linux/dim: Move implementation to .c files

Moved all logic from dim.h and net_dim.h to dim.c and net_dim.c.
This is both more structurally appealing and would allow to only
expose externally used functions.

Signed-off-by: Tal Gilboa <talgi@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
This commit is contained in:
Tal Gilboa 2019-01-10 17:33:17 +02:00 committed by Saeed Mahameed
parent 8960b38932
commit 4f75da3666
17 changed files with 546 additions and 344 deletions

View file

@ -6,20 +6,49 @@
#include <linux/module.h>
/**
* Number of events between DIM iterations.
* Causes a moderation of the algorithm run.
*/
#define DIM_NEVENTS 64
/* more than 10% difference */
/**
* Is a difference between values justifies taking an action.
* We consider 10% difference as significant.
*/
#define IS_SIGNIFICANT_DIFF(val, ref) \
(((100UL * abs((val) - (ref))) / (ref)) > 10)
#define BIT_GAP(bits, end, start) ((((end) - (start)) + BIT_ULL(bits)) \
& (BIT_ULL(bits) - 1))
/**
* Calculate the gap between two values.
* Take wrap-around and variable size into consideration.
*/
#define BIT_GAP(bits, end, start) ((((end) - (start)) + BIT_ULL(bits)) \
& (BIT_ULL(bits) - 1))
/**
* Structure for CQ moderation values.
* Used for communications between DIM and its consumer.
*
* @usec: CQ timer suggestion (by DIM)
* @pkts: CQ packet counter suggestion (by DIM)
* @cq_period_mode: CQ priod count mode (from CQE/EQE)
*/
struct dim_cq_moder {
u16 usec;
u16 pkts;
u8 cq_period_mode;
};
/**
* Structure for DIM sample data.
* Used for communications between DIM and its consumer.
*
* @time: Sample timestamp
* @pkt_ctr: Number of packets
* @byte_ctr: Number of bytes
* @event_ctr: Number of events
*/
struct dim_sample {
ktime_t time;
u32 pkt_ctr;
@ -27,13 +56,36 @@ struct dim_sample {
u16 event_ctr;
};
/**
* Structure for DIM stats.
* Used for holding current measured rates.
*
* @ppms: Packets per msec
* @bpms: Bytes per msec
* @epms: Events per msec
*/
struct dim_stats {
int ppms; /* packets per msec */
int bpms; /* bytes per msec */
int epms; /* events per msec */
int ppms;
int bpms;
int epms;
};
struct dim { /* Dynamic Interrupt Moderation */
/**
* Main structure for dynamic interrupt moderation (DIM).
* Used for holding all information about a specific DIM instance.
*
* @state: Algorithm state (see below)
* @prev_stats: Measured rates from previous iteration (for comparison)
* @start_sample: Sampled data at start of current iteration
* @work: Work to perform on action required
* @profile_ix: Current moderation profile
* @mode: CQ period count mode
* @tune_state: Algorithm tuning state (see below)
* @steps_right: Number of steps taken towards higher moderation
* @steps_left: Number of steps taken towards lower moderation
* @tired: Parking depth counter
*/
struct dim {
u8 state;
struct dim_stats prev_stats;
struct dim_sample start_sample;
@ -46,18 +98,49 @@ struct dim { /* Dynamic Interrupt Moderation */
u8 tired;
};
/**
* enum dim_cq_period_mode
*
* These are the modes for CQ period count.
*
* @DIM_CQ_PERIOD_MODE_START_FROM_EQE: Start counting from EQE
* @DIM_CQ_PERIOD_MODE_START_FROM_CQE: Start counting from CQE (implies timer reset)
* @DIM_CQ_PERIOD_NUM_MODES: Number of modes
*/
enum {
DIM_CQ_PERIOD_MODE_START_FROM_EQE = 0x0,
DIM_CQ_PERIOD_MODE_START_FROM_CQE = 0x1,
DIM_CQ_PERIOD_NUM_MODES
};
/**
* enum dim_state
*
* These are the DIM algorithm states.
* These will determine if the algorithm is in a valid state to start an iteration.
*
* @DIM_START_MEASURE: This is the first iteration (also after applying a new profile)
* @DIM_MEASURE_IN_PROGRESS: Algorithm is already in progress - check if
* need to perform an action
* @DIM_APPLY_NEW_PROFILE: DIM consumer is currently applying a profile - no need to measure
*/
enum {
DIM_START_MEASURE,
DIM_MEASURE_IN_PROGRESS,
DIM_APPLY_NEW_PROFILE,
};
/**
* enum dim_tune_state
*
* These are the DIM algorithm tune states.
* These will determine which action the algorithm should perform.
