diff options
Diffstat (limited to 'drivers/infiniband/hw/qib')
29 files changed, 475 insertions, 611 deletions
diff --git a/drivers/infiniband/hw/qib/qib.h b/drivers/infiniband/hw/qib/qib.h index 769a1d9..c0b72a6 100644 --- a/drivers/infiniband/hw/qib/qib.h +++ b/drivers/infiniband/hw/qib/qib.h @@ -171,7 +171,9 @@ struct qib_ctxtdata { /* how many alloc_pages() chunks in rcvegrbuf_pages */ u32 rcvegrbuf_chunks; /* how many egrbufs per chunk */ - u32 rcvegrbufs_perchunk; + u16 rcvegrbufs_perchunk; + /* ilog2 of above */ + u16 rcvegrbufs_perchunk_shift; /* order for rcvegrbuf_pages */ size_t rcvegrbuf_size; /* rcvhdrq size (for freeing) */ @@ -221,6 +223,9 @@ struct qib_ctxtdata { /* ctxt rcvhdrq head offset */ u32 head; u32 pkt_count; + /* lookaside fields */ + struct qib_qp *lookaside_qp; + u32 lookaside_qpn; /* QPs waiting for context processing */ struct list_head qp_wait_list; }; @@ -807,6 +812,10 @@ struct qib_devdata { * supports, less gives more pio bufs/ctxt, etc. */ u32 cfgctxts; + /* + * number of ctxts available for PSM open + */ + u32 freectxts; /* * hint that we should update pioavailshadow before @@ -936,7 +945,9 @@ struct qib_devdata { /* chip address space used by 4k pio buffers */ u32 align4k; /* size of each rcvegrbuffer */ - u32 rcvegrbufsize; + u16 rcvegrbufsize; + /* log2 of above */ + u16 rcvegrbufsize_shift; /* localbus width (1, 2,4,8,16,32) from config space */ u32 lbus_width; /* localbus speed in MHz */ @@ -994,12 +1005,6 @@ struct qib_devdata { /* control high-level access to EEPROM */ struct mutex eep_lock; uint64_t traffic_wds; - /* active time is kept in seconds, but logged in hours */ - atomic_t active_time; - /* Below are nominal shadow of EEPROM, new since last EEPROM update */ - uint8_t eep_st_errs[QIB_EEP_LOG_CNT]; - uint8_t eep_st_new_errs[QIB_EEP_LOG_CNT]; - uint16_t eep_hrs; /* * masks for which bits of errs, hwerrs that cause * each of the counters to increment. @@ -1012,6 +1017,8 @@ struct qib_devdata { u8 psxmitwait_supported; /* cycle length of PS* counters in HW (in picoseconds) */ u16 psxmitwait_check_rate; + /* high volume overflow errors defered to tasklet */ + struct tasklet_struct error_tasklet; }; /* hol_state values */ @@ -1214,8 +1221,7 @@ int qib_twsi_blk_rd(struct qib_devdata *dd, int dev, int addr, void *buffer, int qib_twsi_blk_wr(struct qib_devdata *dd, int dev, int addr, const void *buffer, int len); void qib_get_eeprom_info(struct qib_devdata *); -int qib_update_eeprom_log(struct qib_devdata *dd); -void qib_inc_eeprom_err(struct qib_devdata *dd, u32 eidx, u32 incr); +#define qib_inc_eeprom_err(dd, eidx, incr) void qib_dump_lookup_output_queue(struct qib_devdata *); void qib_force_pio_avail_update(struct qib_devdata *); void qib_clear_symerror_on_linkup(unsigned long opaque); @@ -1415,6 +1421,10 @@ extern struct mutex qib_mutex; qib_get_unit_name((dd)->unit), ##__VA_ARGS__); \ } while (0) +#define qib_dev_warn(dd, fmt, ...) \ + dev_warn(&(dd)->pcidev->dev, "%s: " fmt, \ + qib_get_unit_name((dd)->unit), ##__VA_ARGS__) + #define qib_dev_porterr(dd, port, fmt, ...) \ do { \ dev_err(&(dd)->pcidev->dev, "%s: IB%u:%u " fmt, \ @@ -1433,6 +1443,7 @@ extern struct mutex qib_mutex; struct qib_hwerror_msgs { u64 mask; const char *msg; + size_t sz; }; #define QLOGIC_IB_HWE_MSG(a, b) { .mask = a, .msg = b } diff --git a/drivers/infiniband/hw/qib/qib_diag.c b/drivers/infiniband/hw/qib/qib_diag.c index 204c4dd..9892456 100644 --- a/drivers/infiniband/hw/qib/qib_diag.c +++ b/drivers/infiniband/hw/qib/qib_diag.c @@ -46,6 +46,7 @@ #include <linux/pci.h> #include <linux/poll.h> #include <linux/vmalloc.h> +#include <linux/export.h> #include <linux/fs.h> #include <linux/uaccess.h> diff --git a/drivers/infiniband/hw/qib/qib_driver.c b/drivers/infiniband/hw/qib/qib_driver.c index 23e584f..c90a55f 100644 --- a/drivers/infiniband/hw/qib/qib_driver.c +++ b/drivers/infiniband/hw/qib/qib_driver.c @@ -37,6 +37,7 @@ #include <linux/delay.h> #include <linux/netdevice.h> #include <linux/vmalloc.h> +#include <linux/module.h> #include "qib.h" @@ -279,10 +280,10 @@ bail: */ static inline void *qib_get_egrbuf(const struct qib_ctxtdata *rcd, u32 etail) { - const u32 chunk = etail / rcd->rcvegrbufs_perchunk; - const u32 idx = etail % rcd->rcvegrbufs_perchunk; + const u32 chunk = etail >> rcd->rcvegrbufs_perchunk_shift; + const u32 idx = etail & ((u32)rcd->rcvegrbufs_perchunk - 1); - return rcd->rcvegrbuf[chunk] + idx * rcd->dd->rcvegrbufsize; + return rcd->rcvegrbuf[chunk] + (idx << rcd->dd->rcvegrbufsize_shift); } /* @@ -310,7 +311,6 @@ static u32 qib_rcv_hdrerr(struct qib_ctxtdata *rcd, struct qib_pportdata *ppd, u32 opcode; u32 psn; int diff; - unsigned long flags; /* Sanity check packet */ if (tlen < 24) @@ -365,7 +365,6 @@ static u32 qib_rcv_hdrerr(struct qib_ctxtdata *rcd, struct qib_pportdata *ppd, switch (qp->ibqp.qp_type) { case IB_QPT_RC: - spin_lock_irqsave(&qp->s_lock, flags); ruc_res = qib_ruc_check_hdr( ibp, hdr, @@ -373,11 +372,8 @@ static u32 qib_rcv_hdrerr(struct qib_ctxtdata *rcd, struct qib_pportdata *ppd, qp, be32_to_cpu(ohdr->bth[0])); if (ruc_res) { - spin_unlock_irqrestore(&qp->s_lock, - flags); goto unlock; } - spin_unlock_irqrestore(&qp->s_lock, flags); /* Only deal with RDMA Writes for now */ if (opcode < @@ -547,6 +543,15 @@ move_along: updegr = 0; } } + /* + * Notify qib_destroy_qp() if it is waiting + * for lookaside_qp to finish. + */ + if (rcd->lookaside_qp) { + if (atomic_dec_and_test(&rcd->lookaside_qp->refcount)) + wake_up(&rcd->lookaside_qp->wait); + rcd->lookaside_qp = NULL; + } rcd->head = l; rcd->pkt_count += i; diff --git a/drivers/infiniband/hw/qib/qib_eeprom.c b/drivers/infiniband/hw/qib/qib_eeprom.c index 92d9cfe..6359c2f 100644 --- a/drivers/infiniband/hw/qib/qib_eeprom.c +++ b/drivers/infiniband/hw/qib/qib_eeprom.c @@ -263,189 +263,9 @@ void qib_get_eeprom_info(struct qib_devdata *dd) qib_dev_err(dd, "Board SN %s did not pass functional " "test: %s\n", dd->serial, ifp->if_comment); - memcpy(&dd->eep_st_errs, &ifp->if_errcntp, QIB_EEP_LOG_CNT); - /* - * Power-on (actually "active") hours are kept as little-endian value - * in EEPROM, but as seconds in a (possibly as small as 24-bit) - * atomic_t while running. - */ - atomic_set(&dd->active_time, 0); - dd->eep_hrs = ifp->if_powerhour[0] | (ifp->if_powerhour[1] << 8); - done: vfree(buf); bail:; } -/** - * qib_update_eeprom_log - copy active-time and error counters to eeprom - * @dd: the qlogic_ib device - * - * Although the time is kept as seconds in the qib_devdata struct, it is - * rounded to hours for re-write, as we have only 16 bits in EEPROM. - * First-cut code reads whole (expected) struct qib_flash, modifies, - * re-writes. Future direction: read/write only what we need, assuming - * that the EEPROM had to have been "good enough" for driver init, and - * if not, we aren't making it worse. - * - */ -int qib_update_eeprom_log(struct qib_devdata *dd) -{ - void *buf; - struct qib_flash *ifp; - int len, hi_water; - uint32_t new_time, new_hrs; - u8 csum; - int ret, idx; - unsigned long flags; - - /* first, check if we actually need to do anything. */ - ret = 0; - for (idx = 0; idx < QIB_EEP_LOG_CNT; ++idx) { - if (dd->eep_st_new_errs[idx]) { - ret = 1; - break; - } - } - new_time = atomic_read(&dd->active_time); - - if (ret == 0 && new_time < 3600) - goto bail; - - /* - * The quick-check above determined that there is something worthy - * of logging, so get current contents and do a more detailed idea. - * read full flash, not just currently used part, since it may have - * been written with a newer definition - */ - len = sizeof(struct qib_flash); - buf = vmalloc(len); - ret = 1; - if (!buf) { - qib_dev_err(dd, "Couldn't allocate memory to read %u " - "bytes from eeprom for logging\n", len); - goto bail; - } - - /* Grab semaphore and read current EEPROM. If we get an - * error, let go, but if not, keep it until we finish write. - */ - ret = mutex_lock_interruptible(&dd->eep_lock); - if (ret) { - qib_dev_err(dd, "Unable to acquire EEPROM for logging\n"); - goto free_bail; - } - ret = qib_twsi_blk_rd(dd, dd->twsi_eeprom_dev, 0, buf, len); - if (ret) { - mutex_unlock(&dd->eep_lock); - qib_dev_err(dd, "Unable read EEPROM for logging\n"); - goto free_bail; - } - ifp = (struct qib_flash *)buf; - - csum = flash_csum(ifp, 0); - if (csum != ifp->if_csum) { - mutex_unlock(&dd->eep_lock); - qib_dev_err(dd, "EEPROM cks err (0x%02X, S/B 0x%02X)\n", - csum, ifp->if_csum); - ret = 1; - goto free_bail; - } - hi_water = 0; - spin_lock_irqsave(&dd->eep_st_lock, flags); - for (idx = 0; idx < QIB_EEP_LOG_CNT; ++idx) { - int new_val = dd->eep_st_new_errs[idx]; - if (new_val) { - /* - * If we have seen any errors, add to EEPROM values - * We need to saturate at 0xFF (255) and we also - * would need to adjust the checksum if we were - * trying to minimize EEPROM traffic - * Note that we add to actual current count in EEPROM, - * in case it was altered while we were running. - */ - new_val += ifp->if_errcntp[idx]; - if (new_val > 0xFF) - new_val = 0xFF; - if (ifp->if_errcntp[idx] != new_val) { - ifp->if_errcntp[idx] = new_val; - hi_water = offsetof(struct qib_flash, - if_errcntp) + idx; - } - /* - * update our shadow (used to minimize EEPROM - * traffic), to match what we are about to write. - */ - dd->eep_st_errs[idx] = new_val; - dd->eep_st_new_errs[idx] = 0; - } - } - /* - * Now update active-time. We would like to round to the nearest hour - * but unless atomic_t are sure to be proper signed ints we cannot, - * because we need to account for what we "transfer" to EEPROM and - * if we log an hour at 31 minutes, then we would need to set - * active_time to -29 to accurately count the _next_ hour. - */ - if (new_time >= 3600) { - new_hrs = new_time / 3600; - atomic_sub((new_hrs * 3600), &dd->active_time); - new_hrs += dd->eep_hrs; - if (new_hrs > 0xFFFF) - new_hrs = 0xFFFF; - dd->eep_hrs = new_hrs; - if ((new_hrs & 0xFF) != ifp->if_powerhour[0]) { - ifp->if_powerhour[0] = new_hrs & 0xFF; - hi_water = offsetof(struct qib_flash, if_powerhour); - } - if ((new_hrs >> 8) != ifp->if_powerhour[1]) { - ifp->if_powerhour[1] = new_hrs >> 8; - hi_water = offsetof(struct qib_flash, if_powerhour) + 1; - } - } - /* - * There is a tiny possibility that we could somehow fail to write - * the EEPROM after updating our shadows, but problems from holding - * the spinlock too long are a much bigger issue. - */ - spin_unlock_irqrestore(&dd->eep_st_lock, flags); - if (hi_water) { - /* we made some change to the data, uopdate cksum and write */ - csum = flash_csum(ifp, 1); - ret = eeprom_write_with_enable(dd, 0, buf, hi_water + 1); - } - mutex_unlock(&dd->eep_lock); - if (ret) - qib_dev_err(dd, "Failed updating EEPROM\n"); - -free_bail: - vfree(buf); -bail: - return ret; -} - -/** - * qib_inc_eeprom_err - increment one of the four error counters - * that are logged to EEPROM. - * @dd: the qlogic_ib device - * @eidx: 0..3, the counter to increment - * @incr: how much to add - * - * Each counter is 8-bits, and saturates at 255 (0xFF). They - * are copied to the EEPROM (aka flash) whenever qib_update_eeprom_log() - * is called, but it can only be called in a context that allows sleep. - * This function can be called even at interrupt level. - */ -void qib_inc_eeprom_err(struct qib_devdata *dd, u32 eidx, u32 incr) -{ - uint new_val; - unsigned long flags; - - spin_lock_irqsave(&dd->eep_st_lock, flags); - new_val = dd->eep_st_new_errs[eidx] + incr; - if (new_val > 255) - new_val = 255; - dd->eep_st_new_errs[eidx] = new_val; - spin_unlock_irqrestore(&dd->eep_st_lock, flags); -} diff --git a/drivers/infiniband/hw/qib/qib_file_ops.c b/drivers/infiniband/hw/qib/qib_file_ops.c index 406fca5..a740324 100644 --- a/drivers/infiniband/hw/qib/qib_file_ops.c +++ b/drivers/infiniband/hw/qib/qib_file_ops.c @@ -43,6 +43,7 @@ #include <linux/jiffies.h> #include <asm/pgtable.h> #include <linux/delay.h> +#include <linux/export.h> #include "qib.h" #include "qib_common.h" @@ -1284,6 +1285,7 @@ static int setup_ctxt(struct qib_pportdata *ppd, int ctxt, strlcpy(rcd->comm, current->comm, sizeof(rcd->comm)); ctxt_fp(fp) = rcd; qib_stats.sps_ctxts++; + dd->freectxts--; ret = 0; goto bail; @@ -1527,6 +1529,7 @@ done_chk_sdma: struct qib_filedata *fd = fp->private_data; const struct qib_ctxtdata *rcd = fd->rcd; const struct qib_devdata *dd = rcd->dd; + unsigned int weight; if (dd->flags & QIB_HAS_SEND_DMA) { fd->pq = qib_user_sdma_queue_create(&dd->pcidev->dev, @@ -1545,8 +1548,8 @@ done_chk_sdma: * it just means that sooner or later we don't recommend * a cpu, and let the scheduler do it's best. */ - if (!ret && cpus_weight(current->cpus_allowed) >= - qib_cpulist_count) { + weight = cpumask_weight(tsk_cpus_allowed(current)); + if (!ret && weight >= qib_cpulist_count) { int cpu; cpu = find_first_zero_bit(qib_cpulist, qib_cpulist_count); @@ -1554,13 +1557,13 @@ done_chk_sdma: __set_bit(cpu, qib_cpulist); fd->rec_cpu_num = cpu; } - } else if (cpus_weight(current->cpus_allowed) == 1 && - test_bit(first_cpu(current->cpus_allowed), + } else if (weight == 1 && + test_bit(cpumask_first(tsk_cpus_allowed(current)), qib_cpulist)) qib_devinfo(dd->pcidev, "%s PID %u affinity " "set to cpu %d; already allocated\n", current->comm, current->pid, - first_cpu(current->cpus_allowed)); + cpumask_first(tsk_cpus_allowed(current))); } mutex_unlock(&qib_mutex); @@ -1791,6 +1794,7 @@ static int qib_close(struct inode *in, struct file *fp) if (dd->pageshadow) unlock_expected_tids(rcd); qib_stats.sps_ctxts--; + dd->freectxts++; } mutex_unlock(&qib_mutex); @@ -1904,8 +1908,9 @@ int qib_set_uevent_bits(struct qib_pportdata *ppd, const int evtbit) struct qib_ctxtdata *rcd; unsigned ctxt; int ret = 0; + unsigned long flags; - spin_lock(&ppd->dd->uctxt_lock); + spin_lock_irqsave(&ppd->dd->uctxt_lock, flags); for (ctxt = ppd->dd->first_user_ctxt; ctxt < ppd->dd->cfgctxts; ctxt++) { rcd = ppd->dd->rcd[ctxt]; @@ -1924,7 +1929,7 @@ int qib_set_uevent_bits(struct qib_pportdata *ppd, const int evtbit) ret = 1; break; } - spin_unlock(&ppd->dd->uctxt_lock); + spin_unlock_irqrestore(&ppd->dd->uctxt_lock, flags); return ret; } diff --git a/drivers/infiniband/hw/qib/qib_iba6120.c b/drivers/infiniband/hw/qib/qib_iba6120.c index 65df26c..2a0f7de 100644 --- a/drivers/infiniband/hw/qib/qib_iba6120.c +++ b/drivers/infiniband/hw/qib/qib_iba6120.c @@ -2674,8 +2674,6 @@ static void qib_get_6120_faststats(unsigned long opaque) spin_lock_irqsave(&dd->eep_st_lock, flags); traffic_wds -= dd->traffic_wds; dd->traffic_wds += traffic_wds; - if (traffic_wds >= QIB_TRAFFIC_ACTIVE_THRESHOLD) - atomic_add(5, &dd->active_time); /* S/B #define */ spin_unlock_irqrestore(&dd->eep_st_lock, flags); qib_chk_6120_errormask(dd); @@ -3275,6 +3273,8 @@ static int init_6120_variables(struct qib_devdata *dd) /* we always allocate at least 2048 bytes for eager buffers */ ret = ib_mtu_enum_to_int(qib_ibmtu); dd->rcvegrbufsize = ret != -1 ? max(ret, 2048) : QIB_DEFAULT_MTU; + BUG_ON(!is_power_of_2(dd->rcvegrbufsize)); + dd->rcvegrbufsize_shift = ilog2(dd->rcvegrbufsize); qib_6120_tidtemplate(dd); diff --git a/drivers/infiniband/hw/qib/qib_iba7220.c b/drivers/infiniband/hw/qib/qib_iba7220.c index 759bb63..4dc04e3 100644 --- a/drivers/infiniband/hw/qib/qib_iba7220.c +++ b/drivers/infiniband/hw/qib/qib_iba7220.c @@ -39,6 +39,7 @@ #include <linux/interrupt.h> #include <linux/pci.h> #include <linux/delay.h> +#include <linux/module.h> #include <linux/io.h> #include <rdma/ib_verbs.h> @@ -2434,6 +2435,7 @@ static int qib_7220_set_ib_cfg(struct qib_pportdata *ppd, int which, u32 val) int lsb, ret = 0, setforce = 0; u16 lcmd, licmd; unsigned long flags; + u32 tmp = 0; switch (which) { case QIB_IB_CFG_LIDLMC: @@ -2467,9 +2469,6 @@ static int qib_7220_set_ib_cfg(struct qib_pportdata *ppd, int which, u32 val) maskr = IBA7220_IBC_WIDTH_MASK; lsb = IBA7220_IBC_WIDTH_SHIFT; setforce = 1; - spin_lock_irqsave(&ppd->lflags_lock, flags); - ppd->lflags |= QIBL_IB_FORCE_NOTIFY; - spin_unlock_irqrestore(&ppd->lflags_lock, flags); break; case QIB_IB_CFG_SPD_ENB: /* set allowed Link speeds */ @@ -2643,6 +2642,28 @@ static int qib_7220_set_ib_cfg(struct qib_pportdata *ppd, int which, u32 val) goto bail; } qib_set_ib_7220_lstate(ppd, lcmd, licmd); + + maskr = IBA7220_IBC_WIDTH_MASK; + lsb = IBA7220_IBC_WIDTH_SHIFT; + tmp = (ppd->cpspec->ibcddrctrl >> lsb) & maskr; + /* If the width active on the chip does not match the + * width in the shadow register, write the new active + * width to the chip. + * We don't have to worry about speed as the speed is taken + * care of by set_7220_ibspeed_fast called by ib_updown. + */ + if (ppd->link_width_enabled-1 != tmp) { + ppd->cpspec->ibcddrctrl &= ~(maskr << lsb); + ppd->cpspec->ibcddrctrl |= + (((u64)(ppd->link_width_enabled-1) & maskr) << + lsb); + qib_write_kreg(dd, kr_ibcddrctrl, + ppd->cpspec->ibcddrctrl); + qib_write_kreg(dd, kr_scratch, 0); + spin_lock_irqsave(&ppd->lflags_lock, flags); + ppd->lflags |= QIBL_IB_FORCE_NOTIFY; + spin_unlock_irqrestore(&ppd->lflags_lock, flags); + } goto bail; case QIB_IB_CFG_HRTBT: /* set Heartbeat off/enable/auto */ @@ -3271,8 +3292,6 @@ static void qib_get_7220_faststats(unsigned long opaque) spin_lock_irqsave(&dd->eep_st_lock, flags); traffic_wds -= dd->traffic_wds; dd->traffic_wds += traffic_wds; - if (traffic_wds >= QIB_TRAFFIC_ACTIVE_THRESHOLD) - atomic_add(5, &dd->active_time); /* S/B #define */ spin_unlock_irqrestore(&dd->eep_st_lock, flags); done: mod_timer(&dd->stats_timer, jiffies + HZ * ACTIVITY_TIMER); @@ -4067,6 +4086,8 @@ static int qib_init_7220_variables(struct qib_devdata *dd) /* we always allocate at least 2048 bytes for eager buffers */ ret = ib_mtu_enum_to_int(qib_ibmtu); dd->rcvegrbufsize = ret != -1 ? max(ret, 2048) : QIB_DEFAULT_MTU; + BUG_ON(!is_power_of_2(dd->rcvegrbufsize)); + dd->rcvegrbufsize_shift = ilog2(dd->rcvegrbufsize); qib_7220_tidtemplate(dd); diff --git a/drivers/infiniband/hw/qib/qib_iba7322.c b/drivers/infiniband/hw/qib/qib_iba7322.c index 49e4a58..44180c6 100644 --- a/drivers/infiniband/hw/qib/qib_iba7322.c +++ b/drivers/infiniband/hw/qib/qib_iba7322.c @@ -40,6 +40,7 @@ #include <linux/delay.h> #include <linux/io.h> #include <linux/jiffies.h> +#include <linux/module.h> #include <rdma/ib_verbs.h> #include <rdma/ib_smi.h> @@ -114,6 +115,10 @@ static ushort qib_singleport; module_param_named(singleport, qib_singleport, ushort, S_IRUGO); MODULE_PARM_DESC(singleport, "Use only IB port 1; more per-port buffer space"); +static ushort qib_krcvq01_no_msi; +module_param_named(krcvq01_no_msi, qib_krcvq01_no_msi, ushort, S_IRUGO); +MODULE_PARM_DESC(krcvq01_no_msi, "No MSI for kctx < 2"); + /* * Receive header queue sizes */ @@ -397,7 +402,6 @@ MODULE_PARM_DESC(txselect, \ #define crp_txdroppedpkt CREG_IDX(TxDroppedPktCnt) #define crp_txhdrerr CREG_IDX(TxHeadersErrCnt) #define crp_txlenerr CREG_IDX(TxLenErrCnt) -#define crp_txlenerr CREG_IDX(TxLenErrCnt) #define crp_txminmaxlenerr CREG_IDX(TxMaxMinLenErrCnt) #define crp_txsdmadesc CREG_IDX(TxSDmaDescCnt) #define crp_txunderrun CREG_IDX(TxUnderrunCnt) @@ -1107,9 +1111,9 @@ static inline u32 read_7322_creg32_port(const struct qib_pportdata *ppd, #define AUTONEG_TRIES 3 /* sequential retries to negotiate DDR */ #define HWE_AUTO(fldname) { .mask = SYM_MASK(HwErrMask, fldname##Mask), \ - .msg = #fldname } + .msg = #fldname , .sz = sizeof(#fldname) } #define HWE_AUTO_P(fldname, port) { .mask = SYM_MASK(HwErrMask, \ - fldname##Mask##_##port), .msg = #fldname } + fldname##Mask##_##port), .msg = #fldname , .sz = sizeof(#fldname) } static const struct qib_hwerror_msgs qib_7322_hwerror_msgs[] = { HWE_AUTO_P(IBSerdesPClkNotDetect, 1), HWE_AUTO_P(IBSerdesPClkNotDetect, 0), @@ -1127,14 +1131,16 @@ static const struct qib_hwerror_msgs qib_7322_hwerror_msgs[] = { HWE_AUTO_P(IBCBusFromSPCParityErr, 