In drivers/xen/xenbus/xenbus_comms.h there is a stale declaration of
xs_input_avail(). Remove it.
Signed-off-by: Juergen Gross <[email protected]>
Reviewed-by: Boris Ostrovsky <[email protected]>
Signed-off-by: Juergen Gross <[email protected]>
When the xenbus driver does some special handling for a Xenstore
command any error condition related to the command should be returned
via an error response instead of letting the related write operation
fail. Otherwise the user land handler might take wrong decisions
assuming the connection to Xenstore is broken.
While at it try to return the same error values xenstored would
return for those cases.
Signed-off-by: Juergen Gross <[email protected]>
Reviewed-by: Boris Ostrovsky <[email protected]>
Signed-off-by: Juergen Gross <[email protected]>
When accessing Xenstore in a transaction the user is specifying a
transaction id which he normally obtained from Xenstore when starting
the transaction. Xenstore is validating a transaction id against all
known transaction ids of the connection the request came in. As all
requests of a domain not being the one where Xenstore lives share
one connection, validation of transaction ids of different users of
Xenstore in that domain should be done by the kernel of that domain
being the multiplexer between the Xenstore users in that domain and
Xenstore.
In order to prohibit one Xenstore user "hijacking" a transaction from
another user the xenbus driver has to verify a given transaction id
against all known transaction ids of the user before forwarding it to
Xenstore.
Signed-off-by: Juergen Gross <[email protected]>
Reviewed-by: Boris Ostrovsky <[email protected]>
Signed-off-by: Juergen Gross <[email protected]>
Instead of requesting a new slot on the ring to the backend early, do
so only after all has been setup for the request to be sent. This
makes error handling easier as we don't need to undo the request id
allocation and ring slot allocation.
Suggested-by: Jan Beulich <[email protected]>
Signed-off-by: Juergen Gross <[email protected]>
Reviewed-by: Boris Ostrovsky <[email protected]>
Signed-off-by: Juergen Gross <[email protected]>
Use xenbus_read_unsigned() instead of xenbus_scanf() when possible.
This requires to change the type of the reads from int to unsigned,
but these cases have been wrong before: negative values are not allowed
for the modified cases.
Signed-off-by: Juergen Gross <[email protected]>
Acked-by: David Vrabel <[email protected]>
Use xenbus_read_unsigned() instead of xenbus_scanf() when possible.
This requires to change the type of the read from int to unsigned,
but this case has been wrong before: negative values are not allowed
for the modified case.
Signed-off-by: Juergen Gross <[email protected]>
Acked-by: David Vrabel <[email protected]>
Use xenbus_read_unsigned() instead of xenbus_scanf() when possible.
This requires to change the type of the reads from int to unsigned,
but these cases have been wrong before: negative values are not allowed
for the modified cases.
Cc: [email protected]
Cc: [email protected]
Signed-off-by: Juergen Gross <[email protected]>
Acked-by: David Vrabel <[email protected]>
Use xenbus_read_unsigned() instead of xenbus_scanf() when possible.
Signed-off-by: Juergen Gross <[email protected]>
Acked-by: David Vrabel <[email protected]>
Use xenbus_read_unsigned() instead of xenbus_scanf() when possible.
This requires to change the type of the read from int to unsigned,
but this case has been wrong before: negative values are not allowed
for the modified case.
Cc: [email protected]
Cc: [email protected]
Signed-off-by: Juergen Gross <[email protected]>
Acked-by: Bjorn Helgaas <[email protected]>
Acked-by: David Vrabel <[email protected]>
Use xenbus_read_unsigned() instead of xenbus_scanf() when possible.
This requires to change the type of some reads from int to unsigned,
but these cases have been wrong before: negative values are not allowed
for the modified cases.
Cc: [email protected]
Signed-off-by: Juergen Gross <[email protected]>
Acked-by: David Vrabel <[email protected]>
Use xenbus_read_unsigned() instead of xenbus_scanf() when possible.
This requires to change the type of some reads from int to unsigned,
but these cases have been wrong before: negative values are not allowed
for the modified cases.
Cc: [email protected]
Cc: [email protected]
Cc: [email protected]
Signed-off-by: Juergen Gross <[email protected]>
Reviewed-by: Paul Durrant <[email protected]>
Acked-by: David Vrabel <[email protected]>
Use xenbus_read_unsigned() instead of xenbus_scanf() when possible.
This requires to change the type of the reads from int to unsigned,
but these cases have been wrong before: negative values are not allowed
for the modified cases.
Cc: [email protected]
Cc: [email protected]
Signed-off-by: Juergen Gross <[email protected]>
Acked-by: David Vrabel <[email protected]>
Use xenbus_read_unsigned() instead of xenbus_scanf() when possible.
This requires to change the type of some reads from int to unsigned,
but these cases have been wrong before: negative values are not allowed
for the modified cases.
