bhoggett wrote:
It does seem to be limited to Mandrake and a System Rescue CD at the moment (but that could also affect other distros - no reported incidents don't mean a clean bill of health). It does seem to be a problem with the LG drives, as a firmware upgrade is said to remove the threat. I gather the affected distros are also addressing their end of the problem.
No reported incidents, plus Mandrake's specification of the -mdk tree (they're taking the blame), does suggest this particular nugget didn't enter the mainline (Linus's or Marcelo's trees), and between that and the publicity, one can hope and assume it won't spread further.
Anyone else is free to make the same 'mistake,' of course. It's an easy one to make; you just write perfectly valid code, following the same popular and generally-accepted specifications as the rest of the world adheres to, then find out a million units in the field were designed to blow up when presented with it.
Even MS could have the same problem, though their codependency with the hardware companies gives them better chances of catching blatant issues before releasing.
(Even then, Windows 98 wouldn't install properly* on a rather large base of Via-chipset-based boards. That was one of the issues that forced 98SE to market, though they used the opportunity to integrate the new Media Player and other monopoly features.)
Hammer said,
My LG DVD/CD-RW combo drive works with Red Hat Linux 9.0 setup…
The CD-Rs are always going to be unaffected, because they have to support the offending command just to function.
From the erratum, which could be worded better:
"The specification does not require an implementation of the FLUSH_CACHE command in the driver, and returning an error (or doing nothing) would have been the correct behaviour for the drive. Likewise, reusing a command is against the specification and LG has reused the FLUSH_CACHE command to modify the firmware of the drive, but they are unwilling to disclose exactly what the command does.
This FLUSH_CACHE command is supposed to be supported only by CD-RW or DVD-RW devices; the LG-based CD-ROM devices are understanding this command as the UPLOAD_FIRMWARE command."
So the code in the Mandrake kernel took what should be a safe shortcut - sends the FLUSH_CACHE command without checking the drive type, expects an ignore/rejection at worst, per the spec; LG designed their CD-ROM readers to go into self-destruct mode instead. It's like someone sold you an Amiga where the 'dir' command formats the hard drive.
Mandrake's changing their code to avoid any unnecessary destruction, while LG have probably been embarassed into compliance on new units/through fixes for the installed ones.
What makes it a mess is that you can't resuscitate a drive yourself *after* the firmware's been wiped.
*
Since someone's going to call me on this; you could obviously get it installed with enough struggle, and OEMs didn't have so much problem, because they only had to create one working disk image to blow onto all affected machines. Original '98 didn't support the Via chips completely/outright, and of course it wouldn't fall back to BIOS-grade access automatically, so you'd run the installer, boot, watch your system freeze - at least, if it was configured like mine - then have to kick it into Safe Mode yourself and install Via's drivers. Not quite as destructive as this fiasco, but another example of 'unforseen hardware problems.' In that case, MS made that assumption that no hardware would exist to supplant the current state of the market (a MS-grade assumption, if there ever was); they've since learned a thing or two, and XP takes a slightly more stable approach in those domains, while MS is extra-sure to twist the arms of manufacturers.
