On Fri, Apr 04, 2008 at 12:20:46PM +0100, Andrew Doran wrote: > SSP or the stack redzone? Have you got a backtrace from it? (hand copied) usbN at pci .. EHCI .. panic: stack overflow detected: terminated breakpoint() panic() __stack_check_fail() usbd_new_device() usb_doattach() config_interrupts_thread() Without looking, that smells more like SSP to me, and (after a few more runs) it happens so reliably at this same place, also like an issue in the usb code. > The rnd code could definitley do with locking. I'll take a look. Awesome. It could do with a number of other things, too, but that's a different discussion. By the way, I finally saw the obvious and realised that one other reason I get more of the low entropy warnings is that fewer of the rnd source devices have been attached at this stage, since they're getting deferred and parallelised. That's an explanation, but also a problem; we really need some early entropy before some other things are initialised. Previously when I looked for some, mixing in tsc's when the rnd sources were attached during config was easy and early enough. I'm sure there are places we could sprinkle some rnd seeding earlier, even if they're a little more intrusive, now that those device events happen later. Initial tsc's on cpu attach, timecounter initial values, .. ? -- Dan.
Attachment:
pgpd92HaY2jo5.pgp
Description: PGP signature