diff options
Diffstat (limited to 'src/spdk/dpdk/lib/librte_sched')
| -rw-r--r-- | src/spdk/dpdk/lib/librte_sched/Makefile | 33 | ||||
| -rw-r--r-- | src/spdk/dpdk/lib/librte_sched/meson.build | 7 | ||||
| -rw-r--r-- | src/spdk/dpdk/lib/librte_sched/rte_approx.c | 167 | ||||
| -rw-r--r-- | src/spdk/dpdk/lib/librte_sched/rte_approx.h | 46 | ||||
| -rw-r--r-- | src/spdk/dpdk/lib/librte_sched/rte_red.c | 129 | ||||
| -rw-r--r-- | src/spdk/dpdk/lib/librte_sched/rte_red.h | 411 | ||||
| -rw-r--r-- | src/spdk/dpdk/lib/librte_sched/rte_sched.c | 2242 | ||||
| -rw-r--r-- | src/spdk/dpdk/lib/librte_sched/rte_sched.h | 447 | ||||
| -rw-r--r-- | src/spdk/dpdk/lib/librte_sched/rte_sched_common.h | 101 | ||||
| -rw-r--r-- | src/spdk/dpdk/lib/librte_sched/rte_sched_version.map | 37 |
10 files changed, 3620 insertions, 0 deletions
diff --git a/src/spdk/dpdk/lib/librte_sched/Makefile b/src/spdk/dpdk/lib/librte_sched/Makefile new file mode 100644 index 00000000..55d9c698 --- /dev/null +++ b/src/spdk/dpdk/lib/librte_sched/Makefile @@ -0,0 +1,33 @@ +# SPDX-License-Identifier: BSD-3-Clause +# Copyright(c) 2010-2014 Intel Corporation + +include $(RTE_SDK)/mk/rte.vars.mk + +# +# library name +# +LIB = librte_sched.a + +CFLAGS += -O3 +CFLAGS += $(WERROR_FLAGS) + +CFLAGS_rte_red.o := -D_GNU_SOURCE + +LDLIBS += -lm +LDLIBS += -lrt +LDLIBS += -lrte_eal -lrte_mempool -lrte_mbuf -lrte_net +LDLIBS += -lrte_timer + +EXPORT_MAP := rte_sched_version.map + +LIBABIVER := 1 + +# +# all source are stored in SRCS-y +# +SRCS-$(CONFIG_RTE_LIBRTE_SCHED) += rte_sched.c rte_red.c rte_approx.c + +# install includes +SYMLINK-$(CONFIG_RTE_LIBRTE_SCHED)-include := rte_sched.h rte_sched_common.h rte_red.h rte_approx.h + +include $(RTE_SDK)/mk/rte.lib.mk diff --git a/src/spdk/dpdk/lib/librte_sched/meson.build b/src/spdk/dpdk/lib/librte_sched/meson.build new file mode 100644 index 00000000..f85d64df --- /dev/null +++ b/src/spdk/dpdk/lib/librte_sched/meson.build @@ -0,0 +1,7 @@ +# SPDX-License-Identifier: BSD-3-Clause +# Copyright(c) 2017 Intel Corporation + +sources = files('rte_sched.c', 'rte_red.c', 'rte_approx.c') +headers = files('rte_sched.h', 'rte_sched_common.h', + 'rte_red.h', 'rte_approx.h') +deps += ['mbuf', 'meter'] diff --git a/src/spdk/dpdk/lib/librte_sched/rte_approx.c b/src/spdk/dpdk/lib/librte_sched/rte_approx.c new file mode 100644 index 00000000..30620b83 --- /dev/null +++ b/src/spdk/dpdk/lib/librte_sched/rte_approx.c @@ -0,0 +1,167 @@ +/* SPDX-License-Identifier: BSD-3-Clause + * Copyright(c) 2010-2014 Intel Corporation + */ + +#include <stdlib.h> + +#include "rte_approx.h" + +/* + * Based on paper "Approximating Rational Numbers by Fractions" by Michal + * Forisek forisek@dcs.fmph.uniba.sk + * + * Given a rational number alpha with 0 < alpha < 1 and a precision d, the goal + * is to find positive integers p, q such that alpha - d < p/q < alpha + d, and + * q is minimal. + * + * http://people.ksp.sk/~misof/publications/2007approx.pdf + */ + +/* fraction comparison: compare (a/b) and (c/d) */ +static inline uint32_t +less(uint32_t a, uint32_t b, uint32_t c, uint32_t d) +{ + return a*d < b*c; +} + +static inline uint32_t +less_or_equal(uint32_t a, uint32_t b, uint32_t c, uint32_t d) +{ + return a*d <= b*c; +} + +/* check whether a/b is a valid approximation */ +static inline uint32_t +matches(uint32_t a, uint32_t b, + uint32_t alpha_num, uint32_t d_num, uint32_t denum) +{ + if (less_or_equal(a, b, alpha_num - d_num, denum)) + return 0; + + if (less(a ,b, alpha_num + d_num, denum)) + return 1; + + return 0; +} + +static inline void +find_exact_solution_left(uint32_t p_a, uint32_t q_a, uint32_t p_b, uint32_t q_b, + uint32_t alpha_num, uint32_t d_num, uint32_t denum, uint32_t *p, uint32_t *q) +{ + uint32_t k_num = denum * p_b - (alpha_num + d_num) * q_b; + uint32_t k_denum = (alpha_num + d_num) * q_a - denum * p_a; + uint32_t k = (k_num / k_denum) + 1; + + *p = p_b + k * p_a; + *q = q_b + k * q_a; +} + +static inline void +find_exact_solution_right(uint32_t p_a, uint32_t q_a, uint32_t p_b, uint32_t q_b, + uint32_t alpha_num, uint32_t d_num, uint32_t denum, uint32_t *p, uint32_t *q) +{ + uint32_t k_num = - denum * p_b + (alpha_num - d_num) * q_b; + uint32_t k_denum = - (alpha_num - d_num) * q_a + denum * p_a; + uint32_t k = (k_num / k_denum) + 1; + + *p = p_b + k * p_a; + *q = q_b + k * q_a; +} + +static int +find_best_rational_approximation(uint32_t alpha_num, uint32_t d_num, uint32_t denum, uint32_t *p, uint32_t *q) +{ + uint32_t p_a, q_a, p_b, q_b; + + /* check assumptions on the inputs */ + if (!((0 < d_num) && (d_num < alpha_num) && (alpha_num < denum) && (d_num + alpha_num < denum))) { + return -1; + } + + /* set initial bounds for the search */ + p_a = 0; + q_a = 1; + p_b = 1; + q_b = 1; + + while (1) { + uint32_t new_p_a, new_q_a, new_p_b, new_q_b; + uint32_t x_num, x_denum, x; + int aa, bb; + + /* compute the number of steps to the left */ + x_num = denum * p_b - alpha_num * q_b; + x_denum = - denum * p_a + alpha_num * q_a; + x = (x_num + x_denum - 1) / x_denum; /* x = ceil(x_num / x_denum) */ + + /* check whether we have a valid approximation */ + aa = matches(p_b + x * p_a, q_b + x * q_a, alpha_num, d_num, denum); + bb = matches(p_b + (x-1) * p_a, q_b + (x - 1) * q_a, alpha_num, d_num, denum); + if (aa || bb) { + find_exact_solution_left(p_a, q_a, p_b, q_b, alpha_num, d_num, denum, p, q); + return 0; + } + + /* update the interval */ + new_p_a = p_b + (x - 1) * p_a ; + new_q_a = q_b + (x - 1) * q_a; + new_p_b = p_b + x * p_a ; + new_q_b = q_b + x * q_a; + + p_a = new_p_a ; + q_a = new_q_a; + p_b = new_p_b ; + q_b = new_q_b; + + /* compute the number of steps to the right */ + x_num = alpha_num * q_b - denum * p_b; + x_denum = - alpha_num * q_a + denum * p_a; + x = (x_num + x_denum - 1) / x_denum; /* x = ceil(x_num / x_denum) */ + + /* check whether we have a valid approximation */ + aa = matches(p_b + x * p_a, q_b + x * q_a, alpha_num, d_num, denum); + bb = matches(p_b + (x - 1) * p_a, q_b + (x - 1) * q_a, alpha_num, d_num, denum); + if (aa || bb) { + find_exact_solution_right(p_a, q_a, p_b, q_b, alpha_num, d_num, denum, p, q); + return 0; + } + + /* update the interval */ + new_p_a = p_b + (x - 1) * p_a; + new_q_a = q_b + (x - 1) * q_a; + new_p_b = p_b + x * p_a; + new_q_b = q_b + x * q_a; + + p_a = new_p_a; + q_a = new_q_a; + p_b = new_p_b; + q_b = new_q_b; + } +} + +int rte_approx(double alpha, double d, uint32_t *p, uint32_t *q) +{ + uint32_t alpha_num, d_num, denum; + + /* Check input arguments */ + if (!((0.0 < d) && (d < alpha) && (alpha < 1.0))) { + return -1; + } + + if ((p == NULL) || (q == NULL)) { + return -2; + } + + /* Compute alpha_num, d_num and denum */ + denum = 1; + while (d < 1) { + alpha *= 10; + d *= 10; + denum *= 10; + } + alpha_num = (uint32_t) alpha; + d_num = (uint32_t) d; + + /* Perform approximation */ + return find_best_rational_approximation(alpha_num, d_num, denum, p, q); +} diff --git a/src/spdk/dpdk/lib/librte_sched/rte_approx.h b/src/spdk/dpdk/lib/librte_sched/rte_approx.h new file mode 100644 index 00000000..0244d98f --- /dev/null +++ b/src/spdk/dpdk/lib/librte_sched/rte_approx.h @@ -0,0 +1,46 @@ +/* SPDX-License-Identifier: BSD-3-Clause + * Copyright(c) 2010-2014 Intel Corporation + */ + +#ifndef __INCLUDE_RTE_APPROX_H__ +#define __INCLUDE_RTE_APPROX_H__ + +#ifdef __cplusplus +extern "C" { +#endif + +/** + * @file + * RTE Rational Approximation + * + * Given a rational number alpha with 0 < alpha < 1 and a precision d, the goal + * is to find positive integers p, q such that alpha - d < p/q < alpha + d, and + * q is minimal. + * + ***/ + +#include <stdint.h> + +/** + * Find best rational approximation + * + * @param alpha + * Rational number to approximate + * @param d + * Precision for the rational approximation + * @param p + * Pointer to pre-allocated space where the numerator of the rational + * approximation will be stored when operation is successful + * @param q + * Pointer to pre-allocated space where the denominator of the rational + * approximation will be stored when operation is successful + * @return + * 0 upon success, error code otherwise + */ +int rte_approx(double alpha, double d, uint32_t *p, uint32_t *q); + +#ifdef __cplusplus +} +#endif + +#endif /* __INCLUDE_RTE_APPROX_H__ */ diff --git a/src/spdk/dpdk/lib/librte_sched/rte_red.c b/src/spdk/dpdk/lib/librte_sched/rte_red.c new file mode 100644 index 00000000..45a452f6 --- /dev/null +++ b/src/spdk/dpdk/lib/librte_sched/rte_red.c @@ -0,0 +1,129 @@ +/* SPDX-License-Identifier: BSD-3-Clause + * Copyright(c) 2010-2014 Intel Corporation + */ + +#include <math.h> +#include "rte_red.h" +#include <rte_random.h> +#include <rte_common.h> + +#ifdef __INTEL_COMPILER +#pragma warning(disable:2259) /* conversion may lose significant bits */ +#endif + +static int rte_red_init_done = 0; /**< Flag to indicate that global initialisation is done */ +uint32_t rte_red_rand_val = 0; /**< Random value cache */ +uint32_t rte_red_rand_seed = 0; /**< Seed for random number generation */ + +/** + * table[i] = log2(1-Wq) * Scale * -1 + * Wq = 1/(2^i) + */ +uint16_t rte_red_log2_1_minus_Wq[RTE_RED_WQ_LOG2_NUM]; + +/** + * table[i] = 2^(i/16) * Scale + */ +uint16_t rte_red_pow2_frac_inv[16]; + +/** + * @brief Initialize tables used to compute average + * queue size when queue is empty. + */ +static void +__rte_red_init_tables(void) +{ + uint32_t i = 0; + double scale = 0.0; + double table_size = 0.0; + + scale = (double)(1 << RTE_RED_SCALING); + table_size = (double)(RTE_DIM(rte_red_pow2_frac_inv)); + + for (i = 0; i < RTE_DIM(rte_red_pow2_frac_inv); i++) { + double m = (double)i; + + rte_red_pow2_frac_inv[i] = (uint16_t) round(scale / pow(2, m / table_size)); + } + + scale = 1024.0; + + RTE_ASSERT(RTE_RED_WQ_LOG2_NUM == RTE_DIM(rte_red_log2_1_minus_Wq)); + + for (i = RTE_RED_WQ_LOG2_MIN; i <= RTE_RED_WQ_LOG2_MAX; i++) { + double n = (double)i; + double Wq = pow(2, -n); + uint32_t index = i - RTE_RED_WQ_LOG2_MIN; + + rte_red_log2_1_minus_Wq[index] = (uint16_t) round(-1.0 * scale * log2(1.0 - Wq)); + /** + * Table entry of zero, corresponds to a Wq of zero + * which is not valid (avg would remain constant no + * matter how long the queue is empty). So we have + * to check for zero and round up to one. + */ + if (rte_red_log2_1_minus_Wq[index] == 0) { + rte_red_log2_1_minus_Wq[index] = 1; + } + } +} + +int +rte_red_rt_data_init(struct rte_red *red) +{ + if (red == NULL) + return -1; + + red->avg = 0; + red->count = 0; + red->q_time = 0; + return 0; +} + +int +rte_red_config_init(struct rte_red_config *red_cfg, + const uint16_t wq_log2, + const uint16_t min_th, + const uint16_t max_th, + const uint16_t maxp_inv) +{ + if (red_cfg == NULL) { + return -1; + } + if (max_th > RTE_RED_MAX_TH_MAX) { + return -2; + } + if (min_th >= max_th) { + return -3; + } + if (wq_log2 > RTE_RED_WQ_LOG2_MAX) { + return -4; + } + if (wq_log2 < RTE_RED_WQ_LOG2_MIN) { + return -5; + } + if (maxp_inv < RTE_RED_MAXP_INV_MIN) { + return -6; + } + if (maxp_inv > RTE_RED_MAXP_INV_MAX) { + return -7; + } + + /** + * Initialize the RED module if not already done + */ + if (!rte_red_init_done) { + rte_red_rand_seed = rte_rand(); + rte_red_rand_val = rte_fast_rand(); + __rte_red_init_tables(); + rte_red_init_done = 1; + } + + red_cfg->min_th = ((uint32_t) min_th) << (wq_log2 + RTE_RED_SCALING); + red_cfg->max_th = ((uint32_t) max_th) << (wq_log2 + RTE_RED_SCALING); + red_cfg->pa_const = (2 * (max_th - min_th) * maxp_inv) << RTE_RED_SCALING; + red_cfg->maxp_inv = maxp_inv; + red_cfg->wq_log2 = wq_log2; + + return 0; +} diff --git a/src/spdk/dpdk/lib/librte_sched/rte_red.h b/src/spdk/dpdk/lib/librte_sched/rte_red.h new file mode 100644 index 00000000..36273cac --- /dev/null +++ b/src/spdk/dpdk/lib/librte_sched/rte_red.h @@ -0,0 +1,411 @@ +/* SPDX-License-Identifier: BSD-3-Clause + * Copyright(c) 2010-2014 Intel Corporation + */ + +#ifndef __RTE_RED_H_INCLUDED__ +#define __RTE_RED_H_INCLUDED__ + +#ifdef __cplusplus +extern "C" { +#endif + +/** + * @file + * RTE Random Early Detection (RED) + * + * + ***/ + +#include <stdint.h> +#include <limits.h> +#include <rte_common.h> +#include <rte_debug.h> +#include <rte_cycles.h> +#include <rte_branch_prediction.h> + +#define RTE_RED_SCALING 10 /**< Fraction size for fixed-point */ +#define RTE_RED_S (1 << 22) /**< Packet size multiplied by number of leaf queues */ +#define RTE_RED_MAX_TH_MAX 1023 /**< Max threshold limit in fixed point format */ +#define RTE_RED_WQ_LOG2_MIN 1 /**< Min inverse filter weight value */ +#define RTE_RED_WQ_LOG2_MAX 12 /**< Max inverse filter weight value */ +#define RTE_RED_MAXP_INV_MIN 1 /**< Min inverse mark probability value */ +#define RTE_RED_MAXP_INV_MAX 255 /**< Max inverse mark probability value */ +#define RTE_RED_2POW16 (1<<16) /**< 2 power 16 */ +#define RTE_RED_INT16_NBITS (sizeof(uint16_t) * CHAR_BIT) +#define RTE_RED_WQ_LOG2_NUM (RTE_RED_WQ_LOG2_MAX - RTE_RED_WQ_LOG2_MIN + 1) + +/** + * Externs + * + */ +extern uint32_t rte_red_rand_val; +extern uint32_t rte_red_rand_seed; +extern uint16_t rte_red_log2_1_minus_Wq[RTE_RED_WQ_LOG2_NUM]; +extern uint16_t rte_red_pow2_frac_inv[16]; + +/** + * RED