Skip navigation

Tag Archives: via

I previously posted about a problem with my Sylvania G netbook (original model GNET13001 with a VIA C7-M CPU, not the Meso) that I recently upgraded to a 32GB KingSpec 1.8″ ZIF SSD, where it would boot Windows XP just fine, but a Windows 7 installation on the same hard drive would do something extremely unusual: the system would not only fail to boot Windows 7, but the BIOS would not even load and execute the MBR at all, skipping the hard drive ENTIRELY in the boot order.

I figured this out late last night, powered by coffee, Doritos, and forum posts to distract me periodically. The major difference between XP and Windows 7 is the partitioning and MBR code, and the Windows 7 MBR code boots up Windows XP just fine, leaving partition tables as the most likely culprit. I modified the MBR assembly code to output a message indicating that the MBR was loaded and executed, and then halt the system. This way I would be certain that the MBR was executing at all and wouldn’t wonder if the code was somehow tripping over something that caused it to kick right back out. With a partition spanning the full disk, my message showed up. That was expected; XP worked fine, and the partition table was set up in the same fashion that it was under XP. Then, to test and see if we had some sort of issue relating to the cylinder/head/sector geometry values in the raw partition table data, I tweaked the values in the table to be slightly different, and saw no change. Then, to see if we were dealing with a problem with the starting sector for partition 1 being 2048 under Windows 7 instead of 63 under XP, I gradually bumped the partition entry up; first to 64, then 100, 200, 400, 1000, 2000, 2047, and 2048. None of these partition starting sectors caused the boot process to fail. Then, I started trimming the ending of the partition. From 31.6GB to 15GB, then to 4GB. No change, it still tried to boot.

I changed the ending cylinder from near 1000 all the way down to 100 and it stopped booting. YES!

From there, it was simply some intelligent logical bisection of the numbers: 200 failed, but 400 worked. Since 256 is 2^8 (powers of two being very significant) I tried 250, 255, and 256, which also failed. A greater jump straight to 300 passed, though; it was at that point that I realized I was dancing around another important number in terms of computer memory: 2,097,152 KB (4,194,304 512-byte sectors) which is also a power of 2, and happened to be right between the sector counts of 256 and 300 cylinders (these equations are in C/H/S format, and I’m ignoring the first 2048 or so skipped sectors just to annoy you):

256*255*63 = 4,112,640 sectors
300*255*63 = 4,819,500 sectors

With this new information, I changed over to LBA (sector) entry instead, and made the TOTAL number of sectors in this partition 4,194,303. It failed to boot.

When I set the total number of sectors in the first partition to 4,194,304 sectors (exactly 2GB), the Sylvania G loaded and executed the MBR normally, where just one sector less would cause the entire hard drive to be skipped in the boot order.

This effectively means that a stock Windows 7 can never boot on this system without major trickery. I consider this a very severe and stupid BIOS bug. What I don’t understand is how it is possible that this check even exists in the first place. (Edit 2017-03-12: I realized it could be for skipping a small “utility partition” like older laptops do, but that’s what the partition table ACTIVE flag is used for already.) The computer would have to read the MBR from the hard drive into memory and intentionally check the first partition’s size for a minimum amount to make this bug happen, so someone either at VIA, Phoenix (it’s a Phoenix BIOS), or Sylvania (Digital Gadgets, Inc.) had to go out of their way to put this bug into place. Worse yet, there are absolutely no BIOS updates in existence for the Sylvania G, and never will be, so this bug is permanent. Fortunately, if your first partition is larger than 2GB in size, you can blissfully ignore this problem, which means that Linux installations using a compatible partitioning scheme, Windows XP, and Windows Vista all should run on the Sylvania G without hitting this problem at all. Windows 7 is effectively not an option, though. In the next few days, I plan to attempt to overcome this bug and get Windows 7 working on the Sylvania G anyway, as a personal challenge if nothing else.

I am left to ponder what other systems might be having this same issue. The BIOS has no business whatsoever attempting to “verify” my partition table of choice, in my opinion, and I want to know at what point and for what purpose this bug/misfeature was introduced. If anyone has a system that seems to skip the hard drive in the boot order with Windows 7, when Windows Vista or especially Windows XP would work just fine, please leave a comment below with your experiences and insights. I’d love to engage in a discussion with anyone else who feels they may have hit this type of bug.

UPDATE: I have verified that the problem boils down to one single check against the LBA “number of sectors” in partition 1 on the drive being equal to 0x400000 (4194304) or higher, and nothing more. I was able to get Windows 7 to boot normally by copying everything from partition 1 into partition 2, deleting partition 1, then using a hex editor to change byte 0x1CC to 0x40. Since all other parts of partition 1’s table entry are zeroed out, it isn’t seen as a valid partition on the disk, but the BIOS is only checking the DWORD value at 0x1CA-0x1CD to be 0x400000 or greater, so it simultaneously bypasses the bug/misfeature in the BIOS and doesn’t confuse the MBR nor Windows 7 at all, as the partition structure is completely valid. If I was adventurous, I’d shuffle the partition down to sector 2048 and reclaim the missing 100MB, but since I’d lose that space under Windows 7’s out-of-the-box partitioning scheme anyway, and it’s 0.1GB of a 31.6GB SSD, that kind of capacity loss isn’t significant at all.

