From: David Laight <david.laight.linux@gmail.com>
To: Jan Engelhardt <ej@inai.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>,
Greg KH <gregkh@linuxfoundation.org>,
Boqun Feng <boqun.feng@gmail.com>,
Miguel Ojeda <miguel.ojeda.sandonis@gmail.com>,
Christoph Hellwig <hch@infradead.org>,
rust-for-linux <rust-for-linux@vger.kernel.org>,
Linus Torvalds <torvalds@linux-foundation.org>,
David Airlie <airlied@gmail.com>,
linux-kernel@vger.kernel.org, ksummit@lists.linux.dev
Subject: Re: C aggregate passing (Rust kernel policy)
Date: Fri, 21 Feb 2025 18:34:37 +0000 [thread overview]
Message-ID: <20250221183437.1e2b5b94@pumpkin> (raw)
In-Reply-To: <n05p910s-r5o3-0n36-5s44-qr769prp69r5@vanv.qr>
On Thu, 20 Feb 2025 16:17:07 +0100 (CET)
Jan Engelhardt <ej@inai.de> wrote:
> On Thursday 2025-02-20 14:23, H. Peter Anvin wrote:
> >
> >People writing C seem to have a real aversion for using structures
> >as values (arguments, return values or assignments) even though that
> >has been valid since at least C90 and can genuinely produce better
> >code in some cases.
>
> The aversion stems from compilers producing "worse" ASM to this
> date, as in this case for example:
>
> ```c
> #include <sys/stat.h>
> extern struct stat fff();
> struct stat __attribute__((noinline)) fff()
> {
> struct stat sb = {};
> stat(".", &sb);
> return sb;
> }
> ```
>
> Build as C++ and C and compare.
>
> $ g++-15 -std=c++23 -O2 -x c++ -c x.c && objdump -Mintel -d x.o
> $ gcc-15 -std=c23 -O2 -c x.c && objdump -Mintel -d x.o
>
> Returning aggregates in C++ is often implemented with a secret extra
> pointer argument passed to the function. The C backend does not
> perform that kind of transformation automatically. I surmise ABI reasons.
Have you really looked at the generated code?
For anything non-trivial if gets truly horrid.
To pass a class by value the compiler has to call the C++ copy-operator to
generate a deep copy prior to the call, and then call the destructor after
the function returns - compare against passing a pointer to an existing
item (and not letting it be written to).
Returning a class member is probably worse and leads to nasty bugs.
In general the called code will have to do a deep copy from the item
being returned and then (quite likely) call the destructor for the
local variable being returned (if a function always returns a specific
local then the caller-provided temporary might be usable).
The calling code now has a temporary local variable that is going
to go out of scope (and be destructed) very shortly - I think the
next sequence point.
So you have lots of constructors, copy-operators and destructors
being called.
Then you get code like:
const char *foo = data.func().c_str();
very easily written looks fine, but foo points to garbage.
I've been going through some c++ code pretty much removing all the
places that classes get returned by value.
You can return a reference - that doesn't go out of scope.
Or, since most of the culprits are short std::string, replace them by char[].
Code is better, shorter, and actually less buggy.
(Apart from the fact that c++ makes it hard to ensure all the non-class
members are initialised.)
As Linus said, most modern ABI pass short structures in one or two registers
(or stack slots).
But aggregate returns are always done by passing a hidden pointer argument.
It is annoying that double-sized integers (u64 on 32bit and u128 on 64bit)
are returned in a register pair - but similar sized structures have to be
returned by value.
It is possible to get around this with #defines that convert the value
to a big integer (etc) - but I don't remember that actually being done.
