Line data Source code
1 : #include "fd_topo.h"
2 :
3 : #include "../metrics/fd_metrics.h"
4 : #include "../../util/pod/fd_pod_format.h"
5 : #include "../../util/wksp/fd_wksp_private.h"
6 : #include "../../util/shmem/fd_shmem_private.h"
7 :
8 : #include <stdio.h>
9 : #include <errno.h>
10 : #include <unistd.h>
11 : #include <sys/stat.h>
12 :
13 : void *
14 : fd_topo_obj_laddr( fd_topo_t const * topo,
15 24 : ulong obj_id ) {
16 24 : fd_topo_obj_t const * obj = &topo->objs[ obj_id ];
17 24 : if( FD_UNLIKELY( obj_id==ULONG_MAX ) ) FD_LOG_CRIT(( "invalid obj_id ULONG_MAX" ));
18 24 : if( FD_UNLIKELY( obj_id>=FD_TOPO_MAX_OBJS ) ) FD_LOG_CRIT(( "invalid obj_id %lu", obj_id ));
19 24 : FD_TEST( obj->id == obj_id );
20 24 : FD_TEST( obj->offset );
21 24 : return (void *)((ulong)topo->workspaces[ obj->wksp_id ].wksp + obj->offset);
22 24 : }
23 :
24 : void
25 : fd_topo_join_workspace( fd_topo_t * topo,
26 : fd_topo_wksp_t * wksp,
27 : int mode,
28 0 : int dump ) {
29 0 : char name[ PATH_MAX ];
30 0 : FD_TEST( fd_cstr_printf_check( name, PATH_MAX, NULL, "%s_%s.wksp", topo->app_name, wksp->name ) );
31 :
32 0 : wksp->wksp = fd_wksp_join( fd_shmem_join( name, mode, dump, NULL, NULL, NULL ) );
33 0 : if( FD_UNLIKELY( !wksp->wksp ) ) FD_LOG_ERR(( "fd_wksp_join failed" ));
34 0 : }
35 :
36 : FD_FN_PURE static int
37 0 : tile_needs_wksp( fd_topo_t const * topo, fd_topo_tile_t const * tile, ulong wksp_id ) {
38 0 : int mode = -1;
39 0 : for( ulong i=0UL; i<tile->uses_obj_cnt; i++ ) {
40 0 : if( FD_UNLIKELY( topo->objs[ tile->uses_obj_id[ i ] ].wksp_id==wksp_id ) ) {
41 0 : mode = fd_int_max( mode, tile->uses_obj_mode[ i ] );
42 0 : }
43 0 : }
44 0 : return mode;
45 0 : }
46 :
47 : void
48 : fd_topo_join_tile_workspaces( fd_topo_t * topo,
49 : fd_topo_tile_t * tile,
50 0 : int core_dump_level ) {
51 0 : for( ulong i=0UL; i<topo->wksp_cnt; i++ ) {
52 0 : int needs_wksp = tile_needs_wksp( topo, tile, i );
53 0 : if( FD_LIKELY( -1!=needs_wksp ) ) {
54 0 : int dump = core_dump_level >= topo->workspaces[ i ].core_dump_level ? 1 : 0;
55 0 : fd_topo_join_workspace( topo, &topo->workspaces[ i ], needs_wksp, dump );
56 0 : }
57 0 : }
58 0 : }
59 :
60 : void
61 : fd_topo_join_workspaces( fd_topo_t * topo,
62 : int mode,
63 0 : int core_dump_level ) {
64 0 : for( ulong i=0UL; i<topo->wksp_cnt; i++ ) {
65 0 : int dump = core_dump_level >= topo->workspaces[ i ].core_dump_level ? 1 : 0;
66 0 : fd_topo_join_workspace( topo, &topo->workspaces[ i ], mode, dump );
67 0 : }
68 0 : }
69 :
70 : void
71 : fd_topo_leave_workspace( fd_topo_t * topo FD_PARAM_UNUSED,
72 0 : fd_topo_wksp_t * wksp ) {
73 0 : if( FD_LIKELY( wksp->wksp ) ) {
74 0 : if( FD_UNLIKELY( fd_wksp_detach( wksp->wksp ) ) ) FD_LOG_ERR(( "fd_wksp_detach failed" ));
75 0 : wksp->wksp = NULL;
76 0 : wksp->known_footprint = 0UL;
77 0 : wksp->total_footprint = 0UL;
78 0 : }
79 0 : }
80 :
81 : void
82 0 : fd_topo_leave_workspaces( fd_topo_t * topo ) {
83 0 : for( ulong i=0UL; i<topo->wksp_cnt; i++ ) {
84 0 : fd_topo_leave_workspace( topo, &topo->workspaces[ i ] );
85 0 : }
86 0 : }
87 :
88 : extern char fd_shmem_private_base[ FD_SHMEM_PRIVATE_BASE_MAX ];
89 :
90 : int
91 : fd_topo_create_workspace( fd_topo_t * topo,
92 : fd_topo_wksp_t * wksp,
93 0 : int update_existing ) {
94 0 : char name[ PATH_MAX ];
95 0 : FD_TEST( fd_cstr_printf_check( name, PATH_MAX, NULL, "%s_%s.wksp", topo->app_name, wksp->name ) );
96 :
97 0 : ulong sub_page_cnt[ 1 ] = { wksp->page_cnt };
98 0 : ulong sub_cpu_idx [ 1 ] = { fd_shmem_cpu_idx( wksp->numa_idx ) };
99 :
100 0 : int err;
101 0 : if( FD_UNLIKELY( update_existing ) ) {
102 0 : err = fd_shmem_update_multi( name, wksp->page_sz, 1, sub_page_cnt, sub_cpu_idx, S_IRUSR | S_IWUSR ); /* logs details */
103 0 : } else {
104 0 : err = fd_shmem_create_multi( name, wksp->page_sz, 1, sub_page_cnt, sub_cpu_idx, S_IRUSR | S_IWUSR ); /* logs details */
105 0 : }
106 0 : if( FD_UNLIKELY( err && errno==ENOMEM ) ) return -1;
107 0 : else if( FD_UNLIKELY( err ) ) FD_LOG_ERR(( "fd_shmem_create_multi failed" ));
108 :
109 0 : void * shmem = fd_shmem_join( name, FD_SHMEM_JOIN_MODE_READ_WRITE, 0, NULL, NULL, NULL ); /* logs details */
110 :
