| Age | Commit message (Collapse) | Author |
|---|
|
The EL2 'vmemmap' array in nVHE Protected mode is currently very sparse:
only memory pages owned by the hypervisor itself have a matching 'struct
hyp_page'. However, as the size of this struct has been reduced
significantly since its introduction, it appears that we can now afford
to back the vmemmap for all of memory.
Having an easily accessible 'struct hyp_page' for every physical page in
memory provides the hypervisor with a simple mechanism to store metadata
(e.g. a refcount) that wouldn't otherwise fit in the very limited number
of software bits available in the host stage-2 page-table entries. This
will be used in subsequent patches when pinning host memory pages for
use by the hypervisor at EL2.
Tested-by: Vincent Donnefort <vdonnefort@google.com>
Signed-off-by: Quentin Perret <qperret@google.com>
Signed-off-by: Will Deacon <will@kernel.org>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221110190259.26861-4-will@kernel.org
|
|
All the contiguous pages used to initialize a 'struct hyp_pool' are
considered coalescable, which means that the hyp page allocator will
actively try to merge them with their buddies on the hyp_put_page() path.
However, using hyp_put_page() on a page that is not part of the inital
memory range given to a hyp_pool() is currently unsupported.
In order to allow dynamically extending hyp pools at run-time, add a
check to __hyp_attach_page() to allow inserting 'external' pages into
the free-list of order 0. This will be necessary to allow lazy donation
of pages from the host to the hypervisor when allocating guest stage-2
page-table pages at EL2.
Tested-by: Vincent Donnefort <vdonnefort@google.com>
Signed-off-by: Quentin Perret <qperret@google.com>
Signed-off-by: Will Deacon <will@kernel.org>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221110190259.26861-3-will@kernel.org
|
|
We will soon need to manipulate 'struct hyp_page' refcounts from outside
page_alloc.c, so move the helpers to a common header file to allow them
to be reused easily.
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Oliver Upton <oliver.upton@linux.dev>
Tested-by: Vincent Donnefort <vdonnefort@google.com>
Signed-off-by: Quentin Perret <qperret@google.com>
Signed-off-by: Will Deacon <will@kernel.org>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221110190259.26861-2-will@kernel.org
|
|
Fix typo in comment (nVHE/VHE).
Signed-off-by: Zhiyuan Dai <daizhiyuan@phytium.com.cn>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/1667737840-702-1-git-send-email-daizhiyuan@phytium.com.cn
|
|
This includes table format and using reST labels for
cross-referencing to vcpu.rst.
Suggested-by: Bagas Sanjaya <bagasdotme@gmail.com>
Signed-off-by: Usama Arif <usama.arif@bytedance.com>
Reviewed-by: Steven Price <steven.price@arm.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221103131210.3603385-1-usama.arif@bytedance.com
|
|
Kernel configs with PAGE_SIZE=64KB and PA_BITS=48 still advertise 52 bit
IPA space on HW that implements LPA. This is by design (admitedly this
is a very unlikely configuration in the real world).
However on such a config, attempting to create a vm with the guest
kernel placed above 48 bits in IPA space results in misbehaviour due to
the hypervisor incorrectly interpretting a faulting IPA.
Fix up PAR_TO_HPFAR() to always take 52 bits out of the PAR rather than
masking to CONFIG_ARM64_PA_BITS. If the system has a smaller implemented
PARange this should be safe because the bits are res0.
A more robust approach would be to discover the IPA size in use by the
page-table and mask based on that, to avoid relying on res0 reading back
as zero. But this information is difficult to access safely from the
code's location, so take the easy way out.
Fixes: bc1d7de8c550 ("kvm: arm64: Add 52bit support for PAR to HPFAR conversoin")
Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>
[maz: commit message fixes]
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221103150507.32948-3-ryan.roberts@arm.com
|
|
For nvhe and protected modes, the hyp stage 1 page-tables were previously
configured to have the same number of VA bits as the kernel's idmap.
