Add cpuid bit definition for overflow recovery. This is needed in the case
where a deferred error has been sent to the guest, a guest process accesses the
poisoned memory, but the machine_check_poll function has not yet handled the
original deferred error. If overflow recovery is not set in this case, when we
handle the uncorrected error from the poisoned memory access, the overflow bit
will be set and will result in the guest being shut down.
By the time the MCE reaches the guest, the overflow has been handled
by the host and has not caused a shutdown, so include the bit unconditionally.
Signed-off-by: John Allen <john.allen@amd.com>
Message-ID: <20240603193622.47156-4-john.allen@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Add cpuid bit definition for the SUCCOR feature. This cpuid bit is required to
be exposed to guests to allow them to handle machine check exceptions on AMD
hosts.
----
v2:
- Add "succor" feature word.
- Add case to kvm_arch_get_supported_cpuid for the SUCCOR feature.
Reported-by: William Roche <william.roche@oracle.com>
Reviewed-by: Joao Martins <joao.m.martins@oracle.com>
Signed-off-by: John Allen <john.allen@amd.com>
Message-ID: <20240603193622.47156-3-john.allen@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
For the most part, AMD hosts can use the same MCE injection code as Intel, but
there are instances where the qemu implementation is Intel specific. First, MCE
delivery works differently on AMD and does not support broadcast. Second,
kvm_mce_inject generates MCEs that include a number of Intel specific status
bits. Modify kvm_mce_inject to properly generate MCEs on AMD platforms.
Reported-by: William Roche <william.roche@oracle.com>
Signed-off-by: John Allen <john.allen@amd.com>
Message-ID: <20240603193622.47156-2-john.allen@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Different versions of PC machine support different maximum vCPUs, and
even different features have limits on the maximum number of vCPUs (
For example, if x2apic is not enabled in the TCG case, the maximum of
255 vCPUs are supported).
It is difficult to list the maximum vCPUs under all restrictions. Thus,
to avoid confusion, avoid mentioning specific maximum vCPU number
limitations here.
Suggested-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Zhao Liu <zhao1.liu@intel.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Message-ID: <20240606085436.2028900-1-zhao1.liu@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
FRED CPU states are managed in 9 new FRED MSRs, in addtion to a few
existing CPU registers and MSRs, e.g., CR4.FRED and MSR_IA32_PL0_SSP.
Save/restore/migrate FRED MSRs if FRED is exposed to the guest.
Tested-by: Shan Kang <shan.kang@intel.com>
Signed-off-by: Xin Li <xin3.li@intel.com>
Message-ID: <20231109072012.8078-7-xin3.li@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Allow VMX nested-exception support to be exposed in KVM guests, thus
nested KVM guests can enumerate it.
Tested-by: Shan Kang <shan.kang@intel.com>
Signed-off-by: Xin Li <xin3.li@intel.com>
Message-ID: <20231109072012.8078-6-xin3.li@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Report secondary vm-exit controls and the VMX controls used to
save/load FRED MSRs.
Tested-by: Shan Kang <shan.kang@intel.com>
Signed-off-by: Xin Li <xin3.li@intel.com>
Message-ID: <20231109072012.8078-5-xin3.li@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The CR4.FRED bit, i.e., CR4[32], is no longer a reserved bit when FRED
is exposed to guests, otherwise it is still a reserved bit.
Tested-by: Shan Kang <shan.kang@intel.com>
Signed-off-by: Xin Li <xin3.li@intel.com>
Reviewed-by: Zhao Liu <zhao1.liu@intel.com>
Message-ID: <20231109072012.8078-3-xin3.li@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
FRED, i.e., the Intel flexible return and event delivery architecture,
defines simple new transitions that change privilege level (ring
transitions).
The new transitions defined by the FRED architecture are FRED event
delivery and, for returning from events, two FRED return instructions.
