add rustc option -Zpacked-stack by fneddy · Pull Request #152432 · rust-lang/rust

fneddy · 2026-02-10T10:19:08Z

this enables -Zpacked-stack just as -mpacked-stack in clang and gcc. packed-stack is needed on s390x for kernel development.

For reference: #151154 and #150766

look at @uweigand s post for full explanation of what this does. Here a wrap-up:

#150766 (comment)

[...]
packed-stack [...] modifies how the compiler-generated function prolog/epilog code makes use of the 160 byte register save area provided by a caller to the callee [...] this variant is not actually incompatible with the ABI - packed-stack and regular functions can freely call each other without ABI issues.
[...]
combination of -mpacked-stack and -mbackchain [...] the location in the stack frame where the backchain link ought to be stored is not available. [...] is not supported at all with the default ABI
[...]
However, in the special case of also using soft-float, our (implied) soft-float ABI provides a different location for the backchain that is compatible with -mpacked-stack, so that combination should be supported
[...]

rustbot · 2026-02-10T10:19:14Z

r? @chenyukang

rustbot has assigned @chenyukang.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

Why was this reviewer chosen?

The reviewer was selected based on:

Owners of files modified in this PR: compiler
compiler expanded to 66 candidates
Random selection from 14 candidates

fneddy · 2026-02-10T10:21:56Z

CC @cuviper @RalfJung @folkertdev @workingjubilee

fneddy · 2026-02-10T10:25:16Z

I'll add tests for assembly output and for target incompatibility.

compiler/rustc_codegen_llvm/src/attributes.rs

compiler/rustc_target/src/spec/mod.rs

compiler/rustc_session/src/session.rs

src/doc/rustc/src/codegen-options/index.md

fneddy · 2026-02-16T09:38:59Z

ok, i think most should be ok now.

compiler/rustc_interface/src/util.rs

RalfJung · 2026-02-16T09:50:38Z

compiler/rustc_interface/src/util.rs

+    if sess.target.arch == Arch::S390x
+        && sess.opts.unstable_opts.packed_stack
+        && sess.unstable_target_features.contains(&Symbol::intern(&"backchain"))
+        && sess.target.abi != Abi::SoftFloat


target.abi is just in charge of cfg(target_abi), it doesn't control the actual ABI. Please check rustc_abi which controls the actual ABI.

changed it to check on the soft-float unstable_target_feature which llvm will evaluate to switch abi for s390x.

I think checking rustc_abi makes most sense.

tests/ui/target-feature/packedstack-combinations.rs

compiler/rustc_interface/src/errors.rs

RalfJung · 2026-02-16T09:54:36Z

compiler/rustc_interface/src/util.rs

+    // this is the first place in the `run_compiler` flow where all this is available
+    if sess.target.arch == Arch::S390x
+        && sess.opts.unstable_opts.packed_stack
+        && sess.unstable_target_features.contains(&Symbol::intern(&"backchain"))


What happens if someone tries to enable backchain just for a particular function via #[target_feature] instead of -Ctarget-feature?

ufff yes we have to check this for every call to llfn_attrs_from_instance at compiler runtime. is it ok to do ....contains(&Symbol::intern(... every time?

from_target_feature_attr is probably a better, higher-level place to check this?

You can add backchain as a pre-interned symbol in compiler/rustc_span/src/symbol.rs so you don't have to re-intern it every time.

I am back on working on this. yes in from_target_feature_attr we can check for the #[target_feature(...)] attributes. BUT unfortunately not for the -Ctarget-feature... attributes ..

Yeah the handling for -Ctarget-feature and #[target_feature] is entirely separate. Maybe it's possible to introduce a new helper that gets called form those 2 places and that performs these kinds of consistency checks?

Ah I see you're now doing the check in packed_stack_attr. That also seems reasonable.

fneddy · 2026-03-20T13:18:25Z

Kind reminder this PR exists and breaks every now and then due to main -changes.