111 0 : void * wkspmem = fd_wksp_new( shmem, name, 0U, wksp->part_max, wksp->total_footprint ); /* logs details */
112 0 : if( FD_UNLIKELY( !wkspmem ) ) FD_LOG_ERR(( "fd_wksp_new failed" ));
113 :
114 0 : fd_wksp_t * join = fd_wksp_join( wkspmem );
115 0 : if( FD_UNLIKELY( !join ) ) FD_LOG_ERR(( "fd_wksp_join failed" ));
116 :
117 : /* Footprint has been predetermined so that this alloc() call must
118 : succeed inside the data region. The difference between total_footprint
119 : and known_footprint is given to "loose" data, that may be dynamically
120 : allocated out of the workspace at runtime. */
121 0 : if( FD_LIKELY( wksp->known_footprint ) ) {
122 0 : ulong offset = fd_wksp_alloc( join, fd_topo_workspace_align(), wksp->known_footprint, 1UL );
123 0 : if( FD_UNLIKELY( !offset ) ) FD_LOG_ERR(( "fd_wksp_alloc failed" ));
124 :
125 : /* gaddr_lo is the start of the workspace data region that can be
126 : given out in response to wksp alloc requests. We rely on an
127 : implicit assumption everywhere that the bytes we are given by
128 : this single allocation will be at gaddr_lo, so that we can find
129 : them, so we verify this here for paranoia in case the workspace
130 : alloc implementation changes. */
131 0 : if( FD_UNLIKELY( fd_ulong_align_up( ((struct fd_wksp_private*)join)->gaddr_lo, fd_topo_workspace_align() ) != offset ) )
132 0 : FD_LOG_ERR(( "wksp gaddr_lo %lu != offset %lu", fd_ulong_align_up( ((struct fd_wksp_private*)join)->gaddr_lo, fd_topo_workspace_align() ), offset ));
133 0 : }
134 :
135 0 : fd_wksp_leave( join );
136 :
137 0 : if( FD_UNLIKELY( fd_shmem_leave( shmem, NULL, NULL ) ) ) /* logs details */
138 0 : FD_LOG_ERR(( "fd_shmem_leave failed" ));
139 :
140 0 : return 0;
141 0 : }
142 :
143 : void
144 : fd_topo_wksp_new( fd_topo_t const * topo,
145 : fd_topo_wksp_t const * wksp,
146 0 : fd_topo_obj_callbacks_t ** callbacks ) {
147 0 : for( ulong i=0UL; i<topo->obj_cnt; i++ ) {
148 0 : fd_topo_obj_t const * obj = &topo->objs[ i ];
149 0 : if( FD_LIKELY( obj->wksp_id!=wksp->id ) ) continue;
150 :
151 0 : for( ulong j=0UL; callbacks[ j ]; j++ ) {
152 0 : if( FD_LIKELY( strcmp( callbacks[ j ]->name, obj->name ) ) ) continue;
153 :
154 0 : long ts = -fd_log_wallclock();
155 0 : if( FD_LIKELY( callbacks[ j ]->new ) ) callbacks[ j ]->new( topo, obj );
156 0 : long elapsed = fd_log_wallclock() + ts;
157 0 : if( FD_UNLIKELY( elapsed>(1000L*1000L*150L ) ) ) FD_LOG_WARNING(( "fd_topo_wksp_new(%s) took %ld ms", obj->name, elapsed/(1000L*1000L) ));
158 0 : else if( FD_UNLIKELY( elapsed>(1000L*1000L*5L ) ) ) FD_LOG_INFO(( "fd_topo_wksp_new(%s) took %ld ms", obj->name, elapsed/(1000L*1000L) ));
159 0 : break;
160 0 : }
161 0 : }
162 0 : }
163 :
164 : void
165 : fd_topo_workspace_fill( fd_topo_t * topo,
166 0 : fd_topo_wksp_t * wksp ) {
167 0 : for( ulong i=0UL; i<topo->link_cnt; i++ ) {
168 0 : fd_topo_link_t * link = &topo->links[ i ];
169 :
170 0 : if( FD_UNLIKELY( topo->objs[ link->mcache_obj_id ].wksp_id!=wksp->id ) ) continue;
171 0 : link->mcache = fd_mcache_join( fd_topo_obj_laddr( topo, link->mcache_obj_id ) );
172 0 : FD_TEST( link->mcache );
173 :
174 0 : if( link->mtu ) {
175 0 : if( FD_UNLIKELY( topo->objs[ link->dcache_obj_id ].wksp_id!=wksp->id ) ) continue;
176 0 : link->dcache = fd_dcache_join( fd_topo_obj_laddr( topo, link->dcache_obj_id ) );
177 0 : FD_TEST( link->dcache );
178 0 : }
179 0 : }
180 :
181 0 : for( ulong i=0UL; i<topo->tile_cnt; i++ ) {
182 0 : fd_topo_tile_t * tile = &topo->tiles[ i ];
183 :
184 0 : if( FD_LIKELY( topo->objs[ tile->metrics_obj_id ].wksp_id==wksp->id ) ) {
185 0 : tile->metrics = fd_metrics_join( fd_topo_obj_laddr( topo, tile->metrics_obj_id ) );
186 0 : FD_TEST( tile->metrics );
187 0 : }
188 :
189 0 : for( ulong j=0UL; j<tile->in_cnt; j++ ) {
190 0 : if( FD_UNLIKELY( topo->objs[ tile->in_link_fseq_obj_id[ j ] ].wksp_id!=wksp->id ) ) continue;
191 0 : tile->in_link_fseq[ j ] = fd_fseq_join( fd_topo_obj_laddr( topo, tile->in_link_fseq_obj_id[ j ] ) );
192 0 : FD_TEST( tile->in_link_fseq[ j ] );
193 0 : }
194 0 : }
195 0 : }
196 :
197 : void
198 : fd_topo_fill_tile( fd_topo_t * topo,
199 0 : fd_topo_tile_t * tile ) {
200 0 : for( ulong i=0UL; i<topo->wksp_cnt; i++ ) {
201 0 : if( FD_UNLIKELY( -1!=tile_needs_wksp( topo, tile, i ) ) )
