Without this change it is not possible to page-out all guest pages, then
trigger a page-in for all pages, and then page-out everything once
again. All pages in the mru list can not be paged out because they
remain active in the internal bitmap of paged pages.
Use the mru list only if the number of paged-out pages is larger than
the mru list. If the number is smaller, start to clear the mru list. In
case the number of paged-out pages drops to zero the mru list and the
internal bitmap will be empty as well.
Also add a new interface for dropped pages. If a gfn was dropped there
is no need to adjust the mru list because dropping a page is not usage
of a page.
Signed-off-by: Olaf Hering <olaf@aepfle.de>
Committed-by: Ian Jackson <ian.jackson.citrix.com>
Introduce a cmdline handling for the pager. This simplifies libxl support,
debug and mru_size are not passed via the environment anymore.
The new interface looks like this:
xenpaging [options] -f <pagefile> -d <domain_id>
options:
-d <domid> --domain=<domid> numerical domain_id of guest. This option is required.
-f <file> --pagefile=<file> pagefile to use. This option is required.
-m <max_memkb> --max_memkb=<max_memkb> maximum amount of memory to handle.
-r <num> --mru_size=<num> number of paged-in pages to keep in memory.
-d --debug enable debug output.
-h --help this output.
Signed-off-by: Olaf Hering <olaf@aepfle.de>
Committed-by: Ian Jackson <ian.jackson.citrix.com>
Subsequent patches will use xenstored to store the numbers of pages
xenpaging is suppose to page-out.
Remove num_pages and use target_pages instead.
Signed-off-by: Olaf Hering <olaf@aepfle.de>
Committed-by: Ian Jackson <ian.jackson.citrix.com>
This change reverses the task of xenpaging. Before this change a fixed number
of pages was paged out. With this change the guest will not have access to
more than the given number of pages at the same time.
Signed-off-by: Olaf Hering <olaf@aepfle.de>
Committed-by: Ian Jackson <ian.jackson.citrix.com>
In preparation of upcoming libxl integration,
move xenpaging binary from /usr/sbin/ to /usr/lib/xen/bin/
Signed-off-by: Olaf Hering <olaf@aepfle.de>
Committed-by: Ian Jackson <ian.jackson.citrix.com>
Nomination of gfns can fail, but may succeed later.
Thats the case for a guest that starts ballooned.
v2:
- print debug when clearing uncosumed happens
Signed-off-by: Olaf Hering <olaf@aepfle.de>
Committed-by: Ian Jackson <ian.jackson.citrix.com>
Remove the if/else logic to exit from the in case a signal arrives.
Update comments.
Signed-off-by: Olaf Hering <olaf@aepfle.de>
Committed-by: Ian Jackson <ian.jackson.citrix.com>
Move page resume loop into its own function.
Move page eviction loop into its own function.
Allocate all possible slots in a paging file to allow growing and
shrinking of the number of paged-out pages. Adjust other places to
iterate over all slots.
This change is required by subsequent patches.
v2:
- check if victims allocation succeeded
Signed-off-by: Olaf Hering <olaf@aepfle.de>
Committed-by: Ian Jackson <ian.jackson.citrix.com>
Remove xc_dominfo_t from paging_t, record only max_pages.
This value is used to setup internal data structures.
Signed-off-by: Olaf Hering <olaf@aepfle.de>
Committed-by: Ian Jackson <ian.jackson.citrix.com>
Move comment about xc_handle to the right place.
Allocate paging early and use calloc.
Signed-off-by: Olaf Hering <olaf@aepfle.de>
Committed-by: Ian Jackson <ian.jackson.citrix.com>
Catch lseek() errors.
Use -1 as return value and let caller read errno.
Remove const casts from buffer pointers, the page is writeable.
Use wrapper for write() which matches the read() prototype.
Remove unused stdarg.h inclusion.
Remove unused macro.
Signed-off-by: Olaf Hering <olaf@aepfle.de>
Committed-by: Ian Jackson <ian.jackson.citrix.com>
v3:
- adjust arguments for xc_mem_paging_enable() failures
v2:
- move changes to file_op() to different patch
Signed-off-by: Olaf Hering <olaf@aepfle.de>
Committed-by: Ian Jackson <ian.jackson.citrix.com>
Remove stale comment.
If a page was populated several times the vcpu is paused and
xenpaging has to unpause it again.
Signed-off-by: Olaf Hering <olaf@aepfle.de>
Committed-by: Ian Jackson <ian.jackson.citrix.com>
Create two new variables called APPEND_ and PREPEND_ to add compile
flags at the beginning or at the end of the search path.
Added a new semantic for user defined compile flags, here is the list
of possible options:
PREPEND_LIB: add libraries to the search path before xen
(before xen installation folders).
PREPEND_INCLUDES: add headers to the search path before xen
(before xen installation folders).
APPEND_LIB: add libraries to the search path at the end
(after all xen installation folders have been added).
APPEND_INCLUDES: add libraries to the search path at the end
(after all xen installation folders have been added).
