Oliver Neukum wrote: > Am Donnerstag, 18. Juli 2002 23:45 schrieben Sie: > >>>is there a deeper reason for the TDs being allocated under a spinlock? >> >>The presence of an annoying hashtable needed for bus_to_virt() >>style mappings. Minimally, stuffing the hashbuckets needs to >>be reentrant, and that's done as part of allocation. > > > This is a problem for ensuring that storage makes progress. > Could allocation be divorced from hashing them ?
Please elaborate ... what's the problem? Yes, those ops could be split, if the hashtable entries were merged into the TDs. (Which would slow down access during IRQ processing on systems where "dma consistent" memory is just "uncacheable", rather than "cache-coherent".) - Dave ------------------------------------------------------- This sf.net email is sponsored by:ThinkGeek Welcome to geek heaven. http://thinkgeek.com/sf _______________________________________________ [EMAIL PROTECTED] To unsubscribe, use the last form field at: https://lists.sourceforge.net/lists/listinfo/linux-usb-devel
