RE: cc: Internal error: Illegal instruction (program as)
Paul A. Howes wrote: All; These two kernel options seem to have solved the problem. Builds now run smoothly and error-free. I read the comments in NOTES about these options and something clicked: I recall that most Pentium processors will only deal with 4 kB pages. Aren't 4 MB pages a feature of the Xeon processor, as it can address much more memory? 4Mb pages were introduced in Intel Pentium along with other V86 mode extensions, to increase Translation Lookaside Buffer (TLB) hit rate. 36-bit address path was implemented in Pentium Pro to access 64Gb address range, though CPU could work in 32-bit mode, too; ASZ[1:0]# signals were used in order to specify address bus width. I suppose P4 behaves somewhat similar, though its bus interface is totally different to GTL\AGTL used in PPro to PIII systems. --- Regards, Rhett __ Do you Yahoo!? Yahoo! Tax Center - forms, calculators, tips, more http://taxes.yahoo.com/ To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-current in the body of the message
RE: cc: Internal error: Illegal instruction (program as)
All; These two kernel options seem to have solved the problem. Builds now run smoothly and error-free. I read the comments in NOTES about these options and something clicked: I recall that most Pentium processors will only deal with 4 kB pages. Aren't 4 MB pages a feature of the Xeon processor, as it can address much more memory? If that is the case, then these options should be the default, and their inverse provided in the kernel configuration file (ENABLE_PSE ENABLE_PG_G). -- Paul A. Howes -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of John Hay Sent: Saturday, February 22, 2003 1:05 PM To: Paul A. Howes Cc: [EMAIL PROTECTED] Subject: Re: cc: Internal error: Illegal instruction (program as) On Sat, Feb 22, 2003 at 11:43:01AM -0500, Paul A. Howes wrote: All, I am receiving some strange errors during a buildworld of 5.0-RELEASE-p2 from 5.0-RELEASE-p1. The location of where the failure varies, but the program that causes the failure is the same every time: as. The errors are a variety of signal 10 and signal 4. I do find an as.core file under /usr/obj, but as is stripped, so there are no debugging symbols or listing that I can provide. The strange thing is that I have been able to successfully build XFree86, KDE, and many other ports on this system. I followed the 4.x-to-5.0 upgrade directions to the letter about a month ago, and have had no major problems before this. Any help would be greatly appreciated! -- Paul A. Howes Hardware Tyan S2099GNNR motherboard Intel P4-2.4/533 Crucial 512 MB PC2100 DIMM Maxtor 20 GB hard drive. I see it is a P4, try adding options DISABLE_PSE and DISABLE_PG_G to your kernel. My 1.8G P4 do the same without them. John -- John Hay -- [EMAIL PROTECTED] / [EMAIL PROTECTED] To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-current in the body of the message To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-current in the body of the message
Re: cc: Internal error: Illegal instruction (program as)
It seems Paul A. Howes wrote: All; These two kernel options seem to have solved the problem. Builds now run smoothly and error-free. I read the comments in NOTES about these options and something clicked: I recall that most Pentium processors will only deal with 4 kB pages. Aren't 4 MB pages a feature of the Xeon processor, as it can address much more memory? If that is the case, then these options should be the default, and their inverse provided in the kernel configuration file (ENABLE_PSE ENABLE_PG_G). Just for the record but my [EMAIL PROTECTED]/533 512MB/DDR does *not* show this problem no matter how hard I beat it. -Søren To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-current in the body of the message
Re: cc: Internal error: Illegal instruction (program as)
On Mon, Feb 24, 2003 at 12:10:46PM +0100, Soeren Schmidt wrote: inverse provided in the kernel configuration file (ENABLE_PSE ENABLE_PG_G). Just for the record but my [EMAIL PROTECTED]/533 512MB/DDR does *not* show this problem no matter how hard I beat it. -Søren Try this: DON'T remove /usr/obj before doing a buildworld, just let it accumulate. It will show up someday (it's not deterministic). Even sh(1) can die during the build along with make(1) and as and gcc. My P4 never showed such behaviour if I properly remove /usr/obj before a build. Jiawei Ye -- Without the userland, the kernel is useless. --inspired by The Tao of Programming To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-current in the body of the message
