diff options
Diffstat (limited to 'drivers/bus/dpaa/base/qbman/qman.c')
| -rw-r--r-- | drivers/bus/dpaa/base/qbman/qman.c | 2497 |
1 files changed, 2497 insertions, 0 deletions
diff --git a/drivers/bus/dpaa/base/qbman/qman.c b/drivers/bus/dpaa/base/qbman/qman.c new file mode 100644 index 00000000..87fec60d --- /dev/null +++ b/drivers/bus/dpaa/base/qbman/qman.c @@ -0,0 +1,2497 @@ +/*- + * This file is provided under a dual BSD/GPLv2 license. When using or + * redistributing this file, you may do so under either license. + * + * BSD LICENSE + * + * Copyright 2008-2016 Freescale Semiconductor Inc. + * Copyright 2017 NXP. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are met: + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * * Neither the name of the above-listed copyright holders nor the + * names of any contributors may be used to endorse or promote products + * derived from this software without specific prior written permission. + * + * GPL LICENSE SUMMARY + * + * ALTERNATIVELY, this software may be distributed under the terms of the + * GNU General Public License ("GPL") as published by the Free Software + * Foundation, either version 2 of that License or (at your option) any + * later version. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" + * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDERS OR CONTRIBUTORS BE + * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR + * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF + * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS + * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN + * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) + * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE + * POSSIBILITY OF SUCH DAMAGE. + */ + +#include "qman.h" +#include <rte_branch_prediction.h> + +/* Compilation constants */ +#define DQRR_MAXFILL 15 +#define EQCR_ITHRESH 4 /* if EQCR congests, interrupt threshold */ +#define IRQNAME "QMan portal %d" +#define MAX_IRQNAME 16 /* big enough for "QMan portal %d" */ +/* maximum number of DQRR entries to process in qman_poll() */ +#define FSL_QMAN_POLL_LIMIT 8 + +/* Lock/unlock frame queues, subject to the "LOCKED" flag. This is about + * inter-processor locking only. Note, FQLOCK() is always called either under a + * local_irq_save() or from interrupt context - hence there's no need for irq + * protection (and indeed, attempting to nest irq-protection doesn't work, as + * the "irq en/disable" machinery isn't recursive...). + */ +#define FQLOCK(fq) \ + do { \ + struct qman_fq *__fq478 = (fq); \ + if (fq_isset(__fq478, QMAN_FQ_FLAG_LOCKED)) \ + spin_lock(&__fq478->fqlock); \ + } while (0) +#define FQUNLOCK(fq) \ + do { \ + struct qman_fq *__fq478 = (fq); \ + if (fq_isset(__fq478, QMAN_FQ_FLAG_LOCKED)) \ + spin_unlock(&__fq478->fqlock); \ + } while (0) + +static inline void fq_set(struct qman_fq *fq, u32 mask) +{ + dpaa_set_bits(mask, &fq->flags); +} + +static inline void fq_clear(struct qman_fq *fq, u32 mask) +{ + dpaa_clear_bits(mask, &fq->flags); +} + +static inline int fq_isset(struct qman_fq *fq, u32 mask) +{ + return fq->flags & mask; +} + +static inline int fq_isclear(struct qman_fq *fq, u32 mask) +{ + return !(fq->flags & mask); +} + +struct qman_portal { + struct qm_portal p; + /* PORTAL_BITS_*** - dynamic, strictly internal */ + unsigned long bits; + /* interrupt sources processed by portal_isr(), configurable */ + unsigned long irq_sources; + u32 use_eqcr_ci_stashing; + u32 slowpoll; /* only used when interrupts are off */ + /* only 1 volatile dequeue at a time */ + struct qman_fq *vdqcr_owned; + u32 sdqcr; + int dqrr_disable_ref; + /* A portal-specific handler for DCP ERNs. If this is NULL, the global + * handler is called instead. + */ + qman_cb_dc_ern cb_dc_ern; + /* When the cpu-affine portal is activated, this is non-NULL */ + const struct qm_portal_config *config; + struct dpa_rbtree retire_table; + char irqname[MAX_IRQNAME]; + /* 2-element array. cgrs[0] is mask, cgrs[1] is snapshot. */ + struct qman_cgrs *cgrs; + /* linked-list of CSCN handlers. */ + struct list_head cgr_cbs; + /* list lock */ + spinlock_t cgr_lock; + /* track if memory was allocated by the driver */ +#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ + /* Keep a shadow copy of the DQRR on LE systems as the SW needs to + * do byte swaps of DQRR read only memory. First entry must be aligned + * to 2 ** 10 to ensure DQRR index calculations based shadow copy + * address (6 bits for address shift + 4 bits for the DQRR size). + */ + struct qm_dqrr_entry shadow_dqrr[QM_DQRR_SIZE] + __attribute__((aligned(1024))); +#endif +}; + +/* Global handler for DCP ERNs. Used when the portal receiving the message does + * not have a portal-specific handler. + */ +static qman_cb_dc_ern cb_dc_ern; + +static cpumask_t affine_mask; +static DEFINE_SPINLOCK(affine_mask_lock); +static u16 affine_channels[NR_CPUS]; +static RTE_DEFINE_PER_LCORE(struct qman_portal, qman_affine_portal); + +static inline struct qman_portal *get_affine_portal(void) +{ + return &RTE_PER_LCORE(qman_affine_portal); +} + +/* This gives a FQID->FQ lookup to cover the fact that we can't directly demux + * retirement notifications (the fact they are sometimes h/w-consumed means that + * contextB isn't always a s/w demux - and as we can't know which case it is + * when looking at the notification, we have to use the slow lookup for all of + * them). NB, it's possible to have multiple FQ objects refer to the same FQID + * (though at most one of them should be the consumer), so this table isn't for + * all FQs - FQs are added when retirement commands are issued, and removed when + * they complete, which also massively reduces the size of this table. + */ +IMPLEMENT_DPAA_RBTREE(fqtree, struct qman_fq, node, fqid); +/* + * This is what everything can wait on, even if it migrates to a different cpu + * to the one whose affine portal it is waiting on. + */ +static DECLARE_WAIT_QUEUE_HEAD(affine_queue); + +static inline int table_push_fq(struct qman_portal *p, struct qman_fq *fq) +{ + int ret = fqtree_push(&p->retire_table, fq); + + if (ret) + pr_err("ERROR: double FQ-retirement %d\n", fq->fqid); + return ret; +} + +static inline void table_del_fq(struct qman_portal *p, struct qman_fq *fq) +{ + fqtree_del(&p->retire_table, fq); +} + +static inline struct qman_fq *table_find_fq(struct qman_portal *p, u32 fqid) +{ + return fqtree_find(&p->retire_table, fqid); +} + +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP +static void **qman_fq_lookup_table; +static size_t qman_fq_lookup_table_size; + +int qman_setup_fq_lookup_table(size_t num_entries) +{ + num_entries++; + /* Allocate 1 more entry since the first entry is not used */ + qman_fq_lookup_table = vmalloc((num_entries * sizeof(void *))); + if (!qman_fq_lookup_table) { + pr_err("QMan: Could not allocate fq lookup table\n"); + return -ENOMEM; + } + memset(qman_fq_lookup_table, 0, num_entries * sizeof(void *)); + qman_fq_lookup_table_size = num_entries; + pr_debug("QMan: Allocated lookup table at %p, entry count %lu\n", + qman_fq_lookup_table, + (unsigned long)qman_fq_lookup_table_size); + return 0; +} + +/* global structure that maintains fq object mapping */ +static DEFINE_SPINLOCK(fq_hash_table_lock); + +static int find_empty_fq_table_entry(u32 *entry, struct qman_fq *fq) +{ + u32 i; + + spin_lock(&fq_hash_table_lock); + /* Can't use index zero because this has special meaning + * in context_b field. + */ + for (i = 1; i < qman_fq_lookup_table_size; i++) { + if (qman_fq_lookup_table[i] == NULL) { + *entry = i; + qman_fq_lookup_table[i] = fq; + spin_unlock(&fq_hash_table_lock); + return 0; + } + } + spin_unlock(&fq_hash_table_lock); + return -ENOMEM; +} + +static void clear_fq_table_entry(u32 entry) +{ + spin_lock(&fq_hash_table_lock); + DPAA_BUG_ON(entry >= qman_fq_lookup_table_size); + qman_fq_lookup_table[entry] = NULL; + spin_unlock(&fq_hash_table_lock); +} + +static inline struct qman_fq *get_fq_table_entry(u32 entry) +{ + DPAA_BUG_ON(entry >= qman_fq_lookup_table_size); + return qman_fq_lookup_table[entry]; +} +#endif + +static inline void cpu_to_hw_fqd(struct qm_fqd *fqd) +{ + /* Byteswap the FQD to HW format */ + fqd->fq_ctrl = cpu_to_be16(fqd->fq_ctrl); + fqd->dest_wq = cpu_to_be16(fqd->dest_wq); + fqd->ics_cred = cpu_to_be16(fqd->ics_cred); + fqd->context_b = cpu_to_be32(fqd->context_b); + fqd->context_a.opaque = cpu_to_be64(fqd->context_a.opaque); + fqd->opaque_td = cpu_to_be16(fqd->opaque_td); +} + +static inline void hw_fqd_to_cpu(struct qm_fqd *fqd) +{ + /* Byteswap the FQD to CPU format */ + fqd->fq_ctrl = be16_to_cpu(fqd->fq_ctrl); + fqd->dest_wq = be16_to_cpu(fqd->dest_wq); + fqd->ics_cred = be16_to_cpu(fqd->ics_cred); + fqd->context_b = be32_to_cpu(fqd->context_b); + fqd->context_a.opaque = be64_to_cpu(fqd->context_a.opaque); +} + +static inline void cpu_to_hw_fd(struct qm_fd *fd) +{ + fd->addr = cpu_to_be40(fd->addr); + fd->status = cpu_to_be32(fd->status); + fd->opaque = cpu_to_be32(fd->opaque); +} + +static inline void hw_fd_to_cpu(struct qm_fd *fd) +{ + fd->addr = be40_to_cpu(fd->addr); + fd->status = be32_to_cpu(fd->status); + fd->opaque = be32_to_cpu(fd->opaque); +} + +/* In the case that slow- and fast-path handling are both done by qman_poll() + * (ie. because there is no interrupt handling), we ought to balance how often + * we do the fast-path poll versus the slow-path poll. We'll use two decrementer + * sources, so we call the fast poll 'n' times before calling the slow poll + * once. The idle decrementer constant is used when the last slow-poll detected + * no work to do, and the busy decrementer constant when the last slow-poll had + * work to do. + */ +#define SLOW_POLL_IDLE 1000 +#define SLOW_POLL_BUSY 10 +static u32 __poll_portal_slow(struct qman_portal *p, u32 is); +static inline unsigned int __poll_portal_fast(struct qman_portal *p, + unsigned int poll_limit); + +/* Portal interrupt handler */ +static irqreturn_t portal_isr(__always_unused int irq, void *ptr) +{ + struct qman_portal *p = ptr; + /* + * The CSCI/CCSCI source is cleared inside __poll_portal_slow(), because + * it could race against a Query Congestion State command also given + * as part of the handling of this interrupt source. We mustn't + * clear it a second time in this top-level function. + */ + u32 clear = QM_DQAVAIL_MASK | (p->irq_sources & + ~(QM_PIRQ_CSCI | QM_PIRQ_CCSCI)); + u32 is = qm_isr_status_read(&p->p) & p->irq_sources; + /* DQRR-handling if it's interrupt-driven */ + if (is & QM_PIRQ_DQRI) + __poll_portal_fast(p, FSL_QMAN_POLL_LIMIT); + /* Handling of anything else that's interrupt-driven */ + clear |= __poll_portal_slow(p, is); + qm_isr_status_clear(&p->p, clear); + return IRQ_HANDLED; +} + +/* This inner version is used privately by qman_create_affine_portal(), as well + * as by the exported qman_stop_dequeues(). + */ +static inline void qman_stop_dequeues_ex(struct qman_portal *p) +{ + if (!(p->dqrr_disable_ref++)) + qm_dqrr_set_maxfill(&p->p, 0); +} + +static int drain_mr_fqrni(struct qm_portal *p) +{ + const struct qm_mr_entry *msg; +loop: + msg = qm_mr_current(p); + if (!msg) { + /* + * if MR was full and h/w had other FQRNI entries to produce, we + * need to allow it time to produce those entries once the + * existing entries are consumed. A worst-case situation + * (fully-loaded system) means h/w sequencers may have to do 3-4 + * other things before servicing the portal's MR pump, each of + * which (if slow) may take ~50 qman cycles (which is ~200 + * processor cycles). So rounding up and then multiplying this + * worst-case estimate by a factor of 10, just to be + * ultra-paranoid, goes as high as 10,000 cycles. NB, we consume + * one entry at a time, so h/w has an opportunity to produce new + * entries well before the ring has been fully consumed, so + * we're being *really* paranoid here. + */ + u64 now, then = mfatb(); + + do { + now = mfatb(); + } while ((then + 10000) > now); + msg = qm_mr_current(p); + if (!msg) + return 0; + } + if ((msg->verb & QM_MR_VERB_TYPE_MASK) != QM_MR_VERB_FQRNI) { + /* We aren't draining anything but FQRNIs */ + pr_err("Found verb 0x%x in MR\n", msg->verb); + return -1; + } + qm_mr_next(p); + qm_mr_cci_consume(p, 1); + goto loop; +} + +static inline int qm_eqcr_init(struct qm_portal *portal, + enum qm_eqcr_pmode pmode, + unsigned int eq_stash_thresh, + int eq_stash_prio) +{ + /* This use of 'register', as well as all other occurrences, is because + * it has been observed to generate much faster code with gcc than is + * otherwise the case. + */ + register struct qm_eqcr *eqcr = &portal->eqcr; + u32 cfg; + u8 pi; + + eqcr->ring = portal->addr.ce + QM_CL_EQCR; + eqcr->ci = qm_in(EQCR_CI_CINH) & (QM_EQCR_SIZE - 1); + qm_cl_invalidate(EQCR_CI); + pi = qm_in(EQCR_PI_CINH) & (QM_EQCR_SIZE - 1); + eqcr->cursor = eqcr->ring + pi; + eqcr->vbit = (qm_in(EQCR_PI_CINH) & QM_EQCR_SIZE) ? + QM_EQCR_VERB_VBIT : 0; + eqcr->available = QM_EQCR_SIZE - 1 - + qm_cyc_diff(QM_EQCR_SIZE, eqcr->ci, pi); + eqcr->ithresh = qm_in(EQCR_ITR); +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + eqcr->busy = 0; + eqcr->pmode = pmode; +#endif + cfg = (qm_in(CFG) & 0x00ffffff) | + (eq_stash_thresh << 28) | /* QCSP_CFG: EST */ + (eq_stash_prio << 26) | /* QCSP_CFG: EP */ + ((pmode & 0x3) << 24); /* QCSP_CFG::EPM */ + qm_out(CFG, cfg); + return 0; +} + +static inline void qm_eqcr_finish(struct qm_portal *portal) +{ + register struct qm_eqcr *eqcr = &portal->eqcr; + u8 pi, ci; + u32 cfg; + + /* + * Disable EQCI stashing because the QMan only + * presents the value it previously stashed to + * maintain coherency. Setting the stash threshold + * to 1 then 0 ensures that QMan has resyncronized + * its internal copy so that the portal is clean + * when it is reinitialized in the future + */ + cfg = (qm_in(CFG) & 0x0fffffff) | + (1 << 28); /* QCSP_CFG: EST */ + qm_out(CFG, cfg); + cfg &= 0x0fffffff; /* stash threshold = 0 */ + qm_out(CFG, cfg); + + pi = qm_in(EQCR_PI_CINH) & (QM_EQCR_SIZE - 1); + ci = qm_in(EQCR_CI_CINH) & (QM_EQCR_SIZE - 1); + + /* Refresh EQCR CI cache value */ + qm_cl_invalidate(EQCR_CI); + eqcr->ci = qm_cl_in(EQCR_CI) & (QM_EQCR_SIZE - 1); + + DPAA_ASSERT(!eqcr->busy); + if (pi != EQCR_PTR2IDX(eqcr->cursor)) + pr_crit("losing uncommitted EQCR entries\n"); + if (ci != eqcr->ci) + pr_crit("missing existing EQCR completions\n"); + if (eqcr->ci != EQCR_PTR2IDX(eqcr->cursor)) + pr_crit("EQCR destroyed unquiesced\n"); +} + +static inline int qm_dqrr_init(struct qm_portal *portal, + __maybe_unused const struct qm_portal_config *config, + enum qm_dqrr_dmode dmode, + __maybe_unused enum qm_dqrr_pmode pmode, + enum qm_dqrr_cmode cmode, u8 max_fill) +{ + register struct qm_dqrr *dqrr = &portal->dqrr; + u32 cfg; + + /* Make sure the DQRR will be idle when we enable */ + qm_out(DQRR_SDQCR, 0); + qm_out(DQRR_VDQCR, 0); + qm_out(DQRR_PDQCR, 0); + dqrr->ring = portal->addr.ce + QM_CL_DQRR; + dqrr->pi = qm_in(DQRR_PI_CINH) & (QM_DQRR_SIZE - 1); + dqrr->ci = qm_in(DQRR_CI_CINH) & (QM_DQRR_SIZE - 1); + dqrr->cursor = dqrr->ring + dqrr->ci; + dqrr->fill = qm_cyc_diff(QM_DQRR_SIZE, dqrr->ci, dqrr->pi); + dqrr->vbit = (qm_in(DQRR_PI_CINH) & QM_DQRR_SIZE) ? + QM_DQRR_VERB_VBIT : 0; + dqrr->ithresh = qm_in(DQRR_ITR); +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + dqrr->dmode = dmode; + dqrr->pmode = pmode; + dqrr->cmode = cmode; +#endif + /* Invalidate every ring entry before beginning */ + for (cfg = 0; cfg < QM_DQRR_SIZE; cfg++) + dccivac(qm_cl(dqrr->ring, cfg)); + cfg = (qm_in(CFG) & 0xff000f00) | + ((max_fill & (QM_DQRR_SIZE - 1)) << 20) | /* DQRR_MF */ + ((dmode & 1) << 18) | /* DP */ + ((cmode & 3) << 16) | /* DCM */ + 0xa0 | /* RE+SE */ + (0 ? 0x40 : 0) | /* Ignore RP */ + (0 ? 0x10 : 0); /* Ignore SP */ + qm_out(CFG, cfg); + qm_dqrr_set_maxfill(portal, max_fill); + return 0; +} + +static inline void qm_dqrr_finish(struct qm_portal *portal) +{ + __maybe_unused register struct qm_dqrr *dqrr = &portal->dqrr; +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + if ((dqrr->cmode != qm_dqrr_cdc) && + (dqrr->ci != DQRR_PTR2IDX(dqrr->cursor))) + pr_crit("Ignoring completed DQRR entries\n"); +#endif +} + +static inline int qm_mr_init(struct qm_portal *portal, + __maybe_unused enum qm_mr_pmode pmode, + enum qm_mr_cmode cmode) +{ + register struct qm_mr *mr = &portal->mr; + u32 cfg; + + mr->ring = portal->addr.ce + QM_CL_MR; + mr->pi = qm_in(MR_PI_CINH) & (QM_MR_SIZE - 1); + mr->ci = qm_in(MR_CI_CINH) & (QM_MR_SIZE - 1); + mr->cursor = mr->ring + mr->ci; + mr->fill = qm_cyc_diff(QM_MR_SIZE, mr->ci, mr->pi); + mr->vbit = (qm_in(MR_PI_CINH) & QM_MR_SIZE) ? QM_MR_VERB_VBIT : 0; + mr->ithresh = qm_in(MR_ITR); +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + mr->pmode = pmode; + mr->cmode = cmode; +#endif + cfg = (qm_in(CFG) & 0xfffff0ff) | + ((cmode & 1) << 8); /* QCSP_CFG:MM */ + qm_out(CFG, cfg); + return 0; +} + +static inline void qm_mr_pvb_update(struct qm_portal *portal) +{ + register struct qm_mr *mr = &portal->mr; + const struct qm_mr_entry *res = qm_cl(mr->ring, mr->pi); + + DPAA_ASSERT(mr->pmode == qm_mr_pvb); + /* when accessing 'verb', use __raw_readb() to ensure that compiler + * inlining doesn't try to optimise out "excess reads". + */ + if ((__raw_readb(&res->verb) & QM_MR_VERB_VBIT) == mr->vbit) { + mr->pi = (mr->pi + 1) & (QM_MR_SIZE - 1); + if (!mr->pi) + mr->vbit ^= QM_MR_VERB_VBIT; + mr->fill++; + res = MR_INC(res); + } + dcbit_ro(res); +} + +static inline +struct qman_portal *qman_create_portal( + struct qman_portal *portal, + const struct qm_portal_config *c, + const struct qman_cgrs *cgrs) +{ + struct qm_portal *p; + char buf[16]; + int ret; + u32 isdr; + + p = &portal->p; + + portal->use_eqcr_ci_stashing = ((qman_ip_rev >= QMAN_REV30) ? 