Kbuild: fix issues with rustc-option
Alice Ryhl
reviewerPATCH v2 1/1 Kbuild: fix issues with rustc-option
Fix a few different compiler errors that cause rustc-option to give wrong results.
If KBUILD_RUSTFLAGS or the flags being tested contain any -Z flags, then the error below is generated. The RUSTC_BOOTSTRAP environment variable is added to fix this error.
error: the option Z is only accepted on the nightly compiler
help: consider switching to a nightly toolchain: rustup default nightly
note: selecting a toolchain with +toolchain arguments require a rustup proxy;
see <https://rust-lang.github.io/rustup/concepts/index.html>
note: for more information about Rust's stability policy, see
<https://doc.rust-lang.org/book/appendix-07-nightly-rust.html#unstable-features>
error: 1 nightly option were parsed
The probe may also fail incorrectly with the below error message. To fix it, the /dev/null argument is replaced with a new rust/probe.rs file that doesn't need even the core part of the standard library.
error[E0463]: can't find crate for std
|
= note: the aarch64-unknown-none target may not be installed
= help: consider downloading the target with rustup target add aarch64-unknown-none = help: consider building the standard library from source with cargo build -Zbuild-std
The -o and --out-dir parameters are altered to fix this warning:
warning: ignoring --out-dir flag due to -o flag
I verified that the Kconfig version of rustc-option doesn't have the same issues.
---
Changes in v2:
- Add export to RUSTC_BOOTSTRAP.
- Fix error about core being missing.
- Fix warning about -o flag.
- Link to v1: https://lore.kernel.org/r/[email protected]
--- rust/probe.rs | 7 +++++++ scripts/Makefile.compiler | 5 +++-- 2 files changed, 10 insertions(+), 2 deletions(-)
rust/probe.rs new file +7 -0
@@ -0,0 +1,7 @@ +//! Nearly empty file passed to rustc-option by Make. +//! +//! The no_core attribute is needed because rustc-option otherwise fails due to +//! not being able to find the core part of the standard library. + +#![feature(no_core)] +#![no_core]
scripts/Makefile.compiler +3 -2
@@ -21,6 +21,7 @@ TMPOUT = $(if $(KBUILD_EXTMOD),$(firstword $(KBUILD_EXTMOD))/).tmp_$$$$ # automatically cleaned up. try-run = $(shell set -e; \ TMP=$(TMPOUT)/tmp; \ + export RUSTC_BOOTSTRAP=1; \ trap "rm -rf $(TMPOUT)" EXIT; \ mkdir -p $(TMPOUT); \ if ($(1)) >/dev/null 2>&1; \
@@ -76,7 +77,7 @@ ld-option = $(call try-run, $(LD) $(KBUILD_LDFLAGS) $(1) -v,$(1),$(2),$(3)) # __rustc-option # Usage: MY_RUSTFLAGS += $(call __rustc-option,$(RUSTC),$(MY_RUSTFLAGS),-Cinstrument-coverage,-Zinstrument-coverage) __rustc-option = $(call try-run,\ - $(1) $(2) $(3) --crate-type=rlib /dev/null --out-dir=$$TMPOUT -o "$$TMP",$(3),$(4)) + $(1) $(2) $(3) --crate-type=rlib $(srctree)/rust/probe.rs --out-dir=$$TMP,$(3),$(4)) # rustc-option # Usage: rustflags-y += $(call rustc-option,-Cinstrument-coverage,-Zinstrument-coverage)
@@ -86,4 +87,4 @@ rustc-option = $(call __rustc-option, $(RUSTC),\ # rustc-option-yn # Usage: flag := $(call rustc-option-yn,-Cinstrument-coverage) rustc-option-yn = $(call try-run,\ - $(RUSTC) $(KBUILD_RUSTFLAGS) $(1) --crate-type=rlib /dev/null --out-dir=$$TMPOUT -o "$$TMP",y,n) + $(RUSTC) $(KBUILD_RUSTFLAGS) $(1) --crate-type=rlib $(srctree)/rust/probe.rs --out-dir=$$TMP,y,n)
---
base-commit: 8cf0b93919e13d1e8d4466eb4080a4c4d9d66d7b
change-id: 20241008-rustc-option-bootstrap-607e5bf3114c
Best regards,
Signature · 1 line
Alice Ryhl <[email protected]>
Thanks for catching this - I had a specialized toolchain enabled which didn't exercise the RUSTC_BOOTSTRAP issue, and without an unexpected failure, didn't see the --out-dir flag conflict after I got the request to add it.
Quoted 95 lines · Alice Ryhl
Fix a few different compiler errors that cause rustc-option to give wrong results.
