When using metadata_thp, allocate tcache bin stacks from base0, which means they
will be placed on huge pages along with other metadata, instead of mixed with
other regular allocations.
In order to do so, modified the base allocator to support limited reuse: freed
tcached stacks (from thread termination) will be returned to base0 and made
available for reuse, but no merging will be attempted since they were bump
allocated out of base blocks. These reused base extents are managed using
separately allocated base edata_t -- they are cached in base->edata_avail when
the extent is all allocated.
One tricky part is, stats updating must be skipped for such reused extents
(since they were accounted for already, and there is no purging for base). This
requires tracking the "if is reused" state explicitly and bypass the stats
updates when allocating from them.
1. add tcache_max and nhbins into tcache_t so that they are per-tcache,
with one auto tcache per thread, it's also per-thread;
2. add mallctl for each thread to set its own tcache_max (of its auto tcache);
3. store the maximum number of items in each bin instead of using a global storage;
4. add tests for the modifications above.
5. Rename `nhbins` and `tcache_maxclass` to `global_do_not_change_nhbins` and `global_do_not_change_tcache_maxclass`.
As reported in #2449, under certain circumstances it's possible to get
stuck in an infinite loop attempting to purge from the HPA. We now
handle this by validating the HPA settings at the end of
configuration parsing and either normalizing them or aborting depending on
if `abort_conf` is set.
For better or worse, Jemalloc has a significant number of global
variables. Making all eligible global variables `static` and/or `const`
at least makes it slightly easier to reason about them, as these
qualifications communicate to the programmer restrictions on their use
without having to `grep` the whole codebase.
Validate that small allocations (i.e. those with `size <= SC_SMALL_MAXCLASS`)
which are sampled for profiling maintain the expected invariants even
though they now take up less space.
We have observed new workload patterns (namely ML training type) that cycle
through oversized allocations frequently, because 1) the dataset might be sparse
which is faster to go through, and 2) GPU accelerated. As a result, the eager
purging from the oversize arena becomes a bottleneck. To offer an easy
solution, allow normal purging of the oversized extents when background threads
are enabled.
This tests the combination of the prof_recent and thread_name features.
Verified that it catches the issue being fixed in this PR.
Also explicitly set thread name in test/unit/prof_recent. This fixes the name
testing when no default thread name is set (e.g. FreeBSD).
The codebase is already very disciplined in making any function which
can be `static`, but there are a few that appear to have slipped through
the cracks.
Decay should not be triggered during reentrant calls (may cause lock order
reversal / deadlocks). Added a delay_trigger flag to the tickers to bypass
decay when rentrancy_level is not zero.
The previous approach managed the thread name in a separate buffer, which causes
races because the thread name update (triggered by new samples) can happen at
the same time as prof dumping (which reads the thread names) -- these two
operations are under separate locks to avoid blocking each other. Implemented
the thread name storage as part of the tdata struct, which resolves the lifetime
issue and also avoids internal alloc / dalloc during prof_sample.
The added hooks hooks.prof_sample and hooks.prof_sample_free are intended to
allow advanced users to track additional information, to enable new ways of
profiling on top of the jemalloc heap profile and sample features.
The sample hook is invoked after the allocation and backtracing, and forwards
the both the allocation and backtrace to the user hook; the sample_free hook
happens before the actual deallocation, and forwards only the ptr and usz to the
hook.
Previously if a thread does only allocations, it stays on the slow path /
minimal initialized state forever. However, dealloc-only is a valid pattern for
dedicated reclamation threads -- this means thread cache is disabled (no batched
flush) for them, which causes high overhead and contention.
Added the condition to fully initialize TSD when a fair amount of dealloc
activities are observed.
Add new runtime option `debug_double_free_max_scan` that specifies the max
number of stack entries to scan in the cache bit when trying to detect the
double free bug (currently debug build only).
Due to a bug in sec initialization, the number of cached size classes
was equal to 198. The bug caused the creation of more than a hundred of
unused bins, although it didn't affect the caching logic.
The option makes the process to exit with error code 1 if a memory leak
is detected. This is useful for implementing automated tools that rely
on leak detection.
Under high concurrency / heavy test load (e.g. using run_tests.sh), the
background thread may not get scheduled for a longer period of time. Retry 100
times max before bailing out.
On deallocation, sampled pointers (specially aligned) get junked and stashed
into tcache (to prevent immediate reuse). The expected behavior is to have
read-after-free corrupted and stopped by the junk-filling, while
write-after-free is checked when flushing the stashed pointers.
Many profiling related tests make assumptions on the profiling settings,
e.g. opt_prof is off by default, and prof_active is default on when opt_prof is
on. However the default settings can be changed via --with-malloc-conf at build
time. Fixing the tests by adding the assumed settings explicitly.
nstime module guarantees monotonic clock update within a single nstime_t. This
means, if two separate nstime_t variables are read and updated separately,
nstime_subtract between them may result in underflow. Fixed by switching to the
time since utility provided by nstime.
Currently used only for guarding purposes, the hint is used to determine
if the allocation is supposed to be frequently reused. For example, it
might urge the allocator to ensure the allocation is cached.
While initially this file contained helper functions for one particular
test, now its usage spread across different test files. Purpose has
shifted towards a collection of handy arena ctl wrappers.
With prof enabled, number of page aligned allocations doesn't match the
number of slab "ends" because prof allocations skew the addresses. It
leads to 'pages' array overflow and hard to debug failures.
