From: Nipun Gupta Date: Tue, 7 Jul 2020 09:22:22 +0000 (+0530) Subject: bus/fslmc: support portal migration X-Git-Url: http://git.droids-corp.org/?a=commitdiff_plain;h=b3bd7a50d100f2328d349277745d012563f0ac93;p=dpdk.git bus/fslmc: support portal migration The patch adds support for portal migration by disabling stashing for the portals which is used in the non-affined threads, or on threads affined to multiple cores Signed-off-by: Nipun Gupta Acked-by: Hemant Agrawal --- diff --git a/doc/guides/rel_notes/release_20_08.rst b/doc/guides/rel_notes/release_20_08.rst index eb16d13ca9..4233d5a1c7 100644 --- a/doc/guides/rel_notes/release_20_08.rst +++ b/doc/guides/rel_notes/release_20_08.rst @@ -142,6 +142,12 @@ New Features * Re-implemented get_fdir_info and get_fdir_stat in private API. +* **Updated NXP dpaa2 ethdev PMD.** + + Updated the NXP dpaa2 ethdev with new features and improvements, including: + + * Added support to use datapath APIs from non-EAL pthread + * **Added DOCSIS protocol to rte_security.** Added support for combined crypto and CRC operations for the DOCSIS protocol diff --git a/drivers/bus/fslmc/portal/dpaa2_hw_dpio.c b/drivers/bus/fslmc/portal/dpaa2_hw_dpio.c index 5a12ff35d1..97be761164 100644 --- a/drivers/bus/fslmc/portal/dpaa2_hw_dpio.c +++ b/drivers/bus/fslmc/portal/dpaa2_hw_dpio.c @@ -53,10 +53,6 @@ static uint32_t io_space_count; /* Variable to store DPAA2 platform type */ uint32_t dpaa2_svr_family; -/* Physical core id for lcores running on dpaa2. */ -/* DPAA2 only support 1 lcore to 1 phy cpu mapping */ -static unsigned int dpaa2_cpu[RTE_MAX_LCORE]; - /* Variable to store DPAA2 DQRR size */ uint8_t dpaa2_dqrr_size; /* Variable to store DPAA2 EQCR size */ @@ -159,7 +155,7 @@ dpaa2_affine_dpio_intr_to_respective_core(int32_t dpio_id, int cpu_id) return; } - cpu_mask = cpu_mask << dpaa2_cpu[cpu_id]; + cpu_mask = cpu_mask << cpu_id; snprintf(command, COMMAND_LEN, "echo %X > /proc/irq/%s/smp_affinity", cpu_mask, token); ret = system(command); @@ -228,17 +224,9 @@ static void dpaa2_dpio_intr_deinit(struct dpaa2_dpio_dev *dpio_dev) #endif static int -dpaa2_configure_stashing(struct dpaa2_dpio_dev *dpio_dev) +dpaa2_configure_stashing(struct dpaa2_dpio_dev *dpio_dev, int cpu_id) { int sdest, ret; - int cpu_id; - - /* Set the Stashing Destination */ - cpu_id = dpaa2_get_core_id(); - if (cpu_id < 0) { - DPAA2_BUS_ERR("Thread not affined to a single core"); - return -1; - } /* Set the STASH Destination depending on Current CPU ID. * Valid values of SDEST are 4,5,6,7. Where, @@ -277,6 +265,7 @@ static void dpaa2_put_qbman_swp(struct dpaa2_dpio_dev *dpio_dev) static struct dpaa2_dpio_dev *dpaa2_get_qbman_swp(void) { struct dpaa2_dpio_dev *dpio_dev = NULL; + int cpu_id; int ret; /* Get DPIO dev handle from list using index */ @@ -292,11 +281,19 @@ static struct dpaa2_dpio_dev *dpaa2_get_qbman_swp(void) DPAA2_BUS_DEBUG("New Portal %p (%d) affined thread - %lu", dpio_dev, dpio_dev->index, syscall(SYS_gettid)); - ret = dpaa2_configure_stashing(dpio_dev); - if (ret) { - DPAA2_BUS_ERR("dpaa2_configure_stashing failed"); - rte_atomic16_clear(&dpio_dev->ref_count); - return NULL; + /* Set the Stashing Destination */ + cpu_id = dpaa2_get_core_id(); + if (cpu_id < 0) { + DPAA2_BUS_WARN("Thread not affined to a single core"); + if (dpaa2_svr_family != SVR_LX2160A) + qbman_swp_update(dpio_dev->sw_portal, 1); + } else { + ret = dpaa2_configure_stashing(dpio_dev, cpu_id); + if (ret) { + DPAA2_BUS_ERR("dpaa2_configure_stashing failed"); + rte_atomic16_clear(&dpio_dev->ref_count); + return NULL; + } } ret = pthread_setspecific(dpaa2_portal_key, (void *)dpio_dev); @@ -363,46 +360,6 @@ static void dpaa2_portal_finish(void *arg) pthread_setspecific(dpaa2_portal_key, NULL); } -/* - * This checks for not supported lcore mappings as well as get the physical - * cpuid for the lcore. - * one lcore can only map to 1 cpu i.e. 1@10-14 not supported. - * one cpu can be mapped to more than one lcores. - */ -static int -dpaa2_check_lcore_cpuset(void) -{ - unsigned int lcore_id, i; - int ret = 0; - - for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) - dpaa2_cpu[lcore_id] = 0xffffffff; - - for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) { - rte_cpuset_t cpuset = rte_lcore_cpuset(lcore_id); - - for (i = 0; i < CPU_SETSIZE; i++) { - if (!CPU_ISSET(i, &cpuset)) - continue; - if (i >= RTE_MAX_LCORE) { - DPAA2_BUS_ERR("ERR:lcore map to core %u (>= %u) not supported", - i, RTE_MAX_LCORE); - ret = -1; - continue; - } - RTE_LOG(DEBUG, EAL, "lcore id = %u cpu=%u\n", - lcore_id, i); - if (dpaa2_cpu[lcore_id] != 0xffffffff) { - DPAA2_BUS_ERR("ERR:lcore map to multi-cpu not supported"); - ret = -1; - continue; - } - dpaa2_cpu[lcore_id] = i; - } - } - return ret; -} - static int dpaa2_create_dpio_device(int vdev_fd, struct vfio_device_info *obj_info, @@ -413,7 +370,6 @@ dpaa2_create_dpio_device(int vdev_fd, struct qbman_swp_desc p_des; struct dpio_attr attr; int ret; - static int check_lcore_cpuset; if (obj_info->num_regions < NUM_DPIO_REGIONS) { DPAA2_BUS_ERR("Not sufficient number of DPIO regions"); @@ -433,13 +389,6 @@ dpaa2_create_dpio_device(int vdev_fd, /* Using single portal for all devices */ dpio_dev->mc_portal = dpaa2_get_mcp_ptr(MC_PORTAL_INDEX); - if (!check_lcore_cpuset) { - check_lcore_cpuset = 1; - - if (dpaa2_check_lcore_cpuset() < 0) - goto err; - } - dpio_dev->dpio = rte_zmalloc(NULL, sizeof(struct fsl_mc_io), RTE_CACHE_LINE_SIZE); if (!dpio_dev->dpio) { diff --git a/drivers/bus/fslmc/qbman/include/fsl_qbman_debug.h b/drivers/bus/fslmc/qbman/include/fsl_qbman_debug.h index 11267d439a..54096e8774 100644 --- a/drivers/bus/fslmc/qbman/include/fsl_qbman_debug.h +++ b/drivers/bus/fslmc/qbman/include/fsl_qbman_debug.h @@ -1,5 +1,6 @@ /* SPDX-License-Identifier: BSD-3-Clause * Copyright (C) 2015 Freescale Semiconductor, Inc. + * Copyright 2020 NXP */ #ifndef _FSL_QBMAN_DEBUG_H #define _FSL_QBMAN_DEBUG_H diff --git a/drivers/bus/fslmc/qbman/include/fsl_qbman_portal.h b/drivers/bus/fslmc/qbman/include/fsl_qbman_portal.h index f820077d2b..eb68c9cab5 100644 --- a/drivers/bus/fslmc/qbman/include/fsl_qbman_portal.h +++ b/drivers/bus/fslmc/qbman/include/fsl_qbman_portal.h @@ -1,7 +1,7 @@ /* SPDX-License-Identifier: BSD-3-Clause * * Copyright (C) 2014 Freescale Semiconductor, Inc. - * Copyright 2015-2019 NXP + * Copyright 2015-2020 NXP * */ #ifndef _FSL_QBMAN_PORTAL_H @@ -44,6 +44,12 @@ extern uint32_t dpaa2_svr_family; */ struct qbman_swp *qbman_swp_init(const struct qbman_swp_desc *d); +/** + * qbman_swp_update() - Update portal cacheability attributes. + * @p: the given qbman swp portal + */ +int qbman_swp_update(struct qbman_swp *p, int stash_off); + /** * qbman_swp_finish() - Create and destroy a functional object representing * the given QBMan portal descriptor. diff --git a/drivers/bus/fslmc/qbman/qbman_portal.c b/drivers/bus/fslmc/qbman/qbman_portal.c index d7ff74c7a2..57f50b0d8a 100644 --- a/drivers/bus/fslmc/qbman/qbman_portal.c +++ b/drivers/bus/fslmc/qbman/qbman_portal.c @@ -1,7 +1,7 @@ /* SPDX-License-Identifier: BSD-3-Clause * * Copyright (C) 2014-2016 Freescale Semiconductor, Inc. - * Copyright 2018-2019 NXP + * Copyright 