FRED event delivery can effect a transition from ring 3 to ring 0, but
it is used also to deliver events incident to ring 0. One FRED
instruction (ERETU) effects a return from ring 0 to ring 3, while the
other (ERETS) returns while remaining in ring 0. Collectively, FRED
event delivery and the FRED return instructions are FRED transitions.
In addition to these transitions, the FRED architecture defines a new
instruction (LKGS) for managing the state of the GS segment register.
The LKGS instruction can be used by 64-bit operating systems that do
not use the new FRED transitions.
WRMSRNS is an instruction that behaves exactly like WRMSR, with the
only difference being that it is not a serializing instruction by
default. Under certain conditions, WRMSRNS may replace WRMSR to improve
performance. FRED uses it to switch RSP0 in a faster manner.
Search for the latest FRED spec in most search engines with this search
pattern:
site:intel.com FRED (flexible return and event delivery) specification
The CPUID feature flag CPUID.(EAX=7,ECX=1):EAX[17] enumerates FRED, and
the CPUID feature flag CPUID.(EAX=7,ECX=1):EAX[18] enumerates LKGS, and
the CPUID feature flag CPUID.(EAX=7,ECX=1):EAX[19] enumerates WRMSRNS.
Add CPUID definitions for FRED/LKGS/WRMSRNS, and expose them to KVM guests.
Because FRED relies on LKGS and WRMSRNS, add that to feature dependency
map.
Tested-by: Shan Kang <shan.kang@intel.com>
Signed-off-by: Xin Li <xin3.li@intel.com>
Message-ID: <20231109072012.8078-2-xin3.li@intel.com>
[Fix order of dependencies, add dependencies from LM to FRED. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
When a macOS Hypervisor.framework call fails which is checked by
assert_hvf_ok(), Qemu exits printing the error value, but not the
location
in the code, as regular assert() macro expansions would.
This change turns assert_hvf_ok() into a macro similar to other
assertions, which expands to a call to the corresponding _impl()
function together with information about the expression that failed
the assertion and its location in the code.
Additionally, stringifying the numeric hv_return_t code is factored
into a helper function that can be reused for diagnostics and debugging
outside of assertions.
Signed-off-by: Phil Dennis-Jordan <phil@philjordan.eu>
Message-ID: <20240605112556.43193-8-phil@philjordan.eu>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
macOS 10.15 introduced the more efficient hv_vcpu_run_until() function
to supersede hv_vcpu_run(). According to the documentation, there is no
longer any reason to use the latter on modern host OS versions, especially
after 11.0 added support for an indefinite deadline.
Observed behaviour of the newer function is that as documented, it exits
much less frequently - and most of the original function’s exits seem to
have been effectively pointless.
Another reason to use the new function is that it is a prerequisite for
using newer features such as in-kernel APIC support. (Not covered by
this patch.)
This change implements the upgrade by selecting one of three code paths
at compile time: two static code paths for the new and old functions
respectively, when building for targets where the new function is either
not available, or where the built executable won’t run on older
platforms lacking the new function anyway. The third code path selects
dynamically based on runtime detected availability of the weakly-linked
symbol.
Signed-off-by: Phil Dennis-Jordan <phil@philjordan.eu>
Message-ID: <20240605112556.43193-7-phil@philjordan.eu>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
When interrupting a vCPU thread, this patch actually tells the hypervisor to
stop running guest code on that vCPU.
Calling hv_vcpu_interrupt actually forces a vCPU exit, analogously to
hv_vcpus_exit on aarch64. Alternatively, if the vCPU thread
is not
running the VM, it will immediately cause an exit when it attempts
to do so.
Previously, hvf_kick_vcpu_thread relied upon hv_vcpu_run returning very
frequently, including many spurious exits, which made it less of a problem that
nothing was actively done to stop the vCPU thread running guest code.
The newer, more efficient hv_vcpu_run_until exits much more rarely, so a true
"kick" is needed before switching to that.
