- Jul 16, 2019
-
-
-
Currently initProfiling gets defined by Profiling.c only if PROFILING is defined. Otherwise the ProfHeap.c defines it. This is just needlessly complicated so in this commit I make Profiling and ProfHeap into properly seperate modules and call their respective init functions from RtsStartup.c.
-
- Jul 14, 2019
-
-
These are unexploded minds as far as the linter is concerned. I don't want to hit in my MRs by mistake! I did this with `sed`, and then rolled back some changes in the docs, config.guess, and the linter itself.
-
- Jul 10, 2019
-
-
These prevent multi-target builds. They were gotten rid of in 3 ways: 1. In the compiler itself, replacing `#if` with runtime `if`. In these cases, we care about the target platform still, but the target platform is dynamic so we must delay the elimination to run time. 2. In the compiler itself, replacing `TARGET` with `HOST`. There was just one bit of this, in some code splitting strings representing lists of paths. These paths are used by GHC itself, and not by the compiled binary. (They are compiler lookup paths, rather than RPATHS or something that does matter to the compiled binary, and thus would legitamentally be target-sensative.) As such, the path-splitting method only depends on where GHC runs and not where code it produces runs. This should have been `HOST` all along. 3. Changing the RTS. The RTS doesn't care about the target platform, full stop. 4. `includes/stg/HaskellMachRegs.h` This file is also included in the genapply executable. This is tricky because the RTS's host platform really is that utility's target platform. so that utility really really isn't multi-target either. But at least it isn't an installed part of GHC, but just a one-off tool when building the RTS. Lying with the `HOST` to a one-off program (genapply) that isn't installed doesn't seem so bad. It's certainly better than the other way around of lying to the RTS though not to genapply. The RTS is more important, and it is installed, *and* this header is installed as part of the RTS.
-
- Jul 05, 2019
-
-
In dumpCensus we switch/case on doHeapProfile twice. The second switch tries to barf on unknown doHeapProfile modes but HEAP_BY_CLOSURE_TYPE is checked by the first switch and not included in the second. So when trying to pass -hT to the profiling rts it barfs. This commit simply merges the two switches into one which fixes this problem.
-
-
- Jul 02, 2019
-
-
-
-
-
This adds lookup logic for _GLOBAL_OFFSET_TABLE_ as well as relocation logic for R_ARM_BASE_PREL and R_ARM_GOT_BREL which the gnu toolchain (gas, gcc, ...) prefers to produce. Apparently recent llvm toolchains will produce those as well.
-
- Jun 28, 2019
-
-
Ben Gamari authored
I'm not entirely sure we are careful about ensuring this; this is a last-ditch check.
-
Here the following changes are introduced: - A read barrier machine op is added to Cmm. - The order in which a closure's fields are read and written is changed. - Memory barriers are added to RTS code to ensure correctness on out-or-order machines with weak memory ordering. Cmm has a new CallishMachOp called MO_ReadBarrier. On weak memory machines, this is lowered to an instruction that ensures memory reads that occur after said instruction in program order are not performed before reads coming before said instruction in program order. On machines with strong memory ordering properties (e.g. X86, SPARC in TSO mode) no such instruction is necessary, so MO_ReadBarrier is simply erased. However, such an instruction is necessary on weakly ordered machines, e.g. ARM and PowerPC. Weam memory ordering has consequences for how closures are observed and mutated. For example, consider a closure that needs to be updated to an indirection. In order for the indirection to be safe for concurrent observers to enter, said observers must read the indirection's info table before they read the indirectee. Furthermore, the entering observer makes assumptions about the closure based on its info table contents, e.g. an INFO_TYPE of IND imples the closure has an indirectee pointer that is safe to follow. When a closure is updated with an indirection, both its info table and its indirectee must be written. With weak memory ordering, these two writes can be arbitrarily reordered, and perhaps even interleaved with other threads' reads and writes (in the absence of memory barrier instructions). Consider this example of a bad reordering: - An updater writes to a closure's info table (INFO_TYPE is now IND). - A concurrent observer branches upon reading the closure's INFO_TYPE as IND. - A concurrent observer reads the closure's indirectee and enters it. (!!!) - An updater writes the closure's indirectee. Here the update to the indirectee comes too late and the concurrent observer has jumped off into the abyss. Speculative execution can also cause us issues, consider: - An observer is about to case on a value in closure's info table. - The observer speculatively reads one or more of closure's fields. - An updater writes to closure's info table. - The observer takes a branch based on the new info table value, but with the old closure fields! - The updater writes to the closure's other fields, but its too late. Because of these effects, reads and writes to a closure's info table must be ordered carefully with respect to reads and writes to the closure's other fields, and memory barriers must be placed to ensure that reads and writes occur in program order. Specifically, updates to a closure must follow the following pattern: - Update the closure's (non-info table) fields. - Write barrier. - Update the closure's info table. Observing a closure's fields must follow the following pattern: - Read the closure's info pointer. - Read barrier. - Read the closure's (non-info table) fields. This patch updates RTS code to obey this pattern. This should fix long-standing SMP bugs on ARM (specifically newer aarch64 microarchitectures supporting out-of-order execution) and PowerPC. This fixes issue #15449. Co-Authored-By:Ben Gamari <ben@well-typed.com>
-
-
- Jun 27, 2019
-
-
It is important that `heapCensus` and `LdvCensusForDead` traverse the same areas. `heapCensus` increases the `not_used` counter which tracks how many closures are live but haven't been used yet. `LdvCensusForDead` increases the `void_total` counter which tracks how many dead closures there are. The `LAG` is then calculated by substracting the `void_total` from `not_used` and so it is essential that `not_used >= void_total`. This fact is checked by quite a few assertions. However, if a program has low maximum residency but allocates a lot in the nursery then these assertions were failing (see #16753 and #15903) because `LdvCensusForDead` was observing dead closures from the nursery which totalled more than the `not_used`. The same closures were not counted by `heapCensus`. Therefore, it seems that the correct fix is to make `LdvCensusForDead` agree with `heapCensus` and not traverse the nursery for dead closures. Fixes #16100 #16753 #15903 #8982
-
It is possible that void_total is exactly equal to not_used and the other assertions for this check for <= rather than <.
