Line data Source code
1 : #define _GNU_SOURCE
2 :
3 : /* Let's say there was a computer, the "leader" computer, that acted as
4 : a bank. Users could send it messages saying they wanted to deposit
5 : money, or transfer it to someone else.
6 :
7 : That's how, for example, Bank of America works but there are problems
8 : with it. One simple problem is: the bank can set your balance to
9 : zero if they don't like you.
10 :
11 : You could try to fix this by having the bank periodically publish the
12 : list of all account balances and transactions. If the customers add
13 : unforgeable signatures to their deposit slips and transfers, then
14 : the bank cannot zero a balance without it being obvious to everyone.
15 :
16 : There's still problems. The bank can't lie about your balance now or
17 : take your money, but it can just not accept deposits on your behalf
18 : by ignoring you.
19 :
20 : You could fix this by getting a few independent banks together, lets
21 : say Bank of America, Bank of England, and Westpac, and having them
22 : rotate who operates the leader computer periodically. If one bank
23 : ignores your deposits, you can just wait and send them to the next
24 : one.
25 :
26 : This is Solana.
27 :
28 : There's still problems of course but they are largely technical. How
29 : do the banks agree who is leader? How do you recover if a leader
30 : misbehaves? How do customers verify the transactions aren't forged?
31 : How do banks receive and publish and verify each others work quickly?
32 : These are the main technical innovations that enable Solana to work
33 : well.
34 :
35 : What about Proof of History?
36 :
37 : One particular niche problem is about the leader schedule. When the
38 : leader computer is moving from one bank to another, the new bank must
39 : wait for the old bank to say it's done and provide a final list of
40 : balances that it can start working off of. But: what if the computer
41 : at the old bank crashes and never says its done?
42 :
43 : Does the new leader just take over at some point? What if the new
44 : leader is malicious, and says the past thousand leaders crashed, and
45 : there have been no transactions for days? How do you check?
46 :
47 : This is what Proof of History solves. Each bank in the network must
48 : constantly do a lot of busywork (compute hashes), even when it is not
49 : leader.
50 :
51 : If the prior thousand leaders crashed, and no transactions happened
52 : in an hour, the new leader would have to show they did about an hour
53 : of busywork for everyone else to believe them.
54 :
55 : A better name for this is proof of skipping. If a leader is skipping
56 : slots (building off of a slot that is not the direct parent), it must
57 : prove that it waited a good amount of time to do so.
58 :
59 : It's not a perfect solution. For one thing, some banks have really
60 : fast computers and can compute a lot of busywork in a short amount of
61 : time, allowing them to skip prior slot(s) anyway. But: there is a
62 : social component that prevents validators from skipping the prior
63 : leader slot. It is easy to detect when this happens and the network
64 : could respond by ignoring their votes or stake.
65 :
66 : You could come up with other schemes: for example, the network could
67 : just use wall clock time. If a new leader publishes a block without
68 : waiting 400 milliseconds for the prior slot to complete, then there
69 : is no "proof of skipping" and the nodes ignore the slot.
70 :
71 : These schemes have a problem in that they are not deterministic
72 : across the network (different computers have different clocks), and
73 : so they will cause frequent forks which are very expensive to
74 : resolve. Even though the proof of history scheme is not perfect,
75 : it is better than any alternative which is not deterministic.
76 :
77 : With all that background, we can now describe at a high level what
78 : this PoH tile actually does,
79 :
80 : (1) Whenever any other leader in the network finishes a slot, and
81 : the slot is determined to be the best one to build off of, this
82 : tile gets "reset" onto that block, the so called "reset slot".
83 :
84 : (2) The tile is constantly doing busy work, hash(hash(hash(...))) on
85 : top of the last reset slot, even when it is not leader.
86 :
87 : (3) When the tile becomes leader, it continues hashing from where it
88 : was. Typically, the prior leader finishes their slot, so the
89 : reset slot will be the parent one, and this tile only publishes
90 : hashes for its own slot. But if prior slots were skipped, then
91 : there might be a whole chain already waiting.
92 :
93 : That's pretty much it. When we are leader, in addition to doing
94 : busywork, we publish ticks and microblocks to the shred tile. A
95 : microblock is a non-empty group of transactions whose hashes are
96 : mixed-in to the chain, while a tick is a periodic stamp of the
97 : current hash, with no transactions (nothing mixed in). We need
98 : to send both to the shred tile, as ticks are important for other
99 : validators to verify in parallel.
100 :
101 : As well, the tile should never become leader for a slot that it has
102 : published anything for, otherwise it may create a duplicate block.
103 :
104 : Some particularly common misunderstandings:
105 :
106 : - PoH is critical to security.
107 :
108 : This largely isn't true. The target hash rate of the network is
109 : so slow (1 hash per 500 nanoseconds) that a malicious leader can
110 : easily catch up if they start from an old hash, and the only
111 : practical attack prevented is the proof of skipping. Most of the
112 : long range attacks in the Solana whitepaper are not relevant.
113 :
114 : - PoH keeps passage of time.
115 :
116 : This is also not true. The way the network keeps time so it can
117 : decide who is leader is that, each leader uses their operating
118 : system clock to time 400 milliseconds and publishes their block
119 : when this timer expires.
120 :
121 : If a leader just hashed as fast as they could, they could publish
122 : a block in tens of milliseconds, and the rest of the network
123 : would happily accept it. This is why the Solana "clock" as
124 : determined by PoH is not accurate and drifts over time.
125 :
126 : - PoH prevents transaction reordering by the leader.
127 :
128 : The leader can, in theory, wait until the very end of their
129 : leader slot to publish anything at all to the network. They can,
130 : in particular, hold all received transactions for 400
131 : milliseconds and then reorder and publish some right at the end
132 : to advantage certain transactions.
133 :
134 : You might be wondering... if all the PoH chain is helping us do is
135 : prove that slots were skipped correctly, why do we need to "mix in"
136 : transactions to the hash value? Or do anything at all for slots
137 : where we don't skip the prior slot?
138 :
139 : It's a good question, and the answer is that this behavior is not
140 : necessary. An ideal implementation of PoH have no concept of ticks
141 : or mixins, and would not be part of the TPU pipeline at all.
142 : Instead, there would be a simple field "skip_proof" on the last
143 : shred we send for a slot, the hash(hash(...)) value. This field
144 : would only be filled in (and only verified by replayers) in cases
145 : where the slot actually skipped a parent.
146 :
147 : Then what is the "clock? In Solana, time is constructed as follows:
148 :
149 : HASHES
150 :
151 : The base unit of time is a hash. Hereafter, any values whose
152 : units are in hashes are called a "hashcnt" to distinguish them
153 : from actual hashed values.
154 :
155 : Agave generally defines a constant duration for each tick
156 : (see below) and then varies the number of hashcnt per tick, but
157 : as we consider the hashcnt the base unit of time, Firedancer and
158 : this PoH implementation defines everything in terms of hashcnt
159 : duration instead.
160 :
161 : In mainnet-beta, testnet, and devnet the hashcnt ticks over
162 : (increments) every 100 nanoseconds. The hashcnt rate is
163 : specified as 500 nanoseconds according to the genesis, but there
164 : are several features which increase the number of hashes per
165 : tick while keeping tick duration constant, which make the time
166 : per hashcnt lower. These features up to and including the
167 : `update_hashes_per_tick6` feature are activated on mainnet-beta,
168 : devnet, and testnet, and are described in the TICKS section
169 : below.
170 :
171 : Other chains and development environments might have a different
172 : hashcnt rate in the genesis, or they might not have activated
173 : the features which increase the rate yet, which we also support.
174 :
175 : In practice, although each validator follows a hashcnt rate of
176 : 100 nanoseconds, the overall observed hashcnt rate of the
177 : network is a little slower than once every 100 nanoseconds,
178 : mostly because there are gaps and clock synchronization issues
179 : during handoff between leaders. This is referred to as clock
180 : drift.
181 :
182 : TICKS
183 :
184 : The leader needs to periodically checkpoint the hash value
185 : associated with a given hashcnt so that they can publish it to
186 : other nodes for verification.
187 :
188 : On mainnet-beta, testnet, and devnet this occurs once every
189 : 62,500 hashcnts, or approximately once every 6.4 microseconds.
190 : This value is determined at genesis time, and according to the
191 : features below, and could be different in development
192 : environments or on other chains which we support.
193 :
194 : Due to protocol limitations, when mixing in transactions to the
195 : proof-of-history chain, it cannot occur on a tick boundary (but
196 : can occur at any other hashcnt).
197 :
198 : Ticks exist mainly so that verification can happen in parallel.
199 : A verifier computer, rather than needing to do hash(hash(...))
200 : all in sequence to verify a proof-of-history chain, can do,
201 :
202 : Core 0: hash(hash(...))
203 : Core 1: hash(hash(...))
204 : Core 2: hash(hash(...))
205 : Core 3: hash(hash(...))
206 : ...
207 :
208 : Between each pair of tick boundaries.
209 :
210 : Solana sometimes calls the current tick the "tick height",
211 : although it makes more sense to think of it as a counter from
212 : zero, it's just the number of ticks since the genesis hash.
213 :
214 : There is a set of features which increase the number of hashcnts
215 : per tick. These are all deployed on mainnet-beta, devnet, and
216 : testnet.
217 :
218 : name: update_hashes_per_tick
219 : id: 3uFHb9oKdGfgZGJK9EHaAXN4USvnQtAFC13Fh5gGFS5B
220 : hashes per tick: 12,500
221 : hashcnt duration: 500 nanos
222 :
223 : name: update_hashes_per_tick2
224 : id: EWme9uFqfy1ikK1jhJs8fM5hxWnK336QJpbscNtizkTU
225 : hashes per tick: 17,500
226 : hashcnt duration: 357.142857143 nanos
227 :
228 : name: update_hashes_per_tick3
229 : id: 8C8MCtsab5SsfammbzvYz65HHauuUYdbY2DZ4sznH6h5
230 : hashes per tick: 27,500
231 : hashcnt duration: 227.272727273 nanos
232 :
233 : name: update_hashes_per_tick4
234 : id: 8We4E7DPwF2WfAN8tRTtWQNhi98B99Qpuj7JoZ3Aikgg
235 : hashes per tick: 47,500
236 : hashcnt duration: 131.578947368 nanos
237 :
238 : name: update_hashes_per_tick5
239 : id: BsKLKAn1WM4HVhPRDsjosmqSg2J8Tq5xP2s2daDS6Ni4
240 : hashes per tick: 57,500
241 : hashcnt duration: 108.695652174 nanos
242 :
243 : name: update_hashes_per_tick6
244 : id: FKu1qYwLQSiehz644H6Si65U5ZQ2cp9GxsyFUfYcuADv
245 : hashes per tick: 62,500
246 : hashcnt duration: 100 nanos
247 :
248 : In development environments, there is a way to configure the
249 : hashcnt per tick to be "none" during genesis, for a so-called
250 : "low power" tick producer. The idea is not to spin cores during
251 : development. This is equivalent to setting the hashcnt per tick
252 : to be 1, and increasing the hashcnt duration to the desired tick
253 : duration.
254 :
255 : SLOTS
256 :
257 : Each leader needs to be leader for a fixed amount of time, which
258 : is called a slot. During a slot, a leader has an opportunity to
259 : receive transactions and produce a block for the network,
260 : although they may miss ("skip") the slot if they are offline or
261 : not behaving.
262 :
263 : In mainnet-beta, testnet, and devnet a slot is 64 ticks, or
264 : 4,000,000 hashcnts, or approximately 400 milliseconds.
265 :
266 : Due to the way the leader schedule is constructed, each leader
267 : is always given at least four (4) consecutive slots in the
268 : schedule. This means when becoming leader you will be leader
269 : for at least 4 slots, or 1.6 seconds.
270 :
271 : It is rare, although can happen that a leader gets more than 4
272 : consecutive slots (eg, 8, or 12), if they are lucky with the
273 : leader schedule generation.
274 :
275 : The number of ticks in a slot is fixed at genesis time, and
276 : could be different for development or other chains, which we
277 : support. There is nothing special about 4 leader slots in a
278 : row, and this might be changed in future, and the proof of
279 : history makes no assumptions that this is the case.
280 :
281 : EPOCHS
282 :
283 : Infrequently, the network needs to do certain housekeeping,
284 : mainly things like collecting rent and deciding on the leader
285 : schedule. The length of an epoch is fixed on mainnet-beta,
286 : devnet and testnet at 420,000 slots, or around ~2 (1.94) days.
287 : This value is fixed at genesis time, and could be different for
288 : other chains including development, which we support. Typically
289 : in development, epochs are every 8,192 slots, or around ~1 hour
290 : (54.61 minutes), although it depends on the number of ticks per
291 : slot and the target hashcnt rate of the genesis as well.
292 :
293 : In development, epochs need not be a fixed length either. There
294 : is a "warmup" option, where epochs start short and grow, which
295 : is useful for quickly warming up stake during development.
296 :
297 : The epoch is important because it is the only time the leader
298 : schedule is updated. The leader schedule is a list of which
299 : leader is leader for which slot, and is generated by a special
300 : algorithm that is deterministic and known to all nodes.
301 :
302 : The leader schedule is computed one epoch in advance, so that
303 : at slot T, we always know who will be leader up until the end
304 : of slot T+EPOCH_LENGTH. Specifically, the leader schedule for
305 : epoch N is computed during the epoch boundary crossing from
306 : N-2 to N-1. For mainnet-beta, the slots per epoch is fixed and
307 : will always be 420,000. */
308 :
309 : #include "../../disco/tiles.h"
310 : #include "../../disco/bundle/fd_bundle_crank.h"
311 : #include "../../disco/pack/fd_pack.h"
312 : #include "../../ballet/sha256/fd_sha256.h"
313 : #include "../../disco/metrics/fd_metrics.h"
314 : #include "../../util/pod/fd_pod.h"
315 : #include "../../disco/shred/fd_shredder.h"
316 : #include "../../disco/keyguard/fd_keyload.h"
317 : #include "../../disco/keyguard/fd_keyswitch.h"
318 : #include "../../disco/metrics/generated/fd_metrics_poh.h"
319 : #include "../../disco/plugin/fd_plugin.h"
320 : #include "../../flamenco/leaders/fd_multi_epoch_leaders.h"
321 :
322 : #include <string.h>
323 :
324 : /* The maximum number of microblocks that pack is allowed to pack into a
325 : single slot. This is not consensus critical, and pack could, if we
326 : let it, produce as many microblocks as it wants, and the slot would
327 : still be valid.
328 :
329 : We have this here instead so that PoH can estimate slot completion,
330 : and keep the hashcnt up to date as pack progresses through packing
331 : the slot. If this upper bound was not enforced, PoH could tick to
332 : the last hash of the slot and have no hashes left to mixin incoming
333 : microblocks from pack, so this upper bound is a coordination
334 : mechanism so that PoH can progress hashcnts while the slot is active,
335 : and know that pack will not need those hashcnts later to do mixins. */
336 0 : #define MAX_MICROBLOCKS_PER_SLOT (32768UL)
337 :
338 : /* When we are hashing in the background in case a prior leader skips
339 : their slot, we need to store the result of each tick hash so we can
340 : publish them when we become leader. The network requires at least
341 : one leader slot to publish in each epoch for the leader schedule to
342 : generate, so in the worst case we might need two full epochs of slots
343 : to store the hashes. (Eg, if epoch T only had a published slot in
344 : position 0 and epoch T+1 only had a published slot right at the end).
345 :
346 : There is a tighter bound: the block data limit of mainnet-beta is
347 : currently FD_PACK_MAX_DATA_PER_BLOCK, or 27,332,342 bytes per slot.
348 : At 48 bytes per tick, it is not possible to publish a slot that skips
349 : 569,424 or more prior slots. */
350 0 : #define MAX_SKIPPED_TICKS (1UL+(FD_PACK_MAX_DATA_PER_BLOCK/48UL))
351 :
352 0 : #define IN_KIND_BANK (0)
353 0 : #define IN_KIND_PACK (1)
354 0 : #define IN_KIND_STAKE (2)
355 :
356 :
357 : typedef struct {
358 : fd_wksp_t * mem;
359 : ulong chunk0;
360 : ulong wmark;
361 : } fd_poh_in_ctx_t;
362 :
363 : typedef struct {
364 : ulong idx;
365 : fd_wksp_t * mem;
366 : ulong chunk0;
367 : ulong wmark;
368 : ulong chunk;
369 : } fd_poh_out_ctx_t;
370 :
371 : typedef struct {
372 : fd_stem_context_t * stem;
373 :
374 : /* Static configuration determined at genesis creation time. See
375 : long comment above for more information. */
376 : ulong tick_duration_ns;
377 : ulong hashcnt_per_tick;
378 : ulong ticks_per_slot;
379 :
380 : /* Derived from the above configuration, but we precompute it. */
381 : double slot_duration_ns;
382 : double hashcnt_duration_ns;
383 : ulong hashcnt_per_slot;
384 : /* Constant, fixed at initialization. The maximum number of
385 : microblocks that the pack tile can publish in each slot. */
386 : ulong max_microblocks_per_slot;
387 :
388 : /* Consensus-critical slot cost limits. */
389 : struct {
390 : ulong slot_max_cost;
391 : ulong slot_max_vote_cost;
392 : ulong slot_max_write_cost_per_acct;
393 : } limits;
394 :
395 : /* The current slot and hashcnt within that slot of the proof of
396 : history, including hashes we have been producing in the background
397 : while waiting for our next leader slot. */
398 : ulong slot;
399 : ulong hashcnt;
400 : ulong cus_used;
401 :
402 : /* When we send a microblock on to the shred tile, we need to tell
403 : it how many hashes there have been since the last microblock, so
404 : this tracks the hashcnt of the last published microblock.