EXTRA_INCLUDES and EXTRA_LIB can still be used, and they will have the
same effect as PREPEND_INCLUDES and PREPEND_LIB.
Signed-off-by: Roger Pau Monne <roger.pau@entel.upc.edu>
Committed-by: Ian Jackson <ian.jackson@eu.citrix.com>
Do munmap() on all mapped pages, not just the first one. Without this
change the gfns backing the remaining pages can not be paged out again
because the page count does not go down to 1. This change was missing
from changeset 23827:d1d6abc1db20.
Signed-off-by: Olaf Hering <olaf@aepfle.de>
Committed-by: Ian Jackson <ian.jackson@eu.citrix.com>
Disallow xenpaging in a PoD guest until coexistance between the two features
is properly implemented.
Signed-off-by: Olaf Hering <olaf@aepfle.de>
Acked-by: Tim Deegan <tim@xen.org>
Committed-by: Tim Deegan <tim@xen.org>
Evict of a nominated gfn must fail if some other process mapped the
page without checking the p2mt of that gfn first.
Add a check to cancel eviction if the page usage count is not 1.
Handle the possible eviction failure in the page-in paths.
After nominate and before evict, something may check the p2mt and call
populate. Handle this case and let the gfn enter the page-in path. The
gfn may still be connected to a mfn, so there is no need to allocate a
new page in prep.
Adjust do_mmu_update to return -ENOENT only if the gfn has entered the
page-in path and if it is not yet connected to a mfn. Otherwise
linux_privcmd_map_foreign_bulk() may loop forever.
Add MEM_EVENT_FLAG_EVICT_FAIL to inform pager that a page-in request for
a possible not-evicted page was sent. xenpaging does currently not need
that flag because failure to evict a gfn will be caught.
Signed-off-by: Olaf Hering <olaf@aepfle.de>
Acked-by: Tim Deegan <tim@xen.org>
Committed-by: Tim Deegan <tim@xen.org>
If the guest happens to read from the gfn while xenpaging is in the process of
evicting the page, the guest may read zeros instead of actual data.
Also if eviction fails the page content will be corrupted and xenpaging wont
attempt to restore the page.
Remove page scrubbing from pager and do it after successful eviction.
Signed-off-by: Olaf Hering <olaf@aepfle.de>
Acked-by: Ian Jackson <ian.jackson@eu.citrix.com>
Acked-by: Tim Deegan <tim@xen.org>
Committed-by: Tim Deegan <tim@xen.org>
Up to now a single ring buffer was used for mem_share, xenpaging and
xen-access. Each helper would have to cooperate and pull only its own
requests from the ring. Unfortunately this was not implemented. And
even if it was, it would make the whole concept fragile because a crash
or early exit of one helper would stall the others.
What happend up to now is that active xenpaging + memory_sharing would
push memsharing requests in the buffer. xenpaging is not prepared for
such requests.
This patch creates an independet ring buffer for mem_share, xenpaging
and xen-access and adds also new functions to enable xenpaging and
xen-access. The xc_mem_event_enable/xc_mem_event_disable functions will
be removed. The various XEN_DOMCTL_MEM_EVENT_* macros were cleaned up.
Due to the removal the API changed, so the SONAME will be changed too.
Signed-off-by: Olaf Hering <olaf@aepfle.de>
Acked-by: Tim Deegan <tim@xen.org>
Acked-by: Ian Jackson <ian.jackson@eu.citrix.com>
Committed-by: Tim Deegan <tim@xen.org>
tools/xenpaging/mem_event.h is only included in xenpaging.h.
Add the contents into that file and remove mem_event.h.
Signed-off-by: Olaf Hering <olaf@aepfle.de>
Committed-by: Ian Jackson <ian.jackson.citrix.com>
Write all paged-out pages back into the guest if the pager is
interrupted by ctrl-c or if it receives SIGTERM.
Signed-off-by: Olaf Hering <olaf@aepfle.de>
Committed-by: Ian Jackson <ian.jackson.citrix.com>
If xenpaging is started manually then no event is sent to xenpaging when
the guest is shutdown or rebooted. Add a watch on the @releaseDomain
node to leave the loop and gracefully shutdown the pager.
Signed-off-by: Olaf Hering <olaf@aepfle.de>
Committed-by: Ian Jackson <ian.jackson.citrix.com>
Unlink pagefile in the signal handler and also in the exit path.
This does not leave a stale pagefile if an error occoured.
Signed-off-by: Olaf Hering <olaf@aepfle.de>
Committed-by: Ian Jackson <ian.jackson.citrix.com>
Pass gfn as integer to xenpaging_populate_page(). xc_map_foreign_pages()
takes a pointer to a list of gfns, but its a const pointer. So writing
the value back to the caller is not needed.
Signed-off-by: Olaf Hering <olaf@aepfle.de>
Committed-by: Ian Jackson <ian.jackson.citrix.com>
Set the starting gfn to somewhere in the middle of the gfn range to
avoid paging during BIOS startup. This can speedup booting of a guest.
