On 10/21/24 2:33 AM, David Hildenbrand wrote:
Am 21.10.24 um 08:48 schrieb Muhammad Usama Anjum:
...
But now comes the tricky part: an architecture defines whether it wants to
(a) Use the asm-generic unistd.h
(b) Use a custom one
E.g.,
$ cat include/uapi/linux/unistd.h
/* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */
#ifndef _LINUX_UNISTD_H_
#define _LINUX_UNISTD_H_
/*
* Include machine specific syscall numbers
*/
#include <asm/unistd.h>
#endif /* _LINUX_UNISTD_H_ */
For example on riscv arch/riscv/include/asm/unistd.h will include arch/riscv/include/uapi/asm/unistd.h which will include "asm-generic/unistd.h".
If you follow the flow on x86, you'll find that it will not include that asm-generic one as default.
So the asm-generic variant only applies if an arch wants to do it in the generic way.
$ find tools -name unistd.h
tools/arch/x86/include/uapi/asm/unistd.h
tools/arch/arc/include/uapi/asm/unistd.h
tools/arch/riscv/include/uapi/asm/unistd.h
tools/arch/hexagon/include/uapi/asm/unistd.h
tools/arch/arm64/include/uapi/asm/unistd.h
tools/arch/loongarch/include/uapi/asm/unistd.h
tools/include/uapi/asm-generic/unistd.h
tools/include/nolibc/unistd.h
Consequently, the asm-generic one should never be used directly.
ohhh, I think I may have inadvertently started this problem! Via a few
commits such as:
a5c6bc590094 ("selftests/mm: remove local __NR_* definitions")
, which did things like this:
-#include <unistd.h>
+#include <asm-generic/unistd.h>
So it seems that it should have been:
-#include <unistd.h>
+#include <asm/unistd.h>
...and each arch's unistd.h needs to be checked to ensure that it is
up to date with all the symbols that kselftests need.
thanks,
--
John Hubbard