If
KBUILD_RUSTFLAGSor the flags being tested contain any -Z flags, then the error below is generated. TheRUSTC_BOOTSTRAPenvironment variable is added to fix this error.error: the option
Zis only accepted on the nightly compiler
help: consider switching to a nightly toolchain:rustup default nightly
note: selecting a toolchain with+toolchainarguments require a rustup proxy;
see <https://rust-lang.github.io/rustup/concepts/index.html>
note: for more information about Rust's stability policy, see
<https://doc.rust-lang.org/book/appendix-07-nightly-rust.html#unstable-features>
error: 1 nightly option were parsedThe probe may also fail incorrectly with the below error message. To fix it, the
/dev/nullargument is replaced with a newrust/probe.rsfile that doesn't need even the core part of the standard library.error[E0463]: can't find crate for
std
|
= note: theaarch64-unknown-nonetarget may not be installed
= help: consider downloading the target withrustup target add aarch64-unknown-none= help: consider building the standard library from source withcargo build -Zbuild-stdThe -o and --out-dir parameters are altered to fix this warning:
warning: ignoring --out-dir flag due to -o flag
I verified that the
Kconfigversion of rustc-option doesn't have the same issues.Fixes c42297438aee ("kbuild: rust: Define probing macros for rustc")Signed-off-by Alice Ryhl <[email protected]>---
Changes in v2:
- AddexporttoRUSTC_BOOTSTRAP.
- Fix error about core being missing.
- Fix warning about -o flag.
- Link to v1: https://lore.kernel.org/r/[email protected]--- rust/probe.rs | 7 +++++++ scripts/Makefile.compiler | 5 +++-- 2 files changed, 10 insertions(+), 2 deletions(-)rust/probe.rs new file +7
-0@@ -0,0 +1,7 @@ +//! Nearly empty file passed to rustc-option by Make. +//! +//! The no_core attribute is needed because rustc-option otherwise fails due to +//! not being able to find the core part of the standard library. + +#![feature(no_core)] +#![no_core] scripts/Makefile.compiler +3
-3@@ -21,6 +21,7 @@ TMPOUT = $(if $(KBUILD_EXTMOD),$(firstword $(KBUILD_EXTMOD))/).tmp_$$$$ # automatically cleaned up. try-run = $(shell set -e; \ TMP=$(TMPOUT)/tmp; \ + export RUSTC_BOOTSTRAP=1; \ trap "rm -rf $(TMPOUT)" EXIT; \ mkdir -p $(TMPOUT); \ if ($(1)) >/dev/null 2>&1; \@@ -76,7 +77,7 @@ ld-option = $(call try-run, $(LD) $(KBUILD_LDFLAGS) $(1) -v,$(1),$(2),$(3)) # __rustc-option # Usage: MY_RUSTFLAGS += $(call __rustc-option,$(RUSTC),$(MY_RUSTFLAGS),-Cinstrument-coverage,-Zinstrument-coverage) __rustc-option = $(call try-run,\ - $(1) $(2) $(3) --crate-type=rlib /dev/null --out-dir=$$TMPOUT -o "$$TMP",$(3),$(4)) + $(1) $(2) $(3) --crate-type=rlib $(srctree)/rust/probe.rs --out-dir=$$TMP,$(3),$(4)) # rustc-option # Usage: rustflags-y += $(call rustc-option,-Cinstrument-coverage,-Zinstrument-coverage)@@ -86,4 +87,4 @@ rustc-option = $(call __rustc-option, $(RUSTC),\ # rustc-option-yn # Usage: flag := $(call rustc-option-yn,-Cinstrument-coverage) rustc-option-yn = $(call try-run,\ - $(RUSTC) $(KBUILD_RUSTFLAGS) $(1) --crate-type=rlib /dev/null --out-dir=$$TMPOUT -o "$$TMP",y,n) + $(RUSTC) $(KBUILD_RUSTFLAGS) $(1) --crate-type=rlib $(srctree)/rust/probe.rs --out-dir=$$TMP,y,n) ---base-commit:
8cf0b93919e13d1e8d4466eb4080a4c4d9d66d7b
change-id: 20241008-rustc-option-bootstrap-607e5bf3114cBest regards,
--
Alice Ryhl <[email protected]>
Err, slight amendment - I think you want -o $$TMP, and not --out-dir $$TMPOUT
You definitely don't want --out-dir $$TMP, but the other two settings above would be defensible.
Quoted 105 lines · Matthew Maurer
Thanks for catching this - I had a specialized toolchain enabled which didn't exercise the
RUSTC_BOOTSTRAPissue, and without an unexpected failure, didn't see the--out-dirflag conflict after I got the request to add it.Reviewed-by Matthew Maurer <[email protected]>Quoted 95 lines · Alice Ryhl
Fix a few different compiler errors that cause rustc-option to give wrong results.