Signed-off-by: Phil Dennis-Jordan <phil@philjordan.eu>
Message-ID: <20240605112556.43193-6-phil@philjordan.eu>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
When using x86 macOS Hypervisor.framework as accelerator, detection of
dirty memory regions is implemented by marking logged memory region
slots as read-only in the EPT, then setting the dirty flag when a
guest write causes a fault. The area marked dirty should then be marked
writable in order for subsequent writes to succeed without a VM exit.
However, dirty bits are tracked on a per-page basis, whereas the fault
handler was marking the whole logged memory region as writable. This
change fixes the fault handler so only the protection of the single
faulting page is marked as dirty.
(Note: the dirty page tracking appeared to work despite this error
because HVF’s hv_vcpu_run() function generated unnecessary EPT fault
exits, which ended up causing the dirty marking handler to run even
when the memory region had been marked RW. When using
hv_vcpu_run_until(), a change planned for a subsequent commit, these
spurious exits no longer occur, so dirty memory tracking malfunctions.)
Additionally, the dirty page is set to permit code execution, the same
as all other guest memory; changing memory protection from RX to RW not
RWX appears to have been an oversight.
Signed-off-by: Phil Dennis-Jordan <phil@philjordan.eu>
Reviewed-by: Roman Bolshakov <roman@roolebo.dev>
Tested-by: Roman Bolshakov <roman@roolebo.dev>
Message-ID: <20240605112556.43193-5-phil@philjordan.eu>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
macOS Hypervisor.framework uses different types for identifying vCPUs, hv_vcpu_t or hv_vcpuid_t, depending on host architecture. They are not just differently named typedefs for the same primitive type, but reference different-width integers.
Instead of using an integer type and casting where necessary, this change introduces a typedef which resolves the active architecture’s hvf typedef. It also removes a now-unnecessary cast.
Signed-off-by: Phil Dennis-Jordan <phil@philjordan.eu>
Reviewed-by: Roman Bolshakov <roman@roolebo.dev>
Tested-by: Roman Bolshakov <roman@roolebo.dev>
Message-ID: <20240605112556.43193-4-phil@philjordan.eu>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
A bunch of function definitions used empty parentheses instead of (void) syntax, yielding the following warning when building with clang on macOS:
warning: a function declaration without a prototype is deprecated in all versions of C [-Wstrict-prototypes]
In addition to fixing these function headers, it also fixes what appears to be a typo causing a variable to be unused after initialisation.
warning: variable 'entry_ctls' set but not used [-Wunused-but-set-variable]
Signed-off-by: Phil Dennis-Jordan <phil@philjordan.eu>
Reviewed-by: Roman Bolshakov <roman@roolebo.dev>
Tested-by: Roman Bolshakov <roman@roolebo.dev>
Message-ID: <20240605112556.43193-3-phil@philjordan.eu>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This patch adds the INVTSC bit to the Hypervisor.framework accelerator's
CPUID bit passthrough allow-list. Previously, specifying +invtsc in the CPU
configuration would fail with the following warning despite the host CPU
advertising the feature:
qemu-system-x86_64: warning: host doesn't support requested feature:
CPUID.80000007H:EDX.invtsc [bit 8]
x86 macOS itself relies on a fixed rate TSC for its own Mach absolute time
timestamp mechanism, so there's no reason we can't enable this bit for guests.
When the feature is enabled, a migration blocker is installed.
Signed-off-by: Phil Dennis-Jordan <phil@philjordan.eu>
Reviewed-by: Roman Bolshakov <roman@roolebo.dev>
Tested-by: Roman Bolshakov <roman@roolebo.dev>
Message-ID: <20240605112556.43193-2-phil@philjordan.eu>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Compiling without system, user, tools or guest-agent fails with the
following error message:
./configure --disable-system --disable-user --disable-tools \
--disable-guest-agent
error message:
/usr/bin/ld: libqemuutil.a.p/util_error-report.c.o: in function `error_printf':
/media/liuzhao/data/qemu-cook/build/../util/error-report.c:38: undefined reference to `error_vprintf'
/usr/bin/ld: libqemuutil.a.p/util_error-report.c.o: in function `vreport':
/media/liuzhao/data/qemu-cook/build/../util/error-report.c:215: undefined reference to `error_vprintf'
collect2: error: ld returned 1 exit status
This is because tests/bench and tests/unit both need qemuutil, which
requires error_vprintf stub when system is disabled.