1 : 0); + /* + * prep the low-level portal struct with the mapped addresses from the + * config, everything that follows depends on it and "config" is more + * for (de)reference + */ + p->addr.ce = c->addr_virt[DPAA_PORTAL_CE]; + p->addr.ci = c->addr_virt[DPAA_PORTAL_CI]; + /* + * If CI-stashing is used, the current defaults use a threshold of 3, + * and stash with high-than-DQRR priority. + */ + if (qm_eqcr_init(p, qm_eqcr_pvb, + portal->use_eqcr_ci_stashing ? 3 : 0, 1)) { + pr_err("Qman EQCR initialisation failed\n"); + goto fail_eqcr; + } + if (qm_dqrr_init(p, c, qm_dqrr_dpush, qm_dqrr_pvb, + qm_dqrr_cdc, DQRR_MAXFILL)) { + pr_err("Qman DQRR initialisation failed\n"); + goto fail_dqrr; + } + if (qm_mr_init(p, qm_mr_pvb, qm_mr_cci)) { + pr_err("Qman MR initialisation failed\n"); + goto fail_mr; + } + if (qm_mc_init(p)) { + pr_err("Qman MC initialisation failed\n"); + goto fail_mc; + } + + /* static interrupt-gating controls */ + qm_dqrr_set_ithresh(p, 0); + qm_mr_set_ithresh(p, 0); + qm_isr_set_iperiod(p, 0); + portal->cgrs = kmalloc(2 * sizeof(*cgrs), GFP_KERNEL); + if (!portal->cgrs) + goto fail_cgrs; + /* initial snapshot is no-depletion */ + qman_cgrs_init(&portal->cgrs[1]); + if (cgrs) + portal->cgrs[0] = *cgrs; + else + /* if the given mask is NULL, assume all CGRs can be seen */ + qman_cgrs_fill(&portal->cgrs[0]); + INIT_LIST_HEAD(&portal->cgr_cbs); + spin_lock_init(&portal->cgr_lock); + portal->bits = 0; + portal->slowpoll = 0; + portal->sdqcr = QM_SDQCR_SOURCE_CHANNELS | QM_SDQCR_COUNT_UPTO3 | + QM_SDQCR_DEDICATED_PRECEDENCE | QM_SDQCR_TYPE_PRIO_QOS | + QM_SDQCR_TOKEN_SET(0xab) | QM_SDQCR_CHANNELS_DEDICATED; + portal->dqrr_disable_ref = 0; + portal->cb_dc_ern = NULL; + sprintf(buf, "qportal-%d", c->channel); + dpa_rbtree_init(&portal->retire_table); + isdr = 0xffffffff; + qm_isr_disable_write(p, isdr); + portal->irq_sources = 0; + qm_isr_enable_write(p, portal->irq_sources); + qm_isr_status_clear(p, 0xffffffff); + snprintf(portal->irqname, MAX_IRQNAME, IRQNAME, c->cpu); + if (request_irq(c->irq, portal_isr, 0, portal->irqname, + portal)) { + pr_err("request_irq() failed\n"); + goto fail_irq; + } + + /* Need EQCR to be empty before continuing */ + isdr &= ~QM_PIRQ_EQCI; + qm_isr_disable_write(p, isdr); + ret = qm_eqcr_get_fill(p); + if (ret) { + pr_err("Qman EQCR unclean\n"); + goto fail_eqcr_empty; + } + isdr &= ~(QM_PIRQ_DQRI | QM_PIRQ_MRI); + qm_isr_disable_write(p, isdr); + if (qm_dqrr_current(p)) { + pr_err("Qman DQRR unclean\n"); + qm_dqrr_cdc_consume_n(p, 0xffff); + } + if (qm_mr_current(p) && drain_mr_fqrni(p)) { + /* special handling, drain just in case it's a few FQRNIs */ + if (drain_mr_fqrni(p)) + goto fail_dqrr_mr_empty; + } + /* Success */ + portal->config = c; + qm_isr_disable_write(p, 0); + qm_isr_uninhibit(p); + /* Write a sane SDQCR */ + qm_dqrr_sdqcr_set(p, portal->sdqcr); + return portal; +fail_dqrr_mr_empty: +fail_eqcr_empty: + free_irq(c->irq, portal); +fail_irq: + kfree(portal->cgrs); + spin_lock_destroy(&portal->cgr_lock); +fail_cgrs: + qm_mc_finish(p); +fail_mc: + qm_mr_finish(p); +fail_mr: + qm_dqrr_finish(p); +fail_dqrr: + qm_eqcr_finish(p); +fail_eqcr: + return NULL; +} + +struct qman_portal *qman_create_affine_portal(const struct qm_portal_config *c, + const struct qman_cgrs *cgrs) +{ + struct qman_portal *res; + struct qman_portal *portal = get_affine_portal(); + /* A criteria for calling this function (from qman_driver.c) is that + * we're already affine to the cpu and won't schedule onto another cpu. + */ + + res = qman_create_portal(portal, c, cgrs); + if (res) { + spin_lock(&affine_mask_lock); + CPU_SET(c->cpu, &affine_mask); + affine_channels[c->cpu] = + c->channel; + spin_unlock(&affine_mask_lock); + } + return res; +} + +static inline +void qman_destroy_portal(struct qman_portal *qm) +{ + const struct qm_portal_config *pcfg; + + /* Stop dequeues on the portal */ + qm_dqrr_sdqcr_set(&qm->p, 0); + + /* + * NB we do this to "quiesce" EQCR. If we add enqueue-completions or + * something related to QM_PIRQ_EQCI, this may need fixing. + * Also, due to the prefetching model used for CI updates in the enqueue + * path, this update will only invalidate the CI cacheline *after* + * working on it, so we need to call this twice to ensure a full update + * irrespective of where the enqueue processing was at when the teardown + * began. + */ + qm_eqcr_cce_update(&qm->p); + qm_eqcr_cce_update(&qm->p); + pcfg = qm->config; + + free_irq(pcfg->irq, qm); + + kfree(qm->cgrs); + qm_mc_finish(&qm->p); + qm_mr_finish(&qm->p); + qm_dqrr_finish(&qm->p); + qm_eqcr_finish(&qm->p); + + qm->config = NULL; + + spin_lock_destroy(&qm->cgr_lock); +} + +const struct qm_portal_config *qman_destroy_affine_portal(void) +{ + /* We don't want to redirect if we're a slave, use "raw" */ + struct qman_portal *qm = get_affine_portal(); + const struct qm_portal_config *pcfg; + int cpu; + + pcfg = qm->config; + cpu = pcfg->cpu; + + qman_destroy_portal(qm); + + spin_lock(&affine_mask_lock); + CPU_CLR(cpu, &affine_mask); + spin_unlock(&affine_mask_lock); + return pcfg; +} + +int qman_get_portal_index(void) +{ + struct qman_portal *p = get_affine_portal(); + return p->config->index; +} + +/* Inline helper to reduce nesting in __poll_portal_slow() */ +static inline void fq_state_change(struct qman_portal *p, struct qman_fq *fq, + const struct qm_mr_entry *msg, u8 verb) +{ + FQLOCK(fq); + switch (verb) { + case QM_MR_VERB_FQRL: + DPAA_ASSERT(fq_isset(fq, QMAN_FQ_STATE_ORL)); + fq_clear(fq, QMAN_FQ_STATE_ORL); + table_del_fq(p, fq); + break; + case QM_MR_VERB_FQRN: + DPAA_ASSERT((fq->state == qman_fq_state_parked) || + (fq->state == qman_fq_state_sched)); + DPAA_ASSERT(fq_isset(fq, QMAN_FQ_STATE_CHANGING)); + fq_clear(fq, QMAN_FQ_STATE_CHANGING); + if (msg->fq.fqs & QM_MR_FQS_NOTEMPTY) + fq_set(fq, QMAN_FQ_STATE_NE); + if (msg->fq.fqs & QM_MR_FQS_ORLPRESENT) + fq_set(fq, QMAN_FQ_STATE_ORL); + else + table_del_fq(p, fq); + fq->state = qman_fq_state_retired; + break; + case QM_MR_VERB_FQPN: + DPAA_ASSERT(fq->state == qman_fq_state_sched); + DPAA_ASSERT(fq_isclear(fq, QMAN_FQ_STATE_CHANGING)); + fq->state = qman_fq_state_parked; + } + FQUNLOCK(fq); +} + +static u32 __poll_portal_slow(struct qman_portal *p, u32 is) +{ + const struct qm_mr_entry *msg; + struct qm_mr_entry swapped_msg; + + if (is & QM_PIRQ_CSCI) { + struct qman_cgrs rr, c; + struct qm_mc_result *mcr; + struct qman_cgr *cgr; + + spin_lock(&p->cgr_lock); + /* + * The CSCI bit must be cleared _before_ issuing the + * Query Congestion State command, to ensure that a long + * CGR State Change callback cannot miss an intervening + * state change. + */ + qm_isr_status_clear(&p->p, QM_PIRQ_CSCI); + qm_mc_start(&p->p); + qm_mc_commit(&p->p, QM_MCC_VERB_QUERYCONGESTION); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + /* mask out the ones I'm not interested in */ + qman_cgrs_and(&rr, (const struct qman_cgrs *) + &mcr->querycongestion.state, &p->cgrs[0]); + /* check previous snapshot for delta, enter/exit congestion */ + qman_cgrs_xor(&c, &rr, &p->cgrs[1]); + /* update snapshot */ + qman_cgrs_cp(&p->cgrs[1], &rr); + /* Invoke callback */ + list_for_each_entry(cgr, &p->cgr_cbs, node) + if (cgr->cb && qman_cgrs_get(&c, cgr->cgrid)) + cgr->cb(p, cgr, qman_cgrs_get(&rr, cgr->cgrid)); + spin_unlock(&p->cgr_lock); + } + + if (is & QM_PIRQ_EQRI) { + qm_eqcr_cce_update(&p->p); + qm_eqcr_set_ithresh(&p->p, 0); + wake_up(&affine_queue); + } + + if (is & QM_PIRQ_MRI) { + struct qman_fq *fq; + u8 verb, num = 0; +mr_loop: + qm_mr_pvb_update(&p->p); + msg = qm_mr_current(&p->p); + if (!msg) + goto mr_done; + swapped_msg = *msg; + hw_fd_to_cpu(&swapped_msg.ern.fd); + verb = msg->verb & QM_MR_VERB_TYPE_MASK; + /* The message is a software ERN iff the 0x20 bit is set */ + if (verb & 0x20) { + switch (verb) { + case QM_MR_VERB_FQRNI: + /* nada, we drop FQRNIs on the floor */ + break; + case QM_MR_VERB_FQRN: + case QM_MR_VERB_FQRL: + /* Lookup in the retirement table */ + fq = table_find_fq(p, + be32_to_cpu(msg->fq.fqid)); + DPAA_BUG_ON(!fq); + fq_state_change(p, fq, &swapped_msg, verb); + if (fq->cb.fqs) + fq->cb.fqs(p, fq, &swapped_msg); + break; + case QM_MR_VERB_FQPN: + /* Parked */ +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP + fq = get_fq_table_entry( + be32_to_cpu(msg->fq.contextB)); +#else + fq = (void *)(uintptr_t) + be32_to_cpu(msg->fq.contextB); +#endif + fq_state_change(p, fq, msg, verb); + if (fq->cb.fqs) + fq->cb.fqs(p, fq, &swapped_msg); + break; + case QM_MR_VERB_DC_ERN: + /* DCP ERN */ + if (p->cb_dc_ern) + p->cb_dc_ern(p, msg); + else if (cb_dc_ern) + cb_dc_ern(p, msg); + else { + static int warn_once; + + if (!warn_once) { + pr_crit("Leaking DCP ERNs!