configuration parameters passed by user + * + */ +struct rte_red_params { + uint16_t min_th; /**< Minimum threshold for queue (max_th) */ + uint16_t max_th; /**< Maximum threshold for queue (max_th) */ + uint16_t maxp_inv; /**< Inverse of packet marking probability maximum value (maxp = 1 / maxp_inv) */ + uint16_t wq_log2; /**< Negated log2 of queue weight (wq = 1 / (2 ^ wq_log2)) */ +}; + +/** + * RED configuration parameters + */ +struct rte_red_config { + uint32_t min_th; /**< min_th scaled in fixed-point format */ + uint32_t max_th; /**< max_th scaled in fixed-point format */ + uint32_t pa_const; /**< Precomputed constant value used for pa calculation (scaled in fixed-point format) */ + uint8_t maxp_inv; /**< maxp_inv */ + uint8_t wq_log2; /**< wq_log2 */ +}; + +/** + * RED run-time data + */ +struct rte_red { + uint32_t avg; /**< Average queue size (avg), scaled in fixed-point format */ + uint32_t count; /**< Number of packets since last marked packet (count) */ + uint64_t q_time; /**< Start of the queue idle time (q_time) */ +}; + +/** + * @brief Initialises run-time data + * + * @param red [in,out] data pointer to RED runtime data + * + * @return Operation status + * @retval 0 success + * @retval !0 error + */ +int +rte_red_rt_data_init(struct rte_red *red); + +/** + * @brief Configures a single RED configuration parameter structure. + * + * @param red_cfg [in,out] config pointer to a RED configuration parameter structure + * @param wq_log2 [in] log2 of the filter weight, valid range is: + * RTE_RED_WQ_LOG2_MIN <= wq_log2 <= RTE_RED_WQ_LOG2_MAX + * @param min_th [in] queue minimum threshold in number of packets + * @param max_th [in] queue maximum threshold in number of packets + * @param maxp_inv [in] inverse maximum mark probability + * + * @return Operation status + * @retval 0 success + * @retval !0 error + */ +int +rte_red_config_init(struct rte_red_config *red_cfg, + const uint16_t wq_log2, + const uint16_t min_th, + const uint16_t max_th, + const uint16_t maxp_inv); + +/** + * @brief Generate random number for RED + * + * Implementation based on: + * http://software.intel.com/en-us/articles/fast-random-number-generator-on-the-intel-pentiumr-4-processor/ + * + * 10 bit shift has been found through empirical tests (was 16). + * + * @return Random number between 0 and (2^22 - 1) + */ +static inline uint32_t +rte_fast_rand(void) +{ + rte_red_rand_seed = (214013 * rte_red_rand_seed) + 2531011; + return rte_red_rand_seed >> 10; +} + +/** + * @brief calculate factor to scale average queue size when queue + * becomes empty + * + * @param wq_log2 [in] where EWMA filter weight wq = 1/(2 ^ wq_log2) + * @param m [in] exponent in the computed value (1 - wq) ^ m + * + * @return computed value + * @retval ((1 - wq) ^ m) scaled in fixed-point format + */ +static inline uint16_t +__rte_red_calc_qempty_factor(uint8_t wq_log2, uint16_t m) +{ + uint32_t n = 0; + uint32_t f = 0; + + /** + * Basic math tells us that: + * a^b = 2^(b * log2(a) ) + * + * in our case: + * a = (1-Wq) + * b = m + * Wq = 1/ (2^log2n) + * + * So we are computing this equation: + * factor = 2 ^ ( m * log2(1-Wq)) + * + * First we are computing: + * n = m * log2(1-Wq) + * + * To avoid dealing with signed numbers log2 values are positive + * but they should be negative because (1-Wq) is always < 1. + * Contents of log2 table values are also scaled for precision. + */ + + n = m * rte_red_log2_1_minus_Wq[wq_log2 - RTE_RED_WQ_LOG2_MIN]; + + /** + * The tricky part is computing 2^n, for this I split n into + * integer part and fraction part. + * f - is fraction part of n + * n - is integer part of original n + * + * Now using basic math we compute 2^n: + * 2^(f+n) = 2^f * 2^n + * 2^f - we use lookup table + * 2^n - can be replaced with bit shift right operations + */ + + f = (n >> 6) & 0xf; + n >>= 10; + + if (n < RTE_RED_SCALING) + return (uint16_t) ((rte_red_pow2_frac_inv[f] + (1 << (n - 1))) >> n); + + return 0; +} + +/** + * @brief Updates queue average in condition when queue is empty + * + * Note: packet is never dropped in this particular case. + * + * @param red_cfg [in] config pointer to a RED configuration parameter structure + * @param red [in,out] data pointer to RED runtime data + * @param time [in] current time stamp + * + * @return Operation status + * @retval 0 enqueue the packet + * @retval 1 drop the packet based on max threshold criterion + * @retval 2 drop the packet based on mark probability criterion + */ +static inline int +rte_red_enqueue_empty(const struct rte_red_config *red_cfg, + struct rte_red *red, + const uint64_t time) +{ + uint64_t time_diff = 0, m = 0; + + RTE_ASSERT(red_cfg != NULL); + RTE_ASSERT(red != NULL); + + red->count ++; + + /** + * We compute avg but we don't compare avg against + * min_th or max_th, nor calculate drop probability + */ + time_diff = time - red->q_time; + + /** + * m is the number of packets that might have arrived while the queue was empty. + * In this case we have time stamps provided by scheduler in byte units (bytes + * transmitted on network port). Such time stamp translates into time units as + * port speed is fixed but such approach simplifies the code. + */ + m = time_diff / RTE_RED_S; + + /** + * Check that m will fit into 16-bit unsigned integer + */ + if (m >= RTE_RED_2POW16) { + red->avg = 0; + } else { + red->avg = (red->avg >> RTE_RED_SCALING) * __rte_red_calc_qempty_factor(red_cfg->wq_log2, (uint16_t) m); + } + + return 0; +} + +/** + * Drop probability (Sally Floyd and Van Jacobson): + * + * pb = (1 / maxp_inv) * (avg - min_th) / (max_th - min_th) + * pa = pb / (2 - count * pb) + * + * + * (1 / maxp_inv) * (avg - min_th) + * --------------------------------- + * max_th - min_th + * pa = ----------------------------------------------- + * count * (1 / maxp_inv) * (avg - min_th) + * 2 - ----------------------------------------- + * max_th - min_th + * + * + * avg - min_th + * pa = ----------------------------------------------------------- + * 2 * (max_th - min_th) * maxp_inv - count * (avg - min_th) + * + * + * We define pa_const as: pa_const = 2 * (max_th - min_th) * maxp_inv. Then: + * + * + * avg - min_th + * pa = ----------------------------------- + * pa_const - count * (avg - min_th) + */ + +/** + * @brief make a decision to drop or enqueue a packet based on mark probability + * criteria + * + * @param red_cfg [in] config pointer to structure defining RED parameters + * @param red [in,out] data pointer to RED runtime data + * + * @return operation status + * @retval 0 enqueue the packet + * @retval 1 drop the packet + */ +static inline int +__rte_red_drop(const struct rte_red_config *red_cfg, struct rte_red *red) +{ + uint32_t pa_num = 0; /* numerator of drop-probability */ + uint32_t pa_den = 0; /* denominator of drop-probability */ + uint32_t pa_num_count = 0; + + pa_num = (red->avg - red_cfg->min_th) >> (red_cfg->wq_log2); + + pa_num_count = red->count * pa_num; + + if (red_cfg->pa_const <= pa_num_count) + return 1; + + pa_den = red_cfg->pa_const - pa_num_count; + + /* If drop, generate and save random number to be used next time */ + if (unlikely((rte_red_rand_val % pa_den) < pa_num)) { + rte_red_rand_val = rte_fast_rand(); + + return 1; + } + + /* No drop */ + return 0; +} + +/** + * @brief Decides if new packet should be enqeued or dropped in queue non-empty case + * + * @param red_cfg [in] config pointer to a RED configuration parameter structure + * @param red [in,out] data pointer to RED runtime data + * @param q [in] current queue size (measured in packets) + * + * @return Operation status + * @retval 0 enqueue the packet + * @retval 1 drop the packet based on max threshold criterion + * @retval 2 drop the packet based on mark probability criterion + */ +static inline int +rte_red_enqueue_nonempty(const struct rte_red_config *red_cfg, + struct rte_red *red, + const unsigned q) +{ + RTE_ASSERT(red_cfg != NULL); + RTE_ASSERT(red != NULL); + + /** + * EWMA filter (Sally Floyd and Van Jacobson): + * avg = (1 - wq) * avg + wq * q + * avg = avg + q * wq - avg * wq + * + * We select: wq = 2^(-n). Let scaled version of avg be: avg_s = avg * 2^(N+n). We get: + * avg_s = avg_s + q * 2^N - avg_s * 2^(-n) + * + * By using shift left/right operations, we get: + * avg_s = avg_s + (q << N) - (avg_s >> n) + * avg_s += (q << N) - (avg_s >> n) + */ + + /* avg update */ + red->avg += (q << RTE_RED_SCALING) - (red->avg >> red_cfg->wq_log2); + + /* avg < min_th: do not mark the packet */ + if (red->avg < red_cfg->min_th) { + red->count ++; + return 0; + } + + /* min_th <= avg < max_th: mark the packet with pa probability */ + if (red->avg < red_cfg->max_th) { + if (!__rte_red_drop(red_cfg, red)) { + red->count ++; + return 0; + } + + red->count = 0; + return 2; + } + + /* max_th <= avg: always mark the packet */ + red->count = 0; + return 1; +} + +/** + * @brief Decides if new packet should be enqeued or dropped + * Updates run time data based on new queue size value. + * Based on new queue average and RED configuration parameters + * gives verdict whether to enqueue or drop the packet. + * + * @param red_cfg [in] config pointer to a RED configuration parameter structure + * @param red [in,out] data pointer to RED runtime data + * @param q [in] updated queue size in packets + * @param time [in] current time stamp + * + * @return Operation status + * @retval 0 enqueue the packet + * @retval 1 drop the packet based on max threshold criteria + * @retval 2 drop the packet based on mark probability criteria + */ +static inline int +rte_red_enqueue(const struct rte_red_config *red_cfg, + struct rte_red *red, + const unsigned q, + const uint64_t time) +{ + RTE_ASSERT(red_cfg != NULL); + RTE_ASSERT(red != NULL); + + if (q != 0) { + return rte_red_enqueue_nonempty(red_cfg, red, q); + } else { + return rte_red_enqueue_empty(red_cfg, red, time); + } +} + +/** + * @brief Callback to records time that queue became empty + * + * @param red [in,out] data pointer to RED runtime data + * @param time [in] current time stamp + */ +static inline void +rte_red_mark_queue_empty(struct rte_red *red, const uint64_t time) +{ + red->q_time = time; +} + +#ifdef __cplusplus +} +#endif + +#endif /* __RTE_RED_H_INCLUDED__ */ diff --git a/src/spdk/dpdk/lib/librte_sched/rte_sched.c b/src/spdk/dpdk/lib/librte_sched/rte_sched.c new file mode 100644 index 00000000..9269e5c7 --- /dev/null +++ b/src/spdk/dpdk/lib/librte_sched/rte_sched.c @@ -0,0 +1,2242 @@ +/* SPDX-License-Identifier: BSD-3-Clause + * Copyright(c) 2010-2014 Intel Corporation + */ + +#include <stdio.h> +#include <string.h> + +#include <rte_common.h> +#include <rte_log.h> +#include <rte_memory.h> +#include <rte_malloc.h> +#include <rte_cycles.h> +#include <rte_prefetch.h> +#include <rte_branch_prediction.h> +#include <rte_mbuf.h> +#include <rte_bitmap.h> +#include <rte_reciprocal.h> + +#include "rte_sched.h" +#include "rte_sched_common.h" +#include "rte_approx.h" + +#ifdef __INTEL_COMPILER +#pragma warning(disable:2259) /* conversion may lose significant bits */ +#endif + +#ifdef RTE_SCHED_VECTOR +#include <rte_vect.h> + +#ifdef RTE_ARCH_X86 +#define SCHED_VECTOR_SSE4 +#elif defined(RTE_MACHINE_CPUFLAG_NEON) +#define SCHED_VECTOR_NEON +#endif + +#endif + +#define RTE_SCHED_TB_RATE_CONFIG_ERR (1e-7) +#define RTE_SCHED_WRR_SHIFT 3 +#define RTE_SCHED_GRINDER_PCACHE_SIZE (64 / RTE_SCHED_QUEUES_PER_PIPE) +#define RTE_SCHED_PIPE_INVALID UINT32_MAX +#define RTE_SCHED_BMP_POS_INVALID UINT32_MAX + +/* Scaling for cycles_per_byte calculation + * Chosen so that minimum rate is 480 bit/sec + */ +#define RTE_SCHED_TIME_SHIFT 8 + +struct rte_sched_subport { + /* Token bucket (TB) */ + uint64_t tb_time; /* time of last update */ + uint32_t tb_period; + uint32_t tb_credits_per_period; + uint32_t tb_size; + uint32_t tb_credits; + + /* Traffic classes (TCs) */ + uint64_t tc_time; /* time of next update */ + uint32_t tc_credits_per_period[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE]; + uint32_t tc_credits[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE]; + uint32_t tc_period; + + /* TC oversubscription */ + uint32_t tc_ov_wm; + uint32_t tc_ov_wm_min; + uint32_t tc_ov_wm_max; + uint8_t tc_ov_period_id; + uint8_t tc_ov; + uint32_t tc_ov_n; + double tc_ov_rate; + + /* Statistics */ + struct rte_sched_subport_stats stats; +}; + +struct rte_sched_pipe_profile { + /* Token bucket (TB) */ + uint32_t tb_period; + uint32_t tb_credits_per_period; + uint32_t tb_size; + + /* Pipe traffic classes */ + uint32_t tc_period; + uint32_t tc_credits_per_period[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE]; + uint8_t tc_ov_weight; + + /* Pipe queues */ + uint8_t wrr_cost[RTE_SCHED_QUEUES_PER_PIPE]; +}; + +struct rte_sched_pipe { + /* Token bucket (TB) */ + uint64_t tb_time; /* time of last update */ + uint32_t tb_credits; + + /* Pipe profile and flags */ + uint32_t profile; + + /* Traffic classes (TCs) */ + uint64_t tc_time; /* time of next update */ + uint32_t tc_credits[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE]; + + /* Weighted Round Robin (WRR) */ + uint8_t wrr_tokens[RTE_SCHED_QUEUES_PER_PIPE]; + + /* TC oversubscription */ + uint32_t tc_ov_credits; + uint8_t tc_ov_period_id; + uint8_t reserved[3]; +} __rte_cache_aligned; + +struct rte_sched_queue { + uint16_t qw; + uint16_t qr; +}; + +struct rte_sched_queue_extra { + struct rte_sched_queue_stats stats; +#ifdef RTE_SCHED_RED + struct rte_red red; +#endif +}; + +enum grinder_state { + e_GRINDER_PREFETCH_PIPE = 0, + e_GRINDER_PREFETCH_TC_QUEUE_ARRAYS, + e_GRINDER_PREFETCH_MBUF, + e_GRINDER_READ_MBUF +}; + +/* + * Path through the scheduler hierarchy used by the scheduler enqueue + * operation to identify the destination queue for the current + * packet. Stored in the field pkt.hash.sched of struct rte_mbuf of + * each packet, typically written by the classification stage and read + * by scheduler enqueue. + */ +struct rte_sched_port_hierarchy { + uint16_t queue:2; /**< Queue ID (0 .. 