Since the workaround is as simple as using a dummy entry where partition 1 is supposed to be, this trick should work for any operating system. Just start partitions at number 2 instead of 1, and apply the change 1 with a hex editor when you’re done, and you can have a tiny “first” partition without consequence! The only time you’ll ever need to redo the hack is if you run something that rewrites that portion of the partition table.

This excerpt is from the Tritech Computer Solutions page called Sylvania G Netbook Tips and Tricks.

Thanks to a helpful developer at the ZSNES forum, we found the solution to the VIA C7 platform speed/timing problems with certain programs.

Edit C:\BOOT.INI and add the /usepmtimer switch to the boot command line for Windows XP. This uses a timer that is stable even when in power management modes, which means that the throttling of the C7 CPU and accompanying chipset do not affect its timing characteristics.

Some applications like ZSNES rely on a very precise and stable timing mechanism (in the case of ZSNES, speed regulation depends on QueryPerformanceCounter) and the normal timer on this platform is unstable in certain power management states. Also, we used WCPUID’s real-time clock display to figure out that the System control panel (sysdm.cpl) reports the frequency wrong on the C7 and that it is in fact running at 1200 MHz when under a full load, despite Windows’ insistence that somehow the CPU is going as low as 198 MHz when the lowest ACPI P-state is 400 MHz.

If you have games on this netbook and are experiencing strange timing jerks or other glitches, you need this simple fix.

UPDATE: I found a solution.

I changed my Sylvania G (original, non-Meso) netbook to Windows XP/Linux dual-boot to test some software I’m working on, and discovered that while Windows XP certainly does boot and run in general on the G, some kind of system timer or timing loop is severely out of whack!  I wanted to use my little G as a portable gaming machine from the Windows XP install, and to my horror, ZSNES couldn’t decide what speed it wanted to run!  Now, I’ve never had a single issue with ZSNES on any computer I’ve ever tried it on, even preferring the Windows port of it over the Linux native one, and not once has a problem existed with ZSNES that I couldn’t find an easy fix for, until now.

I’ve been researching the matter and gathering evidence, and I may have a potential answer to the problem.  Linux requires activation of the VIA C7 Enhanced PowerSaver module e_powersaver to clock the VIA C7 CPU properly between 400 and 1200 MHz; apparently the default speed of the CPU is only 600 MHz instead of 1200 MHz, because Linux installs without e_powersaver and Windows XP report a ~600 MHz processor where a 1.2 GHz one exists.  Here’s the extremely weird part, though: if I check the System control panel shortly after bootup and read the clock speed, sometimes it registers a clock speed of 198 MHz (about 200 MHz) which isn’t even one of the ACPI P-states for the VIA C7-M 1.2 processor.

I’ve unlocked the Windows HAL options (I’ll post how to do that at another time) and switched between ACPI Multiprocessor PC (the default for the image I used) and ACPI Uniprocessor PC and MPS Uniprocessor PC, all of which use the local APIC for IRQ routing but the MPS variant of which doesn’t theoretically touch ACPI.  Nothing seems to have helped.  I have two working theories as to what’s going on here, and how it might be fixed:

  1. A calibration loop in Windows a la BogoMIPS in Linux is being screwed up by the VIA C7, or
  2. The VIA C7’s PowerSaver feature is ignored or incorrectly used by Windows (via generic ACPI P-states) and it’s throwing off some kind of timer that ZSNES relies on for proper emulation of the 65816 CPU and SPC audio processor.

So far, I haven’t found a solution to this problem, and Sylvania’s site is extremely unhelpful, with only Windows drivers and a new version of gOS, but no BIOS updates or further information.  I’m looking into the technical stuff on the VIA c7 now, and it looks like the solution (assuming Windows isn’t doing something sinister) lies in clever manipulation of the C7 model-specific registers (MSRs) that control the processor’s power state.  If ZSNES is mis-calibrating some kind of tight internal timing loop because of some kind of CPU clocking issue, then tweaking the MSRs may be the solution to the problem.  Unfortunately, I’m no Windows developer, so I’m not certain how I should approach the problem.  I don’t think it’s isolated to ZSNES either, but I don’t recall what I saw that justifies that belief.  In any case, I’m working on it.  It’s just one of many pesky projects I’m hitting my head against at the moment.  We’re still working on that remote access software package; in fact, someone found our site and called us, and I had to sort of turn her away.  It’s all a bit behind schedule, and there’s not really much I can do to make things proceed any more quickly.  Stay tuned…

%d bloggers like this: