| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
to reflect the new license.
We understand that people may be surprised that we're moving the header
entirely to discuss the new license. We checked this carefully with the
Foundation's lawyer and we believe this is the correct approach.
Essentially, all code in the project is now made available by the LLVM
project under our new license, so you will see that the license headers
include that license only. Some of our contributors have contributed
code under our old license, and accordingly, we have retained a copy of
our old license notice in the top-level files in each project and
repository.
llvm-svn: 351636
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Summary:
There is currently too much redundancy in the class/variable/* names in Scudo:
- we are in the namespace `__scudo`, so there is no point in having something
named `ScudoX` to end up with a final name of `__scudo::ScudoX`;
- there are a lot of types/* that have `Allocator` in the name, given that
Scudo is an allocator I figure this doubles up as well.
So change a bunch of the Scudo names to make them shorter, less redundant, and
overall simpler. They should still be pretty self explaining (or at least it
looks so to me).
The TSD part will be done in another CL (eg `__scudo::ScudoTSD`).
Reviewers: alekseyshl, eugenis
Reviewed By: alekseyshl
Subscribers: delcypher, #sanitizers, llvm-commits
Differential Revision: https://reviews.llvm.org/D49505
llvm-svn: 337557
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Summary:
The shared TSD model in its current form doesn't scale. Here is an example of
rpc2-benchmark (with default parameters, which is threading heavy) on a 72-core
machines (defaulting to a `CompactSizeClassMap` and no Quarantine):
- with tcmalloc: 337K reqs/sec, peak RSS of 338MB;
- with scudo (exclusive): 321K reqs/sec, peak RSS of 637MB;
- with scudo (shared): 241K reqs/sec, peak RSS of 324MB.
This isn't great, since the exclusive model uses a lot of memory, while the
shared model doesn't even come close to be competitive.
This is mostly due to the fact that we are consistently scanning the TSD pool
starting at index 0 for an available TSD, which can result in a lot of failed
lock attempts, and touching some memory that needs not be touched.
This CL attempts to make things better in most situations:
- first, use a thread local variable on Linux (intead of pthread APIs) to store
the current TSD in the shared model;
- move the locking boolean out of the TSD: this allows the compiler to use a
register and potentially optimize out a branch instead of reading it from the
TSD everytime (we also save a tiny bit of memory per TSD);
- 64-bit atomic operations on 32-bit ARM platforms happen to be expensive: so
store the `Precedence` in a `uptr` instead of a `u64`. We lose some
nanoseconds of precision and we'll wrap around at some point, but the benefit
is worth it;
- change a `CHECK` to a `DCHECK`: this should never happen, but if something is
ever terribly wrong, we'll crash on a near null AV if the TSD happens to be
null;
- based on an idea by dvyukov@, we are implementing a bound random scan for
an available TSD. This requires computing the coprimes for the number of TSDs,
and attempting to lock up to 4 TSDs in an random order before falling back to
the current one. This is obviously slightly more expansive when we have just
2 TSDs (barely noticeable) but is otherwise beneficial. The `Precedence` still
basically corresponds to the moment of the first contention on a TSD. To seed
on random choice, we use the precedence of the current TSD since it is very
likely to be non-zero (since we are in the slow path after a failed `tryLock`)
With those modifications, the benchmark yields to:
- with scudo (shared): 330K reqs/sec, peak RSS of 327MB.
So the shared model for this specific situation not only becomes competitive but
outperforms the exclusive model. I experimented with some values greater than 4
for the number of TSDs to attempt to lock and it yielded a decrease in QPS. Just
sticking with the current TSD is also a tad slower. Numbers on platforms with
less cores (eg: Android) remain similar.
Reviewers: alekseyshl, dvyukov, javed.absar
Reviewed By: alekseyshl, dvyukov
Subscribers: srhines, kristof.beyls, delcypher, llvm-commits, #sanitizers
Differential Revision: https://reviews.llvm.org/D47289
llvm-svn: 334410
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Summary:
See D40657 & D40679 for previous versions of this patch & description.
A couple of things were fixed here to have it not break some bots.
Weak symbols can't be used with `SANITIZER_GO` so the previous version was
breakin TsanGo. I set up some additional local tests and those pass now.
I changed the workaround for the glibc vDSO issue: `__progname` is initialized
after the vDSO and is actually public and of known type, unlike
`__vdso_clock_gettime`. This works better, and with all compilers.
