If my domain loans a page out and then reduces its reservation, is it safe to assume that when that page appears in the rx ring as a result that it will be one of the ones I loaned out? in other words, do pages I return to Xen for use in the rx ring get replaced by other pages, or do they come back to me when they are full of data? thanks ron ------------------------------------------------------- This SF.Net email is sponsored by BEA Weblogic Workshop FREE Java Enterprise J2EE developer tools! Get your free copy of BEA WebLogic Workshop 8.1 today. http://ads.osdn.com/?ad_id=5047&alloc_id=10808&op=click _______________________________________________ Xen-devel mailing list Xen-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/xen-devel
The frames you get from the receive ring are not necessarily the ones you removed from your memory reservation. i.e., you will be replacing with different pages. -- Keir PS. Try to stick close to the Linxu device drivers as much as possible -- the mechanism for granting foreign access to your pages, and for accepting page transfers from other foreign domains, will be changing soon (it''s rather fast and loose at the moment). It''ll be nice if you can take a patch directly from our Linux tree.> If my domain loans a page out and then reduces its reservation, is it safe > to assume that when that page appears in the rx ring as a result that it > will be one of the ones I loaned out? in other words, do pages I return to > Xen for use in the rx ring get replaced by other pages, or do they come > back to me when they are full of data? > > thanks > > ron > > > > ------------------------------------------------------- > This SF.Net email is sponsored by BEA Weblogic Workshop > FREE Java Enterprise J2EE developer tools! > Get your free copy of BEA WebLogic Workshop 8.1 today. > http://ads.osdn.com/?ad_id=5047&alloc_id=10808&op=click > _______________________________________________ > Xen-devel mailing list > Xen-devel@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/xen-devel------------------------------------------------------- This SF.Net email is sponsored by BEA Weblogic Workshop FREE Java Enterprise J2EE developer tools! Get your free copy of BEA WebLogic Workshop 8.1 today. http://ads.osdn.com/?ad_id=5047&alloc_id=10808&op=click _______________________________________________ Xen-devel mailing list Xen-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/xen-devel
> > If my domain loans a page out and then reduces its reservation, is it safe > to assume that when that page appears in the rx ring as a result that it > will be one of the ones I loaned out? in other words, do pages I return to > Xen for use in the rx ring get replaced by other pages, or do they come > back to me when they are full of data?Nope, that wouldn''t be a safe assumption. All the pages given up by different domains go into the driver domain''s free buffer pool that are fed to the hardware. When a packet is received by the Ethernet NIC, it will be DMAed into whatever buffer happens to be the head of the queue. Once the driver domain figures out which domain this packet should be delivered to, it reassigns the page to that domain, so its then able to map the page into its address space. In a world with smarter NICs, we could give each domain its own receive buffer pool and have the NIC do the demultiplex direct into the domain''s buffers. For the moment, the current scheme is the best we can do with ''stock'' hardware. [1] Aside: some researchers have previously claimed that ''page flipping'' is slow as it involves TLB shootdowns on SMP systems. Xen uses a few neat tricks such to avoid the need for shootdowns in the vast majority of cases. It''s not slow at all. Really. Ian [1] Keir and I built such as smart NIC a few years ago. See: http://www.cl.cam.ac.uk/~iap10/gige.ps ------------------------------------------------------- This SF.Net email is sponsored by BEA Weblogic Workshop FREE Java Enterprise J2EE developer tools! Get your free copy of BEA WebLogic Workshop 8.1 today. http://ads.osdn.com/?ad_id=5047&alloc_id=10808&op=click _______________________________________________ Xen-devel mailing list Xen-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/xen-devel
On Wed, 1 Sep 2004, Keir Fraser wrote:> > The frames you get from the receive ring are not necessarily the ones > you removed from your memory reservation. i.e., you will be replacing > with different pages.Something else I''m unclear on then: what if you have a pfn array that is of size x, covering a range of physical frame numbers, and you get back a pfn that has an index such that it won''t fit in that array? I''m a little unclear on how this would work. It looks in the newer stuff like you don''t really resize the array, unless I missed it. I guess I''ll look again, but does the linux loader size the mfn array in all cases so that it has enough entries for all of physical memory, so that it can accomodate any PFN? Anyway, back to board of drawings.> PS. Try to stick close to the Linxu device drivers as much as possible > -- the mechanism for granting foreign access to your pages, and for > accepting page transfers from other foreign domains, will be changing > soon (it''s rather fast and loose at the moment). It''ll be nice if you > can take a patch directly from our Linux tree.yeah, I''m staying as close as I can. Moving .h files over is a script. .c can not be a script for many reasons, and I''ll tell you that linux and bsd drivers are way more similar to each other than they are to Plan 9 drivers, but I can track the xen stuff anyway. ron ------------------------------------------------------- This SF.Net email is sponsored by BEA Weblogic Workshop FREE Java Enterprise J2EE developer tools! Get your free copy of BEA WebLogic Workshop 8.1 today. http://ads.osdn.com/?ad_id=5047&alloc_id=10808&op=click _______________________________________________ Xen-devel mailing list Xen-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/xen-devel