405 :
406 : If we are skipping slots prior to our leader slot, the last_slot
407 : will be quite old, and potentially much larger than the number of
408 : hashcnts in one slot. */
409 : ulong last_slot;
410 : ulong last_hashcnt;
411 :
412 : /* If we have published a tick or a microblock for a particular slot
413 : to the shred tile, we should never become leader for that slot
414 : again, otherwise we could publish a duplicate block.
415 :
416 : This value tracks the max slot that we have published a tick or
417 : microblock for so we can prevent this. */
418 : ulong highwater_leader_slot;
419 :
420 : /* See how this field is used below. If we have sequential leader
421 : slots, we don't reset the expected slot end time between the two,
422 : to prevent clock drift. If we didn't do this, our 2nd slot would
423 : end 400ms + `time_for_replay_to_move_slot_and_reset_poh` after
424 : our 1st, rather than just strictly 400ms. */
425 : int lagged_consecutive_leader_start;
426 : ulong expect_sequential_leader_slot;
427 :
428 : /* There's a race condition ... let's say two banks A and B, bank A
429 : processes some transactions, then releases the account locks, and
430 : sends the microblock to PoH to be stamped. Pack now re-packs the
431 : same accounts with a new microblock, sends to bank B, bank B
432 : executes and sends the microblock to PoH, and this all happens fast
433 : enough that PoH picks the 2nd block to stamp before the 1st. The
434 : accounts database changes now are misordered with respect to PoH so
435 : replay could fail.
436 :
437 : To prevent this race, we order all microblocks and only process
438 : them in PoH in the order they are produced by pack. This is a
439 : little bit over-strict, we just need to ensure that microblocks
440 : with conflicting accounts execute in order, but this is easiest to
441 : implement for now. */
442 : uint expect_pack_idx;
443 :
444 : /* If we have received the slot done message from pack yet. We are
445 : not allowed to fully finish hashing the block until this happens so
446 : that we know which slot the slot_done message is arriving for. */
447 : int slot_done;
448 :
449 : /* Pack and bank tiles need a reference to the bank object with a
450 : slightly different lifetime than current_leader_bank, particularly
451 : when we switch forks in the middle of a leader slot. We need to
452 : make sure we don't free the last reference to the bank while the
453 : pack or bank tiles are still using it. The strange thing is that
454 : bank tiles have no concept of the current slot, but we know they're
455 : done with the bank object when pack's inter-slot bank draining
456 : process is complete. Pack notifies PoH by a frag with
457 : sig==ULONG_MAX on the pack_poh link when the banks are drained, and
458 : the PoH tile must then free the reference on behalf of pack.
459 :
460 : pack_leader_bank is non-NULL when the reference we're holding on
461 : behalf of the pack tile is acquired, and NULL when it is not
462 : acquired. */
463 : void const * pack_leader_bank;
464 :
465 : /* The PoH tile must never drop microblocks that get committed by the
466 : bank, so it needs to always be able to mixin a microblock hash.
467 : Mixing in requires incrementing the hashcnt, so we need to ensure
468 : at all times that there is enough hascnts left in the slot to
469 : mixin whatever future microblocks pack might produce for it.
470 :
471 : This value tracks that. At any time, max_microblocks_per_slot
472 : - microblocks_lower_bound is an upper bound on the maximum number
473 : of microblocks that might still be received in this slot. */
474 : ulong microblocks_lower_bound;
475 :
476 : uchar __attribute__((aligned(32UL))) reset_hash[ 32 ];
477 : uchar __attribute__((aligned(32UL))) hash[ 32 ];
478 :
479 : /* When we are not leader, we need to save the hashes that were
480 : produced in case the prior leader skips. If they skip, we will
481 : replay these skipped hashes into our next leader bank so that
482 : the slot hashes sysvar can be updated correctly, and also publish
483 : them to peer nodes as part of our outgoing shreds. */
484 : uchar skipped_tick_hashes[ MAX_SKIPPED_TICKS ][ 32 ];
485 :
486 : /* The timestamp in nanoseconds of when the reset slot was received.
487 : This is the timestamp we are building on top of to determine when
488 : our next leader slot starts. */
489 : long reset_slot_start_ns;
490 :
491 : /* The timestamp in nanoseconds of when we got the bank for the
492 : current leader slot. */
493 : long leader_bank_start_ns;
494 :
495 : /* The hashcnt corresponding to the start of the current reset slot. */
496 : ulong reset_slot;
497 :
498 : /* The hashcnt at which our next leader slot begins, or ULONG max if
499 : we have no known next leader slot. */
500 : ulong next_leader_slot;
501 :
502 : /* If an in progress frag should be skipped */
503 : int skip_frag;
504 :
505 : ulong max_active_descendant;
506 :
507 : /* If we currently are the leader according the clock AND we have
508 : received the leader bank for the slot from the replay stage,
509 : this value will be non-NULL.
510 :
511 : Note that we might be inside our leader slot, but not have a bank
512 : yet, in which case this will still be NULL.
513 :
514 : It will be NULL for a brief race period between consecutive leader
515 : slots, as we ping-pong back to replay stage waiting for a new bank.
516 :
517 : Agave refers to this as the "working bank". */
518 : void const * current_leader_bank;
519 :
520 : fd_sha256_t * sha256;
521 :
522 : fd_multi_epoch_leaders_t * mleaders;
523 :
524 : /* The last sequence number of an outgoing fragment to the shred tile,
525 : or ULONG max if no such fragment. See fd_keyswitch.h for details
526 : of how this is used. */
527 : ulong shred_seq;
528 :
529 : int halted_switching_key;
530 :
531 : fd_keyswitch_t * keyswitch;
532 : fd_pubkey_t identity_key;
533 :
534 : /* We need a few pieces of information to compute the right addresses
535 : for bundle crank information that we need to send to pack. */
536 : struct {
537 : int enabled;
538 : fd_pubkey_t vote_account;
539 : fd_bundle_crank_gen_t gen[1];
540 : } bundle;
541 :
542 :
543 : /* The Agave client needs to be notified when the leader changes,
544 : so that they can resume the replay stage if it was suspended waiting. */
545 : void * signal_leader_change;
546 :
547 : /* These are temporarily set in during_frag so they can be used in
548 : after_frag once the frag has been validated as not overrun. */
549 : uchar _txns[ USHORT_MAX ];
550 : fd_microblock_trailer_t _microblock_trailer[ 1 ];
551 :
552 : int in_kind[ 64 ];
553 : fd_poh_in_ctx_t in[ 64 ];
554 :
555 : fd_poh_out_ctx_t shred_out[ 1 ];
556 : fd_poh_out_ctx_t pack_out[ 1 ];
557 : fd_poh_out_ctx_t plugin_out[ 1 ];
558 :
559 : fd_histf_t begin_leader_delay[ 1 ];
560 : fd_histf_t first_microblock_delay[ 1 ];
561 : fd_histf_t slot_done_delay[ 1 ];
562 : fd_histf_t bundle_init_delay[ 1 ];
563 :
564 : ulong features_activation_avail;
565 : fd_shred_features_activation_t features_activation[1];
566 :
567 : ulong parent_slot;
568 : uchar parent_block_id[ 32 ];
569 :
570 : uchar __attribute__((aligned(FD_MULTI_EPOCH_LEADERS_ALIGN))) mleaders_mem[ FD_MULTI_EPOCH_LEADERS_FOOTPRINT ];
571 : } fd_poh_ctx_t;
572 :
573 : /* The PoH recorder is implemented in Firedancer but for now needs to
574 : work with Agave, so we have a locking scheme for them to
575 : co-operate.
576 :
577 : This is because the PoH tile lives in the Agave memory address
578 : space and their version of concurrency is locking the PoH recorder
579 : and reading arbitrary fields.
580 :
581 : So we allow them to lock the PoH tile, although with a very bad (for
582 : them) locking scheme. By default, the tile has full and exclusive
583 : access to the data. If part of Agave wishes to read/write they
584 : can either,
585 :
586 : 1. Rewrite their concurrency to message passing based on mcache
587 : (preferred, but not feasible).
588 : 2. Signal to the tile they wish to acquire the lock, by setting
589 : fd_poh_waiting_lock to 1.
590 :
591 : During after_credit, the tile will check if the waiting lock is set
592 : to 1, and if so, set the returned lock to 1, indicating to the waiter
593 : that they may now proceed.
594 :
595 : When the waiter is done reading and writing, they restore the
596 : returned lock value back to zero, and the POH tile continues with its
597 : day. */
598 :
599 : static fd_poh_ctx_t * fd_poh_global_ctx;
600 :
601 : static volatile ulong fd_poh_waiting_lock __attribute__((aligned(128UL)));
602 : static volatile ulong fd_poh_returned_lock __attribute__((aligned(128UL)));
603 :
604 : /* Agave also needs to write to some mcaches, so we trampoline
605 : that via. the PoH tile as well. */
606 :
607 : struct poh_link {
608 : fd_frag_meta_t * mcache;
609 : ulong depth;
610 : ulong tx_seq;
611 :
612 : void * mem;
613 : void * dcache;
614 : ulong chunk0;
615 : ulong wmark;
616 : ulong chunk;
617 :
618 : ulong cr_avail;
619 : ulong rx_cnt;
620 : ulong * rx_fseqs[ 32UL ];
621 : };
622 :
623 : typedef struct poh_link poh_link_t;
624 :
625 : static poh_link_t gossip_dedup;
626 : static poh_link_t stake_out;
627 : static poh_link_t crds_shred;
628 : static poh_link_t replay_resolv;
629 : static poh_link_t executed_txn;
630 :
631 : static poh_link_t replay_plugin;
632 : static poh_link_t gossip_plugin;
633 : static poh_link_t start_progress_plugin;
634 : static poh_link_t vote_listener_plugin;
635 : static poh_link_t validator_info_plugin;
636 :
637 : static void
638 0 : poh_link_wait_credit( poh_link_t * link ) {
639 0 : if( FD_LIKELY( link->cr_avail ) ) return;
640 :
641 0 : while( 1 ) {
642 0 : ulong cr_query = ULONG_MAX;
643 0 : for( ulong i=0UL; i<link->rx_cnt; i++ ) {
644 0 : ulong const * _rx_seq = link->rx_fseqs[ i ];
645 0 : ulong rx_seq = FD_VOLATILE_CONST( *_rx_seq );
646 0 : ulong rx_cr_query = (ulong)fd_long_max( (long)link->depth - fd_long_max( fd_seq_diff( link->tx_seq, rx_seq ), 0L ), 0L );
647 0 : cr_query = fd_ulong_min( rx_cr_query, cr_query );
648 0 : }
649 0 : if( FD_LIKELY( cr_query>0UL ) ) {
650 0 : link->cr_avail = cr_query;
651 0 : break;
652 0 : }
653 0 : FD_SPIN_PAUSE();
654 0 : }
655 0 : }
656 :
657 : static void
658 : poh_link_publish( poh_link_t * link,
659 : ulong sig,
660 : uchar const * data,
661 0 : ulong data_sz ) {
662 0 : while( FD_UNLIKELY( !FD_VOLATILE_CONST( link->mcache ) ) ) FD_SPIN_PAUSE();
663 0 : if( FD_UNLIKELY( !link->mem ) ) return; /* link not enabled, don't publish */
664 0 : poh_link_wait_credit( link );
665 :
666 0 : uchar * dst = (uchar *)fd_chunk_to_laddr( link->mem, link->chunk );
667 0 : fd_memcpy( dst, data, data_sz );
668 0 : ulong tspub = (ulong)fd_frag_meta_ts_comp( fd_tickcount() );
669 0 : fd_mcache_publish( link->mcache, link->depth, link->tx_seq, sig, link->chunk, data_sz, 0UL, 0UL, tspub );
670 0 : link->chunk = fd_dcache_compact_next( link->chunk, data_sz, link->chunk0, link->wmark );
671 0 : link->cr_avail--;
672 0 : link->tx_seq++;
673 0 : }
674 :
675 : static void
676 : poh_link_init( poh_link_t * link,
677 : fd_topo_t * topo,
678 : fd_topo_tile_t * tile,
679 0 : ulong out_idx ) {
680 0 : fd_topo_link_t * topo_link = &topo->links[ tile->out_link_id[ out_idx ] ];
681 0 : fd_topo_wksp_t * wksp = &topo->workspaces[ topo->objs[ topo_link->dcache_obj_id ].wksp_id ];
682 :
683 0 : link->mem = wksp->wksp;
684 0 : link->depth = fd_mcache_depth( topo_link->mcache );
685 0 : link->tx_seq = 0UL;
686 0 : link->dcache = topo_link->dcache;
687 0 : link->chunk0 = fd_dcache_compact_chunk0( wksp->wksp, topo_link->dcache );
688 0 : link->wmark = fd_dcache_compact_wmark ( wksp->wksp, topo_link->dcache, topo_link->mtu );
689 0 : link->chunk = link->chunk0;
690 0 : link->cr_avail = 0UL;
691 0 : link->rx_cnt = 0UL;
692 0 : for( ulong i=0UL; i<topo->tile_cnt; i++ ) {
693 0 : fd_topo_tile_t * _tile = &topo->tiles[ i ];
694 0 : for( ulong j=0UL; j<_tile->in_cnt; j++ ) {
695 0 : if( _tile->in_link_id[ j ]==topo_link->id && _tile->in_link_reliable[ j ] ) {
696 0 : FD_TEST( link->rx_cnt<32UL );
697 0 : link->rx_fseqs[ link->rx_cnt++ ] = _tile->in_link_fseq[ j ];
698 0 : break;
699 0 : }
700 0 : }
701 0 : }
702 0 : FD_COMPILER_MFENCE();
703 0 : link->mcache = topo_link->mcache;
704 0 : FD_COMPILER_MFENCE();
705 0 : FD_TEST( link->mcache );
706 0 : }
707 :
708 : /* To help show correctness, functions that might be called from
709 : Rust, either directly or indirectly, have this fake "attribute"
710 : CALLED_FROM_RUST, which is actually nothing. Calls from Rust
711 : typically execute on threads did not call fd_boot, so they do not
712 : have the typical FD_TL variables. In particular, they cannot use
713 : normal metrics, and their log messages don't have full context.
714 : Additionally, Rust functions marked CALLED_FROM_RUST cannot call back
715 : into a C fd_ext function without causing a deadlock (although the
716 : other Rust fd_ext functions have a similar problem).
717 :
718 : To prevent annotation from polluting the whole codebase, calls to
719 : functions outside this file are manually checked and marked as being
720 : safe at each call rather than annotated. */
721 : #define CALLED_FROM_RUST
722 :
723 : static CALLED_FROM_RUST fd_poh_ctx_t *
724 0 : fd_ext_poh_write_lock( void ) {
725 0 : for(;;) {
726 : /* Acquire the waiter lock to make sure we are the first writer in the queue. */
727 0 : if( FD_LIKELY( !FD_ATOMIC_CAS( &fd_poh_waiting_lock, 0UL, 1UL) ) ) break;
728 0 : FD_SPIN_PAUSE();
729 0 : }
730 0 : FD_COMPILER_MFENCE();
731 0 : for(;;) {
732 : /* Now wait for the tile to tell us we can proceed. */
733 0 : if( FD_LIKELY( FD_VOLATILE_CONST( fd_poh_returned_lock ) ) ) break;
734 0 : FD_SPIN_PAUSE();
735 0 : }
736 0 : FD_COMPILER_MFENCE();
737 0 : return fd_poh_global_ctx;
738 0 : }
739 :
740 : static CALLED_FROM_RUST void
741 0 : fd_ext_poh_write_unlock( void ) {
742 0 : FD_COMPILER_MFENCE();
743 0 : FD_VOLATILE( fd_poh_returned_lock ) = 0UL;
744 0 : }
745 :
746 : /* The PoH tile needs to interact with the Agave address space to
747 : do certain operations that Firedancer hasn't reimplemented yet, a.k.a
748 : transaction execution. We have Agave export some wrapper
749 : functions that we call into during regular tile execution. These do
750 : not need any locking, since they are called serially from the single
751 : PoH tile. */
752 :
753 : extern CALLED_FROM_RUST void fd_ext_bank_acquire( void const * bank );
754 : extern CALLED_FROM_RUST void fd_ext_bank_release( void const * bank );
755 : extern CALLED_FROM_RUST void fd_ext_poh_signal_leader_change( void * sender );
756 : extern void fd_ext_poh_register_tick( void const * bank, uchar const * hash );
757 :
758 : /* fd_ext_poh_initialize is called by Agave on startup to
759 : initialize the PoH tile with some static configuration, and the
760 : initial reset slot and hash which it retrieves from a snapshot.
761 :
762 : This function is called by some random Agave thread, but
763 : it blocks booting of the PoH tile. The tile will spin until it
764 : determines that this initialization has happened.