Add error_vprintf stub into stub_ss for all cases other than disabling
system.
Fixes: 3a15604900 ("stubs: include stubs only if needed")
Reported-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Zhao Liu <zhao1.liu@intel.com>
Message-ID: <20240605152549.1795762-1-zhao1.liu@intel.com>
[Include error-printf.c unconditionally. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Before this commit, scsi-disk accepts a string of arbitrary length for
its "serial" property. However, the value visible on the guest is
actually truncated to 36 characters. This limitation doesn't come from
the SCSI specification, it is an arbitrary limit that was initially
picked as 20 and later bumped to 36 by commit 48b62063.
Similarly, device_id was introduced as a copy of the serial number,
limited to 20 characters, but commit 48b62063 forgot to actually bump
it.
As long as we silently truncate the given string, extending the limit is
actually not a harmless change, but break the guest ABI. This is the
most important reason why commit 48b62063 was really wrong (and it's
also why we can't change device_id to be in sync with the serial number
again and use 36 characters now, it would be another guest ABI
breakage).
In order to avoid future breakage, don't silently truncate the serial
number string any more, but just error out if it would be truncated.
Buglink: https://issues.redhat.com/browse/RHEL-3542
Suggested-by: Peter Krempa <pkrempa@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-ID: <20240604161755.63448-1-kwolf@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
No semantic change, just simpler control flow.
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Detect early unsupported MADV_MERGEABLE and MADV_DONTDUMP, and print a clearer
error message that points to the deficiency of the host.
Cc: Michal Privoznik <mprivozn@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
If memory-backend-{file,ram} has a size that's not aligned to
underlying page size it is not only wasteful, but also may lead
to hard to debug behaviour. For instance, in case
memory-backend-file and hugepages, madvise() and mbind() fail.
Rightfully so, page is the smallest unit they can work with. And
even though an error is reported, the root cause it not very
clear:
qemu-system-x86_64: Couldn't set property 'dump' on 'memory-backend-file': Invalid argument
After this commit:
qemu-system-x86_64: backend 'memory-backend-file' memory size must be multiple of 2 MiB
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Tested-by: Mario Casquero <mcasquer@redhat.com>
Message-ID: <b5b9f9c6bba07879fb43f3c6f496c69867ae3716.1717584048.git.mprivozn@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Not every OS is capable of madvise() or posix_madvise() even. In
that case, errno should be set to ENOSYS as it reflects the cause
better.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: David Hildenbrand <david@redhat.com>
Message-ID: <b381c23bd8f413f1453a2c1a66e0979beaf27433.1717584048.git.mprivozn@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The unspoken premise of qemu_madvise() is that errno is set on
error. And it is mostly the case except for posix_madvise() which
is documented to return either zero (on success) or a positive
error number. This means, we must set errno ourselves. And while
at it, make the function return a negative value on error, just
like other error paths do.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Message-ID: <af17113e7c1f2cc909ffd36d23f5a411b63b8764.1717584048.git.mprivozn@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
On Darwin, posix_madvise() has the same return semantics as plain
madvise() [1]. That's not really what our usage expects.
Fortunately, madvise() is available and preferred anyways so we
may stop detecting posix_madvise() on Darwin.
1: https://opensource.apple.com/source/xnu/xnu-7195.81.3/bsd/man/man2/madvise.2.auto.html
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Message-ID: <00f71753bdeb8c0f049fda05fb63b84bb5502fb3.1717584048.git.mprivozn@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Otherwise, starting any guest on a non-Linux guests results in
qemu-system-arm: Couldn't set property 'merge' on 'memory-backend-ram': Invalid argument
Cc: Michal Privoznik <mprivozn@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The calculation of FrameTemp is done using the size indicated by mo_pushpop()
before being written back to EBP, but the final writeback to EBP is done using
the size indicated by mo_stacksize().
In the case where mo_pushpop() is MO_32 and mo_stacksize() is MO_16 then the
final writeback to EBP is done using MO_16 which can leave junk in the top
16-bits of EBP after executing ENTER.
Change the writeback of EBP to use the same size indicated by mo_pushpop() to
ensure that the full value is written back.
Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Resolves: https://gitlab.com/qemu-project/qemu/-/issues/2198
Message-ID: <20240606095319.229650-5-mark.cave-ayland@ilande.co.uk>
Cc: qemu-stable@nongnu.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
When OS/2 Warp configures its segment descriptors, many of them are configured with
the P flag clear to allow for a fault-on-demand implementation. In the case where
the stack value is POPped into the segment registers, the SP is incremented before
calling gen_helper_load_seg() to validate the segment descriptor:
IN:
0xffef2c0c: 66 07 popl %es
OP:
ld_i32 loc9,env,$0xfffffffffffffff8
sub_i32 loc9,loc9,$0x1
brcond_i32 loc9,$0x0,lt,$L0
st16_i32 loc9,env,$0xfffffffffffffff8
st8_i32 $0x1,env,$0xfffffffffffffffc
---- 0000000000000c0c 0000000000000000
ext16u_i64 loc0,rsp
add_i64 loc0,loc0,ss_base
ext32u_i64 loc0,loc0
qemu_ld_a64_i64 loc0,loc0,noat+un+leul,5
add_i64 loc3,rsp,$0x4
deposit_i64 rsp,rsp,loc3,$0x0,$0x10
extrl_i64_i32 loc5,loc0
call load_seg,$0x0,$0,env,$0x0,loc5
add_i64 rip,rip,$0x2
ext16u_i64 rip,rip
exit_tb $0x0
set_label $L0
exit_tb $0x7fff58000043
If helper_load_seg() generates a fault when validating the segment descriptor then as
the SP has already been incremented, the topmost word of the stack is overwritten by
the arguments pushed onto the stack by the CPU before taking the fault handler. As a
consequence things rapidly go wrong upon return from the fault handler due to the
corrupted stack.
Update the logic for the existing writeback condition so that a POP into the segment
registers also calls helper_load_seg() first before incrementing the SP, so that if a
fault occurs the SP remains unaltered.
Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Resolves: https://gitlab.com/qemu-project/qemu/-/issues/2198
Message-ID: <20240606095319.229650-4-mark.cave-ayland@ilande.co.uk>
Fixes: cc1d28bdbe ("target/i386: move 00-5F opcodes to new decoder", 2024-05-07)
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Instead of directly implementing the writeback using gen_op_st_v(), use the
existing gen_writeback() function.
Suggested-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Message-ID: <20240606095319.229650-3-mark.cave-ayland@ilande.co.uk>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This will make subsequent changes a little easier to read.
Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Message-ID: <20240606095319.229650-2-mark.cave-ayland@ilande.co.uk>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
DISAS_NORETURN suppresses the work normally done by gen_eob(), and therefore
must be used in special cases only. Document them.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
HLT uses DISAS_NORETURN because the corresponding helper calls
cpu_loop_exit(). However, while gen_eob() clears HF_RF_MASK and
synthesizes a #DB exception if single-step is active, none of this is
done by HLT. Note that the single-step trap is generated after the halt
is finished.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
PAUSE uses DISAS_NORETURN because the corresponding helper
calls cpu_loop_exit(). However, while HLT clear HF_INHIBIT_IRQ_MASK
to correctly handle "STI; HLT", the same is missing from PAUSE.
And also gen_eob() clears HF_RF_MASK and synthesizes a #DB exception
if single-step is active; none of this is done by HLT and PAUSE.
Start fixing PAUSE, HLT will follow.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
From vm entry to exit, VMRUN is handled as a single instruction. It
uses DISAS_NORETURN in order to avoid processing TF or RF before
the first instruction executes in the guest. However, the corresponding
handling is missing in vmexit. Add it, and at the same time reorganize
the comments with quotes from the manual about the tasks performed
by a #VMEXIT.
Another gen_eob() task that is missing in VMRUN is preparing the
HF_INHIBIT_IRQ flag for the next instruction, in this case by loading
it from the VMCB control state.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
If the required DR7 (either from the VMCB or from the host save
area) disables a breakpoint that was enabled prior to vmentry
or vmexit, it is left enabled and will trigger EXCP_DEBUG.
This causes a spurious #DB on the next crossing of the breakpoint.
To disable it, vmentry/vmexit must use cpu_x86_update_dr7
to load DR7.
Because cpu_x86_update_dr7 takes a 32-bit argument, check
reserved bits prior to calling cpu_x86_update_dr7, and do the
same for DR6 as well for consistency.
This scenario is tested by the "host_rflags" test in kvm-unit-tests.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
DR7.GD triggers a #DB exception on any access to debug registers.
The GD bit is cleared so that the #DB handler itself can access
the debug registers.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Use decode.c's support for intercepts, doing the check in TCG-generated
code rather than the helper. This is cleaner because it allows removing
the eip_addend argument to helper_pause(), even though it adds a bit of
bloat for opcode 0x90's new decoding function.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Use decode.c's support for intercepts, doing the check in TCG-generated
code rather than the helper. This is cleaner because it allows removing
the eip_addend argument to helper_hlt().
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
ICEBP generates a trap-like exception, while gen_exception() produces
a fault. Resurrect gen_update_eip_next() to implement the desired
semantics.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
When preparing an exception stack frame for a fault exception, the value
pushed for RF is 1. Take that into account. The same should be true
of interrupts for repeated string instructions, but the situation there
is complicated.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
* require x86-64-v2 baseline ISA
* SEV-SNP host support
* fix xsave.flat with TCG
* fixes for CPUID checks done by TCG
-----BEGIN PGP SIGNATURE-----
iQFIBAABCAAyFiEE8TM4V0tmI4mGbHaCv/vSX3jHroMFAmZgKVYUHHBib256aW5p
QHJlZGhhdC5jb20ACgkQv/vSX3jHroPKYgf/QkWrNXdjjD3yAsv5LbJFVTVyCYW3
b4Iax29kEDy8k9wbzfLxOfIk9jXIjmbOMO5ZN9LFiHK6VJxbXslsMh6hm50M3xKe
49X1Rvf9YuVA7KZX+dWkEuqLYI6Tlgj3HaCilYWfXrjyo6hY3CxzkPV/ChmaeYlV
Ad4Y8biifoUuuEK8OTeTlcDWLhOHlFXylG3AXqULsUsXp0XhWJ9juXQ60eATv/W4
eCEH7CSmRhYFu2/rV+IrWFYMnskLRTk1OC1/m6yXGPKOzgnOcthuvQfiUgPkbR/d
llY6Ni5Aaf7+XX3S7Avcyvoq8jXzaaMzOrzL98rxYGDR1sYBYO+4h4ZToA==
=qQeP
-----END PGP SIGNATURE-----
Merge tag 'for-upstream' of https://gitlab.com/bonzini/qemu into staging