David
next prev parent reply other threads:[~2025-02-21 18:34 UTC|newest]
Thread overview: 362+ messages / expand[flat|nested] mbox.gz Atom feed top
2025-02-09 20:56 Rust kernel policy Miguel Ojeda
2025-02-18 16:08 ` Christoph Hellwig
2025-02-18 16:35 ` Jarkko Sakkinen
2025-02-18 16:39 ` Jarkko Sakkinen
2025-02-18 18:08 ` Jarkko Sakkinen
2025-02-18 21:22 ` Boqun Feng
2025-02-19 6:20 ` Jarkko Sakkinen
2025-02-19 6:35 ` Dave Airlie
2025-02-19 11:37 ` Jarkko Sakkinen
2025-02-19 13:25 ` Geert Uytterhoeven
2025-02-19 13:40 ` Jarkko Sakkinen
2025-02-19 7:05 ` Boqun Feng
2025-02-19 11:32 ` Jarkko Sakkinen
2025-02-18 17:36 ` Jiri Kosina
2025-02-20 6:33 ` Christoph Hellwig
2025-02-20 18:40 ` Alexei Starovoitov
2025-02-18 18:46 ` Miguel Ojeda
2025-02-18 21:49 ` H. Peter Anvin
2025-02-18 22:38 ` Dave Airlie
2025-02-18 22:54 ` Miguel Ojeda
2025-02-19 0:58 ` H. Peter Anvin
2025-02-19 3:04 ` Boqun Feng
2025-02-19 5:07 ` NeilBrown
2025-02-19 5:39 ` Greg KH
2025-02-19 15:05 ` Laurent Pinchart
2025-02-20 20:49 ` Lyude Paul
2025-02-21 19:24 ` Laurent Pinchart
2025-02-20 7:03 ` Martin Uecker
2025-02-20 7:10 ` Greg KH
2025-02-20 8:57 ` Martin Uecker
2025-02-20 13:46 ` Dan Carpenter
2025-02-20 14:09 ` Martin Uecker
2025-02-20 14:38 ` H. Peter Anvin
2025-02-20 15:25 ` Dan Carpenter
2025-02-20 15:49 ` Willy Tarreau
2025-02-22 15:30 ` Kent Overstreet
2025-02-20 14:53 ` Greg KH
2025-02-20 15:40 ` Martin Uecker
2025-02-21 0:46 ` Miguel Ojeda
2025-02-21 9:48 ` Dan Carpenter
2025-02-21 16:28 ` Martin Uecker
2025-02-21 17:43 ` Steven Rostedt
2025-02-21 18:07 ` Linus Torvalds
2025-02-21 18:19 ` Steven Rostedt
2025-02-21 18:31 ` Martin Uecker
2025-02-21 19:30 ` Linus Torvalds
2025-02-21 19:59 ` Martin Uecker
2025-02-21 20:11 ` Linus Torvalds
2025-02-22 7:20 ` Martin Uecker
2025-02-21 22:24 ` Steven Rostedt
2025-02-21 23:04 ` Linus Torvalds
2025-02-22 17:53 ` Kent Overstreet
2025-02-22 18:44 ` Linus Torvalds
2025-02-23 16:42 ` David Laight
2025-02-21 23:37 ` Martin Uecker
2025-02-25 22:49 ` David Laight
2025-02-22 18:42 ` Linus Torvalds
2025-02-22 9:45 ` Dan Carpenter
2025-02-22 10:25 ` Martin Uecker
2025-02-22 11:07 ` Greg KH
2025-02-21 18:23 ` Martin Uecker
2025-02-21 22:14 ` Steven Rostedt
2025-03-01 13:22 ` Askar Safin
2025-03-01 13:55 ` Martin Uecker
2025-03-02 6:50 ` Kees Cook
2025-02-21 18:11 ` Theodore Ts'o
2025-02-24 8:12 ` Dan Carpenter
2025-02-20 22:08 ` Paul E. McKenney
2025-02-22 23:42 ` Piotr Masłowski
2025-02-23 8:10 ` Martin Uecker
2025-02-23 23:31 ` comex
2025-02-24 9:08 ` Ventura Jack
2025-02-24 18:03 ` Martin Uecker