Cc: [email protected]
Cc: [email protected]
Signed-off-by: Juergen Gross <[email protected]>
Acked-by: David Vrabel <[email protected]>
Use xenbus_read_unsigned() instead of xenbus_scanf() when possible.
This requires to change the type of one read from int to unsigned,
but this case has been wrong before: negative values are not allowed
for the modified case.
Cc: [email protected]
Cc: [email protected]
Signed-off-by: Juergen Gross <[email protected]>
Acked-by: David Vrabel <[email protected]>
There are multiple instances of code reading an optional unsigned
parameter from Xenstore via xenbus_scanf(). Instead of repeating the
same code over and over add a service function doing the job.
Signed-off-by: Juergen Gross <[email protected]>
Reviewed-by: David Vrabel <[email protected]>
The Xen docs specify several flags which a guest can set to advertise
which values of the xenstore control/shutdown key it will recognize.
This patch adds code to write all the relevant feature-flag keys.
Based-on-patch-by: Paul Durrant <[email protected]>
Signed-off-by: Juergen Gross <[email protected]>
Reviewed-by: David Vrabel <[email protected]>
Reviewed-by: Paul Durrant <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
When running as Xen dom0 a special processor_aggregator driver is
needed. Don't register the standard driver in this case.
Without that check an error message:
"Error: Driver 'processor_aggregator' is already registered,
aborting..."
will be displayed.
Signed-off-by: Juergen Gross <[email protected]>
[ rjw: Minor fixups ]
Signed-off-by: Rafael J. Wysocki <[email protected]>
Support the driver_override scheme introduced with commit 782a985d7a
("PCI: Introduce new device binding path using pci_dev.driver_override")
As pcistub_probe() is called for all devices (it has to check for a
match based on the slot address rather than device type) it has to
check for driver_override set to "pciback" itself.
Up to now for assigning a pci device to pciback you need something like:
echo 0000:07:10.0 > /sys/bus/pci/devices/0000\:07\:10.0/driver/unbind
echo 0000:07:10.0 > /sys/bus/pci/drivers/pciback/new_slot
echo 0000:07:10.0 > /sys/bus/pci/drivers_probe
while with the patch you can use the same mechanism as for similar
drivers like pci-stub and vfio-pci:
echo pciback > /sys/bus/pci/devices/0000\:07\:10.0/driver_override
echo 0000:07:10.0 > /sys/bus/pci/devices/0000\:07\:10.0/driver/unbind
echo 0000:07:10.0 > /sys/bus/pci/drivers_probe
So e.g. libvirt doesn't need special handling for pciback.
Signed-off-by: Juergen Gross <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
The Xen pciback driver has a list of all pci devices it is ready to
seize. There is no check whether a to be added entry already exists.
While this might be no problem in the common case it might confuse
those which consume the list via sysfs.
Modify the handling of this list by not adding an entry which already
exists. As this will be needed later split out the list handling into
a separate function.
Signed-off-by: Juergen Gross <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
The Xen pciback driver maintains a list of all its seized devices.
There are two functions searching the list for a specific device with
basically the same semantics just returning different structures in
case of a match.
Split out the search function.
Signed-off-by: Juergen Gross <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
Instead of open coding it use the threaded irq mechanism in
xen-netback.
Signed-off-by: Juergen Gross <[email protected]>
Acked-by: Wei Liu <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
The default for the Xen hypervisor is to not enable VPMU in order to
avoid security issues. In this case the Linux kernel will issue the
message "Could not initialize VPMU for cpu 0, error -95" which looks
more like an error than a normal state.
Change the message to something less scary in case the hypervisor
returns EOPNOTSUPP or ENOSYS when trying to activate VPMU.
Signed-off-by: Juergen Gross <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
There are two functions with name xen_pmu_init() in the kernel. Rename
the one in drivers/xen/sys-hypervisor.c to avoid shadowing the one in
arch/x86/xen/pmu.c
To avoid the same problem in future rename some more functions.
Signed-off-by: Juergen Gross <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
It is known that PMU isn't working in some virtualized environments.
Modify the message issued in that case to mention why hardware PMU
isn't usable instead of reporting it to be broken.
As a side effect this will correct a little bug in the error message:
The error message was meant to be either of level err or info
depending on the environment (native or virtualized). As the level is
taken from the format string and not the printed string, specifying
it via %s and a conditional argument didn't work the way intended.
Signed-off-by: Juergen Gross <[email protected]>
Cc: Alexander Shishkin <[email protected]>
Cc: Arnaldo Carvalho de Melo <[email protected]>
Cc: Jiri Olsa <[email protected]>
Cc: Linus Torvalds <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Ingo Molnar <[email protected]>
pv_time_ops might be overwritten with xen_time_ops after the
steal_clock operation has been initialized already. To prevent calling
a now uninitialized function pointer add the steal_clock static
initialization to xen_time_ops.
Signed-off-by: Juergen Gross <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
Up to now reading the stolen time of a remote cpu was not possible in a
performant way under Xen. This made support of runqueue steal time via
paravirt_steal_rq_enabled impossible.
