diff mbox series

support: arm: Add libc_do_syscall function to libsupport routines

Message ID 20210315154723.11687-1-lukma@denx.de
State Superseded
Headers show
Series support: arm: Add libc_do_syscall function to libsupport routines | expand

Commit Message

Lukasz Majewski March 15, 2021, 3:47 p.m. UTC
The ARM port requires the __libc_do_syscall function to be able to call
utimensat_time64 syscall required to check if file system supports 64
bit time.

This patch adds the sysdeps/unix/sysv/linux/arm/libc-do-syscall.S to
libsupport routines.

It fixes the following error on ARM 32 bit port:

y2038-glibc/support/support_path_support_time64.c:34: undefined reference to
`__libc_do_syscall'
collect2: error: ld returned 1 exit status
---
 sysdeps/unix/sysv/linux/arm/Makefile | 5 +++++
 1 file changed, 5 insertions(+)

Comments

Florian Weimer March 15, 2021, 3:55 p.m. UTC | #1
* Lukasz Majewski:

> It fixes the following error on ARM 32 bit port:
>
> y2038-glibc/support/support_path_support_time64.c:34: undefined reference to
> `__libc_do_syscall'
> collect2: error: ld returned 1 exit status

Should libsupport use syscall for direct system calls?

Thanks,
Florian
Lukasz Majewski March 15, 2021, 10:14 p.m. UTC | #2
Hi Florian,

> * Lukasz Majewski:
> 
> > It fixes the following error on ARM 32 bit port:
> >
> > y2038-glibc/support/support_path_support_time64.c:34: undefined
> > reference to `__libc_do_syscall'
> > collect2: error: ld returned 1 exit status  
> 
> Should libsupport use syscall for direct system calls?

I'm not sure what is the correct approach here.

In my case - armv7, 32 bit - in support/support_path_support_time64.c
the utimensat_time64 syscall is called, which is wrapped to
INLINE_SYSCALL_CALL(), which calls __libc_do_syscall ASM ARM function.

This function is only defined for i386 and 32 bit ARM (in e.g.
sysdeps/unix/sysv/linux/{arm|i386}/libc-do-syscall.S)

> 
> Thanks,
> Florian
> 




Best regards,

Lukasz Majewski

--

DENX Software Engineering GmbH,      Managing Director: Wolfgang Denk
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-59 Fax: (+49)-8142-66989-80 Email: lukma@denx.de
Florian Weimer March 16, 2021, 5:10 a.m. UTC | #3
* Lukasz Majewski:

> Hi Florian,
>
>> * Lukasz Majewski:
>> 
>> > It fixes the following error on ARM 32 bit port:
>> >
>> > y2038-glibc/support/support_path_support_time64.c:34: undefined
>> > reference to `__libc_do_syscall'
>> > collect2: error: ld returned 1 exit status  
>> 
>> Should libsupport use syscall for direct system calls?
>
> I'm not sure what is the correct approach here.
>
> In my case - armv7, 32 bit - in support/support_path_support_time64.c
> the utimensat_time64 syscall is called, which is wrapped to
> INLINE_SYSCALL_CALL(), which calls __libc_do_syscall ASM ARM function.
>
> This function is only defined for i386 and 32 bit ARM (in e.g.
> sysdeps/unix/sysv/linux/{arm|i386}/libc-do-syscall.S)

Yes, and using the syscall function would avoid this issue.

So the question is, why INLINE_SYSCALL_CALL?

Thanks,
Florian
Lukasz Majewski March 16, 2021, 9:06 a.m. UTC | #4
Hi Florian,

> * Lukasz Majewski:
> 
> > Hi Florian,
> >  
> >> * Lukasz Majewski:
> >>   
> >> > It fixes the following error on ARM 32 bit port:
> >> >
> >> > y2038-glibc/support/support_path_support_time64.c:34: undefined
> >> > reference to `__libc_do_syscall'
> >> > collect2: error: ld returned 1 exit status    
> >> 
> >> Should libsupport use syscall for direct system calls?  
> >
> > I'm not sure what is the correct approach here.
> >
> > In my case - armv7, 32 bit - in
> > support/support_path_support_time64.c the utimensat_time64 syscall
> > is called, which is wrapped to INLINE_SYSCALL_CALL(), which calls
> > __libc_do_syscall ASM ARM function.
> >
> > This function is only defined for i386 and 32 bit ARM (in e.g.
> > sysdeps/unix/sysv/linux/{arm|i386}/libc-do-syscall.S)  
> 
> Yes, and using the syscall function would avoid this issue.
> 
> So the question is, why INLINE_SYSCALL_CALL?

Good question - I guess that Adhemerval would know why it was necessary.

