include/linux/dim.h

   1 /* SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB */
   2 /* Copyright (c) 2019 Mellanox Technologies. */
   3
   4 #ifndef DIM_H
   5 #define DIM_H
   6
   7 #include <linux/module.h>
   8
   9 /**
  10  * Number of events between DIM iterations.
  11  * Causes a moderation of the algorithm run.
  12  */
  13 #define DIM_NEVENTS 64
  14
  15 /**
  16  * Is a difference between values justifies taking an action.
  17  * We consider 10% difference as significant.
  18  */
  19 #define IS_SIGNIFICANT_DIFF(val, ref) \
  20         (((100UL * abs((val) - (ref))) / (ref)) > 10)
  21
  22 /**
  23  * Calculate the gap between two values.
  24  * Take wrap-around and variable size into consideration.
  25  */
  26 #define BIT_GAP(bits, end, start) ((((end) - (start)) + BIT_ULL(bits)) \
  27                 & (BIT_ULL(bits) - 1))
  28
  29 /**
  30  * Structure for CQ moderation values.
  31  * Used for communications between DIM and its consumer.
  32  *
  33  * @usec: CQ timer suggestion (by DIM)
  34  * @pkts: CQ packet counter suggestion (by DIM)
  35  * @cq_period_mode: CQ priod count mode (from CQE/EQE)
  36  */
  37 struct dim_cq_moder {
  38         u16 usec;
  39         u16 pkts;
  40         u16 comps;
  41         u8 cq_period_mode;
  42 };
  43
  44 /**
  45  * Structure for DIM sample data.
  46  * Used for communications between DIM and its consumer.
  47  *
  48  * @time: Sample timestamp
  49  * @pkt_ctr: Number of packets
  50  * @byte_ctr: Number of bytes
  51  * @event_ctr: Number of events
  52  */
  53 struct dim_sample {
  54         ktime_t time;
  55         u32 pkt_ctr;
  56         u32 byte_ctr;
  57         u16 event_ctr;
  58         u32 comp_ctr;
  59 };
  60
  61 /**
  62  * Structure for DIM stats.
  63  * Used for holding current measured rates.
  64  *
  65  * @ppms: Packets per msec
  66  * @bpms: Bytes per msec
  67  * @epms: Events per msec
  68  */
  69 struct dim_stats {
  70         int ppms; /* packets per msec */
  71         int bpms; /* bytes per msec */
  72         int epms; /* events per msec */
  73         int cpms; /* completions per msec */
  74         int cpe_ratio; /* ratio of completions to events */
  75 };
  76
  77 /**
  78  * Main structure for dynamic interrupt moderation (DIM).
  79  * Used for holding all information about a specific DIM instance.
  80  *
  81  * @state: Algorithm state (see below)
  82  * @prev_stats: Measured rates from previous iteration (for comparison)
  83  * @start_sample: Sampled data at start of current iteration
  84  * @work: Work to perform on action required
  85  * @priv: A pointer to the struct that points to dim
  86  * @profile_ix: Current moderation profile
  87  * @mode: CQ period count mode
  88  * @tune_state: Algorithm tuning state (see below)
  89  * @steps_right: Number of steps taken towards higher moderation
  90  * @steps_left: Number of steps taken towards lower moderation
  91  * @tired: Parking depth counter
  92  */
  93 struct dim {
  94         u8 state;
  95         struct dim_stats prev_stats;
  96         struct dim_sample start_sample;
  97         struct dim_sample measuring_sample;
  98         struct work_struct work;
  99         void *priv;
 100         u8 profile_ix;
 101         u8 mode;
 102         u8 tune_state;
 103         u8 steps_right;
 104         u8 steps_left;
 105         u8 tired;
 106 };
 107
 108 /**
 109  * enum dim_cq_period_mode
 110  *
 111  * These are the modes for CQ period count.