202 0 : fd_topo_workspace_fill( topo, &topo->workspaces[ i ] );
203 0 : }
204 0 : }
205 :
206 : void
207 0 : fd_topo_fill( fd_topo_t * topo ) {
208 0 : for( ulong i=0UL; i<topo->wksp_cnt; i++ ) {
209 0 : fd_topo_workspace_fill( topo, &topo->workspaces[ i ] );
210 0 : }
211 0 : }
212 :
213 : FD_FN_CONST static ulong
214 0 : fd_topo_tile_extra_huge_pages( fd_topo_tile_t const * tile ) {
215 : /* Every tile maps an additional set of pages for the stack. */
216 0 : (void)tile;
217 0 : return (FD_TILE_PRIVATE_STACK_SZ/FD_SHMEM_HUGE_PAGE_SZ)+2UL;
218 0 : }
219 :
220 : FD_FN_PURE static ulong
221 0 : fd_topo_tile_extra_normal_pages( fd_topo_tile_t const * tile ) {
222 0 : ulong key_pages = 0UL;
223 0 : if( FD_UNLIKELY( tile->id_keyswitch_obj_id!=ULONG_MAX ) ) {
224 : /* Certain tiles using fd_keyload_load need normal pages to hold
225 : key material. */
226 0 : key_pages += 5UL;
227 0 : }
228 0 : if( FD_UNLIKELY( tile->av_keyswitch_obj_id!=ULONG_MAX ) ) {
229 : /* Certain tiles using fd_keyload_load need normal pages to hold
230 : key material. */
231 0 : key_pages += 5UL;
232 0 : }
233 :
234 0 : if( !strcmp( tile->name, "net" ) ) {
235 : /* net tile uses normal pages to hold xsk rings */
236 :
237 : /* xdp_desc struct is in linux UAPI so its size can be
238 : safely assumed */
239 0 : ulong xdp_desc_sz_bytes = 16UL;
240 0 : ulong xsk_rings_sz_bytes = 0UL;
241 0 : ulong xdp_address_sz_bytes = sizeof(ulong);
242 :
243 : /* rx ring */
244 0 : xsk_rings_sz_bytes += tile->xdp.xdp_rx_queue_size * xdp_desc_sz_bytes;
245 : /* tx ring */
246 0 : xsk_rings_sz_bytes += tile->xdp.xdp_tx_queue_size * xdp_desc_sz_bytes;
247 :
248 : /* completion ring */
249 0 : xsk_rings_sz_bytes += tile->xdp.xdp_tx_queue_size * xdp_address_sz_bytes;
250 : /* free ring */
251 0 : xsk_rings_sz_bytes += tile->xdp.free_ring_depth * xdp_address_sz_bytes;
252 :
253 0 : key_pages += fd_ulong_align_up( xsk_rings_sz_bytes, FD_SHMEM_NORMAL_PAGE_SZ ) / FD_SHMEM_NORMAL_PAGE_SZ;
254 :
255 : /* All 4 rings must store a ring header. This is 320 bytes
256 : per ring as of linux v6.18.3, however could change in
257 : the future so allow up to a full 4KB page per ring to
258 : be safe. */
259 0 : key_pages += 4UL;
260 0 : }
261 :
262 0 : return key_pages;
263 0 : }
264 :
265 : FD_FN_PURE static ulong
266 : fd_topo_mlock_max_tile1( fd_topo_t const * topo,
267 0 : fd_topo_tile_t const * tile ) {
268 0 : ulong tile_mem = 0UL;
269 :
270 0 : for( ulong i=0UL; i<topo->wksp_cnt; i++ ) {
271 0 : if( FD_UNLIKELY( -1!=tile_needs_wksp( topo, tile, i ) ) )
272 0 : tile_mem += topo->workspaces[ i ].page_cnt * topo->workspaces[ i ].page_sz;
273 0 : }
274 :
275 0 : return tile_mem +
276 0 : fd_topo_tile_extra_huge_pages( tile ) * FD_SHMEM_HUGE_PAGE_SZ +
277 0 : fd_topo_tile_extra_normal_pages( tile ) * FD_SHMEM_NORMAL_PAGE_SZ;
278 0 : }
279 :
280 : FD_FN_PURE ulong
281 0 : fd_topo_mlock_max_tile( fd_topo_t const * topo ) {
282 0 : ulong highest_tile_mem = 0UL;
283 0 : for( ulong i=0UL; i<topo->tile_cnt; i++ ) {
284 0 : fd_topo_tile_t const * tile = &topo->tiles[ i ];
285 0 : highest_tile_mem = fd_ulong_max( highest_tile_mem, fd_topo_mlock_max_tile1( topo, tile ) );
286 0 : }
287 :
288 0 : return highest_tile_mem;
289 0 : }
290 :
291 : FD_FN_PURE ulong
292 : fd_topo_gigantic_page_cnt( fd_topo_t const * topo,
293 0 : ulong numa_idx ) {
294 0 : ulong result = 0UL;
295 0 : for( ulong i=0UL; i<topo->wksp_cnt; i++ ) {
296 0 : fd_topo_wksp_t const * wksp = &topo->workspaces[ i ];
297 0 : if( FD_LIKELY( wksp->numa_idx!=numa_idx ) ) continue;
298 :
299 0 : if( FD_LIKELY( wksp->page_sz==FD_SHMEM_GIGANTIC_PAGE_SZ ) ) {
300 0 : result += wksp->page_cnt;
301 0 : }
302 0 : }
303 0 : return result;
304 0 : }
305 :
306 : FD_FN_PURE ulong
307 : fd_topo_huge_page_cnt( fd_topo_t const * topo,
308 : ulong numa_idx,
309 0 : int include_anonymous ) {
310 0 : ulong result = 0UL;
311 0 : for( ulong i=0UL; i<topo->wksp_cnt; i++ ) {
312 0 : fd_topo_wksp_t const * wksp = &topo->workspaces[ i ];
313 0 : if( FD_LIKELY( wksp->numa_idx!=numa_idx ) ) continue;
314 :
315 0 : if( FD_LIKELY( wksp->page_sz==FD_SHMEM_HUGE_PAGE_SZ ) ) {
316 0 : result += wksp->page_cnt;
317 0 : }
318 0 : }
319 :