Re: cc: Internal error: Illegal instruction (program as)
It seems leafy wrote: Try this: DON'T remove /usr/obj before doing a buildworld, just let it accumulate. It will show up someday (it's not deterministic). Even sh(1) can die during the build a long with make(1) and as and gcc. My P4 never showed such behaviour if I properl y remove /usr/obj before a build. Doesn't make any difference, the only way I (so far) has been able to reproduce this is by severely overclocking the CPU and RAM... -Søren To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-current in the body of the message
RE: cc: Internal error: Illegal instruction (program as)
Jiawei, I always remove the contents of /usr/obj prior to performing a buildworld, and I could reproduce this behavior every time. -- Paul A. Howes -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of leafy Sent: Monday, February 24, 2003 7:08 AM To: [EMAIL PROTECTED]; [EMAIL PROTECTED] Subject: Re: cc: Internal error: Illegal instruction (program as) On Mon, Feb 24, 2003 at 12:10:46PM +0100, Soeren Schmidt wrote: inverse provided in the kernel configuration file (ENABLE_PSE ENABLE_PG_G). Just for the record but my [EMAIL PROTECTED]/533 512MB/DDR does *not* show this problem no matter how hard I beat it. -S?en Try this: DON'T remove /usr/obj before doing a buildworld, just let it accumulate. It will show up someday (it's not deterministic). Even sh(1) can die during the build along with make(1) and as and gcc. My P4 never showed such behaviour if I properly remove /usr/obj before a build. Jiawei Ye -- Without the userland, the kernel is useless. --inspired by The Tao of Programming To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-current in the body of the message To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-current in the body of the message
RE: cc: Internal error: Illegal instruction (program as)
Soren, I have a machine that uses a SiS 648, but it has other problems... I'm one of the unlucky individuals who bought the ASUS implementation of this board, and ended up with a flaky P.O.S. I'm seriously thinking of swapping it for a Tyan S2662 (Granite Bay) motherboard, or waiting a few months for the 800 MHz FSB boards and chips to come out. My wife does a great deal of photo restoration, and the faster FSB would be a major improvement over the ASUS board. I haven't bothered to test any of my systems in an over clocked state, as the stability is more important to me, and my work, than wringing out the last MHz of performance. :) -- Paul A. Howes -Original Message- From: Soeren Schmidt [mailto:[EMAIL PROTECTED] Sent: Monday, February 24, 2003 7:15 AM To: [EMAIL PROTECTED] Subject: Re: cc: Internal error: Illegal instruction (program as) Could very well be that, this is a SiS 648 system with DDR333 RAM.. I can provoke something that looks like this problem if I overclock it severely ie run it a ~3Ghz with 233Mhz mem clock... -Søren To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-current in the body of the message
Re: cc: Internal error: Illegal instruction (program as)
It seems Paul A. Howes wrote: Soren, I have a machine that uses a SiS 648, but it has other problems... I'm one of the unlucky individuals who bought the ASUS implementation of this board, and ended up with a flaky P.O.S. Uhm the board here is and ASUS P4S8X and it works like a charm -Søren To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-current in the body of the message
Re: cc: Internal error: Illegal instruction (program as)
On Mon, Feb 24, 2003 at 12:10:46PM +0100, Soeren Schmidt [EMAIL PROTECTED] wrote: These two kernel options seem to have solved the problem. Builds now run smoothly and error-free. I read the comments in NOTES about these options and something clicked: I recall that most Pentium processors will only deal with 4 kB pages. Aren't 4 MB pages a feature of the Xeon processor, as it can address much more memory? If that is the case, then these options should be the default, and their inverse provided in the kernel configuration file (ENABLE_PSE ENABLE_PG_G). Just for the record but my [EMAIL PROTECTED]/533 512MB/DDR does *not* show this problem no matter how hard I beat it. Terry claims the problem is somewhat masked on systems using more memory and particular kernel options. Not my words and claims, but Terry's, however you can try to physically take out memory (if you have two 256MB modules) or limit it at boottime (into which I don't believe, personally, but that's my uneducated faith