summaryrefslogtreecommitdiffstats
path: root/src
Commit message (Collapse)AuthorAgeFilesLines
* trie: Use load8_beu*() for trie_mismatch()HEADmainTavian Barnes3 days1-29/+23
|
* bfstd: Use load8_leu*() for asciinlen()Tavian Barnes3 days1-24/+20
|
* bit: Add the load8_[bl]euN() utilitiesTavian Barnes3 days1-1/+53
|
* trie: Make nibble indices big-endianTavian Barnes3 days1-7/+7
| | | | Otherwise the order doesn't match lexicographical order on bytes.
* trie: New trie_node_size() helperTavian Barnes3 days1-3/+9
|
* trie: Clean up some bounds checkingTavian Barnes4 days1-6/+6
|
* trie: Micro-optimize trie_representative()Tavian Barnes5 days1-4/+5
| | | | | popcount(map & (bit - 1) & mask) has a longer critical path than popcount(map & (bit - 1)) & mask.
* eval: Don't error out in -ls if the time overflowsTavian Barnes9 days1-19/+30
|
* ioq: Use getppid() for the heavy NOP syscallTavian Barnes10 days1-1/+1
| | | | | | | | getpid() could hypothetically be cached (and glibc used to do it). getppid() can't be, because the parent could die and reparent the child to init. Link: https://lore.kernel.org/linux-mm/Z58dIOGlJ3grsAge@casper.infradead.org/
* mtab: Use mount IDs instead of dev_t for bfs_fstype()Tavian Barnes10 days1-2/+2
| | | | | | | | | This fixes -fstype on btrfs subvolumes, which previously reported "unknown" due to their dev_t being potentially different from the mount point. Link: https://savannah.gnu.org/bugs/?50859 Link: https://lwn.net/Articles/866582/
* stat: Expose mount IDs from statx()Tavian Barnes10 days3-2/+40
|
* trie: New trie_{get,set}_{str,mem}() functionsTavian Barnes10 days4-23/+93
|
* diag: Try to make diagnostics signal-saferTavian Barnes10 days1-3/+16
| | | | Link: https://sourceware.org/bugzilla/show_bug.cgi?id=16060
* ioq: Rewrite the spin loop to avoid a warningTavian Barnes13 days1-4/+3
| | | | | | | | | | | | With some GCC versions, --enable-ubsan leads to this warning: src/ioq.c: In function ‘ioq_slot_wait’: src/ioq.c:287:17: warning: ignoring loop annotation 287 | for (int j = 0; j < (1 << i); ++j) { | ^~~ presumably due to UBSan rewriting the shift to check for overflow. Work around this by precomputing the iteration count.
* sighook: Fix sigreset() error handlingTavian Barnes2025-01-201-9/+16
|
* sighook: Don't forget to check `initialized` in sigreset()Tavian Barnes2025-01-181-0/+4
|
* sighook: Fix up siginit() error pathsTavian Barnes2025-01-181-8/+14
|
* xspawn: Unblock signals between fork() and exec()Tavian Barnes2025-01-181-2/+15
| | | | | | Otherwise the child process runs with all signals blocked. Fixes: 423cfa4 ("xspawn: Mask signals before fork()")
* sighook: New sigreset() function to reset all handlersTavian Barnes2025-01-182-3/+45
|
* sighook: Convert siglist to a more general rcu_list typeTavian Barnes2025-01-181-52/+66
|
* build: Feature-detect #pragma nounroll supportTavian Barnes2025-01-171-2/+2
|
* list: Don't use leading underscores for globalsTavian Barnes2025-01-161-6/+6
|
* ioq: Don't unroll the spin loopTavian Barnes2025-01-162-1/+15
|
* ioq: Prefetch slots with a write intentTavian Barnes2025-01-161-1/+1
|
* build: Check for compound literal storage-class supportTavian Barnes2025-01-071-1/+1
| | | | This lets us use them on GCC >= 13, even before C23.
