- 16 Mar, 2020 40 commits
-
-
Sean Christopherson authored
Use the recently introduced KVM CPU caps to propagate VMX-only (kernel) settings to supported CPUID flags. No functional change intended. Reviewed-by:
Vitaly Kuznetsov <vkuznets@redhat.com> Signed-off-by:
Sean Christopherson <sean.j.christopherson@intel.com> Signed-off-by:
Paolo Bonzini <pbonzini@redhat.com>
-
Sean Christopherson authored
Use the recently introduced KVM CPU caps to propagate SVM-only (kernel) settings to supported CPUID flags. Note, there are a few subtleties: - Setting a flag based on a *different* feature is effectively emulation, and must be done at runtime via ->set_supported_cpuid(). - CPUID 0x8000000A.EDX is a feature leaf that was previously not adjusted by kvm_cpu_cap_mask() because all features are hidden by default. Opportunistically add a technically unnecessary break and fix an indentation issue in svm_set_supported_cpuid(). No functional change intended. Reviewed-by: -
Sean Christopherson authored
Calculate the CPUID masks for KVM_GET_SUPPORTED_CPUID at load time using what is effectively a KVM-adjusted copy of boot_cpu_data, or more precisely, the x86_capability array in boot_cpu_data. In terms of KVM support, the vast majority of CPUID feature bits are constant, and *all* feature support is known at KVM load time. Rather than apply boot_cpu_data, which is effectively read-only after init, at runtime, copy it into a KVM-specific array and use *that* to mask CPUID registers. In additional to consolidating the masking, kvm_cpu_caps can be adjusted by SVM/VMX at load time and thus eliminate all feature bit manipulation in ->set_supported_cpuid(). Opportunistically clean up a few warts: - Replace bare "unsigned" with "unsigned int" when a feature flag is captured in a local variable, e.g. f_nx. - Sort the CPUID masks by function, index and register (alphabetically for registers, i.e. EBX comes before ECX/EDX). - Remove the superfluous /* cpuid 7.0.ecx */ comments. No functional change intended. Signed-off-by: -
Sean Christopherson authored
Invert the handling of XSAVES, i.e. set it based on boot_cpu_has() by default, in preparation for adding KVM cpu caps, which will generate the mask at load time before ->xsaves_supported() is ready. No functional change intended. Reviewed-by:
-
Paolo Bonzini authored
The clearing of the GBPAGE CPUID bit for VMX is wrong; support for 1GB pages in EPT has no relationship to whether 1GB pages should be marked as supported in CPUID. This has no ill effect because we're only clearing the bit, but we're not marking 1GB pages as available when EPT is disabled (even though they are actually supported thanks to shadowing). Instead, forcibly enable 1GB pages in the shadow paging case. This also eliminates an instance of the undesirable "unsigned f_* = *_supported ? F(*) : 0" pattern in the common CPUID handling code, and paves the way toward eliminating ->get_lpage_level(). Signed-off-by: -
Sean Christopherson authored
Move the Processor Trace CPUID adjustment into VMX code to eliminate an instance of the undesirable "unsigned f_* = *_supported ? F(*) : 0" pattern in the common CPUID handling code, and to pave the way toward eventually removing ->pt_supported(). No functional change intended. Reviewed-by:
-
Sean Christopherson authored
Move the clearing of the RDTSCP CPUID bit into VMX, which has a separate VMCS control to enable RDTSCP in non-root, to eliminate an instance of the undesirable "unsigned f_* = *_supported ? F(*) : 0" pattern in the common CPUID handling code. Drop ->rdtscp_supported() since CPUID adjustment was the last remaining user. No functional change intended. Reviewed-by:
-
Sean Christopherson authored
Move the setting of the PKU CPUID bit into VMX to eliminate an instance of the undesirable "unsigned f_* = *_supported ? F(*) : 0" pattern in the common CPUID handling code. Drop ->pku_supported(), CPUID adjustment was the only user. Note, some AMD CPUs now support PKU, but SVM doesn't yet support exposing it to a guest. No functional change intended. Reviewed-by:
-
Sean Christopherson authored
Move the CPUID adjustment for UMIP emulation into VMX code to eliminate an instance of the undesirable "unsigned f_* = *_supported ? F(*) : 0" pattern in the common CPUID handling code. No functional change intended. Reviewed-by:
-
Sean Christopherson authored
Move the INVPCID CPUID adjustments into VMX to eliminate an instance of the undesirable "unsigned f_* = *_supported ? F(*) : 0" pattern in the common CPUID handling code. Drop ->invpcid_supported(), CPUID adjustment was the only user. No functional change intended. Reviewed-by:
-
Sean Christopherson authored
Move the MPX CPUID adjustments into VMX to eliminate an instance of the undesirable "unsigned f_* = *_supported ? F(*) : 0" pattern in the common CPUID handling code. Note, to maintain existing behavior, VMX must manually check for kernel support for MPX by querying boot_cpu_has(X86_FEATURE_MPX). Previously, do_cpuid_7_mask() masked MPX based on boot_cpu_data by invoking cpuid_mask() on the associated cpufeatures word, but cpuid_mask() runs prior to executing vmx_set_supported_cpuid(). No functional change intended. Reviewed-by:
-
Sean Christopherson authored
Use the recently introduced cpuid_entry_get_reg() to automatically get the appropriate register when masking a CPUID entry. No functional change intended. Reviewed-by:
-
Sean Christopherson authored
Introduce mutators to modify feature bits in CPUID entries and use the new mutators where applicable. Using the mutators eliminates the need to manually specify the register to modify query at no extra cost and will allow adding runtime consistency checks on the function/index in a future patch. No functional change intended. Reviewed-by:
-
Sean Christopherson authored
Introduce accessors to retrieve feature bits from CPUID entries and use the new accessors where applicable. Using the accessors eliminates the need to manually specify the register to be queried at no extra cost (binary output is identical) and will allow adding runtime consistency checks on the function and index in a future patch. No functional change intended. Reviewed-by:
-
Sean Christopherson authored
Replace "unsigned" with "unsigned int" to make checkpatch and people everywhere a little bit happier, and to avoid propagating the filth when future patches add more cpuid helpers that work with unsigned (ints). No functional change intended. Suggested-by:
-
Sean Christopherson authored
Change the intermediate CPUID output register values from "int" to "u32" to match both hardware and the storage type in struct cpuid_reg. No functional change intended. Reviewed-by:
-
Sean Christopherson authored
Drop the explicit @func param from ->set_supported_cpuid() and instead pull the CPUID function from the relevant entry. This sets the stage for hardening guest CPUID updates in future patches, e.g. allows adding run-time assertions that the CPUID feature being changed is actually a bit in the referenced CPUID entry. No functional change intended. Reviewed-by:
-
Sean Christopherson authored
Clear the output regs for the main CPUID 0x14 leaf (index=0) if Intel PT isn't exposed to the guest. Leaf 0x14 enumerates Intel PT capabilities and should return zeroes if PT is not supported. Incorrectly reporting PT capabilities is essentially a cosmetic error, i.e. doesn't negatively affect any known userspace/kernel, as the existence of PT itself is correctly enumerated via CPUID 0x7. Reviewed-by:
-
Sean Christopherson authored
Expose kvm_mpx_supported() as a static inline so that it can be inlined in kvm_intel.ko. No functional change intended. Reviewed-by:
Xiaoyao Li <xiaoyao.li@intel.com> Reviewed-by:
-
Sean Christopherson authored
Query supported_xcr0 when checking for MPX support instead of invoking ->mpx_supported() and drop ->mpx_supported() as kvm_mpx_supported() was its last user. Rename vmx_mpx_supported() to cpu_has_vmx_mpx() to better align with VMX/VMCS nomenclature. Modify VMX's adjustment of xcr0 to call cpus_has_vmx_mpx() (renamed from vmx_mpx_supported()) directly to avoid reading supported_xcr0 before it's fully configured. No functional change intended. Reviewed-by:
-
Sean Christopherson authored
Add a new global variable, supported_xcr0, to track which xcr0 bits can be exposed to the guest instead of calculating the mask on every call. The supported bits are constant for a given instance of KVM. This paves the way toward eliminating the ->mpx_supported() call in kvm_mpx_supported(), e.g. eliminates multiple retpolines in VMX's nested VM-Enter path, and eventually toward eliminating ->mpx_supported() altogether. No functional change intended. Reviewed-by:
-
Sean Christopherson authored
Add helpers to query which of the (two) supported PT modes is active. The primary motivation is to help document that there is a third PT mode (host-only) that's currently not supported by KVM. As is, it's not obvious that PT_MODE_SYSTEM != !PT_MODE_HOST_GUEST and vice versa, e.g. that "pt_mode == PT_MODE_SYSTEM" and "pt_mode != PT_MODE_HOST_GUEST" are two distinct checks. No functional change intended. Reviewed-by:
-
Sean Christopherson authored
Use __do_cpuid_func()'s common loop iterator, "i", when enumerating the sub-leafs for CPUID 0xD now that the CPUID 0xD loop doesn't need to manual maintain separate counts for the entries index and CPUID index. No functional changed intended. Reviewed-by:
-
Sean Christopherson authored
Drop a "nent >= maxnent" check in kvm_get_cpuid() that's fully redundant now that kvm_get_cpuid() isn't indexing the array to pass an entry to do_cpuid_func(). Reviewed-by:
-
Sean Christopherson authored
Add a struct to hold the array of CPUID entries and its associated metadata when handling KVM_GET_SUPPORTED_CPUID. Lookup and provide the correct entry in do_host_cpuid(), which eliminates the majority of array indexing shenanigans, e.g. entries[i -1], and generally makes the code more readable. The last array indexing holdout is kvm_get_cpuid(), which can't really be avoided without throwing the baby out with the bathwater. No functional change intended. Reviewed-by:
-
Sean Christopherson authored
Refactoring the sub-leaf handling for CPUID 0x4/0x8000001d to eliminate a one-off variable and its associated brackets. No functional change intended. Reviewed-by:
-
Sean Christopherson authored
Declare "i" and "max_idx" at the top of __do_cpuid_func() to consolidate a handful of declarations in various case statements. More importantly, establish the pattern of using max_idx instead of e.g. entry->eax as the loop terminator in preparation for refactoring how entry is handled in __do_cpuid_func(). No functional change intended. Reviewed-by:
-
Sean Christopherson authored
Move the nent vs. maxnent check and nent increment into do_host_cpuid() to consolidate what is now identical code. To signal success vs. failure, return the entry and NULL respectively. A future patch will build on this to also move the entry retrieval into do_host_cpuid(). No functional change intended. Reviewed-by:
-
Sean Christopherson authored
Drop redundant checks when "emulating" SSBD feature across vendors, i.e. advertising the AMD variant when running on an Intel CPU and vice versa. Both SPEC_CTRL_SSBD and AMD_SSBD are already defined in the leaf-specific feature masks and are *not* forcefully set by the kernel, i.e. will already be set in the entry when supported by the host. Functionally, this changes nothing, but the redundant check is confusing, especially when considering future patches that will further differentiate between "real" and "emulated" feature bits. No functional change intended. Reviewed-by:
-
Sean Christopherson authored
Drop the index param from do_cpuid_7_mask() and instead switch on the entry's index, which is guaranteed to be set by do_host_cpuid(). No functional change intended. Reviewed-by:
-
Sean Christopherson authored
Refactor the sub-leaf loop for CPUID 0x7 to move the main leaf out of said loop. The emitted code savings is basically a mirage, as the handling of the main leaf can easily be split to its own helper to avoid code bloat. No functional change intended. Reviewed-by:
-
Sean Christopherson authored
Increment the number of CPUID entries immediately after do_host_cpuid() in preparation for moving the logic into do_host_cpuid(). Handle the rare/impossible case of encountering a bogus sub-leaf by decrementing the number entries on failure. Reviewed-by:
-
Sean Christopherson authored
WARN if the save state size for a valid XCR0-managed sub-leaf is zero, which would indicate a KVM or CPU bug. Add a comment to explain why KVM WARNs so the reader doesn't have to tease out the relevant bits from Intel's SDM and KVM's XCR0/XSS code. Reviewed-by:
-
Sean Christopherson authored
