Re: Instant reboots with CPUTYPE=pentium-m
Bartosz Fabianowski wrote: I use the athlon-xp switch on 3 boxes with no problems all of them running 5.3 What CFLAGS are you using? I have CFLAGS=-O -pipe in my make.conf. Maybe you have optimization turned off and that's making a difference? I reported the problem with the athlon-xp. My /etc/make.conf looks like the following (when I wish to kill the loader and experience instant reboots, that is ;-): CPUTYPE?= athlon-xp CFLAGS= -O -pipe COPTFLAGS= -O -pipe Note that I only see the problem when building the loader as part of a buildworld. Compiling just the boot stuff in /usr/src/sys/boot/ (i.e., without a bootstrapped gcc) results in a loader that works fine. Thus, I think defining CPUTYPE results in a gcc that produces a bad loader. Jon ___ freebsd-stable@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-stable To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: Instant reboots with CPUTYPE=pentium-m
- Original Message - From: Jon Noack [EMAIL PROTECTED] To: Bartosz Fabianowski [EMAIL PROTECTED] Cc: Chris [EMAIL PROTECTED]; [EMAIL PROTECTED]; [EMAIL PROTECTED] Sent: Thursday, January 06, 2005 7:35 PM Subject: Re: Instant reboots with CPUTYPE=pentium-m Bartosz Fabianowski wrote: I use the athlon-xp switch on 3 boxes with no problems all of them running 5.3 What CFLAGS are you using? I have CFLAGS=-O -pipe in my make.conf. Maybe you have optimization turned off and that's making a difference? I reported the problem with the athlon-xp. My /etc/make.conf looks like the following (when I wish to kill the loader and experience instant reboots, that is ;-): CPUTYPE?= athlon-xp CFLAGS= -O -pipe COPTFLAGS= -O -pipe Note that I only see the problem when building the loader as part of a buildworld. Compiling just the boot stuff in /usr/src/sys/boot/ (i.e., without a bootstrapped gcc) results in a loader that works fine. Thus, I think defining CPUTYPE results in a gcc that produces a bad loader. Mentioning athlon-xp sparked my interest... One of my systems is unstable under load when compiled with CPUTYPE?= athlon-xp, but not with i686. This typically manifests itself as segmentation faults or ICE errors at random places during compile jobs. The bootloader seems always fine however. regards, Derkjan ___ freebsd-stable@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-stable To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: Instant reboots with CPUTYPE=pentium-m
Note that I only see the problem when building the loader as part of a buildworld. Compiling just the boot stuff in /usr/src/sys/boot/ (i.e., without a bootstrapped gcc) results in a loader that works fine. Curious. Either what you are seeing is a different problem then or it somehow shows in a different way than on my system. I would be very interested in investigating this further. To find the root of the cause, I guess it would be best to compare the .s files generated for the loader when a) compiling it as part of the world and b) compiling it on its own. In both cases, CPUTYPE=athlon-xp should be set, of course. Also, you should be using the exact same GCC to eliminate the possibility that different GCC binaries behave in different ways. I am not sure how much space this would take, but if you have the space, it would be great if you could do a full buildworld with -save-temps and then a separate build of the loader, with -save-temps again. For me, one of the offending files is: /usr/obj/usr/src/sys/boot/i386/loader/bcache.s You could try diffing that file and if it shows no differences, all the other .s files in that directory. Some file must change when the loader breaks / unbreaks after all. If you can send me the diffs, I will be glad to look into it. - Bartosz ___ freebsd-stable@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-stable To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: Instant reboots with CPUTYPE=pentium-m