\n"); + warn_once = 1; + } + } + break; + default: + pr_crit("Invalid MR verb 0x%02x\n", verb); + } + } else { + /* Its a software ERN */ +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP + fq = get_fq_table_entry(be32_to_cpu(msg->ern.tag)); +#else + fq = (void *)(uintptr_t)be32_to_cpu(msg->ern.tag); +#endif + fq->cb.ern(p, fq, &swapped_msg); + } + num++; + qm_mr_next(&p->p); + goto mr_loop; +mr_done: + qm_mr_cci_consume(&p->p, num); + } + /* + * QM_PIRQ_CSCI/CCSCI has already been cleared, as part of its specific + * processing. If that interrupt source has meanwhile been re-asserted, + * we mustn't clear it here (or in the top-level interrupt handler). + */ + return is & (QM_PIRQ_EQCI | QM_PIRQ_EQRI | QM_PIRQ_MRI); +} + +/* + * remove some slowish-path stuff from the "fast path" and make sure it isn't + * inlined. + */ +static noinline void clear_vdqcr(struct qman_portal *p, struct qman_fq *fq) +{ + p->vdqcr_owned = NULL; + FQLOCK(fq); + fq_clear(fq, QMAN_FQ_STATE_VDQCR); + FQUNLOCK(fq); + wake_up(&affine_queue); +} + +/* + * The only states that would conflict with other things if they ran at the + * same time on the same cpu are: + * + * (i) setting/clearing vdqcr_owned, and + * (ii) clearing the NE (Not Empty) flag. + * + * Both are safe. Because; + * + * (i) this clearing can only occur after qman_set_vdq() has set the + * vdqcr_owned field (which it does before setting VDQCR), and + * qman_volatile_dequeue() blocks interrupts and preemption while this is + * done so that we can't interfere. + * (ii) the NE flag is only cleared after qman_retire_fq() has set it, and as + * with (i) that API prevents us from interfering until it's safe. + * + * The good thing is that qman_set_vdq() and qman_retire_fq() run far + * less frequently (ie. per-FQ) than __poll_portal_fast() does, so the nett + * advantage comes from this function not having to "lock" anything at all. + * + * Note also that the callbacks are invoked at points which are safe against the + * above potential conflicts, but that this function itself is not re-entrant + * (this is because the function tracks one end of each FIFO in the portal and + * we do *not* want to lock that). So the consequence is that it is safe for + * user callbacks to call into any QMan API. + */ +static inline unsigned int __poll_portal_fast(struct qman_portal *p, + unsigned int poll_limit) +{ + const struct qm_dqrr_entry *dq; + struct qman_fq *fq; + enum qman_cb_dqrr_result res; + unsigned int limit = 0; +#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ + struct qm_dqrr_entry *shadow; +#endif + do { + qm_dqrr_pvb_update(&p->p); + dq = qm_dqrr_current(&p->p); + if (!dq) + break; +#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ + /* If running on an LE system the fields of the + * dequeue entry must be swapper. Because the + * QMan HW will ignore writes the DQRR entry is + * copied and the index stored within the copy + */ + shadow = &p->shadow_dqrr[DQRR_PTR2IDX(dq)]; + *shadow = *dq; + dq = shadow; + shadow->fqid = be32_to_cpu(shadow->fqid); + shadow->contextB = be32_to_cpu(shadow->contextB); + shadow->seqnum = be16_to_cpu(shadow->seqnum); + hw_fd_to_cpu(&shadow->fd); +#endif + + if (dq->stat & QM_DQRR_STAT_UNSCHEDULED) { + /* + * VDQCR: don't trust context_b as the FQ may have + * been configured for h/w consumption and we're + * draining it post-retirement. + */ + fq = p->vdqcr_owned; + /* + * We only set QMAN_FQ_STATE_NE when retiring, so we + * only need to check for clearing it when doing + * volatile dequeues. It's one less thing to check + * in the critical path (SDQCR). + */ + if (dq->stat & QM_DQRR_STAT_FQ_EMPTY) + fq_clear(fq, QMAN_FQ_STATE_NE); + /* + * This is duplicated from the SDQCR code, but we + * have stuff to do before *and* after this callback, + * and we don't want multiple if()s in the critical + * path (SDQCR). + */ + res = fq->cb.dqrr(p, fq, dq); + if (res == qman_cb_dqrr_stop) + break; + /* Check for VDQCR completion */ + if (dq->stat & QM_DQRR_STAT_DQCR_EXPIRED) + clear_vdqcr(p, fq); + } else { + /* SDQCR: context_b points to the FQ */ +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP + fq = get_fq_table_entry(dq->contextB); +#else + fq = (void *)(uintptr_t)dq->contextB; +#endif + /* Now let the callback do its stuff */ + res = fq->cb.dqrr(p, fq, dq); + /* + * The callback can request that we exit without + * consuming this entry nor advancing; + */ + if (res == qman_cb_dqrr_stop) + break; + } + /* Interpret 'dq' from a driver perspective. */ + /* + * Parking isn't possible unless HELDACTIVE was set. NB, + * FORCEELIGIBLE implies HELDACTIVE, so we only need to + * check for HELDACTIVE to cover both. + */ + DPAA_ASSERT((dq->stat & QM_DQRR_STAT_FQ_HELDACTIVE) || + (res != qman_cb_dqrr_park)); + /* just means "skip it, I'll consume it myself later on" */ + if (res != qman_cb_dqrr_defer) + qm_dqrr_cdc_consume_1ptr(&p->p, dq, + res == qman_cb_dqrr_park); + /* Move forward */ + qm_dqrr_next(&p->p); + /* + * Entry processed and consumed, increment our counter. The + * callback can request that we exit after consuming the + * entry, and we also exit if we reach our processing limit, + * so loop back only if neither of these conditions is met. + */ + } while (++limit < poll_limit && res != qman_cb_dqrr_consume_stop); + + return limit; +} + +u16 qman_affine_channel(int cpu) +{ + if (cpu < 0) { + struct qman_portal *portal = get_affine_portal(); + + cpu = portal->config->cpu; + } + DPAA_BUG_ON(!CPU_ISSET(cpu, &affine_mask)); + return affine_channels[cpu]; +} + +struct qm_dqrr_entry *qman_dequeue(struct qman_fq *fq) +{ + struct qman_portal *p = get_affine_portal(); + const struct qm_dqrr_entry *dq; +#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ + struct qm_dqrr_entry *shadow; +#endif + + qm_dqrr_pvb_update(&p->p); + dq = qm_dqrr_current(&p->p); + if (!dq) + return NULL; + + if (!(dq->stat & QM_DQRR_STAT_FD_VALID)) { + /* Invalid DQRR - put the portal and consume the DQRR. + * Return NULL to user as no packet is seen. + */ + qman_dqrr_consume(fq, (struct qm_dqrr_entry *)dq); + return NULL; + } + +#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ + shadow = &p->shadow_dqrr[DQRR_PTR2IDX(dq)]; + *shadow = *dq; + dq = shadow; + shadow->fqid = be32_to_cpu(shadow->fqid); + shadow->contextB = be32_to_cpu(shadow->contextB); + shadow->seqnum = be16_to_cpu(shadow->seqnum); + hw_fd_to_cpu(&shadow->fd); +#endif + + if (dq->stat & QM_DQRR_STAT_FQ_EMPTY) + fq_clear(fq, QMAN_FQ_STATE_NE); + + return (struct qm_dqrr_entry *)dq; +} + +void qman_dqrr_consume(struct qman_fq *fq, + struct qm_dqrr_entry *dq) +{ + struct qman_portal *p = get_affine_portal(); + + if (dq->stat & QM_DQRR_STAT_DQCR_EXPIRED) + clear_vdqcr(p, fq); + + qm_dqrr_cdc_consume_1ptr(&p->p, dq, 0); + qm_dqrr_next(&p->p); +} + +int qman_poll_dqrr(unsigned int limit) +{ + struct qman_portal *p = get_affine_portal(); + int ret; + + ret = __poll_portal_fast(p, limit); + return ret; +} + +void qman_poll(void) +{ + struct qman_portal *p = get_affine_portal(); + + if ((~p->irq_sources) & QM_PIRQ_SLOW) { + if (!(p->slowpoll--)) { + u32 is = qm_isr_status_read(&p->p) & ~p->irq_sources; + u32 active = __poll_portal_slow(p, is); + + if (active) { + qm_isr_status_clear(&p->p, active); + p->slowpoll = SLOW_POLL_BUSY; + } else + p->slowpoll = SLOW_POLL_IDLE; + } + } + if ((~p->irq_sources) & QM_PIRQ_DQRI) + __poll_portal_fast(p, FSL_QMAN_POLL_LIMIT); +} + +void qman_stop_dequeues(void) +{ + struct qman_portal *p = get_affine_portal(); + + qman_stop_dequeues_ex(p); +} + +void qman_start_dequeues(void) +{ + struct qman_portal *p = get_affine_portal(); + + DPAA_ASSERT(p->dqrr_disable_ref > 0); + if (!(--p->dqrr_disable_ref)) + qm_dqrr_set_maxfill(&p->p, DQRR_MAXFILL); +} + +void qman_static_dequeue_add(u32 pools) +{ + struct qman_portal *p = get_affine_portal(); + + pools &= p->config->pools; + p->sdqcr |= pools; + qm_dqrr_sdqcr_set(&p->p, p->sdqcr); +} + +void qman_static_dequeue_del(u32 pools) +{ + struct qman_portal *p = get_affine_portal(); + + pools &= p->config->pools; + p->sdqcr &= ~pools; + qm_dqrr_sdqcr_set(&p->p, p->sdqcr); +} + +u32 qman_static_dequeue_get(void) +{ + struct qman_portal *p = get_affine_portal(); + return p->sdqcr; +} + +void qman_dca(struct qm_dqrr_entry *dq, int park_request) +{ + struct qman_portal *p = get_affine_portal(); + + qm_dqrr_cdc_consume_1ptr(&p->p, dq, park_request); +} + +/* Frame queue API */ +static const char *mcr_result_str(u8 result) +{ + switch (result) { + case QM_MCR_RESULT_NULL: + return "QM_MCR_RESULT_NULL"; + case QM_MCR_RESULT_OK: + return "QM_MCR_RESULT_OK"; + case QM_MCR_RESULT_ERR_FQID: + return "QM_MCR_RESULT_ERR_FQID"; + case QM_MCR_RESULT_ERR_FQSTATE: + return "QM_MCR_RESULT_ERR_FQSTATE"; + case QM_MCR_RESULT_ERR_NOTEMPTY: + return "QM_MCR_RESULT_ERR_NOTEMPTY"; + case QM_MCR_RESULT_PENDING: + return "QM_MCR_RESULT_PENDING"; + case QM_MCR_RESULT_ERR_BADCOMMAND: + return "QM_MCR_RESULT_ERR_BADCOMMAND"; + } + return "<unknown MCR result>"; +} + +int qman_create_fq(u32 fqid, u32 flags, struct qman_fq *fq) +{ + struct qm_fqd fqd; + struct qm_mcr_queryfq_np np; + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p; + + if (flags & QMAN_FQ_FLAG_DYNAMIC_FQID) { + int ret = qman_alloc_fqid(&fqid); + + if (ret) + return ret; + } + spin_lock_init(&fq->fqlock); + fq->fqid = fqid; + fq->flags = flags; + fq->state = qman_fq_state_oos; + fq->cgr_groupid = 0; +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP + if (unlikely(find_empty_fq_table_entry(&fq->key, fq))) { + pr_info("Find empty table entry failed\n"); + return -ENOMEM; + } +#endif + if (!