320 : /* The stack huge pages are also placed in the hugetlbfs. */
321 0 : for( ulong i=0UL; i<topo->tile_cnt; i++ ) {
322 0 : result += fd_topo_tile_extra_huge_pages( &topo->tiles[ i ] );
323 0 : }
324 :
325 : /* No anonymous huge pages in use yet. */
326 0 : (void)include_anonymous;
327 :
328 0 : return result;
329 0 : }
330 :
331 : FD_FN_PURE ulong
332 0 : fd_topo_normal_page_cnt( fd_topo_t const * topo ) {
333 0 : ulong result = 0UL;
334 0 : for( ulong i=0UL; i<topo->tile_cnt; i++ ) {
335 0 : result += fd_topo_tile_extra_normal_pages( &topo->tiles[ i ] );
336 0 : }
337 0 : return result;
338 0 : }
339 :
340 : FD_FN_PURE ulong
341 0 : fd_topo_mlock( fd_topo_t const * topo ) {
342 0 : ulong result = 0UL;
343 0 : for( ulong i=0UL; i<topo->wksp_cnt; i++ ) {
344 0 : result += topo->workspaces[ i ].page_cnt * topo->workspaces[ i ].page_sz;
345 0 : }
346 0 : return result;
347 0 : }
348 :
349 : static void
350 0 : fd_topo_mem_sz_string( ulong sz, char out[static 24] ) {
351 0 : if( FD_LIKELY( sz >= FD_SHMEM_GIGANTIC_PAGE_SZ ) ) {
352 0 : FD_TEST( fd_cstr_printf_check( out, 24, NULL, "%.1f GiB", (double)sz / (double)(1UL << 30) ) );
353 0 : } else if( FD_LIKELY( sz >= 1048576 ) ) {
354 0 : FD_TEST( fd_cstr_printf_check( out, 24, NULL, "%.1f MiB", (double)sz / (double)(1UL << 20) ) );
355 0 : } else if( FD_LIKELY( sz >= 1024 ) ) {
356 0 : FD_TEST( fd_cstr_printf_check( out, 24, NULL, "%.1f KiB", (double)sz / (double)(1UL << 10) ) );
357 0 : } else {
358 0 : FD_TEST( fd_cstr_printf_check( out, 24, NULL, "%lu B", sz ) );
359 0 : }
360 0 : }
361 :
362 : /* Style sizes by magnitude so the big numbers pop */
363 :
364 : static char const *
365 : fd_topo_mem_sz_style( ulong sz,
366 : char const * bold,
367 0 : char const * dim ) {
368 0 : if( sz>=(1UL<<30) ) return bold;
369 0 : if( sz< (1UL<<20) ) return dim;
370 0 : return "";
371 0 : }
372 :
373 : void
374 : fd_topo_print_log( int stdout,
375 0 : fd_topo_t * topo ) {
376 0 : char message[ 32UL*4096UL ] = {0}; /* Same as FD_LOG_BUF_SZ */
377 :
378 0 : char * cur = message;
379 0 : ulong remaining = sizeof(message) - 1; /* Leave one character at the end to ensure NUL terminated */
380 :
381 0 : #define PRINT( ... ) do { \
382 0 : int n = snprintf( cur, remaining, __VA_ARGS__ ); \
383 0 : if( FD_UNLIKELY( n < 0 ) ) FD_LOG_ERR(( "snprintf failed" )); \
384 0 : if( FD_UNLIKELY( (ulong)n >= remaining ) ) { \
385 : /* stdout mode flushes and retries, log mode keeps one message */ \
386 0 : if( FD_UNLIKELY( !stdout ) ) FD_LOG_ERR(( "snprintf overflow" )); \
387 0 : *cur = '\0'; \
388 0 : FD_LOG_STDOUT(( "%s", message )); \
389 0 : cur = message; remaining = sizeof(message) - 1; \
390 0 : n = snprintf( cur, remaining, __VA_ARGS__ ); \
391 0 : if( FD_UNLIKELY( n < 0 ) ) FD_LOG_ERR(( "snprintf failed" )); \
392 0 : if( FD_UNLIKELY( (ulong)n >= remaining ) ) FD_LOG_ERR(( "snprintf overflow" )); \
393 0 : } \
394 0 : remaining -= (ulong)n; \
395 0 : cur += n; \
396 0 : } while( 0 )
397 :
398 0 : int color = stdout && fd_log_colorize() && isatty( STDOUT_FILENO );
399 0 : char const * c_bold = color ? "\033[1m" : "";
400 0 : char const * c_dim = color ? "\033[2m" : "";
401 0 : char const * c_normal = color ? "\033[0m" : "";
402 :
403 0 : char const * sep = "";
404 :
405 0 : #define SECTION( ... ) do { \
406 0 : char _title[ 64 ]; \
407 0 : FD_TEST( fd_cstr_printf_check( _title, sizeof(_title), NULL, __VA_ARGS__ ) ); \
408 0 : PRINT( "%s%sāā%s %s%s%s %s", sep, c_dim, c_normal, c_bold, _title, c_normal, c_dim ); \
409 0 : for( ulong _i=strlen( _title ); _i<76UL; _i++ ) PRINT( "ā" ); \
410 0 : PRINT( "%s\n", c_normal ); \
411 0 : sep = "\n"; \
412 0 : } while( 0 )
413 :
414 0 : SECTION( "Summary" );
415 :
416 : /* The logic to compute number of stack pages is taken from
417 : fd_tile_thread.cxx, in function fd_topo_tile_stack_join, and this
418 : should match that. */
419 0 : ulong stack_pages = topo->tile_cnt * FD_SHMEM_HUGE_PAGE_SZ * ((FD_TILE_PRIVATE_STACK_SZ/FD_SHMEM_HUGE_PAGE_SZ)+2UL);
420 :
421 0 : ulong normal_page_bytes = fd_topo_normal_page_cnt( topo ) * FD_SHMEM_NORMAL_PAGE_SZ;
422 0 : ulong total_bytes = fd_topo_mlock( topo ) + stack_pages + normal_page_bytes;
423 :
424 0 : PRINT(" %s%-23s%s %lu\n", c_dim, "Total Tiles", c_normal, topo->tile_cnt );
