If dom0_vcpus_pin is used, make sure the pinning is only done for
dom0 vcpus, instead of for the hardware domain (which might not be
dom0 at all!).
Suggested-by: Jan Beulich <jbeulich@suse.com>
Signed-off-by: Dario Faggioli <dfaggioli@suse.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
Trying to include .ocamldep.make will cause it to be generated if it
doesn't exist.
We do not want this during make clean: we would remove it anyway.
Speeds up make clean.
Before (measured on f732240fd3, July 2022):
```
Parsing /home/edwin/xen2/tools/ocaml/libs/xl/../../../../tools/libs/light/libxl_types.idl
Parsing /home/edwin/xen2/tools/ocaml/libs/xl/../../../../tools/libs/light/libxl_types.idl
Parsing /home/edwin/xen2/tools/ocaml/libs/xl/../../../../tools/libs/light/libxl_types.idl
Parsing /home/edwin/xen2/tools/ocaml/libs/xl/../../../../tools/libs/light/libxl_types.idl
Parsing /home/edwin/xen2/tools/ocaml/libs/xl/../../../../tools/libs/light/libxl_types.idl
Performance counter stats for 'make clean -j8 -s' (5 runs):
4.2233 +- 0.0208 seconds time elapsed ( +- 0.49% )
```
After:
```
perf stat -r 5 --null make clean -j8 -s
Performance counter stats for 'make clean -j8 -s' (5 runs):
2.7325 +- 0.0138 seconds time elapsed ( +- 0.51% )
```
No functional change.
Signed-off-by: Edwin Török <edvin.torok@citrix.com>
Acked-by: Christian Lindig <christian.lindig@citrix.com>
Especially for the use in evtchn_move_pirqs() (called when moving a vCPU
across pCPU-s) and the ones in EOI handling in PCI pass-through code,
serializing perhaps an entire domain isn't helpful when no state (which
isn't e.g. further protected by the per-channel lock) changes.
Unfortunately this implies dropping of lock profiling for this lock,
until r/w locks may get enabled for such functionality.
While ->notify_vcpu_id is now meant to be consistently updated with the
per-channel lock held, an extension applies to ECS_PIRQ: The field is
also guaranteed to not change with the per-domain event lock held for
writing. Therefore the link_pirq_port() call from evtchn_bind_pirq()
could in principle be moved out of the per-channel locked regions, but
this further code churn didn't seem worth it.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Julien Grall <jgrall@amazon.com>
Acked-by: Daniel P. Smith <dpsmith@apertussolutions.com>
When vGIC performs irouter registers emulation, to get the target vCPU
via virq conveniently, Xen doesn't store the irouter value directly,
instead it will use the value (affinities) in irouter to calculate the
target vCPU, and then save the target vCPU in irq rank->vcpu[offset].
When vGIC tries to get the target vCPU, it first calculates the target
vCPU index via
int target = read_atomic(&rank->vcpu[virq & INTERRUPT_RANK_MASK]);
and then it gets the target vCPU via
v->domain->vcpu[target];
When vGIC tries to store irouter for one virq, the target vCPU index
in the rank is computed as
offset &= virq & INTERRUPT_RANK_MASK;
finally it gets the target vCPU via
d->vcpu[read_atomic(&rank->vcpu[offset])];
There is a difference between them while getting the target vCPU index
in the rank. Actually (virq & INTERRUPT_RANK_MASK) would already get
the target vCPU index in the rank, it's wrong to add '&' before '=' when
calculate the offset.
For example, the target vCPU index in the rank should be 6 for virq 38,
but vGIC will get offset=0 when vGIC stores the irouter for this virq,
and finally vGIC will access the wrong target vCPU index in the rank
when updating the irouter.
Fixes: 5d495f4349 ("xen/arm: vgic: Optimize the way to store the target vCPU in the rank")
Signed-off-by: Hongda Deng <Hongda.Deng@arm.com>
Reviewed-by: Julien Grall <jgrall@amazon.com>
Prevent header file from being included more than once by adding ifndef guard.
In order to be close to gnu-efi code
- for x86_64, use the same guard
- for arm64, that there is no guard in gnu-efi, for consistency,
use a similar format and position to the x86_64 guard
Signed-off-by: Xenia Ragiadakou <burzalodowa@gmail.com>
Acked-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Bertrand Marquis <bertrand.marquis@arm.com>
Qemu VM is configured with 2 cpus but the device tree passed has only 1.
Generate a device tree with 2 cpus.
Signed-off-by: Xenia Ragiadakou <burzalodowa@gmail.com>
Reviewed-by: Stefano Stabellini <sstabellini@kernel.org>
Rename the device tree from virt-gicv3 to virt-gicv2 to avoid confusion
since the version of the generic interrupt controller used for this test
is the v2 and not the v3.
Signed-off-by: Xenia Ragiadakou <burzalodowa@gmail.com>
Reviewed-by: Stefano Stabellini <sstabellini@kernel.org>
Remove comment "# Install QEMU" because qemu is not installed, it is taken
from a test-artifacts container.
Change comment "# Busybox Dom0" to "# Busybox" because busybox is not used
only for the Dom0 but also for the DomU.
Signed-off-by: Xenia Ragiadakou <burzalodowa@gmail.com>
Reviewed-by: Stefano Stabellini <sstabellini@kernel.org>
The function idle_loop() is referenced only in domain.c.
Change its linkage from external to internal by adding the storage-class
specifier static to its definitions.
Add the function as a 'fake' input operand to the inline assembly statement,
to make the compiler aware that the function is used.
Fake means that the function is not actually used as an operand by the asm code.
That is because there is not a suitable gcc arm32 asm constraint for labels.
Declare return_to_new_vcpu32() and return_to_new_vcpu64() that are also
referenced by this inline asm statement.
Also, this patch resolves indirectly a MISRA C 2012 Rule 8.4 violation warning.
Signed-off-by: Xenia Ragiadakou <burzalodowa@gmail.com>
[add noreturn]
Signed-off-by: Stefano Stabellini <stefano.stabellini@amd.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Bertrand Marquis <bertrand.marquis@arm.com>
At the moment, xen_second is used to cover the first 2GB of the
virtual address space. With the recent rework of the page-tables,
only the first 1GB region (where Xen resides) is effectively used.
In addition to that, I would like to reshuffle the memory layout.
So Xen mappings may not be anymore in the first 2GB of the virtual
address space.
Therefore, rework xen_second so it only covers the 1GB region where
Xen will reside.
With this change, xen_second doesn't cover anymore the xenheap area
on arm32. So, we first need to add memory to the boot allocator before
setting up the xenheap mappings.
Take the opportunity to update the comments on top of xen_fixmap and
xen_xenmap.
Signed-off-by: Julien Grall <jgrall@amazon.com>
Reviewed-by: Bertrand Marquis <bertrand.marquis@arm.com>
The file xen/arch/mm.c has been growing quite a lot. It now contains
various independent part of the MM subsytem.
One of them is the helpers to map/unmap a page which is only used
by arm32 and protected by CONFIG_ARCH_MAP_DOMAIN_PAGE. Move them in a
new file xen/arch/arm/domain_page.c.
Signed-off-by: Julien Grall <jgrall@amazon.com>
Reviewed-by: Bertrand Marquis <bertrand.marquis@arm.com>
move it to Kconfig.
The define CONFIG_DOMAIN_PAGE indicates whether the architecture provide
helpers to map/unmap a domain page. Rename it to CONFIG_ARCH_MAP_DOMAIN_PAGE
so it is clearer that support for domain page is not something that
can be disabled in Xen.
Take the opportunity to move CONFIG_MAP_DOMAIN_PAGE to Kconfig as this
will soon be necessary to use it in the Makefile.
Signed-off-by: Julien Grall <jgrall@amazon.com>
Reviewed-by: Bertrand Marquis <bertrand.marquis@arm.com> #arm part
At the moment, the domheap mappings initialization is done separately for
the boot CPU and secondary CPUs. The main difference is for the former
the pages are part of Xen binary whilst for the latter they are
dynamically allocated.
It would be good to have a single helper so it is easier to rework
on the domheap is initialized.
For CPU0, we still need to use pre-allocated pages because the
allocators may use domain_map_page(), so we need to have the domheap
area ready first. But we can still delay the initialization to setup_mm().
Introduce a new helper init_domheap_mappings() that will be called
from setup_mm() for the boot CPU and from init_secondary_pagetables()
for secondary CPUs.
Signed-off-by: Julien Grall <jgrall@amazon.com>
Reviewed-by: Bertrand Marquis <bertrand.marquis@arm.com>
Reviewed-by: Luca Fancellu <luca.fancellu@arm.com>
Tested-by: Luca Fancellu <luca.fancellu@arm.com>
At the moment, *_VIRT_END may either point to the address after the end
or the last address of the region.
The lack of consistency make quite difficult to reason with them.
Furthermore, there is a risk of overflow in the case where the address
points past to the end. I am not aware of any cases, so this is only a
latent bug.
Start to solve the problem by removing all the *_VIRT_END exclusively used
by the Arm code and add *_VIRT_SIZE when it is not present.
Take the opportunity to rename BOOT_FDT_SLOT_SIZE to BOOT_FDT_VIRT_SIZE
for better consistency and use _AT(vaddr_t, ).
Also take the opportunity to fix the coding style of the comment touched
in mm.c.
Signed-off-by: Julien Grall <jgrall@amazon.com>
Reviewed-by: Bertrand Marquis <bertrand.marquis@arm.com>
Reviewed-by: Luca Fancellu <luca.fancellu@arm.com>
Tested-By: Luca Fancellu <luca.fancellu@arm.com>
The macro parameter 'p' is used as an expression and needs to be enclosed in
parentheses.
Signed-off-by: Xenia Ragiadakou <burzalodowa@gmail.com>
Reviewed-by: Stefano Stabellini <sstabellini@kernel.org>
Now that we're not building multi.c anymore for 2 and 3 guest levels
when !HVM, there's no point in having these conditionals anymore. (As
somewhat a special case, the last of the removed conditionals really
builds on shadow_mode_external() always returning false when !HVM.) This
way the code becomes a tiny bit more readable.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Acked-by: Andrew Cooper <andrew.cooper3@citrix.com>
Let's use the existing inline wrapper instead of repeating respective
commentary at every site.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Acked-by: Andrew Cooper <andrew.cooper3@citrix.com>
In my (debug) build this amounts to well over 500 bytes of dead code.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Acked-by: Andrew Cooper <andrew.cooper3@citrix.com>
We should not blindly (in a release build) insert the new entry in the
hash if a reference to the guest page cannot be obtained, or else an
excess reference would be put when removing the hash entry again. Crash
the domain in that case instead. The sole caller doesn't further care
about the state of the guest page: All it does is return the
corresponding shadow page (which was obtained successfully before) to
its caller.
To compensate we further need to adjust hash removal: Since the shadow
page already has had its backlink set, domain cleanup code would try to
destroy the shadow, and hence still cause a put_page() without
corresponding get_page(). Leverage that the failed get_page() leads to
no hash insertion, making shadow_hash_delete() no longer assume it will
find the requested entry. Instead return back whether the entry was
found. This way delete_shadow_status() can avoid calling put_page() in
the problem scenario.
For the other caller of shadow_hash_delete() simply reinstate the
otherwise dropped assertion at the call site.
While touching the conditionals in {set,delete}_shadow_status() anyway,
also switch around their two pre-existing parts, to have the cheap one
first (frequently allowing to avoid evaluation of the expensive - due to
evaluate_nospec() - one altogether).
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
Enable CONFIG_STATIC_MEMORY in the existing arm64 build.
Create a new test job, called qemu-smoke-arm64-gcc-staticmem.
Adjust qemu-smoke-arm64.sh script to accomodate the static memory test as a
new test variant. The test variant is determined based on the first argument
passed to the script. For testing static memory, the argument is 'static-mem'.
The test configures DOM1 with a static memory region and adds a check in the
init script.
The check consists in comparing the contents of the /proc/device-tree
memory entry with the static memory range with which DOM1 was configured.
If the memory layout is correct, a message gets printed by DOM1.
At the end of the qemu run, the script searches for the specific message
in the logs and fails if not found.
Signed-off-by: Xenia Ragiadakou <burzalodowa@gmail.com>
Signed-off-by: Stefano Stabellini <stefano.stabellini@amd.com>
Reviewed-by: Penny Zheng <penny.zheng@arm.com>
Reviewed-by: Stefano Stabellini <sstabellini@kernel.org>
The EXPERT config option cannot anymore be selected via the environmental
variable XEN_CONFIG_EXPERT. Remove stale references to XEN_CONFIG_EXPERT
from the automation code.
Signed-off-by: Xenia Ragiadakou <burzalodowa@gmail.com>
Reviewed-by: Stefano Stabellini <sstabellini@kernel.org>
Reuse generic IOMMU device tree bindings to communicate Xen specific
information for the virtio devices for which the restricted memory
access using Xen grant mappings need to be enabled.
Insert "iommus" property pointed to the IOMMU node with "xen,grant-dma"
compatible to all virtio devices which backends are going to run in
non-hardware domains (which are non-trusted by default).
Based on device-tree binding from Linux:
Documentation/devicetree/bindings/iommu/xen,grant-dma.yaml
The example of generated nodes:
xen_iommu {
compatible = "xen,grant-dma";
#iommu-cells = <0x01>;
phandle = <0xfde9>;
};
virtio@2000000 {
compatible = "virtio,mmio";
reg = <0x00 0x2000000 0x00 0x200>;
interrupts = <0x00 0x01 0xf01>;
interrupt-parent = <0xfde8>;
dma-coherent;
iommus = <0xfde9 0x01>;
};
virtio@2000200 {
compatible = "virtio,mmio";
reg = <0x00 0x2000200 0x00 0x200>;
interrupts = <0x00 0x02 0xf01>;
interrupt-parent = <0xfde8>;
dma-coherent;
iommus = <0xfde9 0x01>;
};
Signed-off-by: Oleksandr Tyshchenko <oleksandr_tyshchenko@epam.com>
Reviewed-by: Stefano Stabellini <sstabellini@kernel.org>
Reviewed-by: Anthony PERARD <anthony.perard@citrix.com>
This patch introduces helpers to allocate Virtio MMIO params
(IRQ and memory region) and create specific device node in
the Guest device-tree with allocated params. In order to deal
with multiple Virtio devices, reserve corresponding ranges.
For now, we reserve 1MB for memory regions and 10 SPIs.
As these helpers should be used for every Virtio device attached
to the Guest, call them for Virtio disk(s).
Please note, with statically allocated Virtio IRQs there is
a risk of a clash with a physical IRQs of passthrough devices.
For the first version, it's fine, but we should consider allocating
the Virtio IRQs automatically. Thankfully, we know in advance which
IRQs will be used for passthrough to be able to choose non-clashed
ones.
Signed-off-by: Julien Grall <julien.grall@arm.com>
Signed-off-by: Oleksandr Tyshchenko <oleksandr_tyshchenko@epam.com>
Reviewed-by: Stefano Stabellini <sstabellini@kernel.org>
Reviewed-by: Anthony PERARD <anthony.perard@citrix.com>
This patch adds basic support for configuring and assisting virtio-mmio
based virtio-disk backend (emulator) which is intended to run out of
Qemu and could be run in any domain.
Although the Virtio block device is quite different from traditional
Xen PV block device (vbd) from the toolstack's point of view:
- as the frontend is virtio-blk which is not a Xenbus driver, nothing
written to Xenstore are fetched by the frontend currently ("vdev"
is not passed to the frontend). But this might need to be revised
in future, so frontend data might be written to Xenstore in order to
support hotplugging virtio devices or passing the backend domain id
on arch where the device-tree is not available.
- the ring-ref/event-channel are not used for the backend<->frontend
communication, the proposed IPC for Virtio is IOREQ/DM
it is still a "block device" and ought to be integrated in existing
"disk" handling. So, re-use (and adapt) "disk" parsing/configuration
logic to deal with Virtio devices as well.
For the immediate purpose and an ability to extend that support for
other use-cases in future (Qemu, virtio-pci, etc) perform the following
actions:
- Add new disk backend type (LIBXL_DISK_BACKEND_STANDALONE) and reflect
that in the configuration
- Introduce new disk "specification" and "transport" fields to struct
libxl_device_disk. Both are written to the Xenstore. The transport
field is only used for the specification "virtio" and it assumes
only "mmio" value for now.
- Introduce new "specification" option with "xen" communication
protocol being default value.
- Add new device kind (LIBXL__DEVICE_KIND_VIRTIO_DISK) as current
one (LIBXL__DEVICE_KIND_VBD) doesn't fit into Virtio disk model
An example of domain configuration for Virtio disk:
disk = [ 'phy:/dev/mmcblk0p3, xvda1, backendtype=standalone, specification=virtio']
Nothing has changed for default Xen disk configuration.
Please note, this patch is not enough for virtio-disk to work
on Xen (Arm), as for every Virtio device (including disk) we need
to allocate Virtio MMIO params (IRQ and memory region) and pass
them to the backend, also update Guest device-tree. The subsequent
patch will add these missing bits. For the current patch,
the default "irq" and "base" are just written to the Xenstore.
This is not an ideal splitting, but this way we avoid breaking
the bisectability.
Signed-off-by: Oleksandr Tyshchenko <oleksandr_tyshchenko@epam.com>
Reviewed-by: Anthony PERARD <anthony.perard@citrix.com>
Acked-by: George Dunlap <george.dunlap@citrix.com>
Tested-by: Jiamei Xie <jiamei.xie@arm.com>
Coverity validly points out that the mfn_add() as used was dead code.
Coverity ID: 1507475
Fixes: c1e1564c89 ("IOMMU/x86: perform PV Dom0 mappings in batches")
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Acked-by: Andrew Cooper <andrew.cooper3@citrix.com>
It's hard to imagine a case where an error may legitimately be ignored
here. It's bad enough that in at least one case (set_shadow_status())
the return value was checked only by way of ASSERT()ing.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Acked-by: Julien Grall <jgrall@amazon.com>
As of 8cc5036bc3 ("x86/pv: Fix ABAC cmpxchg() race in
_get_page_type()") this no longer needs passing separately - the type
can now be read from struct page_info, as the call now happens after its
writing.
While there also constify the 2nd parameter.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
The latest Intel manual now says the X2APIC reserved range is only
0x800 to 0x8ff (NOT 0xbff).
This changed between SDM 68 (Nov 2018) and SDM 69 (Jan 2019).
The AMD manual documents 0x800-0x8ff too.
There are non-X2APIC MSRs in the 0x900-0xbff range now:
e.g. 0x981 is IA32_TME_CAPABILITY, an architectural MSR.
The new MSR in this range appears to have been introduced in Icelake,
so this commit should be backported to Xen versions supporting Icelake.
Backport: 4.13+
Signed-off-by: Edwin Török <edvin.torok@citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
GCC with IBT extensions complains:
arch/x86/cpu/vpmu.c:351:15: error: conflicting types for 'vpmu_save_force'; have 'void(void *)' with implied 'nocf_check' attribute
351 | void cf_check vpmu_save_force(void *arg)
| ^~~~~~~~~~~~~~~
In file included from ./arch/x86/include/asm/domain.h:10,
from ./include/xen/domain.h:8,
from ./include/xen/sched.h:11,
from ./include/xen/event.h:12,
from arch/x86/cpu/vpmu.c:23:
./arch/x86/include/asm/vpmu.h:117:6: note: previous declaration of 'vpmu_save_force' with type 'void(void *)'
117 | void vpmu_save_force(void *arg);
| ^~~~~~~~~~~~~~~
Adjust the declaraion.
Fixes: 755087eb9b ("xen/mem_sharing: support forks with active vPMU state")
Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
This is a debug behaviour to identify buggy kernels. Crashing the domain is
the most unhelpful thing to do, because it discards the relevant context.
Instead, inject #GP[0] like other permission errors in x86. In particular,
this lets the kernel provide a backtrace which is more likely to be helpful to
a developer.
As a bugfix, this always injects #GP[0] to current, not l1e_owner. It is not
l1e_owner's fault if dom0 using superpowers triggers an implicit unmap.
Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
Acked-by: Jan Beulich <jbeulich@suse.com>
When this logic was moved, it was moved across the point where nx is
updated to hold the new type for the page. IOW originally it was
equivalent to using x (and perhaps x would better have been used), but
now it isn't anymore. Switch to using x, which then brings things in
line again with the slightly earlier comment there (now) talking about
transitions _from_ writable.
I have to confess though that I cannot make a direct connection between
the reported observed behavior of guests leaving several pages around
with pending general references and the change here. Repeated testing,
nevertheless, confirms the reported issue is no longer there.
This is CVE-2022-33745 / XSA-408.
Reported-by: Charles Arnold <carnold@suse.com>
Fixes: 8cc5036bc3 ("x86/pv: Fix ABAC cmpxchg() race in _get_page_type()")
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
In common/memory.c the ifdef code surrounding ptdom_max_order is
using HAS_PASSTHROUGH instead of CONFIG_HAS_PASSTHROUGH, fix the
problem using the correct macro.
Fixes: e0d44c1f94 ("build: convert HAS_PASSTHROUGH use to Kconfig")
Signed-off-by: Luca Fancellu <luca.fancellu@arm.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
Quite obviously to determine the split condition successive pages'
attributes need to be evaluated, not always those of the initial page.
Fixes: 72b02bc75b ("xen/heap: pass order to free_heap_pages() in heap init")
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Julien Grall <jgrall@amazon.com>
For a .cmd file to be picked up, the respective target needs to be
listed in $(targets). This wasn't the case for hypercall-defs.i, leading
to permanent re-building even on an entirely unchanged tree (because of
the command apparently having changed).
In exchange the target doesn't need naming in $(clean-files) anymore.
Fixes: eca1f00d02 ("xen: generate hypercall interface related code")
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Anthony PERARD <anthony.perard@citrix.com>
This addresses a build failure when CONFIG_DTB_FILE evaluates to a non-
empty string.
Fixes: d07358f2dc ("xen/arm32: head.S: Introduce a macro to load the physical address of a symbol")
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Currently the vPMU state from a parent isn't copied to VM forks. To enable the
vPMU state to be copied to a fork VM we export certain vPMU functions. First,
the vPMU context needs to be allocated for the fork if the parent has one. For
this we introduce vpmu->allocate_context, which has previously only been called
when the guest enables the PMU on itself. Furthermore, we export
vpmu_save_force so that the PMU context can be saved on-demand even if no
context switch took place on the parent's CPU yet. Additionally, we make sure
all relevant configuration MSRs are saved in the vPMU context so the copy is
complete and the fork starts with the same PMU config as the parent.
Signed-off-by: Tamas K Lengyel <tamas.lengyel@intel.com>
Acked-by: Andrew Cooper <andrew.cooper3@citrix.com>
Re-generate goland bindings to reflect changes to libxl_types.idl
from the following commit:
54d8f27d04 tools/libxl: report trusted backend status to frontends
Signed-off-by: Oleksandr Tyshchenko <oleksandr_tyshchenko@epam.com>
Reviewed-by: Anthony PERARD <anthony.perard@citrix.com>
This way intel_iommu_unmap_page() ends up quite a bit more similar to
intel_iommu_map_page().
No functional change intended.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Reviewed-by: Roger Pau Monné <roger.pau@citrix.com>
Reviewed-by: Paul Durrant <paul@xen.org>
When a page table ends up with all contiguous entries (including all
identical attributes), it can be replaced by a superpage entry at the
next higher level. The page table itself can then be scheduled for
freeing.
The adjustment to LEVEL_MASK is merely to avoid leaving a latent trap
for whenever we (and obviously hardware) start supporting 512G mappings.
Note that cache sync-ing is likely more strict than necessary. This is
both to be on the safe side as well as to maintain the pattern of all
updates of (potentially) live tables being accompanied by a flush (if so
needed).
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Reviewed-by: Roger Pau Monné <roger.pau@citrix.com>
Reviewed-by: Paul Durrant <paul@xen.org>
When a page table ends up with all contiguous entries (including all
identical attributes), it can be replaced by a superpage entry at the
next higher level. The page table itself can then be scheduled for
freeing.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Roger Pau Monné <roger.pau@citrix.com>
Reviewed-by: Paul Durrant <paul@xen.org>
When a page table ends up with no present entries left, it can be
replaced by a non-present entry at the next higher level. The page table
itself can then be scheduled for freeing.
Note that while its output isn't used there yet,
pt_update_contig_markers() right away needs to be called in all places
where entries get updated, not just the one where entries get cleared.
Note further that while pt_update_contig_markers() updates perhaps
several PTEs within the table, since these are changes to "avail" bits
only I do not think that cache flushing would be needed afterwards. Such
cache flushing (of entire pages, unless adding yet more logic to me more
selective) would be quite noticable performance-wise (very prominent
during Dom0 boot).
Also note that cache sync-ing is likely more strict than necessary. This
is both to be on the safe side as well as to maintain the pattern of all
updates of (potentially) live tables being accompanied by a flush (if so
needed).
