unofficial mirror of libc-alpha@sourceware.org
 help / color / mirror / Atom feed
From: Adhemerval Zanella Netto via Libc-alpha <libc-alpha@sourceware.org>
To: Paul E Murphy <murphyp@linux.ibm.com>, libc-alpha@sourceware.org
Subject: Re: [PATCH v2 05/11] powerpc64: Add the clone3 wrapper
Date: Tue, 29 Nov 2022 17:43:57 -0300	[thread overview]
Message-ID: <21f073bc-8bb7-7346-fe6d-e1c0e872e8cb@linaro.org> (raw)
In-Reply-To: <b23b5596-f696-d69d-ab33-15ed0c3360e7@linux.ibm.com>



On 28/11/22 13:38, Paul E Murphy wrote:
> 
> 
> On 11/4/22 2:01 PM, Adhemerval Zanella via Libc-alpha wrote:
>> It follows the internal signature:
>>
>>    extern int clone3 (struct clone_args *__cl_args, size_t __size,
>>   int (*__func) (void *__arg), void *__arg);
>>
>> The powerpc64 ABI requires an initial stackframe so the child can
>> store/restore the TOC.  Iit is create prior calling clone3 by
> 
> trivial nit, s/Iit/It/.

Ack.

> 
>> adjusting the stack size (since kernel will compute the stack as
>> stack plus size).
>>
>> Checked on powerpc64-linux-gnu (power8, kernel 6.0) and
>> powerpc64le-linux-gnu (power9, kernel 4.18).
>> ---
>>   .../sysv/linux/powerpc/powerpc64/clone3.S     | 152 ++++++++++++++++++
>>   sysdeps/unix/sysv/linux/powerpc/sysdep.h      |   1 +
>>   2 files changed, 153 insertions(+)
>>   create mode 100644 sysdeps/unix/sysv/linux/powerpc/powerpc64/clone3.S
>>
>> diff --git a/sysdeps/unix/sysv/linux/powerpc/powerpc64/clone3.S b/sysdeps/unix/sysv/linux/powerpc/powerpc64/clone3.S
>> new file mode 100644
>> index 0000000000..0fe2fe91db
>> --- /dev/null
>> +++ b/sysdeps/unix/sysv/linux/powerpc/powerpc64/clone3.S
>> @@ -0,0 +1,152 @@
>> +/* The clone3 syscall wrapper.  Linux/powerpc64 version.
>> +   Copyright (C) 2022 Free Software Foundation, Inc.
>> +
>> +   This file is part of the GNU C Library.
>> +
>> +   The GNU C Library is free software; you can redistribute it and/or
>> +   modify it under the terms of the GNU Lesser General Public
>> +   License as published by the Free Software Foundation; either
>> +   version 2.1 of the License, or (at your option) any later version.
>> +
>> +   The GNU C Library is distributed in the hope that it will be useful,
>> +   but WITHOUT ANY WARRANTY; without even the implied warranty of
>> +   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
>> +   Lesser General Public License for more details.
>> +
>> +   You should have received a copy of the GNU Lesser General Public
>> +   License along with the GNU C Library; if not, see
>> +   <https://www.gnu.org/licenses/>.  */
>> +
>> +#include <sysdep.h>
>> +#define _ERRNO_H        1
>> +#include <bits/errno.h>
>> +
>> +/* The userland implementation is:
>> +   int clone3 (struct clone_args *cl_args, size_t size,
>> +               int (*func)(void *arg), void *arg);
>> +
>> +   the kernel entry is:
>> +   int clone3 (struct clone_args *cl_args, size_t size);
>> +
>> +   The parameters are passed in registers from userland:
>> +   r3: cl_args
>> +   r4: size
>> +   r5: func
>> +   r6: arg  */
>> +
>> +        .text
>> +ENTRY(__clone3)
>> +    CALL_MCOUNT 4
>> +
>> +    /* Sanity checks args.  */
>> +    cmpdi    cr0, r3, 0
>> +    cmpdi    cr1, r5, 0
>> +    cror    cr0*4+eq, cr1*4+eq, cr0*4+eq
>> +    beq    cr0,L(badargs)
>> +
>> +    /* Save some regs in the "red zone".  */
> Is there reason to avoid explicitly stacking a frame in here? Also, does the red zone exist on ELFv1?
> 

I don't think so, although it allows to tail call directly to
__syscall_error.  I used powerpc64 clone.S as example, which does
something similar.