*
* @DIM_PARKING_ON_TOP: Algorithm found a local top point - exit on significant difference
* @DIM_PARKING_TIRED: Algorithm found a deep top point - don't exit if tired > 0
* @DIM_GOING_RIGHT: Algorithm is currently trying higher moderation levels
* @DIM_GOING_LEFT: Algorithm is currently trying lower moderation levels
*/
enum {
DIM_PARKING_ON_TOP,
DIM_PARKING_TIRED,
@ -65,63 +148,95 @@ enum {
DIM_GOING_LEFT,
};
/**
* enum dim_stats_state
*
* These are the DIM algorithm statistics states.
* These will determine the verdict of current iteration.
*
* @DIM_STATS_WORSE: Current iteration shows worse performance than before
* @DIM_STATS_WORSE: Current iteration shows same performance than before
* @DIM_STATS_WORSE: Current iteration shows better performance than before
*/
enum {
DIM_STATS_WORSE,
DIM_STATS_SAME,
DIM_STATS_BETTER,
};
/**
* enum dim_step_result
*
* These are the DIM algorithm step results.
* These describe the result of a step.
*
* @DIM_STEPPED: Performed a regular step
* @DIM_TOO_TIRED: Same kind of step was done multiple times - should go to
* tired parking
* @DIM_ON_EDGE: Stepped to the most left/right profile
*/
enum {
DIM_STEPPED,
DIM_TOO_TIRED,
DIM_ON_EDGE,
};
static inline bool dim_on_top(struct dim *dim)
{
switch (dim->tune_state) {
case DIM_PARKING_ON_TOP:
case DIM_PARKING_TIRED:
return true;
case DIM_GOING_RIGHT:
return (dim->steps_left > 1) && (dim->steps_right == 1);
default: /* DIM_GOING_LEFT */
return (dim->steps_right > 1) && (dim->steps_left == 1);
}
}
/**
* dim_on_top - check if current state is a good place to stop (top location)
* @dim: DIM context
*
* Check if current profile is a good place to park at.
* This will result in reducing the DIM checks frequency as we assume we
* shouldn't probably change profiles, unless traffic pattern wasn't changed.
*/
bool dim_on_top(struct dim *dim);
static inline void dim_turn(struct dim *dim)
{
switch (dim->tune_state) {
case DIM_PARKING_ON_TOP:
case DIM_PARKING_TIRED:
break;
case DIM_GOING_RIGHT:
dim->tune_state = DIM_GOING_LEFT;
dim->steps_left = 0;
break;
case DIM_GOING_LEFT:
dim->tune_state = DIM_GOING_RIGHT;
dim->steps_right = 0;
break;
}
}
/**
* dim_turn - change profile alterning direction
* @dim: DIM context
*
* Go left if we were going right and vice-versa.
* Do nothing if currently parking.
*/
void dim_turn(struct dim *dim);
static inline void dim_park_on_top(struct dim *dim)
{
dim->steps_right = 0;
dim->steps_left = 0;
dim->tired = 0;
dim->tune_state = DIM_PARKING_ON_TOP;
}
/**
* dim_park_on_top - enter a parking state on a top location
* @dim: DIM context
*
* Enter parking state.
* Clear all movement history.
*/
void dim_park_on_top(struct dim *dim);
static inline void dim_park_tired(struct dim *dim)
{
dim->steps_right = 0;
dim->steps_left = 0;
dim->tune_state = DIM_PARKING_TIRED;
}
/**
* dim_park_tired - enter a tired parking state
* @dim: DIM context
*
* Enter parking state.
* Clear all movement history and cause DIM checks frequency to reduce.
*/
void dim_park_tired(struct dim *dim);
/**
* dim_calc_stats - calculate the difference between two samples
* @start: start sample
* @end: end sample
* @curr_stats: delta between samples
*
* Calculate the delta between two samples (in data rates).
* Takes into consideration counter wrap-around.
*/
void dim_calc_stats(struct dim_sample *start, struct dim_sample *end,
struct dim_stats *curr_stats);
/**
* dim_update_sample - set a sample's fields with give values
* @event_ctr: number of events to set
* @packets: number of packets to set
* @bytes: number of bytes to set
* @s: DIM sample
*/
static inline void
dim_update_sample(u16 event_ctr, u64 packets, u64 bytes, struct dim_sample *s)
{
@ -131,23 +246,99 @@ dim_update_sample(u16 event_ctr, u64 packets, u64 bytes, struct dim_sample *s)
s->event_ctr = event_ctr;
}
static inline void
dim_calc_stats(struct dim_sample *start, struct dim_sample *end,
struct dim_stats *curr_stats)
{
/* u32 holds up to 71 minutes, should be enough */
u32 delta_us = ktime_us_delta(end->time, start->time);
u32 npkts = BIT_GAP(BITS_PER_TYPE(u32), end->pkt_ctr, start->pkt_ctr);
u32 nbytes = BIT_GAP(BITS_PER_TYPE(u32), end->byte_ctr,
start->byte_ctr);
/* Net DIM */
if (!delta_us)
return;
/*
* Net DIM profiles:
* There are different set of profiles for each CQ period mode.
