Line data Source code
1 : #include "../../../../disco/tiles.h"
2 :
3 : #include <sys/types.h> /* SEEK_SET */
4 : #include <unistd.h>
5 : #include "generated/cswtch_seccomp.h"
6 :
7 : #include "../../../../disco/metrics/fd_metrics.h"
8 :
9 : #include <fcntl.h>
10 : #include <errno.h>
11 :
12 0 : #define REPORT_INTERVAL_MILLIS (100L)
13 :
14 : typedef struct {
15 : long next_report_nanos;
16 :
17 : ulong tile_cnt;
18 : int status_fds[ FD_TILE_MAX ];
19 : volatile ulong * metrics[ FD_TILE_MAX ];
20 : } fd_cswtch_ctx_t;
21 :
22 : FD_FN_CONST static inline ulong
23 3 : scratch_align( void ) {
24 3 : return 128UL;
25 3 : }
26 :
27 : FD_FN_PURE static inline ulong
28 3 : scratch_footprint( fd_topo_tile_t const * tile ) {
29 3 : (void)tile;
30 3 : ulong l = FD_LAYOUT_INIT;
31 3 : l = FD_LAYOUT_APPEND( l, alignof( fd_cswtch_ctx_t ), sizeof( fd_cswtch_ctx_t ) );
32 3 : return FD_LAYOUT_FINI( l, scratch_align() );
33 3 : }
34 :
35 : static void
36 : before_credit( fd_cswtch_ctx_t * ctx,
37 : fd_stem_context_t * mux,
38 0 : int * charge_busy ) {
39 0 : (void)mux;
40 :
41 0 : long now = fd_log_wallclock();
42 0 : if( FD_UNLIKELY( now<ctx->next_report_nanos ) ) return;
43 0 : ctx->next_report_nanos += REPORT_INTERVAL_MILLIS*1000L*1000L;
44 :
45 0 : *charge_busy = 1;
46 :
47 0 : for( ulong i=0UL; i<ctx->tile_cnt; i++ ) {
48 0 : if( FD_UNLIKELY( -1==lseek( ctx->status_fds[ i ], 0, SEEK_SET ) ) ) FD_LOG_ERR(( "lseek failed (%i-%s)", errno, strerror( errno ) ));
49 :
50 0 : char contents[ 4096 ] = {0};
51 0 : ulong contents_len = 0UL;
52 :
53 0 : int process_died = 0;
54 0 : while( 1 ) {
55 0 : if( FD_UNLIKELY( contents_len>=sizeof( contents ) ) ) FD_LOG_ERR(( "contents overflow" ));
56 0 : long n = read( ctx->status_fds[ i ], contents + contents_len, sizeof( contents ) - contents_len );
57 0 : if( FD_UNLIKELY( -1==n ) ) {
58 0 : if( FD_UNLIKELY( errno==ESRCH ) ) {
59 0 : process_died = 1;
60 0 : break;
61 0 : }
62 0 : FD_LOG_ERR(( "read failed (%i-%s)", errno, strerror( errno ) ));
63 0 : }
64 0 : if( FD_LIKELY( 0==n ) ) break;
65 0 : contents_len += (ulong)n;
66 0 : }
67 :
68 : /* Supervisor is going to bring the whole process tree down if any
69 : of the target PIDs died, so we can ignore this and wait. */
70 0 : if( FD_UNLIKELY( process_died ) ) {
71 0 : FD_LOG_WARNING(( "cannot get context switch metrics for dead tile idx %lu", i ));
72 0 : continue;
73 0 : }
74 :
75 0 : int found_voluntary = 0;
76 0 : int found_involuntary = 0;
77 :
78 0 : char * line = contents;
79 0 : while( 1 ) {
80 0 : char * next_line = strchr( line, '\n' );
81 0 : if( FD_UNLIKELY( NULL==next_line ) ) break;
82 0 : *next_line = '\0';
83 :
84 0 : char * colon = strchr( line, ':' );
85 0 : if( FD_UNLIKELY( NULL==colon ) ) FD_LOG_ERR(( "no colon in line '%s'", line ));