> On Wed, 1 Sep 2004, Keir Fraser wrote: > > > > > The frames you get from the receive ring are not necessarily the ones > > you removed from your memory reservation. i.e., you will be replacing > > with different pages. > > Something else I''m unclear on then: what if you have a pfn array that is > of size x, covering a range of physical frame numbers, and you get back a > pfn that has an index such that it won''t fit in that array? I''m a little > unclear on how this would work. It looks in the newer stuff like you > don''t really resize the array, unless I missed it. I guess I''ll look > again, but does the linux loader size the mfn array in all cases so that > it has enough entries for all of physical memory, so that it can > accomodate any PFN?When you receive a packet you get given the MFN of the page its contained in. It''s purely a matter for the guestos where it chooses to insert it into it''s PFN table. Typically, it will insert it into a slot that was vacated when the guestos passed a free receive buffer down to the driver domain. If you look at the linux code, you''ll see that the data page associated with the skb gets passed down as the free buffer. When a packet is received, the driver hooks in the new page to take the place of the original data page. Ian ------------------------------------------------------- This SF.Net email is sponsored by BEA Weblogic Workshop FREE Java Enterprise J2EE developer tools! Get your free copy of BEA WebLogic Workshop 8.1 today. http://ads.osdn.com/?ad_id=5047&alloc_id=10808&op=click _______________________________________________ Xen-devel mailing list Xen-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/xen-devel
On Wed, 1 Sep 2004, Ian Pratt wrote:> When you receive a packet you get given the MFN of the page its > contained in. It''s purely a matter for the guestos where it chooses to > insert it into it''s PFN table. Typically, it will insert it into a slot > that was vacated when the guestos passed a free receive buffer down to > the driver domain. If you look at the linux code, you''ll see that the > data page associated with the skb gets passed down as the free buffer. > When a packet is received, the driver hooks in the new page to take the > place of the original data page.right, right, right, what was I thinking? It''s obvious in the code. sorry for that silly question. I am still hoping to save the plan 9 kernel physical/kernel virtual scheme, but it may have to die and follow the xenolinux physical to machine and machine to physical mapping strategy. That would certainly simplify a few other things however, namely presenting a single physically contiguous address space to the kernel, instead of my current physically discontiguous space. We''ll see. thanks ron ------------------------------------------------------- This SF.Net email is sponsored by BEA Weblogic Workshop FREE Java Enterprise J2EE developer tools! Get your free copy of BEA WebLogic Workshop 8.1 today. http://ads.osdn.com/?ad_id=5047&alloc_id=10808&op=click _______________________________________________ Xen-devel mailing list Xen-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/xen-devel
Geez, I gotta learn u-diff with bk. Sorry.
python/xen/xend/XendDomainInfo.py
change this:
# Register image handlers.
add_image_handler(''linux'', vm_image_linux)
To this;
# Register image handlers.
add_image_handler(''linux'', vm_image_linux)
add_image_handler(''plan9'', vm_image_plan9)
-------------------------------------------------------
SF email is sponsored by - The IT Product Guide
Read honest & candid reviews on hundreds of IT Products from real users.
Discover which products truly live up to the hype. Start reading now.
http://productguide.itmanagersjournal.com/
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/xen-devel
Forgot to delete a CC: ron ------------------------------------------------------- SF email is sponsored by - The IT Product Guide Read honest & candid reviews on hundreds of IT Products from real users. Discover which products truly live up to the hype. Start reading now. http://productguide.itmanagersjournal.com/ _______________________________________________ Xen-devel mailing list Xen-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/xen-devel
ok, as you might have guessed, I have the ok to release the plan 9 stuff for xen and have sent the code to ian and keir. My bk ignorance meant it was sent as files, not u-diffs, but I am leaning on the mercy of the court as the changes are very small! thanks ron p.s. Xen is just excellent. Boot time for Plan 9 is < 10 seconds. Vmware was, well, somewhat longer. ------------------------------------------------------- SF email is sponsored by - The IT Product Guide Read honest & candid reviews on hundreds of IT Products from real users. Discover which products truly live up to the hype. Start reading now. http://productguide.itmanagersjournal.com/ _______________________________________________ Xen-devel mailing list Xen-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/xen-devel