Skip to contentSkip to conversation

rust: Suppress searching builtin sysroot

PATCH In review Matthew Maurer 4 participants 5 messages Oct 31, 2023 – Dec 13, 2023

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 rust: Suppress searching builtin sysroot

By default, if Rust is passed --target=foo rather than a target.json file, it will infer a default sysroot if that component is installed. As the proposed aarch64 support uses aarch64-unknown-none rather than a target.json file, this is needed to prevent rustc from being confused between the custom kernel sysroot and the pre-installed one.

---

This patch is prompted by the issue I encountered at
https://lore.kernel.org/all/CAGSQo01pOixiPXkW867h4vPUaAjtKtHGKhkV-rpifJvKxAf4Ww@mail.gmail.com/
but should be generically more hermetic even if we don't end up landing that patch.

 rust/Makefile          | 1 +
 scripts/Makefile.build | 1 +
 2 files changed, 2 insertions(+)
rust/Makefile +1 -0
index a27f35f924ec..0403e88e19fd 100644
@@ -400,6 +400,7 @@ quiet_cmd_rustc_library = $(if $(skip_clippy),RUSTC,$(RUSTC_OR_CLIPPY_QUIET)) L
 		--emit=metadata=$(dir $@)$(patsubst %.o,lib%.rmeta,$(notdir $@)) \
 		--crate-type rlib -L$(objtree)/$(obj) \
 		--crate-name $(patsubst %.o,%,$(notdir $@)) $< \
+		--sysroot=/dev/null \
 	$(if $(rustc_objcopy),;$(OBJCOPY) $(rustc_objcopy) $@)
 
 rust-analyzer:
scripts/Makefile.build +1 -0
index 82e3fb19fdaf..6e4ee513cc3c 100644
@@ -275,6 +275,7 @@ rust_common_cmd = \
 	--extern alloc --extern kernel \
 	--crate-type rlib -L $(objtree)/rust/ \
 	--crate-name $(basename $(notdir $@)) \
+	--sysroot=/dev/null \
 	--out-dir $(dir $@) --emit=dep-info=$(depfile)
 
 # `--emit=obj`, `--emit=asm` and `--emit=llvm-ir` imply a single codegen unit
Signature · 1 line
2.42.0.820.g83a721a137-goog
reviewer Reviewed-by on v1 1/1
#1 lore raw
Quoted 8 lines · Matthew Maurer

By default, if Rust is passed --target=foo rather than a target.json file, it will infer a default sysroot if that component is installed. As the proposed aarch64 support uses aarch64-unknown-none rather than a target.json file, this is needed to prevent rustc from being confused between the custom kernel sysroot and the pre-installed one.

I guess these are the only places where we need the sysroot parameter because the other rustc invocations compile for the host target instead?

Tested-by on v1 1/1
#2 lore raw
In reply to Matthew Maurer

[Cc Catalin as well]

Quoted 6 lines · Matthew Maurer

By default, if Rust is passed --target=foo rather than a target.json file, it will infer a default sysroot if that component is installed. As the proposed aarch64 support uses aarch64-unknown-none rather than a target.json file, this is needed to prevent rustc from being confused between the custom kernel sysroot and the pre-installed one.

Quoted 8 lines

---

This patch is prompted by the issue I encountered at
https://lore.kernel.org/all/CAGSQo01pOixiPXkW867h4vPUaAjtKtHGKhkV-rpifJvKxAf4Ww@mail.gmail.com/
but should be generically more hermetic even if we don't end up landing that patch.

I'm able to reproduce the issue you mentioned here after rustup target add aarch64-unknown-none, and your patch fixed that.

Regards,
Boqun

Quoted 31 lines
 rust/Makefile          | 1 +
 scripts/Makefile.build | 1 +
 2 files changed, 2 insertions(+)
rust/Makefile +1 -0
index a27f35f924ec..0403e88e19fd 100644
@@ -400,6 +400,7 @@ quiet_cmd_rustc_library = $(if $(skip_clippy),RUSTC,$(RUSTC_OR_CLIPPY_QUIET)) L
 		--emit=metadata=$(dir $@)$(patsubst %.o,lib%.rmeta,$(notdir $@)) \
 		--crate-type rlib -L$(objtree)/$(obj) \
 		--crate-name $(patsubst %.o,%,$(notdir $@)) $< \
+		--sysroot=/dev/null \
 	$(if $(rustc_objcopy),;$(OBJCOPY) $(rustc_objcopy) $@)
 
 rust-analyzer:
scripts/Makefile.build +1 -0
index 82e3fb19fdaf..6e4ee513cc3c 100644
@@ -275,6 +275,7 @@ rust_common_cmd = \
 	--extern alloc --extern kernel \
 	--crate-type rlib -L $(objtree)/rust/ \
 	--crate-name $(basename $(notdir $@)) \
+	--sysroot=/dev/null \
 	--out-dir $(dir $@) --emit=dep-info=$(depfile)
 