only). Timecounter i8254 frequency 1193182 Hz Timecounter TSC frequency 1699953492 Hz CPU: Intel(R) Celeron(R) CPU 1.70GHz (1699.95-MHz 686-class CPU) Origin = GenuineIntel Id = 0xf13 Stepping = 3 Features=0x3febfbffFPU,VME,DE,PSE,TSC,MSR,PAE,MCE,CX8,APIC,SEP,MTRR,PGE,MCA,CMOV,PAT,PSE36,CLFLUSH,DTS,ACPI,MMX,FXSR,SSE,SSE2,SS,HTT,TM real memory = 132382720 (126 MB) avail memory = 123191296 (117 MB) Initializing GEOMetry subsystem Pentium Pro MTRR support enabled VESA: v3.0, 832k memory, flags:0x1, mode table:0xc043e2e0 (140) VESA: Brookdale-G Graphics Chip Accelerated VGA BIOS npx0: math processor on motherboard npx0: INT 16 interface acpi0: INTEL D845GLLY on motherboard ACPI-0625: *** Info: GPE Block0 defined as GPE0 to GPE31 Using $PIR table, 9 entries at 0xc00f3d20 -- Vallo Kallaste To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-current in the body of the message
Re: cc: Internal error: Illegal instruction (program as)
On Mon, Feb 24, 2003 at 01:21:46PM +0100, Soeren Schmidt wrote: Doesn't make any difference, the only way I (so far) has been able to reproduce this is by severely overclocking the CPU and RAM... -Søren I didn't believe it either at first. But I had kernel #54 before I first got this weird behavior. Ever since I have only kernel #0. If you do buildworld/buildkernel installworld/installkernel twice a day, you might get to it sometime around 20 days. Cheers, Jiawei Ye -- Without the userland, the kernel is useless. --inspired by The Tao of Programming To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-current in the body of the message
Re: cc: Internal error: Illegal instruction (program as)
It seems leafy wrote: Doesn't make any difference, the only way I (so far) has been able to reproduce this is by severely overclocking the CPU and RAM... -Søren I didn't believe it either at first. But I had kernel #54 before I first got thi s weird behavior. Ever since I have only kernel #0. If you do buildworld/buildkernel installworld/installkernel twice a day, you mig ht get to it sometime around 20 days. Well the box has been doing about 50 buildworlds aday in a loop for the last 4-5 days, I guess that should do it no ? :) -Søren To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-current in the body of the message
Re: cc: Internal error: Illegal instruction (program as)
Paul A. Howes wrote: These two kernel options seem to have solved the problem. Builds now run smoothly and error-free. I read the comments in NOTES about these options and something clicked: I recall that most Pentium processors will only deal with 4 kB pages. Aren't 4 MB pages a feature of the Xeon processor, as it can address much more memory? No. The feature is existance-tested before it is used. These processors have a bug when dealing with certain operations relating to interactions of 4K and 4M pages. If that is the case, then these options should be the default, and their inverse provided in the kernel configuration file (ENABLE_PSE ENABLE_PG_G). You will lose around 15% performance from not having them enabled. -- Terry To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-current in the body of the message
Re: cc: Internal error: Illegal instruction (program as)
On Mon, Feb 24, 2003 at 05:15:07PM +0100, Soeren Schmidt wrote: Well the box has been doing about 50 buildworlds aday in a loop for the last 4-5 days, I guess that should do it no ? :) -Søren It should :) But I still fail to see why it does or doesn't behave this way. :( Jiawei -- Without the userland, the kernel is useless. --inspired by The Tao of Programming To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-current in the body of the message
Re: cc: Internal error: Illegal instruction (program as)
leafy wrote: On Mon, Feb 24, 2003 at 12:10:46PM +0100, Soeren Schmidt wrote: inverse provided in the kernel configuration file (ENABLE_PSE ENABLE_PG_G). Just for the record but my [EMAIL PROTECTED]/533 512MB/DDR does *not* show this problem no matter how hard I beat it. Try this: DON'T remove /usr/obj before doing a buildworld, just let it accumulate. It will show up someday (it's not deterministic). Even sh(1) can die during the build along with make(1) and as and gcc. My P4 never showed such behaviour if I properly remove /usr/obj before a build. Also make sure: 1) You have the same amount of physical RAM 2) You are using the same kernel config -- Terry To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-current in the body of the message
Re: cc: Internal error: Illegal instruction (program as)