(flags & QMAN_FQ_FLAG_AS_IS) || (flags & QMAN_FQ_FLAG_NO_MODIFY)) + return 0; + /* Everything else is AS_IS support */ + p = get_affine_portal(); + mcc = qm_mc_start(&p->p); + mcc->queryfq.fqid = cpu_to_be32(fqid); + qm_mc_commit(&p->p, QM_MCC_VERB_QUERYFQ); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPAA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCC_VERB_QUERYFQ); + if (mcr->result != QM_MCR_RESULT_OK) { + pr_err("QUERYFQ failed: %s\n", mcr_result_str(mcr->result)); + goto err; + } + fqd = mcr->queryfq.fqd; + hw_fqd_to_cpu(&fqd); + mcc = qm_mc_start(&p->p); + mcc->queryfq_np.fqid = cpu_to_be32(fqid); + qm_mc_commit(&p->p, QM_MCC_VERB_QUERYFQ_NP); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPAA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCC_VERB_QUERYFQ_NP); + if (mcr->result != QM_MCR_RESULT_OK) { + pr_err("QUERYFQ_NP failed: %s\n", mcr_result_str(mcr->result)); + goto err; + } + np = mcr->queryfq_np; + /* Phew, have queryfq and queryfq_np results, stitch together + * the FQ object from those. + */ + fq->cgr_groupid = fqd.cgid; + switch (np.state & QM_MCR_NP_STATE_MASK) { + case QM_MCR_NP_STATE_OOS: + break; + case QM_MCR_NP_STATE_RETIRED: + fq->state = qman_fq_state_retired; + if (np.frm_cnt) + fq_set(fq, QMAN_FQ_STATE_NE); + break; + case QM_MCR_NP_STATE_TEN_SCHED: + case QM_MCR_NP_STATE_TRU_SCHED: + case QM_MCR_NP_STATE_ACTIVE: + fq->state = qman_fq_state_sched; + if (np.state & QM_MCR_NP_STATE_R) + fq_set(fq, QMAN_FQ_STATE_CHANGING); + break; + case QM_MCR_NP_STATE_PARKED: + fq->state = qman_fq_state_parked; + break; + default: + DPAA_ASSERT(NULL == "invalid FQ state"); + } + if (fqd.fq_ctrl & QM_FQCTRL_CGE) + fq->state |= QMAN_FQ_STATE_CGR_EN; + return 0; +err: + if (flags & QMAN_FQ_FLAG_DYNAMIC_FQID) + qman_release_fqid(fqid); + return -EIO; +} + +void qman_destroy_fq(struct qman_fq *fq, u32 flags __maybe_unused) +{ + /* + * We don't need to lock the FQ as it is a pre-condition that the FQ be + * quiesced. Instead, run some checks. + */ + switch (fq->state) { + case qman_fq_state_parked: + DPAA_ASSERT(flags & QMAN_FQ_DESTROY_PARKED); + case qman_fq_state_oos: + if (fq_isset(fq, QMAN_FQ_FLAG_DYNAMIC_FQID)) + qman_release_fqid(fq->fqid); +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP + clear_fq_table_entry(fq->key); +#endif + return; + default: + break; + } + DPAA_ASSERT(NULL == "qman_free_fq() on unquiesced FQ!"); +} + +u32 qman_fq_fqid(struct qman_fq *fq) +{ + return fq->fqid; +} + +void qman_fq_state(struct qman_fq *fq, enum qman_fq_state *state, u32 *flags) +{ + if (state) + *state = fq->state; + if (flags) + *flags = fq->flags; +} + +int qman_init_fq(struct qman_fq *fq, u32 flags, struct qm_mcc_initfq *opts) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p; + + u8 res, myverb = (flags & QMAN_INITFQ_FLAG_SCHED) ? + QM_MCC_VERB_INITFQ_SCHED : QM_MCC_VERB_INITFQ_PARKED; + + if ((fq->state != qman_fq_state_oos) && + (fq->state != qman_fq_state_parked)) + return -EINVAL; +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + if (unlikely(fq_isset(fq, QMAN_FQ_FLAG_NO_MODIFY))) + return -EINVAL; +#endif + if (opts && (opts->we_mask & QM_INITFQ_WE_OAC)) { + /* And can't be set at the same time as TDTHRESH */ + if (opts->we_mask & QM_INITFQ_WE_TDTHRESH) + return -EINVAL; + } + /* Issue an INITFQ_[PARKED|SCHED] management command */ + p = get_affine_portal(); + FQLOCK(fq); + if (unlikely((fq_isset(fq, QMAN_FQ_STATE_CHANGING)) || + ((fq->state != qman_fq_state_oos) && + (fq->state != qman_fq_state_parked)))) { + FQUNLOCK(fq); + return -EBUSY; + } + mcc = qm_mc_start(&p->p); + if (opts) + mcc->initfq = *opts; + mcc->initfq.fqid = cpu_to_be32(fq->fqid); + mcc->initfq.count = 0; + /* + * If the FQ does *not* have the TO_DCPORTAL flag, context_b is set as a + * demux pointer. Otherwise, the caller-provided value is allowed to + * stand, don't overwrite it. + */ + if (fq_isclear(fq, QMAN_FQ_FLAG_TO_DCPORTAL)) { + dma_addr_t phys_fq; + + mcc->initfq.we_mask |= QM_INITFQ_WE_CONTEXTB; +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP + mcc->initfq.fqd.context_b = fq->key; +#else + mcc->initfq.fqd.context_b = (u32)(uintptr_t)fq; +#endif + /* + * and the physical address - NB, if the user wasn't trying to + * set CONTEXTA, clear the stashing settings. + */ + if (!(mcc->initfq.we_mask & QM_INITFQ_WE_CONTEXTA)) { + mcc->initfq.we_mask |= QM_INITFQ_WE_CONTEXTA; + memset(&mcc->initfq.fqd.context_a, 0, + sizeof(mcc->initfq.fqd.context_a)); + } else { + phys_fq = rte_mem_virt2iova(fq); + qm_fqd_stashing_set64(&mcc->initfq.fqd, phys_fq); + } + } + if (flags & QMAN_INITFQ_FLAG_LOCAL) { + mcc->initfq.fqd.dest.channel = p->config->channel; + if (!(mcc->initfq.we_mask & QM_INITFQ_WE_DESTWQ)) { + mcc->initfq.we_mask |= QM_INITFQ_WE_DESTWQ; + mcc->initfq.fqd.dest.wq = 4; + } + } + mcc->initfq.we_mask = cpu_to_be16(mcc->initfq.we_mask); + cpu_to_hw_fqd(&mcc->initfq.fqd); + qm_mc_commit(&p->p, myverb); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPAA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == myverb); + res = mcr->result; + if (res != QM_MCR_RESULT_OK) { + FQUNLOCK(fq); + return -EIO; + } + if (opts) { + if (opts->we_mask & QM_INITFQ_WE_FQCTRL) { + if (opts->fqd.fq_ctrl & QM_FQCTRL_CGE) + fq_set(fq, QMAN_FQ_STATE_CGR_EN); + else + fq_clear(fq, QMAN_FQ_STATE_CGR_EN); + } + if (opts->we_mask & QM_INITFQ_WE_CGID) + fq->cgr_groupid = opts->fqd.cgid; + } + fq->state = (flags & QMAN_INITFQ_FLAG_SCHED) ? + qman_fq_state_sched : qman_fq_state_parked; + FQUNLOCK(fq); + return 0; +} + +int qman_schedule_fq(struct qman_fq *fq) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p; + + int ret = 0; + u8 res; + + if (fq->state != qman_fq_state_parked) + return -EINVAL; +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + if (unlikely(fq_isset(fq, QMAN_FQ_FLAG_NO_MODIFY))) + return -EINVAL; +#endif + /* Issue a ALTERFQ_SCHED management command */ + p = get_affine_portal(); + + FQLOCK(fq); + if (unlikely((fq_isset(fq, QMAN_FQ_STATE_CHANGING)) || + (fq->state != qman_fq_state_parked))) { + ret = -EBUSY; + goto out; + } + mcc = qm_mc_start(&p->p); + mcc->alterfq.fqid = cpu_to_be32(fq->fqid); + qm_mc_commit(&p->p, QM_MCC_VERB_ALTER_SCHED); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPAA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCR_VERB_ALTER_SCHED); + res = mcr->result; + if (res != QM_MCR_RESULT_OK) { + ret = -EIO; + goto out; + } + fq->state = qman_fq_state_sched; +out: + FQUNLOCK(fq); + + return ret; +} + +int qman_retire_fq(struct qman_fq *fq, u32 *flags) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p; + + int rval; + u8 res; + + if ((fq->state != qman_fq_state_parked) && + (fq->state != qman_fq_state_sched)) + return -EINVAL; +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + if (unlikely(fq_isset(fq, QMAN_FQ_FLAG_NO_MODIFY))) + return -EINVAL; +#endif + p = get_affine_portal(); + + FQLOCK(fq); + if (unlikely((fq_isset(fq, QMAN_FQ_STATE_CHANGING)) || + (fq->state == qman_fq_state_retired) || + (fq->state == qman_fq_state_oos))) { + rval = -EBUSY; + goto out; + } + rval = table_push_fq(p, fq); + if (rval) + goto out; + mcc = qm_mc_start(&p->p); + mcc->alterfq.fqid = cpu_to_be32(fq->fqid); + qm_mc_commit(&p->p, QM_MCC_VERB_ALTER_RETIRE); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPAA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCR_VERB_ALTER_RETIRE); + res = mcr->result; + /* + * "Elegant" would be to treat OK/PENDING the same way; set CHANGING, + * and defer the flags until FQRNI or FQRN (respectively) show up. But + * "Friendly" is to process OK immediately, and not set CHANGING. We do + * friendly, otherwise the caller doesn't necessarily have a fully + * "retired" FQ on return even if the retirement was immediate. However + * this does mean some code duplication between here and + * fq_state_change(). + */ + if (likely(res == QM_MCR_RESULT_OK)) { + rval = 0; + /* Process 'fq' right away, we'll ignore FQRNI */ + if (mcr->alterfq.fqs & QM_MCR_FQS_NOTEMPTY) + fq_set(fq, QMAN_FQ_STATE_NE); + if (mcr->alterfq.fqs & QM_MCR_FQS_ORLPRESENT) + fq_set(fq, QMAN_FQ_STATE_ORL); + else + table_del_fq(p, fq); + if (flags) + *flags = fq->flags; + fq->state = qman_fq_state_retired; + if (fq->cb.fqs) { + /* + * Another issue with supporting "immediate" retirement + * is that we're forced to drop FQRNIs, because by the + * time they're seen it may already be "too late" (the + * fq may have been OOS'd and free()'d already). But if + * the upper layer wants a callback whether it's + * immediate or not, we have to fake a "MR" entry to + * look like an FQRNI... + */ + struct qm_mr_entry msg; + + msg.verb = QM_MR_VERB_FQRNI; + msg.fq.fqs = mcr->alterfq.fqs; + msg.fq.fqid = fq->fqid; +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP + msg.fq.contextB = fq->key; +#else + msg.fq.contextB = (u32)(uintptr_t)fq; +#endif + fq->cb.fqs(p, fq, &msg); + } + } else if (res == QM_MCR_RESULT_PENDING) { + rval = 1; + fq_set(fq, QMAN_FQ_STATE_CHANGING); + } else { + rval = -EIO; + table_del_fq(p, fq); + } +out: + FQUNLOCK(fq); + return rval; +} + +int qman_oos_fq(struct qman_fq *fq) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p; + + int ret = 0; + u8 res; + + if (fq->state != qman_fq_state_retired) + return -EINVAL; +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + if (unlikely(fq_isset(fq, QMAN_FQ_FLAG_NO_MODIFY))) + return -EINVAL; +#endif + p = get_affine_portal(); + FQLOCK(fq); + if (unlikely((fq_isset(fq, QMAN_FQ_STATE_BLOCKOOS)) || + (fq->state != qman_fq_state_retired))) { + ret = -EBUSY; + goto out; + } + mcc = qm_mc_start(&p->p); + mcc->alterfq.fqid = cpu_to_be32(fq->fqid); + qm_mc_commit(&p->p, QM_MCC_VERB_ALTER_OOS); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPAA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCR_VERB_ALTER_OOS); + res = mcr->result; + if (res != QM_MCR_RESULT_OK) { + ret = -EIO; + goto out; + } + fq->state = qman_fq_state_oos; +out: + FQUNLOCK(fq); + return ret; +} + +int qman_fq_flow_control(struct qman_fq *fq, int xon) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p; + + int ret = 0; + u8 res; + u8 myverb; + + if ((fq->state == qman_fq_state_oos) || + (fq->state == qman_fq_state_retired) || + (fq->state == qman_fq_state_parked)) + return -EINVAL; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + if (unlikely(fq_isset(fq, QMAN_FQ_FLAG_NO_MODIFY))) + return -EINVAL; +#endif + /* Issue a ALTER_FQXON or ALTER_FQXOFF management command */ + p = get_affine_portal(); + FQLOCK(fq); + if (unlikely((fq_isset(fq, QMAN_FQ_STATE_CHANGING)) || + (fq->state == qman_fq_state_parked) || + (fq->state == qman_fq_state_oos) || + (fq->state == qman_fq_state_retired))) { + ret = -EBUSY; + goto out; + } + mcc = qm_mc_start(&p->p); + mcc->alterfq.fqid = fq->fqid; + mcc->alterfq.count = 0; + myverb = xon ? QM_MCC_VERB_ALTER_FQXON : QM_MCC_VERB_ALTER_FQXOFF; + + qm_mc_commit(&p->p, myverb); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPAA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == myverb); + + res = mcr->result; + if (res != QM_MCR_RESULT_OK) { + ret = -EIO; + goto out; + } +out: + FQUNLOCK(fq); + return ret; +} + +int qman_query_fq(struct qman_fq *fq, struct qm_fqd *fqd) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p = get_affine_portal(); + + u8 res; + + mcc = qm_mc_start(&p->p); + mcc->queryfq.fqid = cpu_to_be32(fq->fqid); + qm_mc_commit(&p->p, QM_MCC_VERB_QUERYFQ); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPAA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCR_VERB_QUERYFQ); + res = mcr->result; + if (res == QM_MCR_RESULT_OK) + *fqd = mcr->queryfq.fqd; + hw_fqd_to_cpu(fqd); + if (res != QM_MCR_RESULT_OK) + return -EIO; + return 0; +} + +int qman_query_fq_has_pkts(struct qman_fq *fq) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p = get_affine_portal(); + + int ret = 0; + u8 res; + + mcc = qm_mc_start(&p->p); + mcc->queryfq.fqid = cpu_to_be32(fq->fqid); + qm_mc_commit(&p->p, QM_MCC_VERB_QUERYFQ_NP); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + res = mcr->result; + if (res == QM_MCR_RESULT_OK) + ret = !!mcr->queryfq_np.frm_cnt; + return ret; +} + +int qman_query_fq_np(struct qman_fq *fq, struct qm_mcr_queryfq_np *np) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p = get_affine_portal(); + + u8 res; + + mcc = qm_mc_start(&p->p); + mcc->queryfq.fqid = cpu_to_be32(fq->fqid); + qm_mc_commit(&p->p, QM_MCC_VERB_QUERYFQ_NP); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPAA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCR_VERB_QUERYFQ_NP); + res = mcr->result; + if (res == QM_MCR_RESULT_OK) { + *np = mcr->queryfq_np; + np->fqd_link = be24_to_cpu(np->fqd_link); + np->odp_seq = be16_to_cpu(np->odp_seq); + np->orp_nesn = be16_to_cpu(np->orp_nesn); + np->orp_ea_hseq = be16_to_cpu(np->orp_ea_hseq); + np->orp_ea_tseq = be16_to_cpu(np->orp_ea_tseq); + np->orp_ea_hptr = be24_to_cpu(np->orp_ea_hptr); + np->orp_ea_tptr = be24_to_cpu(np->orp_ea_tptr); + np->pfdr_hptr = be24_to_cpu(np->pfdr_hptr); + np->pfdr_tptr = be24_to_cpu(np->pfdr_tptr); + np->ics_surp = be16_to_cpu(np->ics_surp); + np->byte_cnt = be32_to_cpu(np->byte_cnt); + np->frm_cnt = be24_to_cpu(np->frm_cnt); + np->ra1_sfdr = be16_to_cpu(np->ra1_sfdr); + np->ra2_sfdr = be16_to_cpu(np->ra2_sfdr); + np->od1_sfdr = be16_to_cpu(np->od1_sfdr); + np->od2_sfdr = be16_to_cpu(np->od2_sfdr); + np->od3_sfdr = be16_to_cpu(np->od3_sfdr); + } + if (res == QM_MCR_RESULT_ERR_FQID) + return -ERANGE; + else if (res != QM_MCR_RESULT_OK) + return -EIO; + return 0; +} + +int qman_query_wq(u8 query_dedicated, struct qm_mcr_querywq *wq) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p = get_affine_portal(); + + u8 res, myverb; + + myverb = (query_dedicated) ? QM_MCR_VERB_QUERYWQ_DEDICATED : + QM_MCR_VERB_QUERYWQ; + mcc = qm_mc_start(&p->p); + mcc->querywq.channel.id = cpu_to_be16(wq->channel.id); + qm_mc_commit(&p->p, myverb); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPAA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == myverb); + res = mcr->result; + if (res == QM_MCR_RESULT_OK) { + int i, array_len; + + wq->channel.id = be16_to_cpu(mcr->querywq.channel.id); + array_len = ARRAY_SIZE(mcr->querywq.wq_len); + for (i = 0; i < array_len; i++) + wq->wq_len[i] = be32_to_cpu(mcr->querywq.wq_len[i]); + } + if (res != QM_MCR_RESULT_OK) { + pr_err("QUERYWQ failed: %s\n", mcr_result_str(res)); + return -EIO; + } + return 0; +} + +int qman_testwrite_cgr(struct qman_cgr *cgr, u64 i_bcnt, + struct qm_mcr_cgrtestwrite *result) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p = get_affine_portal(); + + u8 res; + + mcc = qm_mc_start(&p->p); + mcc->cgrtestwrite.cgid = cgr->cgrid; + mcc->cgrtestwrite.i_bcnt_hi = (u8)(i_bcnt >> 32); + mcc->cgrtestwrite.i_bcnt_lo = (u32)i_bcnt; + qm_mc_commit(&p->p, QM_MCC_VERB_CGRTESTWRITE); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPAA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCC_VERB_CGRTESTWRITE); + res = mcr->result; + if (res == QM_MCR_RESULT_OK) + *result = mcr->cgrtestwrite; + if (res != QM_MCR_RESULT_OK) { + pr_err("CGR TEST WRITE failed: %s\n", mcr_result_str(res)); + return -EIO; + } + return 0; +} + +int qman_query_cgr(struct qman_cgr *cgr, struct qm_mcr_querycgr *cgrd) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p = get_affine_portal(); + u8 res; + unsigned int i; + + mcc = qm_mc_start(&p->p); + mcc->querycgr.cgid = cgr->cgrid; + qm_mc_commit(&p->p, QM_MCC_VERB_QUERYCGR); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPAA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCC_VERB_QUERYCGR); + res = mcr->result; + if (res == QM_MCR_RESULT_OK) + *cgrd = mcr->querycgr; + if (res != QM_MCR_RESULT_OK) { + pr_err("QUERY_CGR failed: %s\n", mcr_result_str(res)); + return -EIO; + } + cgrd->cgr.wr_parm_g.word = + be32_to_cpu(cgrd->cgr.wr_parm_g.word); + cgrd->cgr.wr_parm_y.word = + be32_to_cpu(cgrd->cgr.wr_parm_y.word); + cgrd->cgr.wr_parm_r.word = + be32_to_cpu(cgrd->cgr.wr_parm_r.word); + cgrd->cgr.cscn_targ = be32_to_cpu(cgrd->cgr.cscn_targ); + cgrd->cgr.__cs_thres = be16_to_cpu(cgrd->cgr.__cs_thres); + for (i = 0; i < ARRAY_SIZE(cgrd->cscn_targ_swp); i++) + cgrd->cscn_targ_swp[i] = + be32_to_cpu(cgrd->cscn_targ_swp[i]); + return 0; +} + +int qman_query_congestion(struct qm_mcr_querycongestion *congestion) +{ + struct qm_mc_result *mcr; + struct qman_portal *p = get_affine_portal(); + u8 res; + unsigned int i; + + qm_mc_start(&p->p); + qm_mc_commit(&p->p, QM_MCC_VERB_QUERYCONGESTION); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPAA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == + QM_MCC_VERB_QUERYCONGESTION); + res = mcr->result; + if (res == QM_MCR_RESULT_OK) + *congestion = mcr->querycongestion; + if (res != QM_MCR_RESULT_OK) { + pr_err("QUERY_CONGESTION failed: %s\n", mcr_result_str(res)); + return -EIO; + } + for (i = 0; i < ARRAY_SIZE(congestion->state.state); i++) + congestion->state.state[i] = + be32_to_cpu(congestion->state.state[i]); + return 0; +} + +int qman_set_vdq(struct qman_fq *fq, u16 num) +{ + struct qman_portal *p = get_affine_portal(); + uint32_t vdqcr; + int ret = -EBUSY; + + vdqcr = QM_VDQCR_EXACT; + vdqcr |= QM_VDQCR_NUMFRAMES_SET(num); + + if ((fq->state != qman_fq_state_parked) && + (fq->state != qman_fq_state_retired)) { + ret = -EINVAL; + goto out; + } + if (fq_isset(fq, QMAN_FQ_STATE_VDQCR)) { + ret = -EBUSY; + goto out; + } + vdqcr = (vdqcr & ~QM_VDQCR_FQID_MASK) | fq->fqid; + + if (!p->vdqcr_owned) { + FQLOCK(fq); + if (fq_isset(fq, QMAN_FQ_STATE_VDQCR)) + goto escape; + fq_set(fq, QMAN_FQ_STATE_VDQCR); + FQUNLOCK(fq); + p->vdqcr_owned = fq; + ret = 0; + } +escape: + if (!ret) + qm_dqrr_vdqcr_set(&p->p, vdqcr); + +out: + return ret; +} + +int qman_volatile_dequeue(struct qman_fq *fq, u32 flags __maybe_unused, + u32 vdqcr) +{ + struct qman_portal *p; + int ret = -EBUSY; + + if ((fq->state != qman_fq_state_parked) && + (fq->state != qman_fq_state_retired)) + return -EINVAL; + if (vdqcr & QM_VDQCR_FQID_MASK) + return -EINVAL; + if (fq_isset(fq, QMAN_FQ_STATE_VDQCR)) + return -EBUSY; + vdqcr = (vdqcr & ~QM_VDQCR_FQID_MASK) | fq->fqid; + + p = get_affine_portal(); + + if (!p->vdqcr_owned) { + FQLOCK(fq); + if (fq_isset(fq, QMAN_FQ_STATE_VDQCR)) + goto escape; + fq_set(fq, QMAN_FQ_STATE_VDQCR); + FQUNLOCK(fq); + p->vdqcr_owned = fq; + ret = 0; + } +escape: + if (ret) + return ret; + + /* VDQCR is set */ + qm_dqrr_vdqcr_set(&p->p, vdqcr); + return 0; +} + +static noinline void update_eqcr_ci(struct qman_portal *p, u8 avail) +{ + if (avail) + qm_eqcr_cce_prefetch(&p->p); + else + qm_eqcr_cce_update(&p->p); +} + +int qman_eqcr_is_empty(void) +{ + struct qman_portal *p = get_affine_portal(); + u8 avail; + + update_eqcr_ci(p, 0); + avail = qm_eqcr_get_fill(&p->p); + return (avail == 0); +} + +void qman_set_dc_ern(qman_cb_dc_ern handler, int affine) +{ + if (affine) { + struct qman_portal *p = get_affine_portal(); + + p->cb_dc_ern = handler; + } else + cb_dc_ern = handler; +} + +static inline struct qm_eqcr_entry *try_p_eq_start(struct qman_portal *p, + struct qman_fq *fq, + const struct qm_fd *fd, + u32 flags) +{ + struct qm_eqcr_entry *eq; + u8 avail; + + if (p->use_eqcr_ci_stashing) { + /* + * The stashing case is easy, only update if we need to in + * order to try and liberate ring entries. + */ + eq = qm_eqcr_start_stash(&p->p); + } else { + /* + * The non-stashing case is harder, need to prefetch ahead of + * time. + */ + avail = qm_eqcr_get_avail(&p->p); + if (avail < 2) + update_eqcr_ci(p, avail); + eq = qm_eqcr_start_no_stash(&p->p); + } + + if (unlikely(!eq)) + return NULL; + + if (flags & QMAN_ENQUEUE_FLAG_DCA) + eq->dca = QM_EQCR_DCA_ENABLE | + ((flags & QMAN_ENQUEUE_FLAG_DCA_PARK) ? + QM_EQCR_DCA_PARK : 0) | + ((flags >> 8) & QM_EQCR_DCA_IDXMASK); + eq->fqid = cpu_to_be32(fq->fqid); +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP + eq->tag = cpu_to_be32(fq->key); +#else + eq->tag = cpu_to_be32((u32)(uintptr_t)fq); +#endif + eq->fd = *fd; + cpu_to_hw_fd(&eq->fd); + return eq; +} + +int qman_enqueue(struct qman_fq *fq, const struct qm_fd *fd, u32 flags) +{ + struct qman_portal *p = get_affine_portal(); + struct qm_eqcr_entry *eq; + + eq = try_p_eq_start(p, fq, fd, flags); + if (!eq) + return -EBUSY; + /* Note: QM_EQCR_VERB_INTERRUPT == QMAN_ENQUEUE_FLAG_WAIT_SYNC */ + qm_eqcr_pvb_commit(&p->p, QM_EQCR_VERB_CMD_ENQUEUE | + (flags & (QM_EQCR_VERB_COLOUR_MASK | QM_EQCR_VERB_INTERRUPT))); + /* Factor the below out, it's used from qman_enqueue_orp() too */ + return 0; +} + +int qman_enqueue_multi(struct qman_fq *fq, + const struct qm_fd *fd, + int frames_to_send) +{ + struct qman_portal *p = get_affine_portal(); + struct qm_portal *portal = &p->p; + + register struct qm_eqcr *eqcr = &portal->eqcr; + struct qm_eqcr_entry *eq = eqcr->cursor, *prev_eq; + + u8 i, diff, old_ci, sent = 0; + + /* Update the available entries if no entry is free */ + if (!eqcr->available) { + old_ci = eqcr->ci; + eqcr->ci = qm_cl_in(EQCR_CI) & (QM_EQCR_SIZE - 1); + diff = qm_cyc_diff(QM_EQCR_SIZE, old_ci, eqcr->ci); + eqcr->available += diff; + if (!diff) + return 0; + } + + /* try to send as many frames as possible */ + while (eqcr->available && frames_to_send--) { + eq->fqid = cpu_to_be32(fq->fqid); +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP + eq->tag = cpu_to_be32(fq->key); +#else + eq->tag = cpu_to_be32((u32)(uintptr_t)fq); +#endif + eq->fd.opaque_addr = fd->opaque_addr; + eq->fd.addr = cpu_to_be40(fd->addr); + eq->fd.status = cpu_to_be32(fd->status); + eq->fd.opaque = cpu_to_be32(fd->opaque); + + eq = (void *)((unsigned long)(eq + 1) & + (~(unsigned long)(QM_EQCR_SIZE << 6))); + eqcr->available--; + sent++; + fd++; + } + lwsync(); + + /* In order for flushes to complete faster, all lines are recorded in + * 32 bit word. + */ + eq = eqcr->cursor; + for (i = 0; i < sent; i++) { + eq->__dont_write_directly__verb = + QM_EQCR_VERB_CMD_ENQUEUE | eqcr->vbit; + prev_eq = eq; + eq = (void *)((unsigned long)(eq + 1) & + (~(unsigned long)(QM_EQCR_SIZE << 6))); + if (unlikely((prev_eq + 1) != eq)) + eqcr->vbit ^= QM_EQCR_VERB_VBIT; + } + + /* We need to flush all the lines but without load/store operations + * between them + */ + eq = eqcr->cursor; + for (i = 0; i < sent; i++) { + dcbf(eq); + eq = (void *)((unsigned long)(eq + 1) & + (~(unsigned long)(QM_EQCR_SIZE << 6))); + } + /* Update cursor for the next call */ + eqcr->cursor = eq; + return sent; +} + +int qman_enqueue_orp(struct qman_fq *fq, const struct qm_fd *fd, u32 flags, + struct qman_fq *orp, u16 orp_seqnum) +{ + struct qman_portal *p = get_affine_portal(); + struct qm_eqcr_entry *eq; + + eq = try_p_eq_start(p, fq, fd, flags); + if (!eq) + return -EBUSY; + /* Process ORP-specifics here */ + if (flags & QMAN_ENQUEUE_FLAG_NLIS) + orp_seqnum |= QM_EQCR_SEQNUM_NLIS; + else { + orp_seqnum &= ~QM_EQCR_SEQNUM_NLIS; + if (flags & QMAN_ENQUEUE_FLAG_NESN) + orp_seqnum |= QM_EQCR_SEQNUM_NESN; + else + /* No need to check 4 QMAN_ENQUEUE_FLAG_HOLE */ + orp_seqnum &= ~QM_EQCR_SEQNUM_NESN; + } + eq->seqnum = cpu_to_be16(orp_seqnum); + eq->orp = cpu_to_be32(orp->fqid); + /* Note: QM_EQCR_VERB_INTERRUPT == QMAN_ENQUEUE_FLAG_WAIT_SYNC */ + qm_eqcr_pvb_commit(&p->p, QM_EQCR_VERB_ORP | + ((flags & (QMAN_ENQUEUE_FLAG_HOLE | QMAN_ENQUEUE_FLAG_NESN)) ? + 0 : QM_EQCR_VERB_CMD_ENQUEUE) | + (flags & (QM_EQCR_VERB_COLOUR_MASK | QM_EQCR_VERB_INTERRUPT))); + + return 0; +} + +int qman_modify_cgr(struct qman_cgr *cgr, u32 flags, + struct qm_mcc_initcgr *opts) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p = get_affine_portal(); + + u8 res; + u8 verb = QM_MCC_VERB_MODIFYCGR; + + mcc = qm_mc_start(&p->p); + if (opts) + mcc->initcgr = *opts; + mcc->initcgr.we_mask = cpu_to_be16(mcc->initcgr.we_mask); + mcc->initcgr.cgr.wr_parm_g.word = + cpu_to_be32(mcc->initcgr.cgr.wr_parm_g.word); + mcc->initcgr.cgr.wr_parm_y.word = + cpu_to_be32(mcc->initcgr.cgr.wr_parm_y.word); + mcc->initcgr.cgr.wr_parm_r.word = + cpu_to_be32(mcc->initcgr.cgr.wr_parm_r.word); + mcc->initcgr.cgr.cscn_targ = cpu_to_be32(mcc->initcgr.cgr.cscn_targ); + mcc->initcgr.cgr.__cs_thres = cpu_to_be16(mcc->initcgr.cgr.__cs_thres); + + mcc->initcgr.cgid = cgr->cgrid; + if (flags & QMAN_CGR_FLAG_USE_INIT) + verb = QM_MCC_VERB_INITCGR; + qm_mc_commit(&p->p, verb); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + + DPAA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == verb); + res = mcr->result; + return (res == QM_MCR_RESULT_OK) ? 0 : -EIO; +} + +#define TARG_MASK(n) (0x80000000 >> (n->config->channel - \ + QM_CHANNEL_SWPORTAL0)) +#define TARG_DCP_MASK(n) (0x80000000 >> (10 + n)) +#define PORTAL_IDX(n) (n->config->channel - QM_CHANNEL_SWPORTAL0) + +int qman_create_cgr(struct qman_cgr *cgr, u32 flags, + struct qm_mcc_initcgr *opts) +{ + struct qm_mcr_querycgr cgr_state; + struct qm_mcc_initcgr local_opts; + int ret; + struct qman_portal *p; + + /* We have to check that the provided CGRID is within the limits of the + * data-structures, for obvious reasons. However we'll let h/w take + * care of determining whether it's within the limits of what exists on + * the SoC. + */ + if (cgr->cgrid >= __CGR_NUM) + return -EINVAL; + + p = get_affine_portal(); + + memset(&local_opts, 0, sizeof(struct qm_mcc_initcgr)); + cgr->chan = p->config->channel; + spin_lock(&p->cgr_lock); + + /* if no opts specified, just add it to the list */ + if (!opts) + goto add_list; + + ret = qman_query_cgr(cgr, &cgr_state); + if (ret) + goto release_lock; + if (opts) + local_opts = *opts; + if ((qman_ip_rev & 0xFF00) >= QMAN_REV30) + local_opts.cgr.cscn_targ_upd_ctrl = + QM_CGR_TARG_UDP_CTRL_WRITE_BIT | PORTAL_IDX(p); + else + /* Overwrite TARG */ + local_opts.cgr.cscn_targ = cgr_state.cgr.cscn_targ | + TARG_MASK(p); + local_opts.we_mask |= QM_CGR_WE_CSCN_TARG; + + /* send init if flags indicate so */ + if (opts && (flags & QMAN_CGR_FLAG_USE_INIT)) + ret = qman_modify_cgr(cgr, QMAN_CGR_FLAG_USE_INIT, &local_opts); + else + ret = qman_modify_cgr(cgr, 0, &local_opts); + if (ret) + goto release_lock; +add_list: + list_add(&cgr->node, &p->cgr_cbs); + + /* Determine if newly added object requires its callback to be called */ + ret = qman_query_cgr(cgr, &cgr_state); + if (ret) { + /* we