This expands the format to include the class of TRC_HW_IRQ.
This means that instead of:
CPU28 1753521436727 (+ 3252) unknown (0x0000000000802008) [ 0x0000006c 0x4605709c 0x4605b682 0x00000000 0x00000000 0x00000000 0x00000000 ]
we now see:
CPU28 1753521436727 (+ 3252) do_irq [ irq = 108, began = 1174761628us, ended = 1174779522us ]
Acked-by: George Dunlap <george.dunlap@eu.citrix.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
After some more thought on the XSA-36 and specifically the comments we
got regarding disabling the IOMMU in this situation altogether making
things worse instead of better, I came to the conclusion that we can
actually restrict the action in affected cases to just disabling
interrupt remapping. That doesn't make the situation worse than prior
to the XSA-36 fixes (where interrupt remapping didn't really protect
domains from one another), but allows at least DMA isolation to still
be utilized.
To do so, disabling of interrupt remapping must be explicitly requested
on the command line - respective checks will then be skipped.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Acked-by: Suravee Suthikulanit <suravee.suthikulpanit@amd.com>
http://www.intel.com/content/www/us/en/chipsets/5520-and-5500-chipset-ioh-specification-update.html
Stepping B-3 has two errata (#47 and #53) related to Interrupt
remapping, to which the workaround is for the BIOS to completely disable
interrupt remapping. These errata are fixed in stepping C-2.
Unfortunately this chipset stepping is very common and many BIOSes are
not disabling interrupt remapping on this stepping . We can detect this in
Xen and prevent Xen from using the problematic interrupt remapping feature.
The Intel 5500/5520/X58 chipset does not support VT-d
Extended Interrupt Mode(EIM). This means the iommu_supports_eim() check
always fails and so x2apic mode cannot be enabled in Xen before this quirk
disables the interrupt remapping feature.
Signed-off-by: Malcolm Crossley <malcolm.crossley@citrix.com>
Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
Gate the function call to check the quirk on interrupt remapping being
requested to get enabled, and upon failure disable the IOMMU to be in
line with what the changes for XSA-36 (plus follow-ups) did.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Acked-by: "Zhang, Xiantao" <xiantao.zhang@intel.com>
... rather than the IOMMU as a whole.
That in turn required to make sure iommu_intremap gets properly
cleared when the respective initialization fails (or isn't being
done at all).
Along with making sure interrupt remapping doesn't get inconsistently
enabled on some IOMMUs and not on others in the VT-d code, this in turn
allowed quite a bit of cleanup on the VT-d side (if desired, that
cleanup could of course be broken out into a separate patch).
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Acked-by: "Zhang, Xiantao" <xiantao.zhang@intel.com>
Makes the following changes to the vnc generation code:
* Simplifies and comments it, making it easier to read and grok
* Throws an error if duplicate values of display are set, rather
than the current very un-intuitive behavior.
Signed-off-by: George Dunlap <george.dunlap@eu.citrix.com>
Acked-by: Ian Jackson <ian.jackson@eu.citrix.com>
Re-start blktap2 IPC select call on interrupt.
We hunted this miserable bug for a long time.
The teardown of a blktap2 tapdisk instance is being carried out
inconsistently up to and including the 4.2.1 release. The
problem appears to be a classic 'Heisenbug' which disappears if a
single function call is added to the tapdisk shutdown path. It
is likely this bug has been in existence for the life of the
blktap2 code.
Control messages to manipulate a tapdisk instance are sent over a
UNIX domain socket. A select call is used on both the read and
write paths to wait on I/O and to set a timeout for the
transmission and reception of the control plane messages.
The existing code fails receipt or transmission of the control message
on any type of error return from the select call. The xl control
process receives an interrupt while waiting in the select call which
in turn causes an error return with SIGINT as the return code.
This prematurely terminates the teardown of the tapdisk instance
leaving it in various states of shutdown. Since multiple messages
are needed to implement a full teardown the tapdisk instance can be
left in various states ranging from fully connected to only the minor
being left allocated.
The fix is straight forward. Check the return code from the
select call and re-try read or write of the control message if
errno is sent to EINTR. The problem manifests itself in the read
path but there appears to be little reason to not add the fix to
the write path as well. Both paths appear to be cut-and-paste
copies of each other.