However, for kernel configs with VA_BITS=52 and where the kernel is
loaded in physical memory below 48 bits, the idmap VA bits is actually
smaller than the kernel's normal stage 1 VA bits. This can lead to
kernel addresses that can't be mapped into the hypervisor, leading to
kvm initialization failure during boot:
kvm [1]: IPA Size Limit: 48 bits
kvm [1]: Cannot map world-switch code
kvm [1]: error initializing Hyp mode: -34
Fix this by ensuring that the hyp stage 1 VA size is the maximum of
what's used for the idmap and the regular kernel stage 1. At the same
time, refactor the code so that the hyp VA bits is only calculated in
one place.
Prior to 7ba8f2b2d652, the idmap was always 52 bits for a 52 VA bits
kernel and therefore the hyp stage1 was also always 52 bits.
Fixes: 7ba8f2b2d652 ("arm64: mm: use a 48-bit ID map when possible on 52-bit VA builds")
Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>
[maz: commit message fixes]
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221103150507.32948-2-ryan.roberts@arm.com
|
|
Add some mix of tests into page_fault_test: memory regions with all the
pairwise combinations of read-only, userfaultfd, and dirty-logging. For
example, writing into a read-only region which has a hole handled with
userfaultfd.
Signed-off-by: Ricardo Koller <ricarkol@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221017195834.2295901-15-ricarkol@google.com
|
|
Add some readonly memslot tests into page_fault_test. Mark the data and/or
page-table memory regions as readonly, perform some accesses, and check
that the right fault is triggered when expected (e.g., a store with no
write-back should lead to an mmio exit).
Signed-off-by: Ricardo Koller <ricarkol@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221017195834.2295901-14-ricarkol@google.com
|
|
Add some dirty logging tests into page_fault_test. Mark the data and/or
page-table memory regions for dirty logging, perform some accesses, and
check that the dirty log bits are set or clean when expected.
Signed-off-by: Ricardo Koller <ricarkol@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221017195834.2295901-13-ricarkol@google.com
|
|
Add some userfaultfd tests into page_fault_test. Punch holes into the
data and/or page-table memslots, perform some accesses, and check that
the faults are taken (or not taken) when expected.
Signed-off-by: Ricardo Koller <ricarkol@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221017195834.2295901-12-ricarkol@google.com
|
|
Add a new test for stage 2 faults when using different combinations of
guest accesses (e.g., write, S1PTW), backing source type (e.g., anon)
and types of faults (e.g., read on hugetlbfs with a hole). The next
commits will add different handling methods and more faults (e.g., uffd
and dirty logging). This first commit starts by adding two sanity checks
for all types of accesses: AF setting by the hw, and accessing memslots
with holes.
Signed-off-by: Ricardo Koller <ricarkol@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221017195834.2295901-11-ricarkol@google.com
|
|
allocations
Now that kvm_vm allows specifying different memslots for code, page tables,
and data, use the appropriate memslot when making allocations in
common/libraty code. Change them accordingly:
- code (allocated by lib/elf) use the CODE memslot
- stacks, exception tables, and other core data pages (like the TSS in x86)
use the DATA memslot
- page tables and the PGD use the PT memslot
- test data (anything allocated with vm_vaddr_alloc()) uses the TEST_DATA
memslot
No functional change intended. All allocators keep using memslot #0.
Cc: Sean Christopherson <seanjc@google.com>
Cc: Andrew Jones <andrew.jones@linux.dev>
Signed-off-by: Ricardo Koller <ricarkol@google.com>
Reviewed-by: Sean Christopherson <seanjc@google.com>
Reviewed-by: Andrew Jones <andrew.jones@linux.dev>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221017195834.2295901-10-ricarkol@google.com
|
|
Refactor virt_arch_pgd_alloc() and vm_vaddr_alloc() in both RISC-V and
aarch64 to fix the alignment of parameters in a couple of calls. This will
make it easier to fix the alignment in a future commit that adds an extra
parameter (that happens to be very long).
No functional change intended.
Suggested-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Ricardo Koller <ricarkol@google.com>
Reviewed-by: Andrew Jones <andrew.jones@linux.dev>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221017195834.2295901-9-ricarkol@google.com
|
|
The vm_create() helpers are hardcoded to place most page types (code,
page-tables, stacks, etc) in the same memslot #0, and always backed with
anonymous 4K. There are a couple of issues with that. First, tests
willing to differ a bit, like placing page-tables in a different backing
source type must replicate much of what's already done by the vm_create()
functions. Second, the hardcoded assumption of memslot #0 holding most
things is spread everywhere; this makes it very hard to change.
