Jan Beulich [Tue, 24 Nov 2015 11:06:26 +0000 (12:06 +0100)]
VMX: allocate VMCS pages from domain heap
There being only very few uses of the virtual address of a VMCS,
convert these cases to establish a mapping and lift the Xen heap
restriction from the VMCS allocation.
Signed-off-by: Jan Beulich <jbeulich@suse.com> Acked-by: Kevin Tian <kevin.tian@intel.com>
Andrew Cooper [Thu, 19 Nov 2015 14:45:41 +0000 (14:45 +0000)]
tools/libxc: Correct XC_DOM_PAGE_SIZE() to return a long long
c/s abdf3c5b "libxc: create p2m list outside of kernel mapping if supported"
introduces a use which Coverity objects to; an int used to mask a uint64_t.
The result needs to be signed to allow ~XC_DOM_PAGE_SIZE() to function
correctly, and long long to function properly in 32bit builds.
Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com> Acked-by: Wei Liu <wei.liu2@citrix.com>
Juergen Gross [Thu, 19 Nov 2015 16:11:08 +0000 (17:11 +0100)]
libxl: correct bug in domain builder regarding page tables for pvh
Commit 81a76e4b12961a9f54f5021809074196dfe6dbba ("libxc: rework of
domain builder's page table handler") dropped a special case for pvh
resulting in page tables being mapped read-only. This led to a panic
of the domain in early boot.
Correct this error.
Signed-off-by: Juergen Gross <jgross@suse.com> Reviewed-by: Wei Liu <wei.liu2@citrix.com> Tested-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Jan Beulich [Fri, 20 Nov 2015 11:38:33 +0000 (12:38 +0100)]
x86/P2M: consolidate handling of types not requiring a valid MFN
As noted regarding the mixture of checks in p2m_pt_set_entry(),
introduce a new P2M type group allowing to be used everywhere we
just care about accepting operations with either a valid MFN or a type
permitting to be used without (valid) MFN.
Note that p2m_mmio_dm is not included in P2M_NO_MFN_TYPES, as for the
intended purpose that one ought to be treated similar to p2m_invalid
(perhaps the two should ultimately get folded anyway).
Note further that PoD superpages now get INVALID_MFN used when creating
page table entries (was _mfn(0) before).
Signed-off-by: Jan Beulich <jbeulich@suse.com> Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com> Reviewed-by: George Dunlap <george.dunlap@citrix.com>
Jan Beulich [Fri, 20 Nov 2015 11:37:37 +0000 (12:37 +0100)]
x86/PoD: tighten conditions for checking super page
Since calling the function isn't cheap, try to avoid the call when we
know up front it won't help; see the code comment for details on those
conditions.
Signed-off-by: Jan Beulich <jbeulich@suse.com> Reviewed-by: George Dunlap <george.dunlap@citrix.com>
Jan Beulich [Thu, 19 Nov 2015 15:46:10 +0000 (16:46 +0100)]
x86/IO-APIC: fix setting of destinations
In commit a85da715cf ("x86/IO-APIC: adjust setting of destinations") I
made a pretty blatant mistake: get_apic_id() can be used there only
when running APICs in physical mode. For both flat and clustered modes
the change was wrong, causing different kinds of boot problems on
affected systems. Don't revert that change though, but use TARGET_CPUS
(equaling cpu_online_map, and with there only being a single online CPU
fulfilling the original commits intention).
Signed-off-by: Jan Beulich <jbeulich@suse.com> Tested-by: Boris Ostrovsky <boris.ostrovsky@oracle.com> Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
Andrew Cooper [Thu, 19 Nov 2015 15:44:59 +0000 (16:44 +0100)]
x86: fixes to LAPIC probing
* Fix (unsafe) assumption that X86_FEATURE_APIC resided in feature word 0.
* All 64bit processors have local APICs; drop the vendor check.
* Unconditionally probe MSR_IA32_APICBASE (safely, to fail more gracefully in
broken situations) and avoid a redundant double rdmsr().
* Avoid repeatedly OR'ing APICBASE_ENABLE and DEFAULT_PHYS_BASE when
attempting to reenable the LAPIC.
Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com> Reviewed-by: Jan Beulich <jbeulich@suse.com>
Jan Beulich [Tue, 17 Nov 2015 12:23:11 +0000 (13:23 +0100)]
ns16550: limit mapped MMIO size
There's no point in mapping more than the memory we actually may need
to touch, and in fact the too large region could actually extend into
another device's one (which currently is benign on x86 since only a
single page gets mapped anyway, but which is a latent bug on ARM
whenever PCI support gets enabled there).
Signed-off-by: Jan Beulich <jbeulich@suse.com> Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com> Acked-by: Ian Campbell <ian.campbell@citrix.com>
Juergen Gross [Thu, 12 Nov 2015 13:43:36 +0000 (14:43 +0100)]
libxc: create p2m list outside of kernel mapping if supported
In case the kernel of a new pv-domU indicates it is supporting a p2m
list outside the initial kernel mapping by specifying INIT_P2M, let
the domain builder allocate the memory for the p2m list from physical
guest memory only and map it to the address the kernel is expecting.
This will enable loading pv-domUs larger than 512 GB.
Signed-off-by: Juergen Gross <jgross@suse.com> Acked-by: Wei Liu <wei.liu2@citrix.com>
Juergen Gross [Thu, 12 Nov 2015 13:43:35 +0000 (14:43 +0100)]
libxc: rework of domain builder's page table handler
In order to prepare a p2m list outside of the initial kernel mapping
do a rework of the domain builder's page table handler. The goal is
to be able to use common helpers for page table allocation and setup
for initial kernel page tables and page tables mapping the p2m list.
This is achieved by supporting multiple mapping areas. The mapped
virtual addresses of the single areas must not overlap, while the
page tables of a new area added might already be partially present.
Especially the top level page table is existing only once, of course.
Currently restrict the number of mappings to 1 because the only mapping
now is the initial mapping created by toolstack. There should not be
behaviour change and guest visible change introduced.
Signed-off-by: Juergen Gross <jgross@suse.com> Reviewed-by: Wei Liu <wei.liu2@citrix.com>
Juergen Gross [Thu, 12 Nov 2015 13:43:34 +0000 (14:43 +0100)]
libxc: split p2m allocation in domain builder from other magic pages
Carve out the p2m list allocation from the .alloc_magic_pages hook of
the domain builder in order to prepare allocating the p2m list outside
of the initial kernel mapping. This will be needed to support loading
domains with huge memory (>512 GB).
Signed-off-by: Juergen Gross <jgross@suse.com> Acked-by: Ian Campbell <ian.campbell@citrix.com> Acked-by: Wei Liu <wei.liu2@citrix.com>
Juergen Gross [Thu, 12 Nov 2015 13:43:33 +0000 (14:43 +0100)]
libxc: create unmapped initrd in domain builder if supported
In case the kernel of a new pv-domU indicates it is supporting an
unmapped initrd, don't waste precious virtual space for the initrd,
but allocate only guest physical memory for it.
Signed-off-by: Juergen Gross <jgross@suse.com> Acked-by: Wei Liu <wei.liu2@citrix.com>
Juergen Gross [Thu, 12 Nov 2015 13:43:31 +0000 (14:43 +0100)]
libxc: introduce domain builder architecture specific data
Reorganize struct xc_dom_image to contain a pointer to domain builder
architecture specific private data. This will abstract the architecture
or domain type specific data from the general used data.
The new area is allocated as soon as the domain type is known.
Signed-off-by: Juergen Gross <jgross@suse.com> Acked-by: Wei Liu <wei.liu2@citrix.com>
Juergen Gross [Thu, 12 Nov 2015 13:43:30 +0000 (14:43 +0100)]
libxc: rename domain builder count_pgtables to alloc_pgtables
Rename the count_pgtables hook of the domain builder to alloc_pgtables
and do the allocation of the guest memory for page tables inside this
hook. This will remove the need for accessing the x86 specific pgtables
member of struct xc_dom_image in the generic domain builder code.
Signed-off-by: Juergen Gross <jgross@suse.com> Acked-by: Wei Liu <wei.liu2@citrix.com>
Juergen Gross [Thu, 12 Nov 2015 13:43:29 +0000 (14:43 +0100)]
xen: add generic flag to elf_dom_parms indicating support of unmapped initrd
Support of an unmapped initrd is indicated by the kernel of the domain
via elf notes. In order not to have to use raw elf data in the tools
for support of an unmapped initrd add a flag to the parsed data area
to indicate the kernel supporting this feature.
Switch using this flag in the hypervisor domain builder.
Cc: andrew.cooper3@citrix.com Cc: jbeulich@suse.com Cc: keir@xen.org Suggested-by: Ian Campbell <ian.campbell@citrix.com> Signed-off-by: Juergen Gross <jgross@suse.com> Acked-by: Jan Beulich <jbeulich@suse.com> Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
Guest memory allocation in the domain builder of libxc is done via
virtual addresses only. In order to be able to support preallocated
areas not virtually mapped reorganize the memory allocator to keep
track of allocated pages globally and in allocated segments.
This requires an interface change of the allocate callback of the
domain builder which currently is using the last mapped virtual
address as a parameter. This is no problem as the only user of this
callback is stubdom/grub/kexec.c using this virtual address to
calculate the last used pfn.
Signed-off-by: Juergen Gross <jgross@suse.com> Acked-by: Wei Liu <wei.liu2@citrix.com>
Jan Beulich [Mon, 16 Nov 2015 12:11:59 +0000 (13:11 +0100)]
x86/IO-APIC: adjust setting of destinations
setup_IO_APIC_irqs() runs before APs get brought up, so using
desc->arch.cpu_mask as best risks it being either empty or having bits
for CPUs other than the BP set. Just use the APIC ID of the only
online CPU directly.
Signed-off-by: Jan Beulich <jbeulich@suse.com> Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
Jan Beulich [Mon, 16 Nov 2015 12:11:08 +0000 (13:11 +0100)]
x86/IO-APIC: fix setup of Xen internally used IRQs (take 2)
..., i.e. namely that of a PCI serial card with an IRQ above the
legacy range. This had got broken by the switch to cpumask_any() in
cpu_mask_to_apicid_phys(). Fix this by allowing all CPUs for that IRQ
(via setup_vector_irq() properly updating a booting CPU's vector_irq[],
thus avoiding "No irq handler for vector" messages and the interrupt
not working).