2018-2020 NXP * */ @@ -82,6 +82,10 @@ qbman_swp_enqueue_ring_mode_cinh_direct(struct qbman_swp *s, const struct qbman_eq_desc *d, const struct qbman_fd *fd); static int +qbman_swp_enqueue_ring_mode_cinh_direct(struct qbman_swp *s, + const struct qbman_eq_desc *d, + const struct qbman_fd *fd); +static int qbman_swp_enqueue_ring_mode_mem_back(struct qbman_swp *s, const struct qbman_eq_desc *d, const struct qbman_fd *fd); @@ -99,6 +103,12 @@ qbman_swp_enqueue_multiple_cinh_direct(struct qbman_swp *s, uint32_t *flags, int num_frames); static int +qbman_swp_enqueue_multiple_cinh_direct(struct qbman_swp *s, + const struct qbman_eq_desc *d, + const struct qbman_fd *fd, + uint32_t *flags, + int num_frames); +static int qbman_swp_enqueue_multiple_mem_back(struct qbman_swp *s, const struct qbman_eq_desc *d, const struct qbman_fd *fd, @@ -118,6 +128,12 @@ qbman_swp_enqueue_multiple_fd_cinh_direct(struct qbman_swp *s, uint32_t *flags, int num_frames); static int +qbman_swp_enqueue_multiple_fd_cinh_direct(struct qbman_swp *s, + const struct qbman_eq_desc *d, + struct qbman_fd **fd, + uint32_t *flags, + int num_frames); +static int qbman_swp_enqueue_multiple_fd_mem_back(struct qbman_swp *s, const struct qbman_eq_desc *d, struct qbman_fd **fd, @@ -135,6 +151,11 @@ qbman_swp_enqueue_multiple_desc_cinh_direct(struct qbman_swp *s, const struct qbman_fd *fd, int num_frames); static int +qbman_swp_enqueue_multiple_desc_cinh_direct(struct qbman_swp *s, + const struct qbman_eq_desc *d, + const struct qbman_fd *fd, + int num_frames); +static int qbman_swp_enqueue_multiple_desc_mem_back(struct qbman_swp *s, const struct qbman_eq_desc *d, const struct qbman_fd *fd, @@ -143,9 +164,12 @@ qbman_swp_enqueue_multiple_desc_mem_back(struct qbman_swp *s, static int qbman_swp_pull_direct(struct qbman_swp *s, struct qbman_pull_desc *d); static int +qbman_swp_pull_cinh_direct(struct qbman_swp *s, struct qbman_pull_desc *d); +static int qbman_swp_pull_mem_back(struct qbman_swp *s, struct qbman_pull_desc *d); const struct qbman_result *qbman_swp_dqrr_next_direct(struct qbman_swp *s); +const struct qbman_result *qbman_swp_dqrr_next_cinh_direct(struct qbman_swp *s); const struct qbman_result *qbman_swp_dqrr_next_mem_back(struct qbman_swp *s); static int @@ -153,6 +177,10 @@ qbman_swp_release_direct(struct qbman_swp *s, const struct qbman_release_desc *d, const uint64_t *buffers, unsigned int num_buffers); static int +qbman_swp_release_cinh_direct(struct qbman_swp *s, + const struct qbman_release_desc *d, + const uint64_t *buffers, unsigned int num_buffers); +static int qbman_swp_release_mem_back(struct qbman_swp *s, const struct qbman_release_desc *d, const uint64_t *buffers, unsigned int num_buffers); @@ -327,6 +355,28 @@ struct qbman_swp *qbman_swp_init(const struct qbman_swp_desc *d) return p; } +int qbman_swp_update(struct qbman_swp *p, int stash_off) +{ + const struct qbman_swp_desc *d = &p->desc; + struct qbman_swp_sys *s = &p->sys; + int ret; + + /* Nothing needs to be done for QBMAN rev > 5000 with fast access */ + if ((qman_version & QMAN_REV_MASK) >= QMAN_REV_5000 + && (d->cena_access_mode == qman_cena_fastest_access)) + return 0; + + ret = qbman_swp_sys_update(s, d, p->dqrr.dqrr_size, stash_off); + if (ret) { + pr_err("qbman_swp_sys_init() failed %d\n", ret); + return ret; + } + + p->stash_off = stash_off; + + return 0; +} + void qbman_swp_finish(struct qbman_swp *p) { #ifdef QBMAN_CHECKING @@ -462,6 +512,27 @@ void qbman_swp_mc_submit(struct qbman_swp *p, void *cmd, uint8_t cmd_verb) #endif } +void qbman_swp_mc_submit_cinh(struct qbman_swp *p, void *cmd, uint8_t cmd_verb) +{ + uint8_t *v = cmd; +#ifdef QBMAN_CHECKING + QBMAN_BUG_ON(!