Fix the above issues by having selftests specify how they want memory to be
laid out. Start by changing ____vm_create() to not create memslot #0; a
test (to come) will specify all memslots used by the VM. Then, add the
vm->memslots[] array to specify the right memslot for different memory
allocators, e.g.,: lib/elf should use the vm->[MEM_REGION_CODE] memslot.
This will be used as a way to specify the page-tables memslots (to be
backed by huge pages for example).
There is no functional change intended. The current commit lays out memory
exactly as before. A future commit will change the allocators to get the
region they should be using, e.g.,: like the page table allocators using
the pt memslot.
Cc: Sean Christopherson <seanjc@google.com>
Cc: Andrew Jones <andrew.jones@linux.dev>
Signed-off-by: Ricardo Koller <ricarkol@google.com>
Reviewed-by: Andrew Jones <andrew.jones@linux.dev>
Reviewed-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221017195834.2295901-8-ricarkol@google.com
|
|
Add the backing_src_type into struct userspace_mem_region. This struct
already stores a lot of info about memory regions, except the backing
source type. This info will be used by a future commit in order to
determine the method for punching a hole.
Reviewed-by: Oliver Upton <oliver.upton@linux.dev>
Signed-off-by: Ricardo Koller <ricarkol@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221017195834.2295901-7-ricarkol@google.com
|
|
Copy bitfield.h from include/linux/bitfield.h. A subsequent change will
make use of some FIELD_{GET,PREP} macros defined in this header.
The header was copied as-is, no changes needed.
Cc: Jakub Kicinski <kuba@kernel.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Reviewed-by: Oliver Upton <oupton@google.com>
Signed-off-by: Ricardo Koller <ricarkol@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221017195834.2295901-6-ricarkol@google.com
|
|
Define macros for memory type indexes and construct DEFAULT_MAIR_EL1
with macros from asm/sysreg.h. The index macros can then be used when
constructing PTEs (instead of using raw numbers).
Reviewed-by: Andrew Jones <andrew.jones@linux.dev>
Reviewed-by: Oliver Upton <oupton@google.com>
Signed-off-by: Ricardo Koller <ricarkol@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221017195834.2295901-5-ricarkol@google.com
|
|
Deleting a memslot (when freeing a VM) is not closing the backing fd,
nor it's unmapping the alias mapping. Fix by adding the missing close
and munmap.
Reviewed-by: Andrew Jones <andrew.jones@linux.dev>
Reviewed-by: Oliver Upton <oupton@google.com>
Reviewed-by: Ben Gardon <bgardon@google.com>
Signed-off-by: Ricardo Koller <ricarkol@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221017195834.2295901-4-ricarkol@google.com
|
|
Add a library function to get the PTE (a host virtual address) of a
given GVA. This will be used in a future commit by a test to clear and
check the access flag of a particular page.
Reviewed-by: Oliver Upton <oliver.upton@linux.dev>
Reviewed-by: Andrew Jones <andrew.jones@linux.dev>
Signed-off-by: Ricardo Koller <ricarkol@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221017195834.2295901-3-ricarkol@google.com
|
|
Move the generic userfaultfd code out of demand_paging_test.c into a
common library, userfaultfd_util. This library consists of a setup and a
stop function. The setup function starts a thread for handling page
faults using the handler callback function. This setup returns a
uffd_desc object which is then used in the stop function (to wait and
destroy the threads).
Reviewed-by: Oliver Upton <oliver.upton@linux.dev>
Reviewed-by: Ben Gardon <bgardon@google.com>
Signed-off-by: Ricardo Koller <ricarkol@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221017195834.2295901-2-ricarkol@google.com
|
|
Currently, the debug-exceptions test always uses only
{break,watch}point#0 and the highest numbered context-aware
breakpoint. Modify the test to use all {break,watch}points and
context-aware breakpoints supported on the system.
Signed-off-by: Reiji Watanabe <reijiw@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221020054202.2119018-10-reijiw@google.com
|
|
Currently, the debug-exceptions test doesn't have a test case for
a linked watchpoint. Add a test case for the linked watchpoint to
the test. The new test case uses the highest numbered context-aware
breakpoint (for Context ID match), and the watchpoint#0, which is
linked to the context-aware breakpoint.
Signed-off-by: Reiji Watanabe <reijiw@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221020054202.2119018-9-reijiw@google.com
|
|
Currently, the debug-exceptions test doesn't have a test case for
a linked breakpoint. Add a test case for the linked breakpoint to
the test. The new test case uses a pair of breakpoints. One is the
higiest numbered context-aware breakpoint (for Context ID match),
and the other one is the breakpoint#0 (for Address Match), which
is linked to the context-aware breakpoint.
Signed-off-by: Reiji Watanabe <reijiw@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221020054202.2119018-8-reijiw@google.com
|
|
Change debug_version() to take the ID_AA64DFR0_EL1 value instead of
vcpu as an argument, and change its callsite to read ID_AA64DFR0_EL1
(and pass it to debug_version()).
Subsequent patches will reuse the register value in the callsite.
No functional change intended.
Signed-off-by: Reiji Watanabe <reijiw@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221020054202.2119018-7-reijiw@google.com
|
|
Currently, debug-exceptions test unnecessarily tracks some test stages
using GUEST_SYNC(). The code for it needs to be updated as test cases
are added or removed. Stop doing the unnecessary stage tracking,
as they are not so useful and are a bit pain to maintain.
Signed-off-by: Reiji Watanabe <reijiw@google.com>
Reviewed-by: Oliver Upton <oliver.upton@linux.dev>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221020054202.2119018-6-reijiw@google.com
|
|
Add helpers to enable breakpoint and watchpoint exceptions.
No functional change intended.
Signed-off-by: Reiji Watanabe <reijiw@google.com>
Reviewed-by: Ricardo Koller <ricarkol@google.com>
Reviewed-by: Oliver Upton <oliver.upton@linux.dev>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221020054202.2119018-5-reijiw@google.com
|
|
Remove the hard-coded {break,watch}point #0 from the guest_code() in
debug-exceptions to allow {break,watch}point number to be specified.
Change reset_debug_state() to zeroing all dbg{b,w}{c,v}r_el0 registers
so that guest_code() can use the function to reset those registers
even when non-zero {break,watch}points are specified for guest_code().
Subsequent patches will add test cases for non-zero {break,watch}points.
Signed-off-by: Reiji Watanabe <reijiw@google.com>
Reviewed-by: Ricardo Koller <ricarkol@google.com>
Reviewed-by: Oliver Upton <oliver.upton@linux.dev>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221020054202.2119018-4-reijiw@google.com
|
|
Introduce helpers in the debug-exceptions test to write to
dbg{b,w}{c,v}r registers. Those helpers will be useful for
test cases that will be added to the test in subsequent patches.
No functional change intended.
Signed-off-by: Reiji Watanabe <reijiw@google.com>
Reviewed-by: Ricardo Koller <ricarkol@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221020054202.2119018-3-reijiw@google.com
|
|
Use FIELD_GET() macro to extract ID register fields for existing
aarch64 selftests code. No functional change intended.
Signed-off-by: Reiji Watanabe <reijiw@google.com>
Reviewed-by: Oliver Upton <oliver.upton@linux.dev>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221020054202.2119018-2-reijiw@google.com
|
|
The memory area in each slot should be aligned to host page size.
Otherwise, the test will fail. For example, the following command
fails with the following messages with 64KB-page-size-host and
4KB-pae-size-guest. It's not user friendly to abort the test.
Lets do something to report the optimal memory slots, instead of
failing the test.
# ./memslot_perf_test -v -s 1000
Number of memory slots: 999
Testing map performance with 1 runs, 5 seconds each
Adding slots 1..999, each slot with 8 pages + 216 extra pages last
==== Test Assertion Failure ====
lib/kvm_util.c:824: vm_adjust_num_guest_pages(vm->mode, npages) == npages
pid=19872 tid=19872 errno=0 - Success
1 0x00000000004065b3: vm_userspace_mem_region_add at kvm_util.c:822
2 0x0000000000401d6b: prepare_vm at memslot_perf_test.c:273
3 (inlined by) test_execute at memslot_perf_test.c:756
4 (inlined by) test_loop at memslot_perf_test.c:994
5 (inlined by) main at memslot_perf_test.c:1073
6 0x0000ffff7ebb4383: ?? ??:0
7 0x00000000004021ff: _start at :?