Cleanup coding style and types there at once.
While doing this I also noticed that io_apic_set_pci_routing() can't
be quite right: It sets up the destination _before_ getting a vector
allocated (which on other than systems using the flat APIC mode
affects the possible destinations), and also didn't restrict affinity
to ->arch.cpu_mask (as established by assign_irq_vector()).
Signed-off-by: Jan Beulich <jbeulich@suse.com> Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
Andrew Cooper [Tue, 10 Nov 2015 10:46:44 +0000 (10:46 +0000)]
tools/ocaml/xb: Correct calculations of data/space the ring
ml_interface_{read,write}() would miscalculate the quantity of
data/space in the ring if it crossed the ring boundary, and incorrectly
return a short read/write.
This causes a protocol stall, as either side of the ring ends up waiting
for what they believe to be the other side needing to take the next
action.
Correct the calculations to cope with crossing the ring boundary.
In addition, correct the error detection. It is a hard error if the
producer index gets more than a ring size ahead of the consumer, or if
the consumer ever overtakes the producer.
Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com> Reviewed-by: Wei Liu <wei.liu2@citrix.com> Reviewed-by: Samuel Thibault <samuel.thibault@ens-lyon.org> Reviewed-by: David Scott <dave@recoil.org>
Jim Fehlig [Fri, 13 Nov 2015 02:40:46 +0000 (19:40 -0700)]
libxl: relax readonly check introduced by XSA-142 fix
The fix for XSA-142 is quite a big hammer, rejecting readonly
disk configuration even when the requested backend is known to
support readonly. While it is true that qemu doesn't support
readonly for emulated IDE or AHCI disks
$ /usr/lib/xen/bin/qemu-system-i386 \
-drive file=/tmp/disk.raw,if=ide,media=disk,format=raw,readonly=on
qemu-system-i386: Can't use a read-only drive
$ /usr/lib/xen/bin/qemu-system-i386 -device ahci,id=ahci0 \
-drive file=/tmp/disk.raw,if=none,id=ahcidisk-0,format=raw,readonly=on \
-device ide-hd,bus=ahci0.0,unit=0,drive=ahcidisk-0
qemu-system-i386: -device ide-hd,bus=ahci0.0,unit=0,drive=ahcidisk-0:
Can't use a read-only drive
Inside a guest using such a disk, the SCSI kernel driver sees write
protect on
[ 7.339232] sd 2:0:1:0: [sdb] Write Protect is on
Also, PV drivers support readonly, but the patch rejects such
configuration even when PV drivers (vdev=xvd*) have been explicitly
specified and creation of an emulated twin is skiped.
This follow-up patch loosens the restriction to reject readonly when
creating an emulated IDE or AHCI disk, but allows it when the backend
is known to support readonly.
Signed-off-by: Jim Fehlig <jfehlig@suse.com> Acked-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com> Acked-by: Ian Campbell <ian.campbell@citrix.com>
Juergen Gross [Fri, 23 Oct 2015 13:05:00 +0000 (15:05 +0200)]
libxc: remove most of tools/libxc/xc_dom_compat_linux.c
In tools/libxc/xc_dom_compat_linux.c xc_linux_build() is the only
domain building function used by an in-tree component (qemu-xen) which
is really necessary.
Remove the other domain building functions and the unused python
wrapper xc.linux_build() referencing one of the to be removed
functions.
Suggested-by: Ian Campbell <ian.campbell@citrix.com> Signed-off-by: Juergen Gross <jgross@suse.com> Acked-by: Wei Liu <wei.liu2@citrix.com> Acked-by: Ian Campbell <ian.campbell@citrix.com>
In Quota.merge, we merge two quota hashtables, orig_quota and mod_quota, putting
the results into dest_quota. These hashtables map domids to the number of
entries currently owned by that domain.
When mod_quota contains an entry for a domid that was not present in orig_quota
(or dest_quota), the call to get_entry caused Quota.merge to raise a Not_found
exception. This propagates back to the client as an ENOENT error, which is not
an appropriate return value from some operations, such as transaction_end.
This situation can arise when a transaction that introduces a domain (hence
calling Quota.add_entry) needs to be coalesced due to concurrent xenstore
activity.
This patch handles the merge in the case where mod_quota contains an entry not
present in orig_quota (or in dest_quota) by treating that hashtable as having
existing value 0.
Signed-off-by: Jonathan Davies <jonathan.davies@citrix.com> Acked-by: Ian Jackson <ian.jackson@eu.citrix.com>
xen/serial: Return actual bytes stored in TX FIFO for OMAP
This is intended to decrease a time spending in transmitter
while waiting for the free space in TX FIFO.
And as result to reduce the impact of hvc on the entire system
running on OMAP5/DRA7XX based platforms.