I use the athlon-xp switch on 3 boxes with no problems all of them running 5.3 On Wed, 05 Jan 2005 22:00:04 +0100, Bartosz Fabianowski [EMAIL PROTECTED] wrote: Hi list, I am still having the problems with instant reboots that I reported [1] a couple of weeks ago. I have a bit more info now and hope that someone can lead me in the right direction. The problem is that setting CPUTYPE=pentium-m in make.conf leads to a corrupt boot loader and a corrupt kernel being generated. This setting for the CPUTYPE has been available on -STABLE since 16th December, when revision 1.40.2.1 of bsd.cpu.mk was committed. I have traced the start of my problems to that exact commit (the commiter is CC'd). On -current, the setting has been available for a longer time and has led to the same problems for some people. Also, it seems that not only pentium-m is broken, but athlon-xp as well. The end of a thread on the -current mailing list discussing this issue is in [2]. Unfortunately, it provides no insight as to where the problem lies. It seems to me that every other part of the world, including GCC itself, is built correctly when CPUTYPE is set to pentium-m. The issue only affects the boot loader and the kernel for some reason. When I changed the CPUTYPE from pentium-m down to pentium3 (essentially just disabling SSE2) and recompiled the boot loader, I instantly got a working loader again. I have attached a diff of the .s files generated for the loader with CPUTYPE=pentium3 and CPUTYPE=pentium-m. I do not see any real changes except for the use of xmm registers when CPUTYPE=pentium-m is set. Does anybody have an idea how to debug this further? I am totally out of ideas and really do not know where to continue looking. I have some time on my hands to go searching for the bug - all I need is some direction. - Bartosz [1] http://lists.freebsd.org/pipermail/freebsd-stable/2004-December/010594.html [2] http://lists.freebsd.org/pipermail/freebsd-current/2004-November/042127.html diff -u loader_dir_p3/bcache.s loader_dir_pm/bcache.s --- loader_dir_p3/bcache.s Wed Jan 5 21:56:29 2005 +++ loader_dir_pm/bcache.s Wed Jan 5 21:56:29 2005 @@ -684,7 +684,7 @@ pushl %edi pushl %esi pushl %ebx - subl$36, %esp + subl$28, %esp movl12(%ebp), %ebx movl16(%ebp), %esi leal-16(%ebp), %eax @@ -692,21 +692,19 @@ calltime movl$0, -20(%ebp) movlbcache_ctl, %eax - movl12(%eax), %eax - movl%eax, -24(%ebp) + movd12(%eax), %xmm0 movl$1, %ecx cmplbcache_nblks, %ecx jae .L99 - movlbcache_ctl, %eax - movl%eax, -36(%ebp) - movl%eax, -28(%ebp) - movlbcache_nblks, %eax - movl%eax, -32(%ebp) + movd%eax, %xmm1 + movl%eax, -24(%ebp) + movlbcache_nblks, %edi + movl%edi, -28(%ebp) .p2align 4,,15 .L103: movl%ecx, %eax sall$4, %eax - movl-36(%ebp), %edi + movd%xmm1, %edi movl4(%eax,%edi), %edx xorl%esi, %edx movl(%eax,%edi), %eax @@ -717,18 +715,17 @@ jmp .L99 .p2align 4,,7 .L101: - movl-28(%ebp), %edx + movl-24(%ebp), %edx movl%ecx, %eax sall$4, %eax - movl-24(%ebp), %edi + movd%xmm0, %edi cmpl%edi, 12(%eax,%edx) jge .L100 - movl12(%eax,%edx), %eax - movl%eax, -24(%ebp) + movd12(%eax,%edx), %xmm0 movl%ecx, -20(%ebp) .L100: incl%ecx - cmpl-32(%ebp), %ecx + cmpl-28(%ebp), %ecx jb .L103 .L99: movlbcache_blksize, %eax @@ -753,7 +750,7 @@ movl%eax, bcache_bcount movlbcache_ctl, %eax movl%edx, 12(%ecx,%eax) - addl$36, %esp + addl$28, %esp popl%ebx popl%esi popl%edi diff -u loader_dir_p3/interp_backslash.s loader_dir_pm/interp_backslash.s --- loader_dir_p3/interp_backslash.sWed Jan 5 21:56:29 2005 +++ loader_dir_pm/interp_backslash.sWed Jan 5 21:56:29 2005 @@ -12,7 +12,7 @@ pushl %edi pushl %esi pushl %ebx - subl$32, %esp + subl$28, %esp movl8(%ebp), %ebx movl$0, %edi movl$0, %esi @@ -211,39 +211,39 @@ subl$55, %eax sall$6, %eax .L31: - movl%eax, -24(%ebp) + movd%eax, %xmm0 movsbl 1(%ebx),%eax leal-48(%eax), %edx - movl%edx, -40(%ebp) - movl-24(%ebp), %edx + movl%edx, -36(%ebp) + movd%xmm0, %edx leal-384(%edx,%eax,8), %eax - cmpl$9, -40(%ebp) + cmpl$9, -36(%ebp) jbe
Re: Instant reboots with CPUTYPE=pentium-m
I use the athlon-xp switch on 3 boxes with no problems all of them running 5.3 What CFLAGS are you using? I have CFLAGS=-O -pipe in my make.conf. Maybe you have optimization turned off and that's making a difference? - Bartosz ___ freebsd-stable@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-stable To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: Instant reboots with CPUTYPE=pentium-m