0), HWE_AUTO(statusValidNoEop), HWE_AUTO(LATriggered), - { .mask = 0 } + { .mask = 0, .sz = 0 } }; #define E_AUTO(fldname) { .mask = SYM_MASK(ErrMask, fldname##Mask), \ - .msg = #fldname } + .msg = #fldname, .sz = sizeof(#fldname) } #define E_P_AUTO(fldname) { .mask = SYM_MASK(ErrMask_0, fldname##Mask), \ - .msg = #fldname } + .msg = #fldname, .sz = sizeof(#fldname) } static const struct qib_hwerror_msgs qib_7322error_msgs[] = { + E_AUTO(RcvEgrFullErr), + E_AUTO(RcvHdrFullErr), E_AUTO(ResetNegated), E_AUTO(HardwareErr), E_AUTO(InvalidAddrErr), @@ -1147,9 +1153,7 @@ static const struct qib_hwerror_msgs qib_7322error_msgs[] = { E_AUTO(SendSpecialTriggerErr), E_AUTO(SDmaWrongPortErr), E_AUTO(SDmaBufMaskDuplicateErr), - E_AUTO(RcvHdrFullErr), - E_AUTO(RcvEgrFullErr), - { .mask = 0 } + { .mask = 0, .sz = 0 } }; static const struct qib_hwerror_msgs qib_7322p_error_msgs[] = { @@ -1159,7 +1163,8 @@ static const struct qib_hwerror_msgs qib_7322p_error_msgs[] = { /* * SDmaHaltErr is not really an error, make it clearer; */ - {.mask = SYM_MASK(ErrMask_0, SDmaHaltErrMask), .msg = "SDmaHalted"}, + {.mask = SYM_MASK(ErrMask_0, SDmaHaltErrMask), .msg = "SDmaHalted", + .sz = 11}, E_P_AUTO(SDmaDescAddrMisalignErr), E_P_AUTO(SDmaUnexpDataErr), E_P_AUTO(SDmaMissingDwErr), @@ -1195,7 +1200,7 @@ static const struct qib_hwerror_msgs qib_7322p_error_msgs[] = { E_P_AUTO(RcvICRCErr), E_P_AUTO(RcvVCRCErr), E_P_AUTO(RcvFormatErr), - { .mask = 0 } + { .mask = 0, .sz = 0 } }; /* @@ -1203,17 +1208,17 @@ static const struct qib_hwerror_msgs qib_7322p_error_msgs[] = { * context */ #define INTR_AUTO(fldname) { .mask = SYM_MASK(IntMask, fldname##Mask), \ - .msg = #fldname } + .msg = #fldname, .sz = sizeof(#fldname) } /* Below generates "auto-message" for interrupts specific to a port */ #define INTR_AUTO_P(fldname) { .mask = MASK_ACROSS(\ SYM_LSB(IntMask, fldname##Mask##_0), \ SYM_LSB(IntMask, fldname##Mask##_1)), \ - .msg = #fldname "_P" } + .msg = #fldname "_P", .sz = sizeof(#fldname "_P") } /* For some reason, the SerDesTrimDone bits are reversed */ #define INTR_AUTO_PI(fldname) { .mask = MASK_ACROSS(\ SYM_LSB(IntMask, fldname##Mask##_1), \ SYM_LSB(IntMask, fldname##Mask##_0)), \ - .msg = #fldname "_P" } + .msg = #fldname "_P", .sz = sizeof(#fldname "_P") } /* * Below generates "auto-message" for interrupts specific to a context, * with ctxt-number appended @@ -1221,7 +1226,7 @@ static const struct qib_hwerror_msgs qib_7322p_error_msgs[] = { #define INTR_AUTO_C(fldname) { .mask = MASK_ACROSS(\ SYM_LSB(IntMask, fldname##0IntMask), \ SYM_LSB(IntMask, fldname##17IntMask)), \ - .msg = #fldname "_C"} + .msg = #fldname "_C", .sz = sizeof(#fldname "_C") } static const struct qib_hwerror_msgs qib_7322_intr_msgs[] = { INTR_AUTO_P(SDmaInt), @@ -1235,11 +1240,12 @@ static const struct qib_hwerror_msgs qib_7322_intr_msgs[] = { INTR_AUTO_P(SendDoneInt), INTR_AUTO(SendBufAvailInt), INTR_AUTO_C(RcvAvail), - { .mask = 0 } + { .mask = 0, .sz = 0 } }; #define TXSYMPTOM_AUTO_P(fldname) \ - { .mask = SYM_MASK(SendHdrErrSymptom_0, fldname), .msg = #fldname } + { .mask = SYM_MASK(SendHdrErrSymptom_0, fldname), \ + .msg = #fldname, .sz = sizeof(#fldname) } static const struct qib_hwerror_msgs hdrchk_msgs[] = { TXSYMPTOM_AUTO_P(NonKeyPacket), TXSYMPTOM_AUTO_P(GRHFail), @@ -1248,7 +1254,7 @@ static const struct qib_hwerror_msgs hdrchk_msgs[] = { TXSYMPTOM_AUTO_P(SLIDFail), TXSYMPTOM_AUTO_P(RawIPV6), TXSYMPTOM_AUTO_P(PacketTooSmall), - { .mask = 0 } + { .mask = 0, .sz = 0 } }; #define IBA7322_HDRHEAD_PKTINT_SHIFT 32 /* interrupt cnt in upper 32 bits */ @@ -1293,7 +1299,7 @@ static void err_decode(char *msg, size_t len, u64 errs, u64 these, lmask; int took, multi, n = 0; - while (msp && msp->mask) { + while (errs && msp && msp->mask) { multi = (msp->mask & (msp->mask - 1)); while (errs & msp->mask) { these = (errs & msp->mask); @@ -1304,9 +1310,14 @@ static void err_decode(char *msg, size_t len, u64 errs, *msg++ = ','; len--; } - took = scnprintf(msg, len, "%s", msp->msg); + BUG_ON(!msp->sz); + /* msp->sz counts the nul */ + took = min_t(size_t, msp->sz - (size_t)1, len); + memcpy(msg, msp->msg, took); len -= took; msg += took; + if (len) + *msg = '\0'; } errs &= ~lmask; if (len && multi) { @@ -1644,6 +1655,14 @@ done: return; } +static void qib_error_tasklet(unsigned long data) +{ + struct qib_devdata *dd = (struct qib_devdata *)data; + + handle_7322_errors(dd); + qib_write_kreg(dd, kr_errmask, dd->cspec->errormask); +} + static void reenable_chase(unsigned long opaque) { struct qib_pportdata *ppd = (struct qib_pportdata *)opaque; @@ -2260,6 +2279,11 @@ 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, @@ -2288,16 +2312,11 @@ static int qib_7322_bringup_serdes(struct qib_pportdata *ppd) SYM_LSB(IBCCtrlA_0, MaxPktLen); ppd->cpspec->ibcctrl_a = ibc; /* without linkcmd or linkinitcmd! */ - /* initially come up waiting for TS1, without sending anything. */ - val = ppd->cpspec->ibcctrl_a | (QLOGIC_IB_IBCC_LINKINITCMD_DISABLE << - QLOGIC_IB_IBCC_LINKINITCMD_SHIFT); - /* * Reset the PCS interface to the serdes (and also ibc, which is still * in reset from above). Writes new value of ibcctrl_a as last step. */ qib_7322_mini_pcs_reset(ppd); - qib_write_kreg(dd, kr_scratch, 0ULL); if (!ppd->cpspec->ibcctrl_b) { unsigned lse = ppd->link_speed_enabled; @@ -2363,17 +2382,20 @@ static int qib_7322_bringup_serdes(struct qib_pportdata *ppd) ppd->cpspec->ibcctrl_a |= SYM_MASK(IBCCtrlA_0, IBLinkEn); set_vls(ppd); + /* initially come up DISABLED, without sending anything. */ + val = ppd->cpspec->ibcctrl_a | (QLOGIC_IB_IBCC_LINKINITCMD_DISABLE << + QLOGIC_IB_IBCC_LINKINITCMD_SHIFT); + qib_write_kreg_port(ppd, krp_ibcctrl_a, val); + qib_write_kreg(dd, kr_scratch, 0ULL); + /* clear the linkinit cmds */ + ppd->cpspec->ibcctrl_a = val & ~SYM_MASK(IBCCtrlA_0, LinkInitCmd); + /* be paranoid against later code motion, etc. */ spin_lock_irqsave(&dd->cspec->rcvmod_lock, flags); ppd->p_rcvctrl |= SYM_MASK(RcvCtrl_0, RcvIBPortEnable); qib_write_kreg_port(ppd, krp_rcvctrl, ppd->p_rcvctrl); spin_unlock_irqrestore(&dd->cspec->rcvmod_lock, flags); - /* Hold the link state machine for mezz boards */ - if (IS_QMH(dd) || IS_QME(dd)) - qib_set_ib_7322_lstate(ppd, 0, - QLOGIC_IB_IBCC_LINKINITCMD_DISABLE); - /* Also enable IBSTATUSCHG interrupt. */ val = qib_read_kreg_port(ppd, krp_errmask); qib_write_kreg_port(ppd, krp_errmask, @@ -2725,8 +2747,10 @@ static noinline void unlikely_7322_intr(struct qib_devdata *dd, u64 istat) unknown_7322_ibits(dd, istat); if (istat & QIB_I_GPIO) unknown_7322_gpio_intr(dd); - if (istat & QIB_I_C_ERROR) - handle_7322_errors(dd); + if (istat & QIB_I_C_ERROR) { + qib_write_kreg(dd, kr_errmask, 0ULL); + tasklet_schedule(&dd->error_tasklet); + } if (istat & INT_MASK_P(Err, 0) && dd->rcd[0]) handle_7322_p_errors(dd->rcd[0]->ppd); if (istat & INT_MASK_P(Err, 1) && dd->rcd[1]) @@ -2833,9 +2857,8 @@ static irqreturn_t qib_7322intr(int irq, void *data) for (i = 0; i < dd->first_user_ctxt; i++) { if (ctxtrbits & rmask) { ctxtrbits &= ~rmask; - if (dd->rcd[i]) { + if (dd->rcd[i]) qib_kreceive(dd->rcd[i], NULL, &npkts); - } } rmask <<= 1; } @@ -3125,6 +3148,8 @@ try_intx: arg = dd->rcd[ctxt]; if (!arg) continue; + if (qib_krcvq01_no_msi && ctxt < 2) + continue; lsb = QIB_I_RCVAVAIL_LSB + ctxt; handler = qib_7322pintr; name = QIB_DRV_NAME " (kctx)"; @@ -3159,6 +3184,8 @@ try_intx: for (i = 0; i < ARRAY_SIZE(redirect); i++) qib_write_kreg(dd, kr_intredirect + i, redirect[i]); dd->cspec->main_int_mask = mask; + tasklet_init(&dd->error_tasklet, qib_error_tasklet, + (unsigned long)dd); bail:; } @@ -4766,8 +4793,6 @@ static void qib_get_7322_faststats(unsigned long opaque) spin_lock_irqsave(&ppd->dd->eep_st_lock, flags); traffic_wds -= ppd->dd->traffic_wds; ppd->dd->traffic_wds += traffic_wds; - if (traffic_wds >= QIB_TRAFFIC_ACTIVE_THRESHOLD) - atomic_add(ACTIVITY_TIMER, &ppd->dd->active_time); spin_unlock_irqrestore(&ppd->dd->eep_st_lock, flags); if (ppd->cpspec->qdr_dfe_on && (ppd->link_speed_active & QIB_IB_QDR) && @@ -5208,6 +5233,8 @@ static int qib_7322_ib_updown(struct qib_pportdata *ppd, int ibup, u64 ibcs) QIBL_IB_AUTONEG_INPROG))) set_7322_ibspeed_fast(ppd, ppd->link_speed_enabled); if (!(ppd->lflags & QIBL_IB_AUTONEG_INPROG)) { + struct qib_qsfp_data *qd = + &ppd->cpspec->qsfp_data; /* unlock the Tx settings, speed may change */ qib_write_kreg_port(ppd, krp_tx_deemph_override, SYM_MASK(IBSD_TX_DEEMPHASIS_OVERRIDE_0, @@ -5215,6 +5242,12 @@ static int qib_7322_ib_updown(struct qib_pportdata *ppd, int ibup, u64 ibcs) qib_cancel_sends(ppd); /* on link down, ensure sane pcs state */ qib_7322_mini_pcs_reset(ppd); + /* schedule the qsfp refresh which should turn the link + off */ + if (ppd->dd->flags & QIB_HAS_QSFP) { + qd->t_insert = get_jiffies_64(); + queue_work(ib_wq, &qd->work); + } spin_lock_irqsave(&ppd->sdma_lock, flags); if (__qib_sdma_running(ppd)) __qib_sdma_process_event(ppd, @@ -5565,43 +5598,79 @@ static void qsfp_7322_event(struct work_struct *work) struct qib_qsfp_data *qd; struct qib_pportdata *ppd; u64 pwrup; + unsigned long flags; int ret; u32 le2; qd = container_of(work, struct qib_qsfp_data, work); ppd = qd->ppd; - pwrup = qd->t_insert + msecs_to_jiffies(QSFP_PWR_LAG_MSEC); + pwrup = qd->t_insert + + msecs_to_jiffies(QSFP_PWR_LAG_MSEC - QSFP_MODPRS_LAG_MSEC); - /* - * Some QSFP's not only do not respond until the full power-up - * time, but may behave badly if we try. So hold off responding - * to insertion. - */ - while (1) { - u64 now = get_jiffies_64(); - if (time_after64(now, pwrup)) - break; - msleep(20); - } - ret = qib_refresh_qsfp_cache(ppd, &qd->cache); - /* - * Need to change LE2 back to defaults if we couldn't - * read the cable type (to handle cable swaps), so do this - * even on failure to read cable information. We don't - * get here for QME, so IS_QME check not needed here. - */ - if (!ret && !ppd->dd->cspec->r1) { - if (QSFP_IS_ACTIVE_FAR(qd->cache.tech)) - le2 = LE2_QME; - else if (qd->cache.atten[1] >= qib_long_atten && - QSFP_IS_CU(qd->cache.tech)) - le2 = LE2_5m; - else + /* Delay for 20 msecs to allow ModPrs resistor to setup */ + mdelay(QSFP_MODPRS_LAG_MSEC); + + if (!qib_qsfp_mod_present(ppd)) { + ppd->cpspec->qsfp_data.modpresent = 0; + /* Set the physical link to disabled */ + qib_set_ib_7322_lstate(ppd, 0, + QLOGIC_IB_IBCC_LINKINITCMD_DISABLE); + spin_lock_irqsave(&ppd->lflags_lock, flags); + ppd->lflags &= ~QIBL_LINKV; + spin_unlock_irqrestore(&ppd->lflags_lock, flags); + } else { + /* + * Some QSFP's not only do not respond until the full power-up + * time, but may behave badly if we try. So hold off responding + * to insertion. + */ + while (1) { + u64 now = get_jiffies_64(); + if (time_after64(now, pwrup)) + break; + msleep(20); + } + + ret = qib_refresh_qsfp_cache(ppd, &qd->cache); + + /* + * Need to change LE2 back to defaults if we couldn't + * read the cable type (to handle cable swaps), so do this + * even on failure to read cable information. We don't + * get here for QME, so IS_QME check not needed here. + */ + if (!ret && !ppd->dd->cspec->r1) { + if (QSFP_IS_ACTIVE_FAR(qd->cache.tech)) + le2 = LE2_QME; + else if (qd->cache.atten[1] >= qib_long_atten && + QSFP_IS_CU(qd->cache.tech)) + le2 = LE2_5m; + else + le2 = LE2_DEFAULT; + } else le2 = LE2_DEFAULT; - } else - le2 = LE2_DEFAULT; - ibsd_wr_allchans(ppd, 13, (le2 << 7), BMASK(9, 7)); - init_txdds_table(ppd, 0); + ibsd_wr_allchans(ppd, 13, (le2 << 7), BMASK(9, 7)); + /* + * We always change parameteters, since we can choose + * values for cables without eeproms, and the cable may have + * changed from a cable with full or partial eeprom content + * to one with partial or no content. + */ + init_txdds_table(ppd, 0); + /* The physical link is being re-enabled only when the + * previous state was DISABLED and the VALID bit is not + * set. This should only happen when the cable has been + * physically pulled. */ + if (!ppd->cpspec->qsfp_data.modpresent && + (ppd->lflags & (QIBL_LINKV | QIBL_IB_LINK_DISABLED))) { + ppd->cpspec->qsfp_data.modpresent = 1; + qib_set_ib_7322_lstate(ppd, 0, + QLOGIC_IB_IBCC_LINKINITCMD_SLEEP); + spin_lock_irqsave(&ppd->lflags_lock, flags); + ppd->lflags |= QIBL_LINKV; + spin_unlock_irqrestore(&ppd->lflags_lock, flags); + } + } } /* @@ -5705,7 +5774,8 @@ static void set_no_qsfp_atten(struct qib_devdata *dd, int change) /* now change the IBC and serdes, overriding generic */ init_txdds_table(ppd, 1); /* Re-enable the physical state machine on mezz boards - * now that the correct settings have been set. */ + * now that the correct settings have been set. + * QSFP boards are handles by the QSFP event handler */ if (IS_QMH(dd) || IS_QME(dd)) qib_set_ib_7322_lstate(ppd, 0, QLOGIC_IB_IBCC_LINKINITCMD_SLEEP); @@ -6183,6 +6253,8 @@ static int qib_init_7322_variables(struct qib_devdata *dd) /* we always allocate at least 2048 bytes for eager buffers */ dd->rcvegrbufsize = max(mtu, 2048); + BUG_ON(!is_power_of_2(dd->rcvegrbufsize)); + dd->rcvegrbufsize_shift = ilog2(dd->rcvegrbufsize); qib_7322_tidtemplate(dd); @@ -6790,6 +6862,10 @@ struct qib_devdata *qib_init_iba7322_funcs(struct pci_dev *pdev, (i >= ARRAY_SIZE(irq_table) && dd->rcd[i - ARRAY_SIZE(irq_table)])) actual_cnt++; + /* reduce by ctxt's < 2 */ + if (qib_krcvq01_no_msi) + actual_cnt -= dd->num_pports; + tabsize = actual_cnt; dd->cspec->msix_entries = kmalloc(tabsize * sizeof(struct msix_entry), GFP_KERNEL); @@ -7121,7 +7197,8 @@ static void find_best_ent(struct qib_pportdata *ppd, } } - /* Lookup serdes setting by cable type and attenuation */ + /* Active cables don't have attenuation so we only set SERDES + * settings to account for the attenuation of the board traces. */ if (!override && QSFP_IS_ACTIVE(qd->tech)) { *sdr_dds = txdds_sdr + ppd->dd->board_atten; *ddr_dds = txdds_ddr + ppd->dd->board_atten; @@ -7438,12 +7515,6 @@ static int serdes_7322_init_new(struct qib_pportdata *ppd) u32 le_val, rxcaldone; int chan, chan_done = (1 << SERDES_CHANS) - 1; - /* - * Initialize the Tx DDS tables. Also done every QSFP event, - * for adapters with QSFP - */ - init_txdds_table(ppd, 0); - /* Clear cmode-override, may be set from older driver */ ahb_mod(ppd->dd, IBSD(ppd->hw_pidx), 5, 10, 0 << 14, 1 << 14); @@ -7629,6 +7700,12 @@ static int serdes_7322_init_new(struct qib_pportdata *ppd) /* VGA output common mode */ ibsd_wr_allchans(ppd, 12, (3 << 2), BMASK(3, 2)); + /* + * Initialize the Tx DDS tables. Also done every QSFP event, + * for adapters with QSFP + */ + init_txdds_table(ppd, 0); + return 0; } diff --git a/drivers/infiniband/hw/qib/qib_init.c b/drivers/infiniband/hw/qib/qib_init.c index a01f3fc..9966ec2 100644 --- a/drivers/infiniband/hw/qib/qib_init.c +++ b/drivers/infiniband/hw/qib/qib_init.c @@ -37,6 +37,7 @@ #include <linux/vmalloc.h> #include <linux/delay.h> #include <linux/idr.h> +#include <linux/module.h> #include "qib.h" #include "qib_common.h" @@ -183,6 +184,9 @@ struct qib_ctxtdata *qib_create_ctxtdata(struct qib_pportdata *ppd, u32 ctxt) rcd->rcvegrbuf_chunks = (rcd->rcvegrcnt + rcd->rcvegrbufs_perchunk - 1) / rcd->rcvegrbufs_perchunk; + BUG_ON(!is_power_of_2(rcd->rcvegrbufs_perchunk)); + rcd->rcvegrbufs_perchunk_shift = + ilog2(rcd->rcvegrbufs_perchunk); } return rcd; } @@ -398,6 +402,7 @@ static void enable_chip(struct qib_devdata *dd) if (rcd) dd->f_rcvctrl(rcd->ppd, rcvmask, i); } + dd->freectxts = dd->cfgctxts - dd->first_user_ctxt; } static void verify_interrupt(unsigned long opaque) @@ -581,10 +586,6 @@ int qib_init(struct qib_devdata *dd, int reinit) continue; } - /* let link come up, and enable IBC */ - spin_lock_irqsave(&ppd->lflags_lock, flags); - ppd->lflags &= ~QIBL_IB_LINK_DISABLED; - spin_unlock_irqrestore(&ppd->lflags_lock, flags); portok++; } @@ -764,7 +765,6 @@ static void qib_shutdown_device(struct qib_devdata *dd) dd->f_quiet_serdes(ppd); } - qib_update_eeprom_log(dd); } /** diff --git a/drivers/infiniband/hw/qib/qib_keys.c b/drivers/infiniband/hw/qib/qib_keys.c index 8fd19a4..ca6e6cf 100644 --- a/drivers/infiniband/hw/qib/qib_keys.c +++ b/drivers/infiniband/hw/qib/qib_keys.c @@ -69,6 +69,10 @@ int qib_alloc_lkey(struct qib_lkey_table *rkt, struct qib_mregion *mr) * unrestricted LKEY. */ rkt->gen++; + /* + * bits are capped in qib_verbs.c to insure enough bits + * for generation number + */ mr->lkey = (r << (32 - ib_qib_lkey_table_size)) | ((((1 << (24 - ib_qib_lkey_table_size)) - 1) & rkt->gen) << 8); diff --git a/drivers/infiniband/hw/qib/qib_mad.c b/drivers/infiniband/hw/qib/qib_mad.c index 8fd3df5..dd2ad6d3 100644 --- a/drivers/infiniband/hw/qib/qib_mad.c +++ b/drivers/infiniband/hw/qib/qib_mad.c @@ -1007,7 +1007,7 @@ 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; @@ -1125,22 +1125,22 @@ static int subn_trap_repress(struct ib_smp *smp, struct ib_device *ibdev, return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED; } -static int pma_get_classportinfo(struct ib_perf *pmp, +static int pma_get_classportinfo(struct ib_pma_mad *pmp, struct ib_device *ibdev) { - struct ib_pma_classportinfo *p = - (struct ib_pma_classportinfo *)pmp->data; + struct ib_class_port_info *p = + (struct ib_class_port_info *)pmp->data; struct qib_devdata *dd = dd_from_ibdev(ibdev); memset(pmp->data, 0, sizeof(pmp->data)); - if (pmp->attr_mod != 0) - pmp->status |= IB_SMP_INVALID_FIELD; + if (pmp->mad_hdr.attr_mod != 0) + pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD; /* Note that AllPortSelect is not valid */ p->base_version = 1; p->class_version = 1; - p->cap_mask = IB_PMA_CLASS_CAP_EXT_WIDTH; + p->capability_mask = IB_PMA_CLASS_CAP_EXT_WIDTH; /* * Set the most significant bit of CM2 to indicate support for * congestion statistics @@ -1154,7 +1154,7 @@ static int pma_get_classportinfo(struct ib_perf *pmp, return reply((struct ib_smp *) pmp); } -static int pma_get_portsamplescontrol(struct ib_perf *pmp, +static int pma_get_portsamplescontrol(struct ib_pma_mad *pmp, struct ib_device *ibdev, u8 port) { struct ib_pma_portsamplescontrol *p = @@ -1169,8 +1169,8 @@ static int pma_get_portsamplescontrol(struct ib_perf *pmp, memset(pmp->data, 0, sizeof(pmp->data)); p->port_select = port_select; - if (pmp->attr_mod != 0 || port_select != port) { - pmp->status |= IB_SMP_INVALID_FIELD; + if (pmp->mad_hdr.attr_mod != 0 || port_select != port) { + pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD; goto bail; } spin_lock_irqsave(&ibp->lock, flags); @@ -1192,7 +1192,7 @@ bail: return reply((struct ib_smp *) pmp); } -static int pma_set_portsamplescontrol(struct ib_perf *pmp, +static int pma_set_portsamplescontrol(struct ib_pma_mad *pmp, struct ib_device *ibdev, u8 port) { struct ib_pma_portsamplescontrol *p = @@ -1205,8 +1205,8 @@ static int pma_set_portsamplescontrol(struct ib_perf *pmp, u8 status, xmit_flags; int ret; - if (pmp->attr_mod != 0 || p->port_select != port) { - pmp->status |= IB_SMP_INVALID_FIELD; + if (pmp->mad_hdr.attr_mod != 0 || p->port_select != port) { + pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD; ret = reply((struct ib_smp *) pmp); goto bail; } @@ -1321,7 +1321,7 @@ static u64 get_cache_hw_sample_counters(struct qib_pportdata *ppd, return ret; } -static int pma_get_portsamplesresult(struct ib_perf *pmp, +static int pma_get_portsamplesresult(struct ib_pma_mad *pmp, struct ib_device *ibdev, u8 port) { struct ib_pma_portsamplesresult *p = @@ -1360,7 +1360,7 @@ static int pma_get_portsamplesresult(struct ib_perf *pmp, return reply((struct ib_smp *) pmp); } -static int pma_get_portsamplesresult_ext(struct ib_perf *pmp, +static int pma_get_portsamplesresult_ext(struct ib_pma_mad *pmp, struct ib_device *ibdev, u8 port) { struct ib_pma_portsamplesresult_ext *p = @@ -1402,7 +1402,7 @@ static int pma_get_portsamplesresult_ext(struct ib_perf *pmp, return reply((struct ib_smp *) pmp); } -static