765 :
766 : signal_leader_change is an opaque Rust object that is used to
767 : tell the replay stage that the leader has changed. It is a
768 : Box::into_raw(Arc::increment_strong(crossbeam::Sender)), so it
769 : has infinite lifetime unless this C code releases the refcnt.
770 :
771 : It can be used with `fd_ext_poh_signal_leader_change` which
772 : will just issue a nonblocking send on the channel. */
773 :
774 : CALLED_FROM_RUST void
775 : fd_ext_poh_initialize( ulong tick_duration_ns, /* See clock comments above, will be 6.4 microseconds for mainnet-beta. */
776 : ulong hashcnt_per_tick, /* See clock comments above, will be 62,500 for mainnet-beta. */
777 : ulong ticks_per_slot, /* See clock comments above, will almost always be 64. */
778 : ulong tick_height, /* The counter (height) of the tick to start hashing on top of. */
779 : uchar const * last_entry_hash, /* Points to start of a 32 byte region of memory, the hash itself at the tick height. */
780 0 : void * signal_leader_change /* See comment above. */ ) {
781 0 : FD_COMPILER_MFENCE();
782 0 : for(;;) {
783 : /* Make sure the ctx is initialized before trying to take the lock. */
784 0 : if( FD_LIKELY( FD_VOLATILE_CONST( fd_poh_global_ctx ) ) ) break;
785 0 : FD_SPIN_PAUSE();
786 0 : }
787 0 : fd_poh_ctx_t * ctx = fd_ext_poh_write_lock();
788 :
789 0 : ctx->slot = tick_height/ticks_per_slot;
790 0 : ctx->hashcnt = 0UL;
791 0 : ctx->cus_used = 0UL;
792 0 : ctx->last_slot = ctx->slot;
793 0 : ctx->last_hashcnt = 0UL;
794 0 : ctx->reset_slot = ctx->slot;
795 0 : ctx->reset_slot_start_ns = fd_log_wallclock(); /* safe to call from Rust */
796 :
797 0 : memcpy( ctx->reset_hash, last_entry_hash, 32UL );
798 0 : memcpy( ctx->hash, last_entry_hash, 32UL );
799 :
800 0 : ctx->signal_leader_change = signal_leader_change;
801 :
802 : /* Static configuration about the clock. */
803 0 : ctx->tick_duration_ns = tick_duration_ns;
804 0 : ctx->hashcnt_per_tick = hashcnt_per_tick;
805 0 : ctx->ticks_per_slot = ticks_per_slot;
806 :
807 : /* Recompute derived information about the clock. */
808 0 : ctx->slot_duration_ns = (double)ticks_per_slot*(double)tick_duration_ns;
809 0 : ctx->hashcnt_duration_ns = (double)tick_duration_ns/(double)hashcnt_per_tick;
810 0 : ctx->hashcnt_per_slot = ticks_per_slot*hashcnt_per_tick;
811 :
812 0 : if( FD_UNLIKELY( ctx->hashcnt_per_tick==1UL ) ) {
813 : /* Low power producer, maximum of one microblock per tick in the slot */
814 0 : ctx->max_microblocks_per_slot = ctx->ticks_per_slot;
815 0 : } else {
816 : /* See the long comment in after_credit for this limit */
817 0 : ctx->max_microblocks_per_slot = fd_ulong_min( MAX_MICROBLOCKS_PER_SLOT, ctx->ticks_per_slot*(ctx->hashcnt_per_tick-1UL) );
818 0 : }
819 :
820 0 : fd_ext_poh_write_unlock();
821 0 : }
822 :
823 : /* fd_ext_poh_acquire_bank gets the current leader bank if there is one
824 : currently active. PoH might think we are leader without having a
825 : leader bank if the replay stage has not yet noticed we are leader.
826 :
827 : The bank that is returned is owned the caller, and must be converted
828 : to an Arc<Bank> by calling Arc::from_raw() on it. PoH increments the
829 : reference count before returning the bank, so that it can also keep
830 : its internal copy.
831 :
832 : If there is no leader bank, NULL is returned. In this case, the
833 : caller should not call `Arc::from_raw()`. */
834 :
835 : CALLED_FROM_RUST void const *
836 0 : fd_ext_poh_acquire_leader_bank( void ) {
837 0 : fd_poh_ctx_t * ctx = fd_ext_poh_write_lock();
838 0 : void const * bank = NULL;
839 0 : if( FD_LIKELY( ctx->current_leader_bank ) ) {
840 : /* Clone refcount before we release the lock. */
841 0 : fd_ext_bank_acquire( ctx->current_leader_bank );
842 0 : bank = ctx->current_leader_bank;
843 0 : }
844 0 : fd_ext_poh_write_unlock();
845 0 : return bank;
846 0 : }
847 :
848 : /* fd_ext_poh_reset_slot returns the slot height one above the last good
849 : (unskipped) slot we are building on top of. This is always a good
850 : known value, and will not be ULONG_MAX. */
851 :
852 : CALLED_FROM_RUST ulong
853 0 : fd_ext_poh_reset_slot( void ) {
854 0 : fd_poh_ctx_t * ctx = fd_ext_poh_write_lock();
855 0 : ulong reset_slot = ctx->reset_slot;
856 0 : fd_ext_poh_write_unlock();
857 0 : return reset_slot;
858 0 : }
859 :
860 : CALLED_FROM_RUST void
861 0 : fd_ext_poh_update_active_descendant( ulong max_active_descendant ) {
862 0 : fd_poh_ctx_t * ctx = fd_ext_poh_write_lock();
863 0 : ctx->max_active_descendant = max_active_descendant;
864 0 : fd_ext_poh_write_unlock();
865 0 : }
866 :
867 : /* fd_ext_poh_reached_leader_slot returns 1 if we have reached a slot
868 : where we are leader. This is used by the replay stage to determine
869 : if it should create a new leader bank descendant of the prior reset
870 : slot block.
871 :
872 : Sometimes, even when we reach our slot we do not return 1, as we are
873 : giving a grace period to the prior leader to finish publishing their
874 : block.
875 :
876 : out_leader_slot is the slot height of the leader slot we reached, and
877 : reset_slot is the slot height of the last good (unskipped) slot we
878 : are building on top of. */
879 :
880 : CALLED_FROM_RUST int
881 : fd_ext_poh_reached_leader_slot( ulong * out_leader_slot,
882 0 : ulong * out_reset_slot ) {
883 0 : fd_poh_ctx_t * ctx = fd_ext_poh_write_lock();
884 :
885 0 : *out_leader_slot = ctx->next_leader_slot;
886 0 : *out_reset_slot = ctx->reset_slot;
887 :
888 0 : if( FD_UNLIKELY( ctx->next_leader_slot==ULONG_MAX ||
889 0 : ctx->slot<ctx->next_leader_slot ) ) {
890 : /* Didn't reach our leader slot yet. */
891 0 : fd_ext_poh_write_unlock();
892 0 : return 0;
893 0 : }
894 :
895 0 : if( FD_UNLIKELY( ctx->halted_switching_key ) ) {
896 : /* Reached our leader slot, but the leader pipeline is halted
897 : because we are switching identity key. */
898 0 : fd_ext_poh_write_unlock();
899 0 : return 0;
900 0 : }
901 :
902 0 : if( FD_LIKELY( ctx->reset_slot==ctx->next_leader_slot ) ) {
903 : /* We were reset onto our leader slot, because the prior leader
904 : completed theirs, so we should start immediately, no need for a
905 : grace period. */
906 0 : fd_ext_poh_write_unlock();
907 0 : return 1;
908 0 : }
909 :
910 0 : long now_ns = fd_log_wallclock();
911 0 : long expected_start_time_ns = ctx->reset_slot_start_ns + (long)((double)(ctx->next_leader_slot-ctx->reset_slot)*ctx->slot_duration_ns);
912 :
913 : /* If a prior leader is still in the process of publishing their slot,
914 : delay ours to let them finish ... unless they are so delayed that
915 : we risk getting skipped by the leader following us. 1.2 seconds
916 : is a reasonable default here, although any value between 0 and 1.6
917 : seconds could be considered reasonable. This is arbitrary and
918 : chosen due to intuition. */
919 :
920 0 : if( FD_UNLIKELY( now_ns<expected_start_time_ns+(long)(3.0*ctx->slot_duration_ns) ) ) {
921 : /* If the max_active_descendant is >= next_leader_slot, we waited
922 : too long and a leader after us started publishing to try and skip
923 : us. Just start our leader slot immediately, we might win ... */
924 :
925 0 : if( FD_LIKELY( ctx->max_active_descendant>=ctx->reset_slot && ctx->max_active_descendant<ctx->next_leader_slot ) ) {
926 : /* If one of the leaders between the reset slot and our leader
927 : slot is in the process of publishing (they have a descendant
928 : bank that is in progress of being replayed), then keep waiting.
929 : We probably wouldn't get a leader slot out before they
930 : finished.
931 :
932 : Unless... we are past the deadline to start our slot by more
933 : than 1.2 seconds, in which case we should probably start it to
934 : avoid getting skipped by the leader behind us. */
935 0 : fd_ext_poh_write_unlock();
936 0 : return 0;
937 0 : }
938 0 : }
939 :
940 0 : fd_ext_poh_write_unlock();
941 0 : return 1;
942 0 : }
943 :
944 : CALLED_FROM_RUST static inline void
945 : publish_plugin_slot_start( fd_poh_ctx_t * ctx,
946 : ulong slot,
947 0 : ulong parent_slot ) {
948 0 : if( FD_UNLIKELY( !ctx->plugin_out->mem ) ) return;
949 :
950 0 : fd_plugin_msg_slot_start_t * slot_start = (fd_plugin_msg_slot_start_t *)fd_chunk_to_laddr( ctx->plugin_out->mem, ctx->plugin_out->chunk );
951 0 : *slot_start = (fd_plugin_msg_slot_start_t){ .slot = slot, .parent_slot = parent_slot };
952 0 : fd_stem_publish( ctx->stem, ctx->plugin_out->idx, FD_PLUGIN_MSG_SLOT_START, ctx->plugin_out->chunk, sizeof(fd_plugin_msg_slot_start_t), 0UL, 0UL, 0UL );
953 0 : ctx->plugin_out->chunk = fd_dcache_compact_next( ctx->plugin_out->chunk, sizeof(fd_plugin_msg_slot_start_t), ctx->plugin_out->chunk0, ctx->plugin_out->wmark );
954 0 : }
955 :
956 : CALLED_FROM_RUST static inline void
957 : publish_plugin_slot_end( fd_poh_ctx_t * ctx,
958 : ulong slot,
959 0 : ulong cus_used ) {
960 0 : if( FD_UNLIKELY( !ctx->plugin_out->mem ) ) return;
961 :
962 0 : fd_plugin_msg_slot_end_t * slot_end = (fd_plugin_msg_slot_end_t *)fd_chunk_to_laddr( ctx->plugin_out->mem, ctx->plugin_out->chunk );
963 0 : *slot_end = (fd_plugin_msg_slot_end_t){ .slot = slot, .cus_used = cus_used };
964 0 : fd_stem_publish( ctx->stem, ctx->plugin_out->idx, FD_PLUGIN_MSG_SLOT_END, ctx->plugin_out->chunk, sizeof(fd_plugin_msg_slot_end_t), 0UL, 0UL, 0UL );
965 0 : ctx->plugin_out->chunk = fd_dcache_compact_next( ctx->plugin_out->chunk, sizeof(fd_plugin_msg_slot_end_t), ctx->plugin_out->chunk0, ctx->plugin_out->wmark );
966 0 : }
967 :
968 : extern int
969 : fd_ext_bank_load_account( void const * bank,
970 : int fixed_root,
971 : uchar const * addr,
972 : uchar * owner,
973 : uchar * data,
974 : ulong * data_sz );
975 :
976 : CALLED_FROM_RUST static void
977 : publish_became_leader( fd_poh_ctx_t * ctx,
978 : ulong slot,
979 0 : ulong epoch ) {
980 0 : double tick_per_ns = fd_tempo_tick_per_ns( NULL );
981 0 : fd_histf_sample( ctx->begin_leader_delay, (ulong)((double)(fd_log_wallclock()-ctx->reset_slot_start_ns)/tick_per_ns) );
982 :
983 0 : if( FD_UNLIKELY( ctx->lagged_consecutive_leader_start ) ) {
984 : /* If we are mirroring Agave behavior, the wall clock gets reset
985 : here so we don't count time spent waiting for a bank to freeze
986 : or replay stage to actually start the slot towards our 400ms.