3) */ + uint16_t traffic_class:2; /**< Traffic class ID (0 .. 3)*/ + uint32_t color:2; /**< Color */ + uint16_t unused:10; + uint16_t subport; /**< Subport ID */ + uint32_t pipe; /**< Pipe ID */ +}; + +struct rte_sched_grinder { + /* Pipe cache */ + uint16_t pcache_qmask[RTE_SCHED_GRINDER_PCACHE_SIZE]; + uint32_t pcache_qindex[RTE_SCHED_GRINDER_PCACHE_SIZE]; + uint32_t pcache_w; + uint32_t pcache_r; + + /* Current pipe */ + enum grinder_state state; + uint32_t productive; + uint32_t pindex; + struct rte_sched_subport *subport; + struct rte_sched_pipe *pipe; + struct rte_sched_pipe_profile *pipe_params; + + /* TC cache */ + uint8_t tccache_qmask[4]; + uint32_t tccache_qindex[4]; + uint32_t tccache_w; + uint32_t tccache_r; + + /* Current TC */ + uint32_t tc_index; + struct rte_sched_queue *queue[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE]; + struct rte_mbuf **qbase[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE]; + uint32_t qindex[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE]; + uint16_t qsize; + uint32_t qmask; + uint32_t qpos; + struct rte_mbuf *pkt; + + /* WRR */ + uint16_t wrr_tokens[RTE_SCHED_QUEUES_PER_TRAFFIC_CLASS]; + uint16_t wrr_mask[RTE_SCHED_QUEUES_PER_TRAFFIC_CLASS]; + uint8_t wrr_cost[RTE_SCHED_QUEUES_PER_TRAFFIC_CLASS]; +}; + +struct rte_sched_port { + /* User parameters */ + uint32_t n_subports_per_port; + uint32_t n_pipes_per_subport; + uint32_t rate; + uint32_t mtu; + uint32_t frame_overhead; + uint16_t qsize[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE]; + uint32_t n_pipe_profiles; + uint32_t pipe_tc3_rate_max; +#ifdef RTE_SCHED_RED + struct rte_red_config red_config[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE][e_RTE_METER_COLORS]; +#endif + + /* Timing */ + uint64_t time_cpu_cycles; /* Current CPU time measured in CPU cyles */ + uint64_t time_cpu_bytes; /* Current CPU time measured in bytes */ + uint64_t time; /* Current NIC TX time measured in bytes */ + struct rte_reciprocal inv_cycles_per_byte; /* CPU cycles per byte */ + + /* Scheduling loop detection */ + uint32_t pipe_loop; + uint32_t pipe_exhaustion; + + /* Bitmap */ + struct rte_bitmap *bmp; + uint32_t grinder_base_bmp_pos[RTE_SCHED_PORT_N_GRINDERS] __rte_aligned_16; + + /* Grinders */ + struct rte_sched_grinder grinder[RTE_SCHED_PORT_N_GRINDERS]; + uint32_t busy_grinders; + struct rte_mbuf **pkts_out; + uint32_t n_pkts_out; + + /* Queue base calculation */ + uint32_t qsize_add[RTE_SCHED_QUEUES_PER_PIPE]; + uint32_t qsize_sum; + + /* Large data structures */ + struct rte_sched_subport *subport; + struct rte_sched_pipe *pipe; + struct rte_sched_queue *queue; + struct rte_sched_queue_extra *queue_extra; + struct rte_sched_pipe_profile *pipe_profiles; + uint8_t *bmp_array; + struct rte_mbuf **queue_array; + uint8_t memory[0] __rte_cache_aligned; +} __rte_cache_aligned; + +enum rte_sched_port_array { + e_RTE_SCHED_PORT_ARRAY_SUBPORT = 0, + e_RTE_SCHED_PORT_ARRAY_PIPE, + e_RTE_SCHED_PORT_ARRAY_QUEUE, + e_RTE_SCHED_PORT_ARRAY_QUEUE_EXTRA, + e_RTE_SCHED_PORT_ARRAY_PIPE_PROFILES, + e_RTE_SCHED_PORT_ARRAY_BMP_ARRAY, + e_RTE_SCHED_PORT_ARRAY_QUEUE_ARRAY, + e_RTE_SCHED_PORT_ARRAY_TOTAL, +}; + +#ifdef RTE_SCHED_COLLECT_STATS + +static inline uint32_t +rte_sched_port_queues_per_subport(struct rte_sched_port *port) +{ + return RTE_SCHED_QUEUES_PER_PIPE * port->n_pipes_per_subport; +} + +#endif + +static inline uint32_t +rte_sched_port_queues_per_port(struct rte_sched_port *port) +{ + return RTE_SCHED_QUEUES_PER_PIPE * port->n_pipes_per_subport * port->n_subports_per_port; +} + +static inline struct rte_mbuf ** +rte_sched_port_qbase(struct rte_sched_port *port, uint32_t qindex) +{ + uint32_t pindex = qindex >> 4; + uint32_t qpos = qindex & 0xF; + + return (port->queue_array + pindex * + port->qsize_sum + port->qsize_add[qpos]); +} + +static inline uint16_t +rte_sched_port_qsize(struct rte_sched_port *port, uint32_t qindex) +{ + uint32_t tc = (qindex >> 2) & 0x3; + + return port->qsize[tc]; +} + +static int +pipe_profile_check(struct rte_sched_pipe_params *params, + uint32_t rate) +{ + uint32_t i; + + /* Pipe parameters */ + if (params == NULL) + return -10; + + /* TB rate: non-zero, not greater than port rate */ + if (params->tb_rate == 0 || + params->tb_rate > rate) + return -11; + + /* TB size: non-zero */ + if (params->tb_size == 0) + return -12; + + /* TC rate: non-zero, less than pipe rate */ + for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) { + if (params->tc_rate[i] == 0 || + params->tc_rate[i] > params->tb_rate) + return -13; + } + + /* TC period: non-zero */ + if (params->tc_period == 0) + return -14; + +#ifdef RTE_SCHED_SUBPORT_TC_OV + /* TC3 oversubscription weight: non-zero */ + if (params->tc_ov_weight == 0) + return -15; +#endif + + /* Queue WRR weights: non-zero */ + for (i = 0; i < RTE_SCHED_QUEUES_PER_PIPE; i++) { + if (params->wrr_weights[i] == 0) + return -16; + } + + return 0; +} + +static int +rte_sched_port_check_params(struct rte_sched_port_params *params) +{ + uint32_t i; + + if (params == NULL) + return -1; + + /* socket */ + if ((params->socket < 0) || (params->socket >= RTE_MAX_NUMA_NODES)) + return -3; + + /* rate */ + if (params->rate == 0) + return -4; + + /* mtu */ + if (params->mtu == 0) + return -5; + + /* n_subports_per_port: non-zero, limited to 16 bits, power of 2 */ + if (params->n_subports_per_port == 0 || + params->n_subports_per_port > 1u << 16 || + !rte_is_power_of_2(params->n_subports_per_port)) + return -6; + + /* n_pipes_per_subport: non-zero, power of 2 */ + if (params->n_pipes_per_subport == 0 || + !rte_is_power_of_2(params->n_pipes_per_subport)) + return -7; + + /* qsize: non-zero, power of 2, + * no bigger than 32K (due to 16-bit read/write pointers) + */ + for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) { + uint16_t qsize = params->qsize[i]; + + if (qsize == 0 || !rte_is_power_of_2(qsize)) + return -8; + } + + /* pipe_profiles and n_pipe_profiles */ + if (params->pipe_profiles == NULL || + params->n_pipe_profiles == 0 || + params->n_pipe_profiles > RTE_SCHED_PIPE_PROFILES_PER_PORT) + return -9; + + for (i = 0; i < params->n_pipe_profiles; i++) { + struct rte_sched_pipe_params *p = params->pipe_profiles + i; + int status; + + status = pipe_profile_check(p, params->rate); + if (status != 0) + return status; + } + + return 0; +} + +static uint32_t +rte_sched_port_get_array_base(struct rte_sched_port_params *params, enum rte_sched_port_array array) +{ + uint32_t n_subports_per_port = params->n_subports_per_port; + uint32_t n_pipes_per_subport = params->n_pipes_per_subport; + uint32_t n_pipes_per_port = n_pipes_per_subport * n_subports_per_port; + uint32_t n_queues_per_port = RTE_SCHED_QUEUES_PER_PIPE * n_pipes_per_subport * n_subports_per_port; + + uint32_t size_subport = n_subports_per_port * sizeof(struct rte_sched_subport); + uint32_t size_pipe = n_pipes_per_port * sizeof(struct rte_sched_pipe); + uint32_t size_queue = n_queues_per_port * sizeof(struct rte_sched_queue); + uint32_t size_queue_extra + = n_queues_per_port * sizeof(struct rte_sched_queue_extra); + uint32_t size_pipe_profiles + = RTE_SCHED_PIPE_PROFILES_PER_PORT * sizeof(struct rte_sched_pipe_profile); + uint32_t size_bmp_array = rte_bitmap_get_memory_footprint(n_queues_per_port); + uint32_t size_per_pipe_queue_array, size_queue_array; + + uint32_t base, i; + + size_per_pipe_queue_array = 0; + for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) { + size_per_pipe_queue_array += RTE_SCHED_QUEUES_PER_TRAFFIC_CLASS + * params->qsize[i] * sizeof(struct rte_mbuf *); + } + size_queue_array = n_pipes_per_port * size_per_pipe_queue_array; + + base = 0; + + if (array == e_RTE_SCHED_PORT_ARRAY_SUBPORT) + return base; + base += RTE_CACHE_LINE_ROUNDUP(size_subport); + + if (array == e_RTE_SCHED_PORT_ARRAY_PIPE) + return base; + base += RTE_CACHE_LINE_ROUNDUP(size_pipe); + + if (array == e_RTE_SCHED_PORT_ARRAY_QUEUE) + return base; + base += RTE_CACHE_LINE_ROUNDUP(size_queue); + + if (array == e_RTE_SCHED_PORT_ARRAY_QUEUE_EXTRA) + return base; + base += RTE_CACHE_LINE_ROUNDUP(size_queue_extra); + + if (array == e_RTE_SCHED_PORT_ARRAY_PIPE_PROFILES) + return base; + base += RTE_CACHE_LINE_ROUNDUP(size_pipe_profiles); + + if (array == e_RTE_SCHED_PORT_ARRAY_BMP_ARRAY) + return base; + base += RTE_CACHE_LINE_ROUNDUP(size_bmp_array); + + if (array == e_RTE_SCHED_PORT_ARRAY_QUEUE_ARRAY) + return base; + base += RTE_CACHE_LINE_ROUNDUP(size_queue_array); + + return base; +} + +uint32_t +rte_sched_port_get_memory_footprint(struct rte_sched_port_params *params) +{ + uint32_t size0, size1; + int status; + + status = rte_sched_port_check_params(params); + if (status != 0) { + RTE_LOG(NOTICE, SCHED, + "Port scheduler params check failed (%d)\n", status); + + return 0; + } + + size0 = sizeof(struct rte_sched_port); + size1 = rte_sched_port_get_array_base(params, e_RTE_SCHED_PORT_ARRAY_TOTAL); + + return size0 + size1; +} + +static void +rte_sched_port_config_qsize(struct rte_sched_port *port) +{ + /* TC 0 */ + port->qsize_add[0] = 0; + port->qsize_add[1] = port->qsize_add[0] + port->qsize[0]; + port->qsize_add[2] = port->qsize_add[1] + port->qsize[0]; + port->qsize_add[3] = port->qsize_add[2] + port->qsize[0]; + + /* TC 1 */ + port->qsize_add[4] = port->qsize_add[3] + port->qsize[0]; + port->qsize_add[5] = port->qsize_add[4] + port->qsize[1]; + port->qsize_add[6] = port->qsize_add[5] + port->qsize[1]; + port->qsize_add[7] = port->qsize_add[6] + port->qsize[1]; + + /* TC 2 */ + port->qsize_add[8] = port->qsize_add[7] + port->qsize[1]; + port->qsize_add[9] = port->qsize_add[8] + port->qsize[2]; + port->qsize_add[10] = port->qsize_add[9] + port->qsize[2]; + port->qsize_add[11] = port->qsize_add[10] + port->qsize[2]; + + /* TC 3 */ + port->qsize_add[12] = port->qsize_add[11] + port->qsize[2]; + port->qsize_add[13] = port->qsize_add[12] + port->qsize[3]; + port->qsize_add[14] = port->qsize_add[13] + port->qsize[3]; + port->qsize_add[15] = port->qsize_add[14] + port->qsize[3]; + + port->qsize_sum = port->qsize_add[15] + port->qsize[3]; +} + +static void +rte_sched_port_log_pipe_profile(struct rte_sched_port *port, uint32_t i) +{ + struct rte_sched_pipe_profile *p = port->pipe_profiles + i; + + RTE_LOG(DEBUG, SCHED, "Low level config for pipe profile %u:\n" + " Token bucket: period = %u, credits per period = %u, size = %u\n" + " Traffic classes: period = %u, credits per period = [%u, %u, %u, %u]\n" + " Traffic class 3 oversubscription: weight = %hhu\n" + " WRR cost: [%hhu, %hhu, %hhu, %hhu], [%hhu, %hhu, %hhu, %hhu], [%hhu, %hhu, %hhu, %hhu], [%hhu, %hhu, %hhu, %hhu]\n", + i, + + /* Token bucket */ + p->tb_period, + p->tb_credits_per_period, + p->tb_size, + + /* Traffic classes */ + p->tc_period, + p->tc_credits_per_period[0], + p->tc_credits_per_period[1], + p->tc_credits_per_period[2], + p->tc_credits_per_period[3], + + /* Traffic class 3 oversubscription */ + p->tc_ov_weight, + + /* WRR */ + p->wrr_cost[ 0], p->wrr_cost[ 1], p->wrr_cost[ 2], p->wrr_cost[ 3], + p->wrr_cost[ 4], p->wrr_cost[ 5], p->wrr_cost[ 6], p->wrr_cost[ 7], + p->wrr_cost[ 8], p->wrr_cost[ 9], p->wrr_cost[10], p->wrr_cost[11], + p->wrr_cost[12], p->wrr_cost[13], p->wrr_cost[14], p->wrr_cost[15]); +} + +static inline uint64_t +rte_sched_time_ms_to_bytes(uint32_t time_ms, uint32_t rate) +{ + uint64_t time = time_ms; + + time = (time * rate) / 1000; + + return time; +} + +static void +rte_sched_pipe_profile_convert(struct rte_sched_pipe_params *src, + struct rte_sched_pipe_profile *dst, + uint32_t rate) +{ + uint32_t i; + + /* Token Bucket */ + if (src->tb_rate == rate) { + dst->tb_credits_per_period = 1; + dst->tb_period = 1; + } else { + double tb_rate = (double) src->tb_rate + / (double) rate; + double d = RTE_SCHED_TB_RATE_CONFIG_ERR; + + rte_approx(tb_rate, d, + &dst->tb_credits_per_period, &dst->tb_period); + } + + dst->tb_size = src->tb_size; + + /* Traffic Classes */ + dst->tc_period = rte_sched_time_ms_to_bytes(src->tc_period, + rate); + + for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) + dst->tc_credits_per_period[i] + = rte_sched_time_ms_to_bytes(src->tc_period, + src->tc_rate[i]); + +#ifdef RTE_SCHED_SUBPORT_TC_OV + dst->tc_ov_weight = src->tc_ov_weight; +#endif + + /* WRR */ + for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) { + uint32_t