int pma_get_portcounters(struct ib_perf *pmp, +static int pma_get_portcounters(struct ib_pma_mad *pmp, struct ib_device *ibdev, u8 port) { struct ib_pma_portcounters *p = (struct ib_pma_portcounters *) @@ -1436,8 +1436,8 @@ static int pma_get_portcounters(struct ib_perf *pmp, memset(pmp->data, 0, sizeof(pmp->data)); p->port_select = port_select; - if (pmp->attr_mod != 0 || port_select != port) - pmp->status |= IB_SMP_INVALID_FIELD; + if (pmp->mad_hdr.attr_mod != 0 || port_select != port) + pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD; if (cntrs.symbol_error_counter > 0xFFFFUL) p->symbol_error_counter = cpu_to_be16(0xFFFF); @@ -1472,7 +1472,7 @@ static int pma_get_portcounters(struct ib_perf *pmp, cntrs.local_link_integrity_errors = 0xFUL; if (cntrs.excessive_buffer_overrun_errors > 0xFUL) cntrs.excessive_buffer_overrun_errors = 0xFUL; - p->lli_ebor_errors = (cntrs.local_link_integrity_errors << 4) | + p->link_overrun_errors = (cntrs.local_link_integrity_errors << 4) | cntrs.excessive_buffer_overrun_errors; if (cntrs.vl15_dropped > 0xFFFFUL) p->vl15_dropped = cpu_to_be16(0xFFFF); @@ -1500,7 +1500,7 @@ static int pma_get_portcounters(struct ib_perf *pmp, return reply((struct ib_smp *) pmp); } -static int pma_get_portcounters_cong(struct ib_perf *pmp, +static int pma_get_portcounters_cong(struct ib_pma_mad *pmp, struct ib_device *ibdev, u8 port) { /* Congestion PMA packets start at offset 24 not 64 */ @@ -1510,7 +1510,7 @@ static int pma_get_portcounters_cong(struct ib_perf *pmp, struct qib_ibport *ibp = to_iport(ibdev, port); struct qib_pportdata *ppd = ppd_from_ibp(ibp); struct qib_devdata *dd = dd_from_ppd(ppd); - u32 port_select = be32_to_cpu(pmp->attr_mod) & 0xFF; + u32 port_select = be32_to_cpu(pmp->mad_hdr.attr_mod) & 0xFF; u64 xmit_wait_counter; unsigned long flags; @@ -1519,9 +1519,9 @@ static int pma_get_portcounters_cong(struct ib_perf *pmp, * SET method ends up calling this anyway. */ if (!dd->psxmitwait_supported) - pmp->status |= IB_SMP_UNSUP_METH_ATTR; + pmp->mad_hdr.status |= IB_SMP_UNSUP_METH_ATTR; if (port_select != port) - pmp->status |= IB_SMP_INVALID_FIELD; + pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD; qib_get_counters(ppd, &cntrs); spin_lock_irqsave(&ppd->ibport_data.lock, flags); @@ -1603,7 +1603,7 @@ static int pma_get_portcounters_cong(struct ib_perf *pmp, cntrs.local_link_integrity_errors = 0xFUL; if (cntrs.excessive_buffer_overrun_errors > 0xFUL) cntrs.excessive_buffer_overrun_errors = 0xFUL; - p->lli_ebor_errors = (cntrs.local_link_integrity_errors << 4) | + p->link_overrun_errors = (cntrs.local_link_integrity_errors << 4) | cntrs.excessive_buffer_overrun_errors; if (cntrs.vl15_dropped > 0xFFFFUL) p->vl15_dropped = cpu_to_be16(0xFFFF); @@ -1613,7 +1613,7 @@ static int pma_get_portcounters_cong(struct ib_perf *pmp, return reply((struct ib_smp *)pmp); } -static int pma_get_portcounters_ext(struct ib_perf *pmp, +static int pma_get_portcounters_ext(struct ib_pma_mad *pmp, struct ib_device *ibdev, u8 port) { struct ib_pma_portcounters_ext *p = @@ -1626,8 +1626,8 @@ static int pma_get_portcounters_ext(struct ib_perf *pmp, memset(pmp->data, 0, sizeof(pmp->data)); p->port_select = port_select; - if (pmp->attr_mod != 0 || port_select != port) { - pmp->status |= IB_SMP_INVALID_FIELD; + if (pmp->mad_hdr.attr_mod != 0 || port_select != port) { + pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD; goto bail; } @@ -1652,7 +1652,7 @@ bail: return reply((struct ib_smp *) pmp); } -static int pma_set_portcounters(struct ib_perf *pmp, +static int pma_set_portcounters(struct ib_pma_mad *pmp, struct ib_device *ibdev, u8 port) { struct ib_pma_portcounters *p = (struct ib_pma_portcounters *) @@ -1715,14 +1715,14 @@ static int pma_set_portcounters(struct ib_perf *pmp, return pma_get_portcounters(pmp, ibdev, port); } -static int pma_set_portcounters_cong(struct ib_perf *pmp, +static int pma_set_portcounters_cong(struct ib_pma_mad *pmp, struct ib_device *ibdev, u8 port) { struct qib_ibport *ibp = to_iport(ibdev, port); struct qib_pportdata *ppd = ppd_from_ibp(ibp); struct qib_devdata *dd = dd_from_ppd(ppd); struct qib_verbs_counters cntrs; - u32 counter_select = (be32_to_cpu(pmp->attr_mod) >> 24) & 0xFF; + u32 counter_select = (be32_to_cpu(pmp->mad_hdr.attr_mod) >> 24) & 0xFF; int ret = 0; unsigned long flags; @@ -1766,7 +1766,7 @@ static int pma_set_portcounters_cong(struct ib_perf *pmp, return ret; } -static int pma_set_portcounters_ext(struct ib_perf *pmp, +static int pma_set_portcounters_ext(struct ib_pma_mad *pmp, struct ib_device *ibdev, u8 port) { struct ib_pma_portcounters *p = (struct ib_pma_portcounters *) @@ -1959,19 +1959,19 @@ static int process_perf(struct ib_device *ibdev, u8 port, struct ib_mad *in_mad, struct ib_mad *out_mad) { - struct ib_perf *pmp = (struct ib_perf *)out_mad; + struct ib_pma_mad *pmp = (struct ib_pma_mad *)out_mad; int ret; *out_mad = *in_mad; - if (pmp->class_version != 1) { - pmp->status |= IB_SMP_UNSUP_VERSION; + if (pmp->mad_hdr.class_version != 1) { + pmp->mad_hdr.status |= IB_SMP_UNSUP_VERSION; ret = reply((struct ib_smp *) pmp); goto bail; } - switch (pmp->method) { + switch (pmp->mad_hdr.method) { case IB_MGMT_METHOD_GET: - switch (pmp->attr_id) { + switch (pmp->mad_hdr.attr_id) { case IB_PMA_CLASS_PORT_INFO: ret = pma_get_classportinfo(pmp, ibdev); goto bail; @@ -1994,13 +1994,13 @@ static int process_perf(struct ib_device *ibdev, u8 port, ret = pma_get_portcounters_cong(pmp, ibdev, port); goto bail; default: - pmp->status |= IB_SMP_UNSUP_METH_ATTR; + pmp->mad_hdr.status |= IB_SMP_UNSUP_METH_ATTR; ret = reply((struct ib_smp *) pmp); goto bail; } case IB_MGMT_METHOD_SET: - switch (pmp->attr_id) { + switch (pmp->mad_hdr.attr_id) { case IB_PMA_PORT_SAMPLES_CONTROL: ret = pma_set_portsamplescontrol(pmp, ibdev, port); goto bail; @@ -2014,7 +2014,7 @@ static int process_perf(struct ib_device *ibdev, u8 port, ret = pma_set_portcounters_cong(pmp, ibdev, port); goto bail; default: - pmp->status |= IB_SMP_UNSUP_METH_ATTR; + pmp->mad_hdr.status |= IB_SMP_UNSUP_METH_ATTR; ret = reply((struct ib_smp *) pmp); goto bail; } @@ -2030,7 +2030,7 @@ static int process_perf(struct ib_device *ibdev, u8 port, goto bail; default: - pmp->status |= IB_SMP_UNSUP_METHOD; + pmp->mad_hdr.status |= IB_SMP_UNSUP_METHOD; ret = reply((struct ib_smp *) pmp); } diff --git a/drivers/infiniband/hw/qib/qib_mad.h b/drivers/infiniband/hw/qib/qib_mad.h index 7840ab5..ecc416c 100644 --- a/drivers/infiniband/hw/qib/qib_mad.h +++ b/drivers/infiniband/hw/qib/qib_mad.h @@ -32,6 +32,8 @@ * SOFTWARE. */ +#include <rdma/ib_pma.h> + #define IB_SMP_UNSUP_VERSION cpu_to_be16(0x0004) #define IB_SMP_UNSUP_METHOD cpu_to_be16(0x0008) #define IB_SMP_UNSUP_METH_ATTR cpu_to_be16(0x000C) @@ -180,109 +182,8 @@ struct ib_vl_weight_elem { #define IB_VLARB_HIGHPRI_0_31 3 #define IB_VLARB_HIGHPRI_32_63 4 -/* - * PMA class portinfo capability mask bits - */ -#define IB_PMA_CLASS_CAP_ALLPORTSELECT cpu_to_be16(1 << 8) -#define IB_PMA_CLASS_CAP_EXT_WIDTH cpu_to_be16(1 << 9) -#define IB_PMA_CLASS_CAP_XMIT_WAIT cpu_to_be16(1 << 12) - -#define IB_PMA_CLASS_PORT_INFO cpu_to_be16(0x0001) -#define IB_PMA_PORT_SAMPLES_CONTROL cpu_to_be16(0x0010) -#define IB_PMA_PORT_SAMPLES_RESULT cpu_to_be16(0x0011) -#define IB_PMA_PORT_COUNTERS cpu_to_be16(0x0012) -#define IB_PMA_PORT_COUNTERS_EXT cpu_to_be16(0x001D) -#define IB_PMA_PORT_SAMPLES_RESULT_EXT cpu_to_be16(0x001E) #define IB_PMA_PORT_COUNTERS_CONG cpu_to_be16(0xFF00) -struct ib_perf { - u8 base_version; - u8 mgmt_class; - u8 class_version; - u8 method; - __be16 status; - __be16 unused; - __be64 tid; - __be16 attr_id; - __be16 resv; - __be32 attr_mod; - u8 reserved[40]; - u8 data[192]; -} __attribute__ ((packed)); - -struct ib_pma_classportinfo { - u8 base_version; - u8 class_version; - __be16 cap_mask; - u8 reserved[3]; - u8 resp_time_value; /* only lower 5 bits */ - union ib_gid redirect_gid; - __be32 redirect_tc_sl_fl; /* 8, 4, 20 bits respectively */ - __be16 redirect_lid; - __be16 redirect_pkey; - __be32 redirect_qp; /* only lower 24 bits */ - __be32 redirect_qkey; - union ib_gid trap_gid; - __be32 trap_tc_sl_fl; /* 8, 4, 20 bits respectively */ - __be16 trap_lid; - __be16 trap_pkey; - __be32 trap_hl_qp; /* 8, 24 bits respectively */ - __be32 trap_qkey; -} __attribute__ ((packed)); - -struct ib_pma_portsamplescontrol { - u8 opcode; - u8 port_select; - u8 tick; - u8 counter_width; /* only lower 3 bits */ - __be32 counter_mask0_9; /* 2, 10 * 3, bits */ - __be16 counter_mask10_14; /* 1, 5 * 3, bits */ - u8 sample_mechanisms; - u8 sample_status; /* only lower 2 bits */ - __be64 option_mask; - __be64 vendor_mask; - __be32 sample_start; - __be32 sample_interval; - __be16 tag; - __be16 counter_select[15]; -} __attribute__ ((packed)); - -struct ib_pma_portsamplesresult { - __be16 tag; - __be16 sample_status; /* only lower 2 bits */ - __be32 counter[15]; -} __attribute__ ((packed)); - -struct ib_pma_portsamplesresult_ext { - __be16 tag; - __be16 sample_status; /* only lower 2 bits */ - __be32 extended_width; /* only upper 2 bits */ - __be64 counter[15]; -} __attribute__ ((packed)); - -struct ib_pma_portcounters { - u8 reserved; - u8 port_select; - __be16 counter_select; - __be16 symbol_error_counter; - u8 link_error_recovery_counter; - u8 link_downed_counter; - __be16 port_rcv_errors; - __be16 port_rcv_remphys_errors; - __be16 port_rcv_switch_relay_errors; - __be16 port_xmit_discards; - u8 port_xmit_constraint_errors; - u8 port_rcv_constraint_errors; - u8 reserved1; - u8 lli_ebor_errors; /* 4, 4, bits */ - __be16 reserved2; - __be16 vl15_dropped; - __be32 port_xmit_data; - __be32 port_rcv_data; - __be32 port_xmit_packets; - __be32 port_rcv_packets; -} __attribute__ ((packed)); - struct ib_pma_portcounters_cong { u8 reserved; u8 reserved1; @@ -297,7 +198,7 @@ struct ib_pma_portcounters_cong { u8 port_xmit_constraint_errors; u8 