Signed-off-by: Oleksandr Tyshchenko <oleksandr.tyshchenko@globallogic.com> CC: Ian Campbell <ian.campbell@citrix.com> CC: Julien Grall <julien.grall@citrix.com> CC: Stefano Stabellini <stefano.stabellini@citrix.com> Cc: Jan Beulich <jbeulich@suse.com> Acked-by: Ian Campbell <ian.campbell@citrix.com>
xen/serial: Move any OMAP specific things to OMAP UART driver
The 8250-uart.h contains extra serial register definitions
for the internal UARTs in TI OMAP SoCs which are used in
OMAP UART driver only.
In order to clean up code move these definitions to omap-uart.c.
Also rename some definitions to follow to the UART_OMAP* prefix.
Signed-off-by: Oleksandr Tyshchenko <oleksandr.tyshchenko@globallogic.com> CC: Ian Campbell <ian.campbell@citrix.com> CC: Julien Grall <julien.grall@citrix.com> CC: Stefano Stabellini <stefano.stabellini@citrix.com> Cc: Jan Beulich <jbeulich@suse.com> Acked-by: Jan Beulich <jbeulich@suse.com> Acked-by: Ian Campbell <ian.campbell@citrix.com>
Try to use "xen-qemuuser-domid$domid" first, then
"xen-qemuuser-shared" and root if everything else fails.
The uids need to be manually created by the user or, more likely, by the
xen package maintainer.
Expose a device_model_user setting in libxl_domain_build_info, so that
opinionated callers, such as libvirt, can set any user they like. Do not
fall back to root if device_model_user is set. Users can also set
device_model_user by hand in the xl domain config file.
QEMU is going to setuid and setgid to the user ID and the group ID of
the specified user, soon after initialization, before starting to deal
with any guest IO.
To actually secure QEMU when running in Dom0, we need at least to
deprivilege the privcmd and xenstore interfaces, this is just the first
step in that direction.
Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com> Acked-by: Ian Campbell <ian.campbell@citrix.com>
Ian Campbell [Thu, 5 Nov 2015 14:46:12 +0000 (14:46 +0000)]
tools: pygrub: if partition table is empty, try treating as a whole disk
pygrub (in identify_disk_image()) detects a DOS style partition table
via the presence of the 0xaa55 signature at the end of the first
sector of the disk.
However this signature is also present in whole-disk configurations
when there is an MBR on the disk. Many filesystems (e.g. ext[234])
include leading padding in their on disk format specifically to enable
this.
So if we think we have a DOS partition table but do not find any
actual partition table entries we may as well try looking at it as a
whole disk image. Worst case is we probe and find there isn't anything
there.
This was reported by Sjors Gielen in Debian bug #745419. The fix was
inspired by a patch by Adi Kriegisch in
https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=745419#27
Tested by genext2fs'ing my /boot into a new raw image (works) and
then:
dd if=/usr/lib/grub/i386-pc/g2ldr.mbr of=img conv=notrunc bs=512 count=1
to add an MBR (with 0xaa55 signature) to it, which after this patch
also works.
Signed-off-by: Ian Campbell <ian.campbell@citrix.com> Acked-by: Wei Liu <wei.liu2@citrix.com> Acked-by: Ian Jackson <ian.jackson@eu.citrix.com> Cc: 745419-forwarded@bugs.debian.org
Ian Campbell [Wed, 11 Nov 2015 13:33:46 +0000 (13:33 +0000)]
tools: migration: Use PRIpfn when printing frame numbers.
This avoids various printf formatting warnings when building on arm32.
While touching the affected lines make them consistently use %#.
Signed-off-by: Ian Campbell <ian.campbell@citrix.com> Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com> Acked-by: Wei Liu <wei.liu2@citrix.com> Cc: Andrew Cooper <andrew.cooper3@citrix.com>
Jan Beulich [Fri, 13 Nov 2015 14:41:47 +0000 (15:41 +0100)]
ns16550: misc minor adjustments
First and foremost: fix documentation: The use of "clock_hz", when
"base_baud" was meant, has taken me several hours (suspecting a more
complicated problem with the PCIe card I've been trying to get
working). At once correct the "gdb" option, which is more like
"console", not like "com<N>".
Next, fix the types of ns_{read,write}_reg(): Especially the former
having had a signed return type so far caused quite interesting effects
when determining to baud rate if "auto" was specified. In that same
code, also avoid dividing by zero when in fact the baud rate was not
previously set up.
Further, accept I/O port based serial PCI cards with a port range wider
than 8 bytes.
Finally, slightly rearrange struct ns16550 to reduce holes.
Signed-off-by: Jan Beulich <jbeulich@suse.com> Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com> Acked-by: Ian Campbell <ian.campbell@citrix.com>
Jan Beulich [Thu, 12 Nov 2015 16:04:10 +0000 (17:04 +0100)]
x86/IO-APIC: fix setup of Xen internally used IRQs
..., i.e. namely that of a PCI serial card with an IRQ above the
legacy range. This had got broken by the switch to cpumask_any() in
cpu_mask_to_apicid_phys(). Fix this by allowing all CPUs for that IRQ
(such that __setup_vector_irq() will properly update a booting CPU's
vector_irq[], avoiding "No irq handler for vector" messages and the
interrupt not working).
While doing this I also noticed that io_apic_set_pci_routing() can't
be quite right: It sets up the destination _before_ getting a vector
allocated (which on other than systems using the flat APIC mode
affects the possible destinations), and also didn't restrict affinity
to ->arch.cpu_mask (as established by assign_irq_vector()).
