Patchwork qemu fails to build with glibc-2.15

login
register
mail settings
Submitter X O
Date July 17, 2012, 9:38 a.m.
Message ID <CAEy7y0dCB+xgvrCg1QNpOPR7jGagqAmw6dAdSaGn2hXZauWO-w@mail.gmail.com>
Download mbox | patch
Permalink /patch/171361/
State New
Headers show

Comments

X O - July 17, 2012, 9:38 a.m.
On Mon, Jul 16, 2012 at 6:35 PM, X O <bexoff@gmail.com> wrote:
> On Mon, Jul 16, 2012 at 5:14 PM, Avi Kivity <avi@redhat.com> wrote:
>> (fixed mailing list)
>>
>> On 07/16/2012 03:37 PM, X O wrote:
[...]
>>> Is there a way to compile qemu with glibc-2.15?
[...]
>>> ~~~
>>>   CC    i386-linux-user/syscall.o
>>> /tmp/qemu-1.1.1/linux-user/syscall.c: In function 'sys_getdents':
>>> /tmp/qemu-1.1.1/linux-user/syscall.c:217:1: error: '__NR_getdents'
>>> undeclared (first use in this function)
>>> /tmp/qemu-1.1.1/linux-user/syscall.c:217:1: note: each undeclared
>>> identifier is reported only once for each function it appears in
>>> /tmp/qemu-1.1.1/linux-user/syscall.c: In function '_llseek':
>>> /tmp/qemu-1.1.1/linux-user/syscall.c:224:1: error: '__NR_lseek'
>>> undeclared (first use in this function)

Please, see below. I know it is nowhere near clean solution, but it is
a quick hack which works for me and might be useful for somebody else.
Feel free to kick me in the right direction for a proper patch.

I know, some #ifdefs would have to take a place; "<asm/unistd_32.h>"
won't be available for all platforms and all glibc versions(?) etc.
etc. At least these are the issues I can think off from top of my
head.

~~~[CUT]~~~
~~~[CUT]~~~

Thanks to all whom replied.

Regards,
bexoff

>>
>> Please report the output of
>>
>>   grep -r __NR_getdents /usr/include/
>>
>> on your system.
>>
>
> Here goes:
>
> ~~~
> bash-3.1$ grep -r __NR_getdents /usr/include/
> /usr/include/bits/syscall.h:#define SYS_getdents __NR_getdents
> /usr/include/bits/syscall.h:#define SYS_getdents64 __NR_getdents64
> /usr/include/bits/syscall.h:#define SYS_getdents __NR_getdents
> /usr/include/bits/syscall.h:#define SYS_getdents64 __NR_getdents64
> /usr/include/asm-generic/unistd_32.h:#define __NR_getdents              141
> /usr/include/asm-generic/unistd_32.h:#define __NR_getdents64            220
> /usr/include/asm-generic/unistd_64.h:#define __NR_getdents
>                  78
> /usr/include/asm-generic/unistd_64.h:__SYSCALL(__NR_getdents, sys_getdents)
> /usr/include/asm-generic/unistd_64.h:#define __NR_getdents64
>                  217
> /usr/include/asm-generic/unistd_64.h:__SYSCALL(__NR_getdents64, sys_getdents64)
> /usr/include/asm-generic/unistd.h:#define __NR_getdents64 61
> /usr/include/asm-generic/unistd.h:__SC_COMP(__NR_getdents64,
> sys_getdents64, compat_sys_getdents64)
> /usr/include/asm-generic/unistd.h:#define __NR_getdents 1065
> /usr/include/asm-generic/unistd.h:__SYSCALL(__NR_getdents, sys_getdents)
> ~~~
>
> btw system is x86-64. I've neglected to mention that and it might have
> been important.
>
> Thanks,
> bexoff
>
> PS: I'm not signed up to qemu-devel (still under consideration),
> please keep me on CC.
>
>>
>>
>> --
>> error compiling committee.c: too many arguments to function
>>
>>
Peter Maydell - July 17, 2012, 9:51 a.m.
On 17 July 2012 10:38, X O <bexoff@gmail.com> wrote:
> On Mon, Jul 16, 2012 at 6:35 PM, X O <bexoff@gmail.com> wrote:
>> On Mon, Jul 16, 2012 at 5:14 PM, Avi Kivity <avi@redhat.com> wrote:
>>> (fixed mailing list)
>>>
>>> On 07/16/2012 03:37 PM, X O wrote:
> [...]
>>>> Is there a way to compile qemu with glibc-2.15?
> [...]
>>>> ~~~
>>>>   CC    i386-linux-user/syscall.o
>>>> /tmp/qemu-1.1.1/linux-user/syscall.c: In function 'sys_getdents':
>>>> /tmp/qemu-1.1.1/linux-user/syscall.c:217:1: error: '__NR_getdents'
>>>> undeclared (first use in this function)
>>>> /tmp/qemu-1.1.1/linux-user/syscall.c:217:1: note: each undeclared
>>>> identifier is reported only once for each function it appears in
>>>> /tmp/qemu-1.1.1/linux-user/syscall.c: In function '_llseek':
>>>> /tmp/qemu-1.1.1/linux-user/syscall.c:224:1: error: '__NR_lseek'
>>>> undeclared (first use in this function)
>
> Please, see below. I know it is nowhere near clean solution, but it is
> a quick hack which works for me and might be useful for somebody else.
> Feel free to kick me in the right direction for a proper patch.
>
> I know, some #ifdefs would have to take a place; "<asm/unistd_32.h>"
> won't be available for all platforms and all glibc versions(?) etc.
> etc. At least these are the issues I can think off from top of my
> head.
>
> ~~~[CUT]~~~
> --- linux-user/syscall.c.orig   2012-07-17 09:45:07.000000000 +0200
> +++ linux-user/syscall.c        2012-07-17 10:36:12.000000000 +0200
> @@ -25,6 +25,7 @@
>  #include <endian.h>
>  #include <errno.h>
>  #include <unistd.h>
> +#include <asm/unistd_32.h>