If
KBUILD_RUSTFLAGSor the flags being tested contain any -Z flags, then the error below is generated. TheRUSTC_BOOTSTRAPenvironment variable is added to fix this error.error: the option
Zis only accepted on the nightly compiler
help: consider switching to a nightly toolchain:rustup default nightly
note: selecting a toolchain with+toolchainarguments require a rustup proxy;
see <https://rust-lang.github.io/rustup/concepts/index.html>
note: for more information about Rust's stability policy, see
<https://doc.rust-lang.org/book/appendix-07-nightly-rust.html#unstable-features>
error: 1 nightly option were parsedThe probe may also fail incorrectly with the below error message. To fix it, the
/dev/nullargument is replaced with a newrust/probe.rsfile that doesn't need even the core part of the standard library.error[E0463]: can't find crate for
std
|
= note: theaarch64-unknown-nonetarget may not be installed
= help: consider downloading the target withrustup target add aarch64-unknown-none= help: consider building the standard library from source withcargo build -Zbuild-stdThe -o and --out-dir parameters are altered to fix this warning:
warning: ignoring --out-dir flag due to -o flag
I verified that the
Kconfigversion of rustc-option doesn't have the same issues.Fixes c42297438aee ("kbuild: rust: Define probing macros for rustc")Signed-off-by Alice Ryhl <[email protected]>---
Changes in v2:
- AddexporttoRUSTC_BOOTSTRAP.
- Fix error about core being missing.
- Fix warning about -o flag.
- Link to v1: https://lore.kernel.org/r/[email protected]--- rust/probe.rs | 7 +++++++ scripts/Makefile.compiler | 5 +++-- 2 files changed, 10 insertions(+), 2 deletions(-)rust/probe.rs new file +7
-0@@ -0,0 +1,7 @@ +//! Nearly empty file passed to rustc-option by Make. +//! +//! The no_core attribute is needed because rustc-option otherwise fails due to +//! not being able to find the core part of the standard library. + +#![feature(no_core)] +#![no_core] scripts/Makefile.compiler +3
-3@@ -21,6 +21,7 @@ TMPOUT = $(if $(KBUILD_EXTMOD),$(firstword $(KBUILD_EXTMOD))/).tmp_$$$$ # automatically cleaned up. try-run = $(shell set -e; \ TMP=$(TMPOUT)/tmp; \ + export RUSTC_BOOTSTRAP=1; \ trap "rm -rf $(TMPOUT)" EXIT; \ mkdir -p $(TMPOUT); \ if ($(1)) >/dev/null 2>&1; \@@ -76,7 +77,7 @@ ld-option = $(call try-run, $(LD) $(KBUILD_LDFLAGS) $(1) -v,$(1),$(2),$(3)) # __rustc-option # Usage: MY_RUSTFLAGS += $(call __rustc-option,$(RUSTC),$(MY_RUSTFLAGS),-Cinstrument-coverage,-Zinstrument-coverage) __rustc-option = $(call try-run,\ - $(1) $(2) $(3) --crate-type=rlib /dev/null --out-dir=$$TMPOUT -o "$$TMP",$(3),$(4)) + $(1) $(2) $(3) --crate-type=rlib $(srctree)/rust/probe.rs --out-dir=$$TMP,$(3),$(4)) # rustc-option # Usage: rustflags-y += $(call rustc-option,-Cinstrument-coverage,-Zinstrument-coverage)@@ -86,4 +87,4 @@ rustc-option = $(call __rustc-option, $(RUSTC),\ # rustc-option-yn # Usage: flag := $(call rustc-option-yn,-Cinstrument-coverage) rustc-option-yn = $(call try-run,\ - $(RUSTC) $(KBUILD_RUSTFLAGS) $(1) --crate-type=rlib /dev/null --out-dir=$$TMPOUT -o "$$TMP",y,n) + $(RUSTC) $(KBUILD_RUSTFLAGS) $(1) --crate-type=rlib $(srctree)/rust/probe.rs --out-dir=$$TMP,y,n) ---base-commit:
8cf0b93919e13d1e8d4466eb4080a4c4d9d66d7b
change-id: 20241008-rustc-option-bootstrap-607e5bf3114cBest regards,
--
Alice Ryhl <[email protected]>
Alice Ryhl
reviewerQuoted 6 lines · Matthew Maurer
Err, slight amendment - I think you want
-o $$TMP, and not--out-dir $$TMPOUTYou definitely don't want
--out-dir $$TMP, but the other two settings above would be defensible.
Ah good point. It seems like the reason that --out-dir $$TMP still works is that it creates a sub-directory. But I agree that this is not what we want.
Alice
Quoted 3 lines · Matthew Maurer
Err, slight amendment - I think you want
-o $$TMP, and not--out-dir $$TMPOUT
We need the --out-dir to avoid temporary files being created in the current working directory.