 112  *
 113  * @DIM_CQ_PERIOD_MODE_START_FROM_EQE: Start counting from EQE
 114  * @DIM_CQ_PERIOD_MODE_START_FROM_CQE: Start counting from CQE (implies timer reset)
 115  * @DIM_CQ_PERIOD_NUM_MODES: Number of modes
 116  */
 117 enum {
 118         DIM_CQ_PERIOD_MODE_START_FROM_EQE = 0x0,
 119         DIM_CQ_PERIOD_MODE_START_FROM_CQE = 0x1,
 120         DIM_CQ_PERIOD_NUM_MODES
 121 };
 122
 123 /**
 124  * enum dim_state
 125  *
 126  * These are the DIM algorithm states.
 127  * These will determine if the algorithm is in a valid state to start an iteration.
 128  *
 129  * @DIM_START_MEASURE: This is the first iteration (also after applying a new profile)
 130  * @DIM_MEASURE_IN_PROGRESS: Algorithm is already in progress - check if
 131  * need to perform an action
 132  * @DIM_APPLY_NEW_PROFILE: DIM consumer is currently applying a profile - no need to measure
 133  */
 134 enum {
 135         DIM_START_MEASURE,
 136         DIM_MEASURE_IN_PROGRESS,
 137         DIM_APPLY_NEW_PROFILE,
 138 };
 139
 140 /**
 141  * enum dim_tune_state
 142  *
 143  * These are the DIM algorithm tune states.
 144  * These will determine which action the algorithm should perform.
 145  *
 146  * @DIM_PARKING_ON_TOP: Algorithm found a local top point - exit on significant difference
 147  * @DIM_PARKING_TIRED: Algorithm found a deep top point - don't exit if tired > 0
 148  * @DIM_GOING_RIGHT: Algorithm is currently trying higher moderation levels
 149  * @DIM_GOING_LEFT: Algorithm is currently trying lower moderation levels
 150  */
 151 enum {
 152         DIM_PARKING_ON_TOP,
 153         DIM_PARKING_TIRED,
 154         DIM_GOING_RIGHT,
 155         DIM_GOING_LEFT,
 156 };
 157
 158 /**
 159  * enum dim_stats_state
 160  *
 161  * These are the DIM algorithm statistics states.
 162  * These will determine the verdict of current iteration.
 163  *
 164  * @DIM_STATS_WORSE: Current iteration shows worse performance than before
 165  * @DIM_STATS_WORSE: Current iteration shows same performance than before
 166  * @DIM_STATS_WORSE: Current iteration shows better performance than before
 167  */
 168 enum {
 169         DIM_STATS_WORSE,
 170         DIM_STATS_SAME,
 171         DIM_STATS_BETTER,
 172 };
 173
 174 /**
 175  * enum dim_step_result
 176  *
 177  * These are the DIM algorithm step results.
 178  * These describe the result of a step.
 179  *
 180  * @DIM_STEPPED: Performed a regular step
 181  * @DIM_TOO_TIRED: Same kind of step was done multiple times - should go to
 182  * tired parking
 183  * @DIM_ON_EDGE: Stepped to the most left/right profile
 184  */
 185 enum {
 186         DIM_STEPPED,
 187         DIM_TOO_TIRED,
 188         DIM_ON_EDGE,
 189 };
 190
 191 /**
 192  *      dim_on_top - check if current state is a good place to stop (top location)
 193  *      @dim: DIM context
 194  *
 195  * Check if current profile is a good place to park at.
 196  * This will result in reducing the DIM checks frequency as we assume we
 197  * shouldn't probably change profiles, unless traffic pattern wasn't changed.
 198  */
 199 bool dim_on_top(struct dim *dim);
 200
 201 /**
 202  *      dim_turn - change profile alterning direction
 203  *      @dim: DIM context
 204  *
 205  * Go left if we were going right and vice-versa.
 206  * Do nothing if currently parking.
 207  */
 208 void dim_turn(struct dim *dim);
 209
 210 /**
 211  *      dim_park_on_top - enter a parking state on a top location
 212  *      @dim: DIM context
 213  *
 214  * Enter parking state.
