Re: [PATCH v6 14/19] arm64:ilp32: add sys_ilp32.c and a separate table (in entry.S) to use it

From: Arnd Bergmann
Date: Tue Nov 17 2015 - 17:05:48 EST


On Wednesday 18 November 2015 00:16:54 Yury Norov wrote:
> From: Andrew Pinski <apinski@xxxxxxxxxx>
>
> Add a separate syscall-table for ILP32, which dispatches either to native
> LP64 system call implementation or to compat-syscalls, as appropriate.

I like it much better than the previous version, thanks for the rework!

However, it seems that you accidentally have a lot more redirects
than you should have:

> +/* Using non-compat syscalls where necessary */
> +#define compat_sys_fadvise64_64 sys_fadvise64_64
> +#define compat_sys_fallocate sys_fallocate
> +#define compat_sys_ftruncate64 sys_ftruncate
> +#define compat_sys_pread64 sys_pread64
> +#define compat_sys_pwrite64 sys_pwrite64
> +#define compat_sys_readahead sys_readahead

Makes sense. These of course all require the respective changes
in glibc as discussed in the thread regarding loff_t handling.

> +#define compat_sys_rt_sigaction sys_rt_sigaction
> +#define compat_sys_shmat sys_shmat

What's special about compat_sys_shmat?

> +#define compat_sys_sync_file_range sys_sync_file_range
> +#define compat_sys_truncate64 sys_truncate
> +#define compat_sys_sigaltstack sys_sigaltstack
> +
> +#define compat_sys_io_getevents sys_io_getevents

io_getevents seems wrong, you are passing the wrong timespec and
aio_context_t here.

> +#define compat_sys_lookup_dcookie sys_lookup_dcookie
> +#define compat_sys_epoll_pwait sys_epoll_pwait

epoll_pwait takes sigset_t, which I'd assume is different between
ilp32 and lp64, so this is probably wrong too, at least on big-endian.

> +#define compat_sys_fcntl64 compat_sys_fcntl

This uses compat_off_t, not compat_loff_t, and needs to be changed.

> +#define compat_sys_signalfd4 sys_signalfd4
> +#define compat_sys_rt_sigsuspend sys_rt_sigsuspend
> +#define compat_sys_rt_sigprocmask sys_rt_sigprocmask
> +#define compat_sys_rt_sigpending sys_rt_sigpending

sigset_t again, all four of these.

> +#define compat_sys_rt_sigqueueinfo sys_rt_sigqueueinfo

this looks ok though, as you have the 64-bit siginfo

> +#define compat_sys_semtimedop sys_semtimedop

timespec again

> +#define compat_sys_rt_tgsigqueueinfo sys_rt_tgsigqueueinfo
> +
> +#define compat_sys_timer_create sys_timer_create
> +#define compat_sys_timer_gettime sys_timer_gettime
> +#define compat_sys_timer_settime sys_timer_settime

timespec again for gettime/settime. create seems fine if you require
the use of the 64-bit sigevent (why?)

> +#define compat_sys_rt_sigtimedwait sys_rt_sigtimedwait

This one probably needs a custom wrapper as you have the 64-bit
siginfo, but the 32-bit sigset and timespec.

> +#define compat_sys_mq_open sys_mq_open
> +#define compat_sys_mq_timedsend sys_mq_timedsend
> +#define compat_sys_mq_timedreceive sys_mq_timedreceive
> +#define compat_sys_mq_getsetattr sys_mq_getsetattr
> +#define compat_sys_mq_open sys_mq_open

You have compat_sys_mq_open twice, and they all look wrong because
you get the wrong struct mq_attr.

> +#define compat_sys_open_by_handle_at sys_open_by_handle_at

The only difference here is the forced O_LARGEFILE, but that
is set by glibc anyway, right?

> +#define compat_sys_clock_adjtime sys_clock_adjtime

wrong timex structure

> +#define compat_sys_openat sys_openat

same as open_by_handle_at

> +#define compat_sys_getdents64 sys_getdents64

glibc uses linux_dirent64 for 32-bit architectures, so this looks wrong

> +#define compat_sys_waitid sys_waitid

This will probably need a separate wrapper to convert rusage but not siginfo

> +#define compat_sys_timer_settime sys_timer_settime
> +#define compat_sys_sched_rr_get_interval sys_sched_rr_get_interval

timespec again

> +#define compat_sys_execveat sys_execveat

This probably gives you the wrong struct user_arg_ptr

> +#define compat_sys_mq_notify sys_mq_notify

ok.

> +#define compat_sys_clock_nanosleep sys_clock_nanosleep
> +#define compat_sys_clock_getres sys_clock_getres

timespec

> +#define sys_lseek sys_llseek

This seems pointless, as there is no sys_lseek

> +asmlinkage long compat_sys_mmap2_wrapper(void);
> +#define sys_mmap2 compat_sys_mmap2_wrapper
> +
> +asmlinkage long compat_sys_fstatfs64_wrapper(void);
> +#define compat_sys_fstatfs64 compat_sys_fstatfs64_wrapper
> +asmlinkage long compat_sys_statfs64_wrapper(void);
> +#define compat_sys_statfs64 compat_sys_statfs64_wrapper

What are the wrappers for again? Maybe add a comment here.

> +#define compat_sys_preadv compat_sys_preadv64
> +#define compat_sys_pwritev compat_sys_pwritev64

wrong iovec.

Arnd

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/