Cheers,
Miguel
Quoted 8 lines · Miguel Ojeda
Quoted 3 lines · Matthew Maurer
Err, slight amendment - I think you want
-o $$TMP, and not--out-dir $$TMPOUTWe need the
--out-dirto avoid temporary files being created in the current working directory.
In that case we should omit -o $$TMP, because rustc is emitting a warning that it's ignoring --out-dir because -o is set.
Quoted 3 lines
Cheers,
Miguel
Quoted 3 lines · Matthew Maurer
In that case we should omit
-o $$TMP, becauserustcis emitting a warning that it's ignoring--out-dirbecause-ois set.
Yes, we need to omit -o or we could use --out-dir together with one of the --emit=X=... types to be more explicit and avoid building the .rlib, which is what I did in the GitHub issue.
IIRC I used obj to run most of the compiler stages just in case that mattered, but perhaps an even simpler one is good enough, e.g. metadata.
I think just --out-dir should be fine and is simpler for this use case. However, apparently outputting to stdout works too, so we could do:
--emit=obj=- - >/dev/null
and avoid the output file altogether. We still most likely need the --out-dir in case temporaries are created.
Cheers,
Miguel
Alice Ryhl
reviewerQuoted 9 lines · Miguel Ojeda
Quoted 3 lines · Matthew Maurer
In that case we should omit
-o $$TMP, becauserustcis emitting a warning that it's ignoring--out-dirbecause-ois set.Yes, we need to omit
-oor we could use--out-dirtogether with one of the--emit=X=...types to be more explicit and avoid building the.rlib, which is what I did in the GitHub issue.
Miguel, can you link this issue? I don't think I saw it.
Quoted 4 lines
IIRC I used
objto run most of the compiler stages just in case that mattered, but perhaps an even simpler one is good enough, e.g.metadata.
Quoted 8 lines
I think just
--out-dirshould be fine and is simpler for this use case. However, apparently outputting to stdout works too, so we could do:--emit=obj=- - >/dev/nulland avoid the output file altogether. We still most likely need the
--out-dirin case temporaries are created.
Masahiro, are you able to clarify how to pass TMPOUT to rustc?
__rustc-option = $(call try-run2,\
$(1) $(2) $(3) --crate-type=rlib $(srctree)/rust/probe.rs--out-dir=$(TMPOUT),$(3),$(4))
Should I use $(TMPOUT) or $$TMPOUT for this case? Right now, only TMP is defined inside try-run. I am assuming that there is a reason for having TMP be defined in try-run, rather than just using $(TMP) everywhere. Does the same reason apply to TMPOUT? Should I add a TMPOUT=$(TMPOUT) inside try-run?
Alice
Quoted 2 lines · Alice Ryhl
Miguel, can you link this issue? I don't think I saw it.
https://github.com/Rust-for-Linux/linux/pull/1087#issuecomment-2218445303
(It was in the Lore message I linked yesterday, sorry, I should have been more explicit)
Quoted 11 lines
Masahiro, are you able to clarify how to pass TMPOUT to rustc?
__rustc-option = $(call try-run2,\ $(1) $(2) $(3) --crate-type=rlib $(srctree)/rust/probe.rs--out-dir=$(TMPOUT),$(3),$(4))
Should I use $(TMPOUT) or $$TMPOUT for this case? Right now, only TMP is defined inside try-run. I am assuming that there is a reason for having TMP be defined in try-run, rather than just using $(TMP) everywhere. Does the same reason apply to TMPOUT? Should I add a TMPOUT=$(TMPOUT) inside try-run?
TMPOUT is defined already in that Makefile, thus you can directly expand it. However, TMP is defined inside the shell function, and thus $$TMP is used so that that script (inside the shell) expands it instead.
This is why Masahiro was saying that the TMPOUT=$(TMPOUT) was unnecessary, i.e. it would work, but we can just expand it directly.
Something like this, combining everything [1] seems to work for me.
i.e. passing the file inline, RUSTC_BOOTSTRAP=1, avoiding an output file, keeping --out-dir for intermediates files. I added using a null sysroot too (and skipping if already given, since that is an error).
I will test it a bit more with KASAN etc.
Cheers,
Miguel
[1]
scripts/Makefile.compiler +0 -0
@@ -76,7 +76,9 @@ ld-option = $(call try-run, $(LD) $(KBUILD_LDFLAGS)
$(1) -v,$(1),$(2),$(3))
# __rustc-option
# Usage: MY_RUSTFLAGS += $(call
__rustc-option,$(RUSTC),$(MY_RUSTFLAGS),-Cinstrument-coverage,-Zinstrument-coverage)
__rustc-option = $(call try-run,\
- $(1) $(2) $(3) --crate-type=rlib /dev/null --out-dir=$$TMPOUT
-o "$$TMP",$(3),$(4))
+ echo '//!\n#![feature(no_core)]#![no_core]' | RUSTC_BOOTSTRAP=1\
+ $(1) --sysroot=/dev/null $(filter-out
--sysroot=/dev/null,$(2)) $(3)\
+ --crate-type=rlib --out-dir=$(TMPOUT) --emit=obj=- -
Quoted 1 line
/dev/null,$(3),$(4))
# rustc-option # Usage: rustflags-y += $(call rustc-option,-Cinstrument-coverage,-Zinstrument-coverage)
Quoted 22 lines · Miguel Ojeda
Quoted 2 lines · Alice Ryhl
Miguel, can you link this issue? I don't think I saw it.
