Date: Tue, 30 Sep 2003 04:20:09 -0400 (EDT) From: Jeff Roberson <jroberson@chesapeake.net> To: Jeff Roberson <jeff@FreeBSD.org> Cc: cvs-all@FreeBSD.org Subject: Re: cvs commit: src/sys/kern kern_synch.c src/sys/sys proc.h src/sys/i386/i386 genassym.c pmap.c swtch.s src/sys/i386/include pcb.h Message-ID: <20030930041512.X99666-100000@mail.chesapeake.net> In-Reply-To: <200309300811.h8U8BaAQ061807@repoman.freebsd.org>
next in thread | previous in thread | raw e-mail | index | archive | help
This improved my 'make kernel' times by 2%. The effect was less with buildkernel since we spent so much of our time waiting on IO, etc. For an operation like pmap_zero_page() the TLB flush is probably half of the cost of the function. On Tue, 30 Sep 2003, Jeff Roberson wrote: > jeff 2003/09/30 01:11:36 PDT > > FreeBSD src repository > > Modified files: > sys/kern kern_synch.c > sys/sys proc.h > sys/i386/i386 genassym.c pmap.c swtch.s > sys/i386/include pcb.h > Log: > - On my Pentium4-M laptop, invalpg takes ~1100 cycles if the page is found in > the TLB and ~1600 if it is not. Therefore, it is more effecient to > invalidate the TLB after operations that use CMAP rather than before. > - So that the tlb is invalidated prior to switching off of a processor, we > must change the switchin functions to switchout functions. > - Remove td_switchout from the thread and move it to the x86 pcb. > - Move the code that calls switchout into swtch.s. These changes make this > optimization truely x86 specific. > > Revision Changes Path > 1.143 +1 -1 src/sys/i386/i386/genassym.c > 1.438 +33 -33 src/sys/i386/i386/pmap.c > 1.143 +6 -1 src/sys/i386/i386/swtch.s > 1.50 +2 -1 src/sys/i386/include/pcb.h > 1.234 +0 -8 src/sys/kern/kern_synch.c > 1.348 +0 -1 src/sys/sys/proc.h >
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20030930041512.X99666-100000>