The rest is the same.
Reviewers: alekseyshl
Reviewed By: alekseyshl
Subscribers: srhines, kubamracek, krytarowski, llvm-commits, #sanitizers
Differential Revision: https://reviews.llvm.org/D41121
llvm-svn: 320594
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
Summary: D40679 broke a couple of builds, reverting while investigating.
Reviewers: alekseyshl
Reviewed By: alekseyshl
Subscribers: srhines, kubamracek, krytarowski, llvm-commits, #sanitizers
Differential Revision: https://reviews.llvm.org/D41088
llvm-svn: 320417
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
allocator [redo]
Summary:
Redo of D40657, which had the initial discussion. The initial code had to move
into a libcdep file, and things had to be shuffled accordingly.
`NanoTime` is a time sink when checking whether or not to release memory to
the OS. While reducing the amount of calls to said function is in the works,
another solution that was found to be beneficial was to use a timing function
that can leverage the vDSO.
We hit a couple of snags along the way, like the fact that the glibc crashes
when clock_gettime is called from a preinit_array, or the fact that
`__vdso_clock_gettime` is mangled (for security purposes) and can't be used
directly, and also that clock_gettime can be intercepted.
The proposed solution takes care of all this as far as I can tell, and
significantly improve performances and some Scudo load tests with memory
reclaiming enabled.
@mcgrathr: please feel free to follow up on
https://reviews.llvm.org/D40657#940857 here. I posted a reply at
https://reviews.llvm.org/D40657#940974.
Reviewers: alekseyshl, krytarowski, flowerhack, mcgrathr, kubamracek
Reviewed By: alekseyshl, krytarowski
Subscribers: #sanitizers, mcgrathr, srhines, llvm-commits, kubamracek
Differential Revision: https://reviews.llvm.org/D40679
llvm-svn: 320409
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Summary:
It was deemed that the salt in the chunk header didn't improve security
significantly (and could actually decrease it). The initial idea was that the
same chunk would different headers on different allocations, allowing for less
predictability. The issue is that gathering the same chunk header with different
salts can give information about the other "secrets" (cookie, pointer), and that
if an attacker leaks a header, they can reuse it anyway for that same chunk
anyway since we don't enforce the salt value.
So we get rid of the salt in the header. This means we also get rid of the
thread local Prng, and that we don't need a global Prng anymore as well. This
makes everything faster.
We reuse those 8 bits to store the `ClassId` of a chunk now (0 for a secondary
based allocation). This way, we get some additional speed gains:
- `ClassId` is computed outside of the locked block;
- `getActuallyAllocatedSize` doesn't need the `GetSizeClass` call;
- same for `deallocatePrimary`;
We add a sanity check at init for this new field (all sanity checks are moved
in their own function, `init` was getting crowded).
Reviewers: alekseyshl, flowerhack
Reviewed By: alekseyshl
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D40796
llvm-svn: 319791
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Summary:
This first part just prepares the grounds for part 2 and doesn't add any new
functionality. It mostly consists of small refactors:
- move the `pthread.h` include higher as it will be used in the headers;
- use `errno.h` in `scudo_allocator.cpp` instead of the sanitizer one, update
the `errno` assignments accordingly (otherwise it creates conflicts on some
platforms due to `pthread.h` including `errno.h`);
- introduce and use `getCurrentTSD` and `setCurrentTSD` for the shared TSD
model code;
Reviewers: alekseyshl
Reviewed By: alekseyshl
Subscribers: llvm-commits, srhines
Differential Revision: https://reviews.llvm.org/D38826
llvm-svn: 315583
|
|
|
Summary:
Previous parts: D38139, D38183.
In this part of the refactor, we abstract the Linux vs Android TSD dissociation
in favor of a Exclusive vs Shared one, allowing for easier platform introduction
and configuration.
Most of this change consist of shuffling the files around to reflect the new
organization.
We introduce `scudo_platform.h` where platform specific definition lie. This
involves the TSD model and the platform specific allocator parameters. In an
upcoming CL, those will be configurable via defines, but we currently stick
with conservative defaults.
Reviewers: alekseyshl, dvyukov
Reviewed By: alekseyshl, dvyukov
Subscribers: srhines, llvm-commits, mgorny
Differential Revision: https://reviews.llvm.org/D38244
llvm-svn: 314224
|