987 :
988 : See extended comments in the config file on this option. */
989 0 : ctx->reset_slot_start_ns = fd_log_wallclock() - (long)((double)(slot-ctx->reset_slot)*ctx->slot_duration_ns);
990 0 : }
991 :
992 0 : fd_bundle_crank_tip_payment_config_t config[1] = { 0 };
993 0 : fd_acct_addr_t tip_receiver_owner[1] = { 0 };
994 :
995 0 : if( FD_UNLIKELY( ctx->bundle.enabled ) ) {
996 0 : long bundle_time = -fd_tickcount();
997 0 : fd_acct_addr_t tip_payment_config[1];
998 0 : fd_acct_addr_t tip_receiver[1];
999 0 : fd_bundle_crank_get_addresses( ctx->bundle.gen, epoch, tip_payment_config, tip_receiver );
1000 :
1001 0 : fd_acct_addr_t _dummy[1];
1002 0 : uchar dummy[1];
1003 :
1004 0 : void const * bank = ctx->current_leader_bank;
1005 :
1006 : /* Calling rust from a C function that is CALLED_FROM_RUST risks
1007 : deadlock. In this case, I checked the load_account function and
1008 : ensured it never calls any C functions that acquire the lock. */
1009 0 : ulong sz1 = sizeof(config), sz2 = 1UL;
1010 0 : int found1 = fd_ext_bank_load_account( bank, 0, tip_payment_config->b, _dummy->b, (uchar *)config, &sz1 );
1011 0 : int found2 = fd_ext_bank_load_account( bank, 0, tip_receiver->b, tip_receiver_owner->b, dummy, &sz2 );
1012 : /* The bundle crank code detects whether the accounts were found by
1013 : whether they have non-zero values (since found and uninitialized
1014 : should be treated the same), so we actually don't really care
1015 : about the value of found{1,2}. */
1016 0 : (void)found1; (void)found2;
1017 0 : bundle_time += fd_tickcount();
1018 0 : fd_histf_sample( ctx->bundle_init_delay, (ulong)bundle_time );
1019 0 : }
1020 :
1021 0 : long slot_start_ns = ctx->reset_slot_start_ns + (long)((double)(slot-ctx->reset_slot)*ctx->slot_duration_ns);
1022 :
1023 : /* No need to check flow control, there are always credits became when we
1024 : are leader, we will not "become" leader again until we are done, so at
1025 : most one frag in flight at a time. */
1026 :
1027 0 : uchar * dst = (uchar *)fd_chunk_to_laddr( ctx->pack_out->mem, ctx->pack_out->chunk );
1028 :
1029 0 : fd_became_leader_t * leader = (fd_became_leader_t *)dst;
1030 0 : leader->slot_start_ns = slot_start_ns;
1031 0 : leader->slot_end_ns = (long)((double)slot_start_ns + ctx->slot_duration_ns);
1032 0 : leader->bank = ctx->current_leader_bank;
1033 0 : leader->max_microblocks_in_slot = ctx->max_microblocks_per_slot;
1034 0 : leader->ticks_per_slot = ctx->ticks_per_slot;
1035 0 : leader->total_skipped_ticks = ctx->ticks_per_slot*(slot-ctx->reset_slot);
1036 0 : leader->epoch = epoch;
1037 0 : leader->bundle->config[0] = config[0];
1038 :
1039 0 : leader->limits.slot_max_cost = ctx->limits.slot_max_cost;
1040 0 : leader->limits.slot_max_vote_cost = ctx->limits.slot_max_vote_cost;
1041 0 : leader->limits.slot_max_write_cost_per_acct = ctx->limits.slot_max_write_cost_per_acct;
1042 :
1043 0 : memcpy( leader->bundle->last_blockhash, ctx->reset_hash, 32UL );
1044 0 : memcpy( leader->bundle->tip_receiver_owner, tip_receiver_owner, 32UL );
1045 :
1046 0 : if( FD_UNLIKELY( leader->ticks_per_slot+leader->total_skipped_ticks>=MAX_SKIPPED_TICKS ) )
1047 0 : FD_LOG_ERR(( "Too many skipped ticks %lu for slot %lu, chain must halt", leader->ticks_per_slot+leader->total_skipped_ticks, slot ));
1048 :
1049 0 : ulong sig = fd_disco_poh_sig( slot, POH_PKT_TYPE_BECAME_LEADER, 0UL );
1050 0 : fd_stem_publish( ctx->stem, ctx->pack_out->idx, sig, ctx->pack_out->chunk, sizeof(fd_became_leader_t), 0UL, 0UL, 0UL );
1051 0 : ctx->pack_out->chunk = fd_dcache_compact_next( ctx->pack_out->chunk, sizeof(fd_became_leader_t), ctx->pack_out->chunk0, ctx->pack_out->wmark );
1052 :
1053 : /* increment refcount for pack's reference to the current leader bank */
1054 0 : if( FD_UNLIKELY( ctx->current_leader_bank ) ) {
1055 0 : ctx->pack_leader_bank = ctx->current_leader_bank;
1056 0 : fd_ext_bank_acquire( ctx->pack_leader_bank );
1057 0 : }
1058 0 : }
1059 :
1060 : /* The PoH tile knows when it should become leader by waiting for its
1061 : leader slot (with the operating system clock). This function is so
1062 : that when it becomes the leader, it can be told what the leader bank
1063 : is by the replay stage. See the notes in the long comment above for
1064 : more on how this works. */
1065 :
1066 : CALLED_FROM_RUST void
1067 : fd_ext_poh_begin_leader( void const * bank,
1068 : ulong slot,
1069 : ulong epoch,
1070 : ulong hashcnt_per_tick,
1071 : ulong cus_block_limit,
1072 : ulong cus_vote_cost_limit,
1073 0 : ulong cus_account_cost_limit ) {
1074 0 : fd_poh_ctx_t * ctx = fd_ext_poh_write_lock();
1075 :
1076 0 : FD_TEST( !ctx->current_leader_bank );
1077 :
1078 0 : if( FD_UNLIKELY( slot!=ctx->slot ) ) FD_LOG_ERR(( "Trying to begin leader slot %lu but we are now on slot %lu", slot, ctx->slot ));
1079 0 : if( FD_UNLIKELY( slot!=ctx->next_leader_slot ) ) FD_LOG_ERR(( "Trying to begin leader slot %lu but next leader slot is %lu", slot, ctx->next_leader_slot ));
1080 :
1081 0 : if( FD_UNLIKELY( ctx->hashcnt_per_tick!=hashcnt_per_tick ) ) {
1082 0 : FD_LOG_WARNING(( "hashes per tick changed from %lu to %lu", ctx->hashcnt_per_tick, hashcnt_per_tick ));
1083 :
1084 : /* Recompute derived information about the clock. */
1085 0 : ctx->hashcnt_duration_ns = (double)ctx->tick_duration_ns/(double)hashcnt_per_tick;
1086 0 : ctx->hashcnt_per_slot = ctx->ticks_per_slot*hashcnt_per_tick;
1087 0 : ctx->hashcnt_per_tick = hashcnt_per_tick;
1088 :
1089 0 : if( FD_UNLIKELY( ctx->hashcnt_per_tick==1UL ) ) {
1090 : /* Low power producer, maximum of one microblock per tick in the slot */
1091 0 : ctx->max_microblocks_per_slot = ctx->ticks_per_slot;
1092 0 : } else {
1093 : /* See the long comment in after_credit for this limit */
1094 0 : ctx->max_microblocks_per_slot = fd_ulong_min( MAX_MICROBLOCKS_PER_SLOT, ctx->ticks_per_slot*(ctx->hashcnt_per_tick-1UL) );
1095 0 : }
1096 :
1097 : /* Discard any ticks we might have done in the interim. They will
1098 : have the wrong number of hashes per tick. We can just catch back
1099 : up quickly if not too many slots were skipped and hopefully
1100 : publish on time. Note that tick production and verification of
1101 : skipped slots is done for the eventual bank that publishes a
1102 : slot, for example:
1103 :
1104 : Reset Slot: 998
1105 : Epoch Transition Slot: 1000
1106 : Leader Slot: 1002
1107 :
1108 : In this case, if a feature changing the hashcnt_per_tick is
1109 : activated in slot 1000, and we are publishing empty ticks for
1110 : slots 998, 999, 1000, and 1001, they should all have the new
1111 : hashes_per_tick number of hashes, rather than the older one, or
1112 : some combination. */
1113 :
1114 0 : FD_TEST( ctx->last_slot==ctx->reset_slot );
1115 0 : FD_TEST( !ctx->last_hashcnt );
1116 0 : ctx->slot = ctx->reset_slot;
1117 0 : ctx->hashcnt = 0UL;
1118 0 : }
1119 :
1120 0 : ctx->current_leader_bank = bank;
1121 0 : ctx->slot_done = 0;
1122 0 : ctx->microblocks_lower_bound = 0UL;
1123 0 : ctx->cus_used = 0UL;
1124 :
1125 0 : ctx->limits.slot_max_cost = cus_block_limit;
1126 0 : ctx->limits.slot_max_vote_cost = cus_vote_cost_limit;
1127 0 : ctx->limits.slot_max_write_cost_per_acct = cus_account_cost_limit;
1128 :
1129 : /* clamp and warn if we are underutilizing CUs */
1130 0 : if( FD_UNLIKELY( ctx->limits.slot_max_cost > FD_PACK_MAX_COST_PER_BLOCK_UPPER_BOUND ) ) {
1131 0 : FD_LOG_WARNING(( "Underutilizing protocol slot CU limit. protocol_limit=%lu validator_limit=%lu", ctx->limits.slot_max_cost, FD_PACK_MAX_COST_PER_BLOCK_UPPER_BOUND ));
1132 0 : ctx->limits.slot_max_cost = FD_PACK_MAX_COST_PER_BLOCK_UPPER_BOUND;
1133 0 : }
1134 0 : if( FD_UNLIKELY( ctx->limits.slot_max_vote_cost > FD_PACK_MAX_VOTE_COST_PER_BLOCK_UPPER_BOUND ) ) {
1135 0 : FD_LOG_WARNING(( "Underutilizing protocol vote CU limit. protocol_limit=%lu validator_limit=%lu", ctx->limits.slot_max_vote_cost, FD_PACK_MAX_VOTE_COST_PER_BLOCK_UPPER_BOUND ));
1136 0 : ctx->limits.slot_max_vote_cost = FD_PACK_MAX_VOTE_COST_PER_BLOCK_UPPER_BOUND;
1137 0 : }
1138 0 : if( FD_UNLIKELY( ctx->limits.slot_max_write_cost_per_acct > FD_PACK_MAX_WRITE_COST_PER_ACCT_UPPER_BOUND ) ) {
1139 0 : FD_LOG_WARNING(( "Underutilizing protocol write CU limit. protocol_limit=%lu validator_limit=%lu", ctx->limits.slot_max_write_cost_per_acct, FD_PACK_MAX_WRITE_COST_PER_ACCT_UPPER_BOUND ));
1140 0 : ctx->limits.slot_max_write_cost_per_acct = FD_PACK_MAX_WRITE_COST_PER_ACCT_UPPER_BOUND;
1141 0 : }
1142 :
1143 : /* We are about to start publishing to the shred tile for this slot
1144 : so update the highwater mark so we never republish in this slot
1145 : again. Also check that the leader slot is greater than the
1146 : highwater, which should have been ensured earlier. */
1147 :
1148 0 : FD_TEST( ctx->highwater_leader_slot==ULONG_MAX || slot>=ctx->highwater_leader_slot );
1149 0 : ctx->highwater_leader_slot = fd_ulong_max( fd_ulong_if( ctx->highwater_leader_slot==ULONG_MAX, 0UL, ctx->highwater_leader_slot ), slot );
1150 :
1151 0 : publish_became_leader( ctx, slot, epoch );
1152 0 : FD_LOG_INFO(( "fd_ext_poh_begin_leader(slot=%lu, highwater_leader_slot=%lu, last_slot=%lu, last_hashcnt=%lu)", slot, ctx->highwater_leader_slot, ctx->last_slot, ctx->last_hashcnt ));
1153 :
1154 0 : fd_ext_poh_write_unlock();
1155 0 : }
1156 :
1157 : /* Determine what the next slot is in the leader schedule is that we are
1158 : leader. Includes the current slot. If we are not leader in what
1159 : remains of the current and next epoch, return ULONG_MAX. */
1160 :
1161 : static inline CALLED_FROM_RUST ulong
1162 0 : next_leader_slot( fd_poh_ctx_t * ctx ) {
1163 : /* If we have published anything in a particular slot, then we
1164 : should never become leader for that slot again. */
1165 0 : ulong min_leader_slot = fd_ulong_max( ctx->slot, fd_ulong_if( ctx->highwater_leader_slot==ULONG_MAX, 0UL, ctx->highwater_leader_slot ) );
1166 0 : return fd_multi_epoch_leaders_get_next_slot( ctx->mleaders, min_leader_slot, &ctx->identity_key );
1167 0 : }
1168 :
1169 : extern int
1170 : fd_ext_admin_rpc_set_identity( uchar const * identity_keypair,
1171 : int require_tower );
1172 :
1173 : static inline int FD_FN_SENSITIVE
1174 : maybe_change_identity( fd_poh_ctx_t * ctx,
1175 0 : int definitely_not_leader ) {
1176 0 : if( FD_UNLIKELY( ctx->halted_switching_key && fd_keyswitch_state_query( ctx->keyswitch )==FD_KEYSWITCH_STATE_UNHALT_PENDING ) ) {
1177 0 : ctx->halted_switching_key = 0;
1178 0 : fd_keyswitch_state( ctx->keyswitch, FD_KEYSWITCH_STATE_COMPLETED );
1179 0 : return 1;
1180 0 : }
1181 :
1182 : /* Cannot change identity while in the middle of a leader slot, else
1183 : poh state machine would become corrupt. */
1184 :
1185 0 : int is_leader = !definitely_not_leader && ctx->next_leader_slot!=ULONG_MAX && ctx->slot>=ctx->next_leader_slot;
1186 0 : if( FD_UNLIKELY( is_leader ) ) return 0;
1187 :
1188 0 : if( FD_UNLIKELY( fd_keyswitch_state_query( ctx->keyswitch )==FD_KEYSWITCH_STATE_SWITCH_PENDING ) ) {
1189 0 : int failed = fd_ext_admin_rpc_set_identity( ctx->keyswitch->bytes, fd_keyswitch_param_query( ctx->keyswitch )==1 );
1190 0 : explicit_bzero( ctx->keyswitch->bytes, 32UL );
1191 0 : FD_COMPILER_MFENCE();
1192 0 : if( FD_UNLIKELY( failed==-1 ) ) {
1193 0 : fd_keyswitch_state( ctx->keyswitch, FD_KEYSWITCH_STATE_FAILED );
1194 0 : return 0;
1195 0 : }
1196 :
1197 0 : memcpy( ctx->identity_key.uc, ctx->keyswitch->bytes+32UL, 32UL );
1198 :
1199 : /* When we switch key, we might have ticked part way through a slot
1200 : that we are now leader in. This violates the contract of the
1201 : tile, that when we become leader, we have not ticked in that slot
1202 : at all. To see why this would be bad, consider the case where we
1203 : have ticked almost to the end, and there isn't enough space left
1204 : to reserve the minimum amount of microblocks needed by pack.
1205 :
1206 : To resolve this, we just reset PoH back to the reset slot, and
1207 : let it try to catch back up quickly. This is OK since the network
1208 : rarely skips. */
1209 0 : ctx->slot = ctx->reset_slot;
1210 0 : ctx->hashcnt = 0UL;
1211 0 : memcpy( ctx->hash, ctx->reset_hash, 32UL );
1212 :
1213 0 : ctx->halted_switching_key = 1;
1214 0 : ctx->keyswitch->result = ctx->shred_seq;
1215 0 : fd_keyswitch_state( ctx->keyswitch, FD_KEYSWITCH_STATE_COMPLETED );
1216 0 : }
1217 :
1218 0 : return 0;
1219 0 : }
1220 :
1221 : static CALLED_FROM_RUST void
1222 0 : no_longer_leader( fd_poh_ctx_t * ctx ) {
1223 0 : if( FD_UNLIKELY( ctx->current_leader_bank ) ) fd_ext_bank_release( ctx->current_leader_bank );
1224 : /* If we stop being leader in a slot, we can never become leader in
1225 : that slot again, and all in-flight microblocks for that slot
1226 : should be dropped. */
1227 0 : ctx->highwater_leader_slot = fd_ulong_max( fd_ulong_if( ctx->highwater_leader_slot==ULONG_MAX, 0UL, ctx->highwater_leader_slot ), ctx->slot );
1228 0 : ctx->current_leader_bank = NULL;
1229 0 : int identity_changed = maybe_change_identity( ctx, 1 );
1230 0 : ctx->next_leader_slot = next_leader_slot( ctx );
1231 0 : if( FD_UNLIKELY( identity_changed ) ) {
1232 0 : FD_LOG_INFO(( "fd_poh_identity_changed(next_leader_slot=%lu)", ctx->next_leader_slot ));
1233 0 : }
1234 :
1235 0 : FD_COMPILER_MFENCE();
1236 0 : fd_ext_poh_signal_leader_change( ctx->signal_leader_change );
1237 0 : FD_LOG_INFO(( "no_longer_leader(next_leader_slot=%lu)", ctx->next_leader_slot ));
1238 0 : }
1239 :
1240 : /* fd_ext_poh_reset is called by the Agave client when a slot on
1241 : the active fork has finished a block and we need to reset our PoH to
1242 : be ticking on top of the block it produced. */
1243 :
1244 : CALLED_FROM_RUST void
1245 : fd_ext_poh_reset( ulong completed_bank_slot, /* The slot that successfully produced a block */
1246 : uchar const * reset_blockhash, /* The hash of the last tick in the produced block */
1247 : ulong hashcnt_per_tick, /* The hashcnt per tick of the bank that completed */
1248 : uchar const * parent_block_id, /* The block id of the parent block */
1249 0 : ulong const * features_activation /* The activation slot of shred-tile features */ ) {
1250 0 : fd_poh_ctx_t * ctx = fd_ext_poh_write_lock();
1251 :
1252 0 : ulong slot_before_reset = ctx->slot;
1253 0 : int leader_before_reset = ctx->slot>=ctx->next_leader_slot;
1254 0 : if( FD_UNLIKELY( leader_before_reset && ctx->current_leader_bank ) ) {
1255 : /* If we were in the middle of a leader slot that we notified pack
1256 : pack to start packing for we can never publish into that slot
1257 : again, mark all in-flight microblocks to be dropped. */
1258 0 : ctx->highwater_leader_slot = fd_ulong_max( fd_ulong_if( ctx->highwater_leader_slot==ULONG_MAX, 0UL, ctx->highwater_leader_slot ), 1UL+ctx->slot );
1259 0 : }
1260 :
1261 0 : ctx->leader_bank_start_ns = fd_log_wallclock(); /* safe to call from Rust */
1262 0 : if( FD_UNLIKELY( ctx->expect_sequential_leader_slot==(completed_bank_slot+1UL) ) ) {
1263 : /* If we are being reset onto a slot, it means some block was fully
1264 : processed, so we reset to build on top of it. Typically we want
1265 : to update the reset_slot_start_ns to the current time, because
1266 : the network will give the next leader 400ms to publish,
1267 : regardless of how long the prior leader took.
1268 :
1269 : But: if we were leader in the prior slot, and the block was our
1270 : own we can do better. We know that the next slot should start
1271 : exactly 400ms after the prior one started, so we can use that as
1272 : the reset slot start time instead. */
1273 0 : ctx->reset_slot_start_ns = ctx->reset_slot_start_ns + (long)((double)((completed_bank_slot+1UL)-ctx->reset_slot)*ctx->slot_duration_ns);
1274 0 : } else {
1275 0 : ctx->reset_slot_start_ns = ctx->leader_bank_start_ns;
1276 0 : }
1277 0 : ctx->expect_sequential_leader_slot = ULONG_MAX;
1278 :
1279 0 : memcpy( ctx->reset_hash, reset_blockhash, 32UL );
1280 0 : memcpy( ctx->hash, reset_blockhash, 32UL );
1281 0 : if( FD_LIKELY( parent_block_id!=NULL ) ) {
1282 0 : ctx->parent_slot = completed_bank_slot;
1283 0 : memcpy( ctx->parent_block_id, parent_block_id, 32UL );
1284 0 : }
1285 0 : ctx->slot = completed_bank_slot+1UL;
1286 0 : ctx->hashcnt = 0UL;
1287 0 : ctx->last_slot = ctx->slot;
1288 0 : ctx->last_hashcnt = 0UL;
1289 0 : ctx->reset_slot = ctx->slot;
1290 :
1291 0 : if( FD_UNLIKELY( ctx->hashcnt_per_tick!=hashcnt_per_tick ) ) {
1292 0 : FD_LOG_WARNING(( "hashes per tick changed from %lu to %lu", ctx->hashcnt_per_tick, hashcnt_per_tick ));
1293 :
1294 : /* Recompute derived information about the clock. */
1295 0 : ctx->hashcnt_duration_ns = (double)ctx->tick_duration_ns/(double)hashcnt_per_tick;
1296 0 : ctx->hashcnt_per_slot = ctx->ticks_per_slot*hashcnt_per_tick;
1297 0 : ctx->hashcnt_per_tick = hashcnt_per_tick;
1298 :
1299 0 : if( FD_UNLIKELY( ctx->hashcnt_per_tick==1UL ) ) {
1300 : /* Low power producer, maximum of one microblock per tick in the slot */
1301 0 : ctx->max_microblocks_per_slot = ctx->ticks_per_slot;
1302 0 : } else {
1303 : /* See the long comment in after_credit for this limit */
1304 0 : ctx->max_microblocks_per_slot = fd_ulong_min( MAX_MICROBLOCKS_PER_SLOT, ctx->ticks_per_slot*(ctx->hashcnt_per_tick-1UL) );
1305 0 : }
1306 0 : }
1307 :
1308 : /* When we reset, we need to allow PoH to tick freely again rather
1309 : than being constrained. If we are leader after the reset, this
1310 : is OK because we won't tick until we get a bank, and the lower
1311 : bound will be reset with the value from the bank. */
1312 0 : ctx->microblocks_lower_bound = ctx->max_microblocks_per_slot;
1313 :
1314 0 : if( FD_UNLIKELY( leader_before_reset ) ) {
1315 : /* No longer have a leader bank if we are reset. Replay stage will
1316 : call back again to give us a new one if we should become leader
1317 : for the reset slot.
