| Age | Commit message (Collapse) | Author |
|---|
|
Normal include order is that linux/foo.h should include asm/foo.h, CFI has it
the wrong way around.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Sami Tolvanen <samitolvanen@google.com>
Link: https://lore.kernel.org/r/20231215092707.231038174@infradead.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
|
|
The first few generations of TDX hardware have an erratum. Triggering
it in Linux requires some kind of kernel bug involving relatively exotic
memory writes to TDX private memory and will manifest via
spurious-looking machine checks when reading the affected memory.
Make an effort to detect these TDX-induced machine checks and spit out
a new blurb to dmesg so folks do not think their hardware is failing.
== Background ==
Virtually all kernel memory accesses operations happen in full
cachelines. In practice, writing a "byte" of memory usually reads a 64
byte cacheline of memory, modifies it, then writes the whole line back.
Those operations do not trigger this problem.
This problem is triggered by "partial" writes where a write transaction
of less than cacheline lands at the memory controller. The CPU does
these via non-temporal write instructions (like MOVNTI), or through
UC/WC memory mappings. The issue can also be triggered away from the
CPU by devices doing partial writes via DMA.
== Problem ==
A partial write to a TDX private memory cacheline will silently "poison"
the line. Subsequent reads will consume the poison and generate a
machine check. According to the TDX hardware spec, neither of these
things should have happened.
To add insult to injury, the Linux machine code will present these as a
literal "Hardware error" when they were, in fact, a software-triggered
issue.
== Solution ==
In the end, this issue is hard to trigger. Rather than do something
rash (and incomplete) like unmap TDX private memory from the direct map,
improve the machine check handler.
Currently, the #MC handler doesn't distinguish whether the memory is
TDX private memory or not but just dump, for instance, below message:
[...] mce: [Hardware Error]: CPU 147: Machine Check Exception: f Bank 1: bd80000000100134
[...] mce: [Hardware Error]: RIP 10:<ffffffffadb69870> {__tlb_remove_page_size+0x10/0xa0}
...
[...] mce: [Hardware Error]: Run the above through 'mcelog --ascii'
[...] mce: [Hardware Error]: Machine check: Data load in unrecoverable area of kernel
[...] Kernel panic - not syncing: Fatal local machine check
Which says "Hardware Error" and "Data load in unrecoverable area of
kernel".
Ideally, it's better for the log to say "software bug around TDX private
memory" instead of "Hardware Error". But in reality the real hardware
memory error can happen, and sadly such software-triggered #MC cannot be
distinguished from the real hardware error. Also, the error message is
used by userspace tool 'mcelog' to parse, so changing the output may
break userspace.
So keep the "Hardware Error". The "Data load in unrecoverable area of
kernel" is also helpful, so keep it too.
Instead of modifying above error log, improve the error log by printing
additional TDX related message to make the log like:
...
[...] mce: [Hardware Error]: Machine check: Data load in unrecoverable area of kernel
[...] mce: [Hardware Error]: Machine Check: TDX private memory error. Possible kernel bug.
Adding this additional message requires determination of whether the
memory page is TDX private memory. There is no existing infrastructure
to do that. Add an interface to query the TDX module to fill this gap.
== Impact ==
This issue requires some kind of kernel bug to trigger.
TDX private memory should never be mapped UC/WC. A partial write
originating from these mappings would require *two* bugs, first mapping
the wrong page, then writing the wrong memory. It would also be
detectable using traditional memory corruption techniques like
DEBUG_PAGEALLOC.
MOVNTI (and friends) could cause this issue with something like a simple
buffer overrun or use-after-free on the direct map. It should also be
detectable with normal debug techniques.
The one place where this might get nasty would be if the CPU read data
then wrote back the same data. That would trigger this problem but
would not, for instance, set off mechanisms like slab redzoning because
it doesn't actually corrupt data.
With an IOMMU at least, the DMA exposure is similar to the UC/WC issue.
TDX private memory would first need to be incorrectly mapped into the
I/O space and then a later DMA to that mapping would actually cause the
poisoning event.
[ dhansen: changelog tweaks ]
Signed-off-by: Kai Huang <kai.huang@intel.com>
Signed-off-by: Dave Hansen <dave.hansen@linux.intel.com>
Reviewed-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Reviewed-by: Yuan Yao <yuan.yao@intel.com>
Reviewed-by: Dave Hansen <dave.hansen@linux.intel.com>
Reviewed-by: Tony Luck <tony.luck@intel.com>
Link: https://lore.kernel.org/all/20231208170740.53979-18-dave.hansen%40intel.com
|
|
TDX memory has integrity and confidentiality protections. Violations of
this integrity protection are supposed to only affect TDX operations and
are never supposed to affect the host kernel itself. In other words,
the host kernel should never, itself, see machine checks induced by the
TDX integrity hardware.
Alas, the first few generations of TDX hardware have an erratum. A
partial write to a TDX private memory cacheline will silently "poison"
the line. Subsequent reads will consume the poison and generate a
machine check. According to the TDX hardware spec, neither of these
things should have happened.
Virtually all kernel memory accesses operations happen in full
cachelines. In practice, writing a "byte" of memory usually reads a 64
byte cacheline of memory, modifies it, then writes the whole line back.
Those operations do not trigger this problem.
This problem is triggered by "partial" writes where a write transaction
of less than cacheline lands at the memory controller. The CPU does
these via non-temporal write instructions (like MOVNTI), or through
UC/WC memory mappings. The issue can also be triggered away from the
CPU by devices doing partial writes via DMA.
With this erratum, there are additional things need to be done. To
prepare for those changes, add a CPU bug bit to indicate this erratum.
Note this bug reflects the hardware thus it is detected regardless of
whether the kernel is built with TDX support or not.
Signed-off-by: Kai Huang <kai.huang@intel.com>
Signed-off-by: Dave Hansen <dave.hansen@linux.intel.com>
Reviewed-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Dave Hansen <dave.hansen@linux.intel.com>
Link: https://lore.kernel.org/all/20231208170740.53979-17-dave.hansen%40intel.com
|
|
Add a synthetic feature flag specifically for first generation Zen
machines. There's need to have a generic flag for all Zen generations so
make X86_FEATURE_ZEN be that flag.