RalfJung · 2026-03-20T14:26:06Z

r? @nikic

this enables packed-stack just as -mpacked-stack in clang and gcc. packed-stack is needed on s390x for kernel development. Co-authored-by: Ralf Jung <post@ralfj.de>

rustbot · 2026-03-31T07:09:45Z

This PR was rebased onto a different main commit. Here's a range-diff highlighting what actually changed.

Rebasing is a normal part of keeping PRs up to date, so no action is needed—this note is just to help reviewers.

fneddy · 2026-03-31T07:27:22Z

@rustbot ready

nikic · 2026-03-31T09:22:51Z

The interaction with LLVM looks fine.

A question I have is whether it is normal to emit errors during codegen attribute lowering?

Do I understand correctly that if -Z packed-stack is used, then any use of #[target_feature(enable = "backchain")] becomes a hard error? That seems quite unusual. (Does anyone have a reference for why s390x has a per-function backchain feature that is not integrated with the usual frame pointer options?)

RalfJung · 2026-03-31T09:35:18Z

Do I understand correctly that if -Z packed-stack is used, then any use of #[target_feature(enable = "backchain")] becomes a hard error? That seems quite unusual.

Yeah it's a hack to deal with a somewhat ill-behaved target.

Can you think of a better way of dealing with this?

nikic · 2026-03-31T11:03:46Z

Do I understand correctly that if -Z packed-stack is used, then any use of #[target_feature(enable = "backchain")] becomes a hard error? That seems quite unusual.

Yeah it's a hack to deal with a somewhat ill-behaved target.

Can you think of a better way of dealing with this?

Not having backchain as a per-function target feature would be one way... Another would be to have packed-stack as part of the target definition.

fneddy · 2026-03-31T11:41:17Z

Options:

disallow `backchain`- per-function

this would be via an explicit check during parsing or apply of codegen_fn_attrs and error if it is set? I don't see a way to mark a target_feature as allowed via target_spec_file AND cli AND NOT inline

`packed_stack`-part of target definition

I think we don't want to add packed-stack to any specific target. A dev could decide to use it with the regular target or with the softfloat target. Also packed-stack is currently NOT part of the target_features in llvm. Its just a frontend-flag that will set the function attribute. So we cannot add packed-stack to IBMZ_FEATURES.

Sidenote

If one tries to feed the backend with a function where backchain and packed-stack and not soft-float is set the backend will panic.

uweigand · 2026-03-31T11:46:34Z

Does anyone have a reference for why s390x has a per-function backchain feature that is not integrated with the usual frame pointer options?

s390x has a frame pointer, which is used for accessing local variables and stack slots in functions using variable-sized stack allocation - the main purpose of frame pointers everywhere. On some other platforms (in particular x86 and arm), the frame pointer can also be used for stack unwinding, but that does not work on s390x, because our frame pointer indicates the bottom of the stack frame (like the stack pointer), not the top of the stack frame. The main reason for this is that some instructions do not support negative displacements, so a pointer to the top of the stack frame would be awkward to use.

Because the frame pointer is not usable for stack unwinding, there's no point in trying to use something like -fno-omit-frame-pointer on s390x. Instead, we have a separate stack backchain feature (which is unrelated to the frame pointer) that can be activated via -mbackchain; this is only used for frame unwinding in certain scenarios where DWARF CFI is not available, and it is not active by default.

uweigand · 2026-03-31T11:59:15Z

Not sure whether the per-function notion is a particular problem, but this aspect isn't really important - users typically enable backchain for the whole compilation unit. The fact that this is modeled in LLVM as a "target feature" is really more of an implementation detail. If this simplifies the Rust logic, I'd also be fine with only supporting a command line option like -Z backchain and prohibit use of the target feature.

The one important use case is the Linux kernel, which is fully built with backchain, packed-stack, and soft-float. This combination must work (and be active for the kernel target).