wrr_cost[RTE_SCHED_QUEUES_PER_TRAFFIC_CLASS]; + uint32_t lcd, lcd1, lcd2; + uint32_t qindex; + + qindex = i * RTE_SCHED_QUEUES_PER_TRAFFIC_CLASS; + + wrr_cost[0] = src->wrr_weights[qindex]; + wrr_cost[1] = src->wrr_weights[qindex + 1]; + wrr_cost[2] = src->wrr_weights[qindex + 2]; + wrr_cost[3] = src->wrr_weights[qindex + 3]; + + lcd1 = rte_get_lcd(wrr_cost[0], wrr_cost[1]); + lcd2 = rte_get_lcd(wrr_cost[2], wrr_cost[3]); + lcd = rte_get_lcd(lcd1, lcd2); + + wrr_cost[0] = lcd / wrr_cost[0]; + wrr_cost[1] = lcd / wrr_cost[1]; + wrr_cost[2] = lcd / wrr_cost[2]; + wrr_cost[3] = lcd / wrr_cost[3]; + + dst->wrr_cost[qindex] = (uint8_t) wrr_cost[0]; + dst->wrr_cost[qindex + 1] = (uint8_t) wrr_cost[1]; + dst->wrr_cost[qindex + 2] = (uint8_t) wrr_cost[2]; + dst->wrr_cost[qindex + 3] = (uint8_t) wrr_cost[3]; + } +} + +static void +rte_sched_port_config_pipe_profile_table(struct rte_sched_port *port, + struct rte_sched_port_params *params) +{ + uint32_t i; + + for (i = 0; i < port->n_pipe_profiles; i++) { + struct rte_sched_pipe_params *src = params->pipe_profiles + i; + struct rte_sched_pipe_profile *dst = port->pipe_profiles + i; + + rte_sched_pipe_profile_convert(src, dst, params->rate); + rte_sched_port_log_pipe_profile(port, i); + } + + port->pipe_tc3_rate_max = 0; + for (i = 0; i < port->n_pipe_profiles; i++) { + struct rte_sched_pipe_params *src = params->pipe_profiles + i; + uint32_t pipe_tc3_rate = src->tc_rate[3]; + + if (port->pipe_tc3_rate_max < pipe_tc3_rate) + port->pipe_tc3_rate_max = pipe_tc3_rate; + } +} + +struct rte_sched_port * +rte_sched_port_config(struct rte_sched_port_params *params) +{ + struct rte_sched_port *port = NULL; + uint32_t mem_size, bmp_mem_size, n_queues_per_port, i, cycles_per_byte; + + /* Check user parameters. Determine the amount of memory to allocate */ + mem_size = rte_sched_port_get_memory_footprint(params); + if (mem_size == 0) + return NULL; + + /* Allocate memory to store the data structures */ + port = rte_zmalloc("qos_params", mem_size, RTE_CACHE_LINE_SIZE); + if (port == NULL) + return NULL; + + /* compile time checks */ + RTE_BUILD_BUG_ON(RTE_SCHED_PORT_N_GRINDERS == 0); + RTE_BUILD_BUG_ON(RTE_SCHED_PORT_N_GRINDERS & (RTE_SCHED_PORT_N_GRINDERS - 1)); + + /* User parameters */ + port->n_subports_per_port = params->n_subports_per_port; + port->n_pipes_per_subport = params->n_pipes_per_subport; + port->rate = params->rate; + port->mtu = params->mtu + params->frame_overhead; + port->frame_overhead = params->frame_overhead; + memcpy(port->qsize, params->qsize, sizeof(params->qsize)); + port->n_pipe_profiles = params->n_pipe_profiles; + +#ifdef RTE_SCHED_RED + for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) { + uint32_t j; + + for (j = 0; j < e_RTE_METER_COLORS; j++) { + /* if min/max are both zero, then RED is disabled */ + if ((params->red_params[i][j].min_th | + params->red_params[i][j].max_th) == 0) { + continue; + } + + if (rte_red_config_init(&port->red_config[i][j], + params->red_params[i][j].wq_log2, + params->red_params[i][j].min_th, + params->red_params[i][j].max_th, + params->red_params[i][j].maxp_inv) != 0) { + return NULL; + } + } + } +#endif + + /* Timing */ + port->time_cpu_cycles = rte_get_tsc_cycles(); + port->time_cpu_bytes = 0; + port->time = 0; + + cycles_per_byte = (rte_get_tsc_hz() << RTE_SCHED_TIME_SHIFT) + / params->rate; + port->inv_cycles_per_byte = rte_reciprocal_value(cycles_per_byte); + + /* Scheduling loop detection */ + port->pipe_loop = RTE_SCHED_PIPE_INVALID; + port->pipe_exhaustion = 0; + + /* Grinders */ + port->busy_grinders = 0; + port->pkts_out = NULL; + port->n_pkts_out = 0; + + /* Queue base calculation */ + rte_sched_port_config_qsize(port); + + /* Large data structures */ + port->subport = (struct rte_sched_subport *) + (port->memory + rte_sched_port_get_array_base(params, + e_RTE_SCHED_PORT_ARRAY_SUBPORT)); + port->pipe = (struct rte_sched_pipe *) + (port->memory + rte_sched_port_get_array_base(params, + e_RTE_SCHED_PORT_ARRAY_PIPE)); + port->queue = (struct rte_sched_queue *) + (port->memory + rte_sched_port_get_array_base(params, + e_RTE_SCHED_PORT_ARRAY_QUEUE)); + port->queue_extra = (struct rte_sched_queue_extra *) + (port->memory + rte_sched_port_get_array_base(params, + e_RTE_SCHED_PORT_ARRAY_QUEUE_EXTRA)); + port->pipe_profiles = (struct rte_sched_pipe_profile *) + (port->memory + rte_sched_port_get_array_base(params, + e_RTE_SCHED_PORT_ARRAY_PIPE_PROFILES)); + port->bmp_array = port->memory + + rte_sched_port_get_array_base(params, e_RTE_SCHED_PORT_ARRAY_BMP_ARRAY); + port->queue_array = (struct rte_mbuf **) + (port->memory + rte_sched_port_get_array_base(params, + e_RTE_SCHED_PORT_ARRAY_QUEUE_ARRAY)); + + /* Pipe profile table */ + rte_sched_port_config_pipe_profile_table(port, params); + + /* Bitmap */ + n_queues_per_port = rte_sched_port_queues_per_port(port); + bmp_mem_size = rte_bitmap_get_memory_footprint(n_queues_per_port); + port->bmp = rte_bitmap_init(n_queues_per_port, port->bmp_array, + bmp_mem_size); + if (port->bmp == NULL) { + RTE_LOG(ERR, SCHED, "Bitmap init error\n"); + return NULL; + } + + for (i = 0; i < RTE_SCHED_PORT_N_GRINDERS; i++) + port->grinder_base_bmp_pos[i] = RTE_SCHED_PIPE_INVALID; + + + return port; +} + +void +rte_sched_port_free(struct rte_sched_port *port) +{ + uint32_t qindex; + uint32_t n_queues_per_port; + + /* Check user parameters */ + if (port == NULL) + return; + + n_queues_per_port = rte_sched_port_queues_per_port(port); + + /* Free enqueued mbufs */ + for (qindex = 0; qindex < n_queues_per_port; qindex++) { + struct rte_mbuf **mbufs = rte_sched_port_qbase(port, qindex); + uint16_t qsize = rte_sched_port_qsize(port, qindex); + struct rte_sched_queue *queue = port->queue + qindex; + uint16_t qr = queue->qr & (qsize - 1); + uint16_t qw = queue->qw & (qsize - 1); + + for (; qr != qw; qr = (qr + 1) & (qsize - 1)) + rte_pktmbuf_free(mbufs[qr]); + } + + rte_bitmap_free(port->bmp); + rte_free(port); +} + +static void +rte_sched_port_log_subport_config(struct rte_sched_port *port, uint32_t i) +{ + struct rte_sched_subport *s = port->subport + i; + + RTE_LOG(DEBUG, SCHED, "Low level config for subport %u:\n" + " Token bucket: period = %u, credits per period = %u, size = %u\n" + " Traffic classes: period = %u, credits per period = [%u, %u, %u, %u]\n" + " Traffic class 3 oversubscription: wm min = %u, wm max = %u\n", + i, + + /* Token bucket */ + s->tb_period, + s->tb_credits_per_period, + s->tb_size, + + /* Traffic classes */ + s->tc_period, + s->tc_credits_per_period[0], + s->tc_credits_per_period[1], + s->tc_credits_per_period[2], + s->tc_credits_per_period[3], + + /* Traffic class 3 oversubscription */ + s->tc_ov_wm_min, + s->tc_ov_wm_max); +} + +int +rte_sched_subport_config(struct rte_sched_port *port, + uint32_t subport_id, + struct rte_sched_subport_params *params) +{ + struct rte_sched_subport *s; + uint32_t i; + + /* Check user parameters */ + if (port == NULL || + subport_id >= port->n_subports_per_port || + params == NULL) + return -1; + + if (params->tb_rate == 0 || params->tb_rate > port->rate) + return -2; + + if (params->tb_size == 0) + return -3; + + for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) { + if (params->tc_rate[i] == 0 || + params->tc_rate[i] > params->tb_rate) + return -4; + } + + if (params->tc_period == 0) + return -5; + + s = port->subport + subport_id; + + /* Token Bucket (TB) */ + if (params->tb_rate == port->rate) { + s->tb_credits_per_period = 1; + s->tb_period = 1; + } else { + double tb_rate = ((double) params->tb_rate) / ((double) port->rate); + double d = RTE_SCHED_TB_RATE_CONFIG_ERR; + + rte_approx(tb_rate, d, &s->tb_credits_per_period, &s->tb_period); + } + + s->tb_size = params->tb_size; + s->tb_time = port->time; + s->tb_credits = s->tb_size / 2; + + /* Traffic Classes (TCs) */ + s->tc_period = rte_sched_time_ms_to_bytes(params->tc_period, port->rate); + for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) { + s->tc_credits_per_period[i] + = rte_sched_time_ms_to_bytes(params->tc_period, + params->tc_rate[i]); + } + s->tc_time = port->time + s->tc_period; + for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) + s->tc_credits[i] = s->tc_credits_per_period[i]; + +#ifdef RTE_SCHED_SUBPORT_TC_OV + /* TC oversubscription */ + s->tc_ov_wm_min = port->mtu; + s->tc_ov_wm_max = rte_sched_time_ms_to_bytes(params->tc_period, + port->pipe_tc3_rate_max); + s->tc_ov_wm = s->tc_ov_wm_max; + s->tc_ov_period_id = 0; + s->tc_ov = 0; + s->tc_ov_n = 0; + s->tc_ov_rate = 0; +#endif + + rte_sched_port_log_subport_config(port, subport_id); + + return 0; +} + +int +rte_sched_pipe_config(struct rte_sched_port *port, + uint32_t subport_id, + uint32_t pipe_id, + int32_t pipe_profile) +{ + struct rte_sched_subport *s; + struct rte_sched_pipe *p; + struct rte_sched_pipe_profile *params; + uint32_t deactivate, profile, i; + + /* Check user parameters */ + profile = (uint32_t) pipe_profile; + deactivate = (pipe_profile < 0); + + if (port == NULL || + subport_id >= port->n_subports_per_port || + pipe_id >= port->n_pipes_per_subport || + (!deactivate && profile >= port->n_pipe_profiles)) + return -1; + + + /* Check that subport configuration is valid */ + s = port->subport + subport_id; + if (s->tb_period == 0) + return -2; + + p = port->pipe + (subport_id * port->n_pipes_per_subport + pipe_id); + + /* Handle the case when pipe already has a valid configuration */ + if (p->tb_time) { + params = port->pipe_profiles + p->profile; + +#ifdef RTE_SCHED_SUBPORT_TC_OV + double subport_tc3_rate = (double) s->tc_credits_per_period[3] + / (double) s->tc_period; + double pipe_tc3_rate = (double) params->tc_credits_per_period[3] + / (double) params->tc_period; + uint32_t tc3_ov = s->tc_ov; + + /* Unplug pipe from its subport */ + s->tc_ov_n -= params->tc_ov_weight; + s->tc_ov_rate -= pipe_tc3_rate; + s->tc_ov = s->tc_ov_rate > subport_tc3_rate; + + if (s->tc_ov != tc3_ov) { + RTE_LOG(DEBUG, SCHED, + "Subport %u TC3 oversubscription is OFF (%.4lf >= %.4lf)\n", + subport_id, subport_tc3_rate, s->tc_ov_rate); + } +#endif + + /* Reset the pipe */ + memset(p, 0, sizeof(struct rte_sched_pipe)); + } + + if (deactivate) + return 0; + + /* Apply the new pipe configuration */ + p->profile = profile; + params = port->pipe_profiles + p->profile; + + /* Token Bucket (TB) */ + p->tb_time = port->time; + p->tb_credits = params->tb_size / 2; + + /* Traffic Classes (TCs) */ + p->tc_time = port->time + params->tc_period; + for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) + p->tc_credits[i] = params->tc_credits_per_period[i]; + +#ifdef RTE_SCHED_SUBPORT_TC_OV + { + /* Subport TC3 oversubscription */ + double subport_tc3_rate = (double) s->tc_credits_per_period[3] + / (double) s->tc_period; + double pipe_tc3_rate = (double) params->tc_credits_per_period[3] + / (double) params->tc_period; + uint32_t tc3_ov = s->tc_ov; + + s->tc_ov_n += params->tc_ov_weight; + s->tc_ov_rate += pipe_tc3_rate; + s->tc_ov = s->tc_ov_rate > subport_tc3_rate; + + if (s->tc_ov != tc3_ov) { + RTE_LOG(DEBUG, SCHED, + "Subport %u TC3 oversubscription is ON (%.4lf < %.4lf)\n", + subport_id, subport_tc3_rate, s->tc_ov_rate); + } + p->tc_ov_period_id = s->tc_ov_period_id; + p->tc_ov_credits = s->tc_ov_wm; + } +#endif + + return 0; +} + +int __rte_experimental +rte_sched_port_pipe_profile_add(struct rte_sched_port *port, + struct rte_sched_pipe_params *params, + uint32_t *pipe_profile_id) +{ + struct rte_sched_pipe_profile *pp; + uint32_t i; + int status; + + /* Port */ + if (port == NULL) + return -1; + + /* Pipe profiles not exceeds the max limit */ + if (port->n_pipe_profiles >= RTE_SCHED_PIPE_PROFILES_PER_PORT) + return -2; + + /* Pipe params */ + status = pipe_profile_check(params, port->rate); + if (status != 0) + return status; + + pp = &port->pipe_profiles[port->n_pipe_profiles]; + rte_sched_pipe_profile_convert(params, pp, port->rate); + + /* Pipe profile not exists */ + for (i = 0; i < port->n_pipe_profiles; i++) + if (memcmp(port->pipe_profiles + i, pp, sizeof(*pp)) == 0) + return -3; + + /* Pipe profile commit */ + *pipe_profile_id = port->n_pipe_profiles; + port->n_pipe_profiles++; + + if (port->pipe_tc3_rate_max < params->tc_rate[3]) + port->pipe_tc3_rate_max = params->tc_rate[3]; + + rte_sched_port_log_pipe_profile(port, *pipe_profile_id); + + return 0; +} + +void +rte_sched_port_pkt_write(struct rte_mbuf *pkt, + uint32_t subport, uint32_t pipe, uint32_t traffic_class, + uint32_t queue, enum rte_meter_color color) +{ + struct rte_sched_port_hierarchy *sched + = (struct rte_sched_port_hierarchy *) &pkt->hash.sched; + + RTE_BUILD_BUG_ON(sizeof(*sched) > sizeof(pkt->hash.sched)); + + sched->color = (uint32_t) color; + sched->subport = subport; + sched->pipe = pipe; + sched->traffic_class = traffic_class; + sched->queue = queue; +} + +void +rte_sched_port_pkt_read_tree_path(const struct rte_mbuf *pkt, + uint32_t *subport, uint32_t *pipe, + uint32_t *traffic_class, uint32_t *queue) +{ + const struct rte_sched_port_hierarchy *sched + = (const struct rte_sched_port_hierarchy *) &pkt->hash.sched; + + *subport = sched->subport; + *pipe = sched->pipe; + *traffic_class = sched->traffic_class; + *queue = sched->queue; +} + +enum rte_meter_color +rte_sched_port_pkt_read_color(const struct rte_mbuf *pkt) +{ + const struct rte_sched_port_hierarchy *sched + = (const struct rte_sched_port_hierarchy *) &pkt->hash.sched; + + return (enum rte_meter_color) sched->color; +} + +int +rte_sched_subport_read_stats(struct rte_sched_port *port, + uint32_t subport_id, + struct rte_sched_subport_stats *stats, + uint32_t *tc_ov) +{ + struct rte_sched_subport *s; + + /* Check user parameters */ + if (port == NULL || subport_id >= port->n_subports_per_port || + stats == NULL || tc_ov == NULL) + return -1; + + s = port->subport + subport_id; + + /* Copy subport stats and clear */ + memcpy(stats, &s->stats, sizeof(struct rte_sched_subport_stats)); + memset(&s->stats, 0, sizeof(struct rte_sched_subport_stats)); + + /* Subport TC oversubscription status */ + *tc_ov = s->tc_ov; + + return 0; +} + +int +rte_sched_queue_read_stats(struct rte_sched_port *port, + uint32_t queue_id, + struct rte_sched_queue_stats *stats, + uint16_t *qlen) +{ + struct rte_sched_queue *q; + struct rte_sched_queue_extra *qe; + + /* Check user parameters */ + if ((port == NULL) || + (queue_id >= rte_sched_port_queues_per_port(port)) || + (stats == NULL) || + (qlen == NULL)) { + return -1; + } + q = port->queue + queue_id; + qe = port->queue_extra + queue_id; + + /* Copy queue stats and clear */ + memcpy(stats, &qe->stats, sizeof(struct rte_sched_queue_stats)); + memset(&qe->stats, 0, sizeof(struct rte_sched_queue_stats)); + + /* Queue length */ + *qlen = q->qw - q->qr; + + return 0; +} + +static inline uint32_t +rte_sched_port_qindex(struct rte_sched_port *port, uint32_t subport, uint32_t pipe, uint32_t traffic_class, uint32_t queue) +{ + uint32_t result; + + result = subport * port->n_pipes_per_subport + pipe; + result = result * RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE + traffic_class; + result = result * RTE_SCHED_QUEUES_PER_TRAFFIC_CLASS + queue; + + return result; +} + +#ifdef RTE_SCHED_DEBUG + +static inline int +rte_sched_port_queue_is_empty(struct rte_sched_port *port, uint32_t qindex) +{ + struct rte_sched_queue *queue = port->queue + qindex; + + return queue->qr == queue->qw; +} + +#endif /* RTE_SCHED_DEBUG */ + +#ifdef RTE_SCHED_COLLECT_STATS + +static inline void +rte_sched_port_update_subport_stats(struct rte_sched_port *port, uint32_t qindex, struct rte_mbuf *pkt) +{ + struct rte_sched_subport *s = port->subport + (qindex / rte_sched_port_queues_per_subport(port)); + uint32_t tc_index = (qindex >> 2) & 0x3; + uint32_t pkt_len = pkt->pkt_len; + + s->stats.n_pkts_tc[tc_index] += 1; + s->stats.n_bytes_tc[tc_index] += pkt_len; +} + +#ifdef RTE_SCHED_RED +static inline void +rte_sched_port_update_subport_stats_on_drop(struct rte_sched_port *port, + uint32_t qindex, + struct rte_mbuf *pkt, uint32_t red) +#else +static inline void +rte_sched_port_update_subport_stats_on_drop(struct rte_sched_port *port, + uint32_t qindex, + struct rte_mbuf *pkt, __rte_unused uint32_t red) +#endif +{ + struct rte_sched_subport *s = port->subport + (qindex / rte_sched_port_queues_per_subport(port)); + uint32_t tc_index = (qindex >> 2) & 0x3; + uint32_t pkt_len = pkt->pkt_len; + + s->stats.n_pkts_tc_dropped[tc_index] += 1; + s->stats.n_bytes_tc_dropped[tc_index] += pkt_len; +#ifdef RTE_SCHED_RED + s->stats.n_pkts_red_dropped[tc_index] += red; +#endif +} + +static inline void +rte_sched_port_update_queue_stats(struct rte_sched_port *port, uint32_t qindex, struct rte_mbuf *pkt) +{ + struct rte_sched_queue_extra *qe = port->queue_extra + qindex; + uint32_t pkt_len = pkt->pkt_len; + + qe->stats.n_pkts += 1; + qe->stats.n_bytes += pkt_len; +} + +#ifdef RTE_SCHED_RED +static inline void +rte_sched_port_update_queue_stats_on_drop(struct rte_sched_port *port, + uint32_t qindex, + struct rte_mbuf *pkt, uint32_t red) +#else +static inline void +rte_sched_port_update_queue_stats_on_drop(struct rte_sched_port *port, + uint32_t qindex, + struct rte_mbuf *pkt, __rte_unused uint32_t red) +#endif +{ + struct rte_sched_queue_extra *qe = port->queue_extra + qindex; + uint32_t pkt_len = pkt->pkt_len; + + qe->stats.n_pkts_dropped += 1; + qe->stats.n_bytes_dropped += pkt_len; +#ifdef RTE_SCHED_RED + qe->stats.n_pkts_red_dropped += red; +#endif +} + +#endif /* RTE_SCHED_COLLECT_STATS */ + +#ifdef RTE_SCHED_RED + +static inline int +rte_sched_port_red_drop(struct rte_sched_port *port, struct rte_mbuf *pkt, uint32_t qindex, uint16_t qlen) +{ + struct rte_sched_queue_extra *qe; + struct rte_red_config *red_cfg; + struct rte_red *red; + uint32_t tc_index; + enum rte_meter_color color; + + tc_index = (qindex >> 2) & 0x3; + color = rte_sched_port_pkt_read_color(pkt); + red_cfg = &port->red_config[tc_index][color]; + + if ((red_cfg->min_th | red_cfg->max_th) == 0) + return 0; + + qe = port->queue_extra + qindex; + red = &qe->red; + + return rte_red_enqueue(red_cfg, red, qlen, port->time); +} + +static inline void +rte_sched_port_set_queue_empty_timestamp(struct rte_sched_port *port, uint32_t qindex) +{ + struct rte_sched_queue_extra *qe = port->queue_extra + qindex; + struct rte_red *red = &qe->red; + + rte_red_mark_queue_empty(red, port->time); +} + +#else + +#define rte_sched_port_red_drop(port, pkt, qindex, qlen) 0 + +#define rte_sched_port_set_queue_empty_timestamp(port, qindex) + +#endif /* RTE_SCHED_RED */ + +#ifdef RTE_SCHED_DEBUG + +static inline void +debug_check_queue_slab(struct rte_sched_port *port, uint32_t bmp_pos, + uint64_t bmp_slab) +{ + uint64_t mask; + uint32_t i, panic; + + if (bmp_slab == 0) + rte_panic("Empty slab at position %u\n", bmp_pos); + + panic = 0; + for (i = 0, mask = 1; i < 64; i++, mask <<= 1) { + if (mask & bmp_slab) { + if (rte_sched_port_queue_is_empty(port, bmp_pos + i)) { + printf("Queue %u (slab offset %u) is empty\n", bmp_pos + i, i); + panic = 1; + } + } + } + + if (panic) + rte_panic("Empty queues in slab 0x%" PRIx64 "starting at position %u\n", + bmp_slab, bmp_pos); +} + +#endif /* RTE_SCHED_DEBUG */ + +static inline uint32_t +rte_sched_port_enqueue_qptrs_prefetch0(struct rte_sched_port *port, + struct rte_mbuf *pkt) +{ + struct rte_sched_queue *q; +#ifdef RTE_SCHED_COLLECT_STATS + struct rte_sched_queue_extra *qe; +#endif + uint32_t subport, pipe, traffic_class, queue, qindex; + + rte_sched_port_pkt_read_tree_path(pkt, &subport, &pipe, &traffic_class, &queue); + + qindex = rte_sched_port_qindex(port, subport, pipe, traffic_class, queue); + q = port->queue + qindex; + rte_prefetch0(q); +#ifdef RTE_SCHED_COLLECT_STATS + qe = port->queue_extra + qindex; + rte_prefetch0(qe); +#endif + + return qindex; +} + +static inline void +rte_sched_port_enqueue_qwa_prefetch0(struct rte_sched_port *port, + uint32_t qindex, struct rte_mbuf **qbase) +{ + struct rte_sched_queue *q; + struct rte_mbuf **q_qw; + uint16_t qsize; + + q = port->queue + qindex; + qsize = rte_sched_port_qsize(port, qindex); + q_qw = qbase + (q->qw & (qsize - 1)); + + rte_prefetch0(q_qw); + rte_bitmap_prefetch0(port->bmp, qindex); +} + +static inline int +rte_sched_port_enqueue_qwa(struct rte_sched_port *port, uint32_t qindex, + struct rte_mbuf **qbase, struct rte_mbuf *pkt) +{ + struct rte_sched_queue *q; + uint16_t qsize; + uint16_t qlen; + + q = port->queue + qindex; + qsize = rte_sched_port_qsize(port, qindex); + qlen = q->qw - q->qr; + + /* Drop the packet (and update drop stats) when queue is full */ + if (unlikely(rte_sched_port_red_drop(port, pkt, qindex, qlen) || + (qlen >= qsize))) { + rte_pktmbuf_free(pkt); +#ifdef RTE_SCHED_COLLECT_STATS + rte_sched_port_update_subport_stats_on_drop(port, qindex, pkt, + qlen < qsize); + rte_sched_port_update_queue_stats_on_drop(port, qindex, pkt, + qlen < qsize); +#endif + return 0; + } + + /* Enqueue packet */ + qbase[q->qw & (qsize - 1)] = pkt; + q->qw++; + + /* Activate queue in the port bitmap */ + rte_bitmap_set(port->bmp, qindex); + + /* Statistics */ +#ifdef RTE_SCHED_COLLECT_STATS + rte_sched_port_update_subport_stats(port, qindex, pkt); + rte_sched_port_update_queue_stats(port, qindex, pkt); +#endif + + return 1; +} + + +/* + * The enqueue function implements a 4-level pipeline with each stage + * processing two different packets. The purpose of using a pipeline + * is to hide the latency of prefetching the data structures. The + * naming convention is presented in the diagram below: + * + * p00 _______ p10 _______ p20 _______ p30 _______ + * ----->| |----->| |----->| |----->| |-----> + * | 0 | | 1 | | 2 | | 3 | + * ----->|_______|----->|_______|----->|_______|----->|_______|-----> + * p01 p11 p21 p31 + * + */ +int +rte_sched_port_enqueue(struct rte_sched_port *port, struct rte_mbuf **pkts, + uint32_t n_pkts) +{ + struct rte_mbuf *pkt00, *pkt01, *pkt10, *pkt11, *pkt20, *pkt21, + *pkt30, *pkt31, *pkt_last; + struct rte_mbuf **q00_base, **q01_base, **q10_base, **q11_base, + **q20_base, **q21_base, **q30_base, **q31_base, **q_last_base; + uint32_t q00, q01, q10, q11, q20, q21, q30, q31, q_last; + uint32_t r00, r01, r10, r11, r20, r21, r30, r31, r_last; + uint32_t result, i; + + result = 0; + + /* + * Less then 6 input packets available, which is not enough to + * feed the pipeline + */ + if (unlikely(n_pkts < 6)) { + struct rte_mbuf **q_base[5]; + uint32_t q[5]; + + /* Prefetch the mbuf structure of each packet */ + for (i = 0; i < n_pkts; i++) + rte_prefetch0(pkts[i]); + + /* Prefetch the queue structure for each queue */ + for (i = 0; i < n_pkts; i++) + q[i] = rte_sched_port_enqueue_qptrs_prefetch0(port, + pkts[i]); + + /* Prefetch the write pointer location of each queue */ + for (i = 0; i < n_pkts; i++) { + q_base[i] = rte_sched_port_qbase(port, q[i]); + rte_sched_port_enqueue_qwa_prefetch0(port, q[i], + q_base[i]); + } + + /* Write each packet to its queue */ + for (i = 0; i < n_pkts; i++) + result += rte_sched_port_enqueue_qwa(port, q[i], + q_base[i], pkts[i]); + + return result; + } + + /* Feed the first 3 stages of the pipeline (6 packets needed) */ + pkt20 = pkts[0]; + pkt21 = pkts[1]; + rte_prefetch0(pkt20); + rte_prefetch0(pkt21); + + pkt10 = pkts[2]; + pkt11 = pkts[3]; + rte_prefetch0(pkt10); + rte_prefetch0(pkt11); + + q20 = rte_sched_port_enqueue_qptrs_prefetch0(port, pkt20); + q21 = rte_sched_port_enqueue_qptrs_prefetch0(port, pkt21); + + pkt00 = pkts[4]; + pkt01 = pkts[5]; + rte_prefetch0(pkt00); + rte_prefetch0(pkt01); + + q10 = rte_sched_port_enqueue_qptrs_prefetch0(port, pkt10); + q11 = rte_sched_port_enqueue_qptrs_prefetch0(port, pkt11); + + q20_base = rte_sched_port_qbase(port, q20); + q21_base = rte_sched_port_qbase(port, q21); + rte_sched_port_enqueue_qwa_prefetch0(port, q20, q20_base); + rte_sched_port_enqueue_qwa_prefetch0(port, q21, q21_base); + + /* Run the pipeline */ + for (i = 6; i < (n_pkts & (~1)); i += 2) { + /* Propagate stage inputs */ + pkt30 = pkt20; + pkt31 = pkt21; + pkt20 = pkt10; + pkt21 = pkt11; + pkt10 = pkt00; + pkt11 = pkt01; + q30 = q20; + q31 = q21; + q20 = q10; + q21 = q11; + q30_base = q20_base; + q31_base = q21_base; + + /* Stage 0: Get packets in */ + pkt00 = pkts[i]; + pkt01 = pkts[i + 1]; + rte_prefetch0(pkt00); + rte_prefetch0(pkt01); + + /* Stage 1: Prefetch queue structure storing queue pointers */ + q10 = rte_sched_port_enqueue_qptrs_prefetch0(port, pkt10); + q11 = rte_sched_port_enqueue_qptrs_prefetch0(port, pkt11); + + /* Stage 2: Prefetch queue write location */ + q20_base = rte_sched_port_qbase(port, q20); + q21_base = rte_sched_port_qbase(port, q21); + rte_sched_port_enqueue_qwa_prefetch0(port, q20, q20_base); + rte_sched_port_enqueue_qwa_prefetch0(port, q21, q21_base); + + /* Stage 3: Write packet to queue and activate queue */ + r30 = rte_sched_port_enqueue_qwa(port, q30, q30_base, pkt30); + r31 = rte_sched_port_enqueue_qwa(port, q31, q31_base, pkt31); + result += r30 + r31; + } + + /* + * Drain the pipeline (exactly 6 packets). + * Handle the last packet in the case + * of an odd number of input packets. + */ + pkt_last = pkts[n_pkts - 1]; + rte_prefetch0(pkt_last); + + q00 = rte_sched_port_enqueue_qptrs_prefetch0(port, pkt00); + q01 = rte_sched_port_enqueue_qptrs_prefetch0(port, pkt01); + + q10_base = rte_sched_port_qbase(port, q10); + q11_base = rte_sched_port_qbase(port, q11); + rte_sched_port_enqueue_qwa_prefetch0(port, q10, q10_base); + rte_sched_port_enqueue_qwa_prefetch0(port, q11, q11_base); + + r20 = rte_sched_port_enqueue_qwa(port, q20, q20_base, pkt20); + r21 = rte_sched_port_enqueue_qwa(port, q21, q21_base, pkt21); + result += r20 + r21; + + q_last = rte_sched_port_enqueue_qptrs_prefetch0(port, pkt_last); + + q00_base = rte_sched_port_qbase(port, q00); + q01_base = rte_sched_port_qbase(port, q01); + rte_sched_port_enqueue_qwa_prefetch0(port, q00, q00_base); + rte_sched_port_enqueue_qwa_prefetch0(port, q01, q01_base); + + r10 = rte_sched_port_enqueue_qwa(port, q10, q10_base, pkt10); + r11 = rte_sched_port_enqueue_qwa(port, q11, q11_base, pkt11); + result += r10 + r11; + + q_last_base = rte_sched_port_qbase(port, q_last); + rte_sched_port_enqueue_qwa_prefetch0(port, q_last, q_last_base); + + r00 = rte_sched_port_enqueue_qwa(port, q00, q00_base, pkt00); + r01 = rte_sched_port_enqueue_qwa(port, q01, q01_base, pkt01); + result += r00 + r01; + + if (n_pkts & 1) { + r_last = rte_sched_port_enqueue_qwa(port, q_last, q_last_base, pkt_last); + result += r_last; + } + + return result; +} + +#ifndef RTE_SCHED_SUBPORT_TC_OV + +static inline void +grinder_credits_update(struct rte_sched_port *port, uint32_t pos) +{ + struct rte_sched_grinder *grinder = port->grinder + pos; + struct rte_sched_subport *subport = grinder->subport; + struct rte_sched_pipe *pipe = grinder->pipe; + struct rte_sched_pipe_profile *params = grinder->pipe_params; + uint64_t n_periods; + + /* Subport TB */ + n_periods = (port->time - subport->tb_time) / subport->tb_period; + subport->tb_credits += n_periods * subport->tb_credits_per_period; + subport->tb_credits = rte_sched_min_val_2_u32(subport->tb_credits, subport->tb_size); + subport->tb_time += n_periods * subport->tb_period; + + /* Pipe TB */ + n_periods = (port->time - pipe->tb_time) / params->tb_period; + pipe->tb_credits += n_periods * params->tb_credits_per_period; + pipe->tb_credits = rte_sched_min_val_2_u32(pipe->tb_credits, params->tb_size); + pipe->tb_time += n_periods * params->tb_period; + + /* Subport TCs */ + if (unlikely(port->time >= subport->tc_time)) { + subport->tc_credits[0] = subport->tc_credits_per_period[0]; + subport->tc_credits[1] = subport->tc_credits_per_period[1]; + subport->tc_credits[2] = subport->tc_credits_per_period[2]; + subport->tc_credits[3] = subport->tc_credits_per_period[3]; + subport->tc_time = port->time + subport->tc_period; + } + + /* Pipe TCs */ + if (unlikely(port->time >= pipe->tc_time)) { + pipe->tc_credits[0] = params->tc_credits_per_period[0]; + pipe->tc_credits[1] = params->tc_credits_per_period[1]; + pipe->tc_credits[2] = params->tc_credits_per_period[2]; + pipe->tc_credits[3] = params->tc_credits_per_period[3]; + pipe->tc_time = port->time + params->tc_period; + } +} + +#else + +static inline uint32_t +grinder_tc_ov_credits_update(struct rte_sched_port *port, uint32_t pos) +{ + struct rte_sched_grinder *grinder = port->grinder + pos; + struct rte_sched_subport *subport = grinder->subport; + uint32_t tc_ov_consumption[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE]; + uint32_t tc_ov_consumption_max; + uint32_t tc_ov_wm = subport->tc_ov_wm; + + if (subport->tc_ov == 0) + return subport->tc_ov_wm_max; + + tc_ov_consumption[0] = subport->tc_credits_per_period[0] - subport->tc_credits[0]; + tc_ov_consumption[1] = subport->tc_credits_per_period[1] - subport->tc_credits[1]; + tc_ov_consumption[2] = subport->tc_credits_per_period[2] - subport->tc_credits[2]; + tc_ov_consumption[3] = subport->tc_credits_per_period[3] - subport->tc_credits[3]; + + tc_ov_consumption_max = subport->tc_credits_per_period[3] - + (tc_ov_consumption[0] + tc_ov_consumption[1] + tc_ov_consumption[2]); + + if (tc_ov_consumption[3] > (tc_ov_consumption_max - port->mtu)) { + tc_ov_wm -= tc_ov_wm >> 7; + if (tc_ov_wm < subport->tc_ov_wm_min) + tc_ov_wm = subport->tc_ov_wm_min; + + return tc_ov_wm; + } + + tc_ov_wm += (tc_ov_wm >> 7) + 1; + if (tc_ov_wm > subport->tc_ov_wm_max) + tc_ov_wm = subport->tc_ov_wm_max; + + return tc_ov_wm; +} + +static inline void +grinder_credits_update(struct rte_sched_port *port, uint32_t pos) +{ + struct rte_sched_grinder *grinder = port->grinder + pos; + struct rte_sched_subport *subport = grinder->subport; + struct rte_sched_pipe *pipe = grinder->pipe; + struct rte_sched_pipe_profile *params = grinder->pipe_params; + uint64_t n_periods; + + /* Subport TB */ + n_periods = (port->time - subport->tb_time) / subport->tb_period; + subport->tb_credits += n_periods * subport->tb_credits_per_period; + subport->tb_credits = rte_sched_min_val_2_u32(subport->tb_credits, subport->tb_size); + subport->tb_time += n_periods * subport->tb_period; + + /* Pipe TB */ + n_periods = (port->time - pipe->tb_time) / params->tb_period; + pipe->tb_credits += n_periods * params->tb_credits_per_period; + pipe->tb_credits = rte_sched_min_val_2_u32(pipe->tb_credits, params->tb_size); + pipe->tb_time += n_periods * params->tb_period; + + /* Subport TCs */ + if (unlikely(port->time >= subport->tc_time)) { + subport->tc_ov_wm = grinder_tc_ov_credits_update(port, pos); + + subport->tc_credits[0] = subport->tc_credits_per_period[0]; + subport->tc_credits[1] = subport->tc_credits_per_period[1]; + subport->tc_credits[2] = subport->tc_credits_per_period[2]; + subport->tc_credits[3] = subport->tc_credits_per_period[3]; + + subport->tc_time = port->time + subport->tc_period; + subport->tc_ov_period_id++; + } + + /* Pipe TCs */ + if (unlikely(port->time >= pipe->tc_time)) { + pipe->tc_credits[0] = params->tc_credits_per_period[0]; + pipe->tc_credits[1] = params->tc_credits_per_period[1]; + pipe->tc_credits[2] = params->tc_credits_per_period[2]; + pipe->tc_credits[3] = params->tc_credits_per_period[3]; + pipe->tc_time = port->time + params->tc_period; + } + + /* Pipe TCs - Oversubscription */ + if (unlikely(pipe->tc_ov_period_id != subport->tc_ov_period_id)) { + pipe->tc_ov_credits = subport->tc_ov_wm * params->tc_ov_weight; + + pipe->tc_ov_period_id = subport->tc_ov_period_id; + } +} + +#endif /* RTE_SCHED_TS_CREDITS_UPDATE, RTE_SCHED_SUBPORT_TC_OV */ + + +#ifndef RTE_SCHED_SUBPORT_TC_OV + +static inline int +grinder_credits_check(struct rte_sched_port *port, uint32_t pos) +{ + struct rte_sched_grinder *grinder = port->grinder + pos; + struct rte_sched_subport *subport = grinder->subport; + struct rte_sched_pipe *pipe = grinder->pipe; + struct rte_mbuf *pkt = grinder->pkt; + uint32_t tc_index = grinder->tc_index; + uint32_t pkt_len = pkt->pkt_len + port->frame_overhead; + uint32_t subport_tb_credits = subport->tb_credits; + uint32_t subport_tc_credits = subport->tc_credits[tc_index]; + uint32_t pipe_tb_credits = pipe->tb_credits; + uint32_t pipe_tc_credits = pipe->tc_credits[tc_index]; + int enough_credits; + + /* Check queue credits */ + enough_credits = (pkt_len <= subport_tb_credits) && + (pkt_len <= subport_tc_credits) && + (pkt_len <= pipe_tb_credits) && + (pkt_len <= pipe_tc_credits); + + if (!enough_credits) + return 0; + + /* Update port credits */ + subport->tb_credits -= pkt_len; + subport->tc_credits[tc_index] -= pkt_len; + pipe->tb_credits -= pkt_len; + pipe->tc_credits[tc_index] -= pkt_len; + + return 1; +} + +#else + +static inline int +grinder_credits_check(struct rte_sched_port *port, uint32_t pos) +{ + struct rte_sched_grinder *grinder = port->grinder + pos; + struct rte_sched_subport *subport = grinder->subport; + struct rte_sched_pipe *pipe = grinder->pipe; + struct rte_mbuf *pkt = grinder->pkt; + uint32_t tc_index = grinder->tc_index; + uint32_t pkt_len = pkt->pkt_len + port->frame_overhead; + uint32_t subport_tb_credits = subport->tb_credits; + uint32_t subport_tc_credits = subport->tc_credits[tc_index]; + uint32_t pipe_tb_credits = pipe->tb_credits; + uint32_t pipe_tc_credits = pipe->tc_credits[tc_index]; + uint32_t pipe_tc_ov_mask1[] = {UINT32_MAX, UINT32_MAX, UINT32_MAX, pipe->tc_ov_credits}; + uint32_t pipe_tc_ov_mask2[] = {0, 0, 0, UINT32_MAX}; + uint32_t pipe_tc_ov_credits = pipe_tc_ov_mask1[tc_index]; + int enough_credits; + + /* Check pipe and subport credits */ + enough_credits = (pkt_len <= subport_tb_credits) && + (pkt_len <= subport_tc_credits) && + (pkt_len <= pipe_tb_credits) && + (pkt_len <= pipe_tc_credits) && + (pkt_len <= pipe_tc_ov_credits); + + if (!enough_credits) + return 0; + + /* Update pipe and subport credits */ + subport->tb_credits -= pkt_len; + subport->tc_credits[tc_index] -= pkt_len; + pipe->tb_credits -= pkt_len; + pipe->tc_credits[tc_index] -= pkt_len; + pipe->tc_ov_credits -= pipe_tc_ov_mask2[tc_index] & pkt_len; + + return 1; +} + +#endif /* RTE_SCHED_SUBPORT_TC_OV */ + + +static inline int +grinder_schedule(struct rte_sched_port *port, uint32_t pos) +{ + struct rte_sched_grinder *grinder = port->grinder + pos; + struct rte_sched_queue *queue = grinder->queue[grinder->qpos]; + struct rte_mbuf *pkt = grinder->pkt; + uint32_t pkt_len = pkt->pkt_len + port->frame_overhead; + + if (!grinder_credits_check(port, pos)) + return 0; + + /* Advance port time */ + port->time += pkt_len; + + /* Send packet */ + port->pkts_out[port->n_pkts_out++] = pkt; + queue->qr++; + grinder->wrr_tokens[grinder->qpos] += pkt_len * grinder->wrr_cost[grinder->qpos]; + if (queue->qr == queue->qw) { + uint32_t qindex = grinder->qindex[grinder->qpos]; + + rte_bitmap_clear(port->bmp, qindex); + grinder->qmask &= ~(1 << grinder->qpos); + grinder->wrr_mask[grinder->qpos] = 0; + rte_sched_port_set_queue_empty_timestamp(port, qindex); + } + + /* Reset pipe loop detection */ + port->pipe_loop = RTE_SCHED_PIPE_INVALID; + grinder->productive = 1; + + return 1; +} + +#ifdef SCHED_VECTOR_SSE4 + +static inline int +grinder_pipe_exists(struct rte_sched_port *port, uint32_t base_pipe) +{ + __m128i index = _mm_set1_epi32(base_pipe); + __m128i pipes = _mm_load_si128((__m128i *)port->grinder_base_bmp_pos); + __m128i res = _mm_cmpeq_epi32(pipes, index); + + pipes = _mm_load_si128((__m128i *)(port->grinder_base_bmp_pos + 4)); + pipes = _mm_cmpeq_epi32(pipes, index); + res = _mm_or_si128(res, pipes); + + if (_mm_testz_si128(res, res)) + return 0; + + return 1; +} + +#elif defined(SCHED_VECTOR_NEON) + +static inline int +grinder_pipe_exists(struct rte_sched_port *port, uint32_t base_pipe) +{ + uint32x4_t index, pipes; + uint32_t *pos = (uint32_t *)port->grinder_base_bmp_pos; + + index = vmovq_n_u32(base_pipe); + pipes = vld1q_u32(pos); + if (!vminvq_u32(veorq_u32(pipes, index))) + return 1; + + pipes = vld1q_u32(pos + 4); + if (!vminvq_u32(veorq_u32(pipes, index))) + return 1; + + return 0; +} + +#else + +static inline int +grinder_pipe_exists(struct rte_sched_port *port, uint32_t base_pipe) +{ + uint32_t i; + + for (i = 0; i < RTE_SCHED_PORT_N_GRINDERS; i++) { + if (port->grinder_base_bmp_pos[i] == base_pipe) + return 1; + } + + return 0; +} + +#endif /* RTE_SCHED_OPTIMIZATIONS */ + +static inline void +grinder_pcache_populate(struct rte_sched_port *port, uint32_t pos, uint32_t bmp_pos, uint64_t bmp_slab) +{ + struct rte_sched_grinder *grinder = port->grinder + pos; + uint16_t w[4]; + + grinder->pcache_w = 0; + grinder->pcache_r = 0; + + w[0] = (uint16_t) bmp_slab; + w[1] = (uint16_t) (bmp_slab >> 16); + w[2] = (uint16_t) (bmp_slab >> 32); + w[3] = (uint16_t) (bmp_slab >> 48); + + grinder->pcache_qmask[grinder->pcache_w] = w[0]; + grinder->pcache_qindex[grinder->pcache_w] = bmp_pos; + grinder->pcache_w += (w[0] != 0); + + grinder->pcache_qmask[grinder->pcache_w] = w[1]; + grinder->pcache_qindex[grinder->pcache_w] = bmp_pos + 16; + grinder->pcache_w += (w[1] != 0); + + grinder->pcache_qmask[grinder->pcache_w] = w[2]; + grinder->pcache_qindex[grinder->pcache_w] = bmp_pos + 32; + grinder->pcache_w += (w[2] != 0); + + grinder->pcache_qmask[grinder->pcache_w] = w[3]; + grinder->pcache_qindex[grinder->pcache_w] = bmp_pos + 48; + grinder->pcache_w += (w[3] != 0); +} + +static inline void +grinder_tccache_populate(struct rte_sched_port *port, uint32_t pos, uint32_t qindex, uint16_t qmask) +{ + struct rte_sched_grinder *grinder = port->grinder + pos; + uint8_t b[4]; + + grinder->tccache_w = 0; + grinder->tccache_r = 0; + + b[0] = (uint8_t) (qmask & 0xF); + b[1] = (uint8_t) ((qmask >> 4) & 0xF); + b[2] = (uint8_t) ((qmask >> 8) & 0xF); + b[3] = (uint8_t) ((qmask >> 12) & 0xF); + + grinder->tccache_qmask[grinder->tccache_w] = b[0]; + grinder->tccache_qindex[grinder->tccache_w] = qindex; + grinder->tccache_w += (b[0] != 0); + + grinder->tccache_qmask[grinder->tccache_w] = b[1]; + grinder->tccache_qindex[grinder->tccache_w] = qindex + 4; + grinder->tccache_w += (b[1] != 0); + + grinder->tccache_qmask[grinder->tccache_w] = b[2]; + grinder->tccache_qindex[grinder->tccache_w] = qindex + 8; + grinder->tccache_w += (b[2] != 0); + + grinder->tccache_qmask[grinder->tccache_w] = b[3]; + grinder->tccache_qindex[grinder->tccache_w] = qindex + 12; + grinder->tccache_w += (b[3] != 0); +} + +static inline int +grinder_next_tc(struct rte_sched_port *port, uint32_t pos) +{ + struct rte_sched_grinder *grinder = port->grinder + pos; + struct rte_mbuf **qbase; + uint32_t qindex; + uint16_t qsize; + + if (grinder->tccache_r == grinder->tccache_w) + return 0; + + qindex = grinder->tccache_qindex[grinder->tccache_r]; + qbase = rte_sched_port_qbase(port, qindex); + qsize = rte_sched_port_qsize(port, qindex); + + grinder->tc_index = (qindex >> 2) & 0x3; + grinder->qmask = grinder->tccache_qmask[grinder->tccache_r]; + grinder->qsize = qsize; + + grinder->qindex[0] = qindex; + grinder->qindex[1] = qindex + 1; + grinder->qindex[2] = qindex + 2; + grinder->qindex[3] = qindex + 3; + + grinder->queue[0] = port->queue + qindex; + grinder->queue[1] = port->queue + qindex + 1; + grinder->queue[2] = port->queue + qindex + 2; + grinder->queue[3] = port->queue + qindex + 3; + + grinder->qbase[0] = qbase; + grinder->qbase[1] = qbase + qsize; + grinder->qbase[2] = qbase + 2 * qsize; + grinder->qbase[3] = qbase + 3 * qsize; + + grinder->tccache_r++; + return 1; +} + +static inline int +grinder_next_pipe(struct rte_sched_port *port, uint32_t pos) +{ + struct