1318 :
1319 : The order is important here, ctx->hashcnt must be updated before
1320 : calling no_longer_leader. */
1321 0 : no_longer_leader( ctx );
1322 0 : }
1323 0 : ctx->next_leader_slot = next_leader_slot( ctx );
1324 0 : FD_LOG_INFO(( "fd_ext_poh_reset(slot=%lu,next_leader_slot=%lu)", ctx->reset_slot, ctx->next_leader_slot ));
1325 :
1326 0 : if( FD_UNLIKELY( ctx->slot>=ctx->next_leader_slot ) ) {
1327 : /* We are leader after the reset... two cases: */
1328 0 : if( FD_LIKELY( ctx->slot==slot_before_reset ) ) {
1329 : /* 1. We are reset onto the same slot we are already leader on.
1330 : This is a common case when we have two leader slots in a
1331 : row, replay stage will reset us to our own slot. No need to
1332 : do anything here, we already sent a SLOT_START. */
1333 0 : FD_TEST( leader_before_reset );
1334 0 : } else {
1335 : /* 2. We are reset onto a different slot. If we were leader
1336 : before, we should first end that slot, then begin the new
1337 : one if we are newly leader now. */
1338 0 : if( FD_LIKELY( leader_before_reset ) ) publish_plugin_slot_end( ctx, slot_before_reset, ctx->cus_used );
1339 0 : else publish_plugin_slot_start( ctx, ctx->next_leader_slot, ctx->reset_slot );
1340 0 : }
1341 0 : } else {
1342 0 : if( FD_UNLIKELY( leader_before_reset ) ) publish_plugin_slot_end( ctx, slot_before_reset, ctx->cus_used );
1343 0 : }
1344 :
1345 : /* There is a subset of FD_SHRED_FEATURES_ACTIVATION_... slots that
1346 : the shred tile needs to be aware of. Since their computation
1347 : requires the bank, we are forced (so far) to receive them here
1348 : from the Rust side, before forwarding them to the shred tile as
1349 : POH_PKT_TYPE_FEAT_ACT_SLOT. This is not elegant, and it should
1350 : be revised in the future (TODO), but it provides a "temporary"
1351 : working solution to handle features activation. */
1352 0 : fd_memcpy( ctx->features_activation->slots, features_activation, sizeof(fd_shred_features_activation_t) );
1353 0 : ctx->features_activation_avail = 1UL;
1354 :
1355 0 : fd_ext_poh_write_unlock();
1356 0 : }
1357 :
1358 : /* Since it can't easily return an Option<Pubkey>, return 1 for Some and
1359 : 0 for None. */
1360 : CALLED_FROM_RUST int
1361 : fd_ext_poh_get_leader_after_n_slots( ulong n,
1362 0 : uchar out_pubkey[ static 32 ] ) {
1363 0 : fd_poh_ctx_t * ctx = fd_ext_poh_write_lock();
1364 0 : ulong slot = ctx->slot + n;
1365 0 : fd_pubkey_t const * leader = fd_multi_epoch_leaders_get_leader_for_slot( ctx->mleaders, slot );
1366 :
1367 0 : int copied = 0;
1368 0 : if( FD_LIKELY( leader ) ) {
1369 0 : memcpy( out_pubkey, leader, 32UL );
1370 0 : copied = 1;
1371 0 : }
1372 0 : fd_ext_poh_write_unlock();
1373 0 : return copied;
1374 0 : }
1375 :
1376 : FD_FN_CONST static inline ulong
1377 0 : scratch_align( void ) {
1378 0 : return 128UL;
1379 0 : }
1380 :
1381 : FD_FN_PURE static inline ulong
1382 0 : scratch_footprint( fd_topo_tile_t const * tile ) {
1383 0 : (void)tile;
1384 0 : ulong l = FD_LAYOUT_INIT;
1385 0 : l = FD_LAYOUT_APPEND( l, alignof( fd_poh_ctx_t ), sizeof( fd_poh_ctx_t ) );
1386 0 : l = FD_LAYOUT_APPEND( l, FD_SHA256_ALIGN, FD_SHA256_FOOTPRINT );
1387 0 : return FD_LAYOUT_FINI( l, scratch_align() );
1388 0 : }
1389 :
1390 : static void
1391 : publish_tick( fd_poh_ctx_t * ctx,
1392 : fd_stem_context_t * stem,
1393 : uchar hash[ static 32 ],
1394 0 : int is_skipped ) {
1395 0 : ulong hashcnt = ctx->hashcnt_per_tick*(1UL+(ctx->last_hashcnt/ctx->hashcnt_per_tick));
1396 :
1397 0 : uchar * dst = (uchar *)fd_chunk_to_laddr( ctx->shred_out->mem, ctx->shred_out->chunk );
1398 :
1399 0 : FD_TEST( ctx->last_slot>=ctx->reset_slot );
1400 0 : fd_entry_batch_meta_t * meta = (fd_entry_batch_meta_t *)dst;
1401 0 : if( FD_UNLIKELY( is_skipped ) ) {
1402 : /* We are publishing ticks for a skipped slot, the reference tick
1403 : and block complete flags should always be zero. */
1404 0 : meta->reference_tick = 0UL;
1405 0 : meta->block_complete = 0;
1406 0 : } else {
1407 0 : meta->reference_tick = hashcnt/ctx->hashcnt_per_tick;
1408 0 : meta->block_complete = hashcnt==ctx->hashcnt_per_slot;
1409 0 : }
1410 :
1411 0 : ulong slot = fd_ulong_if( meta->block_complete, ctx->slot-1UL, ctx->slot );
1412 0 : meta->parent_offset = 1UL+slot-ctx->reset_slot;
1413 :
1414 : /* From poh_reset we received the block_id for ctx->parent_slot.
1415 : Now we're telling shred tile to build on parent: (slot-meta->parent_offset).
1416 : The block_id that we're passing is valid iff the two are the same,
1417 : i.e. ctx->parent_slot == (slot-meta->parent_offset). */
1418 0 : meta->parent_block_id_valid = ctx->parent_slot == (slot-meta->parent_offset);
1419 0 : if( FD_LIKELY( meta->parent_block_id_valid ) ) {
1420 0 : fd_memcpy( meta->parent_block_id, ctx->parent_block_id, 32UL );
1421 0 : }
1422 :
1423 0 : FD_TEST( hashcnt>ctx->last_hashcnt );
1424 0 : ulong hash_delta = hashcnt-ctx->last_hashcnt;
1425 :
1426 0 : dst += sizeof(fd_entry_batch_meta_t);
1427 0 : fd_entry_batch_header_t * tick = (fd_entry_batch_header_t *)dst;
1428 0 : tick->hashcnt_delta = hash_delta;
1429 0 : fd_memcpy( tick->hash, hash, 32UL );
1430 0 : tick->txn_cnt = 0UL;
1431 :
1432 0 : ulong tspub = (ulong)fd_frag_meta_ts_comp( fd_tickcount() );
1433 0 : ulong sz = sizeof(fd_entry_batch_meta_t)+sizeof(fd_entry_batch_header_t);
1434 0 : ulong sig = fd_disco_poh_sig( slot, POH_PKT_TYPE_MICROBLOCK, 0UL );
1435 0 : fd_stem_publish( stem, ctx->shred_out->idx, sig, ctx->shred_out->chunk, sz, 0UL, 0UL, tspub );
1436 0 : ctx->shred_seq = stem->seqs[ ctx->shred_out->idx ];
1437 0 : ctx->shred_out->chunk = fd_dcache_compact_next( ctx->shred_out->chunk, sz, ctx->shred_out->chunk0, ctx->shred_out->wmark );
1438 :
1439 0 : if( FD_UNLIKELY( hashcnt==ctx->hashcnt_per_slot ) ) {
1440 0 : ctx->last_slot++;
1441 0 : ctx->last_hashcnt = 0UL;
1442 0 : } else {
1443 0 : ctx->last_hashcnt = hashcnt;
1444 0 : }
1445 0 : }
1446 :
1447 : static inline void
1448 : publish_features_activation( fd_poh_ctx_t * ctx,
1449 0 : fd_stem_context_t * stem ) {
1450 0 : uchar * dst = (uchar *)fd_chunk_to_laddr( ctx->shred_out->mem, ctx->shred_out->chunk );
1451 0 : fd_shred_features_activation_t * act_data = (fd_shred_features_activation_t *)dst;
1452 0 : fd_memcpy( act_data, ctx->features_activation, sizeof(fd_shred_features_activation_t) );
1453 :
1454 0 : ulong tspub = (ulong)fd_frag_meta_ts_comp( fd_tickcount() );
1455 0 : ulong sz = sizeof(fd_shred_features_activation_t);
1456 0 : ulong sig = fd_disco_poh_sig( ctx->slot, POH_PKT_TYPE_FEAT_ACT_SLOT, 0UL );
1457 0 : fd_stem_publish( stem, ctx->shred_out->idx, sig, ctx->shred_out->chunk, sz, 0UL, 0UL, tspub );
1458 0 : ctx->shred_seq = stem->seqs[ ctx->shred_out->idx ];
1459 0 : ctx->shred_out->chunk = fd_dcache_compact_next( ctx->shred_out->chunk, sz, ctx->shred_out->chunk0, ctx->shred_out->wmark );
1460 0 : }
1461 :
1462 : static inline void
1463 : after_credit( fd_poh_ctx_t * ctx,
1464 : fd_stem_context_t * stem,
1465 : int * opt_poll_in,
1466 0 : int * charge_busy ) {
1467 0 : ctx->stem = stem;
1468 :
1469 0 : FD_COMPILER_MFENCE();
1470 0 : if( FD_UNLIKELY( fd_poh_waiting_lock ) ) {
1471 0 : FD_VOLATILE( fd_poh_returned_lock ) = 1UL;
1472 0 : FD_COMPILER_MFENCE();
1473 0 : for(;;) {
1474 0 : if( FD_UNLIKELY( !FD_VOLATILE_CONST( fd_poh_returned_lock ) ) ) break;
1475 0 : FD_SPIN_PAUSE();
1476 0 : }
1477 0 : FD_COMPILER_MFENCE();
1478 0 : FD_VOLATILE( fd_poh_waiting_lock ) = 0UL;
1479 0 : *opt_poll_in = 0;
1480 0 : *charge_busy = 1;
1481 0 : return;
1482 0 : }
1483 0 : FD_COMPILER_MFENCE();
1484 :
1485 0 : if( FD_UNLIKELY( ctx->features_activation_avail ) ) {
1486 : /* If we have received an update on features_activation, then
1487 : forward them to the shred tile. In principle, this should
1488 : happen at most once per slot. */
1489 0 : publish_features_activation( ctx, stem );
1490 0 : ctx->features_activation_avail = 0UL;
1491 0 : }
1492 :
1493 0 : int is_leader = ctx->next_leader_slot!=ULONG_MAX && ctx->slot>=ctx->next_leader_slot;
1494 0 : if( FD_UNLIKELY( is_leader && !ctx->current_leader_bank ) ) {
1495 : /* If we are the leader, but we didn't yet learn what the leader
1496 : bank object is from the replay stage, do not do any hashing.
1497 :
1498 : This is not ideal, but greatly simplifies the control flow. */
1499 0 : return;
1500 0 : }
1501 :
1502 : /* If we have skipped ticks pending because we skipped some slots to
1503 : become leader, register them now one at a time. */
1504 0 : if( FD_UNLIKELY( is_leader && ctx->last_slot<ctx->slot ) ) {
1505 0 : ulong publish_hashcnt = ctx->last_hashcnt+ctx->hashcnt_per_tick;
1506 0 : ulong tick_idx = (ctx->last_slot*ctx->ticks_per_slot+publish_hashcnt/ctx->hashcnt_per_tick)%MAX_SKIPPED_TICKS;
1507 :
1508 0 : fd_ext_poh_register_tick( ctx->current_leader_bank, ctx->skipped_tick_hashes[ tick_idx ] );
1509 0 : publish_tick( ctx, stem, ctx->skipped_tick_hashes[ tick_idx ], 1 );
1510 :
1511 : /* If we are catching up now and publishing a bunch of skipped
1512 : ticks, we do not want to process any incoming microblocks until
1513 : all the skipped ticks have been published out; otherwise we would
1514 : intersperse skipped tick messages with microblocks. */
1515 0 : *opt_poll_in = 0;
1516 0 : *charge_busy = 1;
1517 0 : return;
1518 0 : }
1519 :
1520 0 : int low_power_mode = ctx->hashcnt_per_tick==1UL;
1521 :
1522 : /* If we are the leader, always leave enough capacity in the slot so
1523 : that we can mixin any potential microblocks still coming from the
1524 : pack tile for this slot. */
1525 0 : ulong max_remaining_microblocks = ctx->max_microblocks_per_slot - ctx->microblocks_lower_bound;
1526 :
1527 : /* We don't want to tick over (finish) the slot until pack tell us
1528 : it's done. If we're waiting on pack, then we clamp to [0, 1] */
1529 0 : if( FD_LIKELY( !ctx->slot_done && is_leader ) ) max_remaining_microblocks = fd_ulong_max( fd_ulong_min( 1UL, max_remaining_microblocks ), max_remaining_microblocks );
1530 :
1531 : /* With hashcnt_per_tick hashes per tick, we actually get
1532 : hashcnt_per_tick-1 chances to mixin a microblock. For each tick
1533 : span that we need to reserve, we also need to reserve the hashcnt
1534 : for the tick, hence the +
1535 : max_remaining_microblocks/(hashcnt_per_tick-1) rounded up.
1536 :
1537 : However, if hashcnt_per_tick is 1 because we're in low power mode,
1538 : this should probably just be max_remaining_microblocks. */
1539 0 : ulong max_remaining_ticks_or_microblocks = max_remaining_microblocks;
1540 0 : if( FD_LIKELY( !low_power_mode ) ) max_remaining_ticks_or_microblocks += (max_remaining_microblocks+ctx->hashcnt_per_tick-2UL)/(ctx->hashcnt_per_tick-1UL);
1541 :
1542 0 : ulong restricted_hashcnt = fd_ulong_if( ctx->hashcnt_per_slot>=max_remaining_ticks_or_microblocks, ctx->hashcnt_per_slot-max_remaining_ticks_or_microblocks, 0UL );
1543 :
1544 0 : ulong min_hashcnt = ctx->hashcnt;
1545 :
1546 0 : if( FD_LIKELY( !low_power_mode ) ) {
1547 : /* Recall that there are two kinds of events that will get published
1548 : to the shredder,
1549 :
1550 : (a) Ticks. These occur every 62,500 (hashcnt_per_tick) hashcnts,
1551 : and there will be 64 (ticks_per_slot) of them in each slot.
1552 :
1553 : Ticks must not have any transactions mixed into the hash.
1554 : This is not strictly needed in theory, but is required by the
1555 : current consensus protocol. They get published here in
1556 : after_credit.
1557 :
1558 : (b) Microblocks. These can occur at any other hashcnt, as long
1559 : as it is not a tick. Microblocks cannot be empty, and must
1560 : have at least one transactions mixed in. These get
1561 : published in after_frag.
1562 :
1563 : If hashcnt_per_tick is 1, then we are in low power mode and the
1564 : following does not apply, since we can mix in transactions at any
1565 : time.
1566 :
1567 : In the normal, non-low-power mode, though, we have to be careful
1568 : to make sure that we do not publish microblocks on tick
1569 : boundaries. To do that, we need to obey two rules:
1570 : (i) after_credit must not leave hashcnt one before a tick
1571 : boundary
1572 : (ii) if after_credit begins one before a tick boundary, it must
1573 : advance hashcnt and publish the tick
1574 :
1575 : There's some interplay between min_hashcnt and restricted_hashcnt
1576 : here, and we need to show that there's always a value of
1577 : target_hashcnt we can pick such that
1578 : min_hashcnt <= target_hashcnt <= restricted_hashcnt.
1579 : We'll prove this by induction for current_slot==0 and
1580 : is_leader==true, since all other slots should be the same.
1581 :
1582 : Let m_j and r_j be the min_hashcnt and restricted_hashcnt
1583 : (respectively) for the jth call to after_credit in a slot. We
1584 : want to show that for all values of j, it's possible to pick a
1585 : value h_j, the value of target_hashcnt for the jth call to
1586 : after_credit (which is also the value of hashcnt after
1587 : after_credit has completed) such that m_j<=h_j<=r_j.
1588 :
1589 : Additionally, let T be hashcnt_per_tick and N be ticks_per_slot.
1590 :
1591 : Starting with the base case, j==0. m_j=0, and
1592 : r_0 = N*T - max_microblocks_per_slot
1593 : - ceil(max_microblocks_per_slot/(T-1)).
1594 :
1595 : This is monotonic decreasing in max_microblocks_per_slot, so it
1596 : achieves its minimum when max_microblocks_per_slot is its
1597 : maximum.
1598 : r_0 >= N*T - N*(T-1) - ceil( (N*(T-1))/(T-1))
1599 : = N*T - N*(T-1)-N = 0.