Soeren Schmidt wrote: It seems Paul A. Howes wrote: I have a machine that uses a SiS 648, but it has other problems... I'm one of the unlucky individuals who bought the ASUS implementation of this board, and ended up with a flaky P.O.S. Uhm the board here is and ASUS P4S8X and it works like a charm You guys aren't listening again. It's a CPU implementation bug. -- Terry To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-current in the body of the message
Re: cc: Internal error: Illegal instruction (program as)
It seems Terry Lambert wrote: Soeren Schmidt wrote: It seems Paul A. Howes wrote: I have a machine that uses a SiS 648, but it has other problems... I'm one of the unlucky individuals who bought the ASUS implementation of this board, and ended up with a flaky P.O.S. Uhm the board here is and ASUS P4S8X and it works like a charm You guys aren't listening again. It's a CPU implementation bug. Well, and you aren't telling again :( So, if you have the info you claim to have, it should be very easy to tell us exactly what conditions make this error surface, so we can test and find out if this is for real once and for all.. If not, well I dont need to tell you what to do then do I ? -Søren To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-current in the body of the message
Re: cc: Internal error: Illegal instruction (program as)
Soeren Schmidt wrote: You guys aren't listening again. It's a CPU implementation bug. Well, and you aren't telling again :( I sent you private email. If you are willing to agree to non-disclosure, you can know, too. Also again So, if you have the info you claim to have, it should be very easy to tell us exactly what conditions make this error surface, so we can test and find out if this is for real once and for all.. I have already provided a description, under non-disclosure, to Bosko, and he has already created a patch that works around the problem without disclosing what the problem is that it's working around, or why the workaround works, which is satisfactory to me. I have no problem with people having to crib code from FreeBSD, and use it in a particular way, in order to get the workaround. -- Terry To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-current in the body of the message
Re: [Re: cc: Internal error: Illegal instruction (program as)
On Sat, Feb 22, 2003 at 11:18:42PM -0800, Terry Lambert [EMAIL PROTECTED] wrote: I thought this was the default in 5.x GENERIC; has someone turned these options off in the default config?!? I certainly haven't seen changes to locore.s, pmap.c, and machdep.c that would fix the problem by working around the CPU bug. Can't say anything for Paul's case, he probably had custom kernel then? If I can remember peter did the options conditional for CPU type, so only P4 owners get'em at boot time. Check the Feb 12 commit logs. -- Vallo Kallaste To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-current in the body of the message
Re: [Re: cc: Internal error: Illegal instruction (program as)
Vallo Kallaste wrote: On Sat, Feb 22, 2003 at 11:18:42PM -0800, Terry Lambert [EMAIL PROTECTED] wrote: I thought this was the default in 5.x GENERIC; has someone turned these options off in the default config?!? I certainly haven't seen changes to locore.s, pmap.c, and machdep.c that would fix the problem by working around the CPU bug. Can't say anything for Paul's case, he probably had custom kernel then? If I can remember peter did the options conditional for CPU type, so only P4 owners get'em at boot time. Check the Feb 12 commit logs. That's too bad, since it's not a P4 specific problem. AMD and other Intel processors will also have the same problem (AMD seems to have lifted the logic circuit designs directly). The problem has definitely been around since the P3 days (the P3 is where I saw it first). The only thing that's going to be a factor is amount of memory in the system, and certain tuning parameters, in which case the workaround is accidental. -- Terry To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-current in the body of the message
Re: [Re: cc: Internal error: Illegal instruction (program as)
I thought this was the default in 5.x GENERIC; has someone turned these options off in the default config?!? I certainly haven't seen changes to locore.s, pmap.c, and machdep.c that would fix the problem by working around the CPU bug. Can't say anything for Paul's case, he probably had custom kernel then? If I can remember peter did the options conditional for CPU type, so only P4 owners get'em at boot time. Check the Feb 12 commit logs. He did add code to do it, but disabled it again in the next commit. See rev 1.386 and 1.387 of sys/i386/i386/pmap.c. Note the _nots at the end of the #ifdefs. John -- John Hay -- [EMAIL PROTECTED] / [EMAIL PROTECTED] To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-current in the body of the message