Signed-off-by: Dr. Greg Wettstein <greg@enjellic.com>
Acked-by: Ian Jackson <ian.jackson@eu.citrix.com>
Some actions in APEI ERST and EINJ tables are optional, for example,
ACPI_EINJ_BEGIN_OPERATION action is used to do some preparation for
error injection, and firmware may choose to do nothing here. While
some other actions are mandatory, for example, firmware must provide
ACPI_EINJ_GET_ERROR_TYPE implementation.
Original implementation treats all actions as optional (that is, can
have no instructions), that may cause issue if firmware does not
provide some mandatory actions. To fix this, this patch adds
apei_exec_run_optional, which should be used for optional actions.
The original apei_exec_run should be used for mandatory actions.
Signed-off-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Tested-by: Andrew Cooper <andrew.cooper3@citrix.com>
This causes deadlocks during early boot on hardware with broken/buggy
APEI implementations, such as a Dell Poweredge 2950 with the latest
currently available BIOS.
Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
Don't use goto or another special error path, as handling the error
case in normal flow is quite simple.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Commit 2d8a282 ("x86/HPET: fix FSB interrupt masking") may cause the
HPET event to occur while its interrupt is masked. In that case we need
to "manually" deliver the event.
Unfortunately this requires the locking to be changed: For one, it was
always bogus for handle_hpet_broadcast() to use spin_unlock_irq() - the
function is being called from an interrupt handler, and hence shouldn't
blindly re-enable interrupts (this should be left up to the generic
interrupt handling code). And with the event handler wanting to acquire
the lock for two of its code regions, we must not enter it with the
lock already held. Hence move the locking into
hpet_{attach,detach}_channel(), permitting the lock to be dropped by
set_channel_irq_affinity() (which is a tail call of those functions).
Reported-by: Sander Eikelenboom <linux@eikelenboom.it>
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Tested-by: Sander Eikelenboom <linux@eikelenboom.it>
Acked-by: Keir Fraser <keir@xen.org>
Recent changes which have made their way into xen-4.2 stable have pushed the
runtime of construct_dom0() over 5 seconds, which has caused regressions in
XenServer testing because of our 5 second watchdog.
The root cause is that amd_iommu_dom0_init() does not process softirqs and in
particular the nmi_timer which causes the watchdog to decide that no useful
progress is being made.
This patch adds periodic calls to process_pending_softirqs() at the same
interval as the Intel variant of this function. The server which was failing
with the watchdog test now boots reliably with a timeout of 1 second.
Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
While being unable to reproduce the "No irq handler for vector ..."
messages observed on other systems, the change done by 5dc3fd2 ('x86:
extend diagnostics for "No irq handler for vector" messages') appears
to point at the lack of masking - at least I can't see what else might
be wrong with the HPET MSI code that could trigger these warnings.
While at it, also adjust the message printed by aforementioned commit
to not pointlessly insert spaces - we don't need aligned tabular output
here.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Acked-by: Keir Fraser <keir@xen.org>
Since pointer overflow is undefined behavior in C, some compilers such
as clang optimize away the check !((ptr) + 1) in the macro IS_NIL().
This patch fixes the issue by casting the pointer type to uintptr_t,
the operations of which are well-defined.
Signed-off-by: Xi Wang <xi@mit.edu>
With that, we also need to avoid the overflow in NIL().
Note that either part of the change results in the respective macros to
become unsuitable for use with "void".
Signed-off-by: Jan Beulich <jbeulich@suse.com>
libxl_create.c: In function ‘libxl__domain_build_info_setdefault’:
libxl_create.c:109: error: ‘info’ undeclared (first use in this function)
libxl_create.c:109: error: (Each undeclared identifier is reported only once
libxl_create.c:109: error: for each function it appears in.)
cc1: warnings being treated as errors
libxl_create.c:108: error: suggest explicit braces to avoid ambiguous ‘else’
libxl_create.c: At top level:
libxl_create.c:141: error: expected identifier or ‘(’ before ‘if’
...
Fix is to insert the missing opening brace and s/info/b_info/ in one spot.
Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Acked-by: Ian Jackson <ian.jackson@eu.citrix.com>
Committed-by: Ian Jackson <ian.jackson@eu.citrix.com>
By storing the inverted IRQ number in vector_irq[], we may be able to
spot which IRQ a vector was used for most recently, thus hopefully
permitting to understand why these messages trigger on certain systems.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Acked-by: Keir Fraser <keir@xen.org>
These calls ought always to succeed. Assert that they do rather than
ignoring the return value.
Signed-off-by: Tim Deegan <tim@xen.org>
Acked-by: Aravindh Puthiyaparambil <aravindh@virtuata.com>
This call ought always to succeed. Assert that it does rather than
ignoring the return value.
Signed-off-by: Tim Deegan <tim@xen.org>
Acked-by: Andres Lagar-Cavilla <andres@lagarcavilla.org>
Acked-by: Jan Beulich <jbeulich@suse.com>
AFAICT in all these cases we have the p2m lock and have just checked
that the p2m trie is populated so the call should succeed. Make it
explicit with ASSERT() rather than just ignoring the result.
Signed-off-by: Tim Deegan <tim@xen.org>
Acked-by: Jan Beulich <jbeulich@suse.com>
Even if the error propagates up through the p2m ops to the caller,
it'll look like ENOMEM, which won't be obviously a shadow-pool problem.
Warn on the console, once per domain.
Reported-by: Jan Beulich <jbeulich@suse.com>
Signed-off-by: Tim Deegan <tim@xen.org>
Acked-by: Jan Beulich <jbeulich@suse.com>
The stat in libxl__domain_build_info_setdefault's default device model
logic works to fall back to qemu-xen-traditional whenever the
executable for qemu-xen is not found.
We are going to use qemu-xen-traditional in more cases, so break this
check out into its own if statement.
Also add a pair of braces to make the if() statement symmetrical.
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>
Acked-by: Ian Campbell <ian.campbell@citrix.com>
When adding a disk don't launch multiple tapdisk instances for the
same disk, if transaction fails in device_disk_add reuse the same
tapdisk for further tries instead of creating a new instance each
time a transaction fails.
Reported-by: Darren Shepherd <darren.s.shepherd@gmail.com>
Signed-off-by: Roger Pau Monne <roger.pau@citrix.com>
Tested-by: Darren Shepherd <darren.s.shepherd@gmail.com>
I initially added hypervisor-new and confirmed via /proc/device-model
that the content is the same before changing it to drop and replace
an existing node.
NB: There is an ambiguity in the compatibility property.
linux/arch/arm/boot/dts/xenvm-4.2.dts says "xen,xen-4.2" while
Documentation/devicetree/bindings/arm/xen.txt says "xen,xen-4.3". I have
used the actual hypervisor version as discussed in
http://marc.info/?l=xen-devel&m=135963416631423
Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Acked-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
The bootloader should populate /chosen/modules/module@<N>/ for each
module it wishes to pass to the hypervisor. The content of these nodes
is described in docs/misc/arm/device-tree/booting.txt
Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Acked-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
If a node does not have #*-cells then the parent's value should be
used. Currently we were asssuming zero which is useless.
Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Acked-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Use a list of pointers to simplify the handling of 32- vs 64-bit.
Also on ARM the section name is ".init_array" and not ".ctors".
Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Acked-by: Keir Fraser <keir@xen.org>
[ ijc -- tweak whitespace per Frediano's comment ]
As the bits indicating validity of the ADDR and MISC bank MSRs may be
injected in a way that isn't consistent with what the underlying
hardware implements (while the bank must be valid for injection to
work, the auxiliary MSRs may not be implemented - and hence cause #GP
upon access - if the hardware never sets the corresponding valid bits.
Consequently we need to do the clearing writes only if no value was
interposed for the respective MSR (which also makes sense the other way
around: there's no point in clearing a hardware register when all data
read came from software). Of course this all requires the injection
tool to do things in a consistent way (but that had been a requirement
before already).
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Tested-by: Ren Yongjie <yongjie.ren@intel.com>
Acked-by: Liu Jinsong <jinsong.liu@intel.com>
This patch passes the orginal cpuid bits for X86_FEATURE_DTES64 (64-bit
DS Area) and X86_FEATURE_DSCPL (CPL Qualified Debug Store) to the guest
when the BTS feature is switched on. I forgot this when I did this BTS
emulation.