Signed-off-by: Jan Beulich <jbeulich@suse.com> Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
Jan Beulich [Thu, 12 Nov 2015 16:03:20 +0000 (17:03 +0100)]
x86/event: correct debug event generation
RIP is not a linear address, and hence should not on its own be subject
to GVA -> GFN translation. Once at it, move all of the (perhaps
expensive) operations in the two functions into their main if()'s body,
and improve the error code passed to the translation function.
Signed-off-by: Jan Beulich <jbeulich@suse.com> Acked-by: Razvan Cojocaru <rcojocaru@bitdefender.com> Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
Jan Beulich [Thu, 12 Nov 2015 16:02:35 +0000 (17:02 +0100)]
x86: #PF error code adjustments
Add a definition for the (for now unused) protection key related error
code bit, moving our own custom ones out of the way. In the course of
checking the uses of the latter I realized that while right now they
can only get set on their own, callers would better not depend on that
property and check just for the bit rather than matching the entire
value.
Jan Beulich [Thu, 12 Nov 2015 16:01:53 +0000 (17:01 +0100)]
x86/traps: honor EXT bit in error codes
The specification does not explicitly limit the use of this bit to
exceptions that can have selector style error codes, so to be on the
safe side we should deal with it being set even on error codes formally
documented to be always zero (if they're indeed always zero, the change
is simply dead code in those cases).
Introduce and use (where suitable) X86_XEC_* constants to make the code
easier to read.
To match the placement of the "hardware_trap" label, the "hardware_gp"
one gets moved slightly too.
Signed-off-by: Jan Beulich <jbeulich@suse.com> Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
Jan Beulich [Thu, 12 Nov 2015 16:01:04 +0000 (17:01 +0100)]
x86/SVM: don't exceed segment limit when fetching instruction bytes
Also consistently use the vmcb local variable whenever possible.
Signed-off-by: Jan Beulich <jbeulich@suse.com> Reviewed-by: Boris Ostrovsky <boris.ostrovsky@oracle.com> Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
Jan Beulich [Thu, 12 Nov 2015 16:00:31 +0000 (17:00 +0100)]
x86/HVM: unify and fix #UD intercept
The SVM and VMX versions really were identical, so instead of fixing
the same issue in two places, fold them at once. The issue fixed is the
missing seg:off -> linear translation of the current code address.
Signed-off-by: Jan Beulich <jbeulich@suse.com> Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
This new elfnote contains the 32bit entry point into the kernel. Xen will
use this entry point in order to launch the guest kernel in 32bit protected
mode with paging disabled.
Shannon Zhao [Tue, 10 Nov 2015 11:08:29 +0000 (12:08 +0100)]
efi: fix booting failure with UEFI on ARM
Commit 9fd08b4 (efi: split out efi_get_gop()) splits out the
codes getting the pointer to GOP as efi_get_gop(), but it doesn't
initialize the variable handles and gop to NULL like what the original
codes do. This will cause booting failure on ARM while printing below
logs:
Xen 4.7-unstable (c/s Tue Oct 13 14:40:28 2015 +0100 git:7a92036) EFI loader
Synchronous Exception at 0x00000000FECB021C
Signed-off-by: Shannon Zhao <shannon.zhao@linaro.org> Acked-by: Jan Beulich <jbeulich@suse.com> Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
Riku Voipio [Tue, 10 Nov 2015 11:07:55 +0000 (12:07 +0100)]
symbols.c: avoid warn_unused_result build failure on fgets()
In commit:
d37d63d symbols: prefix static symbols with their source file names
An unchecked fgets was added. This causes a compile error at least
on ubuntu utopic:
symbols.c: In function 'read_symbol':
symbols.c:181:3: error: ignoring return value of 'fgets', declared with
attribute warn_unused_result [-Werror=unused-result]
fgets(str, 500, in); /* discard rest of line */
^
Paper over the warning by checking the return value in the if statement.
Signed-off-by: Riku Voipio <riku.voipio@linaro.org> Acked-by: Jan Beulich <jbeulich@suse.com>
Jan Beulich [Tue, 10 Nov 2015 11:03:08 +0000 (12:03 +0100)]
x86/HVM: always intercept #AC and #DB
Both being benign exceptions, and both being possible to get triggered
by exception delivery, this is required to prevent a guest from locking
up a CPU (resulting from no other VM exits occurring once getting into
such a loop).
The specific scenarios:
1) #AC may be raised during exception delivery if the handler is set to
be a ring-3 one by a 32-bit guest, and the stack is misaligned.
This is CVE-2015-5307 / XSA-156.
Reported-by: Benjamin Serebrin <serebrin@google.com>
2) #DB may be raised during exception delivery when a breakpoint got
placed on a data structure involved in delivering the exception. This
can result in an endless loop when a 64-bit guest uses a non-zero IST
for the vector 1 IDT entry, but even without use of IST the time it
takes until a contributory fault would get raised (results depending
on the handler) may be quite long.
This is CVE-2015-8104 / XSA-156.
