sorry......i think i know who is the caller:
i86pc/ml/locore.s:
1553
1554 /*
1555 * XXX - may want a fast path that avoids sys_rtt_common in the
1556 * most common case.
1557 */
1558 ALTENTRY(_sys_rtt)
1559 CLI(%rax) /* disable interrupts */
1560 ALTENTRY(_sys_rtt_ints_disabled)
1561 movq %rsp, %rdi /* pass rp to sys_rtt_common */
1562 call sys_rtt_common /* do common sys_rtt tasks */
1563 testq %rax, %rax /* returning to userland? */
1564 jz sr_sup
1565
1566 /*
1567 * Return to user
1568 */
1569 ASSERT_UPCALL_MASK_IS_SET
1570 cmpw $UCS_SEL, REGOFF_CS(%rsp) /* test for native
(64-bit) lwp? */
1571 je sys_rtt_syscall
1572
1573 /*
1574 * Return to 32-bit userland
1575 */
1576 ALTENTRY(sys_rtt_syscall32)
1577 USER32_POP
1578
the above is difficult to read, the below is far easier:
lwp_rtt+0x55: movq 0x20(%rsp),%rsi
lwp_rtt+0x5a: movq 0x40(%rsp),%rdi
lwp_rtt+0x5f: call +0x1843b4 <post_syscall>
lwp_rtt+0x64: movq %cr0,%rdi
lwp_rtt+0x67: orq $0x8,%rdi
lwp_rtt+0x6b: movq %rdi,%cr0
_sys_rtt: cli
_sys_rtt_ints_disabled: movq %rsp,%rdi
_sys_rtt_ints_disabled+3: call +0x2b431 <sys_rtt_common>
_sys_rtt_ints_disabled+8: testq %rax,%rax
_sys_rtt_ints_disabled+0xb: je +0x91 <sr_sup>
_sys_rtt_ints_disabled+0x11: cmpw $0x53,0xd0(%rsp)
_sys_rtt_ints_disabled+0x1a: je +0x28 <sys_rtt_syscall>
sys_rtt_syscall32: movl 0x10(%rsp),%edi
sys_rtt_syscall32+4: movl 0x18(%rsp),%esi
sys_rtt_syscall32+8: movl 0x20(%rsp),%edx
sys_rtt_syscall32+0xc: movl 0x28(%rsp),%ecx
sys_rtt_syscall32+0x10: movl 0x40(%rsp),%eax
sys_rtt_syscall32+0x14: movl 0x48(%rsp),%ebx
sys_rtt_syscall32+0x18: movl 0x50(%rsp),%ebp
so i guessed the processing will flow through from lwp_rtt, to
sys_rtt, to sys_rtt_syscall32 etc etc.....and calling sys_rtt_common()
from in between.
but the intent is still vague for me?
On Thu, Nov 19, 2009 at 4:38 PM, Peter Teoh <htmldevelo...@gmail.com> wrote:
> Writing the "most generic" dtrace script possible (please correct me
> if wrong :-)):
>
> pid$target:::entry,
> pid$target:::return
> / execname == "YYYYY" /
> {
> printf("u:");
> }
>
> fbt::*:entry,
> fbt::*:return
> / execname == "YYYYY" /
> {
> printf("k:");
> }
>
> I managed to trace the transition from user to kernel and vice versa:
>
> 1 6847 sys_rtt_common:return k:
> 1 67621 mutex_lock:return u:
> 1 6846 sys_rtt_common:entry k:
> 1 6847 sys_rtt_common:return k:
> 1 62685 _malloc_unlocked:entry u:
> 1 6846 sys_rtt_common:entry k:
> 1 6847 sys_rtt_common:return k:
> 1 62693 cleanfree:entry u:
> 1 6846 sys_rtt_common:entry k:
> 1 5746 trap:entry k:
> 1 6940 do_interrupt:entry k:
> 1 6842 cpu_idle_exit:entry k:
> 1 6843 cpu_idle_exit:return k:
> 1 26906 apic_intr_enter:entry k:
> 1 6190 psm_get_cpu_id:entry k:
> 1 6191 psm_get_cpu_id:return k:
> 1 26907 apic_intr_enter:return k:
> 1 3700 hilevel_intr_prolog:entry k:
>
> the thing is that i keep seeing this sys_rtt_common() everywhere.
> what is the purpose of this function? the code is not understood by
> me either:
>
> http://cvs.opensolaris.org/source/xref/onnv/onnv-gate/usr/src/uts/i86pc/os/intr.c
>
> 994 /*
> 995 * Common tasks always done by _sys_rtt, called with
> interrupts disabled.
> 996 * Returns 1 if returning to userland, 0 if returning to system mode.
> 997 */
> 998 int
> 999 sys_rtt_common(struct regs *rp)
> 1000 {
>
> I cannot answer - who call sys_rtt_common()? from any userspace API
> (cleanfree() for example), how is the transition to kernel's
> sys_rtt_common() done?
>
> SYSENTER? (what code)?
>
> and inside the function - why is it sometime returning back to
> userspace, and sometime continuing processing in kernelspace?
>
> --
> Regards,
> Peter Teoh
>
--
Regards,
Peter Teoh
_______________________________________________
opensolaris-code mailing list
opensolaris-code@opensolaris.org
http://mail.opensolaris.org/mailman/listinfo/opensolaris-code