1600 : Thus, m_0 <= r_0, as desired.
1601 :
1602 :
1603 :
1604 : Then, for the inductive step, assume there exists h_j such that
1605 : m_j<=h_j<=r_j, and we want to show that there exists h_{j+1},
1606 : which is the same as showing m_{j+1}<=r_{j+1}.
1607 :
1608 : Let a_j be 1 if we had a microblock immediately following the jth
1609 : call to after_credit, and 0 otherwise. Then hashcnt at the start
1610 : of the (j+1)th call to after_frag is h_j+a_j.
1611 : Also, set b_{j+1}=1 if we are in the case covered by rule (ii)
1612 : above during the (j+1)th call to after_credit, i.e. if
1613 : (h_j+a_j)%T==T-1. Thus, m_{j+1} = h_j + a_j + b_{j+1}.
1614 :
1615 : If we received an additional microblock, then
1616 : max_remaining_microblocks goes down by 1, and
1617 : max_remaining_ticks_or_microblocks goes down by either 1 or 2,
1618 : which means restricted_hashcnt goes up by either 1 or 2. In
1619 : particular, it goes up by 2 if the new value of
1620 : max_remaining_microblocks (at the start of the (j+1)th call to
1621 : after_credit) is congruent to 0 mod T-1. Let b'_{j+1} be 1 if
1622 : this condition is met and 0 otherwise. If we receive a
1623 : done_packing message, restricted_hashcnt can go up by more, but
1624 : we can ignore that case, since it is less restrictive.
1625 : Thus, r_{j+1}=r_j+a_j+b'_{j+1}.
1626 :
1627 : If h_j < r_j (strictly less), then h_j+a_j < r_j+a_j. And thus,
1628 : since b_{j+1}<=b'_{j+1}+1, just by virtue of them both being
1629 : binary,
1630 : h_j + a_j + b_{j+1} < r_j + a_j + b'_{j+1} + 1,
1631 : which is the same (for integers) as
1632 : h_j + a_j + b_{j+1} <= r_j + a_j + b'_{j+1},
1633 : m_{j+1} <= r_{j+1}
1634 :
1635 : On the other hand, if h_j==r_j, this is easy unless b_{j+1}==1,
1636 : which can also only happen if a_j==1. Then (h_j+a_j)%T==T-1,
1637 : which means there's an integer k such that
1638 :
1639 : h_j+a_j==(ticks_per_slot-k)*T-1
1640 : h_j ==ticks_per_slot*T - k*(T-1)-1 - k-1
1641 : ==ticks_per_slot*T - (k*(T-1)+1) - ceil( (k*(T-1)+1)/(T-1) )
1642 :
1643 : Since h_j==r_j in this case, and
1644 : r_j==(ticks_per_slot*T) - max_remaining_microblocks_j - ceil(max_remaining_microblocks_j/(T-1)),
1645 : we can see that the value of max_remaining_microblocks at the
1646 : start of the jth call to after_credit is k*(T-1)+1. Again, since
1647 : a_j==1, then the value of max_remaining_microblocks at the start
1648 : of the j+1th call to after_credit decreases by 1 to k*(T-1),
1649 : which means b'_{j+1}=1.
1650 :
1651 : Thus, h_j + a_j + b_{j+1} == r_j + a_j + b'_{j+1}, so, in
1652 : particular, h_{j+1}<=r_{j+1} as desired. */
1653 0 : min_hashcnt += (ulong)(min_hashcnt%ctx->hashcnt_per_tick == (ctx->hashcnt_per_tick-1UL)); /* add b_{j+1}, enforcing rule (ii) */
1654 0 : }
1655 : /* Now figure out how many hashes are needed to "catch up" the hash
1656 : count to the current system clock, and clamp it to the allowed
1657 : range. */
1658 0 : long now = fd_log_wallclock();
1659 0 : ulong target_hashcnt;
1660 0 : if( FD_LIKELY( !is_leader ) ) {
1661 0 : target_hashcnt = (ulong)((double)(now - ctx->reset_slot_start_ns) / ctx->hashcnt_duration_ns) - (ctx->slot-ctx->reset_slot)*ctx->hashcnt_per_slot;
1662 0 : } else {
1663 : /* We might have gotten very behind on hashes, but if we are leader
1664 : we want to catch up gradually over the remainder of our leader
1665 : slot, not all at once right now. This helps keep the tile from
1666 : being oversubscribed and taking a long time to process incoming
1667 : microblocks. */
1668 0 : long expected_slot_start_ns = ctx->reset_slot_start_ns + (long)((double)(ctx->slot-ctx->reset_slot)*ctx->slot_duration_ns);
1669 0 : double actual_slot_duration_ns = ctx->slot_duration_ns<(double)(ctx->leader_bank_start_ns - expected_slot_start_ns) ? 0.0 : ctx->slot_duration_ns - (double)(ctx->leader_bank_start_ns - expected_slot_start_ns);
1670 0 : double actual_hashcnt_duration_ns = actual_slot_duration_ns / (double)ctx->hashcnt_per_slot;
1671 0 : target_hashcnt = fd_ulong_if( actual_hashcnt_duration_ns==0.0, restricted_hashcnt, (ulong)((double)(now - ctx->leader_bank_start_ns) / actual_hashcnt_duration_ns) );
1672 0 : }
1673 : /* Clamp to [min_hashcnt, restricted_hashcnt] as above */
1674 0 : target_hashcnt = fd_ulong_max( fd_ulong_min( target_hashcnt, restricted_hashcnt ), min_hashcnt );
1675 :
1676 : /* The above proof showed that it was always possible to pick a value
1677 : of target_hashcnt, but we still have a lot of freedom in how to
1678 : pick it. It simplifies the code a lot if we don't keep going after
1679 : a tick in this function. In particular, we want to publish at most
1680 : 1 tick in this call, since otherwise we could consume infinite
1681 : credits to publish here. The credits are set so that we should
1682 : only ever publish one tick during this loop. Also, all the extra
1683 : stuff (leader transitions, publishing ticks, etc.) we have to do
1684 : happens at tick boundaries, so this lets us consolidate all those
1685 : cases.
1686 :
1687 : Mathematically, since the current value of hashcnt is h_j+a_j, the
1688 : next tick (advancing a full tick if we're currently at a tick) is
1689 : t_{j+1} = T*(floor( (h_j+a_j)/T )+1). We need to show that if we set
1690 : h'_{j+1} = min( h_{j+1}, t_{j+1} ), it is still valid.
1691 :
1692 : First, h'_{j+1} <= h_{j+1} <= r_{j+1}, so we're okay in that
1693 : direction.
1694 :
1695 : Next, observe that t_{j+1}>=h_j + a_j + 1, and recall that b_{j+1}
1696 : is 0 or 1. So then,
1697 : t_{j+1} >= h_j+a_j+b_{j+1} = m_{j+1}.
1698 :
1699 : We know h_{j+1) >= m_{j+1} from before, so then h'_{j+1} >=
1700 : m_{j+1}, as desired. */
1701 :
1702 0 : ulong next_tick_hashcnt = ctx->hashcnt_per_tick * (1UL+(ctx->hashcnt/ctx->hashcnt_per_tick));
1703 0 : target_hashcnt = fd_ulong_min( target_hashcnt, next_tick_hashcnt );
1704 :
1705 : /* We still need to enforce rule (i). We know that min_hashcnt%T !=
1706 : T-1 because of rule (ii). That means that if target_hashcnt%T ==
1707 : T-1 at this point, target_hashcnt > min_hashcnt (notice the
1708 : strict), so target_hashcnt-1 >= min_hashcnt and is thus still a
1709 : valid choice for target_hashcnt. */
1710 0 : target_hashcnt -= (ulong)( (!low_power_mode) & ((target_hashcnt%ctx->hashcnt_per_tick)==(ctx->hashcnt_per_tick-1UL)) );
1711 :
1712 0 : FD_TEST( target_hashcnt >= ctx->hashcnt );
1713 0 : FD_TEST( target_hashcnt >= min_hashcnt );
1714 0 : FD_TEST( target_hashcnt <= restricted_hashcnt );
1715 :
1716 0 : if( FD_UNLIKELY( ctx->hashcnt==target_hashcnt ) ) return; /* Nothing to do, don't publish a tick twice */
1717 :
1718 0 : *charge_busy = 1;
1719 :
1720 0 : if( FD_LIKELY( ctx->hashcnt<target_hashcnt ) ) {
1721 0 : fd_sha256_hash_32_repeated( ctx->hash, ctx->hash, target_hashcnt-ctx->hashcnt );
1722 0 : ctx->hashcnt = target_hashcnt;
1723 0 : }
1724 :
1725 0 : if( FD_UNLIKELY( ctx->hashcnt==ctx->hashcnt_per_slot ) ) {
1726 0 : ctx->slot++;
1727 0 : ctx->hashcnt = 0UL;
1728 0 : }
1729 :
1730 0 : if( FD_UNLIKELY( !is_leader && !(ctx->hashcnt%ctx->hashcnt_per_tick ) ) ) {
1731 : /* We finished a tick while not leader... save the current hash so
1732 : it can be played back into the bank when we become the leader. */
1733 0 : ulong tick_idx = (ctx->slot*ctx->ticks_per_slot+ctx->hashcnt/ctx->hashcnt_per_tick)%MAX_SKIPPED_TICKS;
1734 0 : fd_memcpy( ctx->skipped_tick_hashes[ tick_idx ], ctx->hash, 32UL );
1735 :
1736 0 : ulong initial_tick_idx = (ctx->last_slot*ctx->ticks_per_slot+ctx->last_hashcnt/ctx->hashcnt_per_tick)%MAX_SKIPPED_TICKS;
1737 0 : if( FD_UNLIKELY( tick_idx==initial_tick_idx ) ) FD_LOG_ERR(( "Too many skipped ticks from slot %lu to slot %lu, chain must halt", ctx->last_slot, ctx->slot ));
1738 0 : }
1739 :
1740 0 : if( FD_UNLIKELY( is_leader && !(ctx->hashcnt%ctx->hashcnt_per_tick) ) ) {
1741 : /* We ticked while leader... tell the leader bank. */
1742 0 : fd_ext_poh_register_tick( ctx->current_leader_bank, ctx->hash );
1743 :
1744 : /* And send an empty microblock (a tick) to the shred tile. */
1745 0 : publish_tick( ctx, stem, ctx->hash, 0 );
1746 0 : }
1747 :
1748 0 : if( FD_UNLIKELY( !is_leader && ctx->slot>=ctx->next_leader_slot ) ) {
1749 : /* We ticked while not leader and are now leader... transition
1750 : the state machine. */
1751 0 : publish_plugin_slot_start( ctx, ctx->next_leader_slot, ctx->reset_slot );
1752 0 : FD_LOG_INFO(( "fd_poh_ticked_into_leader(slot=%lu, reset_slot=%lu)", ctx->next_leader_slot, ctx->reset_slot ));
1753 0 : }
1754 :
1755 0 : if( FD_UNLIKELY( is_leader && ctx->slot>ctx->next_leader_slot ) ) {
1756 : /* We ticked while leader and are no longer leader... transition
1757 : the state machine. */
1758 0 : FD_TEST( !max_remaining_microblocks );
1759 0 : publish_plugin_slot_end( ctx, ctx->next_leader_slot, ctx->cus_used );
1760 0 : FD_LOG_INFO(( "fd_poh_ticked_outof_leader(slot=%lu)", ctx->next_leader_slot ));
1761 :
1762 0 : no_longer_leader( ctx );
1763 0 : ctx->expect_sequential_leader_slot = ctx->slot;
1764 :
1765 0 : double tick_per_ns = fd_tempo_tick_per_ns( NULL );
1766 0 : fd_histf_sample( ctx->slot_done_delay, (ulong)((double)(fd_log_wallclock()-ctx->reset_slot_start_ns)/tick_per_ns) );
1767 0 : ctx->next_leader_slot = next_leader_slot( ctx );
1768 :
1769 0 : if( FD_UNLIKELY( ctx->slot>=ctx->next_leader_slot ) ) {
1770 : /* We finished a leader slot, and are immediately leader for the
1771 : following slot... transition. */
1772 0 : publish_plugin_slot_start( ctx, ctx->next_leader_slot, ctx->next_leader_slot-1UL );
1773 0 : FD_LOG_INFO(( "fd_poh_ticked_into_leader(slot=%lu, reset_slot=%lu)", ctx->next_leader_slot, ctx->next_leader_slot-1UL ));
1774 0 : }
1775 0 : }
1776 0 : }
1777 :
1778 : static inline void
1779 0 : during_housekeeping( fd_poh_ctx_t * ctx ) {
1780 0 : if( FD_UNLIKELY( maybe_change_identity( ctx, 0 ) ) ) {
1781 0 : ctx->next_leader_slot = next_leader_slot( ctx );
1782 0 : FD_LOG_INFO(( "fd_poh_identity_changed(next_leader_slot=%lu)", ctx->next_leader_slot ));
1783 :
1784 : /* Signal replay to check if we are leader again, in-case it's stuck
1785 : because everything already replayed. */
1786 0 : FD_COMPILER_MFENCE();
1787 0 : fd_ext_poh_signal_leader_change( ctx->signal_leader_change );
1788 0 : }
1789 0 : }
1790 :
1791 : static inline void
1792 0 : metrics_write( fd_poh_ctx_t * ctx ) {
1793 0 : FD_MHIST_COPY( POH, BEGIN_LEADER_DELAY_SECONDS, ctx->begin_leader_delay );
1794 0 : FD_MHIST_COPY( POH, FIRST_MICROBLOCK_DELAY_SECONDS, ctx->first_microblock_delay );
1795 0 : FD_MHIST_COPY( POH, SLOT_DONE_DELAY_SECONDS, ctx->slot_done_delay );
1796 0 : FD_MHIST_COPY( POH, BUNDLE_INITIALIZE_DELAY_SECONDS, ctx->bundle_init_delay );
1797 0 : }
1798 :
1799 : static int
1800 : before_frag( fd_poh_ctx_t * ctx,
1801 : ulong in_idx,
1802 : ulong seq,
1803 0 : ulong sig ) {
1804 0 : (void)seq;
1805 :
1806 0 : if( FD_LIKELY( ctx->in_kind[ in_idx ]!=IN_KIND_BANK && ctx->in_kind[ in_idx ]!=IN_KIND_PACK ) ) return 0;
1807 :
1808 0 : if( FD_UNLIKELY( sig==ULONG_MAX ) ) {
1809 : /* Banks are drained, release pack's owenership of the current bank */
1810 0 : if( FD_UNLIKELY( ctx->pack_leader_bank ) ) fd_ext_bank_release( ctx->pack_leader_bank );
1811 0 : ctx->pack_leader_bank = NULL;
1812 0 : return 1; /* discard */
1813 0 : }
1814 :
1815 0 : uint pack_idx = (uint)fd_disco_bank_sig_pack_idx( sig );
1816 0 : FD_TEST( ((int)(pack_idx-ctx->expect_pack_idx))>=0L );
1817 0 : if( FD_UNLIKELY( pack_idx!=ctx->expect_pack_idx ) ) return -1;
1818 0 : ctx->expect_pack_idx++;
1819 :
1820 0 : return 0;
1821 0 : }
1822 :
1823 : static inline void
1824 : during_frag( fd_poh_ctx_t * ctx,
1825 : ulong in_idx,
1826 : ulong seq FD_PARAM_UNUSED,
1827 : ulong sig,
1828 : ulong chunk,
1829 : ulong sz,
1830 0 : ulong ctl FD_PARAM_UNUSED ) {
1831 0 : ctx->skip_frag = 0;
1832 :
1833 0 : if( FD_UNLIKELY( ctx->in_kind[ in_idx ]==IN_KIND_STAKE ) ) {
1834 0 : if( FD_UNLIKELY( chunk<ctx->in[ in_idx ].chunk0 || chunk>ctx->in[ in_idx ].wmark ) )
1835 0 : FD_LOG_ERR(( "chunk %lu %lu corrupt, not in range [%lu,%lu]", chunk, sz,
1836 0 : ctx->in[ in_idx ].chunk0, ctx->in[ in_idx ].wmark ));
1837 :
1838 0 : uchar const * dcache_entry = fd_chunk_to_laddr_const( ctx->in[ in_idx ].mem, chunk );
1839 0 : fd_multi_epoch_leaders_stake_msg_init( ctx->mleaders, fd_type_pun_const( dcache_entry ) );
1840 0 : return;
1841 0 : }
1842 :
1843 0 : ulong slot;
1844 0 : switch( ctx->in_kind[ in_idx ] ) {
1845 0 : case IN_KIND_BANK:
1846 0 : case IN_KIND_PACK: {
1847 0 : slot = fd_disco_bank_sig_slot( sig );
1848 0 : break;
1849 0 : }
1850 0 : default:
1851 0 : FD_LOG_ERR(( "unexpected in_kind %d", ctx->in_kind[ in_idx ] ));
1852 0 : }
1853 :
1854 : /* The following sequence is possible...
1855 :
1856 : 1. We become leader in slot 10
1857 : 2. While leader, we switch to a fork that is on slot 8, where
1858 : we are leader
1859 : 3. We get the in-flight microblocks for slot 10
1860 :
1861 : These in-flight microblocks need to be dropped, so we check
1862 : against the high water mark (highwater_leader_slot) rather than
1863 : the current hashcnt here when determining what to drop.