Replying to myself, I have found the problem and filed a but report: http://www.freebsd.org/cgi/query-pr.cgi?pr=75898 Thanks for all the help everybody, - Bartosz ___ freebsd-stable@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-stable To unsubscribe, send any mail to [EMAIL PROTECTED]
Instant reboots with CPUTYPE=pentium-m
Hi list, I am still having the problems with instant reboots that I reported [1] a couple of weeks ago. I have a bit more info now and hope that someone can lead me in the right direction. The problem is that setting CPUTYPE=pentium-m in make.conf leads to a corrupt boot loader and a corrupt kernel being generated. This setting for the CPUTYPE has been available on -STABLE since 16th December, when revision 1.40.2.1 of bsd.cpu.mk was committed. I have traced the start of my problems to that exact commit (the commiter is CC'd). On -current, the setting has been available for a longer time and has led to the same problems for some people. Also, it seems that not only pentium-m is broken, but athlon-xp as well. The end of a thread on the -current mailing list discussing this issue is in [2]. Unfortunately, it provides no insight as to where the problem lies. It seems to me that every other part of the world, including GCC itself, is built correctly when CPUTYPE is set to pentium-m. The issue only affects the boot loader and the kernel for some reason. When I changed the CPUTYPE from pentium-m down to pentium3 (essentially just disabling SSE2) and recompiled the boot loader, I instantly got a working loader again. I have attached a diff of the .s files generated for the loader with CPUTYPE=pentium3 and CPUTYPE=pentium-m. I do not see any real changes except for the use of xmm registers when CPUTYPE=pentium-m is set. Does anybody have an idea how to debug this further? I am totally out of ideas and really do not know where to continue looking. I have some time on my hands to go searching for the bug - all I need is some direction. - Bartosz [1] http://lists.freebsd.org/pipermail/freebsd-stable/2004-December/010594.html [2] http://lists.freebsd.org/pipermail/freebsd-current/2004-November/042127.html diff -u loader_dir_p3/bcache.s loader_dir_pm/bcache.s --- loader_dir_p3/bcache.s Wed Jan 5 21:56:29 2005 +++ loader_dir_pm/bcache.s Wed Jan 5 21:56:29 2005 @@ -684,7 +684,7 @@ pushl %edi pushl %esi pushl %ebx - subl$36, %esp + subl$28, %esp movl12(%ebp), %ebx movl16(%ebp), %esi leal-16(%ebp), %eax @@ -692,21 +692,19 @@ calltime movl$0, -20(%ebp) movlbcache_ctl, %eax - movl12(%eax), %eax - movl%eax, -24(%ebp) + movd12(%eax), %xmm0 movl$1, %ecx cmplbcache_nblks, %ecx jae .L99 - movlbcache_ctl, %eax - movl%eax, -36(%ebp) - movl%eax, -28(%ebp) - movlbcache_nblks, %eax - movl%eax, -32(%ebp) + movd%eax, %xmm1 + movl%eax, -24(%ebp) + movlbcache_nblks, %edi + movl%edi, -28(%ebp) .p2align 4,,15 .L103: movl%ecx, %eax sall$4, %eax - movl-36(%ebp), %edi + movd%xmm1, %edi movl4(%eax,%edi), %edx xorl%esi, %edx movl(%eax,%edi), %eax @@ -717,18 +715,17 @@ jmp .L99 .p2align 4,,7 .L101: - movl-28(%ebp), %edx + movl-24(%ebp), %edx movl%ecx, %eax sall$4, %eax - movl-24(%ebp), %edi + movd%xmm0, %edi cmpl%edi, 12(%eax,%edx) jge .L100 - movl12(%eax,%edx), %eax - movl%eax, -24(%ebp) + movd12(%eax,%edx), %xmm0 movl%ecx, -20(%ebp) .L100: incl%ecx - cmpl-32(%ebp), %ecx + cmpl-28(%ebp), %ecx jb .L103 .L99: movlbcache_blksize, %eax @@ -753,7 +750,7 @@ movl%eax, bcache_bcount movlbcache_ctl, %eax movl%edx, 12(%ecx,%eax) - addl$36, %esp + addl$28, %esp popl%ebx popl%esi popl%edi diff -u loader_dir_p3/interp_backslash.s loader_dir_pm/interp_backslash.s --- loader_dir_p3/interp_backslash.sWed Jan 5 21:56:29 2005 +++ loader_dir_pm/interp_backslash.sWed Jan 5 21:56:29 2005 @@ -12,7 +12,7 @@ pushl %edi pushl %esi pushl %ebx - subl$32, %esp + subl$28, %esp movl8(%ebp), %ebx movl$0, %edi movl$0, %esi @@ -211,39 +211,39 @@ subl$55, %eax sall$6, %eax .L31: - movl%eax, -24(%ebp) + movd%eax, %xmm0 movsbl 1(%ebx),%eax leal-48(%eax), %edx - movl%edx, -40(%ebp) - movl-24(%ebp), %edx + movl%edx, -36(%ebp) + movd%xmm0, %edx leal-384(%edx,%eax,8), %eax - cmpl$9, -40(%ebp) + cmpl$9, -36(%ebp) jbe .L37 movsbl 1(%ebx),%eax leal-97(%eax), %edx - movl%edx, -40(%ebp) - movl-24(%ebp), %edx + movl%edx, -36(%ebp) + movd%xmm0, %edx leal-696(%edx,%eax,8), %eax -