* virtio-blk: remove SCSI passthrough functionality
* require x86-64-v2 baseline ISA
* SEV-SNP host support
* fix xsave.flat with TCG
* fixes for CPUID checks done by TCG
# -----BEGIN PGP SIGNATURE-----
#
# iQFIBAABCAAyFiEE8TM4V0tmI4mGbHaCv/vSX3jHroMFAmZgKVYUHHBib256aW5p
# QHJlZGhhdC5jb20ACgkQv/vSX3jHroPKYgf/QkWrNXdjjD3yAsv5LbJFVTVyCYW3
# b4Iax29kEDy8k9wbzfLxOfIk9jXIjmbOMO5ZN9LFiHK6VJxbXslsMh6hm50M3xKe
# 49X1Rvf9YuVA7KZX+dWkEuqLYI6Tlgj3HaCilYWfXrjyo6hY3CxzkPV/ChmaeYlV
# Ad4Y8biifoUuuEK8OTeTlcDWLhOHlFXylG3AXqULsUsXp0XhWJ9juXQ60eATv/W4
# eCEH7CSmRhYFu2/rV+IrWFYMnskLRTk1OC1/m6yXGPKOzgnOcthuvQfiUgPkbR/d
# llY6Ni5Aaf7+XX3S7Avcyvoq8jXzaaMzOrzL98rxYGDR1sYBYO+4h4ZToA==
# =qQeP
# -----END PGP SIGNATURE-----
# gpg: Signature made Wed 05 Jun 2024 02:01:10 AM PDT
# gpg: using RSA key F13338574B662389866C7682BFFBD25F78C7AE83
# gpg: issuer "pbonzini@redhat.com"
# gpg: Good signature from "Paolo Bonzini <bonzini@gnu.org>" [full]
# gpg: aka "Paolo Bonzini <pbonzini@redhat.com>" [full]
* tag 'for-upstream' of https://gitlab.com/bonzini/qemu: (46 commits)
hw/i386: Add support for loading BIOS using guest_memfd
hw/i386/sev: Use guest_memfd for legacy ROMs
memory: Introduce memory_region_init_ram_guest_memfd()
i386/sev: Allow measured direct kernel boot on SNP
i386/sev: Reorder struct declarations
i386/sev: Extract build_kernel_loader_hashes
i386/sev: Enable KVM_HC_MAP_GPA_RANGE hcall for SNP guests
i386/kvm: Add KVM_EXIT_HYPERCALL handling for KVM_HC_MAP_GPA_RANGE
i386/sev: Invoke launch_updata_data() for SNP class
i386/sev: Invoke launch_updata_data() for SEV class
hw/i386/sev: Add support to encrypt BIOS when SEV-SNP is enabled
i386/sev: Add support for SNP CPUID validation
i386/sev: Add support for populating OVMF metadata pages
hw/i386/sev: Add function to get SEV metadata from OVMF header
i386/sev: Set CPU state to protected once SNP guest payload is finalized
i386/sev: Add handling to encrypt/finalize guest launch data
i386/sev: Add the SNP launch start context
i386/sev: Update query-sev QAPI format to handle SEV-SNP
i386/sev: Add a class method to determine KVM VM type for SNP guests
i386/sev: Don't return launch measurements for SEV-SNP guests
...
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
When guest_memfd is enabled, the BIOS is generally part of the initial
encrypted guest image and will be accessed as private guest memory. Add
the necessary changes to set up the associated RAM region with a
guest_memfd backend to allow for this.
Current support centers around using -bios to load the BIOS data.
Support for loading the BIOS via pflash requires additional enablement
since those interfaces rely on the use of ROM memory regions which make
use of the KVM_MEM_READONLY memslot flag, which is not supported for
guest_memfd-backed memslots.
Signed-off-by: Michael Roth <michael.roth@amd.com>
Signed-off-by: Pankaj Gupta <pankaj.gupta@amd.com>
Message-ID: <20240530111643.1091816-29-pankaj.gupta@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Current SNP guest kernels will attempt to access these regions with
with C-bit set, so guest_memfd is needed to handle that. Otherwise,
kvm_convert_memory() will fail when the guest kernel tries to access it
and QEMU attempts to call KVM_SET_MEMORY_ATTRIBUTES to set these ranges
to private.