port_rcv_constraint_errors; u8 reserved2; - u8 lli_ebor_errors; /* 4, 4, bits */ + u8 link_overrun_errors; /* LocalLink: 7:4, BufferOverrun: 3:0 */ __be16 reserved3; __be16 vl15_dropped; __be64 port_xmit_data; @@ -316,49 +217,11 @@ struct ib_pma_portcounters_cong { /* number of 4nsec cycles equaling 2secs */ #define QIB_CONG_TIMER_PSINTERVAL 0x1DCD64EC -#define IB_PMA_SEL_SYMBOL_ERROR cpu_to_be16(0x0001) -#define IB_PMA_SEL_LINK_ERROR_RECOVERY cpu_to_be16(0x0002) -#define IB_PMA_SEL_LINK_DOWNED cpu_to_be16(0x0004) -#define IB_PMA_SEL_PORT_RCV_ERRORS cpu_to_be16(0x0008) -#define IB_PMA_SEL_PORT_RCV_REMPHYS_ERRORS cpu_to_be16(0x0010) -#define IB_PMA_SEL_PORT_XMIT_DISCARDS cpu_to_be16(0x0040) -#define IB_PMA_SEL_LOCAL_LINK_INTEGRITY_ERRORS cpu_to_be16(0x0200) -#define IB_PMA_SEL_EXCESSIVE_BUFFER_OVERRUNS cpu_to_be16(0x0400) -#define IB_PMA_SEL_PORT_VL15_DROPPED cpu_to_be16(0x0800) -#define IB_PMA_SEL_PORT_XMIT_DATA cpu_to_be16(0x1000) -#define IB_PMA_SEL_PORT_RCV_DATA cpu_to_be16(0x2000) -#define IB_PMA_SEL_PORT_XMIT_PACKETS cpu_to_be16(0x4000) -#define IB_PMA_SEL_PORT_RCV_PACKETS cpu_to_be16(0x8000) - #define IB_PMA_SEL_CONG_ALL 0x01 #define IB_PMA_SEL_CONG_PORT_DATA 0x02 #define IB_PMA_SEL_CONG_XMIT 0x04 #define IB_PMA_SEL_CONG_ROUTING 0x08 -struct ib_pma_portcounters_ext { - u8 reserved; - u8 port_select; - __be16 counter_select; - __be32 reserved1; - __be64 port_xmit_data; - __be64 port_rcv_data; - __be64 port_xmit_packets; - __be64 port_rcv_packets; - __be64 port_unicast_xmit_packets; - __be64 port_unicast_rcv_packets; - __be64 port_multicast_xmit_packets; - __be64 port_multicast_rcv_packets; -} __attribute__ ((packed)); - -#define IB_PMA_SELX_PORT_XMIT_DATA cpu_to_be16(0x0001) -#define IB_PMA_SELX_PORT_RCV_DATA cpu_to_be16(0x0002) -#define IB_PMA_SELX_PORT_XMIT_PACKETS cpu_to_be16(0x0004) -#define IB_PMA_SELX_PORT_RCV_PACKETS cpu_to_be16(0x0008) -#define IB_PMA_SELX_PORT_UNI_XMIT_PACKETS cpu_to_be16(0x0010) -#define IB_PMA_SELX_PORT_UNI_RCV_PACKETS cpu_to_be16(0x0020) -#define IB_PMA_SELX_PORT_MULTI_XMIT_PACKETS cpu_to_be16(0x0040) -#define IB_PMA_SELX_PORT_MULTI_RCV_PACKETS cpu_to_be16(0x0080) - /* * The PortSamplesControl.CounterMasks field is an array of 3 bit fields * which specify the N'th counter's capabilities. See ch. 16.1.3.2. diff --git a/drivers/infiniband/hw/qib/qib_pcie.c b/drivers/infiniband/hw/qib/qib_pcie.c index 891cc2f..97a8bdf 100644 --- a/drivers/infiniband/hw/qib/qib_pcie.c +++ b/drivers/infiniband/hw/qib/qib_pcie.c @@ -35,6 +35,7 @@ #include <linux/delay.h> #include <linux/vmalloc.h> #include <linux/aer.h> +#include <linux/module.h> #include "qib.h" @@ -255,7 +256,7 @@ int qib_pcie_params(struct qib_devdata *dd, u32 minw, u32 *nent, u16 linkstat, speed; int pos = 0, pose, ret = 1; - pose = pci_find_capability(dd->pcidev, PCI_CAP_ID_EXP); + pose = pci_pcie_cap(dd->pcidev); if (!pose) { qib_dev_err(dd, "Can't find PCI Express capability!\n"); /* set up something... */ @@ -509,7 +510,7 @@ static int qib_tune_pcie_coalesce(struct qib_devdata *dd) qib_devinfo(dd->pcidev, "Parent not root\n"); return 1; } - ppos = pci_find_capability(parent, PCI_CAP_ID_EXP); + ppos = pci_pcie_cap(parent); if (!ppos) return 1; if (parent->vendor != 0x8086) @@ -578,14 +579,14 @@ static int qib_tune_pcie_caps(struct qib_devdata *dd) qib_devinfo(dd->pcidev, "Parent not root\n"); goto bail; } - ppos = pci_find_capability(parent, PCI_CAP_ID_EXP); + ppos = pci_pcie_cap(parent); if (ppos) { pci_read_config_word(parent, ppos + PCI_EXP_DEVCAP, &pcaps); pci_read_config_word(parent, ppos + PCI_EXP_DEVCTL, &pctl); } else goto bail; /* Find out supported and configured values for endpoint (us) */ - epos = pci_find_capability(dd->pcidev, PCI_CAP_ID_EXP); + epos = pci_pcie_cap(dd->pcidev); if (epos) { pci_read_config_word(dd->pcidev, epos + PCI_EXP_DEVCAP, &ecaps); pci_read_config_word(dd->pcidev, epos + PCI_EXP_DEVCTL, &ectl); diff --git a/drivers/infiniband/hw/qib/qib_qp.c b/drivers/infiniband/hw/qib/qib_qp.c index e16751f..7e7e16f 100644 --- a/drivers/infiniband/hw/qib/qib_qp.c +++ b/drivers/infiniband/hw/qib/qib_qp.c @@ -34,6 +34,7 @@ #include <linux/err.h> #include <linux/vmalloc.h> +#include <linux/jhash.h> #include "qib.h" @@ -204,6 +205,13 @@ static void free_qpn(struct qib_qpn_table *qpt, u32 qpn) clear_bit(qpn & BITS_PER_PAGE_MASK, map->page); } +static inline unsigned qpn_hash(struct qib_ibdev *dev, u32 qpn) +{ + return jhash_1word(qpn, dev->qp_rnd) & + (dev->qp_table_size - 1); +} + + /* * Put the QP into the hash table. * The hash table holds a reference to the QP. @@ -211,22 +219,23 @@ static void free_qpn(struct qib_qpn_table *qpt, u32 qpn) static void insert_qp(struct qib_ibdev *dev, struct qib_qp *qp) { struct qib_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num); - unsigned n = qp->ibqp.qp_num % dev->qp_table_size; unsigned long flags; + unsigned n = qpn_hash(dev, qp->ibqp.qp_num); spin_lock_irqsave(&dev->qpt_lock, flags); + atomic_inc(&qp->refcount); if (qp->ibqp.qp_num == 0) - ibp->qp0 = qp; + rcu_assign_pointer(ibp->qp0, qp); else if (qp->ibqp.qp_num == 1) - ibp->qp1 = qp; + rcu_assign_pointer(ibp->qp1, qp); else { qp->next = dev->qp_table[n]; - dev->qp_table[n] = qp; + rcu_assign_pointer(dev->qp_table[n], qp); } - atomic_inc(&qp->refcount); spin_unlock_irqrestore(&dev->qpt_lock, flags); + synchronize_rcu(); } /* @@ -236,29 +245,32 @@ static void insert_qp(struct qib_ibdev *dev, struct qib_qp *qp) static void remove_qp(struct qib_ibdev *dev, struct qib_qp *qp) { struct qib_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num); - struct qib_qp *q, **qpp; + unsigned n = qpn_hash(dev, qp->ibqp.qp_num); unsigned long flags; - qpp = &dev->qp_table[qp->ibqp.qp_num % dev->qp_table_size]; - spin_lock_irqsave(&dev->qpt_lock, flags); if (ibp->qp0 == qp) { - ibp->qp0 = NULL; atomic_dec(&qp->refcount); + rcu_assign_pointer(ibp->qp0, NULL); } else if (ibp->qp1 == qp) { - ibp->qp1 = NULL; atomic_dec(&qp->refcount); - } else + rcu_assign_pointer(ibp->qp1, NULL); + } else { + struct qib_qp *q, **qpp; + + qpp = &dev->qp_table[n]; for (; (q = *qpp) != NULL; qpp = &q->next) if (q == qp) { - *qpp = qp->next; - qp->next = NULL; atomic_dec(&qp->refcount); + rcu_assign_pointer(*qpp, qp->next); + qp->next = NULL; break; } + } spin_unlock_irqrestore(&dev->qpt_lock, flags); + synchronize_rcu(); } /** @@ -280,21 +292,24 @@ unsigned qib_free_all_qps(struct qib_devdata *dd) if (!qib_mcast_tree_empty(ibp)) qp_inuse++; - if (ibp->qp0) + rcu_read_lock(); + if (rcu_dereference(ibp->qp0)) qp_inuse++; - if (ibp->qp1) + if (rcu_dereference(ibp->qp1)) qp_inuse++; + rcu_read_unlock(); } spin_lock_irqsave(&dev->qpt_lock, flags); for (n = 0; n < dev->qp_table_size; n++) { qp = dev->qp_table[n]; - dev->qp_table[n] = NULL; + rcu_assign_pointer(dev->qp_table[n], NULL); for (; qp; qp = qp->next) qp_inuse++; } spin_unlock_irqrestore(&dev->qpt_lock, flags); + synchronize_rcu(); return qp_inuse; } @@ -309,25 +324,28 @@ unsigned qib_free_all_qps(struct qib_devdata *dd) */ struct qib_qp *qib_lookup_qpn(struct qib_ibport *ibp, u32 qpn) { - struct qib_ibdev *dev = &ppd_from_ibp(ibp)->dd->verbs_dev; - unsigned long flags; - struct qib_qp *qp; + struct qib_qp *qp = NULL; - spin_lock_irqsave(&dev->qpt_lock, flags); + if (unlikely(qpn <= 1)) { + rcu_read_lock(); + if (qpn == 0) + qp = rcu_dereference(ibp->qp0); + else + qp = rcu_dereference(ibp->qp1); + } else { + struct qib_ibdev *dev = &ppd_from_ibp(ibp)->dd->verbs_dev; + unsigned n = qpn_hash(dev, qpn); - if (qpn == 0) - qp = ibp->qp0; - else if (qpn == 1) - qp = ibp->qp1; - else - for (qp = dev->qp_table[qpn % dev->qp_table_size]; qp; - qp = qp->next) + rcu_read_lock(); + for (qp = dev->qp_table[n]; rcu_dereference(qp); qp = qp->next) if (qp->ibqp.qp_num == qpn) break; + } if (qp) - atomic_inc(&qp->refcount); + if (unlikely(!atomic_inc_not_zero(&qp->refcount))) + qp = NULL; - spin_unlock_irqrestore(&dev->qpt_lock, flags); + rcu_read_unlock(); return qp; } @@ -765,8 +783,10 @@ int qib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, } } - if (attr_mask & IB_QP_PATH_MTU) + if (attr_mask & IB_QP_PATH_MTU) { qp->path_mtu = pmtu; + qp->pmtu = ib_mtu_enum_to_int(pmtu); + } if (attr_mask & IB_QP_RETRY_CNT) { qp->s_retry_cnt = attr->retry_cnt; @@ -781,8 +801,12 @@ int qib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, if (attr_mask & IB_QP_MIN_RNR_TIMER) qp->r_min_rnr_timer = attr->min_rnr_timer; - if (attr_mask & IB_QP_TIMEOUT) + if (attr_mask & IB_QP_TIMEOUT) { qp->timeout = attr->timeout; + qp->timeout_jiffies = + usecs_to_jiffies((4096UL * (1UL << qp->timeout)) / + 1000UL); + } if (attr_mask & IB_QP_QKEY) qp->qkey = attr->qkey; @@ -1013,6 +1037,10 @@ struct ib_qp *qib_create_qp(struct ib_pd *ibpd, ret = ERR_PTR(-ENOMEM); goto bail_swq; } + RCU_INIT_POINTER(qp->next, NULL); + qp->timeout_jiffies = + usecs_to_jiffies((4096UL * (1UL << qp->timeout)) / + 1000UL); if (init_attr->srq) sz = 0; else { diff --git a/drivers/infiniband/hw/qib/qib_qsfp.c b/drivers/infiniband/hw/qib/qib_qsfp.c index 3374a52..fa71b1e 100644 --- a/drivers/infiniband/hw/qib/qib_qsfp.c +++ b/drivers/infiniband/hw/qib/qib_qsfp.c @@ -273,18 +273,12 @@ int qib_refresh_qsfp_cache(struct qib_pportdata *ppd, struct qib_qsfp_cache *cp) int ret; int idx; u16 cks; - u32 mask; u8 peek[4]; /* ensure sane contents on invalid reads, for cable swaps */ memset(cp, 0, sizeof(*cp)); - mask = QSFP_GPIO_MOD_PRS_N; - if (ppd->hw_pidx) - mask <<= QSFP_GPIO_PORT2_SHIFT; - - ret = ppd->dd->f_gpio_mod(ppd->dd, 0, 0, 0); - if (ret & mask) { + if (!qib_qsfp_mod_present(ppd)) { ret = -ENODEV; goto bail; } @@ -444,6 +438,19 @@ const char * const qib_qsfp_devtech[16] = { static const char *pwr_codes = "1.5W2.0W2.5W3.5W"; +int qib_qsfp_mod_present(struct qib_pportdata *ppd) +{ + u32 mask; + int ret; + + mask = QSFP_GPIO_MOD_PRS_N << + (ppd->hw_pidx * QSFP_GPIO_PORT2_SHIFT); + ret = ppd->dd->f_gpio_mod(ppd->dd, 0, 0, 0); + + return !