Signed-off-by: Dietmar Hahn <dietmar.hahn@ts.fujitsu.com>
In the Linux kernel, these two git commits:
- f594065faf4f9067c2283a34619fc0714e79a98d
ACPI: Add fixups for AMD P-state figures
- 9855d8ce41a7801548a05d844db2f46c3e810166
ACPI: Check MSR valid bit before using P-state frequencies
Try to fix the the issue that "some AMD systems may round the
frequencies in ACPI tables to 100MHz boundaries. We can obtain the real
frequencies from MSRs, so add a quirk to fix these frequencies up
on AMD systems." (from f594065..)
In discussion (around 9855d8..) "it turned out that indeed real
HW/BIOSes may choose to not set the valid bit and thus mark the
P-state as invalid. So this could be considered a fix for broken
BIOSes." (from 9855d8..)
which is great for Linux. Unfortunatly the Linux kernel, when
it tries to do the RDMSR under Xen it fails to get the right
value (it gets zero) as Xen traps it and returns zero. Hence
when dom0 uploads the P-states they will be unmodified and
we should take care of updating the frequencies with the right
values.
I've tested it under Dell Inc. PowerEdge T105 /0RR825, BIOS 1.3.2
08/20/2008 where this quirk can be observed (x86 == 0x10, model == 2).
Also on other AMD (x86 == 0x12, A8-3850; x86 = 0x14, AMD E-350) to
make sure the quirk is not applied there.
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Acked-by: stefan.bader@canonical.com
Do the MSR access here (and while at it, also the one reading
MSR_PSTATE_CUR_LIMIT) on the target CPU, and bound the loop over
amd_fixup_frequency() by max_hw_pstate (matching the one in
powernow_cpufreq_cpu_init()).
Signed-off-by: Jan Beulich <jbeulich@suse.com>
When guests memory consumption is volatile (multiple guests
ballooning up/down) we are presented with the problem of
being able to determine exactly how much memory there is
for allocation of new guests without negatively impacting
existing guests. Note that the existing models (xapi, xend)
drive the memory consumption from the tool-stack and assume
that the guest will eventually hit the memory target. Other
models, such as the dynamic memory utilized by tmem, do this
differently - the guest drivers the memory consumption (up
to the d->max_pages ceiling). With dynamic memory model, the
guest frequently can balloon up and down as it sees fit.
This presents the problem to the toolstack that it does not
know atomically how much free memory there is (as the information
gets stale the moment the d->tot_pages information is provided
to the tool-stack), and hence when starting a guest can fail
during the memory creation process. Especially if the process
is done in parallel. In a nutshell what we need is a atomic
value of all domains tot_pages during the allocation of guests.
Naturally holding a lock for such a long time is unacceptable.
Hence the goal of this hypercall is to attempt to atomically and very
quickly determine if there are sufficient pages available in the
system and, if so, "set aside" that quantity of pages for future
allocations by that domain. Unlike an existing hypercall such as
increase_reservation or populate_physmap, specific physical
pageframes are not assigned to the domain because this
cannot be done sufficiently quickly (especially for very large
allocations in an arbitrarily fragmented system) and so the
existing mechanisms result in classic time-of-check-time-of-use
(TOCTOU) races. One can think of claiming as similar to a
"lazy" allocation, but subsequent hypercalls are required
to do the actual physical pageframe allocation.
Note that one of effects of this hypercall is that from the
perspective of other running guests - suddenly there is
a new guest occupying X amount of pages. This means that when
we try to balloon up they will hit the system-wide ceiling of
available free memory (if the total sum of the existing d->max_pages
>= host memory). This is OK - as that is part of the overcommit.
What we DO NOT want to do is dictate their ceiling should be
(d->max_pages) as that is risky and can lead to guests OOM-ing.
It is something the guest needs to figure out.
In order for a toolstack to "get" information about whether
a domain has a claim and, if so, how large, and also for
the toolstack to measure the total system-wide claim, a
second subop has been added and exposed through domctl
and libxl (see "xen: XENMEM_claim_pages: xc").
== Alternative solutions ==
There has been a variety of discussion whether the problem
hypercall is solving can be done in user-space, such as:
- For all the existing guest, set their d->max_pages temporarily
to d->tot_pages and create the domain. This forces those
domains to stay at their current consumption level (fyi, this is what
the tmem freeze call is doing). The disadvantage of this is
that needlessly forces the guests to stay at the memory usage
instead of allowing it to decide the optimal target.
