On Fri, 12 Dec 2008 21:24:11 +0100, Andi Kleen <andi@xxxxxxxxxxxxxx> wrote: > > In the current interface, each data item is copied twice: one is to > > the allocated memory from user space (via copy_from_user), and another > > For such large copies it is better to use multiple smaller (e.g. 4K) > copy user, that gives better real time preempt latencies. Each cfu has a > cond_resched(), but only one, not multiple times in the inner loop. For the function in question, the buffer memory can be divided into a smaller size (at least to 4K bytes) since the buffer is repeatedly used for small objects, where the copy_from_user (and a copy_to_user) is only once in each cycle. So, just reducing the allocation size of the buffer seems good; it is likely able to avoid both large preempt latencies and large memory allocation, which also can leave off the use of vmalloc. > > is to on-memory structures or to buffers/pages from the allocated > > memory. > > It depends how performance critical it is. > > One way for example is to grab the user pages using get_user_pages() > and then reference those pages directly using kmap(). > But you would be at the mercy of the user process not modifying in > parallel then. Normally it is safer to work from copies in kernel > space to avoid races. As long as it doesn't happen too often a few > copies are also usually not a problem. I wouldn't worry about them > unless you see them prominently in profiler logs. > > -Andi I got it. If need arises, then I'll recall get_user_pages(). At present, there is likely no need to do like that. Thank you for the informative advises. With regards, Ryusuke -- To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