1864 :
1865 : We know if the slot is lower than the high water mark it's from a stale
1866 : leader slot, because we will not become leader for the same slot twice
1867 : even if we are reset back in time (to prevent duplicate blocks). */
1868 0 : int is_frag_for_prior_leader_slot = slot<ctx->highwater_leader_slot;
1869 :
1870 0 : if( FD_UNLIKELY( ctx->in_kind[ in_idx ]==IN_KIND_PACK ) ) {
1871 : /* We now know the real amount of microblocks published, so set an
1872 : exact bound for once we receive them. */
1873 0 : ctx->skip_frag = 1;
1874 0 : if( FD_UNLIKELY( is_frag_for_prior_leader_slot ) ) return;
1875 :
1876 0 : FD_TEST( ctx->microblocks_lower_bound<=ctx->max_microblocks_per_slot );
1877 0 : fd_done_packing_t const * done_packing = fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk );
1878 0 : FD_LOG_INFO(( "done_packing(slot=%lu,seen_microblocks=%lu,microblocks_in_slot=%lu)",
1879 0 : ctx->slot,
1880 0 : ctx->microblocks_lower_bound,
1881 0 : done_packing->microblocks_in_slot ));
1882 0 : ctx->slot_done = 1;
1883 0 : ctx->microblocks_lower_bound += ctx->max_microblocks_per_slot - done_packing->microblocks_in_slot;
1884 0 : return;
1885 0 : } else {
1886 0 : if( FD_UNLIKELY( chunk<ctx->in[ in_idx ].chunk0 || chunk>ctx->in[ in_idx ].wmark || sz>USHORT_MAX ) )
1887 0 : FD_LOG_ERR(( "chunk %lu %lu corrupt, not in range [%lu,%lu]", chunk, sz, ctx->in[ in_idx ].chunk0, ctx->in[ in_idx ].wmark ));
1888 :
1889 0 : uchar * src = (uchar *)fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk );
1890 :
1891 0 : fd_memcpy( ctx->_txns, src, sz-sizeof(fd_microblock_trailer_t) );
1892 0 : fd_memcpy( ctx->_microblock_trailer, src+sz-sizeof(fd_microblock_trailer_t), sizeof(fd_microblock_trailer_t) );
1893 :
1894 0 : ctx->skip_frag = is_frag_for_prior_leader_slot;
1895 0 : }
1896 0 : }
1897 :
1898 : static void
1899 : publish_microblock( fd_poh_ctx_t * ctx,
1900 : fd_stem_context_t * stem,
1901 : ulong slot,
1902 : ulong hashcnt_delta,
1903 0 : ulong txn_cnt ) {
1904 0 : uchar * dst = (uchar *)fd_chunk_to_laddr( ctx->shred_out->mem, ctx->shred_out->chunk );
1905 0 : FD_TEST( slot>=ctx->reset_slot );
1906 0 : fd_entry_batch_meta_t * meta = (fd_entry_batch_meta_t *)dst;
1907 0 : meta->parent_offset = 1UL+slot-ctx->reset_slot;
1908 0 : meta->reference_tick = (ctx->hashcnt/ctx->hashcnt_per_tick) % ctx->ticks_per_slot;
1909 0 : meta->block_complete = !ctx->hashcnt;
1910 :
1911 : /* Refer to publish_tick() for details on meta->parent_block_id_valid. */
1912 0 : meta->parent_block_id_valid = ctx->parent_slot == (slot-meta->parent_offset);
1913 0 : if( FD_LIKELY( meta->parent_block_id_valid ) ) {
1914 0 : fd_memcpy( meta->parent_block_id, ctx->parent_block_id, 32UL );
1915 0 : }
1916 :
1917 0 : dst += sizeof(fd_entry_batch_meta_t);
1918 0 : fd_entry_batch_header_t * header = (fd_entry_batch_header_t *)dst;
1919 0 : header->hashcnt_delta = hashcnt_delta;
1920 0 : fd_memcpy( header->hash, ctx->hash, 32UL );
1921 :
1922 0 : dst += sizeof(fd_entry_batch_header_t);
1923 0 : ulong payload_sz = 0UL;
1924 0 : ulong included_txn_cnt = 0UL;
1925 0 : for( ulong i=0UL; i<txn_cnt; i++ ) {
1926 0 : fd_txn_p_t * txn = (fd_txn_p_t *)(ctx->_txns + i*sizeof(fd_txn_p_t));
1927 0 : if( FD_UNLIKELY( !(txn->flags & FD_TXN_P_FLAGS_EXECUTE_SUCCESS) ) ) continue;
1928 :
1929 0 : fd_memcpy( dst, txn->payload, txn->payload_sz );
1930 0 : payload_sz += txn->payload_sz;
1931 0 : dst += txn->payload_sz;
1932 0 : included_txn_cnt++;
1933 0 : }
1934 0 : header->txn_cnt = included_txn_cnt;
1935 :
1936 : /* We always have credits to publish here, because we have a burst
1937 : value of 3 credits, and at most we will publish_tick() once and
1938 : then publish_became_leader() once, leaving one credit here to
1939 : publish the microblock. */
1940 0 : ulong tspub = (ulong)fd_frag_meta_ts_comp( fd_tickcount() );
1941 0 : ulong sz = sizeof(fd_entry_batch_meta_t)+sizeof(fd_entry_batch_header_t)+payload_sz;
1942 0 : ulong new_sig = fd_disco_poh_sig( slot, POH_PKT_TYPE_MICROBLOCK, 0UL );
1943 0 : fd_stem_publish( stem, ctx->shred_out->idx, new_sig, ctx->shred_out->chunk, sz, 0UL, 0UL, tspub );
1944 0 : ctx->shred_seq = stem->seqs[ ctx->shred_out->idx ];
1945 0 : ctx->shred_out->chunk = fd_dcache_compact_next( ctx->shred_out->chunk, sz, ctx->shred_out->chunk0, ctx->shred_out->wmark );
1946 0 : }
1947 :
1948 : static inline void
1949 : after_frag( fd_poh_ctx_t * ctx,
1950 : ulong in_idx,
1951 : ulong seq,
1952 : ulong sig,
1953 : ulong sz,
1954 : ulong tsorig,
1955 : ulong tspub,
1956 0 : fd_stem_context_t * stem ) {
1957 0 : (void)in_idx;
1958 0 : (void)seq;
1959 0 : (void)tsorig;
1960 0 : (void)tspub;
1961 :
1962 0 : if( FD_UNLIKELY( ctx->skip_frag ) ) return;
1963 :
1964 0 : if( FD_UNLIKELY( ctx->in_kind[ in_idx ]==IN_KIND_STAKE ) ) {
1965 0 : fd_multi_epoch_leaders_stake_msg_fini( ctx->mleaders );
1966 : /* It might seem like we do not need to do state transitions in and
1967 : out of being the leader here, since leader schedule updates are
1968 : always one epoch in advance (whether we are leader or not would
1969 : never change for the currently executing slot) but this is not
1970 : true for new ledgers when the validator first boots. We will
1971 : likely be the leader in slot 1, and get notified of the leader
1972 : schedule for that slot while we are still in it.
1973 :
1974 : For safety we just handle both transitions, in and out, although
1975 : the only one possible should be into leader. */
1976 0 : ulong next_leader_slot_after_frag = next_leader_slot( ctx );
1977 :
1978 0 : int currently_leader = ctx->slot>=ctx->next_leader_slot;
1979 0 : int leader_after_frag = ctx->slot>=next_leader_slot_after_frag;
1980 :
1981 0 : FD_LOG_INFO(( "stake_update(before_leader=%lu,after_leader=%lu)",
1982 0 : ctx->next_leader_slot,
1983 0 : next_leader_slot_after_frag ));
1984 :
1985 0 : ctx->next_leader_slot = next_leader_slot_after_frag;
1986 0 : if( FD_UNLIKELY( currently_leader && !leader_after_frag ) ) {
1987 : /* Shouldn't ever happen, otherwise we need to do a state
1988 : transition out of being leader. */
1989 0 : FD_LOG_ERR(( "stake update caused us to no longer be leader in an active slot" ));
1990 0 : }
1991 :
1992 : /* Nothing to do if we transition into being leader, since it
1993 : will just get picked up by the regular tick loop. */
1994 0 : if( FD_UNLIKELY( !currently_leader && leader_after_frag ) ) {
1995 0 : publish_plugin_slot_start( ctx, next_leader_slot_after_frag, ctx->reset_slot );
1996 0 : }
1997 :
1998 0 : return;
1999 0 : }
2000 :
2001 0 : if( FD_UNLIKELY( !ctx->microblocks_lower_bound ) ) {
2002 0 : double tick_per_ns = fd_tempo_tick_per_ns( NULL );
2003 0 : fd_histf_sample( ctx->first_microblock_delay, (ulong)((double)(fd_log_wallclock()-ctx->reset_slot_start_ns)/tick_per_ns) );
2004 0 : }
2005 :
2006 0 : ulong target_slot = fd_disco_bank_sig_slot( sig );
2007 :
2008 0 : if( FD_UNLIKELY( target_slot!=ctx->next_leader_slot || target_slot!=ctx->slot ) ) {
2009 0 : FD_LOG_ERR(( "packed too early or late target_slot=%lu, current_slot=%lu. highwater_leader_slot=%lu",
2010 0 : target_slot, ctx->slot, ctx->highwater_leader_slot ));
2011 0 : }
2012 :
2013 0 : FD_TEST( ctx->current_leader_bank );
2014 0 : FD_TEST( ctx->microblocks_lower_bound<ctx->max_microblocks_per_slot );
2015 0 : ctx->microblocks_lower_bound += 1UL;
2016 :
2017 0 : ulong txn_cnt = (sz-sizeof(fd_microblock_trailer_t))/sizeof(fd_txn_p_t);
2018 0 : fd_txn_p_t * txns = (fd_txn_p_t *)(ctx->_txns);
2019 0 : ulong executed_txn_cnt = 0UL;
2020 0 : ulong cus_used = 0UL;
2021 0 : for( ulong i=0UL; i<txn_cnt; i++ ) {
2022 : /* It's important that we check if a transaction is included in the
2023 : block with FD_TXN_P_FLAGS_EXECUTE_SUCCESS since
2024 : actual_consumed_cus may have a nonzero value for excluded
2025 : transactions used for monitoring purposes */
2026 0 : if( FD_LIKELY( txns[ i ].flags & FD_TXN_P_FLAGS_EXECUTE_SUCCESS ) ) {
2027 0 : executed_txn_cnt++;
2028 0 : cus_used += txns[ i ].bank_cu.actual_consumed_cus;
2029 0 : }
2030 0 : }
2031 :
2032 : /* We don't publish transactions that fail to execute. If all the
2033 : transactions failed to execute, the microblock would be empty,
2034 : causing agave to think it's a tick and complain. Instead, we just
2035 : skip the microblock and don't hash or update the hashcnt. */
2036 0 : if( FD_UNLIKELY( !executed_txn_cnt ) ) return;
2037 :
2038 0 : uchar data[ 64 ];
2039 0 : fd_memcpy( data, ctx->hash, 32UL );
2040 0 : fd_memcpy( data+32UL, ctx->_microblock_trailer->hash, 32UL );
2041 0 : fd_sha256_hash( data, 64UL, ctx->hash );
2042 :
2043 0 : ctx->hashcnt++;
2044 0 : FD_TEST( ctx->hashcnt>ctx->last_hashcnt );
2045 0 : ulong hashcnt_delta = ctx->hashcnt - ctx->last_hashcnt;
2046 :
2047 : /* The hashing loop above will never leave us exactly one away from
2048 : crossing a tick boundary, so this increment will never cause the
2049 : current tick (or the slot) to change, except in low power mode
2050 : for development, in which case we do need to register the tick
2051 : with the leader bank. We don't need to publish the tick since
2052 : sending the microblock below is the publishing action. */
2053 0 : if( FD_UNLIKELY( !(ctx->hashcnt%ctx->hashcnt_per_slot ) ) ) {
2054 0 : ctx->slot++;
2055 0 : ctx->hashcnt = 0UL;
2056 0 : }
2057 :
2058 0 : ctx->last_slot = ctx->slot;
2059 0 : ctx->last_hashcnt = ctx->hashcnt;
2060 :
2061 0 : ctx->cus_used += cus_used;
2062 :
2063 0 : if( FD_UNLIKELY( !(ctx->hashcnt%ctx->hashcnt_per_tick ) ) ) {
2064 0 : fd_ext_poh_register_tick( ctx->current_leader_bank, ctx->hash );
2065 0 : if( FD_UNLIKELY( ctx->slot>ctx->next_leader_slot ) ) {
2066 : /* We ticked while leader and are no longer leader... transition
2067 : the state machine. */
2068 0 : publish_plugin_slot_end( ctx, ctx->next_leader_slot, ctx->cus_used );
2069 :
2070 0 : no_longer_leader( ctx );
2071 :
2072 0 : if( FD_UNLIKELY( ctx->slot>=ctx->next_leader_slot ) ) {
2073 : /* We finished a leader slot, and are immediately leader for the
2074 : following slot... transition. */
2075 0 : publish_plugin_slot_start( ctx, ctx->next_leader_slot, ctx->next_leader_slot-1UL );
2076 0 : }
2077 0 : }
2078 0 : }
2079 :
2080 0 : publish_microblock( ctx, stem, target_slot, hashcnt_delta, txn_cnt );
2081 0 : }
2082 :
2083 : static void
2084 : privileged_init( fd_topo_t * topo,
2085 0 : fd_topo_tile_t * tile ) {
2086 0 : void * scratch = fd_topo_obj_laddr( topo, tile->tile_obj_id );
2087 :
2088 0 : FD_SCRATCH_ALLOC_INIT( l, scratch );
2089 0 : fd_poh_ctx_t * ctx = FD_SCRATCH_ALLOC_APPEND( l, alignof( fd_poh_ctx_t ), sizeof( fd_poh_ctx_t ) );
2090 :
2091 0 : if( FD_UNLIKELY( !strcmp( tile->poh.identity_key_path, "" ) ) )
2092 0 : FD_LOG_ERR(( "identity_key_path not set" ));
2093 :
2094 0 : const uchar * identity_key = fd_keyload_load( tile->poh.identity_key_path, /* pubkey only: */ 1 );
2095 0 : fd_memcpy( ctx->identity_key.uc, identity_key, 32UL );
2096 :
2097 0 : if( FD_UNLIKELY( !tile->poh.bundle.vote_account_path[0] ) ) {
2098 0 : tile->poh.bundle.enabled = 0;
2099 0 : }
2100 0 : if( FD_UNLIKELY( tile->poh.bundle.enabled ) ) {
2101 0 : if( FD_UNLIKELY( !fd_base58_decode_32( tile->poh.bundle.vote_account_path, ctx->bundle.vote_account.uc ) ) ) {
2102 0 : const uchar * vote_key = fd_keyload_load( tile->poh.bundle.vote_account_path, /* pubkey only: */ 1 );
2103 0 : fd_memcpy( ctx->bundle.vote_account.uc, vote_key, 32UL );
2104 0 : }
2105 0 : }
2106 0 : }
2107 :
2108 : /* The Agave client needs to communicate to the shred tile what
2109 : the shred version is on boot, but shred tile does not live in the
2110 : same address space, so have the PoH tile pass the value through
2111 : via. a shared memory ulong. */
2112 :
2113 : static volatile ulong * fd_shred_version;
2114 :
2115 : void
2116 0 : fd_ext_shred_set_shred_version( ulong shred_version ) {
2117 0 : while( FD_UNLIKELY( !fd_shred_version ) ) FD_SPIN_PAUSE();
2118 0 : *fd_shred_version = shred_version;
2119 0 : }
2120 :
2121 : void
2122 : fd_ext_poh_publish_gossip_vote( uchar * data,
2123 : ulong data_len,
2124 : uint source_ipv4,
2125 0 : uchar * pubkey ) {
2126 0 : (void)pubkey;
2127 0 : uchar txn_with_header[ FD_TPU_RAW_MTU ];
2128 0 : fd_txn_m_t * txnm = (fd_txn_m_t *)txn_with_header;
2129 0 : *txnm = (fd_txn_m_t) { 0UL };
2130 0 : txnm->payload_sz = (ushort)data_len;
2131 0 : txnm->source_ipv4 = source_ipv4;
2132 0 : txnm->source_tpu = FD_TXN_M_TPU_SOURCE_GOSSIP;