2025-02-20 12:28 ` Jan Engelhardt
2025-02-20 12:37 ` Greg KH
2025-02-20 13:23 ` H. Peter Anvin
2025-02-20 13:51 ` Willy Tarreau
2025-02-20 15:17 ` C aggregate passing (Rust kernel policy) Jan Engelhardt
2025-02-20 16:46 ` Linus Torvalds
2025-02-20 20:34 ` H. Peter Anvin
2025-02-21 8:31 ` HUANG Zhaobin
2025-02-21 18:34 ` David Laight [this message]
2025-02-21 19:12 ` Linus Torvalds
2025-02-21 20:07 ` comex
2025-02-21 21:45 ` David Laight
2025-02-22 6:32 ` Willy Tarreau
2025-02-22 6:37 ` Willy Tarreau
2025-02-22 8:41 ` David Laight
2025-02-22 9:11 ` Willy Tarreau
2025-02-21 20:06 ` Jan Engelhardt
2025-02-21 20:23 ` Laurent Pinchart
2025-02-21 20:24 ` Laurent Pinchart
2025-02-21 22:02 ` David Laight
2025-02-21 22:13 ` Bart Van Assche
2025-02-22 5:56 ` comex
2025-02-21 20:26 ` Linus Torvalds
2025-02-21 22:19 ` henrychurchill
2025-02-21 22:52 ` henrychurchill
2025-02-20 22:13 ` Rust kernel policy Paul E. McKenney
2025-02-21 5:19 ` Felipe Contreras
2025-02-21 5:36 ` Boqun Feng
2025-02-21 5:59 ` Felipe Contreras
2025-02-21 7:04 ` Dave Airlie
2025-02-24 20:27 ` Felipe Contreras
2025-02-24 20:37 ` Boqun Feng
2025-02-26 2:42 ` Felipe Contreras
2025-02-22 16:04 ` Kent Overstreet
2025-02-22 17:10 ` Ventura Jack
2025-02-22 17:34 ` Kent Overstreet
2025-02-23 2:08 ` Bart Van Assche
2025-02-19 5:53 ` Alexey Dobriyan
2025-02-19 5:59 ` Dave Airlie
2025-02-22 18:46 ` Kent Overstreet
2025-02-19 12:37 ` Miguel Ojeda
2025-02-20 11:26 ` Askar Safin
2025-02-20 12:33 ` vpotach
2025-02-19 18:52 ` Kees Cook
2025-02-19 19:08 ` Steven Rostedt
2025-02-19 19:17 ` Kees Cook
2025-02-19 20:27 ` Jason Gunthorpe
2025-02-19 20:46 ` Steven Rostedt
2025-02-19 20:52 ` Bart Van Assche
2025-02-19 21:07 ` Steven Rostedt
2025-02-20 16:05 ` Jason Gunthorpe
2025-02-20 8:13 ` Jarkko Sakkinen
2025-02-20 8:16 ` Jarkko Sakkinen
2025-02-20 11:57 ` Fiona Behrens
2025-02-20 14:07 ` Jarkko Sakkinen
2025-02-21 10:19 ` Jarkko Sakkinen
2025-02-22 12:10 ` Miguel Ojeda
2025-03-04 11:17 ` Fiona Behrens
2025-03-04 17:48 ` Jarkko Sakkinen
2025-02-20 9:55 ` Leon Romanovsky
2025-02-19 19:33 ` H. Peter Anvin
2025-02-20 6:32 ` Alexey Dobriyan
2025-02-20 6:53 ` Greg KH
2025-02-20 8:44 ` Alexey Dobriyan
2025-02-20 13:53 ` Willy Tarreau
2025-02-20 16:04 ` Jason Gunthorpe
2025-02-20 12:01 ` H. Peter Anvin
2025-02-20 12:13 ` H. Peter Anvin
2025-02-20 23:42 ` Miguel Ojeda
2025-02-22 15:21 ` Kent Overstreet
2025-02-20 6:42 ` Christoph Hellwig
2025-02-20 23:44 ` Miguel Ojeda
2025-02-21 15:24 ` Simona Vetter
2025-02-22 12:10 ` Miguel Ojeda
2025-02-26 13:17 ` Fiona Behrens
2025-02-21 0:39 ` Linus Torvalds
