diff options
author | Glenn Elliott <gelliott@cs.unc.edu> | 2012-08-20 17:28:55 -0400 |
---|---|---|
committer | Glenn Elliott <gelliott@cs.unc.edu> | 2012-08-27 15:07:17 -0400 |
commit | 077aaecac31331b65442275843932314049a2ceb (patch) | |
tree | 6b24a59521a0f5e3853b7667669b54211b03b272 | |
parent | 9a19f35c9c287cb8abd5bcf276ae8d1a3e876907 (diff) |
EDF priority tie-breaks.wip-robust-tie-break
Instead of tie-breaking by PID (which is a static
priority tie-break), we can tie-break by other
job-level-unique parameters. This is desirable
because tasks are equaly affected by tardiness
since static priority tie-breaks cause tasks
with greater PID values to experience the most
tardiness.
There are four tie-break methods:
1) Lateness. If two jobs, J_{1,i} and J_{2,j} of
tasks T_1 and T_2, respectively, have equal
deadlines, we favor the job of the task that
had the worst lateness for jobs J_{1,i-1} and
J_{2,j-1}.
Note: Unlike tardiness, lateness may be less than
zero. This occurs when a job finishes before its
deadline.
2) Normalized Lateness. The same as #1, except
lateness is first normalized by each task's
relative deadline. This prevents tasks with short
relative deadlines and small execution requirements
from always losing tie-breaks.
3) Hash. The job tuple (PID, Job#) is used to
generate a hash. Hash values are then compared.
A job has ~50% chance of winning a tie-break
with respect to another job.
Note: Emperical testing shows that some jobs
can have +/- ~1.5% advantage in tie-breaks.
Linux's built-in hash function is not totally
a uniform hash.
4) PIDs. PID-based tie-break used in prior
versions of Litmus.
Conflicts:
litmus/edf_common.c
-rw-r--r-- | include/litmus/fpmath.h | 145 | ||||
-rw-r--r-- | include/litmus/litmus.h | 2 | ||||
-rw-r--r-- | include/litmus/rt_param.h | 6 | ||||
-rw-r--r-- | litmus/Kconfig | 46 | ||||
-rw-r--r-- | litmus/edf_common.c | 110 | ||||
-rw-r--r-- | litmus/jobs.c | 8 |
6 files changed, 297 insertions, 20 deletions
diff --git a/include/litmus/fpmath.h b/include/litmus/fpmath.h new file mode 100644 index 000000000000..04d4bcaeae96 --- /dev/null +++ b/include/litmus/fpmath.h | |||
@@ -0,0 +1,145 @@ | |||
1 | #ifndef __FP_MATH_H__ | ||
2 | #define __FP_MATH_H__ | ||
3 | |||
4 | #ifndef __KERNEL__ | ||
5 | #include <stdint.h> | ||
6 | #define abs(x) (((x) < 0) ? -(x) : x) | ||
7 | #endif | ||
8 | |||
9 | // Use 64-bit because we want to track things at the nanosecond scale. | ||
10 | // This can lead to very large numbers. | ||
11 | typedef int64_t fpbuf_t; | ||
12 | typedef struct | ||
13 | { | ||
14 | fpbuf_t val; | ||
15 | } fp_t; | ||
16 | |||
17 | #define FP_SHIFT 10 | ||
18 | #define ROUND_BIT (FP_SHIFT - 1) | ||
19 | |||
20 | #define _fp(x) ((fp_t) {x}) | ||
21 | |||
22 | #ifdef __KERNEL__ | ||
23 | static const fp_t LITMUS_FP_ZERO = {.val = 0}; | ||
24 | static const fp_t LITMUS_FP_ONE = {.val = (1 << FP_SHIFT)}; | ||
25 | #endif | ||
26 | |||
27 | static inline fp_t FP(fpbuf_t x) | ||
28 | { | ||
29 | return _fp(((fpbuf_t) x) << FP_SHIFT); | ||
30 | } | ||
31 | |||
32 | /* divide two integers to obtain a fixed point value */ | ||
33 | static inline fp_t _frac(fpbuf_t a, fpbuf_t b) | ||
34 | { | ||
35 | return _fp(FP(a).val / (b)); | ||
36 | } | ||
37 | |||
38 | static inline fpbuf_t _point(fp_t x) | ||
39 | { | ||
40 | return (x.val % (1 << FP_SHIFT)); | ||
41 | |||
