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