>> +#ifdef USE_PPC_SCV
>> +    std    r28, -24(r1)
>> +    cfi_offset (r28, -24)
>> +#endif
>> +    std    r29, -16(r1)
>> +    std    r30, -8(r1)
>> +    cfi_offset (r29, -16)
>> +    cfi_offset (r30, -8)
>> +
>> +    /* Save fn and args across syscall.  */
> trivial nit, s/fn/func/ s/args/arg/.

Ack.

> 
>> +    mr    r30, r5        /* Function in r30.  */
>> +    mr    r29, r6        /* Argument in r29.  */
>> +
>> +    /* End FDE now, because in the child the unwind info will be
>> +       wrong.  */
>> +    cfi_endproc
>> +
>> +    /* Setup a minimum stack fram for child.  It needs to first calculate
> s/fram/frame/.

Ack.

> 
>> +       the effective stack address, write the start NULL frame-pointer,
>> +       and update the frame size in the input cl_args.  */
>> +    ld    r7, 40(r3)    /* Load stack value.  */
>> +    ld    r8, 48(r3)    /* Load stack_size value.  */
>> +    li    r10, 0
>> +    add    r7, r7, r8    /* Calculate effective stack address.  */
>> +    std    r10, -FRAME_MIN_SIZE_PARM(r7)
> This is clearing the backchain pointer? Maybe a small comment here.

Yes, since this stack will be the initial one.  I changed the comment to:

  /* Setup a minimum stack frame for child.  It needs to first calculate
     the effective stack address, write the start empty backchain pointer,
     and update the frame size in the input cl_args.  */

To make it more clear.

> 
>> +    addi    r8, r8, -FRAME_MIN_SIZE_PARM
>> +    std    r8, 48(r3)
>> +
>> +    /* Do the system call, the kernel expects:
>> +       r0: system call numer
>> +       r3: cl_args
>> +       r4: size  */
>> +    li    r0, SYS_ify(clone3)
>> +#ifdef USE_PPC_SCV
>> +    CHECK_SCV_SUPPORT r28 0f
>> +    /* This is equivalent to DO_CALL_SCV, but we cannot use the macro here
>> +       because it uses CFI directives and we just called cfi_endproc.  */
>> +    mflr     r9
>> +    std     r9, FRAME_LR_SAVE(r1)
>> +    .machine "push"
>> +    .machine "power9"
>> +    scv     0
>> +    .machine "pop"
>> +    ld     r9, FRAME_LR_SAVE(r1)
>> +    mtlr     r9
>> +
>> +    /* When using scv, error is indicated by negative r3.  */
> With scv an, an error is a value -4095 <= x < 0. This test should be more specific.

Ack, I changed the comment (btw clone.S has the same mistake).

> 
>> +    cmpdi    cr1, r3, 0
>> +    b    1f
>> +#endif
>> +0:      DO_CALL_SC
>> +
>> +    /* With sc, error is indicated by cr0.SO.  */
>> +    cmpdi    cr1, r3, 0
>> +    crandc    cr1*4+eq, cr1*4+eq, cr0*4+so > +
>> +1:    bne-    cr1,L(parent)
>> +
>> +    /* Child, load the function and arguments.  */
>> +    std    r2, FRAME_TOC_SAVE(r1)
>> +    PPC64_LOAD_FUNCPTR r30
>> +    mr    r3, r29
>> +    bctrl
>> +    ld    r2, FRAME_TOC_SAVE(r1)
>> +
>> +    li    r0, SYS_ify(exit)
>> +    DO_CALL_SC
> Minor nit, this should also use scv if supported.

Ack (btw clone.S also use issues DO_CALL_SC).

>> +    /* We won't ever get here but provide a nop so that the linker
>> +       will insert a toc adjusting stub if necessary.  */
>> +    nop
>> +
>> +L(badargs):
>> +    cfi_startproc
>> +    li    r3, EINVAL
>> +    TAIL_CALL_SYSCALL_ERROR
>> +
>> +L(parent):
>> +    /* Check if svc is available.  */
> s/svc/scv/.

Ack.