(p->mc.check != swp_mc_can_submit)); +#endif + /* TBD: "|=" is going to hurt performance. Need to move as many fields + * out of word zero, and for those that remain, the "OR" needs to occur + * at the caller side. This debug check helps to catch cases where the + * caller wants to OR but has forgotten to do so. + */ + QBMAN_BUG_ON((*v & cmd_verb) != *v); + dma_wmb(); + *v = cmd_verb | p->mc.valid_bit; + qbman_cinh_write_complete(&p->sys, QBMAN_CENA_SWP_CR, cmd); + clean(cmd); +#ifdef QBMAN_CHECKING + p->mc.check = swp_mc_can_poll; +#endif +} + void *qbman_swp_mc_result(struct qbman_swp *p) { uint32_t *ret, verb; @@ -500,6 +571,27 @@ void *qbman_swp_mc_result(struct qbman_swp *p) return ret; } +void *qbman_swp_mc_result_cinh(struct qbman_swp *p) +{ + uint32_t *ret, verb; +#ifdef QBMAN_CHECKING + QBMAN_BUG_ON(p->mc.check != swp_mc_can_poll); +#endif + ret = qbman_cinh_read_shadow(&p->sys, + QBMAN_CENA_SWP_RR(p->mc.valid_bit)); + /* Remove the valid-bit - + * command completed iff the rest is non-zero + */ + verb = ret[0] & ~QB_VALID_BIT; + if (!verb) + return NULL; + p->mc.valid_bit ^= QB_VALID_BIT; +#ifdef QBMAN_CHECKING + p->mc.check = swp_mc_can_start; +#endif + return ret; +} + /***********/ /* Enqueue */ /***********/ @@ -640,6 +732,16 @@ static inline void qbman_write_eqcr_am_rt_register(struct qbman_swp *p, QMAN_RT_MODE); } +static void memcpy_byte_by_byte(void *to, const void *from, size_t n) +{ + const uint8_t *src = from; + volatile uint8_t *dest = to; + size_t i; + + for (i = 0; i < n; i++) + dest[i] = src[i]; +} + static int qbman_swp_enqueue_array_mode_direct(struct qbman_swp *s, const struct qbman_eq_desc *d, @@ -754,7 +856,7 @@ static int qbman_swp_enqueue_ring_mode_cinh_direct( return -EBUSY; } - p = qbman_cena_write_start_wo_shadow(&s->sys, + p = qbman_cinh_write_start_wo_shadow(&s->sys, QBMAN_CENA_SWP_EQCR(s->eqcr.pi & half_mask)); memcpy(&p[1], &cl[1], 28); memcpy(&p[8], fd, sizeof(*fd)); @@ -762,8 +864,6 @@ static int qbman_swp_enqueue_ring_mode_cinh_direct( /* Set the verb byte, have to substitute in the valid-bit */ p[0] = cl[0] | s->eqcr.pi_vb; - qbman_cena_write_complete_wo_shadow(&s->sys, - QBMAN_CENA_SWP_EQCR(s->eqcr.pi & half_mask)); s->eqcr.pi++; s->eqcr.pi &= full_mask; s->eqcr.available--; @@ -815,7 +915,10 @@ static int qbman_swp_enqueue_ring_mode(struct qbman_swp *s, const struct qbman_eq_desc *d, const struct qbman_fd *fd) { - return qbman_swp_enqueue_ring_mode_ptr(s, d, fd); + if (!s->stash_off) + return qbman_swp_enqueue_ring_mode_ptr(s, d, fd); + else + return qbman_swp_enqueue_ring_mode_cinh_direct(s, d, fd); } int qbman_swp_enqueue(struct qbman_swp *s, const struct qbman_eq_desc *d, @@ -1025,7 +1128,12 @@ int qbman_swp_enqueue_multiple(struct qbman_swp *s, uint32_t *flags, int num_frames) { - return qbman_swp_enqueue_multiple_ptr(s, d, fd, flags, num_frames); + if (!s->stash_off) + return qbman_swp_enqueue_multiple_ptr(s, d, fd, flags, + num_frames); + else + return qbman_swp_enqueue_multiple_cinh_direct(s, d, fd, flags, + num_frames); } static int qbman_swp_enqueue_multiple_fd_direct(struct qbman_swp *s, @@ -1233,7 +1341,12 @@ int qbman_swp_enqueue_multiple_fd(struct qbman_swp *s, uint32_t *flags, int num_frames) { - return qbman_swp_enqueue_multiple_fd_ptr(s, d, fd, flags, num_frames); + if (!s->stash_off) + return qbman_swp_enqueue_multiple_fd_ptr(s, d, fd, flags, + num_frames); + else + return qbman_swp_enqueue_multiple_fd_cinh_direct(s, d, fd, + flags, num_frames); } static int