diff options
author | Patrick McHardy <kaber@trash.net> | 2007-03-16 04:18:42 -0400 |
---|---|---|
committer | David S. Miller <davem@sunset.davemloft.net> | 2007-04-26 01:26:04 -0400 |
commit | 641b9e0e8b7f96425da6ce98f3361e3af0baee29 (patch) | |
tree | 2315fed3b4fd9df52a52464b9b1ce1561d403a87 /net/sched | |
parent | ddc7b8e32b22fe8b45d306b7d99472d4b560add6 (diff) |
[NET_SCHED]: Use ktime as clocksource
Get rid of the manual clock source selection mess and use ktime. Also
use a scalar representation, which allows to clean up pkt_sched.h a bit
more and results in less ktime_to_ns() calls in most cases.
The PSCHED_US2JIFFIE/PSCHED_JIFFIE2US macros are implemented quite
inefficient by this patch, following patches will convert all qdiscs
to hrtimers and get rid of them entirely.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Diffstat (limited to 'net/sched')
-rw-r--r-- | net/sched/Kconfig | 56 | ||||
-rw-r--r-- | net/sched/sch_api.c | 77 | ||||
-rw-r--r-- | net/sched/sch_hfsc.c | 31 |
3 files changed, 5 insertions, 159 deletions
diff --git a/net/sched/Kconfig b/net/sched/Kconfig index f4544dd86476..475df8449be9 100644 --- a/net/sched/Kconfig +++ b/net/sched/Kconfig | |||
@@ -46,62 +46,6 @@ config NET_SCH_FIFO | |||
46 | 46 | ||
47 | if NET_SCHED | 47 | if NET_SCHED |
48 | 48 | ||
49 | choice | ||
50 | prompt "Packet scheduler clock source" | ||
51 | default NET_SCH_CLK_GETTIMEOFDAY | ||
52 | ---help--- | ||
53 | Packet schedulers need a monotonic clock that increments at a static | ||
54 | rate. The kernel provides several suitable interfaces, each with | ||
55 | different properties: | ||
56 | |||
57 | - high resolution (us or better) | ||
58 | - fast to read (minimal locking, no i/o access) | ||
59 | - synchronized on all processors | ||
60 | - handles cpu clock frequency changes | ||
61 | |||
62 | but nothing provides all of the above. | ||
63 | |||
64 | config NET_SCH_CLK_JIFFIES | ||
65 | bool "Timer interrupt" | ||
66 | ---help--- | ||
67 | Say Y here if you want to use the timer interrupt (jiffies) as clock | ||
68 | source. This clock source is fast, synchronized on all processors and | ||
69 | handles cpu clock frequency changes, but its resolution is too low | ||
70 | for accurate shaping except at very low speed. | ||
71 | |||
72 | config NET_SCH_CLK_GETTIMEOFDAY | ||
73 | bool "gettimeofday" | ||
74 | ---help--- | ||
75 | Say Y here if you want to use gettimeofday as clock source. This clock | ||
76 | source has high resolution, is synchronized on all processors and | ||
77 | handles cpu clock frequency changes, but it is slow. | ||
78 | |||
79 | Choose this if you need a high resolution clock source but can't use | ||
80 | the CPU's cycle counter. | ||
81 | |||
82 | # don't allow on SMP x86 because they can have unsynchronized TSCs. | ||
83 | # gettimeofday is a good alternative | ||
84 | config NET_SCH_CLK_CPU | ||
85 | bool "CPU cycle counter" | ||
86 | depends on ((X86_TSC || X86_64) && !SMP) || ALPHA || SPARC64 || PPC64 || IA64 | ||
87 | ---help--- | ||
88 | Say Y here if you want to use the CPU's cycle counter as clock source. | ||
89 | This is a cheap and high resolution clock source, but on some | ||
90 | architectures it is not synchronized on all processors and doesn't | ||
91 | handle cpu clock frequency changes. | ||
92 | |||
93 | The useable cycle counters are: | ||
94 | |||
95 | x86/x86_64 - Timestamp Counter | ||
96 | alpha - Cycle Counter | ||
97 | sparc64 - %ticks register | ||
98 | ppc64 - Time base | ||
99 | ia64 - Interval Time Counter | ||
100 | |||
101 | Choose this if your CPU's cycle counter is working properly. | ||