> 
>> +    cmpdi cr1, r28, 0
>> +
>> +    /* Parent.  Restore registers & return.  */
>> +#ifdef USE_PPC_SCV
>> +    cfi_offset (r28, -24)
>> +    ld    r28, -24(r1)
>> +    cfi_restore (r28)
>> +#endif
>> +    cfi_offset (r29,-16)
>> +    cfi_offset (r30,-8)
>> +    ld    r29, -16(r1)
>> +    ld    r30, -8(r1)
>> +    cfi_restore (r29)
>> +    cfi_restore (r30)
>> +
>> +#ifdef USE_PPC_SCV
>> +    beq    cr1, 0f
>> +    RET_SCV
>> +    b    1f
>> +#endif
>> +0:    RET_SC
>> +1:    TAIL_CALL_SYSCALL_ERROR
>> +
>> +PSEUDO_END (__clone3)
>> +
>> +libc_hidden_def (__clone3)
>> +weak_alias (__clone3, clone3)
>> diff --git a/sysdeps/unix/sysv/linux/powerpc/sysdep.h b/sysdeps/unix/sysv/linux/powerpc/sysdep.h
>> index 9e44818978..af63b3974b 100644
>> --- a/sysdeps/unix/sysv/linux/powerpc/sysdep.h
>> +++ b/sysdeps/unix/sysv/linux/powerpc/sysdep.h
>> @@ -214,6 +214,7 @@
>>   #if defined(__PPC64__) || defined(__powerpc64__)
>>   #define HAVE_CLOCK_GETRES64_VSYSCALL    "__kernel_clock_getres"
>>   #define HAVE_CLOCK_GETTIME64_VSYSCALL    "__kernel_clock_gettime"
>> +#define HAVE_CLONE3_WRAPPER        1
>>   #else
>>   #define HAVE_CLOCK_GETRES_VSYSCALL    "__kernel_clock_getres"
>>   #define HAVE_CLOCK_GETTIME_VSYSCALL    "__kernel_clock_gettime"

  reply	other threads:[~2022-11-29 20:44 UTC|newest]

Thread overview: 22+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2022-11-04 19:01 [PATCH v2 00/11] Optimize posix_spawn signal setup with clone3 Adhemerval Zanella via Libc-alpha
2022-11-04 19:01 ` [PATCH v2 01/11] linux: Do not reset signal handler in posix_spawn if it is already SIG_DFL Adhemerval Zanella via Libc-alpha
2023-01-11 21:27   ` Carlos O'Donell via Libc-alpha
2022-11-04 19:01 ` [PATCH v2 02/11] linux: Extend internal clone3 documentation Adhemerval Zanella via Libc-alpha
2022-11-07 16:40   ` Szabolcs Nagy via Libc-alpha
2023-01-11 21:35   ` Carlos O'Donell via Libc-alpha
2022-11-04 19:01 ` [PATCH v2 03/11] Linux: Do not align the stack for __clone3 Adhemerval Zanella via Libc-alpha
2023-01-11 21:43   ` Carlos O'Donell via Libc-alpha
2022-11-04 19:01 ` [PATCH v2 04/11] linux: Add clone3 CLONE_CLEAR_SIGHAND optimization to posix_spawn Adhemerval Zanella via Libc-alpha
2023-01-11 21:44   ` Carlos O'Donell via Libc-alpha
2022-11-04 19:01 ` [PATCH v2 05/11] powerpc64: Add the clone3 wrapper Adhemerval Zanella via Libc-alpha
2022-11-28 16:38   ` Paul E Murphy via Libc-alpha
2022-11-29 20:43     ` Adhemerval Zanella Netto via Libc-alpha [this message]
2022-11-04 19:01 ` [PATCH v2 06/11] aarch64: " Adhemerval Zanella via Libc-alpha
2022-11-07 16:14   ` Szabolcs Nagy via Libc-alpha
2022-11-04 19:01 ` [PATCH v2 07/11] s390x: " Adhemerval Zanella via Libc-alpha
2022-11-04 19:01 ` [PATCH v2 08/11] riscv: " Adhemerval Zanella via Libc-alpha
2022-11-04 19:01 ` [PATCH v2 09/11] arm: " Adhemerval Zanella via Libc-alpha
2022-11-04 19:01 ` [PATCH v2 10/11] mips: " Adhemerval Zanella via Libc-alpha
2022-11-04 19:01 ` [PATCH v2 11/11] Linux: optimize clone3 internal usage Adhemerval Zanella via Libc-alpha
2023-01-11 21:45   ` Carlos O'Donell via Libc-alpha
2023-01-11 21:24 ` [PATCH v2 00/11] Optimize posix_spawn signal setup with clone3 Carlos O'Donell via Libc-alpha

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

  List information: https://www.gnu.org/software/libc/involved.html

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=21f073bc-8bb7-7346-fe6d-e1c0e872e8cb@linaro.org \
    --to=libc-alpha@sourceware.org \
    --cc=adhemerval.zanella@linaro.org \
    --cc=murphyp@linux.ibm.com \
    /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 a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).