[PATCH 0/3] fix several compilation and linking errors with old toolchains

[PATCH 0/3] fix several compilation and linking errors with old toolchains

[Sebastian Smolorz]

Sebastian Smolorz (3):
  lib/cobalt: Do not call glibc's {recv|send}mmsg if it is too old
  cobalt/time: add missing include for old toolchains
  utils/analogy: fix compilation of analogy_calibrate with old
    toolchains

 include/cobalt/time.h             | 1 +
 lib/cobalt/wrappers.c             | 8 ++++++++
 utils/analogy/analogy_calibrate.c | 1 -
 3 files changed, 9 insertions(+), 1 deletion(-)

-- 
2.7.4

[PATCH 1/3] lib/cobalt: Do not call glibc's {recv|send}mmsg if it is too old

[Sebastian Smolorz]

Fix linking error when using glibc versions which do not provide
recvmmsg or sendmmsg.

Signed-off-by: Sebastian Smolorz <sebastian.smolorz@gmx.de>
---
 lib/cobalt/wrappers.c | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/lib/cobalt/wrappers.c b/lib/cobalt/wrappers.c
index 20ad63a..951cd15 100644
--- a/lib/cobalt/wrappers.c
+++ b/lib/cobalt/wrappers.c
@@ -260,7 +260,11 @@ __weak
 int __real_recvmmsg(int fd, struct mmsghdr *msgvec, unsigned int vlen,
 		    unsigned int flags, struct timespec *timeout)
 {
+#if (__GLIBC__ == 2) && (__GLIBC_MINOR__ < 12)
+	return -ENOSYS;
+#else
 	return recvmmsg(fd, msgvec, vlen, flags, timeout);
+#endif
 }
 
 __weak
@@ -273,7 +277,11 @@ __weak
 int __real_sendmmsg(int fd, struct mmsghdr *msgvec, unsigned int vlen,
 		    unsigned int flags)
 {
+#if (__GLIBC__ == 2) && (__GLIBC_MINOR__ < 14)
+	return -ENOSYS;
+#else
 	return sendmmsg(fd, msgvec, vlen, flags);
+#endif
 }
 
 __weak
-- 
2.7.4

[PATCH 2/3] cobalt/time: add missing include for old toolchains

[Sebastian Smolorz]

Without this patch compilation fails with

lib/cobalt/clock.c:251: error: conflicting types for '__wrap_clock_adjtime'
include/cobalt/time.h:43: error: previous declaration of '__wrap_clock_adjtime' was here
lib/cobalt/clock.c:251: error: conflicting types for '__cobalt_clock_adjtime'
include/cobalt/time.h:43: error: previous declaration of '__cobalt_clock_adjtime' was here

Signed-off-by: Sebastian Smolorz <sebastian.smolorz@gmx.de>
---
 include/cobalt/time.h | 1 +
 1 file changed, 1 insertion(+)

diff --git a/include/cobalt/time.h b/include/cobalt/time.h
index 9bafe34..6c65c77 100644
--- a/include/cobalt/time.h
+++ b/include/cobalt/time.h
@@ -23,6 +23,7 @@
 
 /* Re-read in case we came from selective __need* block. */
 #include_next <time.h>
+#include <sys/timex.h>
 #include <cobalt/wrappers.h>
 #include <cobalt/uapi/time.h>
 
-- 
2.7.4

[PATCH 3/3] utils/analogy: fix compilation of analogy_calibrate with old toolchains

[Sebastian Smolorz]

The previous commit introduced a fix for old toolchains by including
sys/timex.h in include/cobalt/time.h. However, that breaks the
compilation of analogy_calibrate. Removing #include <sys/time.h>
from analogy_calibrate.c fixes compilation again.

Signed-off-by: Sebastian Smolorz <sebastian.smolorz@gmx.de>
---
 utils/analogy/analogy_calibrate.c | 1 -
 1 file changed, 1 deletion(-)

diff --git a/utils/analogy/analogy_calibrate.c b/utils/analogy/analogy_calibrate.c
index c5990a7..b927ee2 100644
--- a/utils/analogy/analogy_calibrate.c
+++ b/utils/analogy/analogy_calibrate.c
@@ -20,7 +20,6 @@
  * Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
  */
 
-#include <sys/time.h>
 #include <sys/resource.h>
 #include <getopt.h>
 #include <pthread.h>
-- 
2.7.4

Re: [PATCH 1/3] lib/cobalt: Do not call glibc's {recv|send}mmsg if it is too old

[Jan Kiszka]