With the addition of an appropriate hypervisor interface this is now
possible, so add the support.
Signed-off-by: Juergen Gross <[email protected]>
Reviewed-by: Stefano Stabellini <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
The pv_time_ops structure contains a function pointer for the
"steal_clock" functionality used only by KVM and Xen on ARM. Xen on x86
uses its own mechanism to account for the "stolen" time a thread wasn't
able to run due to hypervisor scheduling.
Add support in Xen arch independent time handling for this feature by
moving it out of the arm arch into drivers/xen and remove the x86 Xen
hack.
Signed-off-by: Juergen Gross <[email protected]>
Reviewed-by: Boris Ostrovsky <[email protected]>
Reviewed-by: Stefano Stabellini <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
xen_cleanhighmap() is operating on level2_kernel_pgt only. The upper
bound of the loop setting non-kernel-image entries to zero should not
exceed the size of level2_kernel_pgt.
Reported-by: Linus Torvalds <[email protected]>
Signed-off-by: Juergen Gross <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
Instead of having two functions for cycling through the E820 map in
order to count to be remapped pages and remap them later, just use one
function with a caller supplied sub-function called for each region to
be processed. This eliminates the possibility of a mismatch between
both loops which showed up in certain configurations.
Suggested-by: Ed Swierk <[email protected]>
Signed-off-by: Juergen Gross <[email protected]>
Cc: <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
When adding more than one LUN to a frontend a warning for a failed
assignment is issued in dom0 for each already existing LUN. Avoid this
warning by checking for a LUN already existing when existence is
allowed (scsiback_do_add_lun() called with try == 1).
As the LUN existence check is needed now for a third time, factor it
out into a function. This in turn leads to a more or less complete
rewrite of scsiback_del_translation_entry() which will now return a
proper error code in case of failure.
Signed-off-by: Juergen Gross <[email protected]>
Reviewed-by: Boris Ostrovsky <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
When adding a new frontend to xen-scsiback don't decrement the number
of active frontends in case of no error. Doing so results in a failure
when trying to remove the xen-pvscsi nexus even if no domain is using
it.
Signed-off-by: Juergen Gross <[email protected]>
Reviewed-by: Boris Ostrovsky <[email protected]>
Cc: [email protected]
Signed-off-by: David Vrabel <[email protected]>
Saving and restoring lapic vectors in lapic_suspend() and
lapic_resume() is not consistent: the thmr vector saving is
guarded by a different config option than the restore part. The
cmci vector isn't handled at all.
Those inconsistencies are not very critical, as the missing cmci
vector will be set via mce resume handling, the wrong config
option used for restoring the thmr vector can't be configured
differently than the one which should be used.
Nevertheless correct the thmr vector restore and add cmci vector
handling.
Signed-off-by: Juergen Gross <[email protected]>
Acked-by: Borislav Petkov <[email protected]>
Cc: Andy Lutomirski <[email protected]>
Cc: Borislav Petkov <[email protected]>
Cc: Brian Gerst <[email protected]>
Cc: Denys Vlasenko <[email protected]>
Cc: H. Peter Anvin <[email protected]>
Cc: Linus Torvalds <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
[ Minor code edits. ]
Signed-off-by: Ingo Molnar <[email protected]>
HYPERVISOR_memory_op() is defined to return an "int" value. This is
wrong, as the Xen hypervisor will return "long".
The sub-function XENMEM_maximum_reservation returns the maximum
number of pages for the current domain. An int will overflow for a
domain configured with 8TB of memory or more.
Correct this by using the correct type.
Signed-off-by: Juergen Gross <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
Instead of using physical addresses for accounting of extra memory
areas available for ballooning switch to pfns as this is much less
error prone regarding partial pages.
Reported-by: Roger Pau Monné <[email protected]>
Tested-by: Roger Pau Monné <[email protected]>
Signed-off-by: Juergen Gross <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
When a pv-domain (including dom0) is started it tries to size it's
p2m list according to the maximum possible memory amount it ever can
achieve. Limit the initial maximum memory size to the architectural
limit of the hardware in order to avoid overflows during remapping
of memory.
This problem will occur when dom0 is started with an initial memory
size being a multiple of 1GB, but without specifying it's maximum
memory size. The kernel must be configured without
CONFIG_XEN_BALLOON_MEMORY_HOTPLUG for the problem to happen.
Reported-by: Roger Pau Monné <[email protected]>
Tested-by: Roger Pau Monné <[email protected]>
Signed-off-by: Juergen Gross <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
Commit b1c9f169047b ("xen: split counting of extra memory pages...")
introduced an error when dom0 was started with limited memory occurring
only on some hardware.
The problem arises in case dom0 is started with initial memory and
maximum memory being the same. The kernel must be configured without
CONFIG_XEN_BALLOON_MEMORY_HOTPLUG for the problem to happen. If all
of this is true and the E820 map of the machine is sparse (some areas
are not covered) then the machine might crash early in the boot
process.