RalfJung · 2026-03-31T12:01:00Z

I don't see a way to mark a target_feature as allowed via target_spec_file AND cli AND NOT inline

Yeah this is currently not a thing. Conceptually it's not that hard to add, but doing it in a way that the target feature table does not become a lot more verbose could be tricky.

This is all still unstable obviously so we could also just leave a note on the backchain tracking issue that this should probably not be stabilized for per-function use, only for -C target-feature. (EDIT: I added a note in #150259.)

nikic · 2026-03-31T13:00:31Z

Because the frame pointer is not usable for stack unwinding, there's no point in trying to use something like -fno-omit-frame-pointer on s390x. Instead, we have a separate stack backchain feature (which is unrelated to the frame pointer) that can be activated via -mbackchain; this is only used for frame unwinding in certain scenarios where DWARF CFI is not available, and it is not active by default.

I think in practice, "I want a frame pointer" really means "I want non-DWARF unwind", which is why I consider frame pointers and backchain to be the same thing, even if they technically aren't... But I guess the ship here has sailed in terms of option naming.

nikic · 2026-03-31T13:01:10Z

@bors r+

As everything involved here is unstable anyway, I think this is fine to land as-is...

rust-bors · 2026-03-31T13:01:13Z

📌 Commit f39fa9e has been approved by nikic

It is now in the queue for this repository.

…uwer Rollup of 4 pull requests Successful merges: - #150752 (Update libc to v0.2.183) - #152432 (add rustc option -Zpacked-stack) - #154634 (Use `Hcx`/`hcx` consistently for `StableHashingContext`.) - #154635 (./x run miri: default to edition 2021)

@uweigand

Rollup merge of #152432 - fneddy:s390x_packed_stack_attribute, r=nikic add rustc option -Zpacked-stack this enables `-Zpacked-stack` just as `-mpacked-stack` in clang and gcc. packed-stack is needed on s390x for kernel development. For reference: #151154 and #150766 look at @uweigand s post for full explanation of what this does. Here a wrap-up: #150766 (comment) > [...] > packed-stack [...] modifies how the compiler-generated function prolog/epilog code makes use of the 160 byte register save area provided by a caller to the callee [...] this variant is not actually incompatible with the ABI - packed-stack and regular functions can freely call each other without ABI issues. > [...] > combination of -mpacked-stack and -mbackchain [...] the location in the stack frame where the backchain link ought to be stored is not available. [...] is not supported at all with the default ABI > [...] > However, in the special case of also using soft-float, our (implied) soft-float ABI provides a different location for the backchain that is compatible with -mpacked-stack, so that combination should be supported > [...]

…uwer Rollup of 4 pull requests Successful merges: - rust-lang/rust#150752 (Update libc to v0.2.183) - rust-lang/rust#152432 (add rustc option -Zpacked-stack) - rust-lang/rust#154634 (Use `Hcx`/`hcx` consistently for `StableHashingContext`.) - rust-lang/rust#154635 (./x run miri: default to edition 2021)

rustbot assigned chenyukang Feb 10, 2026

fneddy marked this pull request as draft February 10, 2026 10:24

rustbot added S-waiting-on-author Status: This is awaiting some action (such as code changes or more information) from the author. and removed S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. labels Feb 10, 2026

This comment has been minimized.

Sign in to view

bjorn3 reviewed Feb 10, 2026

View reviewed changes

compiler/rustc_codegen_llvm/src/attributes.rs Show resolved Hide resolved

fneddy force-pushed the s390x_packed_stack_attribute branch from 5ddcabb to 4ebec9d Compare February 10, 2026 11:29

RalfJung reviewed Feb 10, 2026

View reviewed changes

compiler/rustc_target/src/spec/mod.rs Outdated Show resolved Hide resolved

compiler/rustc_session/src/session.rs Show resolved Hide resolved

src/doc/rustc/src/codegen-options/index.md Outdated Show resolved Hide resolved

fneddy mentioned this pull request Feb 11, 2026

remove redundant backchain attribute in codegen #152486

Merged

fneddy marked this pull request as ready for review February 13, 2026 11:19

rustbot added S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. and removed S-waiting-on-author Status: This is awaiting some action (such as code changes or more information) from the author. labels Feb 13, 2026

This comment has been minimized.