rte_sched_grinder *grinder = port->grinder + pos; + uint32_t pipe_qindex; + uint16_t pipe_qmask; + + if (grinder->pcache_r < grinder->pcache_w) { + pipe_qmask = grinder->pcache_qmask[grinder->pcache_r]; + pipe_qindex = grinder->pcache_qindex[grinder->pcache_r]; + grinder->pcache_r++; + } else { + uint64_t bmp_slab = 0; + uint32_t bmp_pos = 0; + + /* Get another non-empty pipe group */ + if (unlikely(rte_bitmap_scan(port->bmp, &bmp_pos, &bmp_slab) <= 0)) + return 0; + +#ifdef RTE_SCHED_DEBUG + debug_check_queue_slab(port, bmp_pos, bmp_slab); +#endif + + /* Return if pipe group already in one of the other grinders */ + port->grinder_base_bmp_pos[pos] = RTE_SCHED_BMP_POS_INVALID; + if (unlikely(grinder_pipe_exists(port, bmp_pos))) + return 0; + + port->grinder_base_bmp_pos[pos] = bmp_pos; + + /* Install new pipe group into grinder's pipe cache */ + grinder_pcache_populate(port, pos, bmp_pos, bmp_slab); + + pipe_qmask = grinder->pcache_qmask[0]; + pipe_qindex = grinder->pcache_qindex[0]; + grinder->pcache_r = 1; + } + + /* Install new pipe in the grinder */ + grinder->pindex = pipe_qindex >> 4; + grinder->subport = port->subport + (grinder->pindex / port->n_pipes_per_subport); + grinder->pipe = port->pipe + grinder->pindex; + grinder->pipe_params = NULL; /* to be set after the pipe structure is prefetched */ + grinder->productive = 0; + + grinder_tccache_populate(port, pos, pipe_qindex, pipe_qmask); + grinder_next_tc(port, pos); + + /* Check for pipe exhaustion */ + if (grinder->pindex == port->pipe_loop) { + port->pipe_exhaustion = 1; + port->pipe_loop = RTE_SCHED_PIPE_INVALID; + } + + return 1; +} + + +static inline void +grinder_wrr_load(struct rte_sched_port *port, uint32_t pos) +{ + struct rte_sched_grinder *grinder = port->grinder + pos; + struct rte_sched_pipe *pipe = grinder->pipe; + struct rte_sched_pipe_profile *pipe_params = grinder->pipe_params; + uint32_t tc_index = grinder->tc_index; + uint32_t qmask = grinder->qmask; + uint32_t qindex; + + qindex = tc_index * 4; + + grinder->wrr_tokens[0] = ((uint16_t) pipe->wrr_tokens[qindex]) << RTE_SCHED_WRR_SHIFT; + grinder->wrr_tokens[1] = ((uint16_t) pipe->wrr_tokens[qindex + 1]) << RTE_SCHED_WRR_SHIFT; + grinder->wrr_tokens[2] = ((uint16_t) pipe->wrr_tokens[qindex + 2]) << RTE_SCHED_WRR_SHIFT; + grinder->wrr_tokens[3] = ((uint16_t) pipe->wrr_tokens[qindex + 3]) << RTE_SCHED_WRR_SHIFT; + + grinder->wrr_mask[0] = (qmask & 0x1) * 0xFFFF; + grinder->wrr_mask[1] = ((qmask >> 1) & 0x1) * 0xFFFF; + grinder->wrr_mask[2] = ((qmask >> 2) & 0x1) * 0xFFFF; + grinder->wrr_mask[3] = ((qmask >> 3) & 0x1) * 0xFFFF; + + grinder->wrr_cost[0] = pipe_params->wrr_cost[qindex]; + grinder->wrr_cost[1] = pipe_params->wrr_cost[qindex + 1]; + grinder->wrr_cost[2] = pipe_params->wrr_cost[qindex + 2]; + grinder->wrr_cost[3] = pipe_params->wrr_cost[qindex + 3]; +} + +static inline void +grinder_wrr_store(struct rte_sched_port *port, uint32_t pos) +{ + struct rte_sched_grinder *grinder = port->grinder + pos; + struct rte_sched_pipe *pipe = grinder->pipe; + uint32_t tc_index = grinder->tc_index; + uint32_t qindex; + + qindex = tc_index * 4; + + pipe->wrr_tokens[qindex] = (grinder->wrr_tokens[0] & grinder->wrr_mask[0]) + >> RTE_SCHED_WRR_SHIFT; + pipe->wrr_tokens[qindex + 1] = (grinder->wrr_tokens[1] & grinder->wrr_mask[1]) + >> RTE_SCHED_WRR_SHIFT; + pipe->wrr_tokens[qindex + 2] = (grinder->wrr_tokens[2] & grinder->wrr_mask[2]) + >> RTE_SCHED_WRR_SHIFT; + pipe->wrr_tokens[qindex + 3] = (grinder->wrr_tokens[3] & grinder->wrr_mask[3]) + >> RTE_SCHED_WRR_SHIFT; +} + +static inline void +grinder_wrr(struct rte_sched_port *port, uint32_t pos) +{ + struct rte_sched_grinder *grinder = port->grinder + pos; + uint16_t wrr_tokens_min; + + grinder->wrr_tokens[0] |= ~grinder->wrr_mask[0]; + grinder->wrr_tokens[1] |= ~grinder->wrr_mask[1]; + grinder->wrr_tokens[2] |= ~grinder->wrr_mask[2]; + grinder->wrr_tokens[3] |= ~grinder->wrr_mask[3]; + + grinder->qpos = rte_min_pos_4_u16(grinder->wrr_tokens); + wrr_tokens_min = grinder->wrr_tokens[grinder->qpos]; + + grinder->wrr_tokens[0] -= wrr_tokens_min; + grinder->wrr_tokens[1] -= wrr_tokens_min; + grinder->wrr_tokens[2] -= wrr_tokens_min; + grinder->wrr_tokens[3] -= wrr_tokens_min; +} + + +#define grinder_evict(port, pos) + +static inline void +grinder_prefetch_pipe(struct rte_sched_port *port, uint32_t pos) +{ + struct rte_sched_grinder *grinder = port->grinder + pos; + + rte_prefetch0(grinder->pipe); + rte_prefetch0(grinder->queue[0]); +} + +static inline void +grinder_prefetch_tc_queue_arrays(struct rte_sched_port *port, uint32_t pos) +{ + struct rte_sched_grinder *grinder = port->grinder + pos; + uint16_t qsize, qr[4]; + + qsize = grinder->qsize; + qr[0] = grinder->queue[0]->qr & (qsize - 1); + qr[1] = grinder->queue[1]->qr & (qsize - 1); + qr[2] = grinder->queue[2]->qr & (qsize - 1); + qr[3] = grinder->queue[3]->qr & (qsize - 1); + + rte_prefetch0(grinder->qbase[0] + qr[0]); + rte_prefetch0(grinder->qbase[1] + qr[1]); + + grinder_wrr_load(port, pos); + grinder_wrr(port, pos); + + rte_prefetch0(grinder->qbase[2] + qr[2]); + rte_prefetch0(grinder->qbase[3] + qr[3]); +} + +static inline void +grinder_prefetch_mbuf(struct rte_sched_port *port, uint32_t pos) +{ + struct rte_sched_grinder *grinder = port->grinder + pos; + uint32_t qpos = grinder->qpos; + struct rte_mbuf **qbase = grinder->qbase[qpos]; + uint16_t qsize = grinder->qsize; + uint16_t qr = grinder->queue[qpos]->qr & (qsize - 1); + + grinder->pkt = qbase[qr]; + rte_prefetch0(grinder->pkt); + + if (unlikely((qr & 0x7) == 7)) { + uint16_t qr_next = (grinder->queue[qpos]->qr + 1) & (qsize - 1); + + rte_prefetch0(qbase + qr_next); + } +} + +static inline uint32_t +grinder_handle(struct rte_sched_port *port, uint32_t pos) +{ + struct rte_sched_grinder *grinder = port->grinder + pos; + + switch (grinder->state) { + case e_GRINDER_PREFETCH_PIPE: + { + if (grinder_next_pipe(port, pos)) { + grinder_prefetch_pipe(port, pos); + port->busy_grinders++; + + grinder->state = e_GRINDER_PREFETCH_TC_QUEUE_ARRAYS; + return 0; + } + + return 0; + } + + case e_GRINDER_PREFETCH_TC_QUEUE_ARRAYS: + { + struct rte_sched_pipe *pipe = grinder->pipe; + + grinder->pipe_params = port->pipe_profiles + pipe->profile; + grinder_prefetch_tc_queue_arrays(port, pos); + grinder_credits_update(port, pos); + + grinder->state = e_GRINDER_PREFETCH_MBUF; + return 0; + } + + case e_GRINDER_PREFETCH_MBUF: + { + grinder_prefetch_mbuf(port, pos); + + grinder->state = e_GRINDER_READ_MBUF; + return 0; + } + + case e_GRINDER_READ_MBUF: + { + uint32_t result = 0; + + result = grinder_schedule(port, pos); + + /* Look for next packet within the same TC */ + if (result && grinder->qmask) { + grinder_wrr(port, pos); + grinder_prefetch_mbuf(port, pos); + + return 1; + } + grinder_wrr_store(port, pos); + + /* Look for another active TC within same pipe */ + if (grinder_next_tc(port, pos)) { + grinder_prefetch_tc_queue_arrays(port, pos); + + grinder->state = e_GRINDER_PREFETCH_MBUF; + return result; + } + + if (grinder->productive == 0 && + port->pipe_loop == RTE_SCHED_PIPE_INVALID) + port->pipe_loop = grinder->pindex; + + grinder_evict(port, pos); + + /* Look for another active pipe */ + if (grinder_next_pipe(port, pos)) { + grinder_prefetch_pipe(port, pos); + + grinder->state = e_GRINDER_PREFETCH_TC_QUEUE_ARRAYS; + return result; + } + + /* No active pipe found */ + port->busy_grinders--; + + grinder->state = e_GRINDER_PREFETCH_PIPE; + return result; + } + + default: + rte_panic("Algorithmic error (invalid state)\n"); + return 0; + } +} + +static inline void +rte_sched_port_time_resync(struct rte_sched_port *port) +{ + uint64_t cycles = rte_get_tsc_cycles(); + uint64_t cycles_diff = cycles - port->time_cpu_cycles; + uint64_t bytes_diff; + + /* Compute elapsed time in bytes */ + bytes_diff = rte_reciprocal_divide(cycles_diff << RTE_SCHED_TIME_SHIFT, + port->inv_cycles_per_byte); + + /* Advance port time */ + port->time_cpu_cycles = cycles; + port->time_cpu_bytes += bytes_diff; + if (port->time < port->time_cpu_bytes) + port->time = port->time_cpu_bytes; + + /* Reset pipe loop detection */ + port->pipe_loop = RTE_SCHED_PIPE_INVALID; +} + +static inline int +rte_sched_port_exceptions(struct rte_sched_port *port, int second_pass) +{ + int exceptions; + + /* Check if any exception flag is set */ + exceptions = (second_pass && port->busy_grinders == 0) || + (port->pipe_exhaustion == 1); + + /* Clear exception flags */ + port->pipe_exhaustion = 0; + + return exceptions; +} + +int +rte_sched_port_dequeue(struct rte_sched_port *port, struct rte_mbuf **pkts, uint32_t n_pkts) +{ + uint32_t i, count; + + port->pkts_out = pkts; + port->n_pkts_out = 0; + + rte_sched_port_time_resync(port); + + /* Take each queue in the grinder one step further */ + for (i = 0, count = 0; ; i++) { + count += grinder_handle(port, i & (RTE_SCHED_PORT_N_GRINDERS - 1)); + if ((count == n_pkts) || + rte_sched_port_exceptions(port, i >= RTE_SCHED_PORT_N_GRINDERS)) { + break; + } + } + + return count; +} diff --git a/src/spdk/dpdk/lib/librte_sched/rte_sched.h b/src/spdk/dpdk/lib/librte_sched/rte_sched.h new file mode 100644 index 00000000..84fa896d --- /dev/null +++ b/src/spdk/dpdk/lib/librte_sched/rte_sched.h @@ -0,0 +1,447 @@ +/* SPDX-License-Identifier: BSD-3-Clause + * Copyright(c) 2010-2014 Intel Corporation + */ + +#ifndef __INCLUDE_RTE_SCHED_H__ +#define __INCLUDE_RTE_SCHED_H__ + +#ifdef __cplusplus +extern "C" { +#endif + +/** + * @file + * RTE Hierarchical Scheduler + * + * The hierarchical scheduler prioritizes the transmission of packets + * from different users and traffic classes according to the Service + * Level Agreements (SLAs) defined for the current network node. + * + * The scheduler supports thousands of packet queues grouped under a + * 5-level hierarchy: + * 1. Port: + * - Typical usage: output Ethernet port; + * - Multiple ports are scheduled in round robin order with + * equal priority; + * 2. Subport: + * - Typical usage: group of users; + * - Traffic shaping using the token bucket algorithm + * (one bucket per subport); + * - Upper limit enforced per traffic class at subport level; + * - Lower priority traffic classes able to reuse subport + * bandwidth currently unused by higher priority traffic + * classes of the same subport; + * - When any subport traffic class is oversubscribed + * (configuration time event), the usage of subport member + * pipes with high demand for thattraffic class pipes is + * truncated to a dynamically adjusted value with no + * impact to low demand pipes; + * 3. Pipe: + * - Typical usage: individual user/subscriber; + * - Traffic shaping using the token bucket algorithm + * (one bucket per pipe); + * 4. Traffic class: + * - Traffic classes of the same pipe handled in strict + * priority order; + * - Upper limit enforced per traffic class at the pipe level; + * - Lower priority traffic classes able to reuse pipe + * bandwidth currently unused by higher priority traffic + * classes of the same pipe; + * 5. Queue: + * - Typical usage: queue hosting packets from one or + * multiple connections of same traffic class belonging to + * the same user; + * - Weighted Round Robin (WRR) is used to service the + * queues within same pipe traffic class. + * + */ + +#include <sys/types.h> +#include <rte_compat.h> +#include <rte_mbuf.h> +#include <rte_meter.h> + +/** Random Early Detection (RED) */ +#ifdef RTE_SCHED_RED +#include "rte_red.h" +#endif + +/** Number of traffic classes per pipe (as well as subport). + * Cannot be changed. + */ +#define RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE 4 + +/** Number of queues per pipe traffic class. Cannot be changed. */ +#define RTE_SCHED_QUEUES_PER_TRAFFIC_CLASS 4 + +/** Number of queues per pipe. */ +#define RTE_SCHED_QUEUES_PER_PIPE \ + (RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE * \ + RTE_SCHED_QUEUES_PER_TRAFFIC_CLASS) + +/** Maximum number of pipe profiles that can be defined per port. + * Compile-time configurable. + */ +#ifndef RTE_SCHED_PIPE_PROFILES_PER_PORT +#define RTE_SCHED_PIPE_PROFILES_PER_PORT 256 +#endif + +/* + * Ethernet framing overhead. Overhead fields per Ethernet frame: + * 1. Preamble: 7 bytes; + * 2. Start of Frame Delimiter (SFD): 1 byte; + * 3. Frame Check Sequence (FCS): 4 bytes; + * 4. Inter Frame Gap (IFG): 12 bytes. + * + * The FCS is considered overhead only if not included in the packet + * length (field pkt_len of struct rte_mbuf). + */ +#ifndef RTE_SCHED_FRAME_OVERHEAD_DEFAULT +#define RTE_SCHED_FRAME_OVERHEAD_DEFAULT 24 +#endif + +/* + * Subport configuration parameters. The period and credits_per_period + * parameters are measured in bytes, with one byte meaning the time + * duration associated with the transmission of one byte on the + * physical medium of the output port, with pipe or pipe traffic class + * rate (measured as percentage of output port rate) determined as + * credits_per_period divided by period. One credit represents one + * byte. + */ +struct rte_sched_subport_params { + /* Subport token bucket */ + uint32_t tb_rate; /**< Rate (measured in bytes per second) */ + uint32_t tb_size; /**< Size (measured