((ret & mask) >> + ((ppd->hw_pidx * QSFP_GPIO_PORT2_SHIFT) + 3)); +} + /* * Initialize structures that control access to QSFP. Called once per port * on cards that support QSFP. @@ -452,7 +459,6 @@ void qib_qsfp_init(struct qib_qsfp_data *qd, void (*fevent)(struct work_struct *)) { u32 mask, highs; - int pins; struct qib_devdata *dd = qd->ppd->dd; @@ -474,19 +480,6 @@ void qib_qsfp_init(struct qib_qsfp_data *qd, udelay(20); /* Generous RST dwell */ dd->f_gpio_mod(dd, mask, mask, mask); - /* Spec says module can take up to two seconds! */ - mask = QSFP_GPIO_MOD_PRS_N; - if (qd->ppd->hw_pidx) - mask <<= QSFP_GPIO_PORT2_SHIFT; - - /* Do not try to wait here. Better to let event handle it */ - pins = dd->f_gpio_mod(dd, 0, 0, 0); - if (pins & mask) - goto bail; - /* We see a module, but it may be unwise to look yet. Just schedule */ - qd->t_insert = get_jiffies_64(); - queue_work(ib_wq, &qd->work); -bail: return; } diff --git a/drivers/infiniband/hw/qib/qib_qsfp.h b/drivers/infiniband/hw/qib/qib_qsfp.h index c109bbd..46002a9 100644 --- a/drivers/infiniband/hw/qib/qib_qsfp.h +++ b/drivers/infiniband/hw/qib/qib_qsfp.h @@ -34,6 +34,7 @@ #define QSFP_DEV 0xA0 #define QSFP_PWR_LAG_MSEC 2000 +#define QSFP_MODPRS_LAG_MSEC 20 /* * Below are masks for various QSFP signals, for Port 1. @@ -177,10 +178,12 @@ struct qib_qsfp_data { struct work_struct work; struct qib_qsfp_cache cache; u64 t_insert; + u8 modpresent; }; extern int qib_refresh_qsfp_cache(struct qib_pportdata *ppd, struct qib_qsfp_cache *cp); +extern int qib_qsfp_mod_present(struct qib_pportdata *ppd); extern void qib_qsfp_init(struct qib_qsfp_data *qd, void (*fevent)(struct work_struct *)); extern void qib_qsfp_deinit(struct qib_qsfp_data *qd); diff --git a/drivers/infiniband/hw/qib/qib_rc.c b/drivers/infiniband/hw/qib/qib_rc.c index eca0c41..894afac 100644 --- a/drivers/infiniband/hw/qib/qib_rc.c +++ b/drivers/infiniband/hw/qib/qib_rc.c @@ -59,8 +59,7 @@ static void start_timer(struct qib_qp *qp) qp->s_flags |= QIB_S_TIMER; qp->s_timer.function = rc_timeout; /* 4.096 usec. * (1 << qp->timeout) */ - qp->s_timer.expires = jiffies + - usecs_to_jiffies((4096UL * (1UL << qp->timeout)) / 1000UL); + qp->s_timer.expires = jiffies + qp->timeout_jiffies; add_timer(&qp->s_timer); } @@ -239,7 +238,7 @@ int qib_make_rc_req(struct qib_qp *qp) u32 len; u32 bth0; u32 bth2; - u32 pmtu = ib_mtu_enum_to_int(qp->path_mtu); + u32 pmtu = qp->pmtu; char newreq; unsigned long flags; int ret = 0; @@ -272,13 +271,9 @@ int qib_make_rc_req(struct qib_qp *qp) goto bail; } wqe = get_swqe_ptr(qp, qp->s_last); - while (qp->s_last != qp->s_acked) { - qib_send_complete(qp, wqe, IB_WC_SUCCESS); - if (++qp->s_last >= qp->s_size) - qp->s_last = 0; - wqe = get_swqe_ptr(qp, qp->s_last); - } - qib_send_complete(qp, wqe, IB_WC_WR_FLUSH_ERR); + qib_send_complete(qp, wqe, qp->s_last != qp->s_acked ? + IB_WC_SUCCESS : IB_WC_WR_FLUSH_ERR); + /* will get called again */ goto done; } @@ -1519,9 +1514,7 @@ read_middle: * 4.096 usec. * (1 << qp->timeout) */ qp->s_flags |= QIB_S_TIMER; - mod_timer(&qp->s_timer, jiffies + - usecs_to_jiffies((4096UL * (1UL << qp->timeout)) / - 1000UL)); + mod_timer(&qp->s_timer, jiffies + qp->timeout_jiffies); if (qp->s_flags & QIB_S_WAIT_ACK) { qp->s_flags &= ~QIB_S_WAIT_ACK; qib_schedule_send(qp); @@ -1732,7 +1725,7 @@ static int qib_rc_rcv_error(struct qib_other_headers *ohdr, * same request. */ offset = ((psn - e->psn) & QIB_PSN_MASK) * - ib_mtu_enum_to_int(qp->path_mtu); + qp->pmtu; len = be32_to_cpu(reth->length); if (unlikely(offset + len != e->rdma_sge.sge_length)) goto unlock_done; @@ -1876,7 +1869,7 @@ void qib_rc_rcv(struct qib_ctxtdata *rcd, struct qib_ib_header *hdr, u32 psn; u32 pad; struct ib_wc wc; - u32 pmtu = ib_mtu_enum_to_int(qp->path_mtu); + u32 pmtu = qp->pmtu; int diff; struct ib_reth *reth; unsigned long flags; @@ -1892,10 +1885,8 @@ void qib_rc_rcv(struct qib_ctxtdata *rcd, struct qib_ib_header *hdr, } opcode = be32_to_cpu(ohdr->bth[0]); - spin_lock_irqsave(&qp->s_lock, flags); if (qib_ruc_check_hdr(ibp, hdr, has_grh, qp, opcode)) - goto sunlock; - spin_unlock_irqrestore(&qp->s_lock, flags); + return; psn = be32_to_cpu(ohdr->bth[2]); opcode >>= 24; @@ -1955,8 +1946,6 @@ void qib_rc_rcv(struct qib_ctxtdata *rcd, struct qib_ib_header *hdr, break; } - memset(&wc, 0, sizeof wc); - if (qp->state == IB_QPS_RTR && !(qp->r_flags & QIB_R_COMM_EST)) { qp->r_flags |= QIB_R_COMM_EST; if (qp->ibqp.event_handler) { @@ -2009,16 +1998,19 @@ send_middle: goto rnr_nak; qp->r_rcv_len = 0; if (opcode == OP(SEND_ONLY)) - goto send_last; - /* FALLTHROUGH */ + goto no_immediate_data; + /* FALLTHROUGH for SEND_ONLY_WITH_IMMEDIATE */ case OP(SEND_LAST_WITH_IMMEDIATE): send_last_imm: wc.ex.imm_data = ohdr->u.imm_data; hdrsize += 4; wc.wc_flags = IB_WC_WITH_IMM; - /* FALLTHROUGH */ + goto send_last; case OP(SEND_LAST): case OP(RDMA_WRITE_LAST): +no_immediate_data: + wc.wc_flags = 0; + wc.ex.imm_data = 0; send_last: /* Get the number of bytes the message was padded by. */ pad = (be32_to_cpu(ohdr->bth[0]) >> 20) & 3; @@ -2051,6 +2043,12 @@ send_last: wc.src_qp = qp->remote_qpn; wc.slid = qp->remote_ah_attr.dlid; wc.sl = qp->remote_ah_attr.sl; + /* zero fields that are N/A */ + wc.vendor_err = 0; + wc.pkey_index = 0; + wc.dlid_path_bits = 0; + wc.port_num = 0; + wc.csum_ok = 0; /* Signal completion event if the solicited bit is set. */ qib_cq_enter(to_icq(qp->ibqp.recv_cq), &wc, (ohdr->bth[0] & @@ -2089,7 +2087,7 @@ send_last: if (opcode == OP(RDMA_WRITE_FIRST)) goto send_middle; else if (opcode == OP(RDMA_WRITE_ONLY)) - goto send_last; + goto no_immediate_data; ret = qib_get_rwqe(qp, 1); if (ret < 0) goto nack_op_err; diff --git a/drivers/infiniband/hw/qib/qib_ruc.c b/drivers/infiniband/hw/qib/qib_ruc.c index eb78d93..b4b37e4 100644 --- a/drivers/infiniband/hw/qib/qib_ruc.c +++ b/drivers/infiniband/hw/qib/qib_ruc.c @@ -260,12 +260,15 @@ static int gid_ok(union ib_gid *gid, __be64 gid_prefix, __be64 id) /* * - * This should be called with the QP s_lock held. + * This should be called with the QP r_lock held. + * + * The s_lock will be acquired around the qib_migrate_qp() call. */ int qib_ruc_check_hdr(struct qib_ibport *ibp, struct qib_ib_header *hdr, int has_grh, struct qib_qp *qp, u32 bth0) { __be64 guid; + unsigned long flags; if (qp->s_mig_state == IB_MIG_ARMED && (bth0 & IB_BTH_MIG_REQ)) { if (!has_grh) { @@ -295,7 +298,9 @@ int qib_ruc_check_hdr(struct qib_ibport *ibp, struct qib_ib_header *hdr, if (be16_to_cpu(hdr->lrh[3]) != qp->alt_ah_attr.dlid || ppd_from_ibp(ibp)->port != qp->alt_ah_attr.port_num) goto err; + spin_lock_irqsave(&qp->s_lock, flags); qib_migrate_qp(qp); + spin_unlock_irqrestore(&qp->s_lock, flags); } else { if (!has_grh) { if (qp->remote_ah_attr.ah_flags & IB_AH_GRH) diff --git a/drivers/infiniband/hw/qib/qib_sd7220.c b/drivers/infiniband/hw/qib/qib_sd7220.c index e9f9f8b..de1a4b2 100644 --- a/drivers/infiniband/hw/qib/qib_sd7220.c +++ b/drivers/infiniband/hw/qib/qib_sd7220.c @@ -38,6 +38,7 @@ #include <linux/pci.h> #include <linux/delay.h> +#include <linux/module.h> #include <linux/firmware.h> #include "qib.h" diff --git a/drivers/infiniband/hw/qib/qib_sdma.c b/drivers/infiniband/hw/qib/qib_sdma.c index cad4449..12a9604 100644 --- a/drivers/infiniband/hw/qib/qib_sdma.c +++ b/drivers/infiniband/hw/qib/qib_sdma.c @@ -32,6 +32,7 @@ #include <linux/spinlock.h> #include <linux/netdevice.h> +#include <linux/moduleparam.h> #include "qib.h" #include "qib_common.h" diff --git a/drivers/infiniband/hw/qib/qib_srq.c b/drivers/infiniband/hw/qib/qib_srq.c index c3ec8ef..d623593 100644 --- a/drivers/infiniband/hw/qib/qib_srq.c +++ b/drivers/infiniband/hw/qib/qib_srq.c @@ -107,6 +107,11 @@ struct ib_srq *qib_create_srq(struct ib_pd *ibpd, u32 sz; struct ib_srq *ret; + if (srq_init_attr->srq_type != IB_SRQT_BASIC) { + ret = ERR_PTR(-ENOSYS); + goto done; + } + if (srq_init_attr->attr.max_sge == 0 || srq_init_attr->attr.max_sge > ib_qib_max_srq_sges || srq_init_attr->attr.max_wr == 0 || diff --git a/drivers/infiniband/hw/qib/qib_sysfs.c b/drivers/infiniband/hw/qib/qib_sysfs.c index d50a33f..c97224a 100644 --- a/drivers/infiniband/hw/qib/qib_sysfs.c +++ b/drivers/infiniband/hw/qib/qib_sysfs.c @@ -507,6 +507,17 @@ static ssize_t show_nctxts(struct device *device, dd->first_user_ctxt); } +static ssize_t show_nfreectxts(struct device *device, + struct device_attribute *attr, char *buf) +{ + struct qib_ibdev *dev = + container_of(device, struct qib_ibdev, ibdev.dev); + struct qib_devdata *dd = dd_from_dev(dev); + + /* Return the number of free user ports (contexts) available. */ + return scnprintf(buf, PAGE_SIZE, "%u\n", dd->freectxts); +} + static ssize_t show_serial(struct device *device, struct device_attribute *attr, char *buf) { @@ -539,28 +550,6 @@ bail: return ret < 0 ? ret : count; } -static ssize_t show_logged_errs(struct device *device, - struct device_attribute *attr, char *buf) -{ - struct qib_ibdev *dev = - container_of(device, struct qib_ibdev, ibdev.dev); - struct qib_devdata *dd = dd_from_dev(dev); - int idx, count; - - /* force consistency with actual EEPROM */ - if (qib_update_eeprom_log(dd) != 0) - return -ENXIO; - - count = 0; - for (idx = 0; idx < QIB_EEP_LOG_CNT; ++idx) { - count += scnprintf(buf + count, PAGE_SIZE - count, "%d%c", - dd->eep_st_errs[idx], - idx == (QIB_EEP_LOG_CNT - 1) ? '\n' : ' '); - } - - return count; -} - /* * Dump tempsense regs. in decimal, to ease shell-scripts. */ @@ -604,9 +593,9 @@ static DEVICE_ATTR(hca_type, S_IRUGO, show_hca, NULL); static DEVICE_ATTR(board_id, S_IRUGO, show_hca, NULL); static DEVICE_ATTR(version, S_IRUGO, show_version, NULL); static DEVICE_ATTR(nctxts, S_IRUGO, show_nctxts, NULL); +static DEVICE_ATTR(nfreectxts, S_IRUGO, show_nfreectxts, NULL); static DEVICE_ATTR(serial, S_IRUGO, show_serial, NULL); static DEVICE_ATTR(boardversion, S_IRUGO, show_boardversion, NULL); -static