* sighook: Add some missing fatal signalsTavian Barnes2024-12-241-5/+14
|
* sighook: Preserve the exact siginfo_t in reraise() on LinuxTavian Barnes2024-12-241-3/+16
| | | | Link: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=66dd34ad31e5963d72a700ec3f2449291d322921
* sighook: Always re-raise faults on macOSTavian Barnes2024-12-221-0/+7
| | | | | | | | | | macOS always fills in si_code for SIG{BUS,ILL,SEGV} as if it were a real hardware fault, so returning from the handler is not guaranteed to re- trigger the signal. Fixes: aecdabb ("sighook: Return instead of re-raising for faults") Link: https://chromium-review.googlesource.com/c/crashpad/crashpad/+/2386463/4/util/posix/signals_test.cc Link: https://github.com/chromium/crashpad/commit/e0d8a0aa01ac176804077f1f128ccc894c098f79
* sanity: New sanitize_resize() functionTavian Barnes2024-12-162-16/+33
| | | | | This wraps __sanitizer_annotate_contiguous_container() to give byte-precise tracking of usable allocation sizes with ASan.
* sanity: Get rid of sanitize_ignore()Tavian Barnes2024-12-162-11/+6
| | | | | It's probably nicer to avoid evaluating expensive arguments when not sanitizing, rather than relying on the optimizer to clean them up.
* ioq: Refactor ioq_ring_probe_flags()Tavian Barnes2024-12-041-11/+13
|
* ioq: Try spinning before blocking in ioq_slot_wait()Tavian Barnes2024-12-031-1/+17
|
* ioq: Submit and pop requests in batchesTavian Barnes2024-12-033-17/+42
| | | | | The new ioq_submit() function is now necessary to call to ensure the pending request batch is flushed.
* ioq: Add more io_uring setup flagsTavian Barnes2024-12-031-20/+55
|
* ioq: Refactor io_uring submission logicTavian Barnes2024-12-031-94/+145
|
* ioq: Use ioq_batch for both pushing and poppingTavian Barnes2024-12-031-47/+84
|
* ioq: Add a hash function between slots and monitorsTavian Barnes2024-12-031-1/+11
| | | | | | This helps avoid situations where multiple waiters block on different slots using the same monitor, which happened more often than expected due to correlations caused by batching.
* ioq: Prefetch pointers before popping themTavian Barnes2024-12-032-8/+10
| | | | | Also, cache-align struct ioq_ent to avoid false sharing when two workers are handling neighbouring requests.
* ioq: Add an ioq_nop() operation for benchmarkingTavian Barnes2024-12-023-0/+62
|
* xtime: New timespec utility functionsTavian Barnes2024-12-023-23/+95
|
* sighook: New SH_ONESHOT flagTavian Barnes2024-12-022-1/+21
|
* bfstd: Add more strto*() wrappersTavian Barnes2024-12-022-24/+69
|
* ioq: Set the worker thread names to ioq-%dTavian Barnes2024-11-271-2/+8
|
* thread: New thread_setname() functionTavian Barnes2024-11-272-0/+17
|
* sighook: Return instead of re-raising for faultsTavian Barnes2024-11-231-5/+11
| | | | | | | | | | | This makes the death look exactly like it would if we didn't handle the signal at all. Coredumps will point at the right instruction, segfaults will get logged in dmesg again, etc. Technically POSIX says this is undefined, but if we get a fault, we've already done something undefined anyway ;) Link: https://github.com/rust-lang/rust/issues/26458
* sighook: Check for SA_RESTART before using itTavian Barnes2024-11-231-1/+7
| | | | QNX doesn't support SA_RESTART.
* bfstd: makedev() takes 3 arguments on QNXTavian Barnes2024-11-231-1/+3
|
* prelude: Define _QNX_SOURCETavian Barnes2024-11-231-0/+5
|
* ioq: Enable IORING_SETUP_SUBMIT_ALLTavian Barnes2024-11-201-2/+18
|
* color: Squelch a -Wmaybe-uninitialized warning on non-LTO buildsTavian Barnes2024-11-201-1/+1
|
generated by cgit v1.2.3 (git 2.39.1) at 2025-02-17 01:12:04 +0000