vmx_complete_interrupts() does too much, split it up:
- vmx_vcpu_run() gets the "cache important vmcs fields" part
- a new vmx_complete_atomic_exit() gets the parts that must be done atomically
- a new vmx_recover_nmi_blocking() does what its name says
- vmx_complete_interrupts() retains the event injection recovery code
This helps in reducing the work done in atomic context.
Signed-off-by: Avi Kivity <avi@redhat.com>
Instead of blindly attempting to inject an event before each guest entry,
check for a possible event first in vcpu->requests. Sites that can trigger
event injection are modified to set KVM_REQ_EVENT:
- interrupt, nmi window opening
- ppr updates
- i8259 output changes
- local apic irr changes
- rflags updates
- gif flag set
- event set on exit
This improves non-injecting entry performance, and sets the stage for
non-atomic injection.
Signed-off-by: Avi Kivity <avi@redhat.com>
Commit "KVM: MMU: Make tdp_enabled a mmu-context parameter" made real-mode
set ->direct_map, and changed the code that merges in the memory type depend
on direct_map instead of tdp_enabled. However, in this case what really
matters is tdp, not direct_map, since tdp changes the pte format regardless
of whether the mapping is direct or not.
As a result, real-mode shadow mappings got corrupted with ept memory types.
The result was a huge slowdown, likely due to the cache being disabled.
Change it back as the simplest fix for the regression (real fix is to move
all that to vmx code, and not use tdp_enabled as a synonym for ept).
Signed-off-by: Avi Kivity <avi@redhat.com>
This patch fixes a bug in KVM where it _always_ reports the
support of the SVM feature to userspace. But KVM only
supports SVM on AMD hardware and only when it is enabled in
the kernel module. This patch fixes the wrong reporting.
Cc: stable@kernel.org
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
This patch implements the reporting of the nested paging
feature support to userspace.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
This patch adds INTR and NMI intercepts to the list of
expected intercepts with an exit_int_info set. While this
can't happen on bare metal it is architectural legal and may
happen with KVMs SVM emulation.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
This patch adds code to initialize the Nested Nested Paging
MMU context when the L1 guest executes a VMRUN instruction
and has nested paging enabled in its VMCB.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
This patch adds the helper functions which will be used in
the mmu context for handling nested nested page faults.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
With Nested Paging emulation the NX state between the two
MMU contexts may differ. To make sure that always the right
fault error code is recorded this patch moves the NX state
into struct kvm_mmu so that the code can distinguish between
L1 and L2 NX state.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Currently the KVM softmmu implementation can not shadow a 32
bit legacy or PAE page table with a long mode page table.
This is a required feature for nested paging emulation
because the nested page table must alway be in host format.
So this patch implements the missing pieces to allow long
mode page tables for page table types.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
This patch factors out the direct-mapping paths of the
mmu_alloc_roots function into a seperate function. This
makes it a lot easier to avoid all the unnecessary checks
done in the shadow path which may break when running direct.
In fact, this patch already fixes a problem when running PAE
guests on a PAE shadow page table.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
This function is implemented to load the pdptr pointers of
the currently running guest (l1 or l2 guest). Therefore it
takes care about the current paging mode and can read pdptrs
out of l2 guest physical memory.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
This function need to be able to load the pdptrs from any
mmu context currently in use. So change this function to
take an kvm_mmu parameter to fit these needs.
As a side effect this patch also moves the cached pdptrs
from vcpu_arch into the kvm_mmu struct.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
KVM currently ignores fetch faults in the instruction
emulator. With nested-npt we could have such faults. This
patch adds the code to handle these.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
This patch implements logic to make sure that either a
page-fault/page-fault-vmexit or a nested-page-fault-vmexit
is propagated back to the guest.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
This patch introduces the init_kvm_nested_mmu() function
which is used to re-initialize the nested mmu when the l2
guest changes its paging mode.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
This patch introduces the kvm_read_guest_page_x86 function
which reads from the physical memory of the guest. If the
guest is running in guest-mode itself with nested paging
enabled it will read from the guest's guest physical memory
instead.
The patch also changes changes the code to use this function
where it is necessary.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
This patch uses kvm_read_guest_page_tdp to make the
walk_addr_generic functions suitable for two-level page
table walking.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
This patch adds a function which can read from the guests
physical memory or from the guest's guest physical memory.