2025-02-21 12:16 ` Danilo Krummrich
2025-02-21 15:59 ` Steven Rostedt
2025-02-23 18:03 ` Laurent Pinchart
2025-02-23 18:31 ` Linus Torvalds
2025-02-26 16:05 ` Jason Gunthorpe
2025-02-26 19:32 ` Linus Torvalds
2025-02-19 8:05 ` Dan Carpenter
2025-02-19 14:14 ` James Bottomley
2025-02-19 14:30 ` Geert Uytterhoeven
2025-02-19 14:46 ` Martin K. Petersen
2025-02-19 14:51 ` Bartosz Golaszewski
2025-02-19 15:15 ` James Bottomley
2025-02-19 15:33 ` Willy Tarreau
2025-02-19 15:45 ` Laurent Pinchart
2025-02-19 15:46 ` James Bottomley
2025-02-19 15:56 ` Willy Tarreau
2025-02-19 16:07 ` Laurent Pinchart
2025-02-19 16:15 ` Willy Tarreau
2025-02-19 16:32 ` Laurent Pinchart
2025-02-19 16:34 ` Willy Tarreau
2025-02-19 16:33 ` Steven Rostedt
2025-02-19 16:47 ` Andrew Lunn
2025-02-19 18:22 ` Jarkko Sakkinen
2025-02-20 6:26 ` Alexey Dobriyan
2025-02-20 15:37 ` Steven Rostedt
2025-02-19 17:00 ` Martin K. Petersen
2025-02-19 15:13 ` Steven Rostedt
2025-02-19 14:05 ` James Bottomley
2025-02-19 15:08 ` Miguel Ojeda
2025-02-19 16:03 ` James Bottomley
2025-02-19 16:44 ` Miguel Ojeda
2025-02-19 17:06 ` Theodore Ts'o
2025-02-20 23:40 ` Miguel Ojeda
2025-02-22 15:03 ` Kent Overstreet
2025-02-20 16:03 ` James Bottomley
2025-02-20 23:47 ` Miguel Ojeda
2025-02-20 6:48 ` Christoph Hellwig
2025-02-20 12:56 ` James Bottomley
-- strict thread matches above, loose matches on Subject: below --
2025-02-22 10:06 C aggregate passing (Rust kernel policy) Ventura Jack
2025-02-22 14:15 ` Gary Guo
2025-02-22 15:03 ` Ventura Jack
2025-02-22 18:54 ` Kent Overstreet
2025-02-22 19:18 ` Linus Torvalds
2025-02-22 20:00 ` Kent Overstreet
2025-02-22 20:54 ` H. Peter Anvin
2025-02-22 21:22 ` Kent Overstreet
2025-02-22 21:46 ` Linus Torvalds
2025-02-22 22:34 ` Kent Overstreet
2025-02-22 23:56 ` Jan Engelhardt
2025-02-22 22:12 ` David Laight
2025-02-22 22:46 ` Kent Overstreet
2025-02-22 23:50 ` H. Peter Anvin
2025-02-23 0:06 ` Kent Overstreet
2025-02-22 21:22 ` Linus Torvalds
2025-02-23 15:30 ` Ventura Jack
2025-02-23 16:28 ` David Laight
2025-02-24 0:27 ` Gary Guo
2025-02-24 9:57 ` Ventura Jack
2025-02-24 10:31 ` Benno Lossin
2025-02-24 12:21 ` Ventura Jack
2025-02-24 12:47 ` Benno Lossin
2025-02-24 16:57 ` Ventura Jack
2025-02-24 22:03 ` Benno Lossin
2025-02-24 23:04 ` Ventura Jack
2025-02-25 22:38 ` Benno Lossin
2025-02-25 22:47 ` Miguel Ojeda
2025-02-25 23:03 ` Benno Lossin
2025-02-24 12:58 ` Theodore Ts'o
2025-02-24 14:47 ` Miguel Ojeda
2025-02-24 14:54 ` Miguel Ojeda
2025-02-24 16:42 ` Philip Herron
2025-02-25 15:55 ` Ventura Jack
2025-02-25 17:30 ` Arthur Cohen