Fixes: 30fa92832f40 ("x86/CPU/AMD: Add ZenX generations flags")
Suggested-by: Brian Gerst <brgerst@gmail.com>
Suggested-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de>
Link: https://lore.kernel.org/r/dc3835e3-0731-4230-bbb9-336bbe3d042b@amd.com
|
|
Teach sparse about __seg_fs and __seg_gs named address space
qualifiers to to avoid warnings about unexpected keyword at
the end of cast operator.
Reported-by: kernel test robot <lkp@intel.com>
Acked-by: Luc Van Oostenryck <luc.vanoostenryck@gmail.com>
Signed-off-by: Uros Bizjak <ubizjak@gmail.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20231204210320.114429-3-ubizjak@gmail.com
Closes: https://lore.kernel.org/oe-kbuild-all/202310080853.UhMe5iWa-lkp@intel.com/
|
|
Version generation for "const_pcpu_hot" symbol failed because genksyms
doesn't know the __seg_gs keyword. Effectively revert commit 4604c052b84d
("x86/percpu: Declare const_pcpu_hot as extern const variable") and
use this_cpu_read_const() instead to avoid "sparse: dereference of
noderef expression" warning when reading const_pcpu_hot.
Signed-off-by: Uros Bizjak <ubizjak@gmail.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Link: https://lore.kernel.org/r/20231204210320.114429-1-ubizjak@gmail.com
|
|
Issue a short message once, on the first try to load a 32-bit process to
save people time when wondering why it won't load and trying to execute
it, would say:
-bash: ./strsep32: cannot execute binary file: Exec format error
Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de>
Reviewed-by: Nikolay Borisov <nik.borisov@suse.com>
Link: https://lore.kernel.org/r/20231130155213.1407-1-bp@alien8.de
|
|
Now that paravirt is using the alternatives patching infrastructure,
remove the paravirt patching code.
Signed-off-by: Juergen Gross <jgross@suse.com>
Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/r/20231210062138.2417-6-jgross@suse.com
|
|
Instead of stacking alternative and paravirt patching, use the new
ALT_FLAG_CALL flag to switch those mixed calls to pure alternative
handling.
Eliminate the need to be careful regarding the sequence of alternative
and paravirt patching.
[ bp: Touch up commit message. ]
Signed-off-by: Juergen Gross <jgross@suse.com>
Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/r/20231210062138.2417-5-jgross@suse.com
|
|
In order to prepare replacing of paravirt patching with alternative
patching, add the capability to replace an indirect call with a direct
one.
This is done via a new flag ALT_FLAG_CALL as the target of the CALL
instruction needs to be evaluated using the value of the location
addressed by the indirect call.
For convenience, add a macro for a default CALL instruction. In case it
is being used without the new flag being set, it will result in a BUG()
when being executed. As in most cases, the feature used will be
X86_FEATURE_ALWAYS so add another macro ALT_CALL_ALWAYS usable for the
flags parameter of the ALTERNATIVE macros.
For a complete replacement, handle the special cases of calling a nop
function and an indirect call of NULL the same way as paravirt does.
[ bp: Massage commit message, fixup the debug output and clarify flow
more. ]
Co-developed-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Juergen Gross <jgross@suse.com>
Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de>
Link: https://lore.kernel.org/r/20231210062138.2417-4-jgross@suse.com
|
|
As a preparation for replacing paravirt patching completely by
alternative patching, move some backend functions and #defines to
the alternatives code and header.
Signed-off-by: Juergen Gross <jgross@suse.com>
Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de>
Link: https://lore.kernel.org/r/20231129133332.31043-3-jgross@suse.com
|
|
Introduce the macro ALT_NOT_XEN as a short form of
ALT_NOT(X86_FEATURE_XENPV).
No functional changes.
Suggested-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Juergen Gross <jgross@suse.com>
Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de>
Link: https://lore.kernel.org/r/20231129133332.31043-2-jgross@suse.com
|
|
The TDX module global metadata provides system-wide information about
the module.
TL;DR:
Use the TDH.SYS.RD SEAMCALL to tell if the module is good or not.
Long Version:
1) Only initialize TDX module with version 1.5 and later
TDX module 1.0 has some compatibility issues with the later versions of
module, as documented in the "Intel TDX module ABI incompatibilities
between TDX1.0 and TDX1.5" spec. Don't bother with module versions that
do not have a stable ABI.
2) Get the essential global metadata for module initialization
TDX reports a list of "Convertible Memory Region" (CMR) to tell the
kernel which memory is TDX compatible. The kernel needs to build a list
of memory regions (out of CMRs) as "TDX-usable" memory and pass them to
the TDX module. The kernel does this by constructing a list of "TD
Memory Regions" (TDMRs) to cover all these memory regions and passing
them to the TDX module.
Each TDMR is a TDX architectural data structure containing the memory
region that the TDMR covers, plus the information to track (within this
TDMR):
a) the "Physical Address Metadata Table" (PAMT) to track each TDX
memory page's status (such as which TDX guest "owns" a given page,
and
b) the "reserved areas" to tell memory holes that cannot be used as
TDX memory.
The kernel needs to get below metadata from the TDX module to build the
list of TDMRs:
a) the maximum number of supported TDMRs
b) the maximum number of supported reserved areas per TDMR and,
c) the PAMT entry size for each TDX-supported page size.
== Implementation ==
The TDX module has two modes of fetching the metadata: a one field at
a time, or all in one blob. Use the field at a time for now. It is
slower, but there just are not enough fields now to justify the
complexity of extra unpacking.
The err_free_tdxmem=>out_put_tdxmem goto looks wonky by itself. But
it is the first of a bunch of error handling that will get stuck at
its site.
[ dhansen: clean up changelog and add a struct to map between
the TDX module fields and 'struct tdx_tdmr_sysinfo' ]
Signed-off-by: Kai Huang <kai.huang@intel.com>
Signed-off-by: Dave Hansen <dave.hansen@linux.intel.com>
Link: https://lore.kernel.org/all/20231208170740.53979-8-dave.hansen%40intel.com
|
|
There are essentially two steps to get the TDX module ready:
1) Get each CPU ready to run TDX
2) Set up the shared TDX module data structures
Introduce and export (to KVM) the infrastructure to do both of these
pieces at runtime.
