Find out supported and configured values for endpoint (us) */
- pcie_capability_read_word(dd->pcidev, PCI_EXP_DEVCAP, &ecaps);
- pcie_capability_read_word(dd->pcidev, PCI_EXP_DEVCTL, &ectl);
+ ep_mpss = dd->pcidev->pcie_mpss;
+ ep_mps = ffs(pcie_get_mps(dd->pcidev)) - 8;
- ret = 0;
/* Find max payload supported by root, endpoint */
- rc_sup = fld2val(pcaps, PCI_EXP_DEVCAP_PAYLOAD);
- ep_sup = fld2val(ecaps, PCI_EXP_DEVCAP_PAYLOAD);
- if (rc_sup > ep_sup)
- rc_sup = ep_sup;
-
- rc_cur = fld2val(pctl, PCI_EXP_DEVCTL_PAYLOAD);
- ep_cur = fld2val(ectl, PCI_EXP_DEVCTL_PAYLOAD);
+ if (rc_mpss > ep_mpss)
+ rc_mpss = ep_mpss;
/* If Supported greater than limit in module param, limit it */
- if (rc_sup > (qib_pcie_caps & 7))
- rc_sup = qib_pcie_caps & 7;
+ if (rc_mpss > (qib_pcie_caps & 7))
+ rc_mpss = qib_pcie_caps & 7;
/* If less than (allowed, supported), bump root payload */
- if (rc_sup > rc_cur) {
- rc_cur = rc_sup;
- pctl = (pctl & ~PCI_EXP_DEVCTL_PAYLOAD) |
- val2fld(rc_cur, PCI_EXP_DEVCTL_PAYLOAD);
- pcie_capability_write_word(parent, PCI_EXP_DEVCTL, pctl);
+ if (rc_mpss > rc_mps) {
+ rc_mps = rc_mpss;
+ pcie_set_mps(parent, 128 << rc_mps);
}
/* If less than (allowed, supported), bump endpoint payload */
- if (rc_sup > ep_cur) {
- ep_cur = rc_sup;
- ectl = (ectl & ~PCI_EXP_DEVCTL_PAYLOAD) |
- val2fld(ep_cur, PCI_EXP_DEVCTL_PAYLOAD);
- pcie_capability_write_word(dd->pcidev, PCI_EXP_DEVCTL, ectl);
+ if (rc_mpss > ep_mps) {
+ ep_mps = rc_mpss;
+ pcie_set_mps(dd->pcidev, 128 << ep_mps);
}
/*
@@ -636,26 +601,22 @@ static int qib_tune_pcie_caps(struct qib_devdata *dd)
* No field for max supported, but PCIe spec limits it to 4096,
* which is code '5' (log2(4096) - 7)
*/
- rc_sup = 5;
- if (rc_sup > ((qib_pcie_caps >> 4) & 7))
- rc_sup = (qib_pcie_caps >> 4) & 7;
- rc_cur = fld2val(pctl, PCI_EXP_DEVCTL_READRQ);
- ep_cur = fld2val(ectl, PCI_EXP_DEVCTL_READRQ);
-
- if (rc_sup > rc_cur) {
- rc_cur = rc_sup;
- pctl = (pctl & ~PCI_EXP_DEVCTL_READRQ) |
- val2fld(rc_cur, PCI_EXP_DEVCTL_READRQ);
- pcie_capability_write_word(parent, PCI_EXP_DEVCTL, pctl);
+ max_mrrs = 5;
+ if (max_mrrs > ((qib_pcie_caps >> 4) & 7))
+ max_mrrs = (qib_pcie_caps >> 4) & 7;
+
+ max_mrrs = 128 << max_mrrs;
+ rc_mrrs = pcie_get_readrq(parent);
+ ep_mrrs = pcie_get_readrq(dd->pcidev);
+
+ if (max_mrrs > rc_mrrs) {
+ rc_mrrs = max_mrrs;
+ pcie_set_readrq(parent, rc_mrrs);
}
- if (rc_sup > ep_cur) {
- ep_cur = rc_sup;
- ectl = (ectl & ~PCI_EXP_DEVCTL_READRQ) |
- val2fld(ep_cur, PCI_EXP_DEVCTL_READRQ);