An example E820 map triggering the problem looks like this:
[ 0.000000] e820: BIOS-provided physical RAM map:
[ 0.000000] BIOS-e820: [mem 0x0000000000000000-0x000000000009d7ff] usable
[ 0.000000] BIOS-e820: [mem 0x000000000009d800-0x000000000009ffff] reserved
[ 0.000000] BIOS-e820: [mem 0x00000000000e0000-0x00000000000fffff] reserved
[ 0.000000] BIOS-e820: [mem 0x0000000000100000-0x00000000cf7fafff] usable
[ 0.000000] BIOS-e820: [mem 0x00000000cf7fb000-0x00000000cf95ffff] reserved
[ 0.000000] BIOS-e820: [mem 0x00000000cf960000-0x00000000cfb62fff] ACPI NVS
[ 0.000000] BIOS-e820: [mem 0x00000000cfb63000-0x00000000cfd14fff] usable
[ 0.000000] BIOS-e820: [mem 0x00000000cfd15000-0x00000000cfd61fff] ACPI NVS
[ 0.000000] BIOS-e820: [mem 0x00000000cfd62000-0x00000000cfd6cfff] ACPI data
[ 0.000000] BIOS-e820: [mem 0x00000000cfd6d000-0x00000000cfd6ffff] ACPI NVS
[ 0.000000] BIOS-e820: [mem 0x00000000cfd70000-0x00000000cfd70fff] usable
[ 0.000000] BIOS-e820: [mem 0x00000000cfd71000-0x00000000cfea8fff] reserved
[ 0.000000] BIOS-e820: [mem 0x00000000cfea9000-0x00000000cfeb9fff] ACPI NVS
[ 0.000000] BIOS-e820: [mem 0x00000000cfeba000-0x00000000cfecafff] reserved
[ 0.000000] BIOS-e820: [mem 0x00000000cfecb000-0x00000000cfecbfff] ACPI NVS
[ 0.000000] BIOS-e820: [mem 0x00000000cfecc000-0x00000000cfedbfff] reserved
[ 0.000000] BIOS-e820: [mem 0x00000000cfedc000-0x00000000cfedcfff] ACPI NVS
[ 0.000000] BIOS-e820: [mem 0x00000000cfedd000-0x00000000cfeddfff] reserved
[ 0.000000] BIOS-e820: [mem 0x00000000cfede000-0x00000000cfee3fff] ACPI NVS
[ 0.000000] BIOS-e820: [mem 0x00000000cfee4000-0x00000000cfef6fff] reserved
[ 0.000000] BIOS-e820: [mem 0x00000000cfef7000-0x00000000cfefffff] usable
[ 0.000000] BIOS-e820: [mem 0x00000000e0000000-0x00000000efffffff] reserved
[ 0.000000] BIOS-e820: [mem 0x00000000fec00000-0x00000000fec00fff] reserved
[ 0.000000] BIOS-e820: [mem 0x00000000fec10000-0x00000000fec10fff] reserved
[ 0.000000] BIOS-e820: [mem 0x00000000fed00000-0x00000000fed00fff] reserved
[ 0.000000] BIOS-e820: [mem 0x00000000fed40000-0x00000000fed44fff] reserved
[ 0.000000] BIOS-e820: [mem 0x00000000fed61000-0x00000000fed70fff] reserved
[ 0.000000] BIOS-e820: [mem 0x00000000fed80000-0x00000000fed8ffff] reserved
[ 0.000000] BIOS-e820: [mem 0x00000000ff000000-0x00000000ffffffff] reserved
[ 0.000000] BIOS-e820: [mem 0x0000000100001000-0x000000020effffff] usable
In this case the area a0000-dffff isn't present in the map. This will
confuse the memory setup of the domain when remapping the memory from
such holes to populated areas.
To avoid the problem the accounting of to be remapped memory has to
count such holes in the E820 map as well.
Reported-by: Boris Ostrovsky <[email protected]>
Signed-off-by: Juergen Gross <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
Commit b1c9f169047b ("xen: split counting of extra memory pages...")
introduced an error when dom0 was started with limited memory.
The problem arises in case dom0 is started with initial memory and
maximum memory being the same and exactly a multiple of 1 GB. The
kernel must be configured without CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
for the problem to happen. In this case it will crash very early
during boot due to the virtual mapped p2m list not being large
enough to be able to remap any memory:
(XEN) Freed 304kB init memory.
mapping kernel into physical memory
about to get started...
(XEN) traps.c:459:d0v0 Unhandled invalid opcode fault/trap [#6] on VCPU 0 [ec=0000]
(XEN) domain_crash_sync called from entry.S: fault at ffff82d080229a93 create_bounce_frame+0x12b/0x13a
(XEN) Domain 0 (vcpu#0) crashed on cpu#0:
(XEN) ----[ Xen-4.5.2-pre x86_64 debug=n Not tainted ]----
(XEN) CPU: 0
(XEN) RIP: e033:[<ffffffff81d120cb>]
(XEN) RFLAGS: 0000000000000206 EM: 1 CONTEXT: pv guest (d0v0)
(XEN) rax: ffffffff81db2000 rbx: 000000004d000000 rcx: 0000000000000000
(XEN) rdx: 000000004d000000 rsi: 0000000000063000 rdi: 000000004d063000
(XEN) rbp: ffffffff81c03d78 rsp: ffffffff81c03d28 r8: 0000000000023000
(XEN) r9: 00000001040ff000 r10: 0000000000007ff0 r11: 0000000000000000
(XEN) r12: 0000000000063000 r13: 000000000004d000 r14: 0000000000000063
(XEN) r15: 0000000000000063 cr0: 0000000080050033 cr4: 00000000000006f0
(XEN) cr3: 0000000105c0f000 cr2: ffffc90000268000
(XEN) ds: 0000 es: 0000 fs: 0000 gs: 0000 ss: e02b cs: e033
(XEN) Guest stack trace from rsp=ffffffff81c03d28:
(XEN) 0000000000000000 0000000000000000 ffffffff81d120cb 000000010000e030
(XEN) 0000000000010006 ffffffff81c03d68 000000000000e02b ffffffffffffffff
(XEN) 0000000000000063 000000000004d063 ffffffff81c03de8 ffffffff81d130a7
(XEN) ffffffff81c03de8 000000000004d000 00000001040ff000 0000000000105db1
(XEN) 00000001040ff001 000000000004d062 ffff8800092d6ff8 0000000002027000
(XEN) ffff8800094d8340 ffff8800092d6ff8 00003ffffffff000 ffff8800092d7ff8
(XEN) ffffffff81c03e48 ffffffff81d13c43 ffff8800094d8000 ffff8800094d9000
(XEN) 0000000000000000 ffff8800092d6000 00000000092d6000 000000004cfbf000
(XEN) 00000000092d6000 00000000052d5442 0000000000000000 0000000000000000
(XEN) ffffffff81c03ed8 ffffffff81d185c1 0000000000000000 0000000000000000
(XEN) ffffffff81c03e78 ffffffff810f8ca4 ffffffff81c03ed8 ffffffff8171a15d
(XEN) 0000000000000010 ffffffff81c03ee8 0000000000000000 0000000000000000
(XEN) ffffffff81f0e402 ffffffffffffffff ffffffff81dae900 0000000000000000
(XEN) 0000000000000000 0000000000000000 ffffffff81c03f28 ffffffff81d0cf0f
(XEN) 0000000000000000 0000000000000000 0000000000000000 ffffffff81db82e0
(XEN) 0000000000000000 0000000000000000 0000000000000000 0000000000000000
(XEN) ffffffff81c03f38 ffffffff81d0c603 ffffffff81c03ff8 ffffffff81d11c86
(XEN) 0300000100000032 0000000000000005 0000000000000020 0000000000000000
(XEN) 0000000000000000 0000000000000000 0000000000000000 0000000000000000
(XEN) 0000000000000000 0000000000000000 0000000000000000 0000000000000000
(XEN) Domain 0 crashed: rebooting machine in 5 seconds.
This can be avoided by allocating aneough space for the p2m to cover
the maximum memory of dom0 plus the identity mapped holes required
for PCI space, BIOS etc.
Reported-by: Boris Ostrovsky <[email protected]>
Signed-off-by: Juergen Gross <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
Cleanup by removing arch/x86/xen/p2m.h as it isn't needed any more.
Most definitions in this file are used in p2m.c only. Move those into
p2m.c.
set_phys_range_identity() is already declared in
arch/x86/include/asm/xen/page.h, add __init annotation there.
MAX_REMAP_RANGES isn't used at all, just delete it.
The only define left is P2M_PER_PAGE which is moved to page.h as well.
Signed-off-by: Juergen Gross <[email protected]>
Acked-by: Konrad Rzeszutek Wilk <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
64 bit pv-domains under Xen are limited to 512 GB of RAM today. The
main reason has been the 3 level p2m tree, which was replaced by the
virtual mapped linear p2m list. Parallel to the p2m list which is
being used by the kernel itself there is a 3 level mfn tree for usage
by the Xen tools and eventually for crash dump analysis. For this tree
the linear p2m list can serve as a replacement, too. As the kernel
can't know whether the tools are capable of dealing with the p2m list
instead of the mfn tree, the limit of 512 GB can't be dropped in all
cases.
This patch replaces the hard limit by a kernel parameter which tells
the kernel to obey the 512 GB limit or not. The default is selected by
a configuration parameter which specifies whether the 512 GB limit
should be active per default for domUs (domain save/restore/migration
and crash dump analysis are affected).
Memory above the domain limit is returned to the hypervisor instead of
being identity mapped, which was wrong anyway.