== Per-CPU TDX Initialization ==
Track the initialization status of each CPU with a per-cpu variable.
This avoids failures in the case of KVM module reloads and handles cases
where CPUs come online later.
Generally, the per-cpu SEAMCALLs happen first. But there's actually one
global call that has to happen before _any_ others (TDH_SYS_INIT). It's
analogous to the boot CPU having to do a bit of extra work just because
it happens to be the first one. Track if _any_ CPU has done this call
and then only actually do it during the first per-cpu init.
== Shared TDX Initialization ==
Create the global state function (tdx_enable()) as a simple placeholder.
The TODO list will be pared down as functionality is added.
Use a state machine protected by mutex to make sure the work in
tdx_enable() will only be done once. This avoids failures if the KVM
module is reloaded.
A CPU must be made ready to run TDX before it can participate in
initializing the shared parts of the module. Any caller of tdx_enable()
need to ensure that it can never run on a CPU which is not ready to
run TDX. It needs to be wary of CPU hotplug, preemption and the
VMX enabling state of any CPU on which it might run.
== Why runtime instead of boot time? ==
The TDX module can be initialized only once in its lifetime. Instead
of always initializing it at boot time, this implementation chooses an
"on demand" approach to initialize TDX until there is a real need (e.g
when requested by KVM). This approach has below pros:
1) It avoids consuming the memory that must be allocated by kernel and
given to the TDX module as metadata (~1/256th of the TDX-usable memory),
and also saves the CPU cycles of initializing the TDX module (and the
metadata) when TDX is not used at all.
2) The TDX module design allows it to be updated while the system is
running. The update procedure shares quite a few steps with this "on
demand" initialization mechanism. The hope is that much of "on demand"
mechanism can be shared with a future "update" mechanism. A boot-time
TDX module implementation would not be able to share much code with the
update mechanism.
3) Making SEAMCALL requires VMX to be enabled. Currently, only the KVM
code mucks with VMX enabling. If the TDX module were to be initialized
separately from KVM (like at boot), the boot code would need to be
taught how to muck with VMX enabling and KVM would need to be taught how
to cope with that. Making KVM itself responsible for TDX initialization
lets the rest of the kernel stay blissfully unaware of VMX.
[ dhansen: completely reorder/rewrite changelog ]
Signed-off-by: Kai Huang <kai.huang@intel.com>
Signed-off-by: Dave Hansen <dave.hansen@linux.intel.com>
Reviewed-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Reviewed-by: Nikolay Borisov <nik.borisov@suse.com>
Reviewed-by: Dave Hansen <dave.hansen@linux.intel.com>
Link: https://lore.kernel.org/all/20231208170740.53979-6-dave.hansen%40intel.com
|
|
The SEAMCALLs involved during the TDX module initialization are not
expected to fail. In fact, they are not expected to return any non-zero
code (except the "running out of entropy error", which can be handled
internally already).
Add yet another set of SEAMCALL wrappers, which treats all non-zero
return code as error, to support printing SEAMCALL error upon failure
for module initialization. Note the TDX module initialization doesn't
use the _saved_ret() variant thus no wrapper is added for it.
SEAMCALL assembly can also return kernel-defined error codes for three
special cases: 1) TDX isn't enabled by the BIOS; 2) TDX module isn't
loaded; 3) CPU isn't in VMX operation. Whether they can legally happen
depends on the caller, so leave to the caller to print error message
when desired.
Also convert the SEAMCALL error codes to the kernel error codes in the
new wrappers so that each SEAMCALL caller doesn't have to repeat the
conversion.
[ dhansen: Align the register dump with show_regs(). Zero-pad the
contents, split on two lines and use consistent spacing. ]
Signed-off-by: Kai Huang <kai.huang@intel.com>
Signed-off-by: Dave Hansen <dave.hansen@linux.intel.com>
Reviewed-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Reviewed-by: Kuppuswamy Sathyanarayanan <sathyanarayanan.kuppuswamy@linux.intel.com>
Reviewed-by: Dave Hansen <dave.hansen@linux.intel.com>
Link: https://lore.kernel.org/all/20231208170740.53979-5-dave.hansen%40intel.com
|
|
Some SEAMCALLs use the RDRAND hardware and can fail for the same reasons
as RDRAND. Use the kernel RDRAND retry logic for them.
There are three __seamcall*() variants. Do the SEAMCALL retry in common
code and add a wrapper for each of them.
Signed-off-by: Kai Huang <kai.huang@intel.com>
Signed-off-by: Dave Hansen <dave.hansen@linux.intel.com>
Reviewed-by: Kirill A. Shutemov <kirll.shutemov@linux.intel.com>
Reviewed-by: Kuppuswamy Sathyanarayanan <sathyanarayanan.kuppuswamy@linux.intel.com>
Reviewed-by: Dave Hansen <dave.hansen@linux.intel.com>
Link: https://lore.kernel.org/all/20231208170740.53979-4-dave.hansen%40intel.com
|
|
TDX supports 4K, 2M and 1G page sizes. The corresponding values are
defined by the TDX module spec and used as TDX module ABI. Currently,
they are used in try_accept_one() when the TDX guest tries to accept a
page. However currently try_accept_one() uses hard-coded magic values.
Define TDX supported page sizes as macros and get rid of the hard-coded
values in try_accept_one(). TDX host support will need to use them too.
Signed-off-by: Kai Huang <kai.huang@intel.com>
Signed-off-by: Dave Hansen <dave.hansen@linux.intel.com>
Reviewed-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Reviewed-by: Dave Hansen <dave.hansen@linux.intel.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Link: https://lore.kernel.org/all/20231208170740.53979-2-dave.hansen%40intel.com
|
|
Intel Trust Domain Extensions (TDX) protects guest VMs from malicious
host and certain physical attacks. A CPU-attested software module
called 'the TDX module' runs inside a new isolated memory range as a
trusted hypervisor to manage and run protected VMs.
Pre-TDX Intel hardware has support for a memory encryption architecture
called MKTME. The memory encryption hardware underpinning MKTME is also
used for Intel TDX. TDX ends up "stealing" some of the physical address
space from the MKTME architecture for crypto-protection to VMs. The
BIOS is responsible for partitioning the "KeyID" space between legacy
MKTME and TDX. The KeyIDs reserved for TDX are called 'TDX private
KeyIDs' or 'TDX KeyIDs' for short.
During machine boot, TDX microcode verifies that the BIOS programmed TDX
private KeyIDs consistently and correctly programmed across all CPU
packages. The MSRs are locked in this state after verification. This
is why MSR_IA32_MKTME_KEYID_PARTITIONING gets used for TDX enumeration:
it indicates not just that the hardware supports TDX, but that all the
boot-time security checks passed.
The TDX module is expected to be loaded by the BIOS when it enables TDX,
but the kernel needs to properly initialize it before it can be used to
create and run any TDX guests. The TDX module will be initialized by
the KVM subsystem when KVM wants to use TDX.
Detect platform TDX support by detecting TDX private KeyIDs.
The TDX module itself requires one TDX KeyID as the 'TDX global KeyID'
to protect its metadata. Each TDX guest also needs a TDX KeyID for its
own protection. Just use the first TDX KeyID as the global KeyID and
leave the rest for TDX guests. If no TDX KeyID is left for TDX guests,
disable TDX as initializing the TDX module alone is useless.
[ dhansen: add X86_FEATURE, replace helper function ]
Signed-off-by: Kai Huang <kai.huang@intel.com>
Signed-off-by: Dave Hansen <dave.hansen@linux.intel.com>
Reviewed-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Reviewed-by: Isaku Yamahata <isaku.yamahata@intel.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Dave Hansen <dave.hansen@linux.intel.com>
Reviewed-by: Kuppuswamy Sathyanarayanan <sathyanarayanan.kuppuswamy@linux.intel.com>
Link: https://lore.kernel.org/all/20231208170740.53979-1-dave.hansen%40intel.com
|
|
There is no real reason to have a separate ASM entry point implementation
for the legacy INT 0x80 syscall emulation on 64-bit.
IDTENTRY provides all the functionality needed with the only difference
that it does not:
- save the syscall number (AX) into pt_regs::orig_ax
- set pt_regs::ax to -ENOSYS
Both can be done safely in the C code of an IDTENTRY before invoking any of
the syscall related functions which depend on this convention.
Aside of ASM code reduction this prepares for detecting and handling a
local APIC injected vector 0x80.
[ kirill.shutemov: More verbose comments ]
Suggested-by: Linus Torvalds <torvalds@linuxfoundation.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Dave Hansen <dave.hansen@linux.intel.com>
Reviewed-by: Borislav Petkov (AMD) <bp@alien8.de>
Cc: <stable@vger.kernel.org> # v6.0+
|
|
The INT 0x80 instruction is used for 32-bit x86 Linux syscalls. The
kernel expects to receive a software interrupt as a result of the INT
0x80 instruction. However, an external interrupt on the same vector
triggers the same handler.
The kernel interprets an external interrupt on vector 0x80 as a 32-bit
system call that came from userspace.
A VMM can inject external interrupts on any arbitrary vector at any
time. This remains true even for TDX and SEV guests where the VMM is
untrusted.
Put together, this allows an untrusted VMM to trigger int80 syscall
handling at any given point. The content of the guest register file at
that moment defines what syscall is triggered and its arguments. It
opens the guest OS to manipulation from the VMM side.
Disable 32-bit emulation by default for TDX and SEV. User can override
it with the ia32_emulation=y command line option.
[ dhansen: reword the changelog ]
Reported-by: Supraja Sridhara <supraja.sridhara@inf.ethz.ch>
Reported-by: Benedict Schlüter <benedict.schlueter@inf.ethz.ch>
Reported-by: Mark Kuhne <mark.kuhne@inf.ethz.ch>
Reported-by: Andrin Bertschi <andrin.bertschi@inf.ethz.ch>
Reported-by: Shweta Shinde <shweta.shinde@inf.ethz.ch>
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Dave Hansen <dave.hansen@linux.intel.com>
Reviewed-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Borislav Petkov (AMD) <bp@alien8.de>
Cc: <stable@vger.kernel.org> # v6.0+: 1da5c9b x86: Introduce ia32_enabled()
Cc: <stable@vger.kernel.org> # v6.0+
|
|
Hyper-V emulation in KVM is a fairly big chunk and in some cases it may be
desirable to not compile it in to reduce module sizes as well as the attack
surface. Introduce CONFIG_KVM_HYPERV option to make it possible.
Note, there's room for further nVMX/nSVM code optimizations when
!CONFIG_KVM_HYPERV, this will be done in follow-up patches.
Reorganize Makefile a bit so all CONFIG_HYPERV and CONFIG_KVM_HYPERV files
are grouped together.
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Reviewed-by: Maxim Levitsky <mlevitsk@redhat.com>
Tested-by: Jeremi Piotrowski <jpiotrowski@linux.microsoft.com>
Link: https://lore.kernel.org/r/20231205103630.1391318-13-vkuznets@redhat.com
Signed-off-by: Sean Christopherson <seanjc@google.com>
|
|
Hyper-V partition assist page is used when KVM runs on top of Hyper-V and
is not used for Windows/Hyper-V guests on KVM, this means that 'hv_pa_pg'
placement in 'struct kvm_hv' is unfortunate. As a preparation to making
Hyper-V emulation optional, move 'hv_pa_pg' to 'struct kvm_arch' and put it
under CONFIG_HYPERV.
While on it, introduce hv_get_partition_assist_page() helper to allocate
partition assist page. Move the comment explaining why we use a single page
for all vCPUs from VMX and expand it a bit.
No functional change intended.
