This patch is an amalgamation of the work done by Olaf Hering <olaf@aepfle.de>
and our work.
It combines logic changes to simplify the memory event API, as well as
leveraging wait queues to deal with extreme conditions in which too many
events are generated by a guest vcpu.
In order to generate a new event, a slot in the ring is claimed. If a
guest vcpu is generating the event and there is no space, it is put on a
wait queue. If a foreign vcpu is generating the event and there is no
space, the vcpu is expected to retry its operation. If an error happens
later, the function returns the claimed slot via a cancel operation.
Thus, the API has only four calls: claim slot, cancel claimed slot, put
request in the ring, consume the response.
With all these mechanisms, no guest events are lost.
Our testing includes 1. ballooning down 512 MiBs; 2. using mem access n2rwx, in
which every page access in a four-vCPU guest results in an event, with no vCPU
pausing, and the four vCPUs touching all RAM. No guest events were lost in
either case, and qemu-dm had no mapping problems.
Signed-off-by: Adin Scannell <adin@scannell.ca>
Signed-off-by: Andres Lagar-Cavilla <andres@lagarcavilla.org>
Signed-off-by: Olaf Hering <olaf@aepfle.de>
Signed-off-by: Tim Deegan <tim@xen.org>
Committed-by: Tim Deegan <tim@xen.org>
Also a couple of missing is_hvm_domain() checks.
Further properly pass the PCI segment in a call to pci_get_pdev().
Signed-off-by: Jan Beulich <jbeulich@suse.com>
This patch can improve Xen performance:
1. Basically, the "delay method" can achieve 11% overall performance
boost for SPECvirt than original credit scheduler.
2. We have tried 1ms delay and 10ms delay, there is no big difference
between these two configurations. (1ms is enough to achieve a good
performance)
3. We have compared different load level response time/latency (low,
high, peak), "delay method" didn't bring very much response time
increase.
4. 1ms delay can reduce 30% context switch at peak performance, where
produces the benefits. (int sched_ratelimit_us = 1000 is the
recommended setting)
Signed-off-by: Hui Lv <hui.lv@intel.com>
Signed-off-by: George Dunlap <george.dunlap@eu.citrix.com>
Acked-by: George Dunlap <george.dunlap@eu.citrix.com>
Committed-by: Keir Fraser <keir@xen.org>
Dealing with interrupts from VT-d IOMMU(s) is deferred to a
softirq-tasklet, raised by the actual IRQ handler. Since a new
interrupt is not generated, even if further faults occur, until we
cleared all the pending ones, there's no need of disabling IRQs, as
the hardware does it by its own. Notice that this may cause the log
to overflow, but none of the existing entry will be overwritten.
Signed-off-by: Dario Faggioli <dario.faggioli@citrix.com>
Committed-by: Keir Fraser <keir@xen.org>
No change to the generated code because this path isn't used yet.
Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Acked-by: Ian Jackson <ian.jackson.citrix.com>
Committed-by: Ian Jackson <ian.jackson.citrix.com>
At once also remove the duplicate forced inclusion of
public/xen-compat.h there (it's already done through cppflags-y) and
convert --include to the canonical -include in said c/s' adjustments
to CFLAGS-y and AFLAGS-y.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
sched_sedf.c used o have its own mechanism for producing tracing-alike
kind of information (domain block, wakeup, etc.). Nowadays, with an
even not so high number of pCPUs/vCPUs, just trying to enable this
makes the serial console completely unusable, produces tons of very
hard to parse and interpreet logging and can easily livelock
Dom0. Moreover, pretty much the same result this is struggling to get
to, is better achieved by enabling the scheduler-related tracing
events, as it is for the other schedulers (say, credit or credit2).
For all these reasons, this removes that machinery completely. While
at it, check in some cosmetics that harmonize the comments withim
themself and with the rest of the code base.
Signed-off-by: Dario Faggioli <dario.faggioli@citrix.com>
Acked-by: George Dunlap <george.dunlap@eu.citrix.com>
Several of the error exits from libxl_ctx_alloc leaked the context
struct itself and sometimes other resources too.
Fix this by using the standard "rc = ERROR_FOO; goto out" error
handling style throughout.
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>
Acked-by: Ian Campbell <ian.campbell@citrix.com>
Committed-by: Ian Jackson <Ian.Jackson@eu.citrix.com>
Previously the only official interface for the ctx lock was the
CTX_LOCK and CTX_UNLOCK convenience macros, which assume and use "ctx"
from the surrounding scope.