Number of guest pages is not compatible with the host. Try npages=16
Report the optimal memory slots instead of failing the test when
the memory area in each slot isn't aligned to host page size. With
this applied, the optimal memory slots is reported.
# ./memslot_perf_test -v -s 1000
Number of memory slots: 999
Testing map performance with 1 runs, 5 seconds each
Memslot count too high for this test, decrease the cap (max is 514)
Signed-off-by: Gavin Shan <gshan@redhat.com>
Reviewed-by: Maciej S. Szmigiero <maciej.szmigiero@oracle.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221020071209.559062-7-gshan@redhat.com
|
|
The addresses and sizes passed to vm_userspace_mem_region_add() and
madvise() should be aligned to host page size, which can be 64KB on
aarch64. So it's wrong by passing additional fixed 4KB memory area
to various tests.
Fix it by passing additional fixed 64KB memory area to various tests.
We also add checks to ensure that none of host/guest page size exceeds
64KB. MEM_TEST_MOVE_SIZE is fixed up to 192KB either.
With this, the following command works fine on 64KB-page-size-host and
4KB-page-size-guest.
# ./memslot_perf_test -v -s 512
Signed-off-by: Gavin Shan <gshan@redhat.com>
Reviewed-by: Maciej S. Szmigiero <maciej.szmigiero@oracle.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221020071209.559062-6-gshan@redhat.com
|
|
The test case is obviously broken on aarch64 because non-4KB guest
page size is supported. The guest page size on aarch64 could be 4KB,
16KB or 64KB.
This supports variable guest page size, mostly for aarch64.
- The host determines the guest page size when virtual machine is
created. The value is also passed to guest through the synchronization
area.
- The number of guest pages are unknown until the virtual machine
is to be created. So all the related macros are dropped. Instead,
their values are dynamically calculated based on the guest page
size.
- The static checks on memory sizes and pages becomes dependent
on guest page size, which is unknown until the virtual machine
is about to be created. So all the static checks are converted
to dynamic checks, done in check_memory_sizes().
- As the address passed to madvise() should be aligned to host page,
the size of page chunk is automatically selected, other than one
page.
- MEM_TEST_MOVE_SIZE has fixed and non-working 64KB. It will be
consolidated in next patch. However, the comments about how
it's calculated has been correct.
- All other changes included in this patch are almost mechanical
replacing '4096' with 'guest_page_size'.
Signed-off-by: Gavin Shan <gshan@redhat.com>
Reviewed-by: Maciej S. Szmigiero <maciej.szmigiero@oracle.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221020071209.559062-5-gshan@redhat.com
|
|
prepare_vm() is called in every iteration and run. The allowed memory
slots (KVM_CAP_NR_MEMSLOTS) are probed for multiple times. It's not
free and unnecessary.
Move the probing logic for the allowed memory slots to parse_args()
for once, which is upper layer of prepare_vm().
No functional change intended.
Signed-off-by: Gavin Shan <gshan@redhat.com>
Reviewed-by: Maciej S. Szmigiero <maciej.szmigiero@oracle.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221020071209.559062-4-gshan@redhat.com
|
|
There are two loops in prepare_vm(), which have different conditions.
'slot' is treated as meory slot index in the first loop, but index of
the host virtual address array in the second loop. It makes it a bit
hard to understand the code.
Change the usage of 'slot' in the second loop, to treat it as the
memory slot index either.
No functional change intended.
Signed-off-by: Gavin Shan <gshan@redhat.com>
Reviewed-by: Maciej S. Szmigiero <maciej.szmigiero@oracle.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221020071209.559062-3-gshan@redhat.com
|
|
In prepare_vm(), 'data->nslots' is assigned with 'max_mem_slots - 1'
at the beginning, meaning they are interchangeable.