https://github.com/Rust-for-Linux/linux/pull/1087#issuecomment-2218445303
(It was in the Lore message I linked yesterday, sorry, I should have been more explicit)
Quoted 11 lines
Masahiro, are you able to clarify how to pass TMPOUT to rustc?
__rustc-option = $(call try-run2,\ $(1) $(2) $(3) --crate-type=rlib $(srctree)/rust/probe.rs--out-dir=$(TMPOUT),$(3),$(4))
Should I use $(TMPOUT) or $$TMPOUT for this case? Right now, only TMP is defined inside try-run. I am assuming that there is a reason for having TMP be defined in try-run, rather than just using $(TMP) everywhere. Does the same reason apply to TMPOUT? Should I add a TMPOUT=$(TMPOUT) inside try-run?
Quoted 7 lines
TMPOUTis defined already in thatMakefile, thus you can directly expand it. However,TMPis defined inside theshellfunction, and thus$$TMPis used so that that script (inside theshell) expands it instead.This is why Masahiro was saying that the
TMPOUT=$(TMPOUT)was unnecessary, i.e. it would work, but we can just expand it directly.
Yes. I like --out-dir=$(TMPOUT)
Signature · 2 lines
Best Regards Masahiro Yamada
Quoted 62 lines · Miguel Ojeda
Quoted 2 lines · Alice Ryhl
Miguel, can you link this issue? I don't think I saw it.
https://github.com/Rust-for-Linux/linux/pull/1087#issuecomment-2218445303
(It was in the Lore message I linked yesterday, sorry, I should have been more explicit)
Quoted 11 lines
Masahiro, are you able to clarify how to pass TMPOUT to rustc?
__rustc-option = $(call try-run2,\ $(1) $(2) $(3) --crate-type=rlib $(srctree)/rust/probe.rs--out-dir=$(TMPOUT),$(3),$(4))
Should I use $(TMPOUT) or $$TMPOUT for this case? Right now, only TMP is defined inside try-run. I am assuming that there is a reason for having TMP be defined in try-run, rather than just using $(TMP) everywhere. Does the same reason apply to TMPOUT? Should I add a TMPOUT=$(TMPOUT) inside try-run?
TMPOUTis defined already in thatMakefile, thus you can directly expand it. However,TMPis defined inside theshellfunction, and thus$$TMPis used so that that script (inside theshell) expands it instead.This is why Masahiro was saying that the
TMPOUT=$(TMPOUT)was unnecessary, i.e. it would work, but we can just expand it directly.Something like this, combining everything [1] seems to work for me.
i.e. passing the file inline,
RUSTC_BOOTSTRAP=1, avoiding an output file, keeping--out-dirfor intermediates files. I added using a null sysroot too (and skipping if already given, since that is an error).I will test it a bit more with KASAN etc.
Cheers,
Miguel[1]
scripts/Makefile.compiler +3
-3@@ -76,7 +76,9 @@ ld-option = $(call try-run, $(LD) $(KBUILD_LDFLAGS) $(1) -v,$(1),$(2),$(3)) # __rustc-option # Usage: MY_RUSTFLAGS += $(call __rustc-option,$(RUSTC),$(MY_RUSTFLAGS),-Cinstrument-coverage,-Zinstrument-coverage) __rustc-option = $(call try-run,\ - $(1) $(2) $(3) --crate-type=rlib /dev/null --out-dir=$$TMPOUT -o "$$TMP",$(3),$(4)) + echo '//!\n#![feature(no_core)]#![no_core]' | RUSTC_BOOTSTRAP=1\ + $(1) --sysroot=/dev/null $(filter-out --sysroot=/dev/null,$(2)) $(3)\ + --crate-type=rlib --out-dir=$(TMPOUT) --emit=obj=- -Quoted 1 line
/dev/null,$(3),$(4))
Quoted 4 lines
# rustc-option # Usage: rustflags-y += $(call rustc-option,-Cinstrument-coverage,-Zinstrument-coverage)
Could you please add a comment line to remember the future clean-up?
e.g.
# TODO: remove RUSTC_BOOTSTRAP=1 when we raise the minimum GNU Make version to 4.4
I also like the commit description to record that RUSTC_BOOTSTRAP=1 is needed for GNU Make prier to commit 98da874c4303 ("[SV 10593] Export variables to $(shell ...) commands"), i.e., GNU Make 4.3 or older.