425 0 : PRINT(" %s%-23s%s %s%lu GiB + %lu MiB + %lu KiB%s %s(%lu bytes)%s\n",
426 0 : c_dim, "Total Memory Locked", c_normal,
427 0 : c_bold,
428 0 : total_bytes / (1 << 30),
429 0 : (total_bytes % (1 << 30)) / (1 << 20),
430 0 : (total_bytes % (1 << 20)) / (1 << 10),
431 0 : c_normal,
432 0 : c_dim, total_bytes, c_normal );
433 :
434 0 : ulong required_gigantic_pages = 0UL;
435 0 : ulong required_huge_pages = 0UL;
436 :
437 0 : ulong numa_node_cnt = fd_shmem_numa_cnt();
438 0 : for( ulong i=0UL; i<numa_node_cnt; i++ ) {
439 0 : required_gigantic_pages += fd_topo_gigantic_page_cnt( topo, i );
440 0 : required_huge_pages += fd_topo_huge_page_cnt( topo, i, 0 );
441 0 : }
442 0 : PRINT(" %s%-23s%s %lu\n", c_dim, "Required Gigantic Pages", c_normal, required_gigantic_pages );
443 0 : PRINT(" %s%-23s%s %lu\n", c_dim, "Required Huge Pages", c_normal, required_huge_pages );
444 0 : PRINT(" %s%-23s%s %lu\n", c_dim, "Required Normal Pages", c_normal, fd_topo_normal_page_cnt( topo ) );
445 0 : if( FD_UNLIKELY( numa_node_cnt>1UL ) ) {
446 0 : for( ulong i=0UL; i<numa_node_cnt; i++ ) {
447 0 : char label[ 40 ];
448 0 : FD_TEST( fd_cstr_printf_check( label, sizeof(label), NULL, "NUMA node %lu", i ) );
449 0 : PRINT(" %s%-23s%s %lu gigantic, %lu huge\n", c_dim, label, c_normal, fd_topo_gigantic_page_cnt( topo, i ), fd_topo_huge_page_cnt( topo, i, 0 ) );
450 0 : }
451 0 : }
452 :
453 0 : if( FD_UNLIKELY( topo->agave_affinity_cnt>0UL ) ) {
454 0 : char agave_affinity[4096];
455 0 : ulong offset = 0UL;
456 0 : for ( ulong i = 0UL; i < topo->agave_affinity_cnt; i++ ) {
457 0 : ulong sz;
458 0 : if( FD_LIKELY( i != 0UL )) FD_TEST( fd_cstr_printf_check( agave_affinity+offset, 4096-offset, &sz, ", %lu", topo->agave_affinity_cpu_idx[ i ] ) );
459 0 : else FD_TEST( fd_cstr_printf_check( agave_affinity+offset, 4096-offset, &sz, "%lu", topo->agave_affinity_cpu_idx[ i ] ) );
460 0 : offset += sz;
461 0 : }
462 0 : PRINT( " %s%-23s%s %s\n", c_dim, "Agave Affinity", c_normal, agave_affinity );
463 0 : }
464 :
465 0 : SECTION( "Workspaces (%lu)", topo->wksp_cnt );
466 0 : PRINT( " %s%3s %10s %-13s %5s %-8s %4s %12s %12s%s\n", c_dim, "ID", "SIZE", "NAME", "PAGES", "PAGE SZ", "NUMA", "FOOTPRINT", "LOOSE", c_normal );
467 0 : for( ulong i=0UL; i<topo->wksp_cnt; i++ ) {
468 0 : fd_topo_wksp_t * wksp = &topo->workspaces[ i ];
469 :
470 0 : ulong sz = wksp->page_sz * wksp->page_cnt;
471 0 : char size[ 24 ];
472 0 : fd_topo_mem_sz_string( sz, size );
473 0 : PRINT( " %3lu %s%10s%s %-13s %5lu %-8s %4lu %12lu %s%12lu%s\n",
474 0 : i, fd_topo_mem_sz_style( sz, c_bold, c_dim ), size, c_normal,
475 0 : wksp->name, wksp->page_cnt, fd_shmem_page_sz_to_cstr( wksp->page_sz ), wksp->numa_idx,
476 0 : wksp->known_footprint, c_dim, wksp->total_footprint - wksp->known_footprint, c_normal );
477 0 : }
478 :
479 0 : SECTION( "Objects (%lu)", topo->obj_cnt );
480 0 : PRINT( " %s%4s %10s %-13s %-13s %4s %12s %s%s\n", c_dim, "ID", "SIZE", "WORKSPACE", "OBJECT", "WKSP", "OFFSET", "PROPERTIES", c_normal );
481 0 : for( ulong i=0UL; i<topo->obj_cnt; i++ ) {
482 0 : fd_topo_obj_t * obj = &topo->objs[ i ];
483 :
484 0 : char size[ 24 ];
485 0 : fd_topo_mem_sz_string( obj->footprint, size );
486 0 : PRINT( " %4lu %s%10s%s %-13s %-13s %4lu %12lu ",
487 0 : i, fd_topo_mem_sz_style( obj->footprint, c_bold, c_dim ), size, c_normal,
488 0 : topo->workspaces[ obj->wksp_id ].name, obj->name,
489 0 : obj->wksp_id, obj->offset );
490 0 : for( fd_pod_iter_t iter=fd_pod_iter_init( fd_pod_queryf_subpod( topo->props, "obj.%lu", obj->id ) );
491 0 : !fd_pod_iter_done( iter );
492 0 : iter=fd_pod_iter_next( iter ) ) {
493 0 : fd_pod_info_t info = fd_pod_iter_info( iter );
494 0 : if( !strcmp( info.key, "seed" ) ) continue; /* key is a cstr (key_sz includes NUL) */
495 0 : PRINT( " %s%.*s=%s", c_dim, (int)info.key_sz, info.key, c_normal );
496 0 : switch( info.val_type ) {
497 0 : case FD_POD_VAL_TYPE_CSTR:
498 0 : PRINT( "%.*s", (int)info.val_sz, (char const *)info.val );
499 0 : break;
500 0 : case FD_POD_VAL_TYPE_ULONG: {
501 0 : ulong val; fd_ulong_svw_dec( info.val, &val );
502 0 : PRINT( "%lu", val );
503 0 : break;
504 0 : }
505 0 : case FD_POD_VAL_TYPE_INT: {
