Mercurial > hg > nginx-quic
annotate src/core/ngx_cpuinfo.c @ 9067:f68fdb017141 quic
QUIC: optimized sending stream response.
When a stream is created by client, it's often the case that nginx will send
immediate response on that stream. An example is HTTP/3 request stream, which
in most cases quickly replies with at least HTTP headers.
QUIC stream init handlers are called from a posted event. Output QUIC
frames are also sent to client from a posted event, called the push event.
If the push event is posted before the stream init event, then output produced
by stream may trigger sending an extra UDP datagram. To address this, push
event is now re-posted when a new stream init event is posted.
An example is handling 0-RTT packets. Client typically sends an init packet
coalesced with a 0-RTT packet. Previously, nginx replied with a padded CRYPTO
datagram, followed by a 1-RTT stream reply datagram. Now CRYPTO and STREAM
packets are coalesced in one reply datagram, which saves bandwidth.
Other examples include coalescing 1-RTT first stream response, and
MAX_STREAMS/STREAM sent in response to ACK/STREAM.
author | Roman Arutyunyan <arut@nginx.com> |
---|---|
date | Mon, 03 Apr 2023 16:17:12 +0400 |
parents | d620f497c50f |
children |
rev | line source |
---|---|
611 | 1 |
2 /* | |
3 * Copyright (C) Igor Sysoev | |
4412 | 4 * Copyright (C) Nginx, Inc. |
611 | 5 */ |
6 | |
7 | |
8 #include <ngx_config.h> | |
9 #include <ngx_core.h> | |
10 | |
11 | |
12 #if (( __i386__ || __amd64__ ) && ( __GNUC__ || __INTEL_COMPILER )) | |
13 | |
14 | |
15 static ngx_inline void ngx_cpuid(uint32_t i, uint32_t *buf); | |
16 | |
17 | |
617 | 18 #if ( __i386__ ) |
19 | |
20 static ngx_inline void | |
21 ngx_cpuid(uint32_t i, uint32_t *buf) | |
22 { | |
23 | |
24 /* | |
25 * we could not use %ebx as output parameter if gcc builds PIC, | |
26 * and we could not save %ebx on stack, because %esp is used, | |
27 * when the -fomit-frame-pointer optimization is specified. | |
28 */ | |
29 | |
30 __asm__ ( | |
31 | |
32 " mov %%ebx, %%esi; " | |
33 | |
34 " cpuid; " | |
651 | 35 " mov %%eax, (%1); " |
36 " mov %%ebx, 4(%1); " | |
37 " mov %%edx, 8(%1); " | |
38 " mov %%ecx, 12(%1); " | |
617 | 39 |
40 " mov %%esi, %%ebx; " | |
41 | |
651 | 42 : : "a" (i), "D" (buf) : "ecx", "edx", "esi", "memory" ); |
617 | 43 } |
44 | |
45 | |
46 #else /* __amd64__ */ | |
47 | |
48 | |
611 | 49 static ngx_inline void |
50 ngx_cpuid(uint32_t i, uint32_t *buf) | |
51 { | |
52 uint32_t eax, ebx, ecx, edx; | |
53 | |
54 __asm__ ( | |
55 | |
56 "cpuid" | |
57 | |
58 : "=a" (eax), "=b" (ebx), "=c" (ecx), "=d" (edx) : "a" (i) ); | |
59 | |
60 buf[0] = eax; | |
61 buf[1] = ebx; | |
62 buf[2] = edx; | |
63 buf[3] = ecx; | |
64 } | |
65 | |
66 | |
617 | 67 #endif |
68 | |
69 | |
611 | 70 /* auto detect the L2 cache line size of modern and widespread CPUs */ |