qbman_swp_enqueue_multiple_desc_direct(struct qbman_swp *s, @@ -1426,7 +1539,13 @@ int qbman_swp_enqueue_multiple_desc(struct qbman_swp *s, const struct qbman_fd *fd, int num_frames) { - return qbman_swp_enqueue_multiple_desc_ptr(s, d, fd, num_frames); + if (!s->stash_off) + return qbman_swp_enqueue_multiple_desc_ptr(s, d, fd, + num_frames); + else + return qbman_swp_enqueue_multiple_desc_cinh_direct(s, d, fd, + num_frames); + } /*************************/ @@ -1574,6 +1693,30 @@ static int qbman_swp_pull_direct(struct qbman_swp *s, return 0; } +static int qbman_swp_pull_cinh_direct(struct qbman_swp *s, + struct qbman_pull_desc *d) +{ + uint32_t *p; + uint32_t *cl = qb_cl(d); + + if (!atomic_dec_and_test(&s->vdq.busy)) { + atomic_inc(&s->vdq.busy); + return -EBUSY; + } + + d->pull.tok = s->sys.idx + 1; + s->vdq.storage = (void *)(size_t)d->pull.rsp_addr_virt; + p = qbman_cinh_write_start_wo_shadow(&s->sys, QBMAN_CENA_SWP_VDQCR); + memcpy_byte_by_byte(&p[1], &cl[1], 12); + + /* Set the verb byte, have to substitute in the valid-bit */ + lwsync(); + p[0] = cl[0] | s->vdq.valid_bit; + s->vdq.valid_bit ^= QB_VALID_BIT; + + return 0; +} + static int qbman_swp_pull_mem_back(struct qbman_swp *s, struct qbman_pull_desc *d) { @@ -1601,7 +1744,10 @@ static int qbman_swp_pull_mem_back(struct qbman_swp *s, int qbman_swp_pull(struct qbman_swp *s, struct qbman_pull_desc *d) { - return qbman_swp_pull_ptr(s, d); + if (!s->stash_off) + return qbman_swp_pull_ptr(s, d); + else + return qbman_swp_pull_cinh_direct(s, d); } /****************/ @@ -1638,7 +1784,10 @@ void qbman_swp_prefetch_dqrr_next(struct qbman_swp *s) */ const struct qbman_result *qbman_swp_dqrr_next(struct qbman_swp *s) { - return qbman_swp_dqrr_next_ptr(s); + if (!s->stash_off) + return qbman_swp_dqrr_next_ptr(s); + else + return qbman_swp_dqrr_next_cinh_direct(s); } const struct qbman_result *qbman_swp_dqrr_next_direct(struct qbman_swp *s) @@ -1718,6 +1867,81 @@ const struct qbman_result *qbman_swp_dqrr_next_direct(struct qbman_swp *s) return p; } +const struct qbman_result *qbman_swp_dqrr_next_cinh_direct(struct qbman_swp *s) +{ + uint32_t verb; + uint32_t response_verb; + uint32_t flags; + const struct qbman_result *p; + + /* Before using valid-bit to detect if something is there, we have to + * handle the case of the DQRR reset bug... + */ + if (s->dqrr.reset_bug) { + /* We pick up new entries by cache-inhibited producer index, + * which means that a non-coherent mapping would require us to + * invalidate and read *only* once that PI has indicated that + * there's an entry here. The first trip around the DQRR ring + * will be much less efficient than all subsequent trips around + * it... + */ + uint8_t pi = qbman_cinh_read(&s->sys, QBMAN_CINH_SWP_DQPI) & + QMAN_DQRR_PI_MASK; + + /* there are new entries if pi != next_idx */ + if (pi == s->dqrr.next_idx) + return NULL; + + /* if next_idx is/was the last ring index, and 'pi' is + * different, we can disable the workaround as all the ring + * entries have now been DMA'd to so valid-bit checking is + * repaired. Note: this logic needs to be based on next_idx + * (which increments one at a time), rather than on pi (which + * can burst and wrap-around between our snapshots of it). + */ + QBMAN_BUG_ON((s->dqrr.dqrr_size - 1) < 0); + if (s->dqrr.next_idx == (s->dqrr.dqrr_size - 1u)) { + pr_debug("DEBUG: next_idx=%d, pi=%d, clear reset bug\n", + s->dqrr.next_idx, pi); + s->dqrr.reset_bug = 0; + } + } + p = qbman_cinh_read_wo_shadow(&s->sys, + QBMAN_CENA_SWP_DQRR(s->dqrr.next_idx)); + + verb = p->dq.verb; + + /* If the valid-bit isn't