506 0 : ulong u; fd_ulong_svw_dec( info.val, &u );
507 0 : PRINT( "%i", (int)fd_long_zz_dec( u ) );
508 0 : break;
509 0 : }
510 0 : }
511 0 : }
512 0 : PRINT( "\n" );
513 0 : }
514 :
515 0 : SECTION( "Links (%lu)", topo->link_cnt );
516 0 : PRINT( " %s%3s %10s %-13s %4s %4s %8s %9s %5s%s\n", c_dim, "ID", "SIZE", "NAME", "KIND", "WKSP", "DEPTH", "MTU", "BURST", c_normal );
517 0 : for( ulong i=0UL; i<topo->link_cnt; i++ ) {
518 0 : fd_topo_link_t * link = &topo->links[ i ];
519 :
520 0 : ulong sz = fd_dcache_req_data_sz( link->mtu, link->depth, link->burst, 1 );
521 0 : char size[ 24 ];
522 0 : fd_topo_mem_sz_string( sz, size );
523 0 : PRINT( " %3lu %s%10s%s %-13s %4lu %4lu %8lu %9lu %5lu\n",
524 0 : i, fd_topo_mem_sz_style( sz, c_bold, c_dim ), size, c_normal,
525 0 : link->name, link->kind_id, topo->objs[ link->mcache_obj_id ].wksp_id, link->depth, link->mtu, link->burst );
526 0 : }
527 :
528 0 : #define PRINTIN( ... ) do { \
529 0 : int n = snprintf( cur_in, remaining_in, __VA_ARGS__ ); \
530 0 : if( FD_UNLIKELY( n < 0 ) ) FD_LOG_ERR(( "snprintf failed" )); \
531 0 : if( FD_UNLIKELY( (ulong)n >= remaining_in ) ) FD_LOG_ERR(( "snprintf overflow" )); \
532 0 : remaining_in -= (ulong)n; \
533 0 : cur_in += n; \
534 0 : } while( 0 )
535 :
536 0 : #define PRINTOUT( ... ) do { \
537 0 : int n = snprintf( cur_out, remaining_out, __VA_ARGS__ ); \
538 0 : if( FD_UNLIKELY( n < 0 ) ) FD_LOG_ERR(( "snprintf failed" )); \
539 0 : if( FD_UNLIKELY( (ulong)n >= remaining_out ) ) FD_LOG_ERR(( "snprintf overflow" )); \
540 0 : remaining_out -= (ulong)n; \
541 0 : cur_out += n; \
542 0 : } while( 0 )
543 :
544 0 : SECTION( "Tiles (%lu)", topo->tile_cnt );
545 0 : PRINT( " %s%3s %10s %-13s %4s %4s %4s %4s %s%s\n", c_dim, "ID", "MLOCK", "NAME", "KIND", "WKSP", "CPU", "NUMA", "LINKS / OBJECTS", c_normal );
546 0 : for( ulong i=0UL; i<topo->tile_cnt; i++ ) {
547 0 : fd_topo_tile_t * tile = &topo->tiles[ i ];
548 :
549 0 : char in[ 256 ] = {0};
550 0 : char * cur_in = in;
551 0 : ulong remaining_in = sizeof( in ) - 1;
552 :
553 0 : for( ulong j=0UL; j<tile->in_cnt; j++ ) {
554 0 : if( FD_LIKELY( j != 0 ) ) PRINTIN( ", " );
555 0 : if( FD_LIKELY( tile->in_link_reliable[ j ] ) ) PRINTIN( "%2lu", tile->in_link_id[ j ] );
556 0 : else PRINTIN( "%2ld", (long)-tile->in_link_id[ j ] );
557 0 : }
558 :
559 0 : char out[ 256 ] = {0};
560 0 : char * cur_out = out;
561 0 : ulong remaining_out = sizeof( out ) - 1;
562 :
563 0 : for( ulong j=0UL; j<tile->out_cnt; j++ ) {
564 0 : if( FD_LIKELY( j != 0 ) ) PRINTOUT( ", " );
565 0 : PRINTOUT( "%2lu", tile->out_link_id[ j ] );
566 0 : }
567 :
568 : /* Determine tile's NUMA node either based on CPU or wksp affinity */
569 0 : ulong tile_numa = 0UL;
570 0 : if( tile->cpu_idx!=ULONG_MAX ) {
571 0 : tile_numa = fd_shmem_numa_idx( tile->cpu_idx );
572 0 : } else {
573 0 : tile_numa = topo->workspaces[ topo->objs[ tile->tile_obj_id ].wksp_id ].numa_idx;
574 0 : }
575 :
576 0 : ulong mlock = fd_topo_mlock_max_tile1( topo, tile );
577 0 : char size[ 24 ];
578 0 : fd_topo_mem_sz_string( mlock, size );
579 0 : PRINT( " %3lu %s%10s%s %-13s %4lu %4lu ",
580 0 : i, fd_topo_mem_sz_style( mlock, c_bold, c_dim ), size, c_normal,
581 0 : tile->name, tile->kind_id, topo->objs[ tile->tile_obj_id ].wksp_id );
582 0 : if( tile->cpu_idx!=ULONG_MAX ) {
583 0 : PRINT( "%4lu", tile->cpu_idx );
584 0 : } else {
585 0 : PRINT( "%s%4s%s", c_dim, "any", c_normal );
586 0 : }
587 :
588 0 : PRINT( " %4lu %sin=%s[%s] %sout=%s[%s] %sobjs=[", tile_numa, c_dim, c_normal, in, c_dim, c_normal, out, c_dim );
589 0 : for( ulong j=0UL; j<tile->uses_obj_cnt; j++ ) {
590 0 : if( FD_LIKELY( j!=0 ) ) PRINT( " " );
591 0 : int is_rw = tile->uses_obj_mode[ j ] == FD_SHMEM_JOIN_MODE_READ_WRITE;
592 0 : PRINT( "%lu:%s", tile->uses_obj_id[ j ], is_rw?"rw":"ro" );
593 0 : }
594 0 : PRINT( "]%s", c_normal );
595 :
596 : #if 0
597 : PRINT( " wksps=[" );
598 : for( ulong j=0UL; j<topo->wksp_cnt; j++ ) {
599 : int mode = tile_needs_wksp( topo, tile, j );
600 : if( FD_UNLIKELY( -1!=mode ) ) {
601 : if( FD_LIKELY( j!=0 ) ) PRINT( " " );
602 : PRINT( "%s:%s", topo->workspaces[ j ].name, mode==FD_SHMEM_JOIN_MODE_READ_WRITE?"rw":"ro" );
603 : }
604 : }
605 : PRINT( "]" );
