On Sat, 2006-12-09 at 18:34 -0800, Pete Zaitcev wrote: > On Sat, 9 Dec 2006 15:09:13 -0600, Erik Jacobson <[EMAIL PROTECTED]> wrote: > > > > Please try to declare u64 timestamp_ns, then copy it into the *ev > > > instead of copying whole *ev. This ought to fix the problem if > > > buffer[] ends aligned to 32 bits or better. > > > > So I took this suggestion for a spin and met with the same result. > > The unaligned access messages are still produced. > > I see. And I see you went a few steps forward with dignosing it: > > > dbg fork after timespec_to_ns call, b4 memcpy > > kernel unaligned access to 0xe000003076b6fbe4, ip=0xa0000001004f1480 > > dbg fork after memcpy, b4 other ev settings... > > > a0000001004f1470 <proc_fork_connector+0x1f0> [MMI] ld8 r40=[r14] > > a0000001004f1476 <proc_fork_connector+0x1f6> ld8 r38=[r38] > > a0000001004f147c <proc_fork_connector+0x1fc> nop.i 0x0;; > > a0000001004f1480 <proc_fork_connector+0x200> [MIB] st8 [r39]=r40 > > a0000001004f1486 <proc_fork_connector+0x206> nop.i 0x0 > > a0000001004f148c <proc_fork_connector+0x20c> br.call.sptk.many > > b0=a0000001000a36c0 <printk>;;
I'm not very familiar with ia64 asm but it looks like its loading and storying 8 bytes at a time for the memcpy(). > It seems rather strange that memcpy gets optimized this way. I could > not have foreseen it. Still, it was worth a try, even if putting 32 > extra bytes on stack and running memcpy on them does not seem too > onerous, for a fork(). Thanks for doing it, and let's go with your > original patch then... if Matt Helsley does not mind. OK, I'll ack the original. > Thank you, > -- Pete I'm shocked memcpy() introduces 8-byte stores that violate architecture alignment rules. Is there any chance this a bug in ia64's memcpy() implementation? I've tried to read it but since I'm not familiar with ia64 asm I can't make out significant parts of it in arch/ia64/lib/memcpy.S. <snip> Cheers, -Matt Helsley - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/