- pcie_capability_write_word(dd->pcidev, PCI_EXP_DEVCTL, ectl);
+ if (max_mrrs > ep_mrrs) {
+ ep_mrrs = max_mrrs;
+ pcie_set_readrq(dd->pcidev, ep_mrrs);
}
-bail:
- return ret;
}
/* End of PCIe capability tuning */
diff --git a/drivers/infiniband/hw/qib/qib_qp.c b/drivers/infiniband/hw/qib/qib_qp.c
index 3cca55b51e5..7fcc150d603 100644
--- a/drivers/infiniband/hw/qib/qib_qp.c
+++ b/drivers/infiniband/hw/qib/qib_qp.c
@@ -585,7 +585,7 @@ int qib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
new_state = attr_mask & IB_QP_STATE ? attr->qp_state : cur_state;
if (!ib_modify_qp_is_ok(cur_state, new_state, ibqp->qp_type,
- attr_mask))
+ attr_mask, IB_LINK_LAYER_UNSPECIFIED))
goto inval;
if (attr_mask & IB_QP_AV) {
@@ -985,7 +985,8 @@ struct ib_qp *qib_create_qp(struct ib_pd *ibpd,
struct ib_qp *ret;
if (init_attr->cap.max_send_sge > ib_qib_max_sges ||
- init_attr->cap.max_send_wr > ib_qib_max_qp_wrs) {
+ init_attr->cap.max_send_wr > ib_qib_max_qp_wrs ||
+ init_attr->create_flags) {
ret = ERR_PTR(-EINVAL);
goto bail;
}
diff --git a/drivers/infiniband/hw/qib/qib_rc.c b/drivers/infiniband/hw/qib/qib_rc.c
index 3ab341320ea..2f2501890c4 100644
--- a/drivers/infiniband/hw/qib/qib_rc.c
+++ b/drivers/infiniband/hw/qib/qib_rc.c
@@ -752,7 +752,7 @@ void qib_send_rc_ack(struct qib_qp *qp)
qib_flush_wc();
qib_sendbuf_done(dd, pbufn);
- ibp->n_unicast_xmit++;
+ this_cpu_inc(ibp->pmastats->n_unicast_xmit);
goto done;
queue_ack:
diff --git a/drivers/infiniband/hw/qib/qib_ruc.c b/drivers/infiniband/hw/qib/qib_ruc.c
index 357b6cfcd46..4c07a8b34ff 100644
--- a/drivers/infiniband/hw/qib/qib_ruc.c
+++ b/drivers/infiniband/hw/qib/qib_ruc.c
@@ -703,6 +703,7 @@ void qib_make_ruc_header(struct qib_qp *qp, struct qib_other_headers *ohdr,
ohdr->bth[0] = cpu_to_be32(bth0);
ohdr->bth[1] = cpu_to_be32(qp->remote_qpn);
ohdr->bth[2] = cpu_to_be32(bth2);
+ this_cpu_inc(ibp->pmastats->n_unicast_xmit);
}
/**
diff --git a/drivers/infiniband/hw/qib/qib_ud.c b/drivers/infiniband/hw/qib/qib_ud.c
index d6c7fe7f88d..aaf7039f8ed 100644
--- a/drivers/infiniband/hw/qib/qib_ud.c
+++ b/drivers/infiniband/hw/qib/qib_ud.c
@@ -57,13 +57,20 @@ static void qib_ud_loopback(struct qib_qp *sqp, struct qib_swqe *swqe)
struct qib_sge *sge;
struct ib_wc wc;
u32 length;
+ enum ib_qp_type sqptype, dqptype;
qp = qib_lookup_qpn(ibp, swqe->wr.wr.ud.remote_qpn);
if (!qp) {
ibp->n_pkt_drops++;
return;
}
- if (qp->ibqp.qp_type != sqp->ibqp.qp_type ||
+
+ sqptype = sqp->ibqp.qp_type == IB_QPT_GSI ?
+ IB_QPT_UD : sqp->ibqp.qp_type;
+ dqptype = qp->ibqp.qp_type == IB_QPT_GSI ?