> 
> Thanks,
> Florian
> 




Best regards,

Lukasz Majewski

--

DENX Software Engineering GmbH,      Managing Director: Wolfgang Denk
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-59 Fax: (+49)-8142-66989-80 Email: lukma@denx.de
Adhemerval Zanella March 16, 2021, 12:46 p.m. UTC | #5
On 16/03/2021 02:10, Florian Weimer wrote:
> * Lukasz Majewski:
> 
>> Hi Florian,
>>
>>> * Lukasz Majewski:
>>>
>>>> It fixes the following error on ARM 32 bit port:
>>>>
>>>> y2038-glibc/support/support_path_support_time64.c:34: undefined
>>>> reference to `__libc_do_syscall'
>>>> collect2: error: ld returned 1 exit status  
>>>
>>> Should libsupport use syscall for direct system calls?
>>
>> I'm not sure what is the correct approach here.
>>
>> In my case - armv7, 32 bit - in support/support_path_support_time64.c
>> the utimensat_time64 syscall is called, which is wrapped to
>> INLINE_SYSCALL_CALL(), which calls __libc_do_syscall ASM ARM function.
>>
>> This function is only defined for i386 and 32 bit ARM (in e.g.
>> sysdeps/unix/sysv/linux/{arm|i386}/libc-do-syscall.S)
> 
> Yes, and using the syscall function would avoid this issue.
> 
> So the question is, why INLINE_SYSCALL_CALL?

The __libc_do_syscall is in fact only used thumb mode used as default,
and the direct syscall is used because there is no direct support
for using 64 bit types on ABI with default 32 bit time.  We might 
eventually remove the INLINE_SYSCALL_CALL once 64 bit time support is 
implemented (so we build the file with _TIME_SIZE=64).

The patch itself looks ok.
Adhemerval Zanella March 16, 2021, 1:17 p.m. UTC | #6
On 16/03/2021 09:46, Adhemerval Zanella wrote:
> 
> 
> On 16/03/2021 02:10, Florian Weimer wrote:
>> * Lukasz Majewski:
>>
>>> Hi Florian,
>>>
>>>> * Lukasz Majewski:
>>>>
>>>>> It fixes the following error on ARM 32 bit port:
>>>>>
>>>>> y2038-glibc/support/support_path_support_time64.c:34: undefined
>>>>> reference to `__libc_do_syscall'
>>>>> collect2: error: ld returned 1 exit status  
>>>>
>>>> Should libsupport use syscall for direct system calls?
>>>
>>> I'm not sure what is the correct approach here.
>>>
>>> In my case - armv7, 32 bit - in support/support_path_support_time64.c
>>> the utimensat_time64 syscall is called, which is wrapped to
>>> INLINE_SYSCALL_CALL(), which calls __libc_do_syscall ASM ARM function.
>>>
>>> This function is only defined for i386 and 32 bit ARM (in e.g.
>>> sysdeps/unix/sysv/linux/{arm|i386}/libc-do-syscall.S)
>>
>> Yes, and using the syscall function would avoid this issue.
>>
>> So the question is, why INLINE_SYSCALL_CALL?
> 
> The __libc_do_syscall is in fact only used thumb mode used as default,
> and the direct syscall is used because there is no direct support
> for using 64 bit types on ABI with default 32 bit time.  We might 
> eventually remove the INLINE_SYSCALL_CALL once 64 bit time support is 
> implemented (so we build the file with _TIME_SIZE=64).
> 
> The patch itself looks ok.
> 

And indeed there is no need to use INLINE_SYSCALL_CALL, syscall(...)
should suffice it won't require to handle the ARM __libc_do_syscall.
Lukasz Majewski March 16, 2021, 2:17 p.m. UTC | #7
Hi Adhemerval,

> On 16/03/2021 02:10, Florian Weimer wrote:
> > * Lukasz Majewski:
> >   
> >> Hi Florian,
> >>  
> >>> * Lukasz Majewski:
> >>>  
> >>>> It fixes the following error on ARM 32 bit port:
> >>>>
> >>>> y2038-glibc/support/support_path_support_time64.c:34: undefined
> >>>> reference to `__libc_do_syscall'
> >>>> collect2: error: ld returned 1 exit status    
> >>>
> >>> Should libsupport use syscall for direct system calls?  
> >>
> >> I'm not sure what is the correct approach here.
> >>
> >> In my case - armv7, 32 bit - in
> >> support/support_path_support_time64.c the utimensat_time64 syscall
> >> is called, which is wrapped to INLINE_SYSCALL_CALL(), which calls
> >> __libc_do_syscall ASM ARM function.
> >>
> >> This function is only defined for i386 and 32 bit ARM (in e.g.
> >> sysdeps/unix/sysv/linux/{arm|i386}/libc-do-syscall.S)  
> > 
> > Yes, and using the syscall function would avoid this issue.
> > 
> > So the question is, why INLINE_SYSCALL_CALL?  
> 
> The __libc_do_syscall is in fact only used thumb mode used as default,
> and the direct syscall is used because there is no direct support
> for using 64 bit types on ABI with default 32 bit time.  We might 
> eventually remove the INLINE_SYSCALL_CALL once 64 bit time support is 
> implemented (so we build the file with _TIME_SIZE=64).
> 
> The patch itself looks ok.