2025-02-26 11:38 ` Ralf Jung
2025-02-24 15:43 ` Miguel Ojeda
2025-02-24 17:24 ` Kent Overstreet
2025-02-25 16:12 ` Alice Ryhl
2025-02-25 17:21 ` Ventura Jack
2025-02-25 17:36 ` Alice Ryhl
2025-02-25 18:16 ` H. Peter Anvin
2025-02-25 20:21 ` Kent Overstreet
2025-02-25 20:37 ` H. Peter Anvin
2025-02-26 13:03 ` Ventura Jack
2025-02-26 13:53 ` Miguel Ojeda
2025-02-26 14:07 ` Ralf Jung
2025-02-26 14:26 ` James Bottomley
2025-02-26 14:37 ` Ralf Jung
2025-02-26 14:39 ` Greg KH
2025-02-26 14:45 ` James Bottomley
2025-02-26 16:00 ` Steven Rostedt
2025-02-26 16:42 ` James Bottomley
2025-02-26 16:47 ` Kent Overstreet
2025-02-26 16:57 ` Steven Rostedt
2025-02-26 17:41 ` Kent Overstreet
2025-02-26 17:47 ` Steven Rostedt
2025-02-26 22:07 ` Josh Poimboeuf
2025-03-02 12:19 ` David Laight
2025-02-26 17:11 ` Miguel Ojeda
2025-02-26 17:42 ` Kent Overstreet
2025-02-26 12:36 ` Ventura Jack
2025-02-26 13:52 ` Miguel Ojeda
2025-02-26 15:21 ` Ventura Jack
2025-02-26 16:06 ` Ralf Jung
2025-02-26 17:49 ` Miguel Ojeda
2025-02-26 18:36 ` Ventura Jack
2025-02-26 14:14 ` Ralf Jung
2025-02-26 15:40 ` Ventura Jack
2025-02-26 16:10 ` Ralf Jung
2025-02-26 16:50 ` Ventura Jack
2025-02-26 21:39 ` Ralf Jung
2025-02-27 15:11 ` Ventura Jack
2025-02-27 15:32 ` Ralf Jung
2025-02-25 18:54 ` Linus Torvalds
2025-02-25 19:47 ` Kent Overstreet
2025-02-25 20:25 ` Linus Torvalds
2025-02-25 20:55 ` Kent Overstreet
2025-02-25 21:24 ` Linus Torvalds
2025-02-25 23:34 ` Kent Overstreet
2025-02-26 11:57 ` Gary Guo
2025-02-27 14:43 ` Ventura Jack
2025-02-26 14:26 ` Ventura Jack
2025-02-25 22:45 ` Miguel Ojeda
2025-02-26 0:05 ` Miguel Ojeda
2025-02-25 22:42 ` Miguel Ojeda
2025-02-26 14:01 ` Ralf Jung
2025-02-26 13:54 ` Ralf Jung
2025-02-26 17:59 ` Linus Torvalds
2025-02-26 19:01 ` Paul E. McKenney
2025-02-26 20:00 ` Martin Uecker
2025-02-26 21:14 ` Linus Torvalds
2025-02-26 21:21 ` Linus Torvalds
2025-02-26 22:54 ` David Laight
2025-02-27 0:35 ` Paul E. McKenney
2025-02-26 21:26 ` Steven Rostedt
2025-02-26 21:37 ` Steven Rostedt
2025-02-26 21:42 ` Linus Torvalds
2025-02-26 21:56 ` Steven Rostedt
2025-02-26 22:13 ` Steven Rostedt
2025-02-26 22:22 ` Linus Torvalds
2025-02-26 22:35 ` Steven Rostedt
2025-02-26 23:18 ` Linus Torvalds
2025-02-26 23:28 ` Steven Rostedt
2025-02-27 0:04 ` Linus Torvalds
2025-02-27 20:47 ` David Laight
2025-02-27 21:33 ` Steven Rostedt
2025-02-28 21:29 ` Paul E. McKenney
2025-02-27 21:41 ` Paul E. McKenney
2025-02-27 22:20 ` David Laight
2025-02-27 22:40 ` Paul E. McKenney
2025-02-28 7:44 ` Ralf Jung
2025-02-28 15:41 ` Kent Overstreet