This will be used in the two-dimensional page table walker.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
This patch adds the functions to do a nested l2_gva to
l1_gpa page table walk.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
This patch introduces the walk_mmu pointer which points to
the mmu-context currently used for gva_to_gpa translations.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
This patch introduces a mmu-callback to translate gpa
addresses in the walk_addr code. This is later used to
translate l2_gpa addresses into l1_gpa addresses.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
This is the first patch in the series towards a generic
walk_addr implementation which could walk two-dimensional
page tables in the end. In this first step the walk_addr
function is renamed into walk_addr_generic which takes a
mmu context as an additional parameter.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
This patch introduces a struct with two new fields in
vcpu_arch for x86:
* fault.address
* fault.error_code
This will be used to correctly propagate page faults back
into the guest when we could have either an ordinary page
fault or a nested page fault. In the case of a nested page
fault the fault-address is different from the original
address that should be walked. So we need to keep track
about the real fault-address.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
This patch changes is_rsvd_bits_set() function prototype to
take only a kvm_mmu context instead of a full vcpu.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Some logic of the init_kvm_softmmu function is required to
build the Nested Nested Paging context. So factor the
required logic into a seperate function and export it.
Also make the whole init path suitable for more than one mmu
context.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
This patch introduces an inject_page_fault function pointer
into struct kvm_mmu which will be used to inject a page
fault. This will be used later when Nested Nested Paging is
implemented.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
This function pointer in the MMU context is required to
implement Nested Nested Paging.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
This patch introduces a special set_tdp_cr3 function pointer
in kvm_x86_ops which is only used for tpd enabled mmu
contexts. This allows to remove some hacks from svm code.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
This is necessary to implement Nested Nested Paging. As a
side effect this allows some cleanups in the SVM nested
paging code.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
This patch changes the tdp_enabled flag from its global
meaning to the mmu-context and renames it to direct_map
there. This is necessary for Nested SVM with emulation of
Nested Paging where we need an extra MMU context to shadow
the Nested Nested Page Table.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
The walk_addr function checks for !is_long_mode in its 64
bit version. But what is meant here is a check for pae
paging. Change the condition to really check for pae paging
so that it also works with nested nested paging.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Some operating systems store data about the host processor at the
time of installation, and when booted on a more uptodate cpu tries
to read MSR_EBC_FREQUENCY_ID. This has been found with XP.
Signed-off-by: Jes Sorensen <Jes.Sorensen@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
This patch changes the rip handling in the vmrun emulation
path from using next_rip to the generic kvm register access
functions.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
This patch implements restoring of the correct rip, rsp, and
rax after the svm emulation in KVM injected a selective_cr0
write intercept into the guest hypervisor. The problem was
that the vmexit is emulated in the instruction emulation
which later commits the registers right after the write-cr0
instruction. So the l1 guest will continue to run with the
l2 rip, rsp and rax resulting in unpredictable behavior.
This patch is not the final word, it is just an easy patch
to fix the issue. The real fix will be done when the
instruction emulator is made aware of nested virtualization.
Until this is done this patch fixes the issue and provides
an easy way to fix this in -stable too.
Cc: stable@kernel.org
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
This patch fixes 32 bit legacy paging with NPT enabled. The
mmu_check_root call on the top-level of the loop causes
root_gfn to take values (in the tdp_enabled path) which are
outside of guest memory. So the mmu_check_root call fails at
some point in the loop interation causing the guest to
tiple-fault.
This patch changes the mmu_check_root calls to the places
where they are really necessary. As a side-effect it
introduces a check for the root of a pae page table too.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
The audit is very high overhead, so we need lower the frequency to assure
the guest is running.
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Both audit_mappings() and audit_sptes_have_rmaps() need to walk vcpu's page
table, so we can do these checking in a spte walking
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Both audit_rmap() and audit_write_protection() need to walk all active sp, so
we can do these checking in a sp walking
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Move the audit code from arch/x86/kvm/mmu.c to arch/x86/kvm/mmu_audit.c
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Add a r/w module parameter named 'mmu_audit', it can control audit
enable/disable:
enable:
echo 1 > /sys/module/kvm/parameters/mmu_audit
disable:
echo 0 > /sys/module/kvm/parameters/mmu_audit
This patch not change the logic
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
MSR_K7_CLK_CTL is a no longer documented MSR, which is only relevant
on said old AMD CPU models. This change returns the expected value,
which the Linux kernel is expecting to avoid writing back the MSR,
plus it ignores all writes to the MSR.