Reviewed-by: Maxim Levitsky <mlevitsk@redhat.com>
Tested-by: Jeremi Piotrowski <jpiotrowski@linux.microsoft.com>
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Link: https://lore.kernel.org/r/20231205103630.1391318-3-vkuznets@redhat.com
Signed-off-by: Sean Christopherson <seanjc@google.com>
|
|
Saving a few bytes of memory per KVM VM is certainly great but what's more
important is the ability to see where the code accesses Xen emulation
context while CONFIG_KVM_XEN is not enabled. Currently, kvm_cpu_get_extint()
is the only such place and it is harmless: kvm_xen_has_interrupt() always
returns '0' when !CONFIG_KVM_XEN.
No functional change intended.
Reviewed-by: Maxim Levitsky <mlevitsk@redhat.com>
Tested-by: Jeremi Piotrowski <jpiotrowski@linux.microsoft.com>
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Link: https://lore.kernel.org/r/20231205103630.1391318-2-vkuznets@redhat.com
Signed-off-by: Sean Christopherson <seanjc@google.com>
|
|
Now that GENERIC_CPU_DEVICES calls arch_register_cpu(), which can be
overridden by the arch code, switch over to this to allow common code
to choose when the register_cpu() call is made.
x86's struct cpus come from struct x86_cpu, which has no other members
or users. Remove this and use the version defined by common code.
This is an intermediate step to the logic being moved to drivers/acpi,
where GENERIC_CPU_DEVICES will do the work when booting with acpi=off.
This patch also has the effect of moving the registration of CPUs from
subsys to driver core initialisation, prior to any initcalls running.
----
Changes since RFC:
* Fixed the second copy of arch_register_cpu() used for non-hotplug
Changes since RFC v2:
* Remove duplicate of the weak generic arch_register_cpu(), spotted
by Jonathan Cameron. Add note about initialisation order change.
Changes since RFC v3:
* Adapt to removal of EXPORT_SYMBOL()s
Signed-off-by: James Morse <james.morse@arm.com>
Reviewed-by: Gavin Shan <gshan@redhat.com>
Signed-off-by: "Russell King (Oracle)" <rmk+kernel@armlinux.org.uk>
Reviewed-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Reviewed-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/r/E1r5R3l-00Cszm-UA@rmk-PC.armlinux.org.uk
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
|
|
It is cheap to take tdp_mmu_pages_lock in all write-side critical sections.
We already do it all the time when zapping with read_lock(), so it is not
a problem to do it from the kvm_tdp_mmu_zap_all() path (aka
kvm_arch_flush_shadow_all(), aka VM destruction and MMU notifier release).
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Link: https://lore.kernel.org/r/20231125083400.1399197-4-pbonzini@redhat.com
Signed-off-by: Sean Christopherson <seanjc@google.com>
|
|
Explicitly track emulated counter events instead of using the common
counter value that's shared with the hardware counter owned by perf.
Bumping the common counter requires snapshotting the pre-increment value
in order to detect overflow from emulation, and the snapshot approach is
inherently flawed.
Snapshotting the previous counter at every increment assumes that there is
at most one emulated counter event per emulated instruction (or rather,
between checks for KVM_REQ_PMU). That's mostly holds true today because
KVM only emulates (branch) instructions retired, but the approach will
fall apart if KVM ever supports event types that don't have a 1:1
relationship with instructions.
And KVM already has a relevant bug, as handle_invalid_guest_state()
emulates multiple instructions without checking KVM_REQ_PMU, i.e. could
miss an overflow event due to clobbering pmc->prev_counter. Not checking
KVM_REQ_PMU is problematic in both cases, but at least with the emulated
counter approach, the resulting behavior is delayed overflow detection,
as opposed to completely lost detection.
Tracking the emulated count fixes another bug where the snapshot approach
can signal spurious overflow due to incorporating both the emulated count
and perf's count in the check, i.e. if overflow is detected by perf, then
KVM's emulation will also incorrectly signal overflow. Add a comment in
the related code to call out the need to process emulated events *after*
pausing the perf event (big kudos to Mingwei for figuring out that
particular wrinkle).
Cc: Mingwei Zhang <mizhang@google.com>
Cc: Roman Kagan <rkagan@amazon.de>
Cc: Jim Mattson <jmattson@google.com>
Cc: Dapeng Mi <dapeng1.mi@linux.intel.com>
Cc: Like Xu <like.xu.linux@gmail.com>
Reviewed-by: Mingwei Zhang <mizhang@google.com>
Link: https://lore.kernel.org/r/20231103230541.352265-7-seanjc@google.com
Signed-off-by: Sean Christopherson <seanjc@google.com>
|
|
Move the common (or at least "ignored") aspects of resetting the vPMU to
common x86 code, along with the stop/release helpers that are no used only
by the common pmu.c.
There is no need to manually handle fixed counters as all_valid_pmc_idx
tracks both fixed and general purpose counters, and resetting the vPMU is
far from a hot path, i.e. the extra bit of overhead to the PMC from the
index is a non-issue.
Zero fixed_ctr_ctrl in common code even though it's Intel specific.
Ensuring it's zero doesn't harm AMD/SVM in any way, and stopping the fixed
counters via all_valid_pmc_idx, but not clearing the associated control
bits, would be odd/confusing.
Make the .reset() hook optional as SVM no longer needs vendor specific
handling.
Cc: stable@vger.kernel.org
Reviewed-by: Dapeng Mi <dapeng1.mi@linux.intel.com>
Link: https://lore.kernel.org/r/20231103230541.352265-2-seanjc@google.com
Signed-off-by: Sean Christopherson <seanjc@google.com>
|
|
const_pcpu_hot is aliased by linker to pcpu_hot, so there is no need
to use the DECLARE_PER_CPU_ALIGNED() macro. Also, declare const_pcpu_hot
as extern to avoid allocating storage space for the aliased structure.
Fixes: ed2f752e0e0a ("x86/percpu: Introduce const-qualified const_pcpu_hot to micro-optimize code generation")
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Uros Bizjak <ubizjak@gmail.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20231130162949.83518-1-ubizjak@gmail.com
Closes: https://lore.kernel.org/oe-kbuild-all/202311302257.tSFtZnly-lkp@intel.com/
|
|
Do it like the rest of the methods using it.
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Cc: Uros Bizjak <ubizjak@gmail.com>
Link: https://lore.kernel.org/r/20231105213731.1878100-3-ubizjak@gmail.com
|
|
Currently thunk debug macros explicitly define %gs: segment register
prefix for their percpu variables. This is not compatible with
!CONFIG_SMP, which requires non-prefixed percpu variables.
Fix call thunks debug macros to use PER_CPU_VAR macro from percpu.h
to conditionally use %gs: segment register prefix, depending on
CONFIG_SMP.
Finally, unify ASM_ prefixed assembly macros with their non-prefixed
variants. With support of %rip-relative relocations in place, call
thunk templates allow %rip-relative addressing, so unified assembly
snippet can be used everywhere.
Signed-off-by: Uros Bizjak <ubizjak@gmail.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20231105213731.1878100-4-ubizjak@gmail.com
|
|
Contrary to alternatives, relocations are currently not supported in
call thunk templates. Re-use the existing infrastructure from
alternative.c to allow %rip-relative relocations when copying call
thunk template from its storage location.
The patch allows unification of ASM_INCREMENT_CALL_DEPTH, which already
uses PER_CPU_VAR macro, with INCREMENT_CALL_DEPTH, used in call thunk
template, which is currently limited to use absolute address.
Reuse existing relocation infrastructure from alternative.c.,
as suggested by Peter Zijlstra.
Signed-off-by: Uros Bizjak <ubizjak@gmail.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20231105213731.1878100-3-ubizjak@gmail.com
|
|
Some C source files define 'asm' statements that use PER_CPU_VAR,
so make PER_CPU_VAR macro available also without __ASSEMBLY__.
Signed-off-by: Uros Bizjak <ubizjak@gmail.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20231105213731.1878100-2-ubizjak@gmail.com
|
|
Declare the kvm_x86_ops hooks used to wire up paravirt TLB flushes when
running under Hyper-V if and only if CONFIG_HYPERV!=n. Wrapping yet more
code with IS_ENABLED(CONFIG_HYPERV) eliminates a handful of conditional
branches, and makes it super obvious why the hooks *might* be valid.
Cc: Vitaly Kuznetsov <vkuznets@redhat.com>
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Link: https://lore.kernel.org/r/20231018192325.1893896-1-seanjc@google.com
Signed-off-by: Sean Christopherson <seanjc@google.com>
|
|
intel_idle_irq() re-enables IRQs very early. As a result, an interrupt
may fire before mwait() is eventually called. If such an interrupt queues
a timer, it may go unnoticed until mwait returns and the idle loop
handles the tick re-evaluation. And monitoring TIF_NEED_RESCHED doesn't
help because a local timer enqueue doesn't set that flag.
The issue is mitigated by the fact that this idle handler is only invoked
for shallow C-states when, presumably, the next tick is supposed to be
close enough. There may still be rare cases though when the next tick
is far away and the selected C-state is shallow, resulting in a timer
getting ignored for a while.
Fix this with using sti_mwait() whose IRQ-reenablement only triggers
upon calling mwait(), dealing with the race while keeping the interrupt
latency within acceptable bounds.
Fixes: c227233ad64c (intel_idle: enable interrupts before C1 on Xeons)
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Rafael J. Wysocki <rafael@kernel.org>
Link: https://lkml.kernel.org/r/20231115151325.6262-3-frederic@kernel.org
|
|
Add a note to make sure we never miss and break the requirements behind
it.
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Rafael J. Wysocki <rafael@kernel.org>
Link: https://lkml.kernel.org/r/20231115151325.6262-2-frederic@kernel.org
|
|
Add X86_FEATURE flags for each Zen generation. They should be used from
now on instead of checking f/m/s.
Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de>
Reviewed-by: Nikolay Borisov <nik.borisov@suse.com>
Acked-by: Thomas Gleixner <tglx@linutronix.de>
Link: http://lore.kernel.org/r/20231120104152.13740-2-bp@alien8.de
|
|
Add support to allow guests to set the new CR4 control bit for LAM and add
implementation to get untagged address for supervisor pointers.
LAM modifies the canonicality check applied to 64-bit linear addresses for
data accesses, allowing software to use of the untranslated address bits for
metadata and masks the metadata bits before using them as linear addresses
to access memory. LAM uses CR4.LAM_SUP (bit 28) to configure and enable LAM
for supervisor pointers. It also changes VMENTER to allow the bit to be set
in VMCS's HOST_CR4 and GUEST_CR4 to support virtualization. Note CR4.LAM_SUP
is allowed to be set even not in 64-bit mode, but it will not take effect
since LAM only applies to 64-bit linear addresses.
Move CR4.LAM_SUP out of CR4_RESERVED_BITS, its reservation depends on vcpu
supporting LAM or not. Leave it intercepted to prevent guest from setting
the bit if LAM is not exposed to guest as well as to avoid vmread every time
when KVM fetches its value, with the expectation that guest won't toggle the
bit frequently.
Set CR4.LAM_SUP bit in the emulated IA32_VMX_CR4_FIXED1 MSR for guests to
allow guests to enable LAM for supervisor pointers in nested VMX operation.
Hardware is not required to do TLB flush when CR4.LAM_SUP toggled, KVM
doesn't need to emulate TLB flush based on it. There's no other features
or vmx_exec_controls connection, and no other code needed in
{kvm,vmx}_set_cr4().
Skip address untag for instruction fetches (which includes branch targets),
operand of INVLPG instructions, and implicit system accesses, all of which
are not subject to untagging. Note, get_untagged_addr() isn't invoked for
implicit system accesses as there is no reason to do so, but check the
flag anyways for documentation purposes.
Signed-off-by: Robert Hoo <robert.hu@linux.intel.com>
Co-developed-by: Binbin Wu <binbin.wu@linux.intel.com>
Signed-off-by: Binbin Wu <binbin.wu@linux.intel.com>
Reviewed-by: Chao Gao <chao.gao@intel.com>
Reviewed-by: Kai Huang <kai.huang@intel.com>
Tested-by: Xuelian Guo <xuelian.guo@intel.com>
Link: https://lore.kernel.org/r/20230913124227.12574-11-binbin.wu@linux.intel.com
Signed-off-by: Sean Christopherson <seanjc@google.com>
|
|
Introduce a new interface get_untagged_addr() to kvm_x86_ops to untag
the metadata from linear address. Call the interface in linearization
of instruction emulator for 64-bit mode.
When enabled feature like Intel Linear Address Masking (LAM) or AMD Upper
Address Ignore (UAI), linear addresses may be tagged with metadata that
needs to be dropped prior to canonicality checks, i.e. the metadata is
ignored.
Introduce get_untagged_addr() to kvm_x86_ops to hide the vendor specific
code, as sadly LAM and UAI have different semantics. Pass the emulator
flags to allow vendor specific implementation to precisely identify the
access type (LAM doesn't untag certain accesses).
Signed-off-by: Binbin Wu <binbin.wu@linux.intel.com>
Reviewed-by: Chao Gao <chao.gao@intel.com>
Tested-by: Xuelian Guo <xuelian.guo@intel.com>
Link: https://lore.kernel.org/r/20230913124227.12574-9-binbin.wu@linux.intel.com
[sean: massage changelog]
Signed-off-by: Sean Christopherson <seanjc@google.com>
|
|
Add HWID and McaType values for new SMCA bank types.
Signed-off-by: Muralidhara M K <muralidhara.mk@amd.com>
Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de>
Link: https://lore.kernel.org/r/20231102114225.2006878-3-muralimk@amd.com
|
|
The long names of the SMCA banks are only used by the MCE decoder
module.
Move them out of the arch code and into the decoder module.
[ bp: Name the long names array "smca_long_names", drop local ptr in
decode_smca_error(), constify arrays. ]
Signed-off-by: Yazen Ghannam <yazen.ghannam@amd.com>
Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de>
Link: https://lore.kernel.org/r/20231118193248.1296798-5-yazen.ghannam@amd.com
|
|
The VDSO functions are defined as globals in the kernel sources but intended
to be called from userspace, so there is no need to declare them in a kernel
side header.
Without a prototype, this now causes warnings such as
arch/mips/vdso/vgettimeofday.c:14:5: error: no previous prototype for '__vdso_clock_gettime' [-Werror=missing-prototypes]
arch/mips/vdso/vgettimeofday.c:28:5: error: no previous prototype for '__vdso_gettimeofday' [-Werror=missing-prototypes]
arch/mips/vdso/vgettimeofday.c:36:5: error: no previous prototype for '__vdso_clock_getres' [-Werror=missing-prototypes]
arch/mips/vdso/vgettimeofday.c:42:5: error: no previous prototype for '__vdso_clock_gettime64' [-Werror=missing-prototypes]
arch/sparc/vdso/vclock_gettime.c:254:1: error: no previous prototype for '__vdso_clock_gettime' [-Werror=missing-prototypes]
arch/sparc/vdso/vclock_gettime.c:282:1: error: no previous prototype for '__vdso_clock_gettime_stick' [-Werror=missing-prototypes]
arch/sparc/vdso/vclock_gettime.c:307:1: error: no previous prototype for '__vdso_gettimeofday' [-Werror=missing-prototypes]
arch/sparc/vdso/vclock_gettime.c:343:1: error: no previous prototype for '__vdso_gettimeofday_stick' [-Werror=missing-prototypes]
Most architectures have already added workarounds for these by adding
declarations somewhere, but since these are all compatible, we should
really just have one copy, with an #ifdef check for the 32-bit vs
64-bit variant and use that everywhere.
Unfortunately, the sparc an um versions are currently incompatible
since they never added support for __vdso_clock_gettime64() in 32-bit
userland. For the moment, I'm leaving this one out, as I can't
easily test it and it requires a larger rework.
Reviewed-by: Vincenzo Frascino <vincenzo.frascino@arm.com>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
|
|
some architectures run into a -Wmissing-prototypes warning
for trap_init()
arch/microblaze/kernel/traps.c:21:6: warning: no previous prototype for 'trap_init' [-Wmissing-prototypes]
Include the right header to avoid this consistently, removing
the extra declarations on m68k and x86 that were added as local
workarounds already.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
|
|
The prototype was hidden in an #ifdef on x86, which causes a warning:
kernel/irq_work.c:72:13: error: no previous prototype for 'arch_irq_work_raise' [-Werror=missing-prototypes]
Some architectures have a working prototype, while others don't.
Fix this by providing it in only one place that is always visible.
Reviewed-by: Alexander Gordeev <agordeev@linux.ibm.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Acked-by: Palmer Dabbelt <palmer@rivosinc.com>
Acked-by: Guo Ren <guoren@kernel.org>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
|
|
On 64-bit platforms, the pfn_to_kaddr() macro requires that the input
value is 64 bits in order to ensure that valid address bits don't get
lost when shifting that input by PAGE_SHIFT to calculate the physical
address to provide a virtual address for.
One such example is in pvalidate_pages() (used by SEV-SNP guests), where
the GFN in the struct used for page-state change requests is a 40-bit
bit-field, so attempts to pass this GFN field directly into
pfn_to_kaddr() ends up causing guest crashes when dealing with addresses
above the 1TB range due to the above.
Fix this issue with SEV-SNP guests, as well as any similar cases that
might cause issues in current/future code, by using an inline function,
instead of a macro, so that the input is implicitly cast to the
expected 64-bit input type prior to performing the shift operation.
While it might be argued that the issue is on the caller side, other
archs/macros have taken similar approaches to deal with instances like
this, such as ARM explicitly casting the input to phys_addr_t:
e48866647b48 ("ARM: 8396/1: use phys_addr_t in pfn_to_kaddr()")
A C inline function is even better though.
[ mingo: Refined the changelog some more & added __always_inline. ]
Fixes: 6c3211796326 ("x86/sev: Add SNP-specific unaccepted memory support")
Suggested-by: Dave Hansen <dave.hansen@intel.com>
Suggested-by: H. Peter Anvin <hpa@zytor.com>
Signed-off-by: Michael Roth <michael.roth@amd.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20231122163700.400507-1-michael.roth@amd.com
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Rik van Riel <riel@surriel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
|
|
Commit 1e8f93e18379 ("x86: Consolidate port I/O helpers") moved some
port I/O helpers to <asm/shared/io.h>, which caused the 'bw' parameter
in the BUILDIO() macro to become unused. Remove it.
Signed-off-by: Yuntao Wang <ytcoode@gmail.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20231123034911.217791-1-ytcoode@gmail.com
|
|
This type predates recorded history in tglx/history.git, making it older
than Feb 5th 2002.
This structure is literally old enough to drink in most juristictions in
the world, and has not been used once in that time.
Lay it to rest in /dev/null.
Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de>
Reviewed-by: Steve Wahl <steve.wahl@hpe.com>
Link: https://lore.kernel.org/r/20231102-x86-apic-v1-3-bf049a2a0ed6@citrix.com
|
|
The type is not used any more.
Replace the constants with plain defines so they can live outside of an
__ASSEMBLY__ block, allowing for more cleanup in subsequent changes.
Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de>
Reviewed-by: Steve Wahl <steve.wahl@hpe.com>
Link: https://lore.kernel.org/r/20231102-x86-apic-v1-2-bf049a2a0ed6@citrix.com
|
|
This field is set to APIC_DELIVERY_MODE_FIXED in all cases, and is read
exactly once. Fold the constant in uv_program_mmr() and drop the field.
Searching for the origin of the stale HyperV comment reveals commit
a31e58e129f7 ("x86/apic: Switch all APICs to Fixed delivery mode") which
notes:
As a consequence of this change, the apic::irq_delivery_mode field is
now pointless, but this needs to be cleaned up in a separate patch.
6 years is long enough for this technical debt to have survived.
[ bp: Fold in
https://lore.kernel.org/r/20231121123034.1442059-1-andrew.cooper3@citrix.com
]
Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de>
Reviewed-by: Steve Wahl <steve.wahl@hpe.com>
Link: https://lore.kernel.org/r/20231102-x86-apic-v1-1-bf049a2a0ed6@citrix.com
|
|
Avoid conflicts, base on fixes.
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
|
|
Introduce several new KVM uAPIs to ultimately create a guest-first memory
subsystem within KVM, a.k.a. guest_memfd. Guest-first memory allows KVM
to provide features, enhancements, and optimizations that are kludgly
or outright impossible to implement in a generic memory subsystem.
The core KVM ioctl() for guest_memfd is KVM_CREATE_GUEST_MEMFD, which
similar to the generic memfd_create(), creates an anonymous file and
returns a file descriptor that refers to it. Again like "regular"
memfd files, guest_memfd files live in RAM, have volatile storage,
and are automatically released when the last reference is dropped.
The key differences between memfd files (and every other memory subystem)
is that guest_memfd files are bound to their owning virtual machine,
cannot be mapped, read, or written by userspace, and cannot be resized.
guest_memfd files do however support PUNCH_HOLE, which can be used to
convert a guest memory area between the shared and guest-private states.
A second KVM ioctl(), KVM_SET_MEMORY_ATTRIBUTES, allows userspace to
specify attributes for a given page of guest memory. In the long term,
it will likely be extended to allow userspace to specify per-gfn RWX
protections, including allowing memory to be writable in the guest
without it also being writable in host userspace.
The immediate and driving use case for guest_memfd are Confidential
(CoCo) VMs, specifically AMD's SEV-SNP, Intel's TDX, and KVM's own pKVM.
For such use cases, being able to map memory into KVM guests without
requiring said memory to be mapped into the host is a hard requirement.
While SEV+ and TDX prevent untrusted software from reading guest private
data by encrypting guest memory, pKVM provides confidentiality and
integrity *without* relying on memory encryption. In addition, with
SEV-SNP and especially TDX, accessing guest private memory can be fatal
to the host, i.e. KVM must be prevent host userspace from accessing
guest memory irrespective of hardware behavior.
Long term, guest_memfd may be useful for use cases beyond CoCo VMs,
for example hardening userspace against unintentional accesses to guest
memory. As mentioned earlier, KVM's ABI uses userspace VMA protections to
define the allow guest protection (with an exception granted to mapping
guest memory executable), and similarly KVM currently requires the guest
mapping size to be a strict subset of the host userspace mapping size.
Decoupling the mappings sizes would allow userspace to precisely map
only what is needed and with the required permissions, without impacting
guest performance.
A guest-first memory subsystem also provides clearer line of sight to
things like a dedicated memory pool (for slice-of-hardware VMs) and
elimination of "struct page" (for offload setups where userspace _never_
needs to DMA from or into guest memory).
guest_memfd is the result of 3+ years of development and exploration;
taking on memory management responsibilities in KVM was not the first,
second, or even third choice for supporting CoCo VMs. But after many
failed attempts to avoid KVM-specific backing memory, and looking at
where things ended up, it is quite clear that of all approaches tried,
guest_memfd is the simplest, most robust, and most extensible, and the
right thing to do for KVM and the kernel at-large.
The "development cycle" for this version is going to be very short;
ideally, next week I will merge it as is in kvm/next, taking this through
the KVM tree for 6.8 immediately after the end of the merge window.
The series is still based on 6.6 (plus KVM changes for 6.7) so it
will require a small fixup for changes to get_file_rcu() introduced in
6.7 by commit 0ede61d8589c ("file: convert to SLAB_TYPESAFE_BY_RCU").
The fixup will be done as part of the merge commit, and most of the text
above will become the commit message for the merge.
Pending post-merge work includes:
- hugepage support
- looking into using the restrictedmem framework for guest memory
- introducing a testing mechanism to poison memory, possibly using
the same memory attributes introduced here
- SNP and TDX support
There are two non-KVM patches buried in the middle of this series:
fs: Rename anon_inode_getfile_secure() and anon_inode_getfd_secure()
mm: Add AS_UNMOVABLE to mark mapping as completely unmovable
The first is small and mostly suggested-by Christian Brauner; the second
a bit less so but it was written by an mm person (Vlastimil Babka).
|