DEVICE_ATTR(logged_errors, S_IRUGO, show_logged_errs, NULL); static DEVICE_ATTR(tempsense, S_IRUGO, show_tempsense, NULL); static DEVICE_ATTR(localbus_info, S_IRUGO, show_localbus_info, NULL); static DEVICE_ATTR(chip_reset, S_IWUSR, NULL, store_chip_reset); @@ -617,9 +606,9 @@ static struct device_attribute *qib_attributes[] = { &dev_attr_board_id, &dev_attr_version, &dev_attr_nctxts, + &dev_attr_nfreectxts, &dev_attr_serial, &dev_attr_boardversion, - &dev_attr_logged_errors, &dev_attr_tempsense, &dev_attr_localbus_info, &dev_attr_chip_reset, diff --git a/drivers/infiniband/hw/qib/qib_tx.c b/drivers/infiniband/hw/qib/qib_tx.c index 7f36454..1bf626c 100644 --- a/drivers/infiniband/hw/qib/qib_tx.c +++ b/drivers/infiniband/hw/qib/qib_tx.c @@ -36,6 +36,7 @@ #include <linux/delay.h> #include <linux/netdevice.h> #include <linux/vmalloc.h> +#include <linux/moduleparam.h> #include "qib.h" diff --git a/drivers/infiniband/hw/qib/qib_uc.c b/drivers/infiniband/hw/qib/qib_uc.c index 32ccf3c..847e7af 100644 --- a/drivers/infiniband/hw/qib/qib_uc.c +++ b/drivers/infiniband/hw/qib/qib_uc.c @@ -51,7 +51,7 @@ int qib_make_uc_req(struct qib_qp *qp) u32 hwords; u32 bth0; u32 len; - u32 pmtu = ib_mtu_enum_to_int(qp->path_mtu); + u32 pmtu = qp->pmtu; int ret = 0; spin_lock_irqsave(&qp->s_lock, flags); @@ -243,13 +243,12 @@ void qib_uc_rcv(struct qib_ibport *ibp, struct qib_ib_header *hdr, int has_grh, void *data, u32 tlen, struct qib_qp *qp) { struct qib_other_headers *ohdr; - unsigned long flags; u32 opcode; u32 hdrsize; u32 psn; u32 pad; struct ib_wc wc; - u32 pmtu = ib_mtu_enum_to_int(qp->path_mtu); + u32 pmtu = qp->pmtu; struct ib_reth *reth; int ret; @@ -263,14 +262,11 @@ void qib_uc_rcv(struct qib_ibport *ibp, struct qib_ib_header *hdr, } opcode = be32_to_cpu(ohdr->bth[0]); - spin_lock_irqsave(&qp->s_lock, flags); if (qib_ruc_check_hdr(ibp, hdr, has_grh, qp, opcode)) - goto sunlock; - spin_unlock_irqrestore(&qp->s_lock, flags); + return; psn = be32_to_cpu(ohdr->bth[2]); opcode >>= 24; - memset(&wc, 0, sizeof wc); /* Compare the PSN verses the expected PSN. */ if (unlikely(qib_cmp24(psn, qp->r_psn) != 0)) { @@ -370,7 +366,7 @@ send_first: } qp->r_rcv_len = 0; if (opcode == OP(SEND_ONLY)) - goto send_last; + goto no_immediate_data; else if (opcode == OP(SEND_ONLY_WITH_IMMEDIATE)) goto send_last_imm; /* FALLTHROUGH */ @@ -389,8 +385,11 @@ send_last_imm: wc.ex.imm_data = ohdr->u.imm_data; hdrsize += 4; wc.wc_flags = IB_WC_WITH_IMM; - /* FALLTHROUGH */ + goto send_last; case OP(SEND_LAST): +no_immediate_data: + wc.ex.imm_data = 0; + wc.wc_flags = 0; send_last: /* Get the number of bytes the message was padded by. */ pad = (be32_to_cpu(ohdr->bth[0]) >> 20) & 3; @@ -418,6 +417,12 @@ last_imm: wc.src_qp = qp->remote_qpn; wc.slid = qp->remote_ah_attr.dlid; wc.sl = qp->remote_ah_attr.sl; + /* zero fields that are N/A */ + wc.vendor_err = 0; + wc.pkey_index = 0; + wc.dlid_path_bits = 0; + wc.port_num = 0; + wc.csum_ok = 0; /* Signal completion event if the solicited bit is set. */ qib_cq_enter(to_icq(qp->ibqp.recv_cq), &wc, (ohdr->bth[0] & @@ -546,6 +551,4 @@ op_err: qib_rc_error(qp, IB_WC_LOC_QP_OP_ERR); return; -sunlock: - spin_unlock_irqrestore(&qp->s_lock, flags); } diff --git a/drivers/infiniband/hw/qib/qib_ud.c b/drivers/infiniband/hw/qib/qib_ud.c index 828609f..c0d93d4 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; diff --git a/drivers/infiniband/hw/qib/qib_user_pages.c b/drivers/infiniband/hw/qib/qib_user_pages.c index 7689e49..2bc1d2b 100644 --- a/drivers/infiniband/hw/qib/qib_user_pages.c +++ b/drivers/infiniband/hw/qib/qib_user_pages.c @@ -74,7 +74,7 @@ static int __qib_get_user_pages(unsigned long start_page, size_t num_pages, goto bail_release; } - current->mm->locked_vm += num_pages; + current->mm->pinned_vm += num_pages; ret = 0; goto bail; @@ -151,7 +151,7 @@ void qib_release_user_pages(struct page **p, size_t num_pages) __qib_release_user_pages(p, num_pages, 1); if (current->mm) { - current->mm->locked_vm -= num_pages; + current->mm->pinned_vm -= num_pages; up_write(¤t->mm->mmap_sem); } } diff --git a/drivers/infiniband/hw/qib/qib_user_sdma.c b/drivers/infiniband/hw/qib/qib_user_sdma.c index 8244208..573b460 100644 --- a/drivers/infiniband/hw/qib/qib_user_sdma.c +++ b/drivers/infiniband/hw/qib/qib_user_sdma.c @@ -284,8 +284,7 @@ static int qib_user_sdma_pin_pages(const struct qib_devdata *dd, int j; int ret; - ret = get_user_pages(current, current->mm, addr, - npages, 0, 1, pages, NULL); + ret = get_user_pages_fast(addr, npages, 0, pages); if (ret != npages) { int i; @@ -830,10 +829,7 @@ int qib_user_sdma_writev(struct qib_ctxtdata *rcd, while (dim) { const int mxp = 8; - down_write(¤t->mm->mmap_sem); ret = qib_user_sdma_queue_pkts(dd, pq, &list, iov, dim, mxp); - up_write(¤t->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 9fab404..c51a6f9 100644 --- a/drivers/infiniband/hw/qib/qib_verbs.c +++ b/drivers/infiniband/hw/qib/qib_verbs.c @@ -35,14 +35,17 @@ #include <rdma/ib_mad.h> #include <rdma/ib_user_verbs.h> #include <linux/io.h> +#include <linux/module.h> #include <linux/utsname.h> #include <linux/rculist.h> #include <linux/mm.h> +#include <linux/random.h> +#include <linux/vmalloc.h> #include "qib.h" #include "qib_common.h" -static unsigned int ib_qib_qp_table_size = 251; +static unsigned int ib_qib_qp_table_size = 256; module_param_named(qp_table_size, ib_qib_qp_table_size, uint, S_IRUGO); MODULE_PARM_DESC(qp_table_size, "QP table size"); @@ -659,17 +662,25 @@ void qib_ib_rcv(struct qib_ctxtdata *rcd, void *rhdr, void *data, u32 tlen) if (atomic_dec_return(&mcast->refcount) <= 1) wake_up(&mcast->wait); } else { - qp = qib_lookup_qpn(ibp, qp_num); - if (!qp) - goto drop; + if (rcd->lookaside_qp) { + if (rcd->lookaside_qpn != qp_num) { + if (atomic_dec_and_test( + &rcd->lookaside_qp->refcount)) + wake_up( + &rcd->lookaside_qp->wait); + rcd->lookaside_qp = NULL; + } + } + if (!rcd->lookaside_qp) { + qp = qib_lookup_qpn(ibp, qp_num); + if (!qp) + goto drop; + rcd->lookaside_qp = qp; + rcd->lookaside_qpn = qp_num; + } else + qp = rcd->lookaside_qp; ibp->n_unicast_rcv++; qib_qp_rcv(rcd, hdr, lnh == QIB_LRH_GRH, data, tlen, qp); - /* - * Notify qib_destroy_qp() if it is waiting - * for us to finish. - */ - if (atomic_dec_and_test(&qp->refcount)) - wake_up(&qp->wait); } return; @@ -1974,6 +1985,8 @@ static void init_ibport(struct qib_pportdata *ppd) ibp->z_excessive_buffer_overrun_errors = cntrs.excessive_buffer_overrun_errors; ibp->z_vl15_dropped = cntrs.vl15_dropped; + RCU_INIT_POINTER(ibp->qp0, NULL); + RCU_INIT_POINTER(ibp->qp1, NULL); } /** @@ -1990,12 +2003,15 @@ int qib_register_ib_device(struct qib_devdata *dd) int ret; dev->qp_table_size = ib_qib_qp_table_size; - dev->qp_table = kzalloc(dev->qp_table_size * sizeof *dev->qp_table, + get_random_bytes(&dev->qp_rnd, sizeof(dev->qp_rnd)); + dev->qp_table = kmalloc(dev->qp_table_size * sizeof *dev->qp_table, GFP_KERNEL); if (!dev->qp_table) { ret = -ENOMEM; goto err_qpt; } + for (i = 0; i < dev->qp_table_size; i++) + RCU_INIT_POINTER(dev->qp_table[i], NULL); for (i = 0; i < dd->num_pports; i++) init_ibport(ppd + i); @@ -2020,10 +2036,16 @@ int qib_register_ib_device(struct qib_devdata *dd) * the LKEY). The remaining bits act as a generation number or tag. */ spin_lock_init(&dev->lk_table.lock); + /* insure generation is at least 4 bits see keys.c */ + if (ib_qib_lkey_table_size > MAX_LKEY_TABLE_BITS) { + qib_dev_warn(dd, "lkey bits %u too large, reduced to %u\n", + ib_qib_lkey_table_size, MAX_LKEY_TABLE_BITS); + ib_qib_lkey_table_size = MAX_LKEY_TABLE_BITS; + } dev->lk_table.max = 1 << ib_qib_lkey_table_size; lk_tab_size = dev->lk_table.max * sizeof(*dev->lk_table.table); dev->lk_table.table = (struct qib_mregion **) - __get_free_pages(GFP_KERNEL, get_order(lk_tab_size)); + vmalloc(lk_tab_size); if (dev->lk_table.table == NULL) { ret = -ENOMEM; goto err_lk; @@ -2193,7 +2215,7 @@ err_tx: sizeof(struct qib_pio_header), dev->pio_hdrs, dev->pio_hdrs_phys); err_hdrs: - free_pages((unsigned long) dev->lk_table.table, get_order(lk_tab_size)); + vfree(dev->lk_table.table); err_lk: kfree(dev->qp_table); err_qpt: @@ -2247,7 +2269,6 @@ void qib_unregister_ib_device(struct qib_devdata *dd) sizeof(struct qib_pio_header), dev->pio_hdrs, dev->pio_hdrs_phys); lk_tab_size = dev->lk_table.max * sizeof(*dev->lk_table.table); - free_pages((unsigned long) dev->lk_table.table, - get_order(lk_tab_size)); + vfree(dev->lk_table.table); kfree(dev->qp_table); } diff --git a/drivers/infiniband/hw/qib/qib_verbs.h b/drivers/infiniband/hw/qib/qib_verbs.h index 95e5b47..66f7f62 100644 --- a/drivers/infiniband/hw/qib/qib_verbs.h +++ b/drivers/infiniband/hw/qib/qib_verbs.h @@ -485,6 +485,7 @@ struct qib_qp { u8 alt_timeout; /* Alternate path timeout for this QP */ u8 port_num; enum ib_mtu path_mtu; + u32 pmtu; /* decoded from path_mtu */ u32 remote_qpn; u32 qkey; /* QKEY for this QP (for UD or RD) */ u32 s_size; /* send work queue size */ @@ -495,6 +496,7 @@ struct qib_qp { u32 s_last; /* last completed entry */ u32 s_ssn; /* SSN of tail entry */ u32 s_lsn; /* limit sequence number (credit) */ + unsigned long timeout_jiffies; /* computed from timeout */ struct qib_swqe *s_wq; /* send work queue */ struct qib_swqe *s_wqe; struct qib_rq r_rq; /* receive work queue */ @@ -620,6 +622,8 @@ struct qib_qpn_table { struct qpn_map map[QPNMAP_ENTRIES]; }; +#define MAX_LKEY_TABLE_BITS 23 + struct qib_lkey_table { spinlock_t lock; /* protect changes in this struct */ u32 next; /* next unused index (speeds search) */ @@ -723,7 +727,8 @@ struct qib_ibdev { dma_addr_t pio_hdrs_phys; /* list of QPs waiting for RNR timer */ spinlock_t pending_lock; /* protect wait lists, PMA counters, etc. */ - unsigned qp_table_size; /* size of the hash table */ + u32 qp_table_size; /* size of the hash table */ + u32 qp_rnd; /* random bytes for hash */ spinlock_t qpt_lock; u32 n_piowait; |