2133 0 : fd_memcpy(txn_with_header + sizeof(fd_txn_m_t), data, data_len);
2134 0 : poh_link_publish( &gossip_dedup, 1UL, txn_with_header, fd_txn_m_realized_footprint( txnm, 0, 0 ) );
2135 0 : }
2136 :
2137 : void
2138 : fd_ext_poh_publish_leader_schedule( uchar * data,
2139 0 : ulong data_len ) {
2140 0 : poh_link_publish( &stake_out, 2UL, data, data_len );
2141 0 : }
2142 :
2143 : void
2144 : fd_ext_poh_publish_cluster_info( uchar * data,
2145 0 : ulong data_len ) {
2146 0 : poh_link_publish( &crds_shred, 2UL, data, data_len );
2147 0 : }
2148 :
2149 : void
2150 0 : fd_ext_poh_publish_executed_txn( uchar const * data ) {
2151 0 : static int lock = 0;
2152 :
2153 : /* Need to lock since the link publisher is not concurrent, and replay
2154 : happens on a thread pool. */
2155 0 : for(;;) {
2156 0 : if( FD_LIKELY( FD_ATOMIC_CAS( &lock, 0, 1 )==0 ) ) break;
2157 0 : FD_SPIN_PAUSE();
2158 0 : }
2159 :
2160 0 : FD_COMPILER_MFENCE();
2161 0 : poh_link_publish( &executed_txn, 0UL, data, 64UL );
2162 0 : FD_COMPILER_MFENCE();
2163 :
2164 0 : FD_VOLATILE(lock) = 0;
2165 0 : }
2166 :
2167 : void
2168 : fd_ext_plugin_publish_replay_stage( ulong sig,
2169 : uchar * data,
2170 0 : ulong data_len ) {
2171 0 : poh_link_publish( &replay_plugin, sig, data, data_len );
2172 0 : }
2173 :
2174 : void
2175 : fd_ext_plugin_publish_genesis_hash( ulong sig,
2176 : uchar * data,
2177 0 : ulong data_len ) {
2178 0 : poh_link_publish( &replay_plugin, sig, data, data_len );
2179 0 : }
2180 :
2181 : void
2182 : fd_ext_plugin_publish_start_progress( ulong sig,
2183 : uchar * data,
2184 0 : ulong data_len ) {
2185 0 : poh_link_publish( &start_progress_plugin, sig, data, data_len );
2186 0 : }
2187 :
2188 : void
2189 : fd_ext_plugin_publish_vote_listener( ulong sig,
2190 : uchar * data,
2191 0 : ulong data_len ) {
2192 0 : poh_link_publish( &vote_listener_plugin, sig, data, data_len );
2193 0 : }
2194 :
2195 : void
2196 : fd_ext_plugin_publish_validator_info( ulong sig,
2197 : uchar * data,
2198 0 : ulong data_len ) {
2199 0 : poh_link_publish( &validator_info_plugin, sig, data, data_len );
2200 0 : }
2201 :
2202 : void
2203 : fd_ext_plugin_publish_periodic( ulong sig,
2204 : uchar * data,
2205 0 : ulong data_len ) {
2206 0 : poh_link_publish( &gossip_plugin, sig, data, data_len );
2207 0 : }
2208 :
2209 : void
2210 : fd_ext_resolv_publish_root_bank( uchar * data,
2211 0 : ulong data_len ) {
2212 0 : poh_link_publish( &replay_resolv, 0UL, data, data_len );
2213 0 : }
2214 :
2215 : void
2216 : fd_ext_resolv_publish_completed_blockhash( uchar * data,
2217 0 : ulong data_len ) {
2218 0 : poh_link_publish( &replay_resolv, 1UL, data, data_len );
2219 0 : }
2220 :
2221 : static inline fd_poh_out_ctx_t
2222 : out1( fd_topo_t const * topo,
2223 : fd_topo_tile_t const * tile,
2224 0 : char const * name ) {
2225 0 : ulong idx = ULONG_MAX;
2226 :
2227 0 : for( ulong i=0UL; i<tile->out_cnt; i++ ) {
2228 0 : fd_topo_link_t const * link = &topo->links[ tile->out_link_id[ i ] ];
2229 0 : if( !strcmp( link->name, name ) ) {
2230 0 : if( FD_UNLIKELY( idx!=ULONG_MAX ) ) FD_LOG_ERR(( "tile %s:%lu had multiple output links named %s but expected one", tile->name, tile->kind_id, name ));
2231 0 : idx = i;
2232 0 : }
2233 0 : }
2234 :
2235 0 : if( FD_UNLIKELY( idx==ULONG_MAX ) ) FD_LOG_ERR(( "tile %s:%lu had no output link named %s", tile->name, tile->kind_id, name ));
2236 :
2237 0 : void * mem = topo->workspaces[ topo->objs[ topo->links[ tile->out_link_id[ idx ] ].dcache_obj_id ].wksp_id ].wksp;
2238 0 : ulong chunk0 = fd_dcache_compact_chunk0( mem, topo->links[ tile->out_link_id[ idx ] ].dcache );
2239 0 : ulong wmark = fd_dcache_compact_wmark ( mem, topo->links[ tile->out_link_id[ idx ] ].dcache, topo->links[ tile->out_link_id[ idx ] ].mtu );
2240 :
2241 0 : return (fd_poh_out_ctx_t){ .idx = idx, .mem = mem, .chunk0 = chunk0, .wmark = wmark, .chunk = chunk0 };
2242 0 : }
2243 :
2244 : static void
2245 : unprivileged_init( fd_topo_t * topo,
2246 0 : fd_topo_tile_t * tile ) {
2247 0 : void * scratch = fd_topo_obj_laddr( topo, tile->tile_obj_id );
2248 :
2249 0 : FD_SCRATCH_ALLOC_INIT( l, scratch );
2250 0 : fd_poh_ctx_t * ctx = FD_SCRATCH_ALLOC_APPEND( l, alignof( fd_poh_ctx_t ), sizeof( fd_poh_ctx_t ) );
2251 0 : void * sha256 = FD_SCRATCH_ALLOC_APPEND( l, FD_SHA256_ALIGN, FD_SHA256_FOOTPRINT );
2252 :
2253 0 : #define NONNULL( x ) (__extension__({ \
2254 0 : __typeof__((x)) __x = (x); \
2255 0 : if( FD_UNLIKELY( !__x ) ) FD_LOG_ERR(( #x " was unexpectedly NULL" )); \
2256 0 : __x; }))
2257 :
2258 0 : ctx->mleaders = NONNULL( fd_multi_epoch_leaders_join( fd_multi_epoch_leaders_new( ctx->mleaders_mem ) ) );
2259 0 : ctx->sha256 = NONNULL( fd_sha256_join( fd_sha256_new( sha256 ) ) );
2260 0 : ctx->current_leader_bank = NULL;
2261 0 : ctx->pack_leader_bank = NULL;
2262 0 : ctx->signal_leader_change = NULL;
2263 :
2264 0 : ctx->shred_seq = ULONG_MAX;
2265 0 : ctx->halted_switching_key = 0;
2266 0 : ctx->keyswitch = fd_keyswitch_join( fd_topo_obj_laddr( topo, tile->keyswitch_obj_id ) );
2267 0 : FD_TEST( ctx->keyswitch );
2268 :
2269 0 : ctx->slot = 0UL;
2270 0 : ctx->hashcnt = 0UL;
2271 0 : ctx->last_hashcnt = 0UL;
2272 0 : ctx->highwater_leader_slot = ULONG_MAX;
2273 0 : ctx->next_leader_slot = ULONG_MAX;
2274 0 : ctx->reset_slot = ULONG_MAX;
2275 :
2276 0 : ctx->lagged_consecutive_leader_start = tile->poh.lagged_consecutive_leader_start;
2277 0 : ctx->expect_sequential_leader_slot = ULONG_MAX;
2278 :
2279 0 : ctx->slot_done = 1;
2280 0 : ctx->expect_pack_idx = 0U;
2281 0 : ctx->microblocks_lower_bound = 0UL;
2282 :
2283 0 : ctx->max_active_descendant = 0UL;
2284 :
2285 0 : if( FD_UNLIKELY( tile->poh.bundle.enabled ) ) {
2286 0 : ctx->bundle.enabled = 1;
2287 0 : NONNULL( fd_bundle_crank_gen_init( ctx->bundle.gen, (fd_acct_addr_t const *)tile->poh.bundle.tip_distribution_program_addr,
2288 0 : (fd_acct_addr_t const *)tile->poh.bundle.tip_payment_program_addr,
2289 0 : (fd_acct_addr_t const *)ctx->bundle.vote_account.uc,
2290 0 : (fd_acct_addr_t const *)ctx->bundle.vote_account.uc, "NAN", 0UL ) ); /* last three arguments are properly bogus */
2291 0 : } else {
2292 0 : ctx->bundle.enabled = 0;
2293 0 : }
2294 :
2295 0 : ulong poh_shred_obj_id = fd_pod_query_ulong( topo->props, "poh_shred", ULONG_MAX );
2296 0 : FD_TEST( poh_shred_obj_id!=ULONG_MAX );
2297 :
2298 0 : fd_shred_version = fd_fseq_join( fd_topo_obj_laddr( topo, poh_shred_obj_id ) );
2299 0 : FD_TEST( fd_shred_version );
2300 :
2301 0 : poh_link_init( &gossip_dedup, topo, tile, out1( topo, tile, "gossip_dedup" ).idx );
2302 0 : poh_link_init( &stake_out, topo, tile, out1( topo, tile, "stake_out" ).idx );
2303 0 : poh_link_init( &crds_shred, topo, tile, out1( topo, tile, "crds_shred" ).idx );
2304 0 : poh_link_init( &replay_resolv, topo, tile, out1( topo, tile, "replay_resol" ).idx );
2305 0 : poh_link_init( &executed_txn, topo, tile, out1( topo, tile, "executed_txn" ).idx );
2306 :
2307 0 : if( FD_LIKELY( tile->poh.plugins_enabled ) ) {
2308 0 : poh_link_init( &replay_plugin, topo, tile, out1( topo, tile, "replay_plugi" ).idx );
2309 0 : poh_link_init( &gossip_plugin, topo, tile, out1( topo, tile, "gossip_plugi" ).idx );
2310 0 : poh_link_init( &start_progress_plugin, topo, tile, out1( topo, tile, "startp_plugi" ).idx );
2311 0 : poh_link_init( &vote_listener_plugin, topo, tile, out1( topo, tile, "votel_plugin" ).idx );
2312 0 : poh_link_init( &validator_info_plugin, topo, tile, out1( topo, tile, "valcfg_plugi" ).idx );
2313 0 : } else {
2314 : /* Mark these mcaches as "available", so the system boots, but the
2315 : memory is not set so nothing will actually get published via.
2316 : the links. */
2317 0 : FD_COMPILER_MFENCE();
2318 0 : replay_plugin.mcache = (fd_frag_meta_t*)1;
2319 0 : gossip_plugin.mcache = (fd_frag_meta_t*)1;
2320 0 : start_progress_plugin.mcache = (fd_frag_meta_t*)1;
2321 0 : vote_listener_plugin.mcache = (fd_frag_meta_t*)1;
2322 0 : validator_info_plugin.mcache = (fd_frag_meta_t*)1;
2323 0 : FD_COMPILER_MFENCE();
2324 0 : }
2325 :
2326 0 : FD_LOG_INFO(( "PoH waiting to be initialized by Agave client... %lu %lu", fd_poh_waiting_lock, fd_poh_returned_lock ));
2327 0 : FD_VOLATILE( fd_poh_global_ctx ) = ctx;
2328 0 : FD_COMPILER_MFENCE();
2329 0 : for(;;) {
2330 0 : if( FD_LIKELY( FD_VOLATILE_CONST( fd_poh_waiting_lock ) ) ) break;
2331 0 : FD_SPIN_PAUSE();
2332 0 : }
2333 0 : FD_VOLATILE( fd_poh_waiting_lock ) = 0UL;
2334 0 : FD_VOLATILE( fd_poh_returned_lock ) = 1UL;
2335 0 : FD_COMPILER_MFENCE();
2336 0 : for(;;) {
2337 0 : if( FD_UNLIKELY( !FD_VOLATILE_CONST( fd_poh_returned_lock ) ) ) break;
2338 0 : FD_SPIN_PAUSE();
2339 0 : }
2340 0 : FD_COMPILER_MFENCE();
2341 :
2342 0 : if( FD_UNLIKELY( ctx->reset_slot==ULONG_MAX ) ) FD_LOG_ERR(( "PoH was not initialized by Agave client" ));
2343 :
2344 0 : fd_histf_join( fd_histf_new( ctx->begin_leader_delay, FD_MHIST_SECONDS_MIN( POH, BEGIN_LEADER_DELAY_SECONDS ),
2345 0 : FD_MHIST_SECONDS_MAX( POH, BEGIN_LEADER_DELAY_SECONDS ) ) );
2346 0 : fd_histf_join( fd_histf_new( ctx->first_microblock_delay, FD_MHIST_SECONDS_MIN( POH, FIRST_MICROBLOCK_DELAY_SECONDS ),
2347 0 : FD_MHIST_SECONDS_MAX( POH, FIRST_MICROBLOCK_DELAY_SECONDS ) ) );
2348 0 : fd_histf_join( fd_histf_new( ctx->slot_done_delay, FD_MHIST_SECONDS_MIN( POH, SLOT_DONE_DELAY_SECONDS ),
2349 0 : FD_MHIST_SECONDS_MAX( POH, SLOT_DONE_DELAY_SECONDS ) ) );
2350 :
2351 0 : fd_histf_join( fd_histf_new( ctx->bundle_init_delay, FD_MHIST_SECONDS_MIN( POH, BUNDLE_INITIALIZE_DELAY_SECONDS ),
2352 0 : FD_MHIST_SECONDS_MAX( POH, BUNDLE_INITIALIZE_DELAY_SECONDS ) ) );
2353 :
2354 0 : for( ulong i=0UL; i<tile->in_cnt; i++ ) {
2355 0 : fd_topo_link_t * link = &topo->links[ tile->in_link_id[ i ] ];
2356 0 : fd_topo_wksp_t * link_wksp = &topo->workspaces[ topo->objs[ link->dcache_obj_id ].wksp_id ];
2357 :
2358 0 : ctx->in[ i ].mem = link_wksp->wksp;
2359 0 : ctx->in[ i ].chunk0 = fd_dcache_compact_chunk0( ctx->in[ i ].mem, link->dcache );
2360 0 : ctx->in[ i ].wmark = fd_dcache_compact_wmark ( ctx->in[ i ].mem, link->dcache, link->mtu );
2361 :
2362 0 : if( !strcmp( link->name, "stake_out" ) ) {
2363 0 : ctx->in_kind[ i ] = IN_KIND_STAKE;
2364 0 : } else if( !strcmp( link->name, "pack_poh" ) ) {
2365 0 : ctx->in_kind[ i ] = IN_KIND_PACK;
2366 0 : } else if( !strcmp( link->name, "bank_poh" ) ) {
2367 0 : ctx->in_kind[ i ] = IN_KIND_BANK;
2368 0 : } else {
2369 0 : FD_LOG_ERR(( "unexpected input link name %s", link->name ));
2370 0 : }
2371 0 : }
2372 :
2373 0 : *ctx->shred_out = out1( topo, tile, "poh_shred" );
2374 0 : *ctx->pack_out = out1( topo, tile, "poh_pack" );
2375 0 : ctx->plugin_out->mem = NULL;
2376 0 : if( FD_LIKELY( tile->poh.plugins_enabled ) ) {
2377 0 : *ctx->plugin_out = out1( topo, tile, "poh_plugin" );
2378 0 : }
2379 :
2380 0 : ctx->features_activation_avail = 0UL;
2381 0 : for( ulong i=0UL; i<FD_SHRED_FEATURES_ACTIVATION_SLOT_CNT; i++ )
2382 0 : ctx->features_activation->slots[i] = FD_SHRED_FEATURES_ACTIVATION_SLOT_DISABLED;
2383 :
2384 0 : ulong scratch_top = FD_SCRATCH_ALLOC_FINI( l, 1UL );
2385 0 : if( FD_UNLIKELY( scratch_top > (ulong)scratch + scratch_footprint( tile ) ) )
2386 0 : FD_LOG_ERR(( "scratch overflow %lu %lu %lu", scratch_top - (ulong)scratch - scratch_footprint( tile ), scratch_top, (ulong)scratch + scratch_footprint( tile ) ));
2387 0 : }
2388 :
2389 : /* One tick, one microblock, one plugin slot end, one plugin slot start,
2390 : one leader update, and one features activation. */
2391 0 : #define STEM_BURST (6UL)
2392 :
2393 : /* See explanation in fd_pack */
2394 0 : #define STEM_LAZY (128L*3000L)
2395 :
2396 0 : #define STEM_CALLBACK_CONTEXT_TYPE fd_poh_ctx_t
2397 0 : #define STEM_CALLBACK_CONTEXT_ALIGN alignof(fd_poh_ctx_t)
2398 :
2399 0 : #define STEM_CALLBACK_DURING_HOUSEKEEPING during_housekeeping
2400 0 : #define STEM_CALLBACK_METRICS_WRITE metrics_write
2401 0 : #define STEM_CALLBACK_AFTER_CREDIT after_credit
2402 0 : #define STEM_CALLBACK_BEFORE_FRAG before_frag
2403 0 : #define STEM_CALLBACK_DURING_FRAG during_frag
2404 0 : #define STEM_CALLBACK_AFTER_FRAG after_frag
2405 :
2406 : #include "../../disco/stem/fd_stem.c"
2407 :
2408 : fd_topo_run_tile_t fd_tile_poh = {
2409 : .name = "poh",
2410 : .populate_allowed_seccomp = NULL,
2411 : .populate_allowed_fds = NULL,
2412 : .scratch_align = scratch_align,
2413 : .scratch_footprint = scratch_footprint,
2414 : .privileged_init = privileged_init,
2415 : .unprivileged_init = unprivileged_init,
2416 : .run = stem_run,
2417 : };
|