The kernel configuration parameter to specify the maximum size of a
domain can be deleted, as it is not relevant any more.
Signed-off-by: Juergen Gross <[email protected]>
Acked-by: Konrad Rzeszutek Wilk <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
Check whether the hypervisor supplied p2m list is placed at a location
which is conflicting with the target E820 map. If this is the case
relocate it to a new area unused up to now and compliant to the E820
map.
As the p2m list might by huge (up to several GB) and is required to be
mapped virtually, set up a temporary mapping for the copied list.
For pvh domains just delete the p2m related information from start
info instead of reserving the p2m memory, as we don't need it at all.
For 32 bit kernels adjust the memblock_reserve() parameters in order
to cover the page tables only. This requires to memblock_reserve() the
start_info page on it's own.
Signed-off-by: Juergen Gross <[email protected]>
Acked-by: Konrad Rzeszutek Wilk <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
Some special pages containing interfaces to xen are being reserved
implicitly only today. The memblock_reserve() call to reserve them is
meant to reserve the p2m list supplied by xen. It is just reserving
not only the p2m list itself, but some more pages up to the start of
the xen built page tables.
To be able to move the p2m list to another pfn range, which is needed
for support of huge RAM, this memblock_reserve() must be split up to
cover all affected reserved pages explicitly.
The affected pages are:
- start_info page
- xenstore ring (might be missing, mfn is 0 in this case)
- console ring (not for initial domain)
Signed-off-by: Juergen Gross <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
During early boot as Xen pv domain the kernel needs to map some page
tables supplied by the hypervisor read only. This is needed to be
able to relocate some data structures conflicting with the physical
memory map especially on systems with huge RAM (above 512GB).
Provide the function early_memremap_ro() to provide this read only
mapping.
Signed-off-by: Juergen Gross <[email protected]>
Acked-by: Konrad Rzeszutek Wilk <[email protected]>
Acked-by: Vlastimil Babka <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
Check whether the initrd is placed at a location which is conflicting
with the target E820 map. If this is the case relocate it to a new
area unused up to now and compliant to the E820 map.
Signed-off-by: Juergen Gross <[email protected]>
Reviewed-by: David Vrabel <[email protected]>
Acked-by: Konrad Rzeszutek Wilk <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
Check whether the page tables built by the domain builder are at
memory addresses which are in conflict with the target memory map.
If this is the case just panic instead of running into problems
later.
Signed-off-by: Juergen Gross <[email protected]>
Acked-by: Konrad Rzeszutek Wilk <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
Checks whether the pre-allocated memory of the loaded kernel is in
conflict with the target memory map. If this is the case, just panic
instead of run into problems later, as there is nothing we can do
to repair this situation.
Signed-off-by: Juergen Gross <[email protected]>
Reviewed-by: David Vrabel <[email protected]>
Acked-by: Konrad Rzeszutek Wilk <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
For being able to relocate pre-allocated data areas like initrd or
p2m list it is mandatory to find a contiguous memory area which is
not yet in use and doesn't conflict with the memory map we want to
be in effect.
In case such an area is found reserve it at once as this will be
required to be done in any case.
Signed-off-by: Juergen Gross <[email protected]>
Reviewed-by: David Vrabel <[email protected]>
Acked-by: Konrad Rzeszutek Wilk <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
Provide a service routine to check a physical memory area against the
E820 map. The routine will return false if the complete area is RAM
according to the E820 map and true otherwise.
Signed-off-by: Juergen Gross <[email protected]>
Reviewed-by: David Vrabel <[email protected]>
Acked-by: Konrad Rzeszutek Wilk <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
Memory pages in the initial memory setup done by the Xen hypervisor
conflicting with the target E820 map are remapped. In order to do this
those pages are counted and remapped in xen_set_identity_and_remap().
Split the counting from the remapping operation to be able to setup
the needed memory sizes in time but doing the remap operation at a
later time. This enables us to simplify the interface to
xen_set_identity_and_remap() as the number of remapped and released
pages is no longer needed here.
Finally move the remapping further down to prepare relocating
conflicting memory contents before the memory might be clobbered by
xen_set_identity_and_remap(). This requires to not destroy the Xen
E820 map when the one for the system is being constructed.
Signed-off-by: Juergen Gross <[email protected]>
Acked-by: Konrad Rzeszutek Wilk <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
Instead of using a function local static e820 map in xen_memory_setup()
and calling various functions in the same source with the map as a
parameter use a map directly accessible by all functions in the source.
Signed-off-by: Juergen Gross <[email protected]>
Reviewed-by: David Vrabel <[email protected]>
Acked-by: Konrad Rzeszutek Wilk <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
Direct Xen to place the initial P->M table outside of the initial
mapping, as otherwise the 1G (implementation) / 2G (theoretical)
restriction on the size of the initial mapping limits the amount
of memory a domain can be handed initially.
