Keir Fraser [Fri, 29 Aug 2008 15:16:33 +0000 (16:16 +0100)]
vtd: fix Dom0 S3 when VT-d is enabled.
Now if VT-d is enabled, when Dom0 does S3, Xen doesn't suspend/resume
the IOMMU states.
The patch adds the missing invocations of
iommu_suspend()/iommu_resume() and makes some nesessary fixes:
1) In iommu_set_root_entry(), we should not re-allocate the root-entry
when Xen returns from S3;
2) Define the array iommu_state correctly (u8 -> u32);
3) Only save/restore the necessary IOMMU registers.
The patch was tested on Weybridge.
NOTE: if we have some HVM guests which have assigned devices, and we
want to do Dom S3, we MUST do HVM S3 for each HVM guest first.
Namely, the steps are:
1) Use HVM S3 in guest to suspend the guest (for example, in Linux HVM
guest, this is "echo mem > /sys/power/state");
2) Use Dom0 S3 to suspend Dom0: "echo mem > /sys/power/state";
3) At some point, Dom0 resumes from S3 (for example, by pressing the
power button or using the acpi alarm);
4) Use "xm trigger HVM_DOM_ID s3resume" to resume every HVM guest
which were suspended previously.
Keir Fraser [Fri, 29 Aug 2008 15:11:05 +0000 (16:11 +0100)]
x86: constify microcode hypercall argument
Linux 2.6.27 marks the data pointer in its firmware struct 'const',
and hence, to avoid a compiler warning, Xen's microcode update
interface should be properly properly constified too.
This is a cleanup patch. It combines destroyDomain() method
into destroy() method, then _prepare_phantom_paths() method and
_cleanup_phantom_devs() method are called only once.
Keir Fraser [Wed, 27 Aug 2008 14:02:38 +0000 (15:02 +0100)]
tools/xenmon/Makefile: Move LDFLAGS after $<
gcc expects libraries needed for object files to be specified after
the object. Linking usually does not fail, unless it is optimized (for
instance, using -Wl,-as-needed).
The related Gentoo bug is 135145 [ https://bugs.gentoo.org/135145 ].
From: Robert Buchholz <rbu@gentoo.org> Signed-off-by: Keir Fraser <keir.fraser@citrix.com>
Keir Fraser [Wed, 27 Aug 2008 13:57:23 +0000 (14:57 +0100)]
ioemu: Add SEEK_STAT in all non-error cases except sleep request,
since our seek times are virtually 0. Although this flag is
deprecated by newer ATA standards, Linux at least checks it for CD
drives, and would timeout in some cases if we do not set it.
Signed-off-by: Samuel Thibault <samuel.thibault@eu.citrix.com>
Keir Fraser [Wed, 27 Aug 2008 13:53:39 +0000 (14:53 +0100)]
hvm: Use main memory for video memory.
When creating an HVM domain, if e.g. another domain is created before
qemu allocates video memory, the extra 8MB memory ballooning is not
available any more, because it got consumed by the other domain.
This fixes it by taking video memory from the main memory:
- make hvmloader use e820_malloc to reserve some of the main memory
and notify ioemu of its address through the Xen platform PCI card.
- add XENMAPSPACE_mfn to the xen_add_to_physmap memory op, to allow
ioemu to move the MFNs between the original position and the PCI
mapping, when LFB acceleration is disabled/enabled
- add a remove_from_physmap memory op, to allow ioemu to unmap it
completely for the case of old guests with acceleration disabled.
- add xc_domain_memory_translate_gpfn_list to libxc to allow ioemu to
get the MFNs of the video memory.
- have xend save the PCI memory space instead of ioemu: if a memory
page is there, the guest can access it like usual memory, so xend
can safely be responsible to save it. The extra benefit is that
live migration will apply the logdirty optimization there too.
- handle old saved images, populating the video memory from ioemu if
really needed.
Signed-off-by: Samuel Thibault <samuel.thibault@eu.citrix.com>
Keir Fraser [Wed, 27 Aug 2008 09:11:33 +0000 (10:11 +0100)]
x86: Assert in_irq() while processing guest-bound interrupts.
Actually a fair amount of infrastructure is onvolved these days, and
we'd like subsystems we call into know that we're in IRQ context so
they can handle that or BUG/ASSERT if appropriate.
Keir Fraser [Wed, 27 Aug 2008 08:46:38 +0000 (09:46 +0100)]
PyGRUB: fix menu flicker
To avoid flickers, PyGRUB has to avoid spurious ncurses refresh as
much as possible, in particular before the complete screen is drawn,
and eventually call doupdate last.
Signed-off-by: Samuel Thibault <samuel.thibault@eu.citrix.com>
Keir Fraser [Tue, 26 Aug 2008 13:11:39 +0000 (14:11 +0100)]
vtd: do FLR before xc.domain_destroy()
When we "xm destroy" a guest with assigned devices, in
xen/common/domain.c:domain_kill(), we relinquish guest's memory first,
then invoke domain_destroy() -> complete_domain_destroy() ->
arch_domain_destroy() -> pci_release_devices()/iommu_domain_destroy().
Between relinquish_memory() and domain_destroy(), a device may be
issuing DMA request and access guest's relinquished memory.
We can do_FLR() to quiesce devices before xc.domain_destroy().
Keir Fraser [Mon, 25 Aug 2008 13:03:42 +0000 (14:03 +0100)]
x86: Time rendezvous function does not need to disable/enable IRQs.
Since rendezvous happens within the IPI handler, no need to specify
'wait' to on_each_cpu().
Isaku Yamahata [Mon, 25 Aug 2008 10:04:37 +0000 (19:04 +0900)]
[IA64] introduce compile time option to disable optvfault.
Debugging optvfault path is extremely difficult because
it was run with psr.ic = 0.
Make it easy to differentiate the bug in optvfault path from
other normal execution path with psr.ic = 0 by introducing compile
time option to disable optvfault.
Isaku Yamahata [Mon, 25 Aug 2008 10:04:37 +0000 (19:04 +0900)]
[IA64] fix stack unwinder.
The stack unwinder can be called in interrupt context. On the other
hand xmalloc() can't be called in interrupt context.
Don't call xmalloc() if in the interrupt context.
Isaku Yamahata [Mon, 25 Aug 2008 10:04:37 +0000 (19:04 +0900)]
[IA64] fix PAL call in physical mode.
Now firmware code is mapped at its own virtual address space
which is same to Linux. So the calculation from virutal
address to physical address should be same to the linux one.
In practice firmware is located in low physical address area
so that masking 60 bit instead of 61 bits doesn't make difference.
I just checked and with the patch below,
xen/arch/ia64/linux-xen/pal.S becomes the same as
arch/ia64/kernel/pal.S from linux 2.6.21.
Keir Fraser [Tue, 19 Aug 2008 14:56:31 +0000 (15:56 +0100)]
x86_emulate: Do not request emulation of REP instructions beyond the
point at which the index register (SI/DI) wraps. This can cause a
discontinuity in the address range accessed by the repeated
instruction.
Keir Fraser [Tue, 19 Aug 2008 09:51:41 +0000 (10:51 +0100)]
vtd: Don't do FLR when we create domain or hotplug device into guest.
VT-d hot-plug is done in such a way: 1) xend sends 'pci-ins' command
to ioemu via xenstore; 2) ioemu creates pass-through device and raises
SCI to guest OS; guest OS loads device driver, and device driver
initializes the hot-plugged device; 3) xend does FLR. There is a race
condition about 2) and 3). If 3) runs slow, we have troubles. Thanks
for Yuji Shimada's locating the issues.
We can work around the issues by not doing FLR when we create domain
or hotplug device into guest, namely, we only do FLR when we destroy
domain or hotplug device from guest.
Signed-off-by: Dexuan Cui <dexuan.cui@intel.com> Signed-off-by: Dongxiao Xu <dongxiao.xu@intel.com>
Keir Fraser [Fri, 15 Aug 2008 13:59:41 +0000 (14:59 +0100)]
x86, hvm: Fix softtsc for AMD-V
The softtsc code for AMD does not update the ip; enabling it on AMD-V
results in the domain spinning on the RDTSC instruction. The patch is
against xen-unstable 18331.
Keir Fraser [Thu, 14 Aug 2008 09:18:08 +0000 (10:18 +0100)]
Fix restore crash with certain guest memory sizes
This fixes a crash when restoring guests with certain memory sizes (eg
223MB). The ROUNDUP in the call to memcpy, was the main offender, but
I didn't care for the inplace resizing of the p2m entries either.
Keir Fraser [Thu, 14 Aug 2008 09:14:32 +0000 (10:14 +0100)]
x86, shadow: Fix OOS on domain crash.
I couldn't reproduce the Nevada crash on my testbox, but this should
fix the first Xen crash that was seen in the Nevada HVM (bugzilla
#1322).
What I think most probably happened there is that the set_l2e call in
shadow_get_and_create_l1e() has tried to resync a page, but somehow we
weren't unable to remove the shadow (the real bug we should actually
look after). sh_resync() then removes the page from the OOS hash and
later in the page fault path we find the gw.l1mfn to be still OOS, so
we try to update the snapshot and the bug happens.
This should fix this and other unlikely (like sh_unsync() failing to
remove for hash collision the current gw.l1mfn) cases.
projects / xen.git / log