diff mbox series

cargo: Increase RUST_MIN_STACK size to avoid SIGSEGV in DEBUG builds only.

Message ID 20241030111601.284064-1-Yash.Shinde@windriver.com
State New
Headers show
Series cargo: Increase RUST_MIN_STACK size to avoid SIGSEGV in DEBUG builds only. | expand

Commit Message

Yash Shinde Oct. 30, 2024, 11:16 a.m. UTC 
From: Yash Shinde <Yash.Shinde@windriver.com>

For some Rust crates, like building p384 in DEBUG mode, a SIGSEGV
is encountered as follows:

     note: rustc unexpectedly overflowed its stack! this is a bug
     note: maximum backtrace depth reached, frames may have been lost
     note: we would appreciate a report at https://github.com/rust-lang/rust
     note: backtrace dumped due to SIGSEGV! resuming signal
     error: could not compile `p384` (lib)

WARNING: Backtrace (BB generated script):
        #1: oe_cargo_build, /home/poky/build/tmp/work/core2-32-wrs-linux/cargo/1.79.0/temp/run.do_compile.3262358, line 193
        #2: cargo_do_compile, /home/poky/build/tmp/work/core2-32-wrs-linux/cargo/1.79.0/temp/run.do_compile.3262358, line 165
        #3: do_compile, /home/poky/build/tmp/work/core2-32-wrs-linux/cargo/1.79.0/temp/run.do_compile.3262358, line 160
        #4: main, /home/poky/build/tmp/work/core2-32-wrs-linux/cargo/1.79.0/temp/run.do_compile.3262358, line 206

Setting RUST_MIN_STACK to 8MB conditionally for DEBUG builds only handles
the stack overflow and SIGSEGV.

References: https://github.com/esp-rs/rust/issues/214
            https://github.com/rust-lang/rust/issues/122357
            https://github.com/rust-lang/rust/pull/122847#issue-2201254359
            https://github.com/rust-lang/rust/pull/122847/commits

Signed-off-by: Yash Shinde <Yash.Shinde@windriver.com>
---
 meta/recipes-devtools/rust/cargo_1.79.0.bb | 3 +++
 1 file changed, 3 insertions(+)

Comments

Yash Shinde Oct. 30, 2024, 11:20 a.m. UTC | #1 
Adding the log file for more details.
Randy MacLeod Oct. 30, 2024, 2:24 p.m. UTC | #2 
On 2024-10-30 7:16 a.m., Yash.Shinde@windriver.com wrote:
> From: Yash Shinde<Yash.Shinde@windriver.com>
>
> For some Rust crates, like building p384 in DEBUG mode, a SIGSEGV
> is encountered as follows:
>
>       note: rustc unexpectedly overflowed its stack! this is a bug
>       note: maximum backtrace depth reached, frames may have been lost
>       note: we would appreciate a report athttps://github.com/rust-lang/rust
>       note: backtrace dumped due to SIGSEGV! resuming signal
>       error: could not compile `p384` (lib)
>
> WARNING: Backtrace (BB generated script):
>          #1: oe_cargo_build, /home/poky/build/tmp/work/core2-32-wrs-linux/cargo/1.79.0/temp/run.do_compile.3262358, line 193
>          #2: cargo_do_compile, /home/poky/build/tmp/work/core2-32-wrs-linux/cargo/1.79.0/temp/run.do_compile.3262358, line 165
>          #3: do_compile, /home/poky/build/tmp/work/core2-32-wrs-linux/cargo/1.79.0/temp/run.do_compile.3262358, line 160
>          #4: main, /home/poky/build/tmp/work/core2-32-wrs-linux/cargo/1.79.0/temp/run.do_compile.3262358, line 206
>
> Setting RUST_MIN_STACK to 8MB conditionally for DEBUG builds only handles
> the stack overflow and SIGSEGV.

s/to 8MB/from 2MB to 8MB/

Best to explain what the current default value in your commit log in 
case the rust devs increase the limit some day.