* There are different set of profiles for RX/TX CQs.
* Each profile size must be of NET_DIM_PARAMS_NUM_PROFILES
*/
#define NET_DIM_PARAMS_NUM_PROFILES 5
#define NET_DIM_DEFAULT_RX_CQ_MODERATION_PKTS_FROM_EQE 256
#define NET_DIM_DEFAULT_TX_CQ_MODERATION_PKTS_FROM_EQE 128
#define NET_DIM_DEF_PROFILE_CQE 1
#define NET_DIM_DEF_PROFILE_EQE 1
curr_stats->ppms = DIV_ROUND_UP(npkts * USEC_PER_MSEC, delta_us);
curr_stats->bpms = DIV_ROUND_UP(nbytes * USEC_PER_MSEC, delta_us);
curr_stats->epms = DIV_ROUND_UP(DIM_NEVENTS * USEC_PER_MSEC,
delta_us);
#define NET_DIM_RX_EQE_PROFILES { \
{1, NET_DIM_DEFAULT_RX_CQ_MODERATION_PKTS_FROM_EQE}, \
{8, NET_DIM_DEFAULT_RX_CQ_MODERATION_PKTS_FROM_EQE}, \
{64, NET_DIM_DEFAULT_RX_CQ_MODERATION_PKTS_FROM_EQE}, \
{128, NET_DIM_DEFAULT_RX_CQ_MODERATION_PKTS_FROM_EQE}, \
{256, NET_DIM_DEFAULT_RX_CQ_MODERATION_PKTS_FROM_EQE}, \
}
#define NET_DIM_RX_CQE_PROFILES { \
{2, 256}, \
{8, 128}, \
{16, 64}, \
{32, 64}, \
{64, 64} \
}
#define NET_DIM_TX_EQE_PROFILES { \
{1, NET_DIM_DEFAULT_TX_CQ_MODERATION_PKTS_FROM_EQE}, \
{8, NET_DIM_DEFAULT_TX_CQ_MODERATION_PKTS_FROM_EQE}, \
{32, NET_DIM_DEFAULT_TX_CQ_MODERATION_PKTS_FROM_EQE}, \
{64, NET_DIM_DEFAULT_TX_CQ_MODERATION_PKTS_FROM_EQE}, \
{128, NET_DIM_DEFAULT_TX_CQ_MODERATION_PKTS_FROM_EQE} \
}
#define NET_DIM_TX_CQE_PROFILES { \
{5, 128}, \
{8, 64}, \
{16, 32}, \
{32, 32}, \
{64, 32} \
}
static const struct dim_cq_moder
rx_profile[DIM_CQ_PERIOD_NUM_MODES][NET_DIM_PARAMS_NUM_PROFILES] = {
NET_DIM_RX_EQE_PROFILES,
NET_DIM_RX_CQE_PROFILES,
};
static const struct dim_cq_moder
tx_profile[DIM_CQ_PERIOD_NUM_MODES][NET_DIM_PARAMS_NUM_PROFILES] = {
NET_DIM_TX_EQE_PROFILES,
NET_DIM_TX_CQE_PROFILES,
};
/**
* net_dim_get_rx_moderation - provide a CQ moderation object for the given RX profile
* @cq_period_mode: CQ period mode
* @ix: Profile index
*/
struct dim_cq_moder net_dim_get_rx_moderation(u8 cq_period_mode, int ix);
/**
* net_dim_get_def_rx_moderation - provide the default RX moderation
* @cq_period_mode: CQ period mode
*/
struct dim_cq_moder net_dim_get_def_rx_moderation(u8 cq_period_mode);
/**
* net_dim_get_tx_moderation - provide a CQ moderation object for the given TX profile
* @cq_period_mode: CQ period mode
* @ix: Profile index
*/
struct dim_cq_moder net_dim_get_tx_moderation(u8 cq_period_mode, int ix);
/**
* net_dim_get_def_tx_moderation - provide the default TX moderation
* @cq_period_mode: CQ period mode
*/
struct dim_cq_moder net_dim_get_def_tx_moderation(u8 cq_period_mode);
/**
* net_dim - main DIM algorithm entry point
* @dim: DIM instance information
* @end_sample: Current data measurement
*
* Called by the consumer.
* This is the main logic of the algorithm, where data is processed in order to decide on next
* required action.
*/
void net_dim(struct dim *dim, struct dim_sample end_sample);
#endif /* DIM_H */