 # `--emit=obj`, `--emit=asm` and `--emit=llvm-ir` imply a single codegen unit
Signature · 1 line
2.42.0.820.g83a721a137-goog
#3 lore raw
Quoted 18 lines · Boqun Feng

[Cc Catalin as well]

Quoted 6 lines · Matthew Maurer

By default, if Rust is passed --target=foo rather than a target.json file, it will infer a default sysroot if that component is installed. As the proposed aarch64 support uses aarch64-unknown-none rather than a target.json file, this is needed to prevent rustc from being confused between the custom kernel sysroot and the pre-installed one.

Quoted 8 lines

---

This patch is prompted by the issue I encountered at
https://lore.kernel.org/all/CAGSQo01pOixiPXkW867h4vPUaAjtKtHGKhkV-rpifJvKxAf4Ww@mail.gmail.com/
but should be generically more hermetic even if we don't end up landing that patch.

Quoted 5 lines

I'm able to reproduce the issue you mentioned here after rustup target add aarch64-unknown-none, and your patch fixed that.

Looks like I spoke too soon... we need another fix onto this:

rust/Makefile +1 -0
index eda32ff757b1..c573d8881686 100644
@@ -178,6 +178,7 @@ quiet_cmd_rustdoc_test_kernel = RUSTDOC TK $<
                --extern build_error --extern macros \
                --extern bindings --extern uapi \
                --no-run --crate-name kernel -Zunstable-options \
+               --sysroot=/dev/null \
                --test-builder $(objtree)/scripts/rustdoc_test_builder \
                $< $(rustdoc_test_kernel_quiet); \
        $(objtree)/scripts/rustdoc_test_gen

, this is for kernel doc test as kunit test.

Needless to say, I also have tested with that and confirm it fixes the issue.

Regards,
Boqun

Quoted 34 lines

Regards,
Boqun

Quoted 31 lines
 rust/Makefile          | 1 +
 scripts/Makefile.build | 1 +
 2 files changed, 2 insertions(+)
rust/Makefile +1 -0
index a27f35f924ec..0403e88e19fd 100644
@@ -400,6 +400,7 @@ quiet_cmd_rustc_library = $(if $(skip_clippy),RUSTC,$(RUSTC_OR_CLIPPY_QUIET)) L
 		--emit=metadata=$(dir $@)$(patsubst %.o,lib%.rmeta,$(notdir $@)) \
 		--crate-type rlib -L$(objtree)/$(obj) \
 		--crate-name $(patsubst %.o,%,$(notdir $@)) $< \
+		--sysroot=/dev/null \
 	$(if $(rustc_objcopy),;$(OBJCOPY) $(rustc_objcopy) $@)
 
 rust-analyzer:
scripts/Makefile.build +1 -0
index 82e3fb19fdaf..6e4ee513cc3c 100644
@@ -275,6 +275,7 @@ rust_common_cmd = \
 	--extern alloc --extern kernel \
 	--crate-type rlib -L $(objtree)/rust/ \
 	--crate-name $(basename $(notdir $@)) \
+	--sysroot=/dev/null \
 	--out-dir $(dir $@) --emit=dep-info=$(depfile)
 
 # `--emit=obj`, `--emit=asm` and `--emit=llvm-ir` imply a single codegen unit
Signature · 1 line
2.42.0.820.g83a721a137-goog
#4 lore raw
In reply to Matthew Maurer
Quoted 8 lines · Matthew Maurer

By default, if Rust is passed --target=foo rather than a target.json file, it will infer a default sysroot if that component is installed. As the proposed aarch64 support uses aarch64-unknown-none rather than a target.json file, this is needed to prevent rustc from being confused between the custom kernel sysroot and the pre-installed one.

Applied to rust-next. I added Boqun's fix for rusttest as well as a fix for rustdoc (which requires a conditional for host crates like macros); and added a couple Links to the arm64 discussions.

I would appreciate extra testing for this one (please use rust-next commits in that case).

Thanks everyone!

Cheers,
Miguel