Signed-off-by: Jes Sorensen <Jes.Sorensen@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
ICW is not a full reset, instead it resets a limited number of registers
in the PIC. Change ICW1 emulation to only reset those registers.
Signed-off-by: Avi Kivity <avi@redhat.com>
x86_emulate_insn() is full of things like
if (rc != X86EMUL_CONTINUE)
goto done;
break;
consolidate all of those at the end of the switch statement.
Signed-off-by: Avi Kivity <avi@redhat.com>
Otherwise EFER_LMA bit is retained across a SIPI reset.
Fixes guest cpu onlining.
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Since commit aad827034e no mmu reinitialization is performed
via init_vmcb.
Zero vcpu->arch.cr0 and pass the reset value as a parameter to
kvm_set_cr0.
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Nothing is checked in count_rmaps(), so remove it
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
There is a bugs in this function, we call gfn_to_pfn() and kvm_mmu_gva_to_gpa_read() in
atomic context(kvm_mmu_audit() is called under the spinlock(mmu_lock)'s protection).
This patch fix it by:
- introduce gfn_to_pfn_atomic instead of gfn_to_pfn
- get the mapping gfn from kvm_mmu_page_get_gfn()
And it adds 'notrap' ptes check in unsync/direct sps
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
The audit code reports some sp not write protected in current code, it's just the
bug in audit_write_protection(), since:
- the invalid sp not need write protected
- using uninitialize local variable('gfn')
- call kvm_mmu_audit() out of mmu_lock's protection
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
The read-only spte also has reverse mapping, so fix the code to check them,
also modify the function name to fit its doing
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
fix:
arch/x86/kvm/mmu.c: In function ‘kvm_mmu_unprotect_page’:
arch/x86/kvm/mmu.c:1741: warning: format ‘%lx’ expects type ‘long unsigned int’, but argument 3 has type ‘gfn_t’
arch/x86/kvm/mmu.c:1745: warning: format ‘%lx’ expects type ‘long unsigned int’, but argument 3 has type ‘gfn_t’
arch/x86/kvm/mmu.c: In function ‘mmu_unshadow’:
arch/x86/kvm/mmu.c:1761: warning: format ‘%lx’ expects type ‘long unsigned int’, but argument 3 has type ‘gfn_t’
arch/x86/kvm/mmu.c: In function ‘set_spte’:
arch/x86/kvm/mmu.c:2005: warning: format ‘%lx’ expects type ‘long unsigned int’, but argument 3 has type ‘gfn_t’
arch/x86/kvm/mmu.c: In function ‘mmu_set_spte’:
arch/x86/kvm/mmu.c:2033: warning: format ‘%lx’ expects type ‘long unsigned int’, but argument 7 has type ‘gfn_t’
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Pit interrupt injection was done by workqueue, so no need to check
pending pit timer in vcpu thread which could lead unnecessary
unblocking of vcpu.
Signed-off-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
The ALU opcode block is very regular; introduce D6ALU() to define decode
flags for 6 instructions at a time.
Suggested by Paolo Bonzini.
Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Many x86 instructions come in byte and word variants distinguished with bit
0 of the opcode. Add macros to aid in defining them.
Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
SrcMemFAddr is not defined with the modrm operand designating a register
instead of a memory address.
Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
x86_emulate_insn() will return 1 if instruction can be restarted
without re-entering a guest.
Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Support prefetch ptes when intercept guest #PF, avoid to #PF by later
access
If we meet any failure in the prefetch path, we will exit it and
not try other ptes to avoid become heavy path
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Kernel time, which advances in discrete steps may progress much slower
than TSC. As a result, when kvmclock is adjusted to a new base, the
apparent time to the guest, which runs at a much higher, nsec scaled
rate based on the current TSC, may have already been observed to have
a larger value (kernel_ns + scaled tsc) than the value to which we are
setting it (kernel_ns + 0).
We must instead compute the clock as potentially observed by the guest
for kernel_ns to make sure it does not go backwards.