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Reviewed-by: Roger Pau Monné <roger.pau@citrix.com>
Reviewed-by: Paul Durrant <paul@xen.org>
When a page table ends up with no present entries left, it can be
replaced by a non-present entry at the next higher level. The page table
itself can then be scheduled for freeing.
Note that while its output isn't used there yet,
pt_update_contig_markers() right away needs to be called in all places
where entries get updated, not just the one where entries get cleared.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Roger Pau Monné <roger.pau@citrix.com>
Reviewed-by: Paul Durrant <paul@xen.org>
Page tables are used for two purposes after allocation: They either
start out all empty, or they are filled to replace a superpage.
Subsequently, to replace all empty or fully contiguous page tables,
contiguous sub-regions will be recorded within individual page tables.
Install the initial set of markers immediately after allocation. Make
sure to retain these markers when further populating a page table in
preparation for it to replace a superpage.
The markers are simply 4-bit fields holding the order value of
contiguous entries. To demonstrate this, if a page table had just 16
entries, this would be the initial (fully contiguous) set of markers:
index 0 1 2 3 4 5 6 7 8 9 A B C D E F
marker 4 0 1 0 2 0 1 0 3 0 1 0 2 0 1 0
"Contiguous" here means not only present entries with successively
increasing MFNs, each one suitably aligned for its slot, and identical
attributes, but also a respective number of all non-present (zero except
for the markers) entries.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Reviewed-by: Roger Pau Monné <roger.pau@citrix.com>
Reviewed-by: Paul Durrant <paul@xen.org>
This is a re-usable helper (kind of a template) which gets introduced
without users so that the individual subsequent patches introducing such
users can get committed independently of one another.
See the comment at the top of the new file. To demonstrate the effect,
if a page table had just 16 entries, this would be the set of markers
for a page table with fully contiguous mappings:
index 0 1 2 3 4 5 6 7 8 9 A B C D E F
marker 4 0 1 0 2 0 1 0 3 0 1 0 2 0 1 0
"Contiguous" here means not only present entries with successively
increasing MFNs, each one suitably aligned for its slot, but also a
respective number of all non-present entries.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Roger Pau Monné <roger.pau@citrix.com>
Reviewed-by: Paul Durrant <paul@xen.org>
... depending on feature availability (and absence of quirks).
Also make the page table dumping function aware of superpages.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Reviewed-by: Roger Pau Monné <roger.pau@citrix.com>
Reviewed-by: Paul Durrant <paul@xen.org>
No separate feature flags exist which would control availability of
these; the only restriction is HATS (establishing the maximum number of
page table levels in general), and even that has a lower bound of 4.
Thus we can unconditionally announce 2M and 1G mappings. (Via non-
default page sizes the implementation in principle permits arbitrary
size mappings, but these require multiple identical leaf PTEs to be
written, which isn't all that different from having to write multiple
consecutive PTEs with increasing frame numbers. IMO that's therefore
beneficial only on hardware where suitable TLBs exist; I'm unaware of
such hardware.)
Note that in principle 512G and 256T mappings could also be supported
right away, but the freeing of page tables (to be introduced in
subsequent patches) when replacing a sufficiently populated tree with a
single huge page would need suitable preemption, which will require
extra work.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Roger Pau Monné <roger.pau@citrix.com>
Reviewed-by: Paul Durrant <paul@xen.org>
Before actually enabling their use, provide a means to suppress it in
case of problems. Note that using the option can also affect the sharing
of page tables in the VT-d / EPT combination: If EPT would use large
page mappings but the option is in effect, page table sharing would be
suppressed (to properly fulfill the admin request).
Requested-by: Roger Pau Monné <roger.pau@citrix.com>
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Reviewed-by: Paul Durrant <paul@xen.org>
For vendor specific code to support superpages we need to be able to
deal with a superpage mapping replacing an intermediate page table (or
hierarchy thereof). Consequently an iommu_alloc_pgtable() counterpart is
needed to free individual page tables while a domain is still alive.
Since the freeing needs to be deferred until after a suitable IOTLB
flush was performed, released page tables get queued for processing by a
tasklet.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Roger Pau Monné <roger.pau@citrix.com>
Reviewed-by: Paul Durrant <paul@xen.org>
Dario Faggioli