Sign in to view

fneddy force-pushed the s390x_packed_stack_attribute branch from 0d6d167 to 0c182af Compare February 13, 2026 12:58

This comment has been minimized.

Sign in to view

fneddy force-pushed the s390x_packed_stack_attribute branch from 0c182af to 7fef6ea Compare February 16, 2026 09:35

This comment has been minimized.

Sign in to view

fneddy force-pushed the s390x_packed_stack_attribute branch from 7fef6ea to 31a40dc Compare February 16, 2026 09:50

RalfJung reviewed Feb 16, 2026

View reviewed changes

This comment has been minimized.

Sign in to view

fneddy force-pushed the s390x_packed_stack_attribute branch from d886fbf to f39c52b Compare February 16, 2026 14:18

fneddy force-pushed the s390x_packed_stack_attribute branch from e9e5f4c to fd39c98 Compare March 20, 2026 13:17

This comment has been minimized.

Sign in to view

rustbot assigned nikic and unassigned davidtwco Mar 20, 2026

This comment has been minimized.

Sign in to view

add rustc option -Zpacked-stack

f39fa9e

this enables packed-stack just as -mpacked-stack in clang and gcc. packed-stack is needed on s390x for kernel development. Co-authored-by: Ralf Jung <post@ralfj.de>

fneddy force-pushed the s390x_packed_stack_attribute branch from fd39c98 to f39fa9e Compare March 31, 2026 07:09

RalfJung mentioned this pull request Mar 31, 2026

Tracking Issue for s390x_target_feature #150259

Open

4 tasks

rust-bors bot added S-waiting-on-bors Status: Waiting on bors to run and complete tests. Bors will change the label on completion. and removed S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. labels Mar 31, 2026

JonathanBrouwer mentioned this pull request Mar 31, 2026

Rollup of 4 pull requests #154638

Merged

rust-bors bot merged commit f883d62 into rust-lang:main Mar 31, 2026
11 checks passed

rustbot added this to the 1.96.0 milestone Mar 31, 2026

Uh oh!

Conversation

fneddy commented Feb 10, 2026 • edited by rustbot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

rustbot commented Feb 10, 2026

Uh oh!

fneddy commented Feb 10, 2026

Uh oh!

fneddy commented Feb 10, 2026

Uh oh!

This comment has been minimized.

Uh oh!

Uh oh!

Uh oh!

Uh oh!

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

fneddy commented Feb 16, 2026

Uh oh!

This comment has been minimized.

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

fneddy Feb 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

This comment has been minimized.

This comment has been minimized.

fneddy commented Mar 20, 2026

Uh oh!

RalfJung commented Mar 20, 2026

Uh oh!

This comment has been minimized.

rustbot commented Mar 31, 2026

Uh oh!

fneddy commented Mar 31, 2026

Uh oh!

nikic commented Mar 31, 2026

Uh oh!

RalfJung commented Mar 31, 2026

Uh oh!

nikic commented Mar 31, 2026

Uh oh!

fneddy commented Mar 31, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Options:

disallow backchain- per-function

packed_stack-part of target definition

Sidenote

Uh oh!

uweigand commented Mar 31, 2026

Uh oh!

uweigand commented Mar 31, 2026

Uh oh!

RalfJung commented Mar 31, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

nikic commented Mar 31, 2026

Uh oh!

fneddy commented Feb 10, 2026 •

edited by rustbot

Loading

fneddy Feb 24, 2026 •

edited

Loading

fneddy commented Mar 31, 2026 •

edited

Loading

disallow `backchain`- per-function

`packed_stack`-part of target definition

RalfJung commented Mar 31, 2026 •

edited

Loading