can't go back, so proceed and return success, but screen + * and wail to the log file. + */ + pr_crit("CGR HW state partially modified\n"); + ret = 0; + goto release_lock; + } + if (cgr->cb && cgr_state.cgr.cscn_en && qman_cgrs_get(&p->cgrs[1], + cgr->cgrid)) + cgr->cb(p, cgr, 1); +release_lock: + spin_unlock(&p->cgr_lock); + return ret; +} + +int qman_create_cgr_to_dcp(struct qman_cgr *cgr, u32 flags, u16 dcp_portal, + struct qm_mcc_initcgr *opts) +{ + struct qm_mcc_initcgr local_opts; + struct qm_mcr_querycgr cgr_state; + int ret; + + if ((qman_ip_rev & 0xFF00) < QMAN_REV30) { + pr_warn("QMan version doesn't support CSCN => DCP portal\n"); + return -EINVAL; + } + /* We have to check that the provided CGRID is within the limits of the + * data-structures, for obvious reasons. However we'll let h/w take + * care of determining whether it's within the limits of what exists on + * the SoC. + */ + if (cgr->cgrid >= __CGR_NUM) + return -EINVAL; + + ret = qman_query_cgr(cgr, &cgr_state); + if (ret) + return ret; + + memset(&local_opts, 0, sizeof(struct qm_mcc_initcgr)); + if (opts) + local_opts = *opts; + + if ((qman_ip_rev & 0xFF00) >= QMAN_REV30) + local_opts.cgr.cscn_targ_upd_ctrl = + QM_CGR_TARG_UDP_CTRL_WRITE_BIT | + QM_CGR_TARG_UDP_CTRL_DCP | dcp_portal; + else + local_opts.cgr.cscn_targ = cgr_state.cgr.cscn_targ | + TARG_DCP_MASK(dcp_portal); + local_opts.we_mask |= QM_CGR_WE_CSCN_TARG; + + /* send init if flags indicate so */ + if (opts && (flags & QMAN_CGR_FLAG_USE_INIT)) + ret = qman_modify_cgr(cgr, QMAN_CGR_FLAG_USE_INIT, + &local_opts); + else + ret = qman_modify_cgr(cgr, 0, &local_opts); + + return ret; +} + +int qman_delete_cgr(struct qman_cgr *cgr) +{ + struct qm_mcr_querycgr cgr_state; + struct qm_mcc_initcgr local_opts; + int ret = 0; + struct qman_cgr *i; + struct qman_portal *p = get_affine_portal(); + + if (cgr->chan != p->config->channel) { + pr_crit("Attempting to delete cgr from different portal than" + " it was create: create 0x%x, delete 0x%x\n", + cgr->chan, p->config->channel); + ret = -EINVAL; + goto put_portal; + } + memset(&local_opts, 0, sizeof(struct qm_mcc_initcgr)); + spin_lock(&p->cgr_lock); + list_del(&cgr->node); + /* + * If there are no other CGR objects for this CGRID in the list, + * update CSCN_TARG accordingly + */ + list_for_each_entry(i, &p->cgr_cbs, node) + if ((i->cgrid == cgr->cgrid) && i->cb) + goto release_lock; + ret = qman_query_cgr(cgr, &cgr_state); + if (ret) { + /* add back to the list */ + list_add(&cgr->node, &p->cgr_cbs); + goto release_lock; + } + /* Overwrite TARG */ + local_opts.we_mask = QM_CGR_WE_CSCN_TARG; + if ((qman_ip_rev & 0xFF00) >= QMAN_REV30) + local_opts.cgr.cscn_targ_upd_ctrl = PORTAL_IDX(p); + else + local_opts.cgr.cscn_targ = cgr_state.cgr.cscn_targ & + ~(TARG_MASK(p)); + ret = qman_modify_cgr(cgr, 0, &local_opts); + if (ret) + /* add back to the list */ + list_add(&cgr->node, &p->cgr_cbs); +release_lock: + spin_unlock(&p->cgr_lock); +put_portal: + return ret; +} + +int qman_shutdown_fq(u32 fqid) +{ + struct qman_portal *p; + struct qm_portal *low_p; + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + u8 state; + int orl_empty, fq_empty, drain = 0; + u32 result; + u32 channel, wq; + u16 dest_wq; + + p = get_affine_portal(); + low_p = &p->p; + + /* Determine the state of the FQID */ + mcc = qm_mc_start(low_p); + mcc->queryfq_np.fqid = cpu_to_be32(fqid); + qm_mc_commit(low_p, QM_MCC_VERB_QUERYFQ_NP); + while (!(mcr = qm_mc_result(low_p))) + cpu_relax(); + DPAA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCR_VERB_QUERYFQ_NP); + state = mcr->queryfq_np.state & QM_MCR_NP_STATE_MASK; + if (state == QM_MCR_NP_STATE_OOS) + return 0; /* Already OOS, no need to do anymore checks */ + + /* Query which channel the FQ is using */ + mcc = qm_mc_start(low_p); + mcc->queryfq.fqid = cpu_to_be32(fqid); + qm_mc_commit(low_p, QM_MCC_VERB_QUERYFQ); + while (!(mcr = qm_mc_result(low_p))) + cpu_relax(); + DPAA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCR_VERB_QUERYFQ); + + /* Need to store these since the MCR gets reused */ + dest_wq = be16_to_cpu(mcr->queryfq.fqd.dest_wq); + channel = dest_wq & 0x7; + wq = dest_wq >> 3; + + switch (state) { + case QM_MCR_NP_STATE_TEN_SCHED: + case QM_MCR_NP_STATE_TRU_SCHED: + case QM_MCR_NP_STATE_ACTIVE: + case QM_MCR_NP_STATE_PARKED: + orl_empty = 0; + mcc = qm_mc_start(low_p); + mcc->alterfq.fqid = cpu_to_be32(fqid); + qm_mc_commit(low_p, QM_MCC_VERB_ALTER_RETIRE); + while (!(mcr = qm_mc_result(low_p))) + cpu_relax(); + DPAA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == + QM_MCR_VERB_ALTER_RETIRE); + result = mcr->result; /* Make a copy as we reuse MCR below */ + + if (result == QM_MCR_RESULT_PENDING) { + /* Need to wait for the FQRN in the message ring, which + * will only occur once the FQ has been drained. In + * order for the FQ to drain the portal needs to be set + * to dequeue from the channel the FQ is scheduled on + */ + const struct qm_mr_entry *msg; + const struct qm_dqrr_entry *dqrr = NULL; + int found_fqrn = 0; + __maybe_unused u16 dequeue_wq = 0; + + /* Flag that we need to drain FQ */ + drain = 1; + + if (channel >= qm_channel_pool1 && + channel < (u16)(qm_channel_pool1 + 15)) { + /* Pool channel, enable the bit in the portal */ + dequeue_wq = (channel - + qm_channel_pool1 + 1) << 4 | wq; + } else if (channel < qm_channel_pool1) { + /* Dedicated channel */ + dequeue_wq = wq; + } else { + pr_info("Cannot recover FQ 0x%x," + " it is scheduled on channel 0x%x", + fqid, channel); + return -EBUSY; + } + /* Set the sdqcr to drain this channel */ + if (channel < qm_channel_pool1) + qm_dqrr_sdqcr_set(low_p, + QM_SDQCR_TYPE_ACTIVE | + QM_SDQCR_CHANNELS_DEDICATED); + else + qm_dqrr_sdqcr_set(low_p, + QM_SDQCR_TYPE_ACTIVE | + QM_SDQCR_CHANNELS_POOL_CONV + (channel)); + while (!found_fqrn) { + /* Keep draining DQRR while checking the MR*/ + qm_dqrr_pvb_update(low_p); + dqrr = qm_dqrr_current(low_p); + while (dqrr) { + qm_dqrr_cdc_consume_1ptr( + low_p, dqrr, 0); + qm_dqrr_pvb_update(low_p); + qm_dqrr_next(low_p); + dqrr = qm_dqrr_current(low_p); + } + /* Process message ring too */ + qm_mr_pvb_update(low_p); + msg = qm_mr_current(low_p); + while (msg) { + if ((msg->verb & + QM_MR_VERB_TYPE_MASK) + == QM_MR_VERB_FQRN) + found_fqrn = 1; + qm_mr_next(low_p); + qm_mr_cci_consume_to_current(low_p); + qm_mr_pvb_update(low_p); + msg = qm_mr_current(low_p); + } + cpu_relax(); + } + } + if (result != QM_MCR_RESULT_OK && + result != QM_MCR_RESULT_PENDING) { + /* error */ + pr_err("qman_retire_fq failed on FQ 0x%x," + " result=0x%x\n", fqid, result); + return -1; + } + if (!(mcr->alterfq.fqs & QM_MCR_FQS_ORLPRESENT)) { + /* ORL had no entries, no need to wait until the + * ERNs come in. + */ + orl_empty = 1; + } + /* Retirement succeeded, check to see if FQ needs + * to be drained. + */ + if (drain || mcr->alterfq.fqs & QM_MCR_FQS_NOTEMPTY) { + /* FQ is Not Empty, drain using volatile DQ commands */ + fq_empty = 0; + do { + const struct qm_dqrr_entry *dqrr = NULL; + u32 vdqcr = fqid | QM_VDQCR_NUMFRAMES_SET(3); + + qm_dqrr_vdqcr_set(low_p, vdqcr); + + /* Wait for a dequeue to occur */ + while (dqrr == NULL) { + qm_dqrr_pvb_update(low_p); + dqrr = qm_dqrr_current(low_p); + if (!dqrr) + cpu_relax(); + } + /* Process the dequeues, making sure to + * empty the ring completely. + */ + while (dqrr) { + if (dqrr->fqid == fqid && + dqrr->stat & QM_DQRR_STAT_FQ_EMPTY) + fq_empty = 1; + qm_dqrr_cdc_consume_1ptr(low_p, + dqrr, 0); + qm_dqrr_pvb_update(low_p); + qm_dqrr_next(low_p); + dqrr = qm_dqrr_current(low_p); + } + } while (fq_empty == 0); + } + qm_dqrr_sdqcr_set(low_p, 0); + + /* Wait for the ORL to have been completely drained */ + while (orl_empty == 0) { + const struct qm_mr_entry *msg; + + qm_mr_pvb_update(low_p); + msg = qm_mr_current(low_p); + while (msg) { + if ((msg->verb & QM_MR_VERB_TYPE_MASK) == + QM_MR_VERB_FQRL) + orl_empty = 1; + qm_mr_next(low_p); + qm_mr_cci_consume_to_current(low_p); + qm_mr_pvb_update(low_p); + msg = qm_mr_current(low_p); + } + cpu_relax(); + } + mcc = qm_mc_start(low_p); + mcc->alterfq.fqid = cpu_to_be32(fqid); + qm_mc_commit(low_p, QM_MCC_VERB_ALTER_OOS); + while (!(mcr = qm_mc_result(low_p))) + cpu_relax(); + DPAA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == + QM_MCR_VERB_ALTER_OOS); + if (mcr->result != QM_MCR_RESULT_OK) { + pr_err( + "OOS after drain Failed on FQID 0x%x, result 0x%x\n", + fqid, mcr->result); + return -1; + } + return 0; + + case QM_MCR_NP_STATE_RETIRED: + /* Send OOS Command */ + mcc = qm_mc_start(low_p); + mcc->alterfq.fqid = cpu_to_be32(fqid); + qm_mc_commit(low_p, QM_MCC_VERB_ALTER_OOS); + while (!(mcr = qm_mc_result(low_p))) + cpu_relax(); + DPAA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == + QM_MCR_VERB_ALTER_OOS); + if (mcr->result) { + pr_err("OOS Failed on FQID 0x%x\n", fqid); + return -1; + } + return 0; + + } + return -1; +} |