Instead, provide libxl__ctx_lock and _unlock functions which can be
used by these convenience macros, and other callers who have
nonstandard requirements.
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>
Acked-by: Ian Campbell <ian.campbell@citrix.com>
Committed-by: Ian Jackson <Ian.Jackson@eu.citrix.com>
Move a lot of
#include <stdfoo.h>
from individual files into libxl_internal.h. This helps avoid
portability mistakes where necessary system headers are omitted from
individual files, and is also of course a convenience when developing.
Also add
#include "libxl_osdeps.h" /* must come before any other headers */
to the top of most libxl*.c files, so that anyone who adds any headers
before libxl_internal.h will put the in the right place.
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>
Acked-by: Ian Campbell <ian.campbell@citrix.com>
Committed-by: Ian Jackson <Ian.Jackson@eu.citrix.com>
Previously LIBXL_INIT_GC was an initialiser, which you were expected
to use like this:
libxl__gc gc = LIBXL_INIT_GC(ctx);
But we are going to want to put things in the gc which are to be
initialised using other macros. That means that LIBXL_INIT_GC has to
become a statement too. So instead, we make it so that it's used like this:
libxl_gc gc;
LIBXL_INIT_GC(gc,ctx);
In fact there are only a couple of callers now,
including GC_INIT which uses this trick:
libxl_gc gc[1];
LIBXL_INIT_GC(gc[0],ctx);
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>
Acked-by: Ian Campbell <ian.campbell@citrix.com>
Committed-by: Ian Jackson <Ian.Jackson@eu.citrix.com>
This was always bogus (xen/config.h should have been used instead) and
is superfluous now that xen/config.h gets included through the compiler
command line.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Acked-by: Keir Fraser <keir@xen.org>
As we expect all source files to include the header as the first thing
anyway, stop doing this by repeating the inclusion in each and every
source file (and in many headers), but rather enforce this uniformly
through the compiler command line.
As a first cleanup step, remove the explicit inclusion from all common
headers. Further cleanup can be done incrementally.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Acked-by: Keir Fraser <keir@xen.org>
In an order-enforcing wrapper for an "external" recursive lock,
we aim to increment/decrement a recurse count and only update the
lock ordering on zero counts.
Unfortunately we incrementing/decrementing the pointer to the
recurse count, rather than the count itself.
Signed-off-by: Andres Lagar-Cavilla <andres@lagarcavilla.org>
Acked-by: Tim Deegan <tim@xen.org>
Committed-by: Tim Deegan <tim@xen.org>
This eliminates pointless prologue code from functions having variable
argument lists (since that way xmm registers can't possibly be passed).
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Acked-by: Keir Fraser <keir@xen.org>
Having it defined unilaterally as 'unsigned long' got me surprised
recently when I tried to use the 'z' printk type modifier, as that is
expected by the compiler to be used only on the type it knows size_t
is supposed to have.
Generally the compiler provides a construct to do this, so use it when
available.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Acked-by: Keir Fraser <keir@xen.org>
Acked-by: Christoph Egger <Christoph.Egger@amd.com>
Not only is it less code to have them consolidated, it also permits
their use virtually everywhere (since config.h is required to be
included everywhere. (Shouldn't we, btw, follow Linux and remove the
explicit inclusion in favor of command line enforced one?)
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Acked-by: Keir Fraser <keir@xen.org>
PPR log and event log share the same interrupt source. Interrupt
handler should check both of them.
Signed-off-by: Wei Wang <wei.wang2@amd.com>
Committed-by: Jan Beulich <jbeulich@suse.com>
Similar to nested paging for SVM, IOMMUv2 supports two level
translations for DMA. This patch enables this feature.
Signed-off-by: Wei Wang <wei.wang2@amd.com>
Committed-by: Jan Beulich <jbeulich@suse.com>
IOMMUv2 writes peripheral page service request (PPR) records into ppr
log to report guest OS pending DMA page requests from ATS devices.
Signed-off-by: Wei Wang <wei.wang2@amd.com>
Committed-by: Jan Beulich <jbeulich@suse.com>
... since it is pointless to have each arch declare them on their own
(and now and the - see ia64 - forget to do so).
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Acked-by: Keir Fraser <keir@xen.org>
24423:069c08b7de46 failed to add a necessary include, and for a long
time kexec.h suffered from a missing structure forward declaration.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Acked-by: Keir Fraser <keir@xen.org>
No functional changes, only code style issues.