Re: cc: Internal error: Illegal instruction (program as)
On Sat, Feb 22, 2003 at 11:43:01AM -0500, Paul A. Howes wrote: All, I am receiving some strange errors during a buildworld of 5.0-RELEASE-p2 from 5.0-RELEASE-p1. The location of where the failure varies, but the program that causes the failure is the same every time: as. The errors are a variety of signal 10 and signal 4. I do find an as.core file under /usr/obj, but as is stripped, so there are no debugging symbols or listing that I can provide. The strange thing is that I have been able to successfully build XFree86, KDE, and many other ports on this system. I followed the 4.x-to-5.0 upgrade directions to the letter about a month ago, and have had no major problems before this. Any help would be greatly appreciated! -- Paul A. Howes Hardware Tyan S2099GNNR motherboard Intel P4-2.4/533 Crucial 512 MB PC2100 DIMM Maxtor 20 GB hard drive. I see it is a P4, try adding options DISABLE_PSE and DISABLE_PG_G to your kernel. My 1.8G P4 do the same without them. John -- John Hay -- [EMAIL PROTECTED] / [EMAIL PROTECTED] To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-current in the body of the message
[Re: cc: Internal error: Illegal instruction (program as)
On Sat, Feb 22, 2003 at 11:43:01AM -0500, Paul A. Howes [EMAIL PROTECTED] wrote: I am receiving some strange errors during a buildworld of 5.0-RELEASE-p2 from 5.0-RELEASE-p1. The location of where the failure varies, but the program that causes the failure is the same every time: as. The errors are a variety of signal 10 and signal 4. I do find an as.core file under /usr/obj, but as is stripped, so there are no debugging symbols or listing that I can provide. The strange thing is that I have been able to successfully build XFree86, KDE, and many other ports on this system. I followed the 4.x-to-5.0 upgrade directions to the letter about a month ago, and have had no major problems before this. Any help would be greatly appreciated! As John Hay suggested, add DISABLE_PSE and DISABLE_PG_G to your kernel configuration and rebuild/install kernel. I was having _exactly_ same behaviour; at the beginning of test runs to narrow the problem a bit I did _large_ ports builds, which ran for 1,5 days.. flawlessly, as you had seen. Then changed test method to parallel (make -j4) buildworld and the problem occasionally appeared from nowhere again. The flags mentioned before will work, as I haven't had any problems after enabling them (months of time now). -- Vallo Kallaste To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-current in the body of the message
Re: [Re: cc: Internal error: Illegal instruction (program as)
Vallo Kallaste wrote: On Sat, Feb 22, 2003 at 11:43:01AM -0500, Paul A. Howes [EMAIL PROTECTED] wrote: I am receiving some strange errors during a buildworld of 5.0-RELEASE-p2 from 5.0-RELEASE-p1. The location of where the failure varies, but the program that causes the failure is the same every time: as. The errors are a variety of signal 10 and signal 4. I do find an as.core file under /usr/obj, but as is stripped, so there are no debugging symbols or listing that I can provide. The strange thing is that I have been able to successfully build XFree86, KDE, and many other ports on this system. I followed the 4.x-to-5.0 upgrade directions to the letter about a month ago, and have had no major problems before this. Any help would be greatly appreciated! As John Hay suggested, add DISABLE_PSE and DISABLE_PG_G to your kernel configuration and rebuild/install kernel. I was having _exactly_ same behaviour; at the beginning of test runs to narrow the problem a bit I did _large_ ports builds, which ran for 1,5 days.. flawlessly, as you had seen. Then changed test method to parallel (make -j4) buildworld and the problem occasionally appeared from nowhere again. The flags mentioned before will work, as I haven't had any problems after enabling them (months of time now). I thought this was the default in 5.x GENERIC; has someone turned these options off in the default config?!? I certainly haven't seen changes to locore.s, pmap.c, and machdep.c that would fix the problem by working around the CPU bug. -- Terry To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-current in the body of the message