Use 'data->nslots' isntead of 'max_mem_slots - 1'. With this, it
becomes easier to move the logic of probing number of slots into
upper layer in subsequent patches.
No functional change intended.
Signed-off-by: Gavin Shan <gshan@redhat.com>
Reviewed-by: Maciej S. Szmigiero <maciej.szmigiero@oracle.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221020071209.559062-2-gshan@redhat.com
|
|
The stage-2 map walker has been made parallel-aware, and as such can be
called while only holding the read side of the MMU lock. Rip out the
conditional locking in user_mem_abort() and instead grab the read lock.
Continue to take the write lock from other callsites to
kvm_pgtable_stage2_map().
Signed-off-by: Oliver Upton <oliver.upton@linux.dev>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221107220033.1895655-1-oliver.upton@linux.dev
|
|
stage2_map_walker_try_leaf() and friends now handle stage-2 PTEs
generically, and perform the correct flush when a table PTE is removed.
Additionally, they've been made parallel-aware, using an atomic break
to take ownership of the PTE.
Stop clearing the PTE in the pre-order callback and instead let
stage2_map_walker_try_leaf() deal with it.
Signed-off-by: Oliver Upton <oliver.upton@linux.dev>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221107220006.1895572-1-oliver.upton@linux.dev
|
|
Convert stage2_map_walker_try_leaf() to use the new break-before-make
helpers, thereby making the handler parallel-aware. As before, avoid the
break-before-make if recreating the existing mapping. Additionally,
retry execution if another vCPU thread is modifying the same PTE.
Signed-off-by: Oliver Upton <oliver.upton@linux.dev>
Reviewed-by: Ben Gardon <bgardon@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221107215934.1895478-1-oliver.upton@linux.dev
|
|
In order to service stage-2 faults in parallel, stage-2 table walkers
must take exclusive ownership of the PTE being worked on. An additional
requirement of the architecture is that software must perform a
'break-before-make' operation when changing the block size used for
mapping memory.
Roll these two concepts together into helpers for performing a
'break-before-make' sequence. Use a special PTE value to indicate a PTE
has been locked by a software walker. Additionally, use an atomic
compare-exchange to 'break' the PTE when the stage-2 page tables are
possibly shared with another software walker. Elide the DSB + TLBI if
the evicted PTE was invalid (and thus not subject to break-before-make).
All of the atomics do nothing for now, as the stage-2 walker isn't fully
ready to perform parallel walks.
Signed-off-by: Oliver Upton <oliver.upton@linux.dev>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221107215855.1895367-1-oliver.upton@linux.dev
|
|
Create a helper to initialize a table and directly call
smp_store_release() to install it (for now). Prepare for a subsequent
change that generalizes PTE writes with a helper.
Signed-off-by: Oliver Upton <oliver.upton@linux.dev>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221107215644.1895162-11-oliver.upton@linux.dev
|
|
The stage2 attr walker is already used for parallel walks. Since commit
f783ef1c0e82 ("KVM: arm64: Add fast path to handle permission relaxation
during dirty logging"), KVM acquires the read lock when
write-unprotecting a PTE. However, the walker only uses a simple store
to update the PTE. This is safe as the only possible race is with
hardware updates to the access flag, which is benign.
However, a subsequent change to KVM will allow more changes to the stage
2 page tables to be done in parallel. Prepare the stage 2 attribute
walker by performing atomic updates to the PTE when walking in parallel.
Signed-off-by: Oliver Upton <oliver.upton@linux.dev>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221107215644.1895162-10-oliver.upton@linux.dev
|
|
Use RCU to safely walk the stage-2 page tables in parallel. Acquire and
release the RCU read lock when traversing the page tables. Defer the
freeing of table memory to an RCU callback. Indirect the calls into RCU
and provide stubs for hypervisor code, as RCU is not available in such a
context.
The RCU protection doesn't amount to much at the moment, as readers are
already protected by the read-write lock (all walkers that free table
memory take the write lock). Nonetheless, a subsequent change will
futher relax the locking requirements around the stage-2 MMU, thereby
depending on RCU.
Signed-off-by: Oliver Upton <oliver.upton@linux.dev>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221107215644.1895162-9-oliver.upton@linux.dev
|
|
The break-before-make sequence is a bit annoying as it opens a window
wherein memory is unmapped from the guest. KVM should replace the PTE
as quickly as possible and avoid unnecessary work in between.