 215  * Clear all movement history.
 216  */
 217 void dim_park_on_top(struct dim *dim);
 218
 219 /**
 220  *      dim_park_tired - enter a tired parking state
 221  *      @dim: DIM context
 222  *
 223  * Enter parking state.
 224  * Clear all movement history and cause DIM checks frequency to reduce.
 225  */
 226 void dim_park_tired(struct dim *dim);
 227
 228 /**
 229  *      dim_calc_stats - calculate the difference between two samples
 230  *      @start: start sample
 231  *      @end: end sample
 232  *      @curr_stats: delta between samples
 233  *
 234  * Calculate the delta between two samples (in data rates).
 235  * Takes into consideration counter wrap-around.
 236  */
 237 void dim_calc_stats(struct dim_sample *start, struct dim_sample *end,
 238                     struct dim_stats *curr_stats);
 239
 240 /**
 241  *      dim_update_sample - set a sample's fields with give values
 242  *      @event_ctr: number of events to set
 243  *      @packets: number of packets to set
 244  *      @bytes: number of bytes to set
 245  *      @s: DIM sample
 246  */
 247 static inline void
 248 dim_update_sample(u16 event_ctr, u64 packets, u64 bytes, struct dim_sample *s)
 249 {
 250         s->time      = ktime_get();
 251         s->pkt_ctr   = packets;
 252         s->byte_ctr  = bytes;
 253         s->event_ctr = event_ctr;
 254 }
 255
 256 /**
 257  *      dim_update_sample_with_comps - set a sample's fields with given
 258  *      values including the completion parameter
 259  *      @event_ctr: number of events to set
 260  *      @packets: number of packets to set
 261  *      @bytes: number of bytes to set
 262  *      @comps: number of completions to set
 263  *      @s: DIM sample
 264  */
 265 static inline void
 266 dim_update_sample_with_comps(u16 event_ctr, u64 packets, u64 bytes, u64 comps,
 267                              struct dim_sample *s)
 268 {
 269         dim_update_sample(event_ctr, packets, bytes, s);
 270         s->comp_ctr = comps;
 271 }
 272
 273 /* Net DIM */
 274
 275 /*
 276  * Net DIM profiles:
 277  *        There are different set of profiles for each CQ period mode.
 278  *        There are different set of profiles for RX/TX CQs.
 279  *        Each profile size must be of NET_DIM_PARAMS_NUM_PROFILES
 280  */
 281 #define NET_DIM_PARAMS_NUM_PROFILES 5
 282 #define NET_DIM_DEFAULT_RX_CQ_MODERATION_PKTS_FROM_EQE 256
 283 #define NET_DIM_DEFAULT_TX_CQ_MODERATION_PKTS_FROM_EQE 128
 284 #define NET_DIM_DEF_PROFILE_CQE 1
 285 #define NET_DIM_DEF_PROFILE_EQE 1
 286
 287 #define NET_DIM_RX_EQE_PROFILES { \
 288         {1,   NET_DIM_DEFAULT_RX_CQ_MODERATION_PKTS_FROM_EQE}, \
 289         {8,   NET_DIM_DEFAULT_RX_CQ_MODERATION_PKTS_FROM_EQE}, \
 290         {64,  NET_DIM_DEFAULT_RX_CQ_MODERATION_PKTS_FROM_EQE}, \
 291         {128, NET_DIM_DEFAULT_RX_CQ_MODERATION_PKTS_FROM_EQE}, \
 292         {256, NET_DIM_DEFAULT_RX_CQ_MODERATION_PKTS_FROM_EQE}, \
 293 }
 294
 295 #define NET_DIM_RX_CQE_PROFILES { \
 296         {2,  256},             \
 297         {8,  128},             \
 298         {16, 64},              \
 299         {32, 64},              \