Thanks for review. I will apply it if Florian don't mind.


Best regards,

Lukasz Majewski

--

DENX Software Engineering GmbH,      Managing Director: Wolfgang Denk
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-59 Fax: (+49)-8142-66989-80 Email: lukma@denx.de
Adhemerval Zanella March 16, 2021, 2:25 p.m. UTC | #8
On 16/03/2021 11:17, Lukasz Majewski wrote:
> Hi Adhemerval,
> 
>> On 16/03/2021 02:10, Florian Weimer wrote:
>>> * Lukasz Majewski:
>>>   
>>>> Hi Florian,
>>>>  
>>>>> * Lukasz Majewski:
>>>>>  
>>>>>> It fixes the following error on ARM 32 bit port:
>>>>>>
>>>>>> y2038-glibc/support/support_path_support_time64.c:34: undefined
>>>>>> reference to `__libc_do_syscall'
>>>>>> collect2: error: ld returned 1 exit status    
>>>>>
>>>>> Should libsupport use syscall for direct system calls?  
>>>>
>>>> I'm not sure what is the correct approach here.
>>>>
>>>> In my case - armv7, 32 bit - in
>>>> support/support_path_support_time64.c the utimensat_time64 syscall
>>>> is called, which is wrapped to INLINE_SYSCALL_CALL(), which calls
>>>> __libc_do_syscall ASM ARM function.
>>>>
>>>> This function is only defined for i386 and 32 bit ARM (in e.g.
>>>> sysdeps/unix/sysv/linux/{arm|i386}/libc-do-syscall.S)  
>>>
>>> Yes, and using the syscall function would avoid this issue.
>>>
>>> So the question is, why INLINE_SYSCALL_CALL?  
>>
>> The __libc_do_syscall is in fact only used thumb mode used as default,
>> and the direct syscall is used because there is no direct support
>> for using 64 bit types on ABI with default 32 bit time.  We might 
>> eventually remove the INLINE_SYSCALL_CALL once 64 bit time support is 
>> implemented (so we build the file with _TIME_SIZE=64).
>>
>> The patch itself looks ok.
> 
> Thanks for review. I will apply it if Florian don't mind.

Please replace the INLINE_SYSCALL_CALL with a simpler syscall(2) so
it won't require to add libc-do-syscall.
Florian Weimer March 16, 2021, 2:28 p.m. UTC | #9
* Lukasz Majewski:

> Thanks for review. I will apply it if Florian don't mind.

I prefer the syscall-based approach because it makes it easier to build
the test outside of the glibc source tree.

Thanks,
Florian
Lukasz Majewski March 16, 2021, 3:36 p.m. UTC | #10
Hi Florian,

> * Lukasz Majewski:
> 
> > Thanks for review. I will apply it if Florian don't mind.  
> 
> I prefer the syscall-based approach because it makes it easier to
> build the test outside of the glibc source tree.

Ok.

> 
> Thanks,
> Florian
> 




Best regards,

Lukasz Majewski

--

DENX Software Engineering GmbH,      Managing Director: Wolfgang Denk
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-59 Fax: (+49)-8142-66989-80 Email: lukma@denx.de
diff mbox series

Patch

diff --git a/sysdeps/unix/sysv/linux/arm/Makefile b/sysdeps/unix/sysv/linux/arm/Makefile
index abdf01f00c..2344e74554 100644
--- a/sysdeps/unix/sysv/linux/arm/Makefile
+++ b/sysdeps/unix/sysv/linux/arm/Makefile
@@ -55,6 +55,11 @@  LDFLAGS-tst-rfc3484-2 += $(common-objpfx)csu/libc-do-syscall.o
 LDFLAGS-tst-rfc3484-3 += $(common-objpfx)csu/libc-do-syscall.o
 endif
 
+ifeq ($(subdir),support)
+libsupport-sysdep_routines += libc-do-syscall
+libsupport-static-only-routines += libc-do-syscall
+endif
+
 abi-variants := soft hard
 
 ifeq (,$(filter $(default-abi),$(abi-variants)))