Presently, the stage-2 map walker tears down a removed table before
installing a block mapping when coalescing a table into a block. As the
removed table is no longer visible to hardware walkers after the
DSB+TLBI, it is possible to move the remaining cleanup to happen after
installing the new PTE.
Reshuffle the stage-2 map walker to install the new block entry in
the pre-order callback. Unwire all of the teardown logic and replace
it with a call to kvm_pgtable_stage2_free_removed() after fixing
the PTE. The post-order visitor is now completely unnecessary, so drop
it. Finally, touch up the comments to better represent the now
simplified map walker.
Note that the call to tear down the unlinked stage-2 is indirected
as a subsequent change will use an RCU callback to trigger tear down.
RCU is not available to pKVM, so there is a need to use different
implementations on pKVM and non-pKVM VMs.
Signed-off-by: Oliver Upton <oliver.upton@linux.dev>
Reviewed-by: Ben Gardon <bgardon@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221107215644.1895162-8-oliver.upton@linux.dev
|
|
Use an opaque type for pteps and require visitors explicitly dereference
the pointer before using. Protecting page table memory with RCU requires
that KVM dereferences RCU-annotated pointers before using. However, RCU
is not available for use in the nVHE hypervisor and the opaque type can
be conditionally annotated with RCU for the stage-2 MMU.
Call the type a 'pteref' to avoid a naming collision with raw pteps. No
functional change intended.
Signed-off-by: Oliver Upton <oliver.upton@linux.dev>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221107215644.1895162-7-oliver.upton@linux.dev
|
|
A subsequent change to KVM will move the tear down of an unlinked
stage-2 subtree out of the critical path of the break-before-make
sequence.
Introduce a new helper for tearing down unlinked stage-2 subtrees.
Leverage the existing stage-2 free walkers to do so, with a deep call
into __kvm_pgtable_walk() as the subtree is no longer reachable from the
root.
Signed-off-by: Oliver Upton <oliver.upton@linux.dev>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221107215644.1895162-6-oliver.upton@linux.dev
|
|
In order to tear down page tables from outside the context of
kvm_pgtable (such as an RCU callback), stop passing a pointer through
kvm_pgtable_walk_data.
No functional change intended.
Signed-off-by: Oliver Upton <oliver.upton@linux.dev>
Reviewed-by: Ben Gardon <bgardon@google.com>
Reviewed-by: Gavin Shan <gshan@redhat.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221107215644.1895162-5-oliver.upton@linux.dev
|
|
As a prerequisite for getting visitors off of struct kvm_pgtable, pass
mm_ops through the visitor context.
No functional change intended.
Signed-off-by: Oliver Upton <oliver.upton@linux.dev>
Reviewed-by: Ben Gardon <bgardon@google.com>
Reviewed-by: Gavin Shan <gshan@redhat.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221107215644.1895162-4-oliver.upton@linux.dev
|
|
Rather than reading the ptep all over the shop, read the ptep once from
__kvm_pgtable_visit() and stick it in the visitor context. Reread the
ptep after visiting a leaf in case the callback installed a new table
underneath.
No functional change intended.
Signed-off-by: Oliver Upton <oliver.upton@linux.dev>
Reviewed-by: Ben Gardon <bgardon@google.com>
Reviewed-by: Gavin Shan <gshan@redhat.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221107215644.1895162-3-oliver.upton@linux.dev
|
|
Passing new arguments by value to the visitor callbacks is extremely
inflexible for stuffing new parameters used by only some of the
visitors. Use a context structure instead and pass the pointer through
to the visitor callback.
While at it, redefine the 'flags' parameter to the visitor to contain
the bit indicating the phase of the walk. Pass the entire set of flags
through the context structure such that the walker can communicate
additional state to the visitor callback.
No functional change intended.
Signed-off-by: Oliver Upton <oliver.upton@linux.dev>
Reviewed-by: Ben Gardon <bgardon@google.com>
Reviewed-by: Gavin Shan <gshan@redhat.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221107215644.1895162-2-oliver.upton@linux.dev
|