On 30.10.18 11:29, Sebastian Smolorz wrote:
> Fix linking error when using glibc versions which do not provide
> recvmmsg or sendmmsg.
> 
> Signed-off-by: Sebastian Smolorz <sebastian.smolorz@gmx.de>
> ---
>   lib/cobalt/wrappers.c | 8 ++++++++
>   1 file changed, 8 insertions(+)
> 
> diff --git a/lib/cobalt/wrappers.c b/lib/cobalt/wrappers.c
> index 20ad63a..951cd15 100644
> --- a/lib/cobalt/wrappers.c
> +++ b/lib/cobalt/wrappers.c
> @@ -260,7 +260,11 @@ __weak
>   int __real_recvmmsg(int fd, struct mmsghdr *msgvec, unsigned int vlen,
>   		    unsigned int flags, struct timespec *timeout)
>   {
> +#if (__GLIBC__ == 2) && (__GLIBC_MINOR__ < 12)
> +	return -ENOSYS;
> +#else
>   	return recvmmsg(fd, msgvec, vlen, flags, timeout);
> +#endif
>   }
>   
>   __weak
> @@ -273,7 +277,11 @@ __weak
>   int __real_sendmmsg(int fd, struct mmsghdr *msgvec, unsigned int vlen,
>   		    unsigned int flags)
>   {
> +#if (__GLIBC__ == 2) && (__GLIBC_MINOR__ < 14)
> +	return -ENOSYS;
> +#else
>   	return sendmmsg(fd, msgvec, vlen, flags);
> +#endif
>   }
>   
>   __weak
> 

Shouldn't we rather probe for the availability of such functions during configure?

Jan

-- 
Siemens AG, Corporate Technology, CT RDA IOT SES-DE
Corporate Competence Center Embedded Linux

Re: [PATCH 1/3] lib/cobalt: Do not call glibc's {recv|send}mmsg if it is too old

[Philippe Gerum]

On 10/31/18 6:45 PM, Jan Kiszka wrote:
> On 30.10.18 11:29, Sebastian Smolorz wrote:
>> Fix linking error when using glibc versions which do not provide
>> recvmmsg or sendmmsg.
>>
>> Signed-off-by: Sebastian Smolorz <sebastian.smolorz@gmx.de>
>> ---
>>   lib/cobalt/wrappers.c | 8 ++++++++
>>   1 file changed, 8 insertions(+)
>>
>> diff --git a/lib/cobalt/wrappers.c b/lib/cobalt/wrappers.c
>> index 20ad63a..951cd15 100644
>> --- a/lib/cobalt/wrappers.c
>> +++ b/lib/cobalt/wrappers.c
>> @@ -260,7 +260,11 @@ __weak
>>   int __real_recvmmsg(int fd, struct mmsghdr *msgvec, unsigned int vlen,
>>               unsigned int flags, struct timespec *timeout)
>>   {
>> +#if (__GLIBC__ == 2) && (__GLIBC_MINOR__ < 12)
>> +    return -ENOSYS;
>> +#else
>>       return recvmmsg(fd, msgvec, vlen, flags, timeout);
>> +#endif
>>   }
>>     __weak
>> @@ -273,7 +277,11 @@ __weak
>>   int __real_sendmmsg(int fd, struct mmsghdr *msgvec, unsigned int vlen,
>>               unsigned int flags)
>>   {
>> +#if (__GLIBC__ == 2) && (__GLIBC_MINOR__ < 14)
>> +    return -ENOSYS;
>> +#else
>>       return sendmmsg(fd, msgvec, vlen, flags);
>> +#endif
>>   }
>>     __weak
>>
> 
> Shouldn't we rather probe for the availability of such functions during
> configure?
> 

Definitely.

-- 
Philippe.

Re: [PATCH 3/3] utils/analogy: fix compilation of analogy_calibrate with old toolchains

[Jan Kiszka]

On 30.10.18 11:29, Sebastian Smolorz wrote:
> The previous commit introduced a fix for old toolchains by including
> sys/timex.h in include/cobalt/time.h. However, that breaks the
> compilation of analogy_calibrate. Removing #include <sys/time.h>
> from analogy_calibrate.c fixes compilation again.
> 

Can you explain why? :)

> Signed-off-by: Sebastian Smolorz <sebastian.smolorz@gmx.de>
> ---
>   utils/analogy/analogy_calibrate.c | 1 -
>   1 file changed, 1 deletion(-)
> 
> diff --git a/utils/analogy/analogy_calibrate.c b/utils/analogy/analogy_calibrate.c
> index c5990a7..b927ee2 100644
> --- a/utils/analogy/analogy_calibrate.c
> +++ b/utils/analogy/analogy_calibrate.c
> @@ -20,7 +20,6 @@
>    * Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
>    */
>   
> -#include <sys/time.h>
>   #include <sys/resource.h>
>   #include <getopt.h>
>   #include <pthread.h>
> 

Please fold patch 3 into 2 so that we do not have a build breakage in the middle.

Jan

-- 
Siemens AG, Corporate Technology, CT RDA IOT SES-DE
Corporate Competence Center Embedded Linux