-
This implements the correct fix for #11627 by skipping over the slop (which is zeroed) rather than adding special case logic for LARGE ARR_WORDS which runs the risk of not performing a correct census by ignoring any subsequent blocks. This approach implements similar logic to that in Sanity.c
-
- Jun 26, 2019
-
-
Siddharth authored
-
- Jun 22, 2019
-
-
When we revert a CAF we must reset the STATIC_LINK field lest the GC might ignore the CAF (e.g. as it carries the STATIC_FLAG_LIST flag) and will consequently overlook references to object code that we are trying to unload. This would result in the reachable object code being unloaded. See Note [CAF lists] and Note [STATIC_LINK fields]. This fixes #16842. Idea-due-to: Phuong Trinh <lolotp@fb.com>
-
- Jun 13, 2019
-
-
`checkUnload` currently doesn't check the info header of static objects. Thus, it may free an `ObjectCode` struct wrongly even if there's still a live static object whose info header lies in a mapped section of that `ObjectCode`. This fixes the issue by adding an appropriate check.
-
- Jun 12, 2019
-
-
This fixes a regression, introduced by 67c422ca, where we mprotect'd the global offset table (GOT) region to PROT_READ before we had finished filling it, resulting in a linker crash. Fixes #16779.
-
-
- Jun 11, 2019
-
-
-
The PLT needs to be located within a close distance of the code calling it under the small memory model. Fixes #16784.
-
This extends mmapForLinker to use the same low-memory mapping strategy used on x86_64 on AArch64. See #16784.
-
-
- Jun 09, 2019
-
-
When pop() returns with `*c == NULL` retainerProfile will immediately return. All other code paths is pop() continue with the next stackElement when this happens so it seems weird to me that TREC_CHUNK we would suddenly abort everything even though the stack might still have elements left to process.
-
Previously these two orthogonal concerns were both implemented in postHeaderEvents which made it difficult to send header events after RTS initialization.
-
- Jun 08, 2019
-
-
- Jun 07, 2019
-
-
This allows a user to observe how long a sampling period lasts so that the time taken can be removed from the profiling output. Fixes #16697
-
- Jun 01, 2019
-
-
- May 31, 2019
-
-
- May 30, 2019
-
-
-
-
We are iterating through all object code for each heap objects when checking whether object code can be unloaded. For large projects in GHCi, this can be very expensive due to the large number of object code that needs to be loaded/unloaded. To speed it up, this arrangess all mapped sections of unloaded object code in a sorted array and use binary search to check if an address location fall on them.
-
As noted in #16701, it is possible that we will find that an object has no segments needing to be mapped. Previously this would result in mmap being called for a zero-length mapping, which would fail. We now simply skip the mmap call in this case; the rest of the logic just works.
-
- May 29, 2019
-
-
- May 27, 2019
-
-
When the number of entries of a cost centre reaches 11 digits, it takes up the whole space reserved for it and the prof file ends up looking like: ... no. entries %time %alloc %time %alloc ... ... 120918 978250 0.0 0.0 0.0 0.0 ... 118891 0 0.0 0.0 73.3 80.8 ... 11890229702412351 8.9 13.5 73.3 80.8 ... 118903 153799689 0.0 0.1 0.0 0.1 ... This results in tooling not being able to parse the .prof file. I realise we have the JSON output as well now, but still it'd be good to fix this little weirdness. Original bug report and full prof file can be seen here: <https://github.com/jaspervdj/profiteur/issues/28>.
-
- May 25, 2019
-
-
- May 22, 2019
-
-
Commit e75a9afd added an `unsigned` cast to account for OSes that have signed `rlim_t` signed. Unfortunately, the `unsigned` cast has the unintended effect of narrowing `rlim_t` to only 4 bytes. This leads to some spurious out of memory crashes (in particular: Haddock crashes with OOM whenn building docs of `ghc`-the-library). In this case, `W_` is a better type to cast to: we know it will be unsigned too and it has the same type as `*len` (so we don't suffer from accidental narrowing).
-
