From mboxrd@z Thu Jan 1 00:00:00 1970 From: Dennis Dalessandro Subject: [PATCH 09/54] staging/rdma/hfi1: Fix per-VL transmit discard counts Date: Wed, 03 Feb 2016 14:31:49 -0800 Message-ID: <20160203223145.5923.63961.stgit@scvm10.sc.intel.com> References: <20160203222512.5923.30980.stgit@scvm10.sc.intel.com> Mime-Version: 1.0 Content-Type: text/plain; charset="utf-8" Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <20160203222512.5923.30980.stgit-9QXIwq+3FY+1XWohqUldA0EOCMrvLtNR@public.gmane.org> Sender: linux-rdma-owner-u79uwXL29TY76Z2rM5mHXA@public.gmane.org To: dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA@public.gmane.org, Mike Marciniszyn , Dean Luick List-Id: linux-rdma@vger.kernel.org From: Mike Marciniszyn Implement per-VL transmit counters. Not all errors can be attributed to a particular VL, so make a best attempt. o Extend the egress error bits used to count toward transmit discard. o When an egress error or send error occur, try to map back to a VL. o Implement a SDMA engine to VL (back) map. o Add per-VL port transmit counters Reviewed-by: Mike Marciniszyn Reviewed-by: Dennis Dalessandro Signed-off-by: Dean Luick --- drivers/staging/rdma/hfi1/chip.c | 138 +++++++++++++++++++++++++++++++------- drivers/staging/rdma/hfi1/hfi.h | 3 + drivers/staging/rdma/hfi1/sdma.c | 6 ++ drivers/staging/rdma/hfi1/sdma.h | 2 + 4 files changed, 123 insertions(+), 26 deletions(-) diff --git a/drivers/staging/rdma/hfi1/chip.c b/drivers/staging/rdma/hfi1/chip.c index 41af05e..108cd48 100644 --- a/drivers/staging/rdma/hfi1/chip.c +++ b/drivers/staging/rdma/hfi1/chip.c @@ -510,6 +510,12 @@ static struct flag_table sdma_err_status_flags[] = { | SEND_DMA_ERR_STATUS_SDMA_CSR_PARITY_ERR_SMASK \ | SEND_DMA_ERR_STATUS_SDMA_PCIE_REQ_TRACKING_UNC_ERR_SMASK) +/* SendEgressErrInfo bits that correspond to a PortXmitDiscard counter */ +#define PORT_DISCARD_EGRESS_ERRS \ + (SEND_EGRESS_ERR_INFO_TOO_LONG_IB_PACKET_ERR_SMASK \ + | SEND_EGRESS_ERR_INFO_VL_MAPPING_ERR_SMASK \ + | SEND_EGRESS_ERR_INFO_VL_ERR_SMASK) + /* * TXE Egress Error flags */ @@ -1481,12 +1487,18 @@ static u64 access_sw_unknown_frame_cnt(const struct cntr_entry *entry, static u64 access_sw_xmit_discards(const struct cntr_entry *entry, void *context, int vl, int mode, u64 data) { - struct hfi1_pportdata *ppd = context; + struct hfi1_pportdata *ppd = (struct hfi1_pportdata *)context; + u64 zero = 0; + u64 *counter; - if (vl != CNTR_INVALID_VL) - return 0; + if (vl == CNTR_INVALID_VL) + counter = &ppd->port_xmit_discards; + else if (vl >= 0 && vl < C_VL_COUNT) + counter = &ppd->port_xmit_discards_vl[vl]; + else + counter = &zero; - return read_write_sw(ppd->dd, &ppd->port_xmit_discards, mode, data); + return read_write_sw(ppd->dd, counter, mode, data); } static u64 access_xmit_constraint_errs(const struct cntr_entry *entry, @@ -5508,12 +5520,14 @@ static void handle_sdma_err(struct hfi1_devdata *dd, u32 unused, u64 reg) } } -static void count_port_inactive(struct hfi1_devdata *dd) +static inline void __count_port_discards(struct hfi1_pportdata *ppd) { - struct hfi1_pportdata *ppd = dd->pport; + incr_cntr64(&ppd->port_xmit_discards); +} - if (ppd->port_xmit_discards < ~(u64)0) - ppd->port_xmit_discards++; +static void count_port_inactive(struct hfi1_devdata *dd) +{ + __count_port_discards(dd->pport); } /* @@ -5525,7 +5539,8 @@ static void count_port_inactive(struct hfi1_devdata *dd) * egress error if more than one packet fails the same integrity check * since we cleared the corresponding bit in SEND_EGRESS_ERR_INFO. */ -static void handle_send_egress_err_info(struct hfi1_devdata *dd) +static void handle_send_egress_err_info(struct hfi1_devdata *dd, + int vl) { struct hfi1_pportdata *ppd = dd->pport; u64 src = read_csr(dd, SEND_EGRESS_ERR_SOURCE); /* read first */ @@ -5540,10 +5555,24 @@ static void handle_send_egress_err_info(struct hfi1_devdata *dd) info, egress_err_info_string(buf, sizeof(buf), info), src); /* Eventually add other counters for each bit */ + if (info & PORT_DISCARD_EGRESS_ERRS) { + int weight, i; - if (info & SEND_EGRESS_ERR_INFO_TOO_LONG_IB_PACKET_ERR_SMASK) { - if (ppd->port_xmit_discards < ~(u64)0) - ppd->port_xmit_discards++; + /* + * Count all, in case multiple bits are set. Reminder: + * since there is only one info register for many sources, + * these may be attributed to the wrong VL if they occur + * too close together. + */ + weight = hweight64(info); + for (i = 0; i < weight; i++) { + __count_port_discards(ppd); + if (vl >= 0 && vl < TXE_NUM_DATA_VL) + incr_cntr64(&ppd->port_xmit_discards_vl[vl]); + else if (vl == 15) + incr_cntr64(&ppd->port_xmit_discards_vl + [C_VL_15]); + } } } @@ -5561,12 +5590,71 @@ static inline int port_inactive_err(u64 posn) * Input value is a bit position within the SEND_EGRESS_ERR_STATUS * register. Does it represent a 'disallowed packet' error? */ -static inline int disallowed_pkt_err(u64 posn) +static inline int disallowed_pkt_err(int posn) { return (posn >= SEES(TX_SDMA0_DISALLOWED_PACKET) && posn <= SEES(TX_SDMA15_DISALLOWED_PACKET)); } +/* + * Input value is a bit position of one of the SDMA engine disallowed + * packet errors. Return which engine. Use of this must be guarded by + * disallowed_pkt_err(). + */ +static inline int disallowed_pkt_engine(int posn) +{ + return posn - SEES(TX_SDMA0_DISALLOWED_PACKET); +} + +/* + * Translate an SDMA engine to a VL. Return -1 if the tranlation cannot + * be done. + */ +static int engine_to_vl(struct hfi1_devdata *dd, int engine) +{ + struct sdma_vl_map *m; + int vl; + + /* range check */ + if (engine < 0 || engine >= TXE_NUM_SDMA_ENGINES) + return -1; + + rcu_read_lock(); + m = rcu_dereference(dd->sdma_map); + vl = m->engine_to_vl[engine]; + rcu_read_unlock(); + + return vl; +} + +/* + * Translate the send context (sofware index) into a VL. Return -1 if the + * translation cannot be done. + */ +static int sc_to_vl(struct hfi1_devdata *dd, int sw_index) +{ + struct send_context_info *sci; + struct send_context *sc; + int i; + + sci = &dd->send_contexts[sw_index]; + + /* there is no information for user (PSM) and ack contexts */ + if (sci->type != SC_KERNEL) + return -1; + + sc = sci->sc; + if (!sc) + return -1; + if (dd->vld[15].sc == sc) + return 15; + for (i = 0; i < num_vls; i++) + if (dd->vld[i].sc == sc) + return i; + + return -1; +} + static void handle_egress_err(struct hfi1_devdata *dd, u32 unused, u64 reg) { u64 reg_copy = reg, handled = 0; @@ -5575,27 +5663,27 @@ static void handle_egress_err(struct hfi1_devdata *dd, u32 unused, u64 reg) if (reg & ALL_TXE_EGRESS_FREEZE_ERR) start_freeze_handling(dd->pport, 0); - if (is_ax(dd) && (reg & - SEND_EGRESS_ERR_STATUS_TX_CREDIT_RETURN_VL_ERR_SMASK) - && (dd->icode != ICODE_FUNCTIONAL_SIMULATOR)) + else if (is_ax(dd) && + (reg & SEND_EGRESS_ERR_STATUS_TX_CREDIT_RETURN_VL_ERR_SMASK) && + (dd->icode != ICODE_FUNCTIONAL_SIMULATOR)) start_freeze_handling(dd->pport, 0); while (reg_copy) { int posn = fls64(reg_copy); - /* - * fls64() returns a 1-based offset, but we generally - * want 0-based offsets. - */ + /* fls64() returns a 1-based offset, we want it zero based */ int shift = posn - 1; + u64 mask = 1ULL << shift; if (port_inactive_err(shift)) { count_port_inactive(dd); - handled |= (1ULL << shift); + handled |= mask; } else if (disallowed_pkt_err(shift)) { - handle_send_egress_err_info(dd); - handled |= (1ULL << shift); + int vl = engine_to_vl(dd, disallowed_pkt_engine(shift)); + + handle_send_egress_err_info(dd, vl); + handled |= mask; } - clear_bit(shift, (unsigned long *)®_copy); + reg_copy &= ~mask; } reg &= ~handled; @@ -5739,7 +5827,7 @@ static void is_sendctxt_err_int(struct hfi1_devdata *dd, send_context_err_status_string(flags, sizeof(flags), status)); if (status & SEND_CTXT_ERR_STATUS_PIO_DISALLOWED_PACKET_ERR_SMASK) - handle_send_egress_err_info(dd); + handle_send_egress_err_info(dd, sc_to_vl(dd, sw_index)); /* * Automatically restart halted kernel contexts out of interrupt diff --git a/drivers/staging/rdma/hfi1/hfi.h b/drivers/staging/rdma/hfi1/hfi.h index 0fe630e..76c369a 100644 --- a/drivers/staging/rdma/hfi1/hfi.h +++ b/drivers/staging/rdma/hfi1/hfi.h @@ -709,8 +709,9 @@ struct hfi1_pportdata { u64 *cntrs; /* port relative synthetic counter buffer */ u64 *scntrs; - /* we synthesize port_xmit_discards from several egress errors */ + /* port_xmit_discards are synthesized from different egress errors */ u64 port_xmit_discards; + u64 port_xmit_discards_vl[C_VL_COUNT]; u64 port_xmit_constraint_errors; u64 port_rcv_constraint_errors; /* count of 'link_err' interrupts from DC */ diff --git a/drivers/staging/rdma/hfi1/sdma.c b/drivers/staging/rdma/hfi1/sdma.c index cb66bd0..52fb22d 100644 --- a/drivers/staging/rdma/hfi1/sdma.c +++ b/drivers/staging/rdma/hfi1/sdma.c @@ -890,6 +890,9 @@ int sdma_map_init(struct hfi1_devdata *dd, u8 port, u8 num_vls, u8 *vl_engines) newmap->actual_vls = num_vls; newmap->vls = roundup_pow_of_two(num_vls); newmap->mask = (1 << ilog2(newmap->vls)) - 1; + /* initialize back-map */ + for (i = 0; i < TXE_NUM_SDMA_ENGINES; i++) + newmap->engine_to_vl[i] = -1; for (i = 0; i < newmap->vls; i++) { /* save for wrap around */ int first_engine = engine; @@ -913,6 +916,9 @@ int sdma_map_init(struct hfi1_devdata *dd, u8 port, u8 num_vls, u8 *vl_engines) /* wrap back to first engine */ engine = first_engine; } + /* assign back-map */ + for (j = 0; j < vl_engines[i]; j++) + newmap->engine_to_vl[first_engine + j] = i; } else { /* just re-use entry without allocating */ newmap->map[i] = newmap->map[i % num_vls]; diff --git a/drivers/staging/rdma/hfi1/sdma.h b/drivers/staging/rdma/hfi1/sdma.h index 0f51c45..1d52d6e 100644 --- a/drivers/staging/rdma/hfi1/sdma.h +++ b/drivers/staging/rdma/hfi1/sdma.h @@ -1087,6 +1087,7 @@ struct sdma_map_elem { /** * struct sdma_map_el - mapping for a vl + * @engine_to_vl - map of an engine to a vl * @list - rcu head for free callback * @mask - vl mask to "mod" the vl to produce an index to map array * @actual_vls - number of vls @@ -1098,6 +1099,7 @@ struct sdma_map_elem { * in turn point to an array of sde's for that vl. */ struct sdma_vl_map { + s8 engine_to_vl[TXE_NUM_SDMA_ENGINES]; struct rcu_head list; u32 mask; u8 actual_vls; -- To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org More majordomo info at http://vger.kernel.org/majordomo-info.html