Signed-off-by: Jan Beulich <jbeulich@suse.com> Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com> Tested-by: Andrew Cooper <andrew.cooper3@citrix.com>
Paul Durrant [Fri, 6 Nov 2015 14:17:00 +0000 (15:17 +0100)]
x86/hvm: make sure stdvga cache cannot be re-enabled
As soon as the cache is disabled, it will become out-of-sync with the
VGA device model and since no mechanism exists to acquire current VRAM
state from the device model, re-enabling it leads to stale data
being seen by the guest.
The problem was introduced by commit 3bbaaec0 ("x86/hvm: unify stdvga
mmio intercept with standard mmio intercept") and can be seen by
deliberately crashing a Windows guest; the BSOD output is corrupted.
This patch changes the existing 'cache' boolean in hvm_hw_stdvga into a
tri-state enum and only allows the state to move from 'uninitialized' to
'enabled'. Once the cache state becomes 'disabled' it will remain so for
the lifetime of the VM.
Signed-off-by: Paul Durrant <paul.durrant@citrix.com> Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
Dario Faggioli [Fri, 6 Nov 2015 14:16:38 +0000 (15:16 +0100)]
sched: fix locking of remove_vcpu() in credit1
In fact, csched_vcpu_remove() (i.e., the credit1
implementation of remove_vcpu()) manipulates runqueues,
so holding the runqueue lock is necessary.
However, the vCPU just can't be on the runqueue, when
the function is called. We can therefore ASSERT() that,
and avoid doing any runqueue manipulations (rather than
adding the runqueue locking around it).
Also, while there, *_lock_irq() (for the private lock) is
enough, there is no need to *_lock_irqsave().
Signed-off-by: Dario Faggioli <dario.faggioli@citrix.com> Reviewed-by: George Dunlap <george.dunlap@citrix.com>
Andrew Cooper [Wed, 4 Nov 2015 16:47:17 +0000 (17:47 +0100)]
x86: cleanup of early cpuid handling
Use register names for variables, rather than their content for leaf 1.
Reduce the number of cpuid instructions issued. Also drop some trailing
whitespace.
No functional change.
Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com> Reviewed-by: Jan Beulich <jbeulich@suse.com>
Harmandeep Kaur [Wed, 4 Nov 2015 16:46:46 +0000 (17:46 +0100)]
credit: remove cpu argument to __runq_insert()
__runq_insert() takes two arguments, cpu and svc. However,
the cpu argument is redundant because we can get all the
information we need about cpu from svc.
Ross Lagerwall [Mon, 2 Nov 2015 11:17:38 +0000 (11:17 +0000)]
xenconsoled: Remove unexpected daemonize behavior
Previously, xenconsoled's daemonize function would do nothing if its
parent process is init (as it is under systemd but not sysv init).
This is confusing. Instead, always daemonize when asked to, but use the
"interactive" switch when running from the systemd service.
Because a pidfile is only written when daemonizing, drop the pidfile
parameters from the service file (systemd keeps track of the pids
anyway).
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com> Acked-by: Ian Jackson <ian.jackson@eu.citrix.com> Acked-by: Wei Liu <wei.liu2@citrix.com>
Dario Faggioli [Wed, 4 Nov 2015 12:03:31 +0000 (13:03 +0100)]
xl: avoid (another) uninitialised use of rc in vcpuset()
Rearange the case when we check the new number of vCPUs
against the number of host pCPUs not to use rc for internal
error reporting. In fact:
- rc was at risk of being used uninitialised;
- rc should only be used for holding libxl error codes.
Signed-off-by: Dario Faggioli <dario.faggioli@citrix.com> Acked-by: Ian Campbell <ian.campbell@citrix.com>
Wei Liu [Wed, 4 Nov 2015 11:32:57 +0000 (11:32 +0000)]
xl: initialise rc before using it in vcpuset
In 5b725e56 (xl: improve return and exit codes of vcpu related
functions), the return value of libxl_cpu_bitmap_alloc was not stored in
rc anymore. Yet the subsequent fprintf still used that.
Reinstate the original implementation, that is, to store return value of
libxl_cpu_bitmap_alloc in rc before using rc.
Signed-off-by: Wei Liu <wei.liu2@citrix.com> Acked-by: Ian Campbell <ian.campbell@citrix.com>
Andrew Cooper [Tue, 3 Nov 2015 17:15:58 +0000 (18:15 +0100)]
x86: query for paddr_bits in early_cpu_detect()
It is __read_mostly, so repeatedly writing to it is suboptiomal. As the
MTRRs have already been set up, nothing good will come from its value
changing across CPUs.
Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com> Reviewed-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Andrew Cooper [Tue, 3 Nov 2015 17:15:15 +0000 (18:15 +0100)]
x86/vmx: replace unqualified ud2 instructions with BUG frames
Using new _ASM_BUGFRAME* internals.
A side effect of complicating the ASM statements is that GCC now chooses to
out-of-line the stub functions, resulting in identical copies being present in
all translation units. As with the stac()/clac() stubs, force them always
inline.
No functional change, other than the failure cases, which now produce a
far more clear error message.
Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com> Acked-by: Kevin Tian <kevin.tian@intel.com>
Andrew Cooper [Tue, 3 Nov 2015 17:14:49 +0000 (18:14 +0100)]
x86/bug: break out the internals of BUG_FRAME()
To allow bug frames can be created inside existing asm() statements. In
order to do so, the current bugframe positional parameters are altered
to be named parameters, to avoid interactions with the parameters of the
existing asm() statement.
No functional change.
Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com> Reviewed-by: Jan Beulich <jbeulich@suse.com>
Jan Beulich [Tue, 3 Nov 2015 17:11:56 +0000 (18:11 +0100)]
x86/HAP: use %pv printk() format where suitable
Signed-off-by: Jan Beulich <jbeulich@suse.com> Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com> Acked-by: George Dunlap <george.dunlap@citrix.com>
Jan Beulich [Tue, 3 Nov 2015 17:07:20 +0000 (18:07 +0100)]
compat: enforce distinguishable file names in symbol table
To make it possible to tell apart the static symbols in files built a
second time for compat guest support, arrange for their source file
names to be prefixed by a suitable path. We can't do this without
explicit .file directives, since gcc has always been stripping paths
from file names handed to the internally generated .file directive.
However, we can leverage __FILE__ if we make sure the second instance
gets compiled out of other than the very directory the wrapper sits in.
Where suitable, remove the long redundant explicit inclusions of
xen/config.h at once.
Signed-off-by: Jan Beulich <jbeulich@suse.com> Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com> Acked-by: Ian Campbell <ian.campbell@citrix.com>
Jan Beulich [Tue, 3 Nov 2015 17:05:35 +0000 (18:05 +0100)]
symbols: prefix static symbols with their source file names
This requires adjustments to the tool generating the symbol table and
its as well as nm's invocation.
Note: Not warning about duplicate symbols in the EFI case for now, as
a binutils bug causes misnamed file name entries to appear in EFI
binaries' symbol tables when the file name is longer than 18 chars.
(Not doing so also avoids other duplicates getting printed twice.)
Signed-off-by: Jan Beulich <jbeulich@suse.com> Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com> Acked-by: Ian Campbell <ian.campbell@citrix.com>
Harmandeep Kaur [Wed, 28 Oct 2015 02:26:24 +0000 (07:56 +0530)]
xl: improve return and exit codes of parse related functions
Turning parsing related functions exit codes towards using the
EXIT_[SUCCESS|FAILURE] constants, instead of instead of arbitrary numbers
or libxl return codes.
- for main_*: arbitrary -> EXIT_SUCCESS|EXIT_FAILURE.
- for internal fucntion: arbitrary -> 0/1.
Don't touch parse_config_data() which is big enough to deserve its own patch.
Harmandeep Kaur [Wed, 28 Oct 2015 02:26:23 +0000 (07:56 +0530)]
xl: improve return and exit codes of parse related functions
Turning cpupools related functions exit codes towards using the
EXIT_[SUCCESS|FAILURE] constants, instead of instead of arbitrary numbers
or libxl return codes.
Harmandeep Kaur [Wed, 28 Oct 2015 02:26:22 +0000 (07:56 +0530)]
xl: improve return and exit codes of vcpu related functions
Turning vcpu manipulation functions exit codes toward using the
EXIT_[SUCCESS|FAILURE] constants, instead of instead of arbitrary numbers
or libxl return codes.
Harmandeep Kaur [Wed, 28 Oct 2015 02:26:21 +0000 (07:56 +0530)]
xl: improve return and exit codes of scheduling related functions
Turning scheduling related functions exit codes towards using the
EXIT_[SUCCESS|FAILURE] constants, instead of instead of arbitrary numbers
or libxl return codes.
- for main_*: arbitrary -> EXIT_SUCCESS|EXIT_FAILURE.
- for internal fucntion: arbitrary -> 0/1.
Harmandeep Kaur [Wed, 28 Oct 2015 02:26:20 +0000 (07:56 +0530)]
xl: convert main() exit codes to EXIT_[SUCCESS|FAILURE]
Turning main() function exit codes towards using the EXIT_[SUCCESS|FAILURE]
constants, instead of instead of arbitrary numbers or libxl return codes.
Also includes a document comment in xl.h stating xl process should always
return EXIT_FOO and main_* can be treated as main() as if they are returning
a process exit status and not a function return value)
Olaf Hering [Thu, 29 Oct 2015 11:02:54 +0000 (11:02 +0000)]
tools/hotplug: xendomains.service conflicts with libvirt
xendomains will manage guests behind libvirts back:
- libvirt starts a guest
- that guest can be "managed" by libvirt and xl at the same time
- when xendomains runs on shutdown it will save the guest using xl
libvirt does not know about this
- when xendomains runs on boot it will restore the saved guest using xl
libvirt does not know about this, it will just fail to manage the
restored guest
To prevent xendomains from interfering with libvirt add a Conflicts= to
xendomains.service. It will cause libvirt to be stopped if xendomains is
started manually with 'systemctl start'.