Whether guests should actually try to access ROM regions in this way (or
need to deal with legacy ROM regions at all), is a separate issue to be
addressed on kernel side, but current SNP guest kernels will exhibit
this behavior and so this handling is needed to allow QEMU to continue
running existing SNP guest kernels.
Signed-off-by: Michael Roth <michael.roth@amd.com>
[pankaj: Added sev_snp_enabled() check]
Signed-off-by: Pankaj Gupta <pankaj.gupta@amd.com>
Message-ID: <20240530111643.1091816-28-pankaj.gupta@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Introduce memory_region_init_ram_guest_memfd() to allocate private
guset memfd on the MemoryRegion initialization. It's for the use case of
TDVF, which must be private on TDX case.
Signed-off-by: Xiaoyao Li <xiaoyao.li@intel.com>
Signed-off-by: Michael Roth <michael.roth@amd.com>
Signed-off-by: Pankaj Gupta <pankaj.gupta@amd.com>
Message-ID: <20240530111643.1091816-4-pankaj.gupta@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
In SNP, the hashes page designated with a specific metadata entry
published in AmdSev OVMF.
Therefore, if the user enabled kernel hashes (for measured direct boot),
QEMU should prepare the content of hashes table, and during the
processing of the metadata entry it copy the content into the designated
page and encrypt it.
Note that in SNP (unlike SEV and SEV-ES) the measurements is done in
whole 4KB pages. Therefore QEMU zeros the whole page that includes the
hashes table, and fills in the kernel hashes area in that page, and then
encrypts the whole page. The rest of the page is reserved for SEV
launch secrets which are not usable anyway on SNP.
If the user disabled kernel hashes, QEMU pre-validates the kernel hashes
page as a zero page.
Signed-off-by: Dov Murik <dovmurik@linux.ibm.com>
Signed-off-by: Michael Roth <michael.roth@amd.com>
Signed-off-by: Pankaj Gupta <pankaj.gupta@amd.com>
Message-ID: <20240530111643.1091816-24-pankaj.gupta@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Move the declaration of PaddedSevHashTable before SevSnpGuest so
we can add a new such field to the latter.
No functional change intended.
Signed-off-by: Dov Murik <dovmurik@linux.ibm.com>
Signed-off-by: Michael Roth <michael.roth@amd.com>
Signed-off-by: Pankaj Gupta <pankaj.gupta@amd.com>
Message-ID: <20240530111643.1091816-23-pankaj.gupta@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Extract the building of the kernel hashes table out from
sev_add_kernel_loader_hashes() to allow building it in
other memory areas (for SNP support).
No functional change intended.
Signed-off-by: Dov Murik <dovmurik@linux.ibm.com>
Signed-off-by: Michael Roth <michael.roth@amd.com>
Signed-off-by: Pankaj Gupta <pankaj.gupta@amd.com>
Message-ID: <20240530111643.1091816-22-pankaj.gupta@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
KVM will forward GHCB page-state change requests to userspace in the
form of KVM_HC_MAP_GPA_RANGE, so make sure the hypercall handling is
enabled for SNP guests.
Signed-off-by: Michael Roth <michael.roth@amd.com>
Signed-off-by: Pankaj Gupta <pankaj.gupta@amd.com>
Message-ID: <20240530111643.1091816-32-pankaj.gupta@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
KVM_HC_MAP_GPA_RANGE will be used to send requests to userspace for
private/shared memory attribute updates requested by the guest.
Implement handling for that use-case along with some basic
infrastructure for enabling specific hypercall events.
Signed-off-by: Michael Roth <michael.roth@amd.com>
Signed-off-by: Pankaj Gupta <pankaj.gupta@amd.com>
Message-ID: <20240530111643.1091816-31-pankaj.gupta@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Invoke as sev_snp_launch_update_data() for SNP object.
Signed-off-by: Pankaj Gupta <pankaj.gupta@amd.com>
Message-ID: <20240530111643.1091816-27-pankaj.gupta@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