102 | |||
103 | endchoice | ||
104 | |||
105 | comment "Queueing/Scheduling" | 49 | comment "Queueing/Scheduling" |
106 | 50 | ||
107 | config NET_SCH_CBQ | 51 | config NET_SCH_CBQ |
diff --git a/net/sched/sch_api.c b/net/sched/sch_api.c index 4a927a5e1fa6..d71bf79eb80b 100644 --- a/net/sched/sch_api.c +++ b/net/sched/sch_api.c | |||
@@ -1175,15 +1175,12 @@ reclassify: | |||
1175 | return -1; | 1175 | return -1; |
1176 | } | 1176 | } |
1177 | 1177 | ||
1178 | static int psched_us_per_tick = 1; | ||
1179 | static int psched_tick_per_us = 1; | ||
1180 | |||
1181 | #ifdef CONFIG_PROC_FS | 1178 | #ifdef CONFIG_PROC_FS |
1182 | static int psched_show(struct seq_file *seq, void *v) | 1179 | static int psched_show(struct seq_file *seq, void *v) |
1183 | { | 1180 | { |
1184 | seq_printf(seq, "%08x %08x %08x %08x\n", | 1181 | seq_printf(seq, "%08x %08x %08x %08x\n", |
1185 | psched_tick_per_us, psched_us_per_tick, | 1182 | (u32)NSEC_PER_USEC, (u32)PSCHED_US2NS(1), |
1186 | 1000000, HZ); | 1183 | 1000000, HZ); |
1187 | 1184 | ||
1188 | return 0; | 1185 | return 0; |
1189 | } | 1186 | } |
@@ -1202,80 +1199,10 @@ static const struct file_operations psched_fops = { | |||
1202 | }; | 1199 | }; |
1203 | #endif | 1200 | #endif |
1204 | 1201 | ||
1205 | #ifdef CONFIG_NET_SCH_CLK_CPU | ||
1206 | psched_tdiff_t psched_clock_per_hz; | ||
1207 | int psched_clock_scale; | ||
1208 | EXPORT_SYMBOL(psched_clock_per_hz); | ||
1209 | EXPORT_SYMBOL(psched_clock_scale); | ||
1210 | |||
1211 | psched_time_t psched_time_base; | ||
1212 | cycles_t psched_time_mark; | ||
1213 | EXPORT_SYMBOL(psched_time_mark); | ||
1214 | EXPORT_SYMBOL(psched_time_base); | ||
1215 | |||
1216 | /* | ||
1217 | * Periodically adjust psched_time_base to avoid overflow | ||
1218 | * with 32-bit get_cycles(). Safe up to 4GHz CPU. | ||
1219 | */ | ||
1220 | static void psched_tick(unsigned long); | ||
1221 | static DEFINE_TIMER(psched_timer, psched_tick, 0, 0); | ||
1222 | |||
1223 | static void psched_tick(unsigned long dummy) | ||
1224 | { | ||
1225 | if (sizeof(cycles_t) == sizeof(u32)) { | ||
1226 | psched_time_t dummy_stamp; | ||
1227 | PSCHED_GET_TIME(dummy_stamp); | ||
1228 | psched_timer.expires = jiffies + 1*HZ; | ||
1229 | add_timer(&psched_timer); | ||
1230 | } | ||
1231 | } | ||
1232 | |||
1233 | int __init psched_calibrate_clock(void) | ||
1234 | { | ||
1235 | psched_time_t stamp, stamp1; | ||
1236 | struct timeval tv, tv1; | ||
1237 | psched_tdiff_t delay; | ||
1238 | long rdelay; | ||
1239 | unsigned long stop; | ||
1240 | |||
1241 | psched_tick(0); | ||
1242 | stop = jiffies + HZ/10; | ||
1243 | PSCHED_GET_TIME(stamp); | ||
1244 | do_gettimeofday(&tv); | ||
1245 | while (time_before(jiffies, stop)) { | ||
1246 | barrier(); | ||
1247 | cpu_relax(); | ||
1248 | } | ||
1249 | PSCHED_GET_TIME(stamp1); | ||
1250 | do_gettimeofday(&tv1); | ||
1251 | |||
1252 | delay = PSCHED_TDIFF(stamp1, stamp); | ||
1253 | rdelay = tv1.tv_usec - tv.tv_usec; | ||
1254 | rdelay += (tv1.tv_sec - tv.tv_sec)*1000000; | ||
1255 | if (rdelay > delay) | ||
1256 | return -1; | ||
1257 | delay /= rdelay; | ||
1258 | psched_tick_per_us = delay; | ||
1259 | while ((delay>>=1) != 0) | ||
1260 | psched_clock_scale++; | ||
1261 | psched_us_per_tick = 1<<psched_clock_scale; | ||
1262 | psched_clock_per_hz = (psched_tick_per_us*(1000000/HZ))>>psched_clock_scale; | ||
1263 | return 0; | ||
1264 | } | ||
1265 | #endif | ||
1266 | |||
1267 | static int __init pktsched_init(void) | 1202 | static int __init pktsched_init(void) |
1268 | { | 1203 | { |