+ IB_QPT_UD : qp->ibqp.qp_type;
+
+ if (dqptype != sqptype ||
!(ib_qib_state_ops[qp->state] & QIB_PROCESS_RECV_OK)) {
ibp->n_pkt_drops++;
goto drop;
@@ -273,11 +280,11 @@ int qib_make_ud_req(struct qib_qp *qp)
ah_attr = &to_iah(wqe->wr.wr.ud.ah)->attr;
if (ah_attr->dlid >= QIB_MULTICAST_LID_BASE) {
if (ah_attr->dlid != QIB_PERMISSIVE_LID)
- ibp->n_multicast_xmit++;
+ this_cpu_inc(ibp->pmastats->n_multicast_xmit);
else
- ibp->n_unicast_xmit++;
+ this_cpu_inc(ibp->pmastats->n_unicast_xmit);
} else {
- ibp->n_unicast_xmit++;
+ this_cpu_inc(ibp->pmastats->n_unicast_xmit);
lid = ah_attr->dlid & ~((1 << ppd->lmc) - 1);
if (unlikely(lid == ppd->lid)) {
/*
diff --git a/drivers/infiniband/hw/qib/qib_user_sdma.c b/drivers/infiniband/hw/qib/qib_user_sdma.c
index d0a0ea0c14d..d2806cae234 100644
--- a/drivers/infiniband/hw/qib/qib_user_sdma.c
+++ b/drivers/infiniband/hw/qib/qib_user_sdma.c
@@ -52,6 +52,17 @@
/* attempt to drain the queue for 5secs */
#define QIB_USER_SDMA_DRAIN_TIMEOUT 500
+/*
+ * track how many times a process open this driver.
+ */
+static struct rb_root qib_user_sdma_rb_root = RB_ROOT;
+
+struct qib_user_sdma_rb_node {
+ struct rb_node node;
+ int refcount;
+ pid_t pid;
+};
+
struct qib_user_sdma_pkt {
struct list_head list; /* list element */
@@ -120,15 +131,60 @@ struct qib_user_sdma_queue {
/* dma page table */
struct rb_root dma_pages_root;
+ struct qib_user_sdma_rb_node *sdma_rb_node;
+
/* protect everything above... */
struct mutex lock;
};
+static struct qib_user_sdma_rb_node *
+qib_user_sdma_rb_search(struct rb_root *root, pid_t pid)
+{
+ struct qib_user_sdma_rb_node *sdma_rb_node;
+ struct rb_node *node = root->rb_node;
+
+ while (node) {
+ sdma_rb_node = container_of(node,
+ struct qib_user_sdma_rb_node, node);
+ if (pid < sdma_rb_node->pid)
+ node = node->rb_left;
+ else if (pid > sdma_rb_node->pid)
+ node = node->rb_right;
+ else
+ return sdma_rb_node;
+ }
+ return NULL;
+}
+
+static int
+qib_user_sdma_rb_insert(struct rb_root *root, struct qib_user_sdma_rb_node *new)
+{
+ struct rb_node **node = &(root->rb_node);
+ struct rb_node *parent = NULL;
+ struct qib_user_sdma_rb_node *got;
+
+ while (*node) {
+ got = container_of(*node, struct qib_user_sdma_rb_node, node);
+ parent = *node;
+ if (new->pid < got->pid)
+ node = &((*node)->rb_left);
+ else if (new->pid > got->pid)
+ node = &((*node)->rb_right);
+ else
+ return 0;
+ }
+
+ rb_link_node(&new->node, parent, node);
+ rb_insert_color(&new->node, root);
+ return 1;
+}
+
struct qib_user_sdma_queue *
qib_user_sdma_queue_create(struct device *dev, int unit, int ctxt, int sctxt)
{
struct qib_user_sdma_queue *pq =
kmalloc(sizeof(struct qib_user_sdma_queue), GFP_KERNEL);
+ struct qib_user_sdma_rb_node *sdma_rb_node;
if (!pq)
goto done;
@@ -138,6 +194,7 @@ qib_user_sdma_queue_create(struct device *dev, int unit, int ctxt, int sctxt)
pq->num_pending = 0;
pq->num_sending = 0;
pq->added = 0;
+ pq->sdma_rb_node = NULL;
INIT_LIST_HEAD(&pq->sent);
spin_lock_init(&pq->sent_lock);
@@ -163,8 +220,30 @@ qib_user_sdma_queue_create(struct device *dev, int unit, int ctxt, int sctxt)
pq->dma_pages_root = RB_ROOT;
+ sdma_rb_node = qib_user_sdma_rb_search(&qib_user_sdma_rb_root,
+ current->pid);
+ if (sdma_rb_node) {
+ sdma_rb_node->refcount++;
+ } else {
+ int ret;
+ sdma_rb_node = kmalloc(sizeof(
+ struct qib_user_sdma_rb_node), GFP_KERNEL);