Signed-off-by: Olaf Hering <olaf@aepfle.de>
Committed-by: Ian Jackson <ian.jackson.citrix.com>
Move num_pages into struct xenpaging.
num_pages will be used by the policy in a subsequent patch.
Also remove a memset, the victims array is allocated with calloc.
Signed-off-by: Olaf Hering <olaf@aepfle.de>
Committed-by: Ian Jackson <ian.jackson.citrix.com>
Remove the local domain_id variable, it is already fetched from
paging->mem_event in other places.
Update the sprintf format string to use unsigned argument.
Signed-off-by: Olaf Hering <olaf@aepfle.de>
Committed-by: Ian Jackson <ian.jackson.citrix.com>
In the unlikely event that xc_mem_paging_resume() fails, do not overwrite the
error with the return value from xc_evtchn_notify()
Signed-off-by: Olaf Hering <olaf@aepfle.de>
Committed-by: Ian Jackson <ian.jackson.citrix.com>
get_request() and put_response() can not fail, remove return value
and update calling functions.
Signed-off-by: Olaf Hering <olaf@aepfle.de>
Committed-by: Ian Jackson <ian.jackson.citrix.com>
The policy uses now a linear algorithm instead of a random one.
Remove the call to srand().
Signed-off-by: Olaf Hering <olaf@aepfle.de>
Committed-by: Ian Jackson <ian.jackson.citrix.com>
do not bounce p2mt back to the hypervisor because p2m_mem_paging_populate()
and p2m_mem_paging_resume() dont make use of p2mt.
Signed-off-by: Olaf Hering <olaf@aepfle.de>
Committed-by: Ian Jackson <ian.jackson.citrix.com>
Doing a one-way channel from Xen to xenpaging is not possible with the
current ring buffer implementation. xenpaging uses the mem_event ring
buffer, which expects request/response pairs to make progress. The
previous patch, which tried to establish a one-way communication from
Xen to xenpaging, stalled the guest once the buffer was filled up with
requests. Correct page-dropping by taking the slow path and let
p2m_mem_paging_resume() consume the response from xenpaging. This makes
room for yet another request/response pair and avoids hanging guests.
Signed-off-by: Olaf Hering <olaf@aepfle.de>
Committed-by: Ian Jackson <ian.jackson.citrix.com>
Finally remove xc.c/xc.h and its references since both are empty now.
Signed-off-by: Olaf Hering <olaf@aepfle.de>
Committed-by: Ian Jackson <ian.jackson.citrix.com>
Move xc_wait_for_event_or_timeout() into xenpaging and massage it a bit for
further changes in subsequent patches.
Include poll.h instead of sys/poll.h.
Signed-off-by: Olaf Hering <olaf@aepfle.de>
Committed-by: Ian Jackson <ian.jackson.citrix.com>
Move xc_mem_paging_flush_ioemu_cache() into xenpaging and massage it a bit to
use the required members from xenpaging_t.
Also update type of rc to match xs_write() return value.
Signed-off-by: Olaf Hering <olaf@aepfle.de>
Committed-by: Ian Jackson <ian.jackson.citrix.com>
A xs_handle is currently used in the xc_mem_paging_flush_ioemu_cache()
function and will be used by a subsequent patch.
Add it to struct xenpaging.
Signed-off-by: Olaf Hering <olaf@aepfle.de>
Committed-by: Ian Jackson <ian.jackson.citrix.com>
Acked-by: Ian Campbell <ian.campbell@citrix.com>
Bitmaps are used in save/restore, xenpaging and blktap2. Merge the code into a
private xc_bitops.h file. All users are single threaded, so locking is not an
issue. The array of bits is handled as volatile because the x86 save/restore
code passes the bitmap to the hypervisor which in turn modifies the bitmap.
blktap2 uses a private bitmap. There was a possible overflow in the
bitmap_size() function, the remainder was not considered.
ia64 save/restore uses a bitmap to send the number of vcpus to the host.
x86 save/restore uses a bitmap to track dirty pages. This bitmap is shared with
the hypervisor. An unused function count_bits() was removed and a new
bitmap_size() function is now used.
xenpaging uses 3 private bitmaps to track the gfns which are in paged-out
state. It had a copy of some Linux bitops.h, which is now obsolete. Also the
BITS_PER_LONG macro was hardcoded to 64 which made it impossible to run 32bit
tools on a 64bit host. Wether this works at all has to be tested, yet.
Signed-off-by: Olaf Hering <olaf@aepfle.de>
Committed-by: Ian Jackson <ian.jackson.citrix.com>
The spinlock code in the pager is a no-op because xenpaging is a single
threaded application. There is no locking when put_response() places a
response into the ringbuffer.
The only locking is inside the hypervisor, where mem_event_put_request() and
mem_event_get_response() lock the ringbuffer to protect multiple vcpus from
each other.
Signed-off-by: Olaf Hering <olaf@aepfle.de>
Committed-by: Ian Jackson <ian.jackson.citrix.com>
Andres Lagar-Cavilla
Olaf Hering
Roger Pau Monne
Tim Deegan