86 :
87 0 : *colon = '\0';
88 0 : char * key = line;
89 0 : char * value = colon + 1;
90 :
91 0 : while( ' '==*value || '\t'==*value ) value++;
92 :
93 0 : if( FD_LIKELY( !strncmp( key, "voluntary_ctxt_switches", 23UL ) ) ) {
94 0 : char * endptr;
95 0 : ulong voluntary_ctxt_switches = strtoul( value, &endptr, 10 );
96 0 : if( FD_UNLIKELY( *endptr!='\0' || voluntary_ctxt_switches==ULONG_MAX ) ) FD_LOG_ERR(( "strtoul failed" ));
97 0 : ctx->metrics[ i ][ FD_METRICS_COUNTER_TILE_CONTEXT_SWITCH_VOLUNTARY_COUNT_OFF ] = voluntary_ctxt_switches;
98 0 : found_voluntary = 1;
99 0 : } else if( FD_LIKELY( !strncmp( key, "nonvoluntary_ctxt_switches", 26UL ) ) ) {
100 0 : char * endptr;
101 0 : ulong involuntary_ctxt_switches = strtoul( value, &endptr, 10 );
102 0 : if( FD_UNLIKELY( *endptr!='\0' || involuntary_ctxt_switches==ULONG_MAX ) ) FD_LOG_ERR(( "strtoul failed" ));
103 0 : ctx->metrics[ i ][ FD_METRICS_COUNTER_TILE_CONTEXT_SWITCH_INVOLUNTARY_COUNT_OFF ] = involuntary_ctxt_switches;
104 0 : found_involuntary = 1;
105 0 : }
106 :
107 0 : line = next_line + 1;
108 0 : }
109 :
110 0 : if( FD_UNLIKELY( !found_voluntary ) ) FD_LOG_ERR(( "voluntary_ctxt_switches not found" ));
111 0 : if( FD_UNLIKELY( !found_involuntary ) ) FD_LOG_ERR(( "nonvoluntary_ctxt_switches not found" ));
112 0 : }
113 0 : }
114 :
115 : static void
116 : privileged_init( fd_topo_t * topo,
117 0 : fd_topo_tile_t * tile ) {
118 0 : void * scratch = fd_topo_obj_laddr( topo, tile->tile_obj_id );
119 :
120 0 : FD_SCRATCH_ALLOC_INIT( l, scratch );
121 0 : fd_cswtch_ctx_t * ctx = FD_SCRATCH_ALLOC_APPEND( l, alignof( fd_cswtch_ctx_t ), sizeof( fd_cswtch_ctx_t ) );
122 :
123 0 : FD_TEST( topo->tile_cnt<FD_TILE_MAX );
124 :
125 0 : ctx->tile_cnt = topo->tile_cnt;
126 0 : for( ulong i=0UL; i<topo->tile_cnt; i++ ) {
127 0 : ulong * metrics = fd_metrics_join( fd_topo_obj_laddr( topo, topo->tiles[ i ].metrics_obj_id ) );
128 :
129 0 : for(;;) {
130 0 : ulong pid, tid;
131 0 : if( FD_UNLIKELY( tile->id==i ) ) {
132 0 : pid = fd_sandbox_getpid();
133 0 : tid = fd_sandbox_gettid();
134 0 : } else {
135 0 : pid = fd_metrics_tile( metrics )[ FD_METRICS_GAUGE_TILE_PID_OFF ];
136 0 : tid = fd_metrics_tile( metrics )[ FD_METRICS_GAUGE_TILE_TID_OFF ];
137 0 : if( FD_UNLIKELY( !pid || !tid ) ) {
138 0 : FD_SPIN_PAUSE();
139 0 : continue;
140 0 : }
141 0 : }
142 :
143 0 : char path[ 64 ];
144 0 : FD_TEST( fd_cstr_printf_check( path, sizeof( path ), NULL, "/proc/%lu/task/%lu/status", pid, tid ) );
145 0 : ctx->status_fds[ i ] = open( path, O_RDONLY );
146 0 : ctx->metrics[ i ] = fd_metrics_tile( metrics );
147 0 : if( FD_UNLIKELY( -1==ctx->status_fds[ i ] ) ) FD_LOG_ERR(( "open failed (%i-%s)", errno, strerror( errno ) ));
148 0 : break;
149 0 : }
150 0 : }
151 0 : }
152 :
153 : static void