One small diff in mem_sharing_gref_to_gfn() looks like a logic change,
but it's not. The flow was confusing, so it's been clarified.
Signed-off-by: Andres Lagar-Cavilla <andres@lagarcavilla.org>
Signed-off-by: Adin Scannell <adin@scannell.ca>
Acked-by: Tim Deegan <tim@xen.org>
Committed-by: Tim Deegan <tim@xen.org>
The only way to page-in a page is now the safe paging_load domctl.
(Unless the page was never paged out in the first place)
Signed-off-by: Andres Lagar-Cavilla <andres@lagarcavilla.org>
Acked-by: Tim Deegan <tim@xen.org>
Committed-by: Tim Deegan <tim@xen.org>
This removes the need for a page to be accessed in order to be pageable
again. A pager can now page-in pages at will with no need to map them
in a separate thread.
Signed-off-by: Andres Lagar-Cavilla <andres@lagarcavilla.org>
Acked-by: Tim Deegan <tim@xen.org>
Committed-by: Tim Deegan <tim@xen.org>
From: Wei, Gang <gang.wei@intel.com>
tboot may be trying to put APs waiting in MWAIT loops before launching
Xen. Xen could check the new flag field in v6 tboot shared page for the
hint. If TB_FLAG_AP_WAKE_SUPPORT bit in flag field is set, Xen BSP have
to write the monitored memory(g_tboot_shared->ap_wake_trigger) to bring
APs out of MWAIT loops. The sipi vector should be written in
g_tboot_shared->ap_wake_addr before waking up APs.
Signed-off-by: Joseph Cihula <joseph.cihula@intel.com>
Signed-off-by: Shane Wang <shane.wang@intel.com>
Signed-off-by: Gang Wei <gang.wei@intel.com>
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Committed-by: Jan Beulich <jbeulich@suse.com>
- use consistent error values (stop mixing of [positive] errno values
with literal -E... ones)
- properly format output
- don't use leading zeros in decimal output
- move printing of average frequency into P-state conditional (rather
than a C-state one)
- don't print some C-state related info when CPU idle management is
disabled in the hypervisor
- use calloc() for array allocations that also got cleared via memset()
so far
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Acked-by: Ian Jackson <ian.jackson@eu.citrix.com>
Committed-by: Ian Jackson <ian.jackson@eu.citrix.com>
Add a simple introduction to FLASK to the xl man page, at the beginning
of the FLASK chapter. Link to the xsm-flask.txt document.
Currently FLASK, TMEM and PCI PASS-THROUGH are defined as =head2 so they
look like sub-chapters of VIRTUAL DEVICE COMMANDS. Make them =head1.
Based on a text written by Daniel De Graaf.
Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Signed-off-by: Daniel De Graaf <dgdegra@tycho.nsa.gov>
Committed-by: Ian Jackson <ian.jackson@eu.citrix.com>
Tell guest about availibilty of xenstoreds XS_RESET_WATCHES function.
Guests can not issue this command unconditionally because some buggy
toolstacks (such as EC2) do not ignore unknown commands properly.
Signed-off-by: Olaf Hering <olaf@aepfle.de>
Committed-by: Ian Jackson <ian.jackson.citrix.com>
Acked-by: Ian Jackson <ian.jackson@eu.citrix.com>
The updated example policy did not include rules to allow managing the
created domains (pause, unpause, destroy); allow these actions.
Signed-off-by: Daniel De Graaf <dgdegra@tycho.nsa.gov>
Committed-by: Ian Jackson <ian.jackson.citrix.com>
If libiconv is detected on the system add -liconv when linking the
libvhd library.
If -liconv is not added when compiling libvhd with libiconv the
following error occours when linking vhd-util and vhd-update:
gcc -o vhd-util vhd-util.o -Llib -lvhd
lib/libvhd.so: undefined reference to `libiconv_open'
lib/libvhd.so: undefined reference to `libiconv_close'
lib/libvhd.so: undefined reference to `libiconv'
Signed-off-by: Roger Pau Monne <roger.pau@entel.upc.edu>
Committed-by: Ian Jackson <ian.jackson.citrix.com>
Acked-by: Ian Jackson <ian.jackson@eu.citrix.com>
Detect if libiconv is present in the system, since we will have to
link against it when using iconv.
Signed-off-by: Roger Pau Monne <roger.pau@entel.upc.edu>
Committed-by: Ian Jackson <ian.jackson.citrix.com>
Acked-by: Ian Jackson <ian.jackson@eu.citrix.com>
Andres Lagar-Cavilla