 300         {64, 64}               \
 301 }
 302
 303 #define NET_DIM_TX_EQE_PROFILES { \
 304         {1,   NET_DIM_DEFAULT_TX_CQ_MODERATION_PKTS_FROM_EQE},  \
 305         {8,   NET_DIM_DEFAULT_TX_CQ_MODERATION_PKTS_FROM_EQE},  \
 306         {32,  NET_DIM_DEFAULT_TX_CQ_MODERATION_PKTS_FROM_EQE},  \
 307         {64,  NET_DIM_DEFAULT_TX_CQ_MODERATION_PKTS_FROM_EQE},  \
 308         {128, NET_DIM_DEFAULT_TX_CQ_MODERATION_PKTS_FROM_EQE}   \
 309 }
 310
 311 #define NET_DIM_TX_CQE_PROFILES { \
 312         {5,  128},  \
 313         {8,  64},  \
 314         {16, 32},  \
 315         {32, 32},  \
 316         {64, 32}   \
 317 }
 318
 319 static const struct dim_cq_moder
 320 rx_profile[DIM_CQ_PERIOD_NUM_MODES][NET_DIM_PARAMS_NUM_PROFILES] = {
 321         NET_DIM_RX_EQE_PROFILES,
 322         NET_DIM_RX_CQE_PROFILES,
 323 };
 324
 325 static const struct dim_cq_moder
 326 tx_profile[DIM_CQ_PERIOD_NUM_MODES][NET_DIM_PARAMS_NUM_PROFILES] = {
 327         NET_DIM_TX_EQE_PROFILES,
 328         NET_DIM_TX_CQE_PROFILES,
 329 };
 330
 331 /**
 332  *      net_dim_get_rx_moderation - provide a CQ moderation object for the given RX profile
 333  *      @cq_period_mode: CQ period mode
 334  *      @ix: Profile index
 335  */
 336 struct dim_cq_moder net_dim_get_rx_moderation(u8 cq_period_mode, int ix);
 337
 338 /**
 339  *      net_dim_get_def_rx_moderation - provide the default RX moderation
 340  *      @cq_period_mode: CQ period mode
 341  */
 342 struct dim_cq_moder net_dim_get_def_rx_moderation(u8 cq_period_mode);
 343
 344 /**
 345  *      net_dim_get_tx_moderation - provide a CQ moderation object for the given TX profile
 346  *      @cq_period_mode: CQ period mode
 347  *      @ix: Profile index
 348  */
 349 struct dim_cq_moder net_dim_get_tx_moderation(u8 cq_period_mode, int ix);
 350
 351 /**
 352  *      net_dim_get_def_tx_moderation - provide the default TX moderation
 353  *      @cq_period_mode: CQ period mode
 354  */
 355 struct dim_cq_moder net_dim_get_def_tx_moderation(u8 cq_period_mode);
 356
 357 /**
 358  *      net_dim - main DIM algorithm entry point
 359  *      @dim: DIM instance information
 360  *      @end_sample: Current data measurement
 361  *
 362  * Called by the consumer.
 363  * This is the main logic of the algorithm, where data is processed in order to decide on next
 364  * required action.
 365  */
 366 void net_dim(struct dim *dim, struct dim_sample end_sample);
 367
 368 /* RDMA DIM */
 369
 370 /*
 371  * RDMA DIM profile:
 372  * profile size must be of RDMA_DIM_PARAMS_NUM_PROFILES.
 373  */
 374 #define RDMA_DIM_PARAMS_NUM_PROFILES 9
 375 #define RDMA_DIM_START_PROFILE 0
 376
 377 /**
 378  * rdma_dim - Runs the adaptive moderation.
 379  * @dim: The moderation struct.
 380  * @completions: The number of completions collected in this round.
 381  *
 382  * Each call to rdma_dim takes the latest amount of completions that
 383  * have been collected and counts them as a new event.
 384  * Once enough events have been collected the algorithm decides a new
 385  * moderation level.
 386  */
 387 void rdma_dim(struct dim *dim, u64 completions);
 388
 389 #endif /* DIM_H */