--
Best Regards
Masahiro Yamada
Quoted 64 lines · Alice Ryhl
Fix a few different compiler errors that cause rustc-option to give wrong results.
If
KBUILD_RUSTFLAGSor the flags being tested contain any -Z flags, then the error below is generated. TheRUSTC_BOOTSTRAPenvironment variable is added to fix this error.error: the option
Zis only accepted on the nightly compiler
help: consider switching to a nightly toolchain:rustup default nightly
note: selecting a toolchain with+toolchainarguments require a rustup proxy;
see <https://rust-lang.github.io/rustup/concepts/index.html>
note: for more information about Rust's stability policy, see
<https://doc.rust-lang.org/book/appendix-07-nightly-rust.html#unstable-features>
error: 1 nightly option were parsedThe probe may also fail incorrectly with the below error message. To fix it, the
/dev/nullargument is replaced with a newrust/probe.rsfile that doesn't need even the core part of the standard library.error[E0463]: can't find crate for
std
|
= note: theaarch64-unknown-nonetarget may not be installed
= help: consider downloading the target withrustup target add aarch64-unknown-none= help: consider building the standard library from source withcargo build -Zbuild-stdThe -o and --out-dir parameters are altered to fix this warning:
warning: ignoring --out-dir flag due to -o flag
I verified that the
Kconfigversion of rustc-option doesn't have the same issues.Fixes c42297438aee ("kbuild: rust: Define probing macros for rustc")Signed-off-by Alice Ryhl <[email protected]>---
Changes in v2:
- AddexporttoRUSTC_BOOTSTRAP.
- Fix error about core being missing.
- Fix warning about -o flag.
- Link to v1: https://lore.kernel.org/r/[email protected]--- rust/probe.rs | 7 +++++++ scripts/Makefile.compiler | 5 +++-- 2 files changed, 10 insertions(+), 2 deletions(-)rust/probe.rs new file +7
-0@@ -0,0 +1,7 @@ +//! Nearly empty file passed to rustc-option by Make. +//! +//! The no_core attribute is needed because rustc-option otherwise fails due to +//! not being able to find the core part of the standard library. + +#![feature(no_core)] +#![no_core] scripts/Makefile.compiler +0
-0
scripts/Makefile.compiler TMPOUT = $(if $(KBUILD_EXTMOD),$(firstword $(KBUILD_EXTMOD))/).tmp_$$$$ · line 24
Quoted 4 lines
scripts/Makefile.compiler +1
-0@@ -21,6 +21,7 @@ TMPOUT = $(if $(KBUILD_EXTMOD),$(firstword $(KBUILD_EXTMOD))/).tmp_$$$$ # automatically cleaned up. try-run = $(shell set -e; \ TMP=$(TMPOUT)/tmp; \ + export RUSTC_BOOTSTRAP=1; \
try-run is not Rust-specific.
Is there any reason why you did not add it
to __rustc-option?
__rustc-option = $(call try-run,\
RUSTC_BOOTSTRAP=1 $(1) $(2) $(3) --crate-type=rlib$(srctree)/rust/probe.rs --out-dir=$$TMP,$(3),$(4))
I guess it is still suspicious because the top-level Makefile exports RUCTC_BOOTSTRAP.
Quoted 28 lines
scripts/Makefile.compiler +2
-3trap "rm -rf $(TMPOUT)" EXIT; \ mkdir -p $(TMPOUT); \ if ($(1)) >/dev/null 2>&1; \@@ -76,7 +77,7 @@ ld-option = $(call try-run, $(LD) $(KBUILD_LDFLAGS) $(1) -v,$(1),$(2),$(3)) # __rustc-option # Usage: MY_RUSTFLAGS += $(call __rustc-option,$(RUSTC),$(MY_RUSTFLAGS),-Cinstrument-coverage,-Zinstrument-coverage) __rustc-option = $(call try-run,\ - $(1) $(2) $(3) --crate-type=rlib /dev/null --out-dir=$$TMPOUT -o "$$TMP",$(3),$(4)) + $(1) $(2) $(3) --crate-type=rlib $(srctree)/rust/probe.rs --out-dir=$$TMP,$(3),$(4)) # rustc-option # Usage: rustflags-y += $(call rustc-option,-Cinstrument-coverage,-Zinstrument-coverage)@@ -86,4 +87,4 @@ rustc-option = $(call __rustc-option, $(RUSTC),\ # rustc-option-yn # Usage: flag := $(call rustc-option-yn,-Cinstrument-coverage) rustc-option-yn = $(call try-run,\ - $(RUSTC) $(KBUILD_RUSTFLAGS) $(1) --crate-type=rlib /dev/null --out-dir=$$TMPOUT -o "$$TMP",y,n) + $(RUSTC) $(KBUILD_RUSTFLAGS) $(1) --crate-type=rlib $(srctree)/rust/probe.rs --out-dir=$$TMP,y,n) ---base-commit:
8cf0b93919e13d1e8d4466eb4080a4c4d9d66d7b
change-id: 20241008-rustc-option-bootstrap-607e5bf3114cBest regards,
--
Alice Ryhl <[email protected]>
Signature · 2 lines
Best Regards Masahiro Yamada
Alice Ryhl
reviewerQuoted 15 lines · Masahiro Yamada
scripts/Makefile.compiler TMPOUT = $(if $(KBUILD_EXTMOD),$(firstword $(KBUILD_EXTMOD))/).tmp_$$$$ · line 24
Quoted 9 lines · Alice Ryhl
scripts/Makefile.compiler +1
-0@@ -21,6 +21,7 @@ TMPOUT = $(if $(KBUILD_EXTMOD),$(firstword $(KBUILD_EXTMOD))/).tmp_$$$$ # automatically cleaned up. try-run = $(shell set -e; \ TMP=$(TMPOUT)/tmp; \ + export RUSTC_BOOTSTRAP=1; \try-run is not Rust-specific.