154 : unprivileged_init( fd_topo_t * topo,
155 0 : fd_topo_tile_t * tile ) {
156 0 : void * scratch = fd_topo_obj_laddr( topo, tile->tile_obj_id );
157 :
158 0 : FD_SCRATCH_ALLOC_INIT( l, scratch );
159 0 : fd_cswtch_ctx_t * ctx = FD_SCRATCH_ALLOC_APPEND( l, alignof( fd_cswtch_ctx_t ), sizeof( fd_cswtch_ctx_t ) );
160 :
161 0 : ctx->next_report_nanos = fd_log_wallclock();
162 :
163 0 : ulong scratch_top = FD_SCRATCH_ALLOC_FINI( l, 1UL );
164 0 : if( FD_UNLIKELY( scratch_top > (ulong)scratch + scratch_footprint( tile ) ) )
165 0 : FD_LOG_ERR(( "scratch overflow %lu %lu %lu", scratch_top - (ulong)scratch - scratch_footprint( tile ), scratch_top, (ulong)scratch + scratch_footprint( tile ) ));
166 0 : }
167 :
168 : static ulong
169 : populate_allowed_seccomp( fd_topo_t const * topo,
170 : fd_topo_tile_t const * tile,
171 : ulong out_cnt,
172 0 : struct sock_filter * out ) {
173 0 : (void)topo;
174 0 : (void)tile;
175 :
176 0 : populate_sock_filter_policy_cswtch( out_cnt, out, (uint)fd_log_private_logfile_fd() );
177 0 : return sock_filter_policy_cswtch_instr_cnt;
178 0 : }
179 :
180 : static ulong
181 : populate_allowed_fds( fd_topo_t const * topo,
182 : fd_topo_tile_t const * tile,
183 : ulong out_fds_cnt,
184 0 : int * out_fds ) {
185 0 : void * scratch = fd_topo_obj_laddr( topo, tile->tile_obj_id );
186 :
187 0 : FD_SCRATCH_ALLOC_INIT( l, scratch );
188 0 : fd_cswtch_ctx_t * ctx = FD_SCRATCH_ALLOC_APPEND( l, alignof( fd_cswtch_ctx_t ), sizeof( fd_cswtch_ctx_t ) );
189 :
190 0 : if( FD_UNLIKELY( out_fds_cnt<2UL+ctx->tile_cnt ) ) FD_LOG_ERR(( "out_fds_cnt %lu", out_fds_cnt ));
191 :
192 0 : ulong out_cnt = 0UL;
193 0 : out_fds[ out_cnt++ ] = 2; /* stderr */
194 0 : if( FD_LIKELY( -1!=fd_log_private_logfile_fd() ) )
195 0 : out_fds[ out_cnt++ ] = fd_log_private_logfile_fd(); /* logfile */
196 0 : for( ulong i=0UL; i<ctx->tile_cnt; i++ )
197 0 : out_fds[ out_cnt++ ] = ctx->status_fds[ i ]; /* /proc/<pid>/task/<tid>/status descriptor */
198 0 : return out_cnt;
199 0 : }
200 :
201 0 : #define STEM_BURST (1UL)
202 :
203 : /* See explanation in fd_pack */
204 0 : #define STEM_LAZY (128L*3000L)
205 :
206 0 : #define STEM_CALLBACK_CONTEXT_TYPE fd_cswtch_ctx_t
207 0 : #define STEM_CALLBACK_CONTEXT_ALIGN alignof(fd_cswtch_ctx_t)
208 :
209 0 : #define STEM_CALLBACK_BEFORE_CREDIT before_credit
210 :
211 : #include "../../../../disco/stem/fd_stem.c"
212 :
213 : fd_topo_run_tile_t fd_tile_cswtch = {
214 : .name = "cswtch",
215 : .populate_allowed_seccomp = populate_allowed_seccomp,
216 : .populate_allowed_fds = populate_allowed_fds,
217 : .scratch_align = scratch_align,
218 : .scratch_footprint = scratch_footprint,
219 : .privileged_init = privileged_init,
220 : .unprivileged_init = unprivileged_init,
221 : .run = stem_run,
222 : };
|