Signed-off-by: Olaf Hering <olaf@aepfle.de> Cc: Ian Jackson <ian.jackson@eu.citrix.com> Cc: Stefano Stabellini <stefano.stabellini@eu.citrix.com> Cc: Ian Campbell <ian.campbell@citrix.com> Cc: Wei Liu <wei.liu2@citrix.com> Acked-by: Wei Liu <wei.liu2@citrix.com>
Dario Faggioli [Mon, 2 Nov 2015 14:33:19 +0000 (15:33 +0100)]
credit1: on vCPU wakeup, kick away current only if makes sense
In fact, when waking up a vCPU, __runq_tickle() is called
to allow the new vCPU to run on a pCPU (which one, depends
on the relationship between the priority of the new vCPU,
and the ones of the vCPUs that are already running).
If there is no idle processor on which the new vCPU can
run (e.g., because of pinning/affinity), we try to migrate
away the vCPU that is currently running on the new vCPU's
processor (i.e., the processor on which the vCPU is waking
up).
Now, trying to migrate a vCPU has the effect of pushing it
through a
running --> offline --> runnable
transition, which, in turn, has the following negative
effects:
1) Credit1 counts that as a wakeup, and it BOOSTs the
vCPU, even if it is a CPU-bound one, which wouldn't
normally have deserved boosting. This can prevent
legit IO-bound vCPUs to get ahold of the processor
until such spurious boosting expires, hurting the
performance!
2) since the vCPU is fails the vcpu_runnable() test
(within the call to csched_schedule() that follows
the wakeup, as a consequence of tickling) the
scheduling rate-limiting mechanism is also fooled,
i.e., the context switch happens even if less than
the minimum execution amount of time passed.
In particular, 1) has been reported to cause the following
issue:
* VM-IO: 1-vCPU pinned to a pCPU, running netperf
* VM-CPU: 1-vCPU pinned the the same pCPU, running a busy
CPU loop
==> Only VM-I/O: throughput is 806.64 Mbps
==> VM-I/O + VM-CPU: throughput is 166.50 Mbps
This patch solves (for the above scenario) the problem
by checking whether or not it makes sense to try to
migrate away the vCPU currently running on the processor.
In fact, if there aren't idle processors where such a vCPU
can execute. attempting the migration is just futile
(harmful, actually!).
With this patch, in the above configuration, results are:
==> Only VM-I/O: throughput is 807.18 Mbps
==> VM-I/O + VM-CPU: throughput is 731.66 Mbps
Reported-by: Kun Suo <ksuo@uccs.edu> Signed-off-by: Dario Faggioli <dario.faggioli@citrix.com> Tested-by: Kun Suo <ksuo@uccs.edu> Acked-by: George Dunlap <george.dunlap@citrix.com>
Jan Beulich [Mon, 2 Nov 2015 14:32:48 +0000 (15:32 +0100)]
x86: make compat_iret() domain crash cases distinguishable
Rather than issuing a (mostly) useless separate message, rely on
domain_crash() providing enough data, and leverage the line number
information it prints.
Signed-off-by: Jan Beulich <jbeulich@suse.com> Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
Jan Beulich [Mon, 2 Nov 2015 14:28:33 +0000 (15:28 +0100)]
libxlu: avoid linker warnings
Recent ld warns about libxenlight.so's dependency libraries not being
available, which can be easily avoided by not just passing the raw
library name on ld's command line.
In the course of checking how things fit together (I originally
suspected the warning to come from the linking of xl) I also noticed a
stray L in SHLIB_libxenguest, which gets removed at once.
Signed-off-by: Jan Beulich <jbeulich@suse.com> Acked-by: Wei Liu <wei.liu2@citrix.com> Acked-by: Ian Campbell <ian.campbell@citrix.com>
Jan Beulich [Mon, 2 Nov 2015 14:26:40 +0000 (15:26 +0100)]
drop get_xen_guest_handle()
Its use in the tools (and its apparent abuse in the hypervisor) are
long gone.
Signed-off-by: Jan Beulich <jbeulich@suse.com> Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com> Reviewed-by: Julien Grall <julien.grall@citrix.com> Acked-by: Ian Campbell <ian.campbell@citrix.com>
Ian Jackson [Wed, 21 Oct 2015 15:18:30 +0000 (16:18 +0100)]
libxl: adjust PoD target by memory fudge, too
PoD guests need to balloon at least as far as required by PoD, or risk
crashing. Currently they don't necessarily know what the right value
is, because our memory accounting is (at the very least) confusing.
Apply the memory limit fudge factor to the in-hypervisor PoD memory
target, too. This will increase the size of the guest's PoD cache by
the fudge factor LIBXL_MAXMEM_CONSTANT (currently 1Mby). This ensures
that even with a slightly-off balloon driver, the guest will be
stable even under memory pressure.
There are two call sites of xc_domain_set_pod_target that need fixing:
The one in libxl_set_memory_target is straightforward.
The one in xc_hvm_build_x86.c:setup_guest is more awkward. Simply
setting the PoD target differently does not work because the various
amounts of memory during domain construction no longer match up.
Instead, we adjust the guest memory target in xenstore (but only for
PoD guests).
This introduces a 1Mby discrepancy between the balloon target of a PoD
guest at boot, and the target set by an apparently-equivalent `xl
mem-set' (or similar) later. This approach is low-risk for a security
fix but we need to fix this up properly in xen.git#staging and
probably also in stable trees.
projects / people / julieng / xen-unstable.git / log