begin quoting Gabriel Sechan as of Tue, Jun 20, 2006 at 01:11:21PM -0500: > >From: Stewart Stremler <[EMAIL PROTECTED]> [snip] > >Send a packet, scan the packet, processor detects the pattern, and > >boom! The blight is bootstrapping into the transcend again. > > Come on Stewart, you're better than that. How does the CPU scan an > incoming packet? A CPU doesn't know wtf a networking card is. All it
How do you know that a CPU doesn't know WTF a networking card is? If Intel makes the CPU and the chipset -- or you get a CPU with ethernet on the main die -- it /could/ know. But you're right in the sense that it doesn't /have/ to know. All it has to watch for is a sequence of bytes. Ethernet, or MyriNET, or serial, or USB, etc. etc. --- just some way to feed a sequence of bytes in to the CPU. > knows is that it recieved an interrupt from a given device, then read and > wrote a few values to a few ports. ....and there's your input stream. Once you have an input stream you can put a recognizer on it; once you have an input stream and a recognizer, you have a trigger. > The most it could do is continuously > scan RAM for the pattern- which would be noticable seconds after ti > started as performance went to 0 and all spare RAM bandwidth was sucked up > by the CPU. Now it's my turn... Come on, Gabe, you're better than that. You wouldn't dedicate the processor to scanning memory; you'd put a recognizer in the cache and scan memory as it's loaded into the CPU, and every so often you'd generate a memory access to a semi-random page. If you steal every 32nd memory access, who would notice? Look at the trouble Gus was having... and he's worked out what his throughput ought to be, and the reaction was, by and large, a shrug. The best place to hide is where nobody will bother to look. > We'd also have boxes randomly crashing all over the place. Oh, so that's why M$ has been training users to ignore crashes! -- _ |\_ \| -- [email protected] http://www.kernel-panic.org/cgi-bin/mailman/listinfo/kplug-list