From:
rust.git on 
Yash Shinde Oct. 31, 2024, 9:35 a.m. UTC | #3 
On 30-10-2024 19:54, Randy MacLeod wrote:
> On 2024-10-30 7:16 a.m., Yash.Shinde@windriver.com wrote:
>> From: Yash Shinde<Yash.Shinde@windriver.com>
>>
>> For some Rust crates, like building p384 in DEBUG mode, a SIGSEGV
>> is encountered as follows:
>>
>>       note: rustc unexpectedly overflowed its stack! this is a bug
>>       note: maximum backtrace depth reached, frames may have been lost
>>       note: we would appreciate a report athttps://github.com/rust-lang/rust
>>       note: backtrace dumped due to SIGSEGV! resuming signal
>>       error: could not compile `p384` (lib)
>>
>> WARNING: Backtrace (BB generated script):
>>          #1: oe_cargo_build, /home/poky/build/tmp/work/core2-32-wrs-linux/cargo/1.79.0/temp/run.do_compile.3262358, line 193
>>          #2: cargo_do_compile, /home/poky/build/tmp/work/core2-32-wrs-linux/cargo/1.79.0/temp/run.do_compile.3262358, line 165
>>          #3: do_compile, /home/poky/build/tmp/work/core2-32-wrs-linux/cargo/1.79.0/temp/run.do_compile.3262358, line 160
>>          #4: main, /home/poky/build/tmp/work/core2-32-wrs-linux/cargo/1.79.0/temp/run.do_compile.3262358, line 206
>>
>> Setting RUST_MIN_STACK to 8MB conditionally for DEBUG builds only handles
>> the stack overflow and SIGSEGV.
>
> s/to 8MB/from 2MB to 8MB/
>
> Best to explain what the current default value in your commit log in 
> case the rust devs increase the limit some day.
>
> From:
> rust.git on 
Randy MacLeod Oct. 31, 2024, 2:02 p.m. UTC | #4 
On 2024-10-31 5:35 a.m., Yash Shinde wrote:
>
> On 30-10-2024 19:54, Randy MacLeod wrote:
>> On 2024-10-30 7:16 a.m., Yash.Shinde@windriver.com wrote:
>>> From: Yash Shinde<Yash.Shinde@windriver.com>
>>>
>>> For some Rust crates, like building p384 in DEBUG mode, a SIGSEGV
>>> is encountered as follows:
>>>
>>>       note: rustc unexpectedly overflowed its stack! this is a bug
>>>       note: maximum backtrace depth reached, frames may have been lost
>>>       note: we would appreciate a report 
>>> athttps://github.com/rust-lang/rust
>>>       note: backtrace dumped due to SIGSEGV! resuming signal
>>>       error: could not compile `p384` (lib)
>>>
>>> WARNING: Backtrace (BB generated script):
>>>          #1: oe_cargo_build, 
>>> /home/poky/build/tmp/work/core2-32-wrs-linux/cargo/1.79.0/temp/run.do_compile.3262358, 
>>> line 193
>>>          #2: cargo_do_compile, 
>>> /home/poky/build/tmp/work/core2-32-wrs-linux/cargo/1.79.0/temp/run.do_compile.3262358, 
>>> line 165
>>>          #3: do_compile, 
>>> /home/poky/build/tmp/work/core2-32-wrs-linux/cargo/1.79.0/temp/run.do_compile.3262358, 
>>> line 160
>>>          #4: main, 
>>> /home/poky/build/tmp/work/core2-32-wrs-linux/cargo/1.79.0/temp/run.do_compile.3262358, 
>>> line 206
>>>
>>> Setting RUST_MIN_STACK to 8MB conditionally for DEBUG builds only 
>>> handles
>>> the stack overflow and SIGSEGV.
>>
>> s/to 8MB/from 2MB to 8MB/
>>
>> Best to explain what the current default value in your commit log in 
>> case the rust devs increase the limit some day.
>>
>> From:
>> rust.git on 
Yash Shinde Nov. 14, 2024, 11:02 a.m. UTC | #5 
On 31-10-2024 19:32, Randy MacLeod wrote:
>
> We discussed this issue and agreed that since it doesn't affect YP AB 
> since there are no DEBUG builds done there
> and more importantly since it's really a work-around rather than a 
> fix, we'll continue to work with upstream p384
> crate owners to resolve the issue. Forcing users to change the stack 
> size, even for debug builds, is not a good policy
> and hopefully some of the code-generation tools that are inputs to 
> p384 can be adjusted to reduce stack usage.
>
> ../Randy
>
This issue is now fixed by removing unused mut variables in p384 crate 
and thus reducing the codegen.
(https://github.com/RustCrypto/elliptic-curves/commit/7652c5842cde60acc1e739ebbba34614cb158f8b)

