On Wed, Jul 4, 2018 at 1:49 PM, Guo Ren [off-list ref] wrote:

quoted

quoted

+#define __NR_set_thread_area   (__NR_arch_specific_syscall + 0)
+__SYSCALL(__NR_set_thread_area, sys_set_thread_area)
+#define __NR_ipc               (__NR_arch_specific_syscall + 1)
+__SYSCALL(__NR_ipc, sys_ipc)
+#define __NR_socketcall                (__NR_arch_specific_syscall + 2)
+__SYSCALL(__NR_socketcall, sys_socketcall)
+#define __NR_ugetrlimit                (__NR_arch_specific_syscall + 3)
+__SYSCALL(__NR_ugetrlimit, sys_getrlimit)
+#define __NR_cacheflush                (__NR_arch_specific_syscall + 4)
+__SYSCALL(__NR_cacheflush, sys_cacheflush)
+#define __NR_sysfs             (__NR_arch_specific_syscall + 5)
+__SYSCALL(__NR_sysfs, sys_sysfs)
+
+__SYSCALL(__NR_fadvise64_64, sys_csky_fadvise64_64)

We definitely don't want ipc, socketcall, ugetrlimit, or sysfs.

Ok, remove them.

quoted

For fadvise64_64, please redefine the symbol name so the
table points at the right entry.

We need exchange the args for abiv1. loff_t is 64bit and abiv1 need
8-bytes align in args.
/*
 * for abiv1 the 64bits args should be even th, So we need mov the advice forward.
 */
SYSCALL_DEFINE4(csky_fadvise64_64,
        int, fd,
        int, advice,
        loff_t, offset,
        loff_t, len)
{
        return sys_fadvise64_64(fd, offset, len, advice);
}

Right, I do understand what it's used for, my point was that you
don't really need a separate system call number for it, just redirect
the entry point using the same trick that nds32 has in
arch/nds32/kernel/syscall_table.c:

#define sys_fadvise64_64 csky_fadvise64_64

quoted

I'm not completely sure about set_thread_area, can you explain
what you need that for?

In abiv1 there is no tls register, so we use "trap 3" for csky_get_tls
defined in arch/csky/kernel/entry.S to get tls.

Also we use set_thread_area to set tls in kernel.

For abiv2 it has r31 for tls-reg, but we still keep the mechanism.

ok.

            Arnd