606 : #endif
607 :
608 0 : if( FD_LIKELY( i != topo->tile_cnt-1 ) ) PRINT( "\n" );
609 0 : }
610 :
611 0 : if( FD_UNLIKELY( stdout ) ) FD_LOG_STDOUT(( "%s\n", message ));
612 0 : else FD_LOG_INFO(( "%s", message ));
613 :
614 0 : #undef PRINT
615 0 : #undef PRINTIN
616 0 : #undef PRINTOUT
617 0 : #undef SECTION
618 0 : }
619 :
620 : /* PRINT_JSON_CSTR streams src (sz bytes, embedded NULs stop early)
621 : as a JSON string literal, escaping quotes, backslashes and control
622 : characters. No fixed intermediate buffer: expansion cannot
623 : truncate regardless of src length or content. */
624 :
625 0 : #define PRINT_JSON_CSTR( src, sz ) do { \
626 0 : char const * _s = (src); \
627 0 : ulong _sz = (sz); \
628 0 : PRINT( "\"" ); \
629 0 : for( ulong _i=0UL; _i<_sz && _s[ _i ]; _i++ ) { \
630 0 : uchar _c = (uchar)_s[ _i ]; \
631 0 : if( FD_UNLIKELY( _c=='"' || _c=='\\' ) ) PRINT( "\\%c", (char)_c ); \
632 0 : else if( FD_UNLIKELY( _c<0x20 ) ) PRINT( "\\u%04x", (uint)_c ); \
633 0 : else PRINT( "%c", (char)_c ); \
634 0 : } \
635 0 : PRINT( "\"" ); \
636 0 : } while( 0 )
637 :
638 : void
639 0 : fd_topo_print_json( fd_topo_t * topo ) {
640 0 : char message[ 32UL*4096UL ] = {0};
641 :
642 0 : char * cur = message;
643 0 : ulong remaining = sizeof(message) - 1;
644 :
645 0 : #define PRINT( ... ) do { \
646 0 : int n = snprintf( cur, remaining, __VA_ARGS__ ); \
647 0 : if( FD_UNLIKELY( n < 0 ) ) FD_LOG_ERR(( "snprintf failed" )); \
648 0 : if( FD_UNLIKELY( (ulong)n >= remaining ) ) { \
649 0 : *cur = '\0'; \
650 0 : FD_LOG_STDOUT(( "%s", message )); \
651 0 : cur = message; remaining = sizeof(message) - 1; \
652 0 : n = snprintf( cur, remaining, __VA_ARGS__ ); \
653 0 : if( FD_UNLIKELY( n < 0 ) ) FD_LOG_ERR(( "snprintf failed" )); \
654 0 : if( FD_UNLIKELY( (ulong)n >= remaining ) ) FD_LOG_ERR(( "snprintf overflow" )); \
655 0 : } \
656 0 : remaining -= (ulong)n; \
657 0 : cur += n; \
658 0 : } while( 0 )
659 :
660 0 : ulong stack_pages = topo->tile_cnt * FD_SHMEM_HUGE_PAGE_SZ * ((FD_TILE_PRIVATE_STACK_SZ/FD_SHMEM_HUGE_PAGE_SZ)+2UL);
661 0 : ulong normal_page_bytes = fd_topo_normal_page_cnt( topo ) * FD_SHMEM_NORMAL_PAGE_SZ;
662 0 : ulong total_bytes = fd_topo_mlock( topo ) + stack_pages + normal_page_bytes;
663 :
664 0 : ulong required_gigantic_pages = 0UL;
665 0 : ulong required_huge_pages = 0UL;
666 0 : ulong numa_node_cnt = fd_shmem_numa_cnt();
667 0 : for( ulong i=0UL; i<numa_node_cnt; i++ ) {
668 0 : required_gigantic_pages += fd_topo_gigantic_page_cnt( topo, i );
669 0 : required_huge_pages += fd_topo_huge_page_cnt( topo, i, 0 );
670 0 : }
671 :
672 0 : PRINT( "{\n \"summary\": {\n" );
673 0 : PRINT( " \"tile_cnt\": %lu,\n", topo->tile_cnt );
674 0 : PRINT( " \"total_memory_locked_bytes\": %lu,\n", total_bytes );
675 0 : PRINT( " \"required_gigantic_pages\": %lu,\n", required_gigantic_pages );
676 0 : PRINT( " \"required_huge_pages\": %lu,\n", required_huge_pages );
677 0 : PRINT( " \"required_normal_pages\": %lu,\n", fd_topo_normal_page_cnt( topo ) );
678 0 : PRINT( " \"numa_nodes\": [" );
679 0 : for( ulong i=0UL; i<numa_node_cnt; i++ ) {
680 0 : PRINT( "%s\n { \"node\": %lu, \"gigantic_pages\": %lu, \"huge_pages\": %lu }", i ? "," : "", i, fd_topo_gigantic_page_cnt( topo, i ), fd_topo_huge_page_cnt( topo, i, 0 ) );
681 0 : }
682 0 : PRINT( "\n ],\n" );
683 0 : PRINT( " \"agave_affinity\": [" );
684 0 : for( ulong i=0UL; i<topo->agave_affinity_cnt; i++ ) PRINT( "%s%lu", i ? ", " : "", topo->agave_affinity_cpu_idx[ i ] );
685 0 : PRINT( "]\n },\n" );
686 :
687 0 : PRINT( " \"workspaces\": [\n" );
688 0 : for( ulong i=0UL; i<topo->wksp_cnt; i++ ) {
689 0 : fd_topo_wksp_t * wksp = &topo->workspaces[ i ];
690 0 : PRINT( " { \"id\": %lu, \"name\": ", i );
691 0 : PRINT_JSON_CSTR( wksp->name, sizeof(wksp->name) );
692 0 : PRINT( ", \"page_cnt\": %lu, \"page_sz\": \"%s\", \"numa_idx\": %lu, \"total_bytes\": %lu, \"footprint\": %lu, \"loose\": %lu }%s\n",
693 0 : wksp->page_cnt, fd_shmem_page_sz_to_cstr( wksp->page_sz ), wksp->numa_idx,
694 0 : wksp->page_cnt * wksp->page_sz, wksp->known_footprint, wksp->total_footprint - wksp->known_footprint,
695 0 : i==topo->wksp_cnt-1UL ? "" : "," );
696 0 : }
697 0 : PRINT( " ],\n" );
698 :
699 0 : PRINT( " \"objects\": [\n" );
700 0 : for( ulong i=0UL; i<topo->obj_cnt; i++ ) {
701 0 : fd_topo_obj_t * obj = &topo->objs[ i ];
702 0 : PRINT( " { \"id\": %lu, \"name\": ", i );
703 0 : PRINT_JSON_CSTR( obj->name, sizeof(obj->name) );
704 0 : PRINT( ", \"wksp\": " );
705 0 : PRINT_JSON_CSTR( topo->workspaces[ obj->wksp_id ].name, sizeof(topo->workspaces[ obj->wksp_id ].name) );
706 0 : PRINT( ", \"wksp_id\": %lu, \"footprint\": %lu, \"offset\": %lu, \"props\": {",
707 0 : obj->wksp_id, obj->footprint, obj->offset );
708 0 : int first = 1;
709 0 : for( fd_pod_iter_t iter=fd_pod_iter_init( fd_pod_queryf_subpod( topo->props, "obj.%lu", obj->id ) );
710 0 : !fd_pod_iter_done( iter );
711 0 : iter=fd_pod_iter_next( iter ) ) {
712 0 : fd_pod_info_t info = fd_pod_iter_info( iter );
713 0 : if( !strcmp( info.key, "seed" ) ) continue; /* key is a cstr (key_sz includes NUL) */
714 0 : switch( info.val_type ) {
715 0 : case FD_POD_VAL_TYPE_CSTR: {
716 0 : PRINT( "%s ", first ? "" : "," );
717 0 : PRINT_JSON_CSTR( info.key, info.key_sz );
718 0 : PRINT( ": " );
719 0 : PRINT_JSON_CSTR( (char const *)info.val, info.val_sz );
720 0 : first = 0;
721 0 : break;
722 0 : }
723 0 : case FD_POD_VAL_TYPE_ULONG: {
724 0 : ulong val; fd_ulong_svw_dec( info.val, &val );
725 0 : PRINT( "%s ", first ? "" : "," );
726 0 : PRINT_JSON_CSTR( info.key, info.key_sz );
727 0 : PRINT( ": %lu", val );
728 0 : first = 0;
729 0 : break;
730 0 : }
731 0 : case FD_POD_VAL_TYPE_INT: {
732 0 : ulong u; fd_ulong_svw_dec( info.val, &u );
733 0 : PRINT( "%s ", first ? "" : "," );
734 0 : PRINT_JSON_CSTR( info.key, info.key_sz );
735 0 : PRINT( ": %i", (int)fd_long_zz_dec( u ) );
736 0 : first = 0;
737 0 : break;
738 0 : }
739 0 : }
740 0 : }
741 0 : PRINT( "%s} }%s\n", first ? "" : " ", i==topo->obj_cnt-1UL ? "" : "," );
742 0 : }
743 0 : PRINT( " ],\n" );
744 :
745 0 : PRINT( " \"links\": [\n" );
746 0 : for( ulong i=0UL; i<topo->link_cnt; i++ ) {
747 0 : fd_topo_link_t * link = &topo->links[ i ];
748 0 : PRINT( " { \"id\": %lu, \"name\": ", i );
749 0 : PRINT_JSON_CSTR( link->name, sizeof(link->name) );
750 0 : PRINT( ", \"kind_id\": %lu, \"wksp_id\": %lu, \"depth\": %lu, \"mtu\": %lu, \"burst\": %lu, \"dcache_bytes\": %lu }%s\n",
751 0 : link->kind_id, topo->objs[ link->mcache_obj_id ].wksp_id, link->depth, link->mtu, link->burst,
752 0 : fd_dcache_req_data_sz( link->mtu, link->depth, link->burst, 1 ),
753 0 : i==topo->link_cnt-1UL ? "" : "," );
754 0 : }
755 0 : PRINT( " ],\n" );
756 :
757 0 : PRINT( " \"tiles\": [\n" );
758 0 : for( ulong i=0UL; i<topo->tile_cnt; i++ ) {
759 0 : fd_topo_tile_t * tile = &topo->tiles[ i ];
760 :
761 0 : ulong tile_numa = 0UL;
762 0 : if( tile->cpu_idx!=ULONG_MAX ) tile_numa = fd_shmem_numa_idx( tile->cpu_idx );
763 0 : else tile_numa = topo->workspaces[ topo->objs[ tile->tile_obj_id ].wksp_id ].numa_idx;
764 :
765 0 : PRINT( " { \"id\": %lu, \"name\": ", i );
766 0 : PRINT_JSON_CSTR( tile->name, sizeof(tile->name) );
767 0 : PRINT( ", \"kind_id\": %lu, \"wksp_id\": %lu, ",
768 0 : tile->kind_id, topo->objs[ tile->tile_obj_id ].wksp_id );
769 0 : if( tile->cpu_idx!=ULONG_MAX ) PRINT( "\"cpu_idx\": %lu, ", tile->cpu_idx );
770 0 : else PRINT( "\"cpu_idx\": null, " );
771 0 : PRINT( "\"numa_idx\": %lu, \"mlock_bytes\": %lu, \"in_links\": [", tile_numa, fd_topo_mlock_max_tile1( topo, tile ) );
772 0 : for( ulong j=0UL; j<tile->in_cnt; j++ ) {
773 0 : PRINT( "%s{ \"link_id\": %lu, \"reliable\": %s }", j ? ", " : "", tile->in_link_id[ j ], tile->in_link_reliable[ j ] ? "true" : "false" );
774 0 : }
775 0 : PRINT( "], \"out_links\": [" );
776 0 : for( ulong j=0UL; j<tile->out_cnt; j++ ) PRINT( "%s%lu", j ? ", " : "", tile->out_link_id[ j ] );
777 0 : PRINT( "], \"objects\": [" );
778 0 : for( ulong j=0UL; j<tile->uses_obj_cnt; j++ ) {
779 0 : int is_rw = tile->uses_obj_mode[ j ] == FD_SHMEM_JOIN_MODE_READ_WRITE;
780 0 : PRINT( "%s{ \"obj_id\": %lu, \"mode\": \"%s\" }", j ? ", " : "", tile->uses_obj_id[ j ], is_rw ? "rw" : "ro" );
781 0 : }
782 0 : PRINT( "] }%s\n", i==topo->tile_cnt-1UL ? "" : "," );
783 0 : }
784 0 : PRINT( " ]\n}" );
785 :
786 0 : #undef PRINT
787 :
788 0 : FD_LOG_STDOUT(( "%s\n", message ));
789 0 : }
|
