> int ret = usb_get_descriptor(dev, USB_DT_DEVICE, 0, &dev->descriptor, > sizeof(dev->descriptor)); > > I'm not convinced that such a call is safe in all circumstances. > dev->descriptor is smaller than a cache line and is not cache line > aligned. When pci_map_single() finally gets called with address it > will invalidate the whole cache line, which could cause problems if > adjacent members of dev are accessed.
Assuming this cacheline sharing issue is real: - if pci_map_single() doesn't flush dirty initial/terminal cache lines before invalidating them, that'd be a clear bug in that implementation: one is supposed to flush dirty data before discarding cache entries. - that specific call is made during enumeration, when no other concurrent modification of that device should be allowed In short, such uses can be safe even given awkward cache models. > Is this reasoning correct or do you think it's better to leave that > code as-is? Leave as-is, I think, unless there's a clear case where the device is being modified during enumeration by something other than khubd (which is likely a bug), causing a problem. - Dave _______________________________________________________________ Don't miss the 2002 Sprint PCS Application Developer's Conference August 25-28 in Las Vegas - http://devcon.sprintpcs.com/adp/index.cfm?source=osdntextlink _______________________________________________ [EMAIL PROTECTED] To unsubscribe, use the last form field at: https://lists.sourceforge.net/lists/listinfo/linux-usb-devel