1269 | struct rtnetlink_link *link_p; | 1204 | struct rtnetlink_link *link_p; |
1270 | 1205 | ||
1271 | #ifdef CONFIG_NET_SCH_CLK_CPU | ||
1272 | if (psched_calibrate_clock() < 0) | ||
1273 | return -1; | ||
1274 | #elif defined(CONFIG_NET_SCH_CLK_JIFFIES) | ||
1275 | psched_tick_per_us = HZ<<PSCHED_JSCALE; | ||
1276 | psched_us_per_tick = 1000000; | ||
1277 | #endif | ||
1278 | |||
1279 | link_p = rtnetlink_links[PF_UNSPEC]; | 1206 | link_p = rtnetlink_links[PF_UNSPEC]; |
1280 | 1207 | ||
1281 | /* Setup rtnetlink links. It is made here to avoid | 1208 | /* Setup rtnetlink links. It is made here to avoid |
diff --git a/net/sched/sch_hfsc.c b/net/sched/sch_hfsc.c index 407c6fb1ba14..f85cfba647f8 100644 --- a/net/sched/sch_hfsc.c +++ b/net/sched/sch_hfsc.c | |||
@@ -195,20 +195,6 @@ struct hfsc_sched | |||
195 | struct timer_list wd_timer; /* watchdog timer */ | 195 | struct timer_list wd_timer; /* watchdog timer */ |
196 | }; | 196 | }; |
197 | 197 | ||
198 | /* | ||
199 | * macros | ||
200 | */ | ||
201 | #ifdef CONFIG_NET_SCH_CLK_GETTIMEOFDAY | ||
202 | #include <linux/time.h> | ||
203 | #undef PSCHED_GET_TIME | ||
204 | #define PSCHED_GET_TIME(stamp) \ | ||
205 | do { \ | ||
206 | struct timeval tv; \ | ||
207 | do_gettimeofday(&tv); \ | ||
208 | (stamp) = 1ULL * USEC_PER_SEC * tv.tv_sec + tv.tv_usec; \ | ||
209 | } while (0) | ||
210 | #endif | ||
211 | |||
212 | #define HT_INFINITY 0xffffffffffffffffULL /* infinite time value */ | 198 | #define HT_INFINITY 0xffffffffffffffffULL /* infinite time value */ |
213 | 199 | ||
214 | 200 | ||
@@ -394,28 +380,17 @@ cftree_update(struct hfsc_class *cl) | |||
394 | * ism: (psched_us/byte) << ISM_SHIFT | 380 | * ism: (psched_us/byte) << ISM_SHIFT |
395 | * dx: psched_us | 381 | * dx: psched_us |
396 | * | 382 | * |
397 | * Clock source resolution (CONFIG_NET_SCH_CLK_*) | 383 | * The clock source resolution with ktime is 1.024us. |
398 | * JIFFIES: for 48<=HZ<=1534 resolution is between 0.63us and 1.27us. | ||
399 | * CPU: resolution is between 0.5us and 1us. | ||
400 | * GETTIMEOFDAY: resolution is exactly 1us. | ||
401 | * | 384 | * |
402 | * sm and ism are scaled in order to keep effective digits. | 385 | * sm and ism are scaled in order to keep effective digits. |
403 | * SM_SHIFT and ISM_SHIFT are selected to keep at least 4 effective | 386 | * SM_SHIFT and ISM_SHIFT are selected to keep at least 4 effective |
404 | * digits in decimal using the following table. | 387 | * digits in decimal using the following table. |
405 | * | 388 | * |
406 | * Note: We can afford the additional accuracy (altq hfsc keeps at most | ||
407 | * 3 effective digits) thanks to the fact that linux clock is bounded | ||
408 | * much more tightly. | ||
409 | * | ||
410 | * bits/sec 100Kbps 1Mbps 10Mbps 100Mbps 1Gbps | 389 | * bits/sec 100Kbps 1Mbps 10Mbps 100Mbps 1Gbps |
411 | * ------------+------------------------------------------------------- | 390 | * ------------+------------------------------------------------------- |
412 | * bytes/0.5us 6.25e-3 62.5e-3 625e-3 6250e-e 62500e-3 | 391 | * bytes/1.024us 12.8e-3 128e-3 1280e-3 12800e-3 128000e-3 |
413 | * bytes/us 12.5e-3 125e-3 1250e-3 12500e-3 125000e-3 | ||
414 | * bytes/1.27us 15.875e-3 158.75e-3 1587.5e-3 15875e-3 158750e-3 | ||
415 | * | 392 | * |
416 | * 0.5us/byte 160 16 1.6 0.16 0.016 | 393 | * 1.024us/byte 78.125 7.8125 0.78125 0.078125 0.0078125 |
417 | * us/byte 80 8 0.8 0.08 0.008 | ||
418 | * 1.27us/byte 63 6.3 0.63 0.063 0.0063 | ||
419 | */ | 394 | */ |
420 | #define SM_SHIFT 20 | 395 | #define SM_SHIFT 20 |
421 | #define ISM_SHIFT 18 | 396 | #define ISM_SHIFT 18 |