Date: Sun, 25 Feb 2001 20:23:32 +1100 (EST) From: Bruce Evans <bde@zeta.org.au> To: Peter Wemm <peter@netplex.com.au> Cc: Jake Burkholder <jake@FreeBSD.org>, cvs-committers@FreeBSD.org, cvs-all@FreeBSD.org Subject: Re: cvs commit: src/sys/i386/i386 db_trace.c exception.s machdep.c trap.c src/sys/i386/include asnames.h Message-ID: <Pine.BSF.4.21.0102251955220.6715-100000@besplex.bde.org> In-Reply-To: <200102250736.f1P7aVi01775@mobile.wemm.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On Sat, 24 Feb 2001, Peter Wemm wrote: > Bruce Evans wrote: > > On Sat, 24 Feb 2001, Jake Burkholder wrote: > > > - Rename the lcall system call handler from Xsyscall to Xlcall_syscall > > > to be more like Xint0x80_syscall and less like c function syscall(). > > > > This is sort of backwards. Xsyscall is supposed to be named like the C > > function syscall(), but it should be used for normal int 0x80 syscalls. > > Well, given that we have two, "Xsyscall" was wrong as it implies that it > was the "normal" one (it is not). Would you rather that Xint0x80syscall > be renamed to Xsyscall instead? Yes. > > > - Reduce code duplication between the int0x80 and lcall handlers by > > > shuffling the elfags into the right place, saving the sizeof the > > > instruction in tf_err and jumping into the common int0x80 code. > > > > This breaks profiling and pessimizes Xsyscall a little. Please back this > > out. > > Are you sure? The X*syscall stubs dont seem to be profiled.. Yes. The FAKE_MCOUNT stuff does it. > > #define IDTVEC(name) ALIGN_TEXT; .globl __CONCAT(X,name); \ > .type __CONCAT(X,name),@function; __CONCAT(X,name): > > SUPERALIGN_TEXT > IDTVEC(int0x80_syscall) > pushl $2 /* sizeof "int 0x80" */ > syscall_with_err_pushed: > subl $4,%esp /* skip over tf_trapno */ > pushal > pushl %ds > pushl %es > pushl %fs > mov $KDSEL,%ax /* switch to kernel segments */ > mov %ax,%ds > mov %ax,%es > mov $KPSEL,%ax > mov %ax,%fs > FAKE_MCOUNT(13*4(%esp)) > call syscall > MEXITCOUNT > jmp doreti > > Obviously I could easily be misunderstanding things, but the FAKE_MCOUNT > pulls the caller off the stack, right? How does this all tie in? > Would this change imply that the cpu time of the lcall version would be > applied to the int0x80 version? It's fairly complicated. The call to mcount must be delayed until enough registers are saved and enough of the frame is set up. Since we jump before doing any mcount stuff, the Xlcall_syscall will be merged with Xint0x80_syscall as far as mcount can see. The frame is uniform (I hope), so 13*4(%esp) gives the correct syscaller's address for both cases; only the address of mcount's caller is messed up. For low resolution profiling, tick counters are updated indpendently and will be reported by gprof as "<spontaneous>" activity in Xlcall_syscall. For high resolution profiling, timestamps are read by mcount, so all activity for Xlcall_syscall will be counted as for Xint0x80_syscall. Bruce To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe cvs-all" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.BSF.4.21.0102251955220.6715-100000>