Quoted 7 lines
Is there any reason why you did not add it
to __rustc-option?__rustc-option = $(call try-run,\ RUSTC_BOOTSTRAP=1 $(1) $(2) $(3) --crate-type=rlib$(srctree)/rust/probe.rs --out-dir=$$TMP,$(3),$(4))
I had an explanation for this in the commit message, but it looks like it got lost when I rewrote it for v2. Anyway, the reason is that I'd have to modify both __rustc-option and rustc-option-yn to do that, and putting it here seemed more future-proof against making the same mistake in any rustc-* commands added in the future.
But I realize that it's not clear-cut. I'm happy to move it if you prefer, or perhaps add a try-run-rust. Let me know what you think.
Quoted 2 lines
I guess it is still suspicious because the top-level
MakefileexportsRUCTC_BOOTSTRAP.
Moving the declaration of RUSTC_BOOTSTRAP to the top of the Makefile seems to fix it. I guess moving it is probably a better solution than adding it in scripts/Makefile.compiler.
Not that I really understand why that is. The existing invocations are in scripts/Makefile.kasan which is invoked after RUSTC_BOOTSTRAP is declared.
Alice
Quoted 25 lines · Alice Ryhl
Quoted 15 lines · Masahiro Yamada
scripts/Makefile.compiler TMPOUT = $(if $(KBUILD_EXTMOD),$(firstword $(KBUILD_EXTMOD))/).tmp_$$$$ · line 24
Quoted 9 lines · Alice Ryhl
scripts/Makefile.compiler +1
-0@@ -21,6 +21,7 @@ TMPOUT = $(if $(KBUILD_EXTMOD),$(firstword $(KBUILD_EXTMOD))/).tmp_$$$$ # automatically cleaned up. try-run = $(shell set -e; \ TMP=$(TMPOUT)/tmp; \ + export RUSTC_BOOTSTRAP=1; \try-run is not Rust-specific.
Quoted 7 lines
Is there any reason why you did not add it
to __rustc-option?__rustc-option = $(call try-run,\ RUSTC_BOOTSTRAP=1 $(1) $(2) $(3) --crate-type=rlib$(srctree)/rust/probe.rs --out-dir=$$TMP,$(3),$(4))
Quoted 5 lines
I had an explanation for this in the commit message, but it looks like it got lost when I rewrote it for v2. Anyway, the reason is that I'd have to modify both __rustc-option and rustc-option-yn to do that, and putting it here seemed more future-proof against making the same mistake in any rustc-* commands added in the future.
One solution is to delete rustc-option-yn since there are no users of it.
Another solution is to refactor the code.
Either way, there is no good reason for code duplication.
If you keep rustc-option-yn, you can rebased v3 on top of this patch:
https://lore.kernel.org/lkml/[email protected]/T/#u
Quoted 10 lines
But I realize that it's not clear-cut. I'm happy to move it if you prefer, or perhaps add a try-run-rust. Let me know what you think.
Quoted 2 lines
I guess it is still suspicious because the top-level
MakefileexportsRUCTC_BOOTSTRAP.Moving the declaration of
RUSTC_BOOTSTRAPto the top of theMakefileseems to fix it. I guess moving it is probably a better solution than adding it inscripts/Makefile.compiler.
I prefer to keep RUSTC_BOOTSTRAP close to other compiler flags.
Quoted 4 lines
Not that I really understand why that is. The existing invocations are in
scripts/Makefile.kasanwhich is invoked afterRUSTC_BOOTSTRAPis declared.
Miguel gave perfect explanation.
https://lore.kernel.org/linux-kbuild/CAK7LNARDkS6uAHcdyZatc2SB7A66TWGfKZWNkYOoa7i3jo3QqA@mail.gmail.com/T/#m899bc321ae80d9c4a904680709c9a53f09e51b9e
Quoted 4 lines
Alice
--
Best Regards
Masahiro Yamada
Alice Ryhl
reviewerQuoted 36 lines · Masahiro Yamada
Quoted 25 lines · Alice Ryhl
Quoted 15 lines · Masahiro Yamada
scripts/Makefile.compiler TMPOUT = $(if $(KBUILD_EXTMOD),$(firstword $(KBUILD_EXTMOD))/).tmp_$$$$ · line 24
Quoted 9 lines · Alice Ryhl
scripts/Makefile.compiler +1
-0@@ -21,6 +21,7 @@ TMPOUT = $(if $(KBUILD_EXTMOD),$(firstword $(KBUILD_EXTMOD))/).tmp_$$$$ # automatically cleaned up. try-run = $(shell set -e; \ TMP=$(TMPOUT)/tmp; \ + export RUSTC_BOOTSTRAP=1; \try-run is not Rust-specific.
Quoted 7 lines
Is there any reason why you did not add it
to __rustc-option?__rustc-option = $(call try-run,\ RUSTC_BOOTSTRAP=1 $(1) $(2) $(3) --crate-type=rlib$(srctree)/rust/probe.rs --out-dir=$$TMP,$(3),$(4))
Quoted 5 lines
I had an explanation for this in the commit message, but it looks like it got lost when I rewrote it for v2. Anyway, the reason is that I'd have to modify both __rustc-option and rustc-option-yn to do that, and putting it here seemed more future-proof against making the same mistake in any rustc-* commands added in the future.
One solution is to delete rustc-option-yn since there are no users of it.
Quoted 7 lines
Another solution is to refactor the code.
Either way, there is no good reason for code duplication.
If you keep rustc-option-yn, you can rebased v3 on top of this patch:
https://lore.kernel.org/lkml/[email protected]/T/#u
I'll rebase on top of that. If we choose to delete rustc-option-yn then I think we should first merge the refactor and then delete it in a follow-up. That way, when someone does need it, they will find the refactored implementation in the git history.
Alice
Quoted 7 lines · Alice Ryhl
rust/probe.rs new file +0
-0
rust/probe.rs line 7
Quoted 7 lines
rust/probe.rs new file +7
-0@@ -0,0 +1,7 @@ +//! Nearly empty file passed to rustc-option by Make. +//! +//! The no_core attribute is needed because rustc-option otherwise fails due to +//! not being able to find the core part of the standard library. + +#![feature(no_core)] +#![no_core]
What I did in the GitHub issue was:
echo '#![feature(no_core)]#![no_core]' | rustc ... -
to avoid a file just for this and so that things were closer (i.e. the comment would be then in the Makefile).
Not sure what Masahiro prefers.
Cheers,
Miguel
Alice Ryhl
reviewerQuoted 33 lines · Alice Ryhl
Fix a few different compiler errors that cause rustc-option to give wrong results.
If
KBUILD_RUSTFLAGSor the flags being tested contain any -Z flags, then the error below is generated. TheRUSTC_BOOTSTRAPenvironment variable is added to fix this error.error: the option
Zis only accepted on the nightly compiler
help: consider switching to a nightly toolchain:rustup default nightly
note: selecting a toolchain with+toolchainarguments require a rustup proxy;
see <https://rust-lang.github.io/rustup/concepts/index.html>
note: for more information about Rust's stability policy, see
<https://doc.rust-lang.org/book/appendix-07-nightly-rust.html#unstable-features>
error: 1 nightly option were parsedThe probe may also fail incorrectly with the below error message. To fix it, the
/dev/nullargument is replaced with a newrust/probe.rsfile that doesn't need even the core part of the standard library.error[E0463]: can't find crate for
std
|
= note: theaarch64-unknown-nonetarget may not be installed
= help: consider downloading the target withrustup target add aarch64-unknown-none= help: consider building the standard library from source withcargo build -Zbuild-stdThe -o and --out-dir parameters are altered to fix this warning:
warning: ignoring --out-dir flag due to -o flag
I verified that the
Kconfigversion of rustc-option doesn't have the same issues.
Quoted 8 lines
Fixes c42297438aee ("kbuild: rust: Define probing macros for rustc")Signed-off-by Alice Ryhl <[email protected]>---
Changes in v2:
- AddexporttoRUSTC_BOOTSTRAP.
- Fix error about core being missing.
- Fix warning about -o flag.
- Link to v1: https://lore.kernel.org/r/[email protected]
v3 sent here:
https://lore.kernel.org/all/[email protected]/
Alice