+ if (!sdma_rb_node)
+ goto err_rb;
+
+ sdma_rb_node->refcount = 1;
+ sdma_rb_node->pid = current->pid;
+
+ ret = qib_user_sdma_rb_insert(&qib_user_sdma_rb_root,
+ sdma_rb_node);
+ BUG_ON(ret == 0);
+ }
+ pq->sdma_rb_node = sdma_rb_node;
+
goto done;
+err_rb:
+ dma_pool_destroy(pq->header_cache);
err_slab:
kmem_cache_destroy(pq->pkt_slab);
err_kfree:
@@ -594,8 +673,7 @@ static int qib_user_sdma_pin_pages(const struct qib_devdata *dd,
else
j = npages;
- ret = get_user_pages(current, current->mm, addr,
- j, 0, 1, pages, NULL);
+ ret = get_user_pages_fast(addr, j, 0, pages);
if (ret != j) {
i = 0;
j = ret;
@@ -1021,8 +1099,13 @@ void qib_user_sdma_queue_destroy(struct qib_user_sdma_queue *pq)
if (!pq)
return;
- kmem_cache_destroy(pq->pkt_slab);
+ pq->sdma_rb_node->refcount--;
+ if (pq->sdma_rb_node->refcount == 0) {
+ rb_erase(&pq->sdma_rb_node->node, &qib_user_sdma_rb_root);
+ kfree(pq->sdma_rb_node);
+ }
dma_pool_destroy(pq->header_cache);
+ kmem_cache_destroy(pq->pkt_slab);
kfree(pq);
}
@@ -1242,26 +1325,52 @@ static int qib_user_sdma_push_pkts(struct qib_pportdata *ppd,
struct qib_user_sdma_queue *pq,
struct list_head *pktlist, int count)
{
- int ret = 0;
unsigned long flags;
if (unlikely(!(ppd->lflags & QIBL_LINKACTIVE)))
return -ECOMM;
- spin_lock_irqsave(&ppd->sdma_lock, flags);
-
- if (unlikely(!__qib_sdma_running(ppd))) {
- ret = -ECOMM;
- goto unlock;
+ /* non-blocking mode */
+ if (pq->sdma_rb_node->refcount > 1) {
+ spin_lock_irqsave(&ppd->sdma_lock, flags);
+ if (unlikely(!__qib_sdma_running(ppd))) {
+ spin_unlock_irqrestore(&ppd->sdma_lock, flags);
+ return -ECOMM;
+ }
+ pq->num_pending += count;
+ list_splice_tail_init(pktlist, &ppd->sdma_userpending);
+ qib_user_sdma_send_desc(ppd, &ppd->sdma_userpending);
+ spin_unlock_irqrestore(&ppd->sdma_lock, flags);
+ return 0;
}
+ /* In this case, descriptors from this process are not
+ * linked to ppd pending queue, interrupt handler
+ * won't update this process, it is OK to directly
+ * modify without sdma lock.
+ */
+
+
pq->num_pending += count;
- list_splice_tail_init(pktlist, &ppd->sdma_userpending);
- qib_user_sdma_send_desc(ppd, &ppd->sdma_userpending);
+ /*
+ * Blocking mode for single rail process, we must
+ * release/regain sdma_lock to give other process
+ * chance to make progress. This is important for
+ * performance.
+ */
+ do {
+ spin_lock_irqsave(&ppd->sdma_lock, flags);
+ if (unlikely(!__qib_sdma_running(ppd))) {
+ spin_unlock_irqrestore(&ppd->sdma_lock, flags);
+ return -ECOMM;
+ }
+ qib_user_sdma_send_desc(ppd, pktlist);
+ if (!list_empty(pktlist))
+ qib_sdma_make_progress(ppd);
+ spin_unlock_irqrestore(&ppd->sdma_lock, flags);
+ } while (!list_empty(pktlist));
-unlock:
- spin_unlock_irqrestore(&ppd->sdma_lock, flags);
- return ret;
+ return 0;
}
int qib_user_sdma_writev(struct qib_ctxtdata *rcd,
@@ -1291,14 +1400,11 @@ int qib_user_sdma_writev(struct qib_ctxtdata *rcd,
qib_user_sdma_queue_clean(ppd, pq);
while (dim) {
- int mxp = 8;
+ int mxp = 1;
int ndesc = 0;
- down_write(&current->mm->mmap_sem);
ret = qib_user_sdma_queue_pkts(dd, ppd, pq,
iov, dim, &list, &mxp, &ndesc);
- up_write(&current->mm->mmap_sem);
-
if (ret < 0)
goto done_unlock;
else {
diff --git a/drivers/infiniband/hw/qib/qib_verbs.c b/drivers/infiniband/hw/qib/qib_verbs.c
index 092b0bb1bb7..9bcfbd84298 100644
--- a/drivers/infiniband/hw/qib/qib_verbs.c