Signed-off-by: Zachary Amsden <zamsden@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
The scale_delta function for shift / multiply with 31-bit
precision moves to a common header so it can be used by both
kernel and kvm module.
Signed-off-by: Zachary Amsden <zamsden@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
If there are active VCPUs which are marked as belonging to
a particular hardware CPU, request a clock sync for them when
enabling hardware; the TSC could be desynchronized on a newly
arriving CPU, and we need to recompute guests system time
relative to boot after a suspend event.
This covers both cases.
Note that it is acceptable to take the spinlock, as either
no other tasks will be running and no locks held (BSP after
resume), or other tasks will be guaranteed to drop the lock
relatively quickly (AP on CPU_STARTING).
Noting we now get clock synchronization requests for VCPUs
which are starting up (or restarting), it is tempting to
attempt to remove the arch/x86/kvm/x86.c CPU hot-notifiers
at this time, however it is not correct to do so; they are
required for systems with non-constant TSC as the frequency
may not be known immediately after the processor has started
until the cpufreq driver has had a chance to run and query
the chipset.
Updated: implement better locking semantics for hardware_enable
Removed the hack of dropping and retaking the lock by adding the
semantic that we always hold kvm_lock when hardware_enable is
called. The one place that doesn't need to worry about it is
resume, as resuming a frozen CPU, the spinlock won't be taken.
Signed-off-by: Zachary Amsden <zamsden@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Make the match of TSC find TSC writes that are close to each other
instead of perfectly identical; this allows the compensator to also
work in migration / suspend scenarios.
Signed-off-by: Zachary Amsden <zamsden@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Add a helper function to compute the kernel time and convert nanoseconds
back to CPU specific cycles. Note that these must not be called in preemptible
context, as that would mean the kernel could enter software suspend state,
which would cause non-atomic operation.
Also, convert the KVM_SET_CLOCK / KVM_GET_CLOCK ioctls to use the kernel
time helper, these should be bootbased as well.
Signed-off-by: Zachary Amsden <zamsden@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
When CPUs with unstable TSCs enter deep C-state, TSC may stop
running. This causes us to require resynchronization. Since
we can't tell when this may potentially happen, we assume the
worst by forcing re-compensation for it at every point the VCPU
task is descheduled.
Signed-off-by: Zachary Amsden <zamsden@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Move the TSC control logic from the vendor backends into x86.c
by adding adjust_tsc_offset to x86 ops. Now all TSC decisions
can be done in one place.
Signed-off-by: Zachary Amsden <zamsden@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
If creating an SMP guest with unstable host TSC, issue a warning
Signed-off-by: Zachary Amsden <zamsden@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
This simplifies much of the init code; we can now simply always
call tsc_khz_changed, optionally passing it a new value, or letting
it figure out the existing value (while interrupts are disabled, and
thus, by inference from the rule, not raceful against CPU hotplug or
frequency updates, which will issue IPIs to the local CPU to perform
this very same task).
Signed-off-by: Zachary Amsden <zamsden@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Attempt to synchronize TSCs which are reset to the same value. In the
case of a reliable hardware TSC, we can just re-use the same offset, but
on non-reliable hardware, we can get closer by adjusting the offset to
match the elapsed time.
Signed-off-by: Zachary Amsden <zamsden@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Also, ensure that the storing of the offset and the reading of the TSC
are never preempted by taking a spinlock. While the lock is overkill
now, it is useful later in this patch series.
Signed-off-by: Zachary Amsden <zamsden@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Change svm / vmx to be the same internally and write TSC offset
instead of bare TSC in helper functions. Isolated as a single
patch to contain code movement.
Signed-off-by: Zachary Amsden <zamsden@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
This is used only by the VMX code, and is not done properly;
if the TSC is indeed backwards, it is out of sync, and will
need proper handling in the logic at each and every CPU change.
For now, drop this test during init as misguided.
Signed-off-by: Zachary Amsden <zamsden@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
commit ad05c88266b4cce1c820928ce8a0fb7690912ba1
(KVM: create aggregate kvm_total_used_mmu_pages value)
introduce percpu counter kvm_total_used_mmu_pages but never
destroy it, this may cause oops when rmmod & modprobe.