As the initial P->M table is copied rather early during boot to
domain private memory and it's initial virtual mapping is dropped,
the easiest way to avoid virtual address conflicts with other
addresses in the kernel is to use a user address area for the
virtual address of the initial P->M table. This allows us to just
throw away the page tables of the initial mapping after the copy
without having to care about address invalidation.
It should be noted that this patch won't enable a pv-domain to USE
more than 512 GB of RAM. It just enables it to be started with a
P->M table covering more memory. This is especially important for
being able to boot a Dom0 on a system with more than 512 GB memory.
Signed-off-by: Juergen Gross <[email protected]>
Based-on-patch-by: Jan Beulich <[email protected]>
Acked-by: Konrad Rzeszutek Wilk <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
In case the Xen tools indicate they don't need the p2m 3 level tree
as they support the virtual mapped linear p2m list, just omit building
the tree.
Signed-off-by: Juergen Gross <[email protected]>
Reviewed-by: David Vrabel <[email protected]>
Acked-by: Konrad Rzeszutek Wilk <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
The virtual address of the linear p2m list should be stored in the
shared info structure read by the Xen tools to be able to support
64 bit pv-domains larger than 512 GB. Additionally the linear p2m
list interface includes a generation count which is changed prior
to and after each mapping change of the p2m list. Reading the
generation count the Xen tools can detect changes of the mappings
and re-read the p2m list eventually.
Signed-off-by: Juergen Gross <[email protected]>
Reviewed-by: David Vrabel <[email protected]>
Acked-by: Konrad Rzeszutek Wilk <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
When dom0 is being ballooned balloon_process() will hold the balloon
mutex until it is finished. This will block e.g. creation of new
domains as the device backends for the new domain need some
autoballooned pages for the ring buffers.
Avoid this by releasing the balloon mutex from time to time during
ballooning. Adjust the comment above balloon_process() regarding
multiple instances of balloon_process().
Instead of open coding it, just use cond_resched().
Signed-off-by: Juergen Gross <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
Commit 25b884a83d ("x86/xen: set
regions above the end of RAM as 1:1") introduced a regression.
To be able to add memory pages which were added via memory hotplug to
a pv domain, the pages must be "invalid" instead of "identity" in the
p2m list before they can be added.
Suggested-by: David Vrabel <[email protected]>
Signed-off-by: Juergen Gross <[email protected]>
Cc: <[email protected]> # 3.16+
Reviewed-by: Daniel Kiper <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
Commit 054954eb05 ("xen: switch to linear
virtual mapped sparse p2m list") introduced a regression regarding to
memory hotplug for a pv-domain: as the virtual space for the p2m list
is allocated for the to be expected memory size of the domain only,
hotplugged memory above that size will not be usable by the domain.
Correct this by using a configurable size for the p2m list in case of
memory hotplug enabled (default supported memory size is 512 GB for
64 bit domains and 4 GB for 32 bit domains).
Signed-off-by: Juergen Gross <[email protected]>
Cc: <[email protected]> # 3.19+
Reviewed-by: Daniel Kiper <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
Up to now the pvscsi frontend hasn't supported domain suspend and
resume. When a domain with an assigned pvscsi device was suspended
and resumed again, it was not able to use the device any more: trying
to do so resulted in hanging processes.
Support suspend and resume of pvscsi devices.
Signed-off-by: Juergen Gross <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
When a xen domain is being restored the LUN state of a pvscsi device
is "Connected" and not "Initialising" as in case of attaching a new
pvscsi LUN.
This must be taken into account when adding a new pvscsi device for
a domain as otherwise the pvscsi LUN won't be connected to the
SCSI target associated with it.
Signed-off-by: Juergen Gross <[email protected]>
Reviewed-by: Boris Ostrovsky <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
Instead of manually list each hypercall in arch/x86/xen/xen-head.S
use the auto generated symbol list.
This also corrects the wrong address of xen_hypercall_mca which was
located 32 bytes higher than it should.
Symbol addresses have been verified to match the correct ones via
objdump output.
Based-on-patch-by: Jan Beulich <[email protected]>
Signed-off-by: Juergen Gross <[email protected]>
Reviewed-by: David Vrabel <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
Instead of manually list all hypervisor calls in arch/x86/xen/trace.c
use the auto generated list.
Signed-off-by: Juergen Gross <[email protected]>
Reviewed-by: David Vrabel <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
The header include/xen/interface/xen.h doesn't contain all definitions
from Xen's version of that header. Update it accordingly.
Signed-off-by: Juergen Gross <[email protected]>
Reviewed-by: David Vrabel <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
Today there are several places in the kernel which build tables
containing one entry for each possible Xen hypercall. Create an
infrastructure to be able to generate these tables at build time.
Based-on-patch-by: Jan Beulich <[email protected]>
Signed-off-by: Juergen Gross <[email protected]>
Reviewed-by: David Vrabel <[email protected]>
Acked-by: Ingo Molnar <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
Using the pvops kernel a NULL pointer dereference was detected on a
large machine (144 processors) when booting as dom0 in
evtchn_fifo_unmask() during assignment of a pirq.