2025-02-28 15:46 ` Boqun Feng
2025-02-28 16:04 ` Kent Overstreet
2025-02-28 16:13 ` Boqun Feng
2025-02-28 16:21 ` Kent Overstreet
2025-02-28 16:40 ` Boqun Feng
2025-03-04 18:12 ` Ralf Jung
2025-02-26 22:27 ` Kent Overstreet
2025-02-26 23:16 ` Linus Torvalds
2025-02-27 0:17 ` Kent Overstreet
2025-02-27 0:26 ` comex
2025-02-27 18:33 ` Ralf Jung
2025-02-27 19:15 ` Linus Torvalds
2025-02-27 19:55 ` Kent Overstreet
2025-02-27 20:28 ` Linus Torvalds
2025-02-28 7:53 ` Ralf Jung
2025-03-06 19:16 ` Ventura Jack
2025-02-27 4:18 ` Martin Uecker
2025-02-27 5:52 ` Linus Torvalds
2025-02-27 6:56 ` Martin Uecker
2025-02-27 14:29 ` Steven Rostedt
2025-02-27 17:35 ` Paul E. McKenney
2025-02-27 18:13 ` Kent Overstreet
2025-02-27 19:10 ` Paul E. McKenney
2025-02-27 18:00 ` Ventura Jack
2025-02-27 18:44 ` Ralf Jung
2025-02-27 14:21 ` Ventura Jack
2025-02-27 15:27 ` H. Peter Anvin
2025-02-28 8:08 ` Ralf Jung
2025-02-28 8:32 ` Martin Uecker
2025-02-26 20:25 ` Kent Overstreet
2025-02-26 20:34 ` Andy Lutomirski
2025-02-26 22:45 ` David Laight
2025-02-22 19:41 ` Miguel Ojeda
2025-02-22 20:49 ` Kent Overstreet
2025-02-26 11:34 ` Ralf Jung
2025-02-26 14:57 ` Ventura Jack
2025-02-26 16:32 ` Ralf Jung
2025-02-26 18:09 ` Ventura Jack
2025-02-26 22:28 ` Ralf Jung
2025-02-26 23:08 ` David Laight
2025-02-27 13:55 ` Ralf Jung
2025-02-27 17:33 ` Ventura Jack
2025-02-27 17:58 ` Ralf Jung
2025-02-27 19:06 ` Ventura Jack
2025-02-27 19:45 ` Ralf Jung
2025-02-27 20:22 ` Kent Overstreet
2025-02-27 22:18 ` David Laight
2025-02-27 23:18 ` Kent Overstreet
2025-02-28 7:38 ` Ralf Jung
2025-02-28 20:48 ` Ventura Jack
2025-02-28 20:41 ` Ventura Jack
2025-02-28 22:13 ` Geoffrey Thomas
2025-03-01 14:19 ` Ventura Jack
2025-03-04 18:24 ` Ralf Jung
2025-03-06 18:49 ` Ventura Jack
2025-02-27 17:58 ` Miguel Ojeda
2025-02-27 19:25 ` Ventura Jack
2025-02-26 19:07 ` Martin Uecker
2025-02-26 19:23 ` Ralf Jung
2025-02-26 20:22 ` Martin Uecker
[not found] <CAFJgqgRZ1w0ONj2wbcczx2=boXYHoLOd=-ke7tHGBAcifSfPUw@mail.gmail.com>
2025-02-25 15:42 ` H. Peter Anvin
2025-02-25 16:45 ` Ventura Jack
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20250221183437.1e2b5b94@pumpkin \
--to=david.laight.linux@gmail.com \
--cc=airlied@gmail.com \
--cc=boqun.feng@gmail.com \
--cc=ej@inai.de \
--cc=gregkh@linuxfoundation.org \
--cc=hch@infradead.org \
--cc=hpa@zytor.com \
--cc=ksummit@lists.linux.dev \
--cc=linux-kernel@vger.kernel.org \
--cc=miguel.ojeda.sandonis@gmail.com \
--cc=rust-for-linux@vger.kernel.org \
--cc=torvalds@linux-foundation.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.