The current stable version of the RustCrypto  p384 crate is v0.13.0, 
which is also the default version for building with Rust.
But, the fix has been merged into the pre-release versions 
"p384/v0.14.0-pre.2" and "p384/v0.14.0-pre.1." of p384 crate.

I have opened a discussion thread in p384 crate: 
https://github.com/RustCrypto/elliptic-curves/discussions/1102
to ask the crate developers either to update p384 crate into new version 
with the fix, or backport it to "v.0.13.0".

We can wait for upstream unless someone wants to merge a patch for 
master now.
Let me know your thoughts on this.

Regards,
Yash
Richard Purdie Nov. 14, 2024, 12:01 p.m. UTC | #6 
On Thu, 2024-11-14 at 16:32 +0530, Yash Shinde via lists.openembedded.org wrote:
> 
> On 31-10-2024 19:32, Randy MacLeod wrote:
> > 
> > We discussed this issue and agreed that since it doesn't affect YP AB 
> > since there are no DEBUG builds done there
> > and more importantly since it's really a work-around rather than a 
> > fix, we'll continue to work with upstream p384
> > crate owners to resolve the issue. Forcing users to change the stack 
> > size, even for debug builds, is not a good policy
> > and hopefully some of the code-generation tools that are inputs to 
> > p384 can be adjusted to reduce stack usage.
> > 
> > ../Randy
> > 
> This issue is now fixed by removing unused mut variables in p384 crate 
> and thus reducing the codegen.
> (https://github.com/RustCrypto/elliptic-curves/commit/7652c5842cde60acc1e739ebbba34614cb158f8b)
> 
> The current stable version of the RustCrypto  p384 crate is v0.13.0, 
> which is also the default version for building with Rust.
> But, the fix has been merged into the pre-release versions 
> "p384/v0.14.0-pre.2" and "p384/v0.14.0-pre.1." of p384 crate.
> 
> I have opened a discussion thread in p384 crate: 
> https://github.com/RustCrypto/elliptic-curves/discussions/1102
> to ask the crate developers either to update p384 crate into new version 
> with the fix, or backport it to "v.0.13.0".
> 
> We can wait for upstream unless someone wants to merge a patch for 
> master now. Let me know your thoughts on this.

Firstly, big thanks for working through this. Getting issues like this
fixed at source in the upstream is infinitely preferable to carrying
weird workarounds.

I think it should be fine to wait for the upstream change to filter
through to us?

Cheers,

Richard
diff mbox series

Patch

diff --git a/meta/recipes-devtools/rust/cargo_1.79.0.bb b/meta/recipes-devtools/rust/cargo_1.79.0.bb
index 123032cdf7..4917b25c0c 100644
--- a/meta/recipes-devtools/rust/cargo_1.79.0.bb
+++ b/meta/recipes-devtools/rust/cargo_1.79.0.bb
@@ -37,6 +37,9 @@  do_cargo_setup_snapshot[vardepsexclude] += "UNINATIVE_LOADER"
 
 do_compile:prepend () {
 	export RUSTC_BOOTSTRAP="1"
+	if [ "${DEBUG_BUILD}" = "1" ]; then
+		export RUST_MIN_STACK=8388608
+	fi
 }
 
 do_install () {