- 02 Aug, 2021 31 commits
-
-
Sean Christopherson authored
Stuff vcpu->arch.apic_base and apic->base_address directly during APIC reset, as opposed to bouncing through kvm_set_apic_base() while fudging the ENABLE bit during creation to avoid the other, unwanted side effects. This is a step towards consolidating the APIC RESET logic across x86, VMX, and SVM. Signed-off-by:
Sean Christopherson <seanjc@google.com> Message-Id: <20210713163324.627647-18-seanjc@google.com> Signed-off-by:
Paolo Bonzini <pbonzini@redhat.com>
-
Sean Christopherson authored
Write vcpu->arch.apic_base directly instead of bouncing through kvm_set_apic_base(). This is a glorified nop, and is a step towards cleaning up the mess that is local APIC creation. When using an in-kernel APIC, kvm_create_lapic() explicitly sets vcpu->arch.apic_base to MSR_IA32_APICBASE_ENABLE to avoid its own kvm_lapic_set_base() call in kvm_lapic_reset() from triggering state changes. That call during RESET exists purely to set apic->base_address to the default base value. As a result, by the time VMX gets control, the only missing piece is the BSP bit being set for the reset BSP. For a userspace APIC, there are no side effects to process (for the APIC). In both cases, the call to kvm_update_cpuid_runtime() is a nop because the vCPU hasn't yet been exposed to userspace, i.e. there can't be any CPUID entries. No functional change intended. Reviewed-by:
Reiji Watanabe <reijiw@google.com> Signed-off-by:
-
Sean Christopherson authored
Set the BSP bit appropriately during local APIC "reset" instead of relying on vendor code to clean up at a later point. This is a step towards consolidating the local APIC, VMX, and SVM xAPIC initialization code. Reviewed-by:
-
Sean Christopherson authored
Don't set the BSP bit in vcpu->arch.apic_base when the local APIC is managed by userspace. Forcing all vCPUs to be BSPs is non-sensical, and was dead code when it was added by commit 97222cc8 ("KVM: Emulate local APIC in kernel"). At the time, kvm_lapic_set_base() was invoked if and only if the local APIC was in-kernel (and it couldn't be called before the vCPU created its APIC). kvm_lapic_set_base() eventually gained generic usage, but the latent bug escaped notice because the only true consumer would be the guest itself in the form of an explicit RDMSRs on APs. Out of Linux, SeaBIOS, and EDK2/OVMF, only OVMF consumes the BSP bit from the APIC_BASE MSR. For the vast majority of usage in OVMF, BSP confusion would be benign. OVMF's BSP election upon SMI rendezvous might be broken, but practically no one runs KVM with an out-of-kernel local APIC, let alone does so while utilizing SMIs with OVMF. Fixes: 97222cc8 ("KVM: Emulate local APIC in kernel") Reviewed-by:
-
Sean Christopherson authored
Make vcpu0 the arbitrary owner of the PIT, as was intended when PIT migration was added by commit 2f599714 ("KVM: migrate PIT timer"). The PIT was unintentionally turned into being owned by the BSP by commit c5af89b6 ("KVM: Introduce kvm_vcpu_is_bsp() function."), and was then unintentionally converted to a shared ownership model when kvm_vcpu_is_bsp() was modified to check the APIC base MSR instead of hardcoding vcpu0 as the BSP. Functionally, this just means the PIT's hrtimer is migrated less often. The real motivation is to remove the usage of kvm_vcpu_is_bsp(), so that more legacy/broken crud can be removed in a future patch. Fixes: 58d269d8 ("KVM: x86: BSP in MSR_IA32_APICBASE is writable") Signed-off-by:
-
Sean Christopherson authored
Remove a BSP APIC update in kvm_lapic_reset() that is a glorified and confusing nop. When the code was originally added, kvm_vcpu_is_bsp() queried kvm->arch.bsp_vcpu, i.e. the intent was to set the BSP bit in the BSP vCPU's APIC. But, stuffing the BSP bit at INIT was wrong since the guest can change its BSP(s); this was fixed by commit 58d269d8 ("KVM: x86: BSP in MSR_IA32_APICBASE is writable"). In other words, kvm_vcpu_is_bsp() is now purely a reflection of vcpu->arch.apic_base.MSR_IA32_APICBASE_BSP, thus the update will always set the current value and kvm_lapic_set_base() is effectively a nop if the new and old values match. The RESET case, which does need to stuff the BSP for the reset vCPU, is handled by vendor code (though this will soon be moved to common code). No functional change intended. Reviewed-by:
-
Sean Christopherson authored
WARN if KVM ends up in a state where it thinks its APIC map needs to be recalculated, but KVM is not emulating the local APIC. This is mostly to document KVM's "rules" in order to provide clarity in future cleanups. Signed-off-by:
-
Sean Christopherson authored
Drop an explicit MMU reset in SVM's vCPU RESET/INIT flow now that the common x86 path correctly handles conditional MMU resets, e.g. if INIT arrives while the vCPU is in 64-bit mode. This reverts commit ebae871a ("kvm: svm: reset mmu on VCPU reset"). Reviewed-by:
-
Sean Christopherson authored
Drop an explicit MMU reset when entering emulated real mode now that the vCPU INIT/RESET path correctly handles conditional MMU resets, e.g. if INIT arrives while the vCPU is in 64-bit mode. Note, while there are multiple other direct calls to vmx_set_cr0(), i.e. paths that change CR0 without invoking kvm_post_set_cr0(), only the INIT emulation can reach enter_rmode(). CLTS emulation only toggles CR.TS, VM-Exit (and late VM-Fail) emulation cannot architecturally transition to Real Mode, and VM-Enter to Real Mode is possible if and only if Unrestricted Guest is enabled (exposed to L1). This effectively reverts commit 8668a3c4 ("KVM: VMX: Reset mmu context when entering real mode") Signed-off-by:
-
Sean Christopherson authored
At vCPU RESET/INIT (mostly RESET), stuff EDX with KVM's hardcoded, default Family-Model-Stepping ID of 0x600 if CPUID.0x1 isn't defined. At RESET, the CPUID lookup is guaranteed to "miss" because KVM emulates RESET before exposing the vCPU to userspace, i.e. userspace can't possibly have done set the vCPU's CPUID model, and thus KVM will always write '0'. At INIT, using 0x600 is less bad than using '0'. While initializing EDX to '0' is _extremely_ unlikely to be noticed by the guest, let alone break the guest, and can be overridden by userspace for the RESET case, using 0x600 is preferable as it will allow consolidating the relevant VMX and SVM RESET/INIT logic in the future. And, digging through old specs suggests that neither Intel nor AMD have ever shipped a CPU that initialized EDX to '0' at RESET. Regarding 0x600 as KVM's default Family, it is a sane default and in many ways the most appropriate. Prior to the 386 implementations, DX was undefined at RESET. With the 386, 486, 586/P5, and 686/P6/Athlon, both Intel and AMD set EDX to 3, 4, 5, and 6 respectively. AMD switched to using '15' as its primary Family with the introduction of AMD64, but Intel has continued using '6' for the last few decades. So, '6' is a valid Family for both Intel and AMD CPUs, is compatible with both 32-bit and 64-bit CPUs (albeit not a perfect fit for 64-bit AMD), and of the common Families (3 - 6), is the best fit with respect to KVM's virtual CPU model. E.g. prior to the P6, Intel CPUs did not have a STI window. Modern operating systems, Linux included, rely on the STI window, e.g. for "safe halt", and KVM unconditionally assumes the virtual CPU has an STI window. Thus enumerating a Family ID of 3, 4, or 5 would be provably wrong. Opportunistically remove a stale comment. Fixes: 66f7b72e ("KVM: x86: Make register state after reset conform to specification") Reviewed-by:
-
Sean Christopherson authored
Do not allow an inexact CPUID "match" when querying the guest's CPUID.0x1 to stuff EDX during INIT. In the common case, where the guest CPU model is an AMD variant, allowing an inexact match is a nop since KVM doesn't emulate Intel's goofy "out-of-range" logic for AMD and Hygon. If the vCPU model happens to be an Intel variant, an inexact match is possible if and only if the max CPUID leaf is precisely '0'. Aside from the fact that there's probably no CPU in existence with a single CPUID leaf, if the max CPUID leaf is '0', that means that CPUID.0.EAX is '0', and thus an inexact match for CPUID.0x1.EAX will also yield '0'. So, with lots of twisty logic, no functional change intended. Reviewed-by:
-
Sean Christopherson authored
Set EDX at RESET/INIT based on the userspace-defined CPUID model when possible, i.e. when CPUID.0x1.EAX is defind by userspace. At RESET/INIT, all CPUs that support CPUID set EDX to the FMS enumerated in CPUID.0x1.EAX. If no CPUID match is found, fall back to KVM's default of 0x600 (Family '6'), which is the least awful approximation of KVM's virtual CPU model. Fixes: 6aa8b732 ("[PATCH] kvm: userspace interface") Signed-off-by:
-
Sean Christopherson authored
Explicitly set GDTR.base and IDTR.base to zero when intializing the VMCB. Functionally this only affects INIT, as the bases are implicitly set to zero on RESET by virtue of the VMCB being zero allocated. Per AMD's APM, GDTR.base and IDTR.base are zeroed after RESET and INIT. Fixes: 04d2cc77 ("KVM: Move main vcpu loop into subarch independent code") Signed-off-by:
-
Sean Christopherson authored
Set L1's LDTR on VM-Exit per the Intel SDM: The host-state area does not contain a selector field for LDTR. LDTR is established as follows on all VM exits: the selector is cleared to 0000H, the segment is marked unusable and is otherwise undefined (although the base address is always canonical). This is likely a benign bug since the LDTR is unusable, as it means the L1 VMM is conditioned to reload its LDTR in order to function properly on bare metal. Fixes: 4704d0be ("KVM: nVMX: Exiting from L2 to L1") Reviewed-by:
-
Sean Christopherson authored
Flush the guest's TLB on INIT, as required by Intel's SDM. Although AMD's APM states that the TLBs are unchanged by INIT, it's not clear that that's correct as the APM also states that the TLB is flush on "External initialization of the processor." Regardless, relying on the guest to be paranoid is unnecessarily risky, while an unnecessary flush is benign from a functional perspective and likely has no measurable impact on guest performance. Note, as of the April 2021 version of Intels' SDM, it also contradicts itself with respect to TLB flushing. The overview of INIT explicitly calls out the TLBs as being invalidated, while a table later in the same section says they are unchanged. 9.1 INITIALIZATION OVERVIEW: The major difference is that during an INIT, the internal caches, MSRs, MTRRs, and x87 FPU state are left unchanged (although, the TLBs and BTB are invalidated as with a hardware reset) Table 9-1: Register Power up Reset INIT Data and Code Cache, TLBs: Invalid[6] Invalid[6] Unchanged Given Core2's erratum[*] about global TLB entries not being flush on INIT, it's safe to assume that the table is simply wrong. AZ28. INIT Does Not Clear Global Entries in the TLB Problem: INIT may not flush a TLB entry when: • The processor is in protected mode with paging enabled and the page global enable flag is set (PGE bit of CR4 register) • G bit for the page table entry is set • TLB entry is present in TLB when INIT occurs • Software may encounter unexpected page fault or incorrect address translation due to a TLB entry erroneously left in TLB after INIT. Workaround: Write to CR3, CR4 (setting bits PSE, PGE or PAE) or CR0 (setting bits PG or PE) registers before writing to memory early in BIOS code to clear all the global entries from TLB. Status: For the steppings affected, see the Summary Tables of Changes. [*] https://www.intel.com/content/dam/support/us/en/documents/processors/mobile/celeron/sb/320121.pdf Fixes: 6aa8b732 ("[PATCH] kvm: userspace interface") Signed-off-by: -
Maxim Levitsky authored
Special case of disabling the APICv on the current vCPU right away in kvm_request_apicv_update doesn't bring much benefit vs raising KVM_REQ_APICV_UPDATE on it instead, since this request will be processed on the next entry to the guest. (the comment about having another #VMEXIT is wrong). It also hides various assumptions that APIVc enable state matches the APICv inhibit state, as this special case only makes those states match on the current vCPU. Previous patches fixed few such assumptions so now it should be safe to drop this special case. Signed-off-by:
Maxim Levitsky <mlevitsk@redhat.com> Message-Id: <20210713142023.106183-5-mlevitsk@redhat.com> Signed-off-by:
-
Paolo Bonzini authored
With the addition of fast page fault support, the TDP-specific MMU has reached feature parity with the original MMU. All my testing in the last few months has been done with the TDP MMU; switch the default on 64-bit machines. Signed-off-by: -
David Matlack authored
Make fast_page_fault interoperate with the TDP MMU by leveraging walk_shadow_page_lockless_{begin,end} to acquire the RCU read lock and introducing a new helper function kvm_tdp_mmu_fast_pf_get_last_sptep to grab the lowest level sptep. Suggested-by:Ben Gardon <bgardon@google.com> Signed-off-by:
David Matlack <dmatlack@google.com> Message-Id: <20210713220957.3493520-5-dmatlack@google.com> Signed-off-by:
-
David Matlack authored
Acquire the RCU read lock in walk_shadow_page_lockless_begin and release it in walk_shadow_page_lockless_end when the TDP MMU is enabled. This should not introduce any functional changes but is used in the following commit to make fast_page_fault interoperate with the TDP MMU. Signed-off-by:
-
David Matlack authored