42 | } | ||
43 | |||
44 | #define fp2str(x) x.val | ||
45 | /*(x.val >> FP_SHIFT), (x.val % (1 << FP_SHIFT)) */ | ||
46 | #define _FP_ "%ld/1024" | ||
47 | |||
48 | static inline fpbuf_t _floor(fp_t x) | ||
49 | { | ||
50 | return x.val >> FP_SHIFT; | ||
51 | } | ||
52 | |||
53 | /* FIXME: negative rounding */ | ||
54 | static inline fpbuf_t _round(fp_t x) | ||
55 | { | ||
56 | return _floor(x) + ((x.val >> ROUND_BIT) & 1); | ||
57 | } | ||
58 | |||
59 | /* multiply two fixed point values */ | ||
60 | static inline fp_t _mul(fp_t a, fp_t b) | ||
61 | { | ||
62 | return _fp((a.val * b.val) >> FP_SHIFT); | ||
63 | } | ||
64 | |||
65 | static inline fp_t _div(fp_t a, fp_t b) | ||
66 | { | ||
67 | #if !defined(__KERNEL__) && !defined(unlikely) | ||
68 | #define unlikely(x) (x) | ||
69 | #define DO_UNDEF_UNLIKELY | ||
70 | #endif | ||
71 | /* try not to overflow */ | ||
72 | if (unlikely( a.val > (2l << ((sizeof(fpbuf_t)*8) - FP_SHIFT)) )) | ||
73 | return _fp((a.val / b.val) << FP_SHIFT); | ||
74 | else | ||
75 | return _fp((a.val << FP_SHIFT) / b.val); | ||
76 | #ifdef DO_UNDEF_UNLIKELY | ||
77 | #undef unlikely | ||
78 | #undef DO_UNDEF_UNLIKELY | ||
79 | #endif | ||
80 | } | ||
81 | |||
82 | static inline fp_t _add(fp_t a, fp_t b) | ||
83 | { | ||
84 | return _fp(a.val + b.val); | ||
85 | } | ||
86 | |||
87 | static inline fp_t _sub(fp_t a, fp_t b) | ||
88 | { | ||
89 | return _fp(a.val - b.val); | ||
90 | } | ||
91 | |||
92 | static inline fp_t _neg(fp_t x) | ||
93 | { | ||
94 | return _fp(-x.val); | ||
95 | } | ||
96 | |||
97 | static inline fp_t _abs(fp_t x) | ||
98 | { | ||
99 | return _fp(abs(x.val)); | ||
100 | } | ||
101 | |||
102 | /* works the same as casting float/double to integer */ | ||
103 | static inline fpbuf_t _fp_to_integer(fp_t x) | ||
104 | { | ||
105 | return _floor(_abs(x)) * ((x.val > 0) ? 1 : -1); | ||
106 | } | ||
107 | |||
108 | static inline fp_t _integer_to_fp(fpbuf_t x) | ||
109 | { | ||
110 | return _frac(x,1); | ||
111 | } | ||
112 | |||
113 | static inline int _leq(fp_t a, fp_t b) | ||
114 | { | ||
115 | return a.val <= b.val; | ||
116 | } | ||
117 | |||
118 | static inline int _geq(fp_t a, fp_t b) | ||
119 | { | ||
120 | return a.val >= b.val; | ||
121 | } | ||
122 | |||
123 | static inline int _lt(fp_t a, fp_t b) | ||
124 | { | ||
125 | return a.val < b.val; | ||
126 | } | ||
127 | |||
128 | static inline int _gt(fp_t a, fp_t b) | ||
129 | { | ||
130 | return a.val > b.val; | ||
131 | } | ||
132 | |||
133 | static inline int _eq(fp_t a, fp_t b) | ||
134 | { | ||
135 | return a.val == b.val; | ||
136 | } | ||
137 | |||
138 | static inline fp_t _max(fp_t a, fp_t b) | ||
139 | { | ||
140 | if (a.val < b.val) | ||
141 | return b; | ||
142 | else | ||
143 | return a; | ||
144 | } | ||
145 | #endif | ||
diff --git a/include/litmus/litmus.h b/include/litmus/litmus.h index 338245abd6ed..807b7888695a 100644 --- a/include/litmus/litmus.h +++ b/include/litmus/litmus.h | |||
@@ -63,7 +63,7 @@ void litmus_exit_task(struct task_struct *tsk); | |||
63 | #define get_exec_time(t) (tsk_rt(t)->job_params.exec_time) | 63 | #define get_exec_time(t) (tsk_rt(t)->job_params.exec_time) |
64 | #define get_deadline(t) (tsk_rt(t)->job_params.deadline) | 64 | #define get_deadline(t) (tsk_rt(t)->job_params.deadline) |
65 | #define get_release(t) (tsk_rt(t)->job_params.release) | 65 | #define get_release(t) (tsk_rt(t)->job_params.release) |