- Account only using d->max_pages of how much free memory there is.
This ignores ballooning changes and any over-commit scenario. This
is similar to the scenario where the sum of all d->max_pages (and
the one to be allocated now) on the host is smaller than the available
free memory. As such it ignores the over-commit problem.
- Provide a ring/FIFO along with event channel to notify an userspace
daemon of guests memory consumption. This daemon can then provide
up-to-date information to the toolstack of how much free memory
there is. This duplicates what the hypervisor is already doing and
introduced latency issues and catching breath for the toolstack as there
might be millions of these updates on heavily used machine. There might
not be any quiescent state ever and the toolstack will heavily consume
CPU cycles and not ever provide up-to-date information.
It has been noted that this claim mechanism solves the
underlying problem (slow failure of domain creation) for
a large class of domains but not all, specifically not
handling (but also not making the problem worse for) PV
domains that specify the "superpages" flag, and 32-bit PV
domains on large RAM systems. These will be addressed at a
later time.
Code overview:
Though the hypercall simply does arithmetic within locks,
some of the semantics in the code may be a bit subtle.
The key variables (d->unclaimed_pages and total_unclaimed_pages)
starts at zero if no claim has yet been staked for any domain.
(Perhaps a better name is "claimed_but_not_yet_possessed" but that's
a bit unwieldy.) If no claim hypercalls are executed, there
should be no impact on existing usage.
When a claim is successfully staked by a domain, it is like a
watermark but there is no record kept of the size of the claim.
Instead, d->unclaimed_pages is set to the difference between
d->tot_pages and the claim. When d->tot_pages increases or decreases,
d->unclaimed_pages atomically decreases or increases. Once
d->unclaimed_pages reaches zero, the claim is satisfied and
d->unclaimed pages stays at zero -- unless a new claim is
subsequently staked.
The systemwide variable total_unclaimed_pages is always the sum
of d->unclaimed_pages, across all domains. A non-domain-
specific heap allocation will fail if total_unclaimed_pages
exceeds free (plus, on tmem enabled systems, freeable) pages.
Claim semantics could be modified by flags. The initial
implementation had three flag, which discerns whether the
caller would like tmem freeable pages to be considered
in determining whether or not the claim can be successfully
staked. This in later patches was removed and there are no
flags.
A claim can be cancelled by requesting a claim with the
number of pages being zero.
A second subop returns the total outstanding claimed pages
systemwide.
Note: Save/restore/migrate may need to be modified,
else it can be documented that all claims are cancelled.
This patch of the proposed XENMEM_claim_pages hypercall/subop, takes
into account review feedback from Jan and Keir and IanC and Matthew Daley,
plus some fixes found via runtime debugging.
Signed-off-by: Dan Magenheimer <dan.magenheimer@oracle.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Acked-by: Tim Deegan <tim@xen.org>
Acked-by: Keir Fraser <keir@xen.org>
Under normal circumstances, snext->credit should never be less than
-CSCHED_MIN_TIMER. However, under some circumstances, a vcpu with low
credits may be allowed to run long enough that its credits are
actually less than -CSCHED_CREDIT_INIT.
(Instances have been observed, for example, where a vcpu with 200us of
credit was allowed to run for 11ms, giving it -10.8ms of credit. Thus
it was still negative even after the reset.)
If this is the case for snext, we simply want to keep moving everyone
up until it is in the black again. This fair because none of the
other vcpus want to run at the moment.
Rather than loop, just detect how many times we want to add
CSCHED_CREDIT_INIT. Try to avoid integer divides and multiplies in
the common case.
Signed-off-by: George Dunlap <george.dunlap@eu.citrix.com>
Konrad Rzeszutek Wilk
Jan Beulich
Malcolm Crossley
Wei Liu
George Dunlap
Dr. Greg Wettstein
fantonifabio@tiscali.it
Huang Ying
Andrew Cooper
Wei Wang
Ian Jackson
Roger Pau Monne
Xi Wang
Ian Campbell
Tim Deegan
Stefano Stabellini
Dietmar Hahn
Dan Magenheimer