Enum values have to be exported to userspace since the formatting is not done in the kernel. Without doing this perf maps RET_PF_FIXED and RET_PF_SPURIOUS to 0, which results in incorrect output: $ perf record -a -e kvmmmu:fast_page_fault --filter "ret==3" -- ./access_tracking_perf_test $ perf script | head -1 [...] new 610006048d25877 spurious 0 fixed 0 <------ should be 1 Fix this by exporting the enum values to userspace with TRACE_DEFINE_ENUM. Fixes: c4371c2a ("KVM: x86/mmu: Return unique RET_PF_* values if the fault was fixed") Signed-off-by:
-
David Matlack authored
fast_page_fault is only called from direct_page_fault where we know the address is a gpa. Fixes: 736c291c ("KVM: x86: Use gpa_t for cr2/gpa to fix TDP support on 32-bit KVM") Reviewed-by:
-
Peter Xu authored
Introduce this safe version of kvm_get_kvm() so that it can be called even during vm destruction. Use it in kvm_debugfs_open() and remove the verbose comment. Prepare to be used elsewhere. Signed-off-by:
Peter Xu <peterx@redhat.com> Message-Id: <20210625153214.43106-3-peterx@redhat.com> [Preserve the comment in kvm_debugfs_open. - Paolo] Signed-off-by:
-
Juergen Gross authored
The file has been moved to arch/x86 long time ago. Time to get rid of non-x86 stuff. Signed-off-by:
Juergen Gross <jgross@suse.com> Message-Id: <20210701154105.23215-3-jgross@suse.com> Signed-off-by:
-
Peter Xu authored
Add a new statistic max_mmu_rmap_size, which stores the maximum size of rmap for the vm. Signed-off-by:
-
Sean Christopherson authored
Return the old SPTE when clearing a SPTE and push the "old SPTE present" check to the caller. Private shadow page support will use the old SPTE in rmap_remove() to determine whether or not there is a linked private shadow page. Signed-off-by:
Sean Christopherson <sean.j.christopherson@intel.com> Signed-off-by:
Isaku Yamahata <isaku.yamahata@intel.com> Reviewed-by:
-
Sean Christopherson authored
Employ a 'continue' to reduce the indentation for linking a new shadow page during __direct_map() in preparation for linking private pages. Signed-off-by:
-
Sean Christopherson authored
Move the kvm_dirty_regs vs. KVM_SYNC_X86_VALID_FIELDS check out of sync_regs() and into its sole caller, kvm_arch_vcpu_ioctl_run(). This allows a future patch to allow synchronizing select state for protected VMs. Signed-off-by:
-
Sean Christopherson authored
Signed-off-by:
-
Sean Christopherson authored
Signed-off-by:
-
Sean Christopherson authored
Export kvm_make_all_cpus_request() and hoist the request helper declarations of request up to the KVM_REQ_* definitions in preparation for adding a "VM bugged" framework. The framework will add KVM_BUG() and KVM_BUG_ON() as alternatives to full BUG()/BUG_ON() for cases where KVM has definitely hit a bug (in itself or in silicon) and the VM is all but guaranteed to be hosed. Marking a VM bugged will trigger a request to all vCPUs to allow arch code to forcefully evict each vCPU from its run loop. Signed-off-by:
-
Sean Christopherson authored
Signed-off-by:
-
- 30 Jul, 2021 1 commit
-
-
Paolo Bonzini authored
Once an exception has been injected, any side effects related to the exception (such as setting CR2 or DR6) have been taked place. Therefore, once KVM sets the VM-entry interruption information field or the AMD EVENTINJ field, the next VM-entry must deliver that exception. Pending interrupts are processed after injected exceptions, so in theory it would not be a problem to use KVM_INTERRUPT when an injected exception is present. However, DOSEMU is using run->ready_for_interrupt_injection to detect interrupt windows and then using KVM_SET_SREGS/KVM_SET_REGS to inject the interrupt manually. For this to work, the interrupt window must be delayed after the completion of the previous event injection. Cc: stable@vger.kernel.org Reported-by:
Stas Sergeev <stsp2@yandex.ru> Tested-by:
-
- 27 Jul, 2021 8 commits
-
-
Paolo Bonzini authored
The arguments to the KVM_CLEAR_DIRTY_LOG ioctl include a pointer, therefore it needs a compat ioctl implementation. Otherwise, 32-bit userspace fails to invoke it on 64-bit kernels; for x86 it might work fine by chance if the padding is zero, but not on big-endian architectures. Reported-by: Thomas Sattler Cc: stable@vger.kernel.org Fixes: 2a31b9db ("kvm: introduce manual dirty log reprotect") Reviewed-by:
-
Li RongQing authored
SMT siblings share caches and other hardware, and busy halt polling will degrade its sibling performance if its sibling is working Sean Christopherson suggested as below: "Rather than disallowing halt-polling entirely, on x86 it should be sufficient to simply have the hardware thread yield to its sibling(s) via PAUSE. It probably won't get back all performance, but I would expect it to be close. This compiles on all KVM architectures, and AFAICT the intended usage of cpu_relax() is identical for all architectures." Suggested-by:
Li RongQing <lirongqing@baidu.com> Message-Id: <20210727111247.55510-1-lirongqing@baidu.com> Signed-off-by:
-
Maxim Levitsky authored
Currently when SVM is enabled in guest CPUID, AVIC is inhibited as soon as the guest CPUID is set. AVIC happens to be fully disabled on all vCPUs by the time any guest entry starts (if after migration the entry can be nested). The reason is that currently we disable avic right away on vCPU from which the kvm_request_apicv_update was called and for this case, it happens to be called on all vCPUs (by svm_vcpu_after_set_cpuid). After we stop doing this, AVIC will end up being disabled only when KVM_REQ_APICV_UPDATE is processed which is after we done switching to the nested guest. Fix this by just using vmcb01 in svm_refresh_apicv_exec_ctrl for avic (which is a right thing to do anyway). Signed-off-by:
-
Maxim Levitsky authored
It is possible that AVIC was requested to be disabled but not yet disabled, e.g if the nested entry is done right after svm_vcpu_after_set_cpuid. Signed-off-by:
-
Maxim Levitsky authored
It is possible for AVIC inhibit and AVIC active state to be mismatched. Currently we disable AVIC right away on vCPU which started the AVIC inhibit request thus this warning doesn't trigger but at least in theory, if svm_set_vintr is called at the same time on multiple vCPUs, the warning can happen. Signed-off-by:
-
Christian Borntraeger authored
commit bc9e9e67 ("KVM: debugfs: Reuse binary stats descriptors") did replace the old definitions with the binary ones. While doing that it missed that some files are names different than the counters. This is especially important for kvm_stat which does have special handling for counters named instruction_*. Fixes: commit bc9e9e67 ("KVM: debugfs: Reuse binary stats descriptors") CC: Jing Zhang <jingzhangos@google.com> Signed-off-by:
Christian Borntraeger <borntraeger@de.ibm.com> Message-Id: <20210726150108.5603-1-borntraeger@de.ibm.com> Signed-off-by:
-
Paolo Bonzini authored
Right now, svm_hv_vmcb_dirty_nested_enlightenments has an incorrect dereference of vmcb->control.reserved_sw before the vmcb is checked for being non-NULL. The compiler is usually sinking the dereference after the check; instead of doing this ourselves in the source, ensure that svm_hv_vmcb_dirty_nested_enlightenments is only called with a non-NULL VMCB. Reported-by:
Dan Carpenter <dan.carpenter@oracle.com> Cc: Vineeth Pillai <viremana@linux.microsoft.com> Signed-off-by:
-
David Matlack authored
This test measures the performance effects of KVM's access tracking. Access tracking is driven by the MMU notifiers test_young, clear_young, and clear_flush_young. These notifiers do not have a direct userspace API, however the clear_young notifier can be triggered by marking a pages as idle in /sys/kernel/mm/page_idle/bitmap. This test leverages that mechanism to enable access tracking on guest memory. To measure performance this test runs a VM with a configurable number of vCPUs that each touch every page in disjoint regions of memory. Performance is measured in the time it takes all vCPUs to finish touching their predefined region. Example invocation: $ ./access_tracking_perf_test -v 8 Testing guest mode: PA-bits:ANY, VA-bits:48, 4K pages guest physical test memory offset: 0xffdfffff000 Populating memory : 1.337752570s Writing to populated memory : 0.010177640s Reading from populated memory : 0.009548239s Mark memory idle : 23.973131748s Writing to idle memory : 0.063584496s Mark memory idle : 24.924652964s Reading from idle memory : 0.062042814s Breaking down the results: * "Populating memory": The time it takes for all vCPUs to perform the first write to every page in their region. * "Writing to populated memory" / "Reading from populated memory": The time it takes for all vCPUs to write and read to every page in their region after it has been populated. This serves as a control for the later results. * "Mark memory idle": The time it takes for every vCPU to mark every page in their region as idle through page_idle. * "Writing to idle memory" / "Reading from idle memory": The time it takes for all vCPUs to write and read to every page in their region after it has been marked idle. This test should be portable across architectures but it is only enabled for x86_64 since that's all I have tested. Reviewed-by:
-