66 | 66 | #define get_lateness(t) (tsk_rt(t)->job_params.lateness) | |
67 | 67 | ||
68 | #define is_hrt(t) \ | 68 | #define is_hrt(t) \ |
69 | (tsk_rt(t)->task_params.cls == RT_CLASS_HARD) | 69 | (tsk_rt(t)->task_params.cls == RT_CLASS_HARD) |
diff --git a/include/litmus/rt_param.h b/include/litmus/rt_param.h index 89ac0dda7d3d..fac939dbd33a 100644 --- a/include/litmus/rt_param.h +++ b/include/litmus/rt_param.h | |||
@@ -110,6 +110,12 @@ struct rt_job { | |||
110 | /* How much service has this job received so far? */ | 110 | /* How much service has this job received so far? */ |
111 | lt_t exec_time; | 111 | lt_t exec_time; |
112 | 112 | ||
113 | /* By how much did the prior job miss its deadline by? | ||
114 | * Value differs from tardiness in that lateness may | ||
115 | * be negative (when job finishes before its deadline). | ||
116 | */ | ||
117 | long long lateness; | ||
118 | |||
113 | /* Which job is this. This is used to let user space | 119 | /* Which job is this. This is used to let user space |
114 | * specify which job to wait for, which is important if jobs | 120 | * specify which job to wait for, which is important if jobs |
115 | * overrun. If we just call sys_sleep_next_period() then we | 121 | * overrun. If we just call sys_sleep_next_period() then we |
diff --git a/litmus/Kconfig b/litmus/Kconfig index 68459d4dca41..48ff3e3c657c 100644 --- a/litmus/Kconfig +++ b/litmus/Kconfig | |||
@@ -79,6 +79,52 @@ config SCHED_CPU_AFFINITY | |||
79 | 79 | ||
80 | Say Yes if unsure. | 80 | Say Yes if unsure. |
81 | 81 | ||
82 | choice | ||
83 | prompt "EDF Tie-Break Behavior" | ||
84 | default EDF_TIE_BREAK_LATENESS_NORM | ||
85 | help | ||
86 | Allows the configuration of tie-breaking behavior when the deadlines | ||
87 | of two EDF-scheduled tasks are equal. | ||
88 | |||
89 | config EDF_TIE_BREAK_LATENESS | ||
90 | bool "Lateness-based Tie Break" | ||
91 | help | ||
92 | Break ties between to jobs, A and B, based upon the lateness of their | ||
93 | prior jobs. The job with the greatest lateness has priority. Note that | ||
94 | lateness has a negative value if the prior job finished before its | ||
95 | deadline. | ||
96 | |||
97 | config EDF_TIE_BREAK_LATENESS_NORM | ||
98 | bool "Normalized Lateness-based Tie Break" | ||
99 | help | ||
100 | Break ties between to jobs, A and B, based upon the lateness, normalized | ||
101 | by relative deadline, their prior jobs. The job with the greatest | ||
102 | normalized lateness has priority. Note that lateness has a negative value | ||
103 | if the prior job finished before its deadline. | ||
104 | |||
105 | Normalized lateness tie-breaks are likely desireable over non-normalized | ||
106 | tie-breaks if the execution times and/or relative deadlines of tasks in a | ||
107 | task set vary greatly. | ||
108 | |||
109 | config EDF_TIE_BREAK_HASH | ||
110 | bool "Hash-based Tie Breaks" | ||
111 | help | ||
112 | Break ties between two jobs, A and B, with equal deadlines by using a | ||
113 | uniform hash; i.e.: hash(A.pid, A.job_num) < hash(B.pid, B.job_num). Job | ||
114 | A has ~50% of winning a given tie-break. | ||
115 | |||
116 | config EDF_PID_TIE_BREAK | ||
117 | bool "PID-based Tie Breaks" | ||
118 | help | ||
119 | Break ties based upon OS-assigned process IDs. Use this option if | ||
120 | required by algorithm's real-time analysis or per-task response-time | ||
121 | jitter must be minimized in overload conditions. | ||
122 | |||
123 | NOTES: | ||
124 | * This tie-breaking method was default in Litmus 2012.2 and before. | ||
125 | |||
126 | endchoice | ||
127 | |||
82 | endmenu | 128 | endmenu |
83 | 129 | ||
84 | menu "Tracing" | 130 | menu "Tracing" |
diff --git a/litmus/edf_common.c b/litmus/edf_common.c index 668737f0fbf9..52205df3ea8b 100644 --- a/litmus/edf_common.c +++ b/litmus/edf_common.c | |||
@@ -14,6 +14,32 @@ | |||
14 | 14 | ||
15 | #include <litmus/edf_common.h> | 15 | #include <litmus/edf_common.h> |
16 | 16 | ||
17 | #ifdef CONFIG_EDF_TIE_BREAK_LATENESS_NORM | ||
18 | #include <litmus/fpmath.h> | ||
19 | #endif | ||
20 | |||
21 | #ifdef CONFIG_EDF_TIE_BREAK_HASH | ||
22 | #include <linux/hash.h> | ||
23 | static inline long edf_hash(struct task_struct *t) | ||
24 | { | ||
25 | /* pid is 32 bits, so normally we would shove that into the | ||
26 | * upper 32-bits and and put the job number in the bottom | ||
27 | * and hash the 64-bit number with hash_64(). Sadly, | ||
28 | * in testing, hash_64() doesn't distribute keys were the | ||
29 | * upper bits are close together (as would be the case with | ||
30 | * pids) and job numbers are equal (as would be the case with | ||
31 | * synchronous task sets with all relative deadlines equal). | ||
32 | * | ||
33 | * A 2006 Linux patch proposed the following solution | ||
34 | * (but for some reason it wasn't accepted...). | ||
35 | * | ||
36 | * At least this workaround works for 32-bit systems as well. | ||
37 | */ | ||
38 | return hash_32(hash_32((u32)tsk_rt(t)->job_params.job_no, 32) ^ t->pid, 32); | ||
39 | } | ||
40 | #endif | ||
41 | |||
42 | |||
17 | /* edf_higher_prio - returns true if first has a higher EDF priority | 43 | /* edf_higher_prio - returns true if first has a higher EDF priority |
18 | * than second. Deadline ties are broken by PID. | 44 | * than second. Deadline ties are broken by PID. |
19 | * | 45 | * |
@@ -63,32 +89,78 @@ int edf_higher_prio(struct task_struct* first, | |||
63 | 89 | ||
64 | #endif | 90 | #endif |
65 | 91 | ||
66 | /* Determine the task with earliest deadline, with | 92 | if (earlier_deadline(first_task, second_task)) { |
67 | * tie-break logic. | ||
68 | */ | ||
69 | if (unlikely(!is_realtime(second_task))) { | ||
70 | return 1; | ||
71 | } | ||
72 | else if (earlier_deadline(first_task, second_task)) { | ||
73 | /* Is the deadline of the first task earlier? | ||
74 | * Then it has higher priority. | ||
75 | */ | ||
76 | return 1; | 93 | return 1; |
77 | } | 94 | } |
78 | else if (get_deadline(first_task) == get_deadline(second_task)) { | 95 | else if (get_deadline(first_task) == get_deadline(second_task)) { |
79 | /* Need to tie break */ | 96 | /* Need to tie break. All methods must set pid_break to 0/1 if |
80 | 97 | * first_task does not have priority over second_task. | |
81 | /* Tie break by pid */ | 98 | */ |
82 | if (first_task->pid < second_task->pid) { | 99 | int pid_break; |
100 | |||
101 | |||
102 | #if defined(CONFIG_EDF_TIE_BREAK_LATENESS) | ||
103 | /* Tie break by lateness. Jobs with greater lateness get | ||
104 | * priority. This should spread tardiness across all tasks, | ||
105 | * especially in task sets where all tasks have the same | ||
106 | * period and relative deadlines. | ||
107 | */ | ||
108 | if (get_lateness(first_task) > get_lateness(second_task)) { | ||
83 | return 1; | 109 | return 1; |
84 | } | 110 | } |
85 | else if (first_task->pid == second_task->pid) { | 111 | pid_break = (get_lateness(first_task) == get_lateness(second_task)); |
86 | /* If the PIDs are the same then the task with the | 112 | |
87 | * inherited priority wins. | 113 | |
88 | */ | 114 | #elif defined(CONFIG_EDF_TIE_BREAK_LATENESS_NORM) |
89 | if (!second_task->rt_param.inh_task) { | 115 | /* Tie break by lateness, normalized by relative deadline. Jobs with |
116 | * greater normalized lateness get priority. | ||
117 | * | ||
118 | * Note: Considered using the algebraically equivalent | ||
119 | * lateness(first)*relative_deadline(second) > | ||
120 | lateness(second)*relative_deadline(first) | ||
121 | * to avoid fixed-point math, but values are prone to overflow if inputs | ||
122 | * are on the order of several seconds, even in 64-bit. | ||
123 | */ | ||
124 | fp_t fnorm = _frac(get_lateness(first_task), | ||
125 | get_rt_relative_deadline(first_task)); | ||
126 | fp_t snorm = _frac(get_lateness(second_task), | ||
127 | get_rt_relative_deadline(second_task)); | ||
128 | if (_gt(fnorm, snorm)) { | ||
129 | return 1; | ||
130 | } | ||
131 | pid_break = _eq(fnorm, snorm); | ||
132 | |||
133 | |||
134 | #elif defined(CONFIG_EDF_TIE_BREAK_HASH) | ||
135 | /* Tie break by comparing hashs of (pid, job#) tuple. There should be | ||
136 | * a 50% chance that first_task has a higher priority than second_task. | ||
137 | */ | ||
138 | long fhash = edf_hash(first_task); | ||
139 | long shash = edf_hash(second_task); | ||
140 | if (fhash < shash) { | ||
141 | return 1; | ||
142 | } | ||
143 | pid_break = (fhash == shash); | ||
144 | #else | ||
145 | |||
146 | |||
147 | /* CONFIG_EDF_PID_TIE_BREAK */ | ||
148 | pid_break = 1; // fall through to tie-break by pid; | ||
149 | #endif | ||
150 | |||
151 | /* Tie break by pid */ | ||
152 | if(pid_break) { | ||
153 | if (first_task->pid < second_task->pid) { | ||
90 | return 1; | 154 | return 1; |
91 | } | 155 | } |
156 | else if (first_task->pid == second_task->pid) { | ||
157 | /* If the PIDs are the same then the task with the | ||
158 | * inherited priority wins. | ||
159 | */ | ||
160 | if (!second_task->rt_param.inh_task) { | ||
161 | return 1; | ||
162 | } | ||
163 | } | ||
92 | } | 164 | } |
93 | } | 165 | } |
94 | return 0; /* fall-through. prio(second_task) > prio(first_task) */ | 166 | return 0; /* fall-through. prio(second_task) > prio(first_task) */ |
diff --git a/litmus/jobs.c b/litmus/jobs.c index bc8246572e54..fb093c03d53d 100644 --- a/litmus/jobs.c +++ b/litmus/jobs.c | |||
@@ -23,6 +23,14 @@ static inline void setup_release(struct task_struct *t, lt_t release) | |||
23 | void prepare_for_next_period(struct task_struct *t) | 23 | void prepare_for_next_period(struct task_struct *t) |
24 | { | 24 | { |
25 | BUG_ON(!t); | 25 | BUG_ON(!t); |
26 | |||
27 | /* Record lateness before we set up the next job's | ||
28 | * release and deadline. Lateness may be negative. | ||
29 | */ | ||
30 | t->rt_param.job_params.lateness = | ||
31 | (long long)litmus_clock() - | ||
32 | (long long)t->rt_param.job_params.deadline; | ||
33 | |||
26 | setup_release(t, get_release(t) + get_rt_period(t)); | 34 | setup_release(t, get_release(t) + get_rt_period(t)); |
27 | } | 35 | } |
28 | 36 | ||