in credits) */ + + /* Subport traffic classes */ + uint32_t tc_rate[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE]; + /**< Traffic class rates (measured in bytes per second) */ + uint32_t tc_period; + /**< Enforcement period for rates (measured in milliseconds) */ +}; + +/** Subport statistics */ +struct rte_sched_subport_stats { + /* Packets */ + uint32_t n_pkts_tc[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE]; + /**< Number of packets successfully written */ + uint32_t n_pkts_tc_dropped[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE]; + /**< Number of packets dropped */ + + /* Bytes */ + uint32_t n_bytes_tc[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE]; + /**< Number of bytes successfully written for each traffic class */ + uint32_t n_bytes_tc_dropped[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE]; + /**< Number of bytes dropped for each traffic class */ + +#ifdef RTE_SCHED_RED + uint32_t n_pkts_red_dropped[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE]; + /**< Number of packets dropped by red */ +#endif +}; + +/* + * Pipe configuration parameters. The period and credits_per_period + * parameters are measured in bytes, with one byte meaning the time + * duration associated with the transmission of one byte on the + * physical medium of the output port, with pipe or pipe traffic class + * rate (measured as percentage of output port rate) determined as + * credits_per_period divided by period. One credit represents one + * byte. + */ +struct rte_sched_pipe_params { + /* Pipe token bucket */ + uint32_t tb_rate; /**< Rate (measured in bytes per second) */ + uint32_t tb_size; /**< Size (measured in credits) */ + + /* Pipe traffic classes */ + uint32_t tc_rate[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE]; + /**< Traffic class rates (measured in bytes per second) */ + uint32_t tc_period; + /**< Enforcement period (measured in milliseconds) */ +#ifdef RTE_SCHED_SUBPORT_TC_OV + uint8_t tc_ov_weight; /**< Weight Traffic class 3 oversubscription */ +#endif + + /* Pipe queues */ + uint8_t wrr_weights[RTE_SCHED_QUEUES_PER_PIPE]; /**< WRR weights */ +}; + +/** Queue statistics */ +struct rte_sched_queue_stats { + /* Packets */ + uint32_t n_pkts; /**< Packets successfully written */ + uint32_t n_pkts_dropped; /**< Packets dropped */ +#ifdef RTE_SCHED_RED + uint32_t n_pkts_red_dropped; /**< Packets dropped by RED */ +#endif + + /* Bytes */ + uint32_t n_bytes; /**< Bytes successfully written */ + uint32_t n_bytes_dropped; /**< Bytes dropped */ +}; + +/** Port configuration parameters. */ +struct rte_sched_port_params { + const char *name; /**< String to be associated */ + int socket; /**< CPU socket ID */ + uint32_t rate; /**< Output port rate + * (measured in bytes per second) */ + uint32_t mtu; /**< Maximum Ethernet frame size + * (measured in bytes). + * Should not include the framing overhead. */ + uint32_t frame_overhead; /**< Framing overhead per packet + * (measured in bytes) */ + uint32_t n_subports_per_port; /**< Number of subports */ + uint32_t n_pipes_per_subport; /**< Number of pipes per subport */ + uint16_t qsize[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE]; + /**< Packet queue size for each traffic class. + * All queues within the same pipe traffic class have the same + * size. Queues from different pipes serving the same traffic + * class have the same size. */ + struct rte_sched_pipe_params *pipe_profiles; + /**< Pipe profile table. + * Every pipe is configured using one of the profiles from this table. */ + uint32_t n_pipe_profiles; /**< Profiles in the pipe profile table */ +#ifdef RTE_SCHED_RED + struct rte_red_params red_params[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE][e_RTE_METER_COLORS]; /**< RED parameters */ +#endif +}; + +/* + * Configuration + * + ***/ + +/** + * Hierarchical scheduler port configuration + * + * @param params + * Port scheduler configuration parameter structure + * @return + * Handle to port scheduler instance upon success or NULL otherwise. + */ +struct rte_sched_port * +rte_sched_port_config(struct rte_sched_port_params *params); + +/** + * Hierarchical scheduler port free + * + * @param port + * Handle to port scheduler instance + */ +void +rte_sched_port_free(struct rte_sched_port *port); + +/** + * @warning + * @b EXPERIMENTAL: this API may change without prior notice. + * + * Hierarchical scheduler pipe profile add + * + * @param port + * Handle to port scheduler instance + * @param params + * Pipe profile parameters + * @param pipe_profile_id + * Set to valid profile id when profile is added successfully. + * @return + * 0 upon success, error code otherwise + */ +int __rte_experimental +rte_sched_port_pipe_profile_add(struct rte_sched_port *port, + struct rte_sched_pipe_params *params, + uint32_t *pipe_profile_id); + +/** + * Hierarchical scheduler subport configuration + * + * @param port + * Handle to port scheduler instance + * @param subport_id + * Subport ID + * @param params + * Subport configuration parameters + * @return + * 0 upon success, error code otherwise + */ +int +rte_sched_subport_config(struct rte_sched_port *port, + uint32_t subport_id, + struct rte_sched_subport_params *params); + +/** + * Hierarchical scheduler pipe configuration + * + * @param port + * Handle to port scheduler instance + * @param subport_id + * Subport ID + * @param pipe_id + * Pipe ID within subport + * @param pipe_profile + * ID of port-level pre-configured pipe profile + * @return + * 0 upon success, error code otherwise + */ +int +rte_sched_pipe_config(struct rte_sched_port *port, + uint32_t subport_id, + uint32_t pipe_id, + int32_t pipe_profile); + +/** + * Hierarchical scheduler memory footprint size per port + * + * @param params + * Port scheduler configuration parameter structure + * @return + * Memory footprint size in bytes upon success, 0 otherwise + */ +uint32_t +rte_sched_port_get_memory_footprint(struct rte_sched_port_params *params); + +/* + * Statistics + * + ***/ + +/** + * Hierarchical scheduler subport statistics read + * + * @param port + * Handle to port scheduler instance + * @param subport_id + * Subport ID + * @param stats + * Pointer to pre-allocated subport statistics structure where the statistics + * counters should be stored + * @param tc_ov + * Pointer to pre-allocated 4-entry array where the oversubscription status for + * each of the 4 subport traffic classes should be stored. + * @return + * 0 upon success, error code otherwise + */ +int +rte_sched_subport_read_stats(struct rte_sched_port *port, + uint32_t subport_id, + struct rte_sched_subport_stats *stats, + uint32_t *tc_ov); + +/** + * Hierarchical scheduler queue statistics read + * + * @param port + * Handle to port scheduler instance + * @param queue_id + * Queue ID within port scheduler + * @param stats + * Pointer to pre-allocated subport statistics structure where the statistics + * counters should be stored + * @param qlen + * Pointer to pre-allocated variable where the current queue length + * should be stored. + * @return + * 0 upon success, error code otherwise + */ +int +rte_sched_queue_read_stats(struct rte_sched_port *port, + uint32_t queue_id, + struct rte_sched_queue_stats *stats, + uint16_t *qlen); + +/** + * Scheduler hierarchy path write to packet descriptor. Typically + * called by the packet classification stage. + * + * @param pkt + * Packet descriptor handle + * @param subport + * Subport ID + * @param pipe + * Pipe ID within subport + * @param traffic_class + * Traffic class ID within pipe (0 .. 3) + * @param queue + * Queue ID within pipe traffic class (0 .. 3) + * @param color + * Packet color set + */ +void +rte_sched_port_pkt_write(struct rte_mbuf *pkt, + uint32_t subport, uint32_t pipe, uint32_t traffic_class, + uint32_t queue, enum rte_meter_color color); + +/** + * Scheduler hierarchy path read from packet descriptor (struct + * rte_mbuf). Typically called as part of the hierarchical scheduler + * enqueue operation. The subport, pipe, traffic class and queue + * parameters need to be pre-allocated by the caller. + * + * @param pkt + * Packet descriptor handle + * @param subport + * Subport ID + * @param pipe + * Pipe ID within subport + * @param traffic_class + * Traffic class ID within pipe (0 .. 3) + * @param queue + * Queue ID within pipe traffic class (0 .. 3) + * + */ +void +rte_sched_port_pkt_read_tree_path(const struct rte_mbuf *pkt, + uint32_t *subport, uint32_t *pipe, + uint32_t *traffic_class, uint32_t *queue); + +enum rte_meter_color +rte_sched_port_pkt_read_color(const struct rte_mbuf *pkt); + +/** + * Hierarchical scheduler port enqueue. Writes up to n_pkts to port + * scheduler and returns the number of packets actually written. For + * each packet, the port scheduler queue to write the packet to is + * identified by reading the hierarchy path from the packet + * descriptor; if the queue is full or congested and the packet is not + * written to the queue, then the packet is automatically dropped + * without any action required from the caller. + * + * @param port + * Handle to port scheduler instance + * @param pkts + * Array storing the packet descriptor handles + * @param n_pkts + * Number of packets to enqueue from the pkts array into the port scheduler + * @return + * Number of packets successfully enqueued + */ +int +rte_sched_port_enqueue(struct rte_sched_port *port, struct rte_mbuf **pkts, uint32_t n_pkts); + +/** + * Hierarchical scheduler port dequeue. Reads up to n_pkts from the + * port scheduler and stores them in the pkts array and returns the + * number of packets actually read. The pkts array needs to be + * pre-allocated by the caller with at least n_pkts entries. + * + * @param port + * Handle to port scheduler instance + * @param pkts + * Pre-allocated packet descriptor array where the packets dequeued + * from the port + * scheduler should be stored + * @param n_pkts + * Number of packets to dequeue from the port scheduler + * @return + * Number of packets successfully dequeued and placed in the pkts array + */ +int +rte_sched_port_dequeue(struct rte_sched_port *port, struct rte_mbuf **pkts, uint32_t n_pkts); + +#ifdef __cplusplus +} +#endif + +#endif /* __INCLUDE_RTE_SCHED_H__ */ diff --git a/src/spdk/dpdk/lib/librte_sched/rte_sched_common.h b/src/spdk/dpdk/lib/librte_sched/rte_sched_common.h new file mode 100644 index 00000000..8c191a9b --- /dev/null +++ b/src/spdk/dpdk/lib/librte_sched/rte_sched_common.h @@ -0,0 +1,101 @@ +/* SPDX-License-Identifier: BSD-3-Clause + * Copyright(c) 2010-2014 Intel Corporation + */ + +#ifndef __INCLUDE_RTE_SCHED_COMMON_H__ +#define __INCLUDE_RTE_SCHED_COMMON_H__ + +#ifdef __cplusplus +extern "C" { +#endif + +#include <stdint.h> +#include <sys/types.h> + +#define __rte_aligned_16 __attribute__((__aligned__(16))) + +static inline uint32_t +rte_sched_min_val_2_u32(uint32_t x, uint32_t y) +{ + return (x < y)? x : y; +} + +#if 0 +static inline uint32_t +rte_min_pos_4_u16(uint16_t *x) +{ + uint32_t pos0, pos1; + + pos0 = (x[0] <= x[1])? 0 : 1; + pos1 = (x[2] <= x[3])? 2 : 3; + + return (x[pos0] <= x[pos1])? pos0 : pos1; +} + +#else + +/* simplified version to remove branches with CMOV instruction */ +static inline uint32_t +rte_min_pos_4_u16(uint16_t *x) +{ + uint32_t pos0 = 0; + uint32_t pos1 = 2; + + if (x[1] <= x[0]) pos0 = 1; + if (x[3] <= x[2]) pos1 = 3; + if (x[pos1] <= x[pos0]) pos0 = pos1; + + return pos0; +} + +#endif + +/* + * Compute the Greatest Common Divisor (GCD) of two numbers. + * This implementation uses Euclid's algorithm: + * gcd(a, 0) = a + * gcd(a, b) = gcd(b, a mod b) + * + */ +static inline uint32_t +rte_get_gcd(uint32_t a, uint32_t b) +{ + uint32_t c; + + if (a == 0) + return b; + if (b == 0) + return a; + + if (a < b) { + c = a; + a = b; + b = c; + } + + while (b != 0) { + c = a % b; + a = b; + b = c; + } + + return a; +} + +/* + * Compute the Lowest Common Denominator (LCD) of two numbers. + * This implementation computes GCD first: + * LCD(a, b) = (a * b) / GCD(a, b) + * + */ +static inline uint32_t +rte_get_lcd(uint32_t a, uint32_t b) +{ + return (a * b) / rte_get_gcd(a, b); +} + +#ifdef __cplusplus +} +#endif + +#endif /* __INCLUDE_RTE_SCHED_COMMON_H__ */ diff --git a/src/spdk/dpdk/lib/librte_sched/rte_sched_version.map b/src/spdk/dpdk/lib/librte_sched/rte_sched_version.map new file mode 100644 index 00000000..72958879 --- /dev/null +++ b/src/spdk/dpdk/lib/librte_sched/rte_sched_version.map @@ -0,0 +1,37 @@ +DPDK_2.0 { + global: + + rte_approx; + rte_red_config_init; + rte_red_log2_1_minus_Wq; + rte_red_pow2_frac_inv; + rte_red_rand_seed; + rte_red_rand_val; + rte_red_rt_data_init; + rte_sched_pipe_config; + rte_sched_port_config; + rte_sched_port_dequeue; + rte_sched_port_enqueue; + rte_sched_port_free; + rte_sched_port_get_memory_footprint; + rte_sched_queue_read_stats; + rte_sched_subport_config; + rte_sched_subport_read_stats; + + local: *; +}; + +DPDK_2.1 { + global: + + rte_sched_port_pkt_write; + rte_sched_port_pkt_read_tree_path; + rte_sched_port_pkt_read_color; + +} DPDK_2.0; + +EXPERIMENTAL { + global: + + rte_sched_port_pipe_profile_add; +}; |