Use correct thread stack limits in jl_init_stack_limits #54639
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Actually these do not really seem to matter much in practice. The one thing that did cause me some issues here, though, was that stack_hi was the low address and stack_lo the high address.
vtjnash
reviewed
May 31, 2024
| *stack_lo = (void*)stackaddr; | ||
| *stack_hi = (void*)__builtin_frame_address(0); | ||
| size_t stacksize = pthread_get_stacksize_np(thread); | ||
| *stack_hi = stackaddr; |
There was a problem hiding this comment.
Don't you meant to switch these, since the malloc region is stackaddr to stackaddr+stacksize?
There was a problem hiding this comment.
Maybe I am misunderstanding you (sorry), but: stacks on all OS/CPU combos supported by Julia grow downwards, the stack starts at the highest address stackaddr and grows downwards towards stackaddr-stacksize (right after which the guard pages comes by default, at least for pthread created stacks with default settings).
I double checked this in darwin_libpthread, glibc and muslc and via real-world experiments on darwin/macos and linux+glibc (no muslc for me
There was a problem hiding this comment.
@vtjnash does my explanation make sense to you? Any other questions?
There was a problem hiding this comment.
Okay, yeah, it is confusing that it is inverted from the values returned by pthread_attr_getstack, but apparently that is the legacy behavior
vtjnash
added a commit
that referenced
this pull request
Aug 16, 2024
Move the registers onto the stack, so that they only are present when the Task is actually switched out, saving memory when the Task is not running yet or already finished. On Linux x86_64 this reduces it from 376 bytes to 184 bytes. Has some additional advantages too, such as copy_stack tasks (e.g. with always_copy_stacks) can migrate to other threads before starting if they are not stick. Also fixes a variable that got mixed up by #54639 and caused always_copy_stacks to abort, since the stack limits were wrong.
vtjnash
added a commit
that referenced
this pull request
Aug 17, 2024
Move the registers onto the stack, so that they only are present when the Task is actually switched out, saving memory when the Task is not running yet or already finished. On Linux x86_64 this reduces it from 376 bytes to 184 bytes. Has some additional advantages too, such as copy_stack tasks (e.g. with always_copy_stacks) can migrate to other threads before starting if they are not stick. Also fixes a variable that got mixed up by #54639 and caused always_copy_stacks to abort, since the stack limits were wrong.
vtjnash
added a commit
that referenced
this pull request
Aug 19, 2024
Move the registers onto the stack, so that they only are present when the Task is actually switched out, saving memory when the Task is not running yet or already finished. On Linux x86_64 this reduces it from 376 bytes to 184 bytes. Has some additional advantages too, such as copy_stack tasks (e.g. with always_copy_stacks) can migrate to other threads before starting if they are not stick. Also fixes a variable that got mixed up by #54639 and caused always_copy_stacks to abort, since the stack limits were wrong.
vtjnash
added a commit
that referenced
this pull request
Aug 20, 2024
Move the registers onto the stack, so that they only are present when the Task is actually switched out, saving memory when the Task is not running yet or already finished. On Linux x86_64 this reduces it from 376 bytes to 184 bytes. Has some additional advantages too, such as copy_stack tasks (e.g. with always_copy_stacks) can migrate to other threads before starting if they are not stick. Also fixes a variable that got mixed up by #54639 and caused always_copy_stacks to abort, since the stack limits were wrong.
vtjnash
added a commit
that referenced
this pull request
Aug 20, 2024
Move the registers onto the stack, so that they only are present when the Task is actually switched out, saving memory when the Task is not running yet or already finished. It makes this mostly just a huge renaming job. On Linux x86_64 this reduces it from 376 bytes to 184 bytes. Has some additional advantages too, such as copy_stack tasks (e.g. with always_copy_stacks) can migrate to other threads before starting if they are not sticky. Also fixes a variable that got mixed up by #54639 and caused always_copy_stacks to abort, since the stack limits were wrong. Also now fixes #43124, though I am not quite confident enough in it to re-enable that test right now.
KristofferC
pushed a commit
that referenced
this pull request
Sep 12, 2024
Move the registers onto the stack, so that they only are present when the Task is actually switched out, saving memory when the Task is not running yet or already finished. It makes this mostly just a huge renaming job. On Linux x86_64 this reduces it from 376 bytes to 184 bytes. Has some additional advantages too, such as copy_stack tasks (e.g. with always_copy_stacks) can migrate to other threads before starting if they are not sticky. Also fixes a variable that got mixed up by #54639 and caused always_copy_stacks to abort, since the stack limits were wrong. Also now fixes #43124, though I am not quite confident enough in it to re-enable that test right now.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Actually these do not really seem to matter much in practice (as explained to me by @vtjnash on Slack, thank you). The one thing that did cause me some grief here, though, was that stack_hi was the low address and stack_lo the high address.
A smaller change to achieve that "hi means high, lo means low" would be to just reverse
stack_hiandstack_lo.