On Mon, Nov 04, 2002 at 08:58:07AM -0500, Andrew Perrin wrote: > On Mon, 4 Nov 2002, Robert Ian Smit wrote: > > > I was surprised that this issue took down the system on Linux. > > I understand, as nate explained, that hardware errors will always > > result in trouble but I expected the kernel to react differently. > > (Or is this a limitation of x86 or the issue you mention?) > > FWIW, I'm skeptical of Nate's claim that excessive I/O errors must bring > down the system. I'm certainly not a kernel hacker, but I see no reason > why the kernel couldn't do what it does in other roughly analogous > situations: decide that the stream is bad and effectively turn it off, > either by killing the process or by redirecting the stream to /dev/null or > something like that. The whole point of a robust, threaded, multitasking > architecture is supposed to be that isolated errors *don't* bring down the > system.
I don't think nate was refering to the literal text stream of errors. I would imagine that hardware problems could get the controller chips into such a state of confusion that a power cycle is required, which would mean rebooting the entire system. -rob
msg10911/pgp00000.pgp
Description: PGP signature