of the expected polarity, nothing there. Note, + * in the DQRR reset bug workaround, we shouldn't need to skip these + * check, because we've already determined that a new entry is available + * and we've invalidated the cacheline before reading it, so the + * valid-bit behaviour is repaired and should tell us what we already + * knew from reading PI. + */ + if ((verb & QB_VALID_BIT) != s->dqrr.valid_bit) + return NULL; + + /* There's something there. Move "next_idx" attention to the next ring + * entry (and prefetch it) before returning what we found. + */ + s->dqrr.next_idx++; + if (s->dqrr.next_idx == s->dqrr.dqrr_size) { + s->dqrr.next_idx = 0; + s->dqrr.valid_bit ^= QB_VALID_BIT; + } + /* If this is the final response to a volatile dequeue command + * indicate that the vdq is no longer busy + */ + flags = p->dq.stat; + response_verb = verb & QBMAN_RESPONSE_VERB_MASK; + if ((response_verb == QBMAN_RESULT_DQ) && + (flags & QBMAN_DQ_STAT_VOLATILE) && + (flags & QBMAN_DQ_STAT_EXPIRED)) + atomic_inc(&s->vdq.busy); + + return p; +} + const struct qbman_result *qbman_swp_dqrr_next_mem_back(struct qbman_swp *s) { uint32_t verb; @@ -2096,6 +2320,37 @@ static int qbman_swp_release_direct(struct qbman_swp *s, return 0; } +static int qbman_swp_release_cinh_direct(struct qbman_swp *s, + const struct qbman_release_desc *d, + const uint64_t *buffers, + unsigned int num_buffers) +{ + uint32_t *p; + const uint32_t *cl = qb_cl(d); + uint32_t rar = qbman_cinh_read(&s->sys, QBMAN_CINH_SWP_RAR); + + pr_debug("RAR=%08x\n", rar); + if (!RAR_SUCCESS(rar)) + return -EBUSY; + + QBMAN_BUG_ON(!num_buffers || (num_buffers > 7)); + + /* Start the release command */ + p = qbman_cinh_write_start_wo_shadow(&s->sys, + QBMAN_CENA_SWP_RCR(RAR_IDX(rar))); + + /* Copy the caller's buffer pointers to the command */ + memcpy_byte_by_byte(&p[2], buffers, num_buffers * sizeof(uint64_t)); + + /* Set the verb byte, have to substitute in the valid-bit and the + * number of buffers. + */ + lwsync(); + p[0] = cl[0] | RAR_VB(rar) | num_buffers; + + return 0; +} + static int qbman_swp_release_mem_back(struct qbman_swp *s, const struct qbman_release_desc *d, const uint64_t *buffers, @@ -2134,7 +2389,11 @@ int qbman_swp_release(struct qbman_swp *s, const uint64_t *buffers, unsigned int num_buffers) { - return qbman_swp_release_ptr(s, d, buffers, num_buffers); + if (!s->stash_off) + return qbman_swp_release_ptr(s, d, buffers, num_buffers); + else + return qbman_swp_release_cinh_direct(s, d, buffers, + num_buffers); } /*******************/ @@ -2157,8 +2416,8 @@ struct qbman_acquire_rslt { uint64_t buf[7]; }; -int qbman_swp_acquire(struct qbman_swp *s, uint16_t bpid, uint64_t *buffers, - unsigned int num_buffers) +static int qbman_swp_acquire_direct(struct qbman_swp *s, uint16_t bpid, + uint64_t *buffers, unsigned int num_buffers) { struct qbman_acquire_desc *p; struct qbman_acquire_rslt *r; @@ -2202,6 +2461,61 @@ int qbman_swp_acquire(struct qbman_swp *s, uint16_t bpid, uint64_t *buffers, return (int)r->num; } +static int qbman_swp_acquire_cinh_direct(struct qbman_swp *s, uint16_t bpid, + uint64_t *buffers, unsigned int num_buffers) +{ + struct qbman_acquire_desc *p; + struct qbman_acquire_rslt *r; + + if (!num_buffers || (num_buffers > 7)) + return -EINVAL; + + /* Start the management command */ + p = qbman_swp_mc_start(s); + + if (!p) + return -EBUSY; + + /* Encode the caller-provided attributes */ + p->bpid = bpid; + p->num = num_buffers; + + /* Complete the management command */ + r = qbman_swp_mc_complete_cinh(s, p, QBMAN_MC_ACQUIRE); + if (!r) { + pr_err("qbman: acquire from BPID %d failed, no response\n", + bpid); + return -EIO; + } + + /* Decode the outcome */ + QBMAN_BUG_ON((r->verb & QBMAN_RESPONSE_VERB_MASK) != QBMAN_MC_ACQUIRE); + + /* Determine success or failure */ + if (r->rslt != QBMAN_MC_RSLT_OK) { + pr_err("Acquire buffers from BPID 0x%x failed, code=0x%02x\n", + bpid, r->rslt); + return -EIO; + } + + QBMAN_BUG_ON(r->num > num_buffers); + + /* Copy the acquired buffers to the caller's array */ + u64_from_le32_copy(buffers, &r->buf[0], r->num); + + return (int)r->num; +} + +int qbman_swp_acquire(struct qbman_swp *s, uint16_t bpid, uint64_t *buffers, + unsigned int num_buffers) +{ + if (!s->stash_off) + return qbman_swp_acquire_direct(s, bpid, buffers, num_buffers); + else + return qbman_swp_acquire_cinh_direct(s, bpid, buffers, + num_buffers); +} + /*****************/ /* FQ management */ /*****************/ diff --git a/drivers/bus/fslmc/qbman/qbman_portal.h b/drivers/bus/fslmc/qbman/qbman_portal.h index 3aaacae52f..1cf7918309 100644 --- a/drivers/bus/fslmc/qbman/qbman_portal.h +++ b/drivers/bus/fslmc/qbman/qbman_portal.h @@ -1,7 +1,7 @@ /* SPDX-License-Identifier: BSD-3-Clause * * Copyright (C) 2014-2016 Freescale Semiconductor, Inc. - * Copyright 2018-2019 NXP + * Copyright 2018-2020 NXP * */ @@ -102,6 +102,7 @@ struct qbman_swp { uint32_t ci; int available; } eqcr; + uint8_t stash_off; }; /* -------------------------- */ @@ -118,7 +119,9 @@ struct qbman_swp { */ void *qbman_swp_mc_start(struct qbman_swp *p); void qbman_swp_mc_submit(struct qbman_swp *p, void *cmd, uint8_t cmd_verb); +void qbman_swp_mc_submit_cinh(struct qbman_swp *p, void *cmd, uint8_t cmd_verb); void *qbman_swp_mc_result(struct qbman_swp *p); +void *qbman_swp_mc_result_cinh(struct qbman_swp *p); /* Wraps up submit + poll-for-result */ static inline void *qbman_swp_mc_complete(struct qbman_swp *swp, void *cmd, @@ -135,6 +138,20 @@ static inline void *qbman_swp_mc_complete(struct qbman_swp *swp, void *cmd, return cmd; } +static inline void *qbman_swp_mc_complete_cinh(struct qbman_swp *swp, void *cmd, + uint8_t cmd_verb) +{ + int loopvar = 1000; + + qbman_swp_mc_submit_cinh(swp, cmd, cmd_verb); + do { + cmd = qbman_swp_mc_result_cinh(swp); + } while (!cmd && loopvar--); + QBMAN_BUG_ON(!loopvar); + + return cmd; +} + /* ---------------------- */ /* Descriptors/cachelines */ /* ---------------------- */ diff --git a/drivers/bus/fslmc/qbman/qbman_sys.h b/drivers/bus/fslmc/qbman/qbman_sys.h index 55449edf38..61f817c477 100644 --- a/drivers/bus/fslmc/qbman/qbman_sys.h +++ b/drivers/bus/fslmc/qbman/qbman_sys.h @@ -1,7 +1,7 @@ /* SPDX-License-Identifier: BSD-3-Clause * * Copyright (C) 2014-2016 Freescale Semiconductor, Inc. - * Copyright 2019 NXP + * Copyright 2019-2020 NXP */ /* qbman_sys_decl.h and qbman_sys.h are the two platform-specific files in the * driver. They are only included via qbman_private.h, which is itself a @@ -190,6 +190,34 @@ static inline void qbman_cinh_write(struct qbman_swp_sys *s, uint32_t offset, #endif } +static inline void *qbman_cinh_write_start_wo_shadow(struct qbman_swp_sys *s, + uint32_t offset) +{ +#ifdef QBMAN_CINH_TRACE + pr_info("qbman_cinh_write_start(%p:%d:0x%03x)\n", + s->addr_cinh, s->idx, offset); +#endif + QBMAN_BUG_ON(offset & 63); + return (s->addr_cinh + offset); +} + +static inline void qbman_cinh_write_complete(struct qbman_swp_sys *s, + uint32_t offset, void *cmd) +{ + const uint32_t *shadow = cmd; + int loop; +#ifdef QBMAN_CINH_TRACE + pr_info("qbman_cinh_write_complete(%p:%d:0x%03x) %p\n", + s->addr_cinh, s->idx, offset, shadow); + hexdump(cmd, 