For the bias to be really meaningful, it should be updated only when
the CPU selected will indeed be returned (and hence used for placing
the vCPU in question).
Signed-off-by: Jan Beulich <jbeulich@novell.com>
This fixes a fairly blatant bug I introduced in c/s 20377:cff23354d026
- I wonder how this went unnoticed for so long.
Signed-off-by: Jan Beulich <jbeulich@novell.com>
This permits suspend/resume to work with 32bit dom0/tools when system
memory extends beyond 160GB (and up to 1TB).
AFAICT the limit to MACH2PHYS_COMPAT_NR_ENTRIES is redundant since
that refers to a limit in 32bit guest compat mappings under 64bit
hypervisors, not userspace where there may be gigabytes of useful
virtual space available for this.
Suggested-by: Ian Campbell <Ian.Campbell@eu.citrix.com>
Signed-off-by: Gianni Tedesco <gianni.tedesco@citrix.com>
c/s 21811:12f0618400de broke this by not properly initializing struct
cpuinfo_x86's x86_num_siblings member (other than Linux, where this is
a global variable, it is being maintained per CPU by Xen).
Hyper-threaded CPUs with CPUID leaf 0xb present had therefore all
cpu_sibling_map-s with just a single bit set, thus improperly feeding
the scheduler.
Signed-off-by: Jan Beulich <jbeulich@novell.com>
cpuid results are possible to be changed now. For example, changing
CR4.OSXSAVE bit or setting MSR XCR_XFEATURE_ENABLED_MASK may change
XSAVE related cpuid leave return values.
The volatile prefix is required to avoid the second cpuid calls
following some possible changing operations being optimized in
incorrect way by compiler.
The sample bug is in xsave_init while debug=3Dn. The second call to
cpuid_count() may be optimized and lead to a BUG_ON case while compare
xsave_cntxt_size with ebx.
Signed-off-by: Wei Gang <gang.wei@intel.com>
Some compiler versions may choose to not inline certain functions, but
the check introduced in c/s 23003:768269c43914 wants .text to be
empty.
Also make sure an eventual error gets properly propagated even on the
first section of an object (.text typically being the first one), and
cover a broader set of sections.
Signed-off-by: Jan Beulich <jbeulich@novell.com>
'unsigned int' is better suited as an array index on x86-64.
'u32' produces better code than 'unsigned long' on x86-64, so use the
former for storing 32-bit values read from the hardware.
this_cpu() uses an implicit smp_processor_id(), and hence using
per_cpu() when the result of smp_processor_id() is already available
is more efficient.
Fold one case of cpu_isset()+cpu_clear() into cpu_test_and_clear().
Drop the unused return value of evt_do_broadcast().
Signed-off-by: Jan Beulich <jbeulich@novell.com>
Acked-by: Wei Gang <gang.wei@intel.com>
Typically there are far less than 32 counters available, and hence
there's no use in wasting the memory on (almost) every system.
Signed-off-by: Jan Beulich <jbeulich@novell.com>
Acked-by: Wei Gang <gang.wei@intel.com>
- separate init and resume code paths (so that the [larger] init parts
can go init .init.* sections)
- drop the separate legacy_hpet_event object, as we can easily re-use
the first slot of hpet_events[] for that purpose (the whole array is
otherwise unused when the legacy code is being used)
- use section placement attributes where reasonable
Signed-off-by: Jan Beulich <jbeulich@novell.com>
Acked-by: Wei Gang <gang.wei@intel.com>
At least the legacy path can enter its interrupt handler callout while
initialization is still in progress - that handler checks whether
->event_handler is non-NULL, and hence all other initialization must
happen before setting this field.
Do the same to the MSI initialization just in case (and to keep the
code in sync).
Signed-off-by: Jan Beulich <jbeulich@novell.com>
This will allow us to use OS specific interfaces to ensure that the
allocated memory is safe for use as a hypercall buffer in the future.
Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
No need to wbinvd before HLT (entering C1) as the processor continues
to participate in the full cache coherency protocol in this sleep
state.
Signed-off-by: Keir Fraser <keir@xen.org>
At some platform (like Xen 7400), when hyperthreading, an offlined
thread may waked spuriously up by its brother, and returning around
the loop. This patch explicitly clflush the cache line in a light
weight way to workaround potential issue. Unlike wbinvd, clflush is
not serializing instruction, hence memory fence is necessary to make
sure all load/store operation visible before flush cache line.
Signed-off-by: Liu, Jinsong <jinsong.liu@intel.com>
Current xen cpu offline logic flush cache too early, which potentially
break cache coherency. wbinvd should be the last ops before cpu going
into dead, otherwise cache may be dirty, i.e, something like setting
an A bit on page tables. Pointed out by Arjan van de Ven.
Signed-off-by: Liu, Jinsong <jinsong.liu@intel.com>
Play dead is a fragile and tricky point of cpu offline logic. For how
to play cpu dead, linux kernel changed several times: Very old kernel
support 3 ways to play cpu dead: mwait, SYSIO, and halt, just like
what cpuidle did when enter C3; Later, it cancel mwait and SYSIO
support, only use halt to play dead; Latest linux 2.6.38 add mwait
support when cpu dead.
This patch cancel SYSIO method when cpu dead, keep same with latest
kernel.
SYSIO is an obsoleted method to enter deep C, with some tricky
hardware behavior, and seldom supported in new platform. Xen
experiment indicate that when cpu dead, SYSIO method would trigger
unknown issue which would bring strange error. We now cancel SYSIO
method when cpu dead, after all, correctness is more important than
power save, and btw new platform use mwait.