71 | |
72 void | |
73 ngx_cpuinfo(void) | |
74 { | |
75 u_char *vendor; | |
2613
b863be280d3b
update cpuid for Core 2 and Atom
Igor Sysoev <igor@sysoev.ru>
parents:
1872
diff
changeset
|
76 uint32_t vbuf[5], cpu[4], model; |
611 | 77 |
78 vbuf[0] = 0; | |
79 vbuf[1] = 0; | |
80 vbuf[2] = 0; | |
81 vbuf[3] = 0; | |
82 vbuf[4] = 0; | |
83 | |
84 ngx_cpuid(0, vbuf); | |
85 | |
86 vendor = (u_char *) &vbuf[1]; | |
87 | |
88 if (vbuf[0] == 0) { | |
89 return; | |
90 } | |
91 | |
92 ngx_cpuid(1, cpu); | |
93 | |
94 if (ngx_strcmp(vendor, "GenuineIntel") == 0) { | |
95 | |
1871
9afb84c7cc61
fix cache line size for Pentium 4
Igor Sysoev <igor@sysoev.ru>
parents:
651
diff
changeset
|
96 switch ((cpu[0] & 0xf00) >> 8) { |
611 | 97 |
98 /* Pentium */ | |
99 case 5: | |
1872
8e88522cb6da
detect L2 cache line size for Intel Core
Igor Sysoev <igor@sysoev.ru>
parents:
1871
diff
changeset
|
100 ngx_cacheline_size = 32; |
8e88522cb6da
detect L2 cache line size for Intel Core
Igor Sysoev <igor@sysoev.ru>
parents:
1871
diff
changeset
|
101 break; |
8e88522cb6da
detect L2 cache line size for Intel Core
Igor Sysoev <igor@sysoev.ru>
parents:
1871
diff
changeset
|
102 |
611 | 103 /* Pentium Pro, II, III */ |
104 case 6: | |
105 ngx_cacheline_size = 32; | |
1872
8e88522cb6da
detect L2 cache line size for Intel Core
Igor Sysoev <igor@sysoev.ru>
parents:
1871
diff
changeset
|
106 |
2613
b863be280d3b
update cpuid for Core 2 and Atom
Igor Sysoev <igor@sysoev.ru>
parents:
1872
diff
changeset
|
107 model = ((cpu[0] & 0xf0000) >> 8) | (cpu[0] & 0xf0); |
b863be280d3b
update cpuid for Core 2 and Atom
Igor Sysoev <igor@sysoev.ru>
parents:
1872
diff
changeset
|
108 |
b863be280d3b
update cpuid for Core 2 and Atom
Igor Sysoev <igor@sysoev.ru>
parents:
1872
diff
changeset
|
109 if (model >= 0xd0) { |
b863be280d3b
update cpuid for Core 2 and Atom
Igor Sysoev <igor@sysoev.ru>
parents:
1872
diff
changeset
|
110 /* Intel Core, Core 2, Atom */ |
1872
8e88522cb6da
detect L2 cache line size for Intel Core
Igor Sysoev <igor@sysoev.ru>
parents:
1871
diff
changeset
|
111 ngx_cacheline_size = 64; |
8e88522cb6da
detect L2 cache line size for Intel Core
Igor Sysoev <igor@sysoev.ru>
parents:
1871
diff
changeset
|
112 } |
8e88522cb6da
detect L2 cache line size for Intel Core
Igor Sysoev <igor@sysoev.ru>
parents:
1871
diff
changeset
|
113 |
611 | 114 break; |
115 | |
116 /* | |
117 * Pentium 4, although its cache line size is 64 bytes, | |
118 * it prefetches up to two cache lines during memory read | |
119 */ | |
120 case 15: | |
121 ngx_cacheline_size = 128; | |
122 break; | |
123 } | |
124 | |
125 } else if (ngx_strcmp(vendor, "AuthenticAMD") == 0) { | |
126 ngx_cacheline_size = 64; | |
127 } | |
128 } | |
129 | |
130 #else | |
131 | |
132 | |
133 void | |
134 ngx_cpuinfo(void) | |
135 { | |
136 } | |
137 | |
138 | |
139 #endif |