On Fri, Oct 18, 2002 at 04:26:44PM +0300, Momchil Velikov wrote:
> Jan> Yes, it should. But then message queue is most appropriate since it can
> Jan> also pass the actual data along. That's what a file descriptor with
> Jan> appropriately implemented poll is.
> >> >>
> >> >> It involves copying. Theoretically it is possible to have the read
> >> >> system call avoid copying for whole overwriten pages (by exchanging
> ^^^^^
>
> >> >> page table entries (and flushing TLBs :-( )), but this may work well
> >> >> on some systems, work not so well on others and not work AT ALL when
> >> >> the source buffer is actually device memory.
> >>
> Jan> No, it won't work, because the read buffer would have to have same
> Jan> alignment as the data in cache.
> >>
> >> I do not understand this ? What do you mean ? What won't work ?
>
> Jan> You said, that read could possibly avoid copying by replacing PTEs. But
> Jan> it counldn't unless the data is appropriately aligned. That is if you
> Jan> read a whole page, starting at page-aligned offset and the buffer is
> Jan> page-aligned, then it will work, but in most other cases it won't.
> Jan> (Exact condition is: if buffer % PAGE_SIZE == offset % PAGE_SIZE, it
> Jan> could be done, otherwise not.)
>
> Every buffer larger than or equal to two page sizes contains at least
> one page aligned on page size boundary, i.e. at least one page frame.
> That page can be remapped. That's why I said "whole overwriten pages".
No, it does not. Lets say PAGE_SIZE is 1 << 12 (as on i386). That means
page-aligned addresses have last 3 hex digits equal to 0.
Let's say you issue a command:
pread(fd, buf=0x80601800, length=0x4000, offset=0);
^^^
Now, in page cache first byte in a file is a first byte on some page.
So the first page of the buffer is partial, address 0x80601800
- 0x80601fff and is to be filled with bytes 0 .. 0x7ff from the first
page in the page cache.
The second page is a complete one, address 0x80602000 - 0x80602fff.
However it's to be filled with bytes 0x800 - 0xfff from the first cache
page plus bytes 0 - 0x7ff from the second one. So they must be copied,
since they are on two different pages.
And so on.
See?
> The first and the last partial pages have to be copied, of course.
-------------------------------------------------------------------------------
Jan 'Bulb' Hudec <bulb@ucw.cz>
--
Kernelnewbies: Help each other learn about the Linux kernel.
Archive: http://mail.nl.linux.org/kernelnewbies/
FAQ: http://kernelnewbies.org/faq/