The event channel in question was the first to need a new entry in
event_array[] in events_fifo.c. Unfortunately xen_irq_info_pirq_setup()
is called with evtchn being 0 for a new pirq and the real event channel
number is assigned to the pirq only during __startup_pirq().
It is mandatory to call xen_evtchn_port_setup() after assigning the
event channel number to the pirq to make sure all memory needed for the
event channel is allocated.
Signed-off-by: Juergen Gross <[email protected]>
Cc: <[email protected]> # 3.14+
Signed-off-by: David Vrabel <[email protected]>
Commit 054954eb05 ("xen: switch to
linear virtual mapped sparse p2m list") introduced an error.
During initialization of the p2m list a p2m identity area mapped by
a complete identity pmd entry has to be split up into smaller chunks
sometimes, if a non-identity pfn is introduced in this area.
If this non-identity pfn is not at index 0 of a p2m page the new
p2m page needed is initialized with wrong identity entries, as the
identity pfns don't start with the value corresponding to index 0,
but with the initial non-identity pfn. This results in weird wrong
mappings.
Correct the wrong initialization by starting with the correct pfn.
Cc: [email protected] # 3.19
Reported-by: Stefan Bader <[email protected]>
Signed-off-by: Juergen Gross <[email protected]>
Tested-by: Stefan Bader <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
A request in the ring buffer mustn't be read after it has been marked
as consumed. Otherwise it might already have been reused by the
frontend without violating the ring protocol.
To avoid inconsistencies in the backend only work on a private copy
of the request. This will ensure a malicious guest not being able to
bypass consistency checks of the backend by modifying an active
request.
Signed-off-by: Juergen Gross <[email protected]>
Cc: <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
Especially on 32 bit kernels memory node ranges are printed with 32 bit
wide addresses only. Use u64 types and %llx specifiers to print full
width of addresses.
Signed-off-by: Juergen Gross <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
The file arch/x86/xen/mmu.c has some functions that can be annotated
with "__init".
Signed-off-by: Juergen Gross <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
Some more functions in arch/x86/xen/setup.c can be made "__init".
xen_ignore_unusable() can be made "static".
Signed-off-by: Juergen Gross <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
In many places in arch/x86/xen/setup.c wrong types are used for
physical addresses (u64 or unsigned long long). Use phys_addr_t
instead.
Use macros already defined instead of open coding them.
Correct some other type mismatches.
Signed-off-by: Juergen Gross <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
Remove extern declarations in arch/x86/xen/setup.c which are either
not used or redundant. Move needed other extern declarations to
xen-ops.h
Signed-off-by: Juergen Gross <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
Commit 281d4078be ("x86: Make page cache mode a real type")
introduced the symbols __cachemode2pte_tbl and __pte2cachemode_tbl and
exported them via EXPORT_SYMBOL_GPL. The exports are part of a
replacement of code which has been EXPORT_SYMBOL before these changes
resulting in build breakage of out-of-tree non-gpl modules.
Change EXPORT_SYMBOL_GPL to EXPORT-SYMBOL for these two symbols.
Fixes: 281d4078be "x86: Make page cache mode a real type"
Reported-and-tested-by: Steven Noonan <[email protected]>
Signed-off-by: Juergen Gross <[email protected]>
Reviewed-by: Toshi Kani <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Thomas Gleixner <[email protected]>
VMWare seems not to emulate the PAT MSR correctly: reaeding
MSR_IA32_CR_PAT returns 0 even after writing another value to it.
Commit bd809af16e triggers this VMWare bug when the kernel is
booted as a VMWare guest.
Detect this bug and don't use the read value if it is 0.
Fixes: bd809af16e "x86: Enable PAT to use cache mode translation tables"
Reported-and-tested-by: Jongman Heo <[email protected]>
Acked-by: Alok N Kataria <[email protected]>
Signed-off-by: Juergen Gross <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Thomas Gleixner <[email protected]>
With the introduction of the linear mapped p2m list setting memory
areas to "invalid" had to be delayed. When doing the invalidation
make sure no zero sized areas are processed.
Signed-off-by: Juegren Gross <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
When converting a pfn to a physical address be sure to use 64 bit
wide types or convert the physical address to a pfn if possible.
Signed-off-by: Juergen Gross <[email protected]>
Tested-by: Boris Ostrovsky <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
When allocating a new pmd for the linear mapped p2m list a check is
done for not introducing another pmd when this just happened on
another cpu. In this case the old pte pointer was returned which
points to the p2m_missing or p2m_identity page. The correct value
would be the pointer to the found new page.
Signed-off-by: Juergen Gross <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
In xen_rebuild_p2m_list() for large areas of invalid or identity
mapped memory the pmd entries on 32 bit systems are initialized
wrong. Correct this error.
Suggested-by: Boris Ostrovsky <[email protected]>
Signed-off-by: Juergen Gross <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
Juergen Gross