Signed-off-by: Liu, Jinsong <jinsong.liu@intel.com>
Before invoking C3, bus master disable / flush cache should be the
last step; After resume from C3, bus master enable should be the first
step;
Signed-off-by: Liu, Jinsong <jinsong.liu@intel.com>
Acked-by: Wei Gang <gang.wei@intel.com>
This is consistent with the expected semantics of a forced device
removal and also avoids a delay when destroying an HVM domain which
either does not support hot unplug (does not respond to SCI) or has
crashed.
Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Committed-by: Ian Jackson <ian.jackson@eu.citrix.com>
While doing development on libvirt libxenlight driver I noticed
that terminating a libxenlight client causes any qemu-dm
processes that were indirectly created by the client to also
terminate. Calling setsid(2) before exec'ing qemu-dm resolves
the issue.
Signed-off-by: Jim Fehlig <jfehlig@novell.com>
Acked-by: Ian Jackson <ian.jackson@eu.citrix.com>
Acked-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Committed-by: Ian Jackson <ian.jackson@eu.citrix.com>
... as Xen doesn't run on such CPUs anyway. Clearly these bits were
particularly odd to have on x86-64.
Signed-off-by: Jan Beulich <jbeulich@novell.com>
Introduce flexarray_append_pair: a simple syntactic sugar to add a pair
of pointers to a flexarray, very useful when adding pairs of strings to
xenstore.
Replace flexarray_vappend calls (and one pair of flexarray_append)
with flexarray_append_pair calls when dealing with flexarrays that are
going to be used with libxl__xs_kvs_of_flexarray:
NULL is a valid pointer value in these cases while flexarray_vappend
uses NULL as vargs terminator, so the old code is buggy.
Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Acked-by: Ian Jackson <ian.jackson@eu.citrix.com>
Committed-by: Ian Jackson <ian.jackson@eu.citrix.com>
In some cases, entire files turned out unnecessary. Of what remains,
move whatever possible into .init.*, and some data items into
.data.read_mostly.
Signed-off-by: Jan Beulich <jbeulich@novell.com>
This also includes the removal of some entirely unused functions.
The patch builds upon the makefile adjustments done in the earlier
sent patch titled "move more kernel decompression bits to .init.*
sections".
Signed-off-by: Jan Beulich <jbeulich@novell.com>
Remove unused and pointless bits from MTRR handling code. Move some
data items into .data.read_mostly. Adjust some types.
Signed-off-by: Jan Beulich <jbeulich@novell.com>
Rather than having per-architecture stubs when ATS support is not
implemented, simply provide inline functions in the common header for
that case.
Signed-off-by: Jan Beulich <jbeulich@novell.com>
Remove unused and pointless bits from IO-APIC handling code. Move
whatever possible into .init.*, and some data items into
.data.read_mostly. Adjust some types.
Signed-off-by: Jan Beulich <jbeulich@novell.com>
Remove unused and pointless bits from APIC handling code. Move some
data items into .data.read_mostly. Adjust some types.
Signed-off-by: Jan Beulich <jbeulich@novell.com>
Remove unused and pointless bits from mpparse.c (and other files where
they are related to it). Of what remains, move whatever possible into
.init.*, and some data items into .data.read_mostly.
Signed-off-by: Jan Beulich <jbeulich@novell.com>
Based on how c/s 22986:076b63b74cf6 changed xen/libelf/Makefile I
suppose this is compatibile with those clang/llvm changes, but I
didn't actually test it.
Signed-off-by: Jan Beulich <jbeulich@novell.com>
... so that finally all mod_lN_entry() functions behave identically,
allowing some cleanup in do_mmu_update() (which no longer needs to
track both an okay status and an error code).
Signed-off-by: Jan Beulich <jbeulich@novell.com>
... so that the guest can actually know the reason for the (hypercall)
failure.
ptwr_do_page_fault() could propagate the error indicator received from
get_page_from_l1e() back to the guest in the high half of the error
code (entry_vector), provided we're sure all existing guests can deal
with that (or indicate so by means of a to-be-added guest feature
flag). Alternatively, a second virtual status register (like CR2)
could be introduced.
Signed-off-by: Jan Beulich <jbeulich@novell.com>
Again, (out-of-memory) errors must not cause hypervisor crashes, and
hence ought to be propagated.
This also adjusts the cache attribute changing loop in
get_page_from_l1e() to not go through an unnecessary iteration. While
this could be considered mere cleanup, it is actually a requirement
for the subsequent now necessary error recovery path.
Also make a few functions static, easing the check for potential
callers needing adjustment.
Signed-off-by: Jan Beulich <jbeulich@novell.com>
Instead, propagate the condition to the caller, all of which also get
adjusted to check for that situation.
Signed-off-by: Jan Beulich <jbeulich@novell.com>
make 3.80 complains about a missing endif if the invocation of
cc-ver-check is itself wrapped in a conditional. Steal how the tools
deal with that.
Signed-off-by: Jan Beulich <jbeulich@novell.com>
When acquiring a transitive grant for copy then the owning domain
needs to be locked down as well as the granting domain. This was being
done, but the unlocking was not. The acquire code now stores the
struct domain * of the owning domain (rather than the domid) in the
active entry in the granting domain. The release code then does the
unlock on the owning domain. Note that I believe I also fixed a bug
where, for non-transitive grants the active entry contained a
reference to the acquiring domain rather than the granting
domain. From my reading of the code this would stop the release code
for transitive grants from terminating its recursion correctly.
Signed-off-by: Paul Durrant <paul.durrant@citrix.com>
Jan Beulich
Gianni Tedesco
Wei Gang
Tim Deegan
Stefano Stabellini
Ian Jackson
Ian Campbell
Keir Fraser
Liu, Jinsong
Jim Fehlig