64); +#endif + for (loop = 15; loop >= 1; loop--) + __raw_writel(shadow[loop], s->addr_cinh + + offset + loop * 4); + lwsync(); + __raw_writel(shadow[0], s->addr_cinh + offset); +} + static inline uint32_t qbman_cinh_read(struct qbman_swp_sys *s, uint32_t offset) { uint32_t reg = __raw_readl(s->addr_cinh + offset); @@ -200,6 +228,35 @@ static inline uint32_t qbman_cinh_read(struct qbman_swp_sys *s, uint32_t offset) return reg; } +static inline void *qbman_cinh_read_shadow(struct qbman_swp_sys *s, + uint32_t offset) +{ + uint32_t *shadow = (uint32_t *)(s->cena + offset); + unsigned int loop; +#ifdef QBMAN_CINH_TRACE + pr_info(" %s (%p:%d:0x%03x) %p\n", __func__, + s->addr_cinh, s->idx, offset, shadow); +#endif + + for (loop = 0; loop < 16; loop++) + shadow[loop] = __raw_readl(s->addr_cinh + offset + + loop * 4); +#ifdef QBMAN_CINH_TRACE + hexdump(shadow, 64); +#endif + return shadow; +} + +static inline void *qbman_cinh_read_wo_shadow(struct qbman_swp_sys *s, + uint32_t offset) +{ +#ifdef QBMAN_CINH_TRACE + pr_info("qbman_cinh_read(%p:%d:0x%03x)\n", + s->addr_cinh, s->idx, offset); +#endif + return s->addr_cinh + offset; +} + static inline void *qbman_cena_write_start(struct qbman_swp_sys *s, uint32_t offset) { @@ -476,6 +533,82 @@ static inline int qbman_swp_sys_init(struct qbman_swp_sys *s, return 0; } +static inline int qbman_swp_sys_update(struct qbman_swp_sys *s, + const struct qbman_swp_desc *d, + uint8_t dqrr_size, + int stash_off) +{ + uint32_t reg; + int i; + int cena_region_size = 4*1024; + uint8_t est = 1; +#ifdef RTE_ARCH_64 + uint8_t wn = CENA_WRITE_ENABLE; +#else + uint8_t wn = CINH_WRITE_ENABLE; +#endif + + if (stash_off) + wn = CINH_WRITE_ENABLE; + + QBMAN_BUG_ON(d->idx < 0); +#ifdef QBMAN_CHECKING + /* We should never be asked to initialise for a portal that isn't in + * the power-on state. (Ie. don't forget to reset portals when they are + * decommissioned!) + */ + reg = qbman_cinh_read(s, QBMAN_CINH_SWP_CFG); + QBMAN_BUG_ON(reg); +#endif + if ((d->qman_version & QMAN_REV_MASK) >= QMAN_REV_5000 + && (d->cena_access_mode == qman_cena_fastest_access)) + memset(s->addr_cena, 0, cena_region_size); + else { + /* Invalidate the portal memory. + * This ensures no stale cache lines + */ + for (i = 0; i < cena_region_size; i += 64) + dccivac(s->addr_cena + i); + } + + if (dpaa2_svr_family == SVR_LS1080A) + est = 0; + + if (s->eqcr_mode == qman_eqcr_vb_array) { + reg = qbman_set_swp_cfg(dqrr_size, wn, + 0, 3, 2, 3, 1, 1, 1, 1, 1, 1); + } else { + if ((d->qman_version & QMAN_REV_MASK) >= QMAN_REV_5000 && + (d->cena_access_mode == qman_cena_fastest_access)) + reg = qbman_set_swp_cfg(dqrr_size, wn, + 1, 3, 2, 0, 1, 1, 1, 1, 1, 1); + else + reg = qbman_set_swp_cfg(dqrr_size, wn, + est, 3, 2, 2, 1, 1, 1, 1, 1, 1); + } + + if ((d->qman_version & QMAN_REV_MASK) >= QMAN_REV_5000 + && (d->cena_access_mode == qman_cena_fastest_access)) + reg |= 1 << SWP_CFG_CPBS_SHIFT | /* memory-backed mode */ + 1 << SWP_CFG_VPM_SHIFT | /* VDQCR read triggered mode */ + 1 << SWP_CFG_CPM_SHIFT; /* CR read triggered mode */ + + qbman_cinh_write(s, QBMAN_CINH_SWP_CFG, reg); + reg = qbman_cinh_read(s, QBMAN_CINH_SWP_CFG); + if (!reg) { + pr_err("The portal %d is not enabled!\n", s->idx); + return -1; + } + + if ((d->qman_version & QMAN_REV_MASK) >= QMAN_REV_5000 + && (d->cena_access_mode == qman_cena_fastest_access)) { + qbman_cinh_write(s, QBMAN_CINH_SWP_EQCR_PI, QMAN_RT_MODE); + qbman_cinh_write(s, QBMAN_CINH_SWP_RCR_PI, QMAN_RT_MODE); + } + + return 0; +} + static inline void qbman_swp_sys_finish(struct qbman_swp_sys *s) { free(s->cena);