This is (obviously) the wrong thing -- we shouldn't need to
include asm/ headers and certainly including 32 bit only
headers will break 64 bit compiles.

My Ubuntu Precise system has (e)glibc-2.15 and it works fine, because:

* qemu linux-user/syscall.c #includes <linux/unistd.h>
* /usr/include/linux/unistd.h #includes <asm/unistd.h>
* /usr/include/asm/unistd.h #includes either asm/unistd_32.h
  or asm/unistd_64.h as appropriate for 32/64 bit compiles.

Is part of this chain broken on your system?

-- PMM
X O - July 17, 2012, 10:01 a.m.
On Tue, Jul 17, 2012 at 11:51 AM, Peter Maydell
<peter.maydell@linaro.org> wrote:
> On 17 July 2012 10:38, X O <bexoff@gmail.com> wrote:
>> On Mon, Jul 16, 2012 at 6:35 PM, X O <bexoff@gmail.com> wrote:
>>> On Mon, Jul 16, 2012 at 5:14 PM, Avi Kivity <avi@redhat.com> wrote:
>>>> (fixed mailing list)
>>>>
>>>> On 07/16/2012 03:37 PM, X O wrote:
>> [...]
>>>>> Is there a way to compile qemu with glibc-2.15?
>> [...]
>>>>> ~~~
>>>>>   CC    i386-linux-user/syscall.o
>>>>> /tmp/qemu-1.1.1/linux-user/syscall.c: In function 'sys_getdents':
>>>>> /tmp/qemu-1.1.1/linux-user/syscall.c:217:1: error: '__NR_getdents'
>>>>> undeclared (first use in this function)
>>>>> /tmp/qemu-1.1.1/linux-user/syscall.c:217:1: note: each undeclared
>>>>> identifier is reported only once for each function it appears in
>>>>> /tmp/qemu-1.1.1/linux-user/syscall.c: In function '_llseek':
>>>>> /tmp/qemu-1.1.1/linux-user/syscall.c:224:1: error: '__NR_lseek'
>>>>> undeclared (first use in this function)
>>
>> Please, see below. I know it is nowhere near clean solution, but it is
>> a quick hack which works for me and might be useful for somebody else.
>> Feel free to kick me in the right direction for a proper patch.
>>
>> I know, some #ifdefs would have to take a place; "<asm/unistd_32.h>"
>> won't be available for all platforms and all glibc versions(?) etc.
>> etc. At least these are the issues I can think off from top of my
>> head.
>>
>> ~~~[CUT]~~~
>> --- linux-user/syscall.c.orig   2012-07-17 09:45:07.000000000 +0200
>> +++ linux-user/syscall.c        2012-07-17 10:36:12.000000000 +0200
>> @@ -25,6 +25,7 @@
>>  #include <endian.h>
>>  #include <errno.h>
>>  #include <unistd.h>
>> +#include <asm/unistd_32.h>
>
> This is (obviously) the wrong thing -- we shouldn't need to
> include asm/ headers and certainly including 32 bit only
> headers will break 64 bit compiles.
>

This is 64bit build.

> My Ubuntu Precise system has (e)glibc-2.15 and it works fine, because:

eglibc ???

>
> * qemu linux-user/syscall.c #includes <linux/unistd.h>
> * /usr/include/linux/unistd.h #includes <asm/unistd.h>

No, it doesn't. It includes <bits/unistd.h> and it "ends" there. It
doesn't work as "expected" anymore(?), resp. beyond glibc-2.15.

> * /usr/include/asm/unistd.h #includes either asm/unistd_32.h
>   or asm/unistd_64.h as appropriate for 32/64 bit compiles.
>
> Is part of this chain broken on your system?
>

Yes, it is "broken" as explained above.

-- bexoff

> -- PMM

Patch

--- linux-user/syscall.c.orig   2012-07-17 09:45:07.000000000 +0200
+++ linux-user/syscall.c        2012-07-17 10:36:12.000000000 +0200
@@ -25,6 +25,7 @@ 
 #include <endian.h>
 #include <errno.h>
 #include <unistd.h>
+#include <asm/unistd_32.h>
 #include <fcntl.h>
 #include <time.h>
 #include <limits.h>
@@ -200,11 +201,6 @@ 
 #define __NR_sys_inotify_add_watch __NR_inotify_add_watch
 #define __NR_sys_inotify_rm_watch __NR_inotify_rm_watch

-#if defined(__alpha__) || defined (__ia64__) || defined(__x86_64__) || \
-    defined(__s390x__)
-#define __NR__llseek __NR_lseek
-#endif
-
 #ifdef __NR_gettid
 _syscall0(int, gettid)
 #else