Signed-off-by: Wei Yongjun <yjwei@cn.fujitsu.com>
Acked-by: Tim Pepper <lnxninja@linux.vnet.ibm.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Latest kvm mmu_shrink code rework makes kernel changes kvm->arch.n_used_mmu_pages/
kvm->arch.n_max_mmu_pages at kvm_mmu_free_page/kvm_mmu_alloc_page, which is called
by kvm_mmu_commit_zap_page. So the kvm->arch.n_used_mmu_pages or
kvm_mmu_available_pages(vcpu->kvm) is unchanged after kvm_mmu_prepare_zap_page(),
This caused kvm_mmu_change_mmu_pages/__kvm_mmu_free_some_pages loops forever.
Moving kvm_mmu_commit_zap_page would make the while loop performs as normal.
Reported-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Xiaotian Feng <dfeng@redhat.com>
Tested-by: Avi Kivity <avi@redhat.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Cc: Tim Pepper <lnxninja@linux.vnet.ibm.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Of slab shrinkers, the VM code says:
* Note that 'shrink' will be passed nr_to_scan == 0 when the VM is
* querying the cache size, so a fastpath for that case is appropriate.
and it *means* it. Look at how it calls the shrinkers:
nr_before = (*shrinker->shrink)(0, gfp_mask);
shrink_ret = (*shrinker->shrink)(this_scan, gfp_mask);
So, if you do anything stupid in your shrinker, the VM will doubly
punish you.
The mmu_shrink() function takes the global kvm_lock, then acquires
every VM's kvm->mmu_lock in sequence. If we have 100 VMs, then
we're going to take 101 locks. We do it twice, so each call takes
202 locks. If we're under memory pressure, we can have each cpu
trying to do this. It can get really hairy, and we've seen lock
spinning in mmu_shrink() be the dominant entry in profiles.
This is guaranteed to optimize at least half of those lock
aquisitions away. It removes the need to take any of the locks
when simply trying to count objects.
A 'percpu_counter' can be a large object, but we only have one
of these for the entire system. There are not any better
alternatives at the moment, especially ones that handle CPU
hotplug.
Signed-off-by: Dave Hansen <dave@linux.vnet.ibm.com>
Signed-off-by: Tim Pepper <lnxninja@linux.vnet.ibm.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Doing this makes the code much more readable. That's
borne out by the fact that this patch removes code. "used"
also happens to be the number that we need to return back to
the slab code when our shrinker gets called. Keeping this
value as opposed to free makes the next patch simpler.
So, 'struct kvm' is kzalloc()'d. 'struct kvm_arch' is a
structure member (and not a pointer) of 'struct kvm'. That
means they start out zeroed. I _think_ they get initialized
properly by kvm_mmu_change_mmu_pages(). But, that only happens
via kvm ioctls.
Another benefit of storing 'used' intead of 'free' is
that the values are consistent from the moment the structure is
allocated: no negative "used" value.
Signed-off-by: Dave Hansen <dave@linux.vnet.ibm.com>
Signed-off-by: Tim Pepper <lnxninja@linux.vnet.ibm.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
arch.n_alloc_mmu_pages is a poor choice of name. This value truly
means, "the number of pages which _may_ be allocated". But,
reading the name, "n_alloc_mmu_pages" implies "the number of allocated
mmu pages", which is dead wrong.
It's really the high watermark, so let's give it a name to match:
nr_max_mmu_pages. This change will make the next few patches
much more obvious and easy to read.
Signed-off-by: Dave Hansen <dave@linux.vnet.ibm.com>
Signed-off-by: Tim Pepper <lnxninja@linux.vnet.ibm.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
"free" is a poor name for this value. In this context, it means,
"the number of mmu pages which this kvm instance should be able to
allocate." But "free" implies much more that the objects are there
and ready for use. "available" is a much better description, especially
when you see how it is calculated.
In this patch, we abstract its use into a function. We'll soon
replace the function's contents by calculating the value in a
different way.
All of the reads of n_free_mmu_pages are taken care of in this
patch. The modification sites will be handled in a patch
later in the series.
Signed-off-by: Dave Hansen <dave@linux.vnet.ibm.com>
Signed-off-by: Tim Pepper <lnxninja@linux.vnet.ibm.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Avi Kivity