+++ b/drivers/infiniband/hw/qib/qib_verbs.c
@@ -662,7 +662,7 @@ void qib_ib_rcv(struct qib_ctxtdata *rcd, void *rhdr, void *data, u32 tlen)
mcast = qib_mcast_find(ibp, &hdr->u.l.grh.dgid);
if (mcast == NULL)
goto drop;
- ibp->n_multicast_rcv++;
+ this_cpu_inc(ibp->pmastats->n_multicast_rcv);
list_for_each_entry_rcu(p, &mcast->qp_list, list)
qib_qp_rcv(rcd, hdr, 1, data, tlen, p->qp);
/*
@@ -678,8 +678,8 @@ void qib_ib_rcv(struct qib_ctxtdata *rcd, void *rhdr, void *data, u32 tlen)
&rcd->lookaside_qp->refcount))
wake_up(
&rcd->lookaside_qp->wait);
- rcd->lookaside_qp = NULL;
- }
+ rcd->lookaside_qp = NULL;
+ }
}
if (!rcd->lookaside_qp) {
qp = qib_lookup_qpn(ibp, qp_num);
@@ -689,7 +689,7 @@ void qib_ib_rcv(struct qib_ctxtdata *rcd, void *rhdr, void *data, u32 tlen)
rcd->lookaside_qpn = qp_num;
} else
qp = rcd->lookaside_qp;
- ibp->n_unicast_rcv++;
+ this_cpu_inc(ibp->pmastats->n_unicast_rcv);
qib_qp_rcv(rcd, hdr, lnh == QIB_LRH_GRH, data, tlen, qp);
}
return;
diff --git a/drivers/infiniband/hw/qib/qib_verbs.h b/drivers/infiniband/hw/qib/qib_verbs.h
index 012e2c7575a..bfc8948fdd3 100644
--- a/drivers/infiniband/hw/qib/qib_verbs.h
+++ b/drivers/infiniband/hw/qib/qib_verbs.h
@@ -150,14 +150,14 @@ struct ib_reth {
__be64 vaddr;
__be32 rkey;
__be32 length;
-} __attribute__ ((packed));
+} __packed;
struct ib_atomic_eth {
__be32 vaddr[2]; /* unaligned so access as 2 32-bit words */
__be32 rkey;
__be64 swap_data;
__be64 compare_data;
-} __attribute__ ((packed));
+} __packed;
struct qib_other_headers {
__be32 bth[3];
@@ -178,7 +178,7 @@ struct qib_other_headers {
__be32 aeth;
struct ib_atomic_eth atomic_eth;
} u;
-} __attribute__ ((packed));
+} __packed;
/*
* Note that UD packets with a GRH header are 8+40+12+8 = 68 bytes
@@ -195,12 +195,12 @@ struct qib_ib_header {
} l;
struct qib_other_headers oth;
} u;
-} __attribute__ ((packed));
+} __packed;
struct qib_pio_header {
__le32 pbc[2];
struct qib_ib_header hdr;
-} __attribute__ ((packed));
+} __packed;
/*
* There is one struct qib_mcast for each multicast GID.
@@ -664,6 +664,13 @@ struct qib_opcode_stats_perctx {
struct qib_opcode_stats stats[128];
};
+struct qib_pma_counters {
+ u64 n_unicast_xmit; /* total unicast packets sent */
+ u64 n_unicast_rcv; /* total unicast packets received */
+ u64 n_multicast_xmit; /* total multicast packets sent */
+ u64 n_multicast_rcv; /* total multicast packets received */
+};
+
struct qib_ibport {
struct qib_qp __rcu *qp0;
struct qib_qp __rcu *qp1;
@@ -680,10 +687,11 @@ struct qib_ibport {
__be64 mkey;
__be64 guids[QIB_GUIDS_PER_PORT - 1]; /* writable GUIDs */
u64 tid; /* TID for traps */
- u64 n_unicast_xmit; /* total unicast packets sent */
- u64 n_unicast_rcv; /* total unicast packets received */
- u64 n_multicast_xmit; /* total multicast packets sent */
- u64 n_multicast_rcv; /* total multicast packets received */
+ struct qib_pma_counters __percpu *pmastats;
+ u64 z_unicast_xmit; /* starting count for PMA */
+ u64 z_unicast_rcv; /* starting count for PMA */
+ u64 z_multicast_xmit; /* starting count for PMA */
+ u64 z_multicast_rcv; /* starting count for PMA */
u64 z_symbol_error_counter; /* starting count for PMA */
u64 z_link_error_recovery_counter; /* starting count for PMA */
u64 z_link_downed_counter; /* starting count for PMA */
generated by cgit v1.2.3-70-g09d2 (git 2.46.0) at 2025-11-04 07:57:34 +0000