Now that sub-leaf 1 is handled separately, verify the next sub-leaf is needed before rejecting KVM_GET_SUPPORTED_CPUID due to an insufficiently sized userspace array. Note, although this is technically a bug, it's not visible to userspace as KVM_GET_SUPPORTED_CPUID is guaranteed to fail on KVM_CPUID_SIGNATURE, which is hardcoded to be added after leaf 0xD. The real motivation for the change is to tightly couple the nent/maxnent and do_host_cpuid() sequences in preparation for future cleanup. Reviewed-by:
-
Sean Christopherson authored
Mov the sub-leaf 1 handling for CPUID 0xD out of the index>0 loop so that the loop only handles index>2. Sub-leafs 2+ have identical semantics, whereas sub-leaf 1 is effectively a feature sub-leaf. Moving sub-leaf 1 out of the loop does duplicate a bit of code, but the nent/maxnent code will be consolidated in a future patch, and duplicating the clear of ECX/EDX is arguably a good thing as the reasons for clearing said registers are completely different. No functional change intended. Reviewed-by:
-
Sean Christopherson authored
Verify that the next sub-leaf of CPUID 0x4 (or 0x8000001d) is valid before rejecting the entire KVM_GET_SUPPORTED_CPUID due to insufficent space in the userspace array. Note, although this is technically a bug, it's not visible to userspace as KVM_GET_SUPPORTED_CPUID is guaranteed to fail on KVM_CPUID_SIGNATURE, which is hardcoded to be added after the affected leafs. The real motivation for the change is to tightly couple the nent/maxnent and do_host_cpuid() sequences in preparation for future cleanup. Reviewed-by:
-
Sean Christopherson authored
Clean up the error handling in kvm_dev_ioctl_get_cpuid(), which has gotten a bit crusty as the function has evolved over the years. Opportunistically hoist the static @funcs declaration to the top of the function to make it more obvious that it's a "static const". No functional change intended. Reviewed-by:
-
Sean Christopherson authored
Refactor the handling of the Centaur-only CPUID leaf to detect the leaf via a runtime query instead of adding a one-off callback in the static array. When the callback was introduced, there were additional fields in the array's structs, and more importantly, retpoline wasn't a thing. No functional change intended. Reviewed-by:
-
Sean Christopherson authored
Move the guts of kvm_dev_ioctl_get_cpuid()'s CPUID func loop to a separate helper to improve code readability and pave the way for future cleanup. No functional change intended. Reviewed-by:
-
Sean Christopherson authored
Fix a long-standing bug that causes KVM to return 0 instead of -E2BIG when userspace's array is insufficiently sized. This technically breaks backwards compatibility, e.g. a userspace with a hardcoded cpuid->nent could theoretically be broken as it would see an error instead of success if cpuid->nent is less than the number of entries required to fully enumerate the host CPU. But, the lowest known cpuid->nent hardcoded by a VMM is 100 (lkvm and selftests), and the limit for current processors on Intel and AMD is well under a 100. E.g. Intel's Icelake server with all the bells and whistles tops out at ~60 entries (variable due to SGX sub-leafs), and AMD's CPUID documentation allows for less than 50. CPUID 0xD sub-leaves on current kernels are capped by the value of KVM_SUPPORTED_XCR0, and therefore so many subleaves cannot have appeared on current kernels. Note, while the Fixes: tag is accurate with respect to the immediate bug, it's likely that similar bugs in KVM_GET_SUPPORTED_CPUID existed prior to the refactoring, e.g. Qemu contains a workaround for the broken KVM_GET_SUPPORTED_CPUID behavior that predates the buggy commit by over two years. The Qemu workaround is also likely the main reason the bug has gone unreported for so long. Qemu hack: commit 76ae317f7c16aec6b469604b1764094870a75470 Author: Mark McLoughlin <markmc@redhat.com> Date: Tue May 19 18:55:21 2009 +0100 kvm: work around supported cpuid ioctl() brokenness KVM_GET_SUPPORTED_CPUID has been known to fail to return -E2BIG when it runs out of entries. Detect this by always trying again with a bigger table if the ioctl() fills the table. Fixes: 831bf664 ("KVM: Refactor and simplify kvm_dev_ioctl_get_supported_cpuid") Reviewed-by:
-
