If you build QEMU with the clang sanitizer enabled, you can see it
fire when running the arm-cpu-features test:
$ QTEST_QEMU_BINARY=./build/arm-clang/qemu-system-aarch64 ./build/arm-clang/tests/qtest/arm-cpu-features
[...]
../../target/arm/cpu64.c:125:19: runtime error: shift exponent 64 is too large for 64-bit type 'unsigned long long'
[...]
This happens because the user can specify some incorrect SVE
properties that result in our calculating a max_vq of 0. We catch
this and error out, but before we do that we calculate
vq_mask = MAKE_64BIT_MASK(0, max_vq);$
and the MAKE_64BIT_MASK() call is only valid for lengths that are
greater than zero, so we hit the undefined behaviour.
Change the logic so that if max_vq is 0 we specifically set vq_mask
to 0 without going via MAKE_64BIT_MASK(). This lets us drop the
max_vq check from the error-exit logic, because if max_vq is 0 then
vq_map must now be 0.
The UB only happens in the case where the user passed us an incorrect
set of SVE properties, so it's not a big problem in practice.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20230704154332.3014896-1-peter.maydell@linaro.org
We're about to move the TCG-only -cpu max configuration code under
CONFIG_TCG. To be able to do that we need to make sure the qtests
still have some cpu configured even when no other accelerator is
available.
Delineate now what is used with TCG-only and what is also used with
qtests to make the subsequent patches cleaner.
Signed-off-by: Fabiano Rosas <farosas@suse.de>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Message-id: 20230426180013.14814-5-farosas@suse.de
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
FEAT_PAN3 adds an EPAN bit to SCTLR_EL1 and SCTLR_EL2, which allows
the PAN bit to make memory non-privileged-read/write if it is
user-executable as well as if it is user-read/write.
Implement this feature and enable it in the AArch64 'max' CPU.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20230331145045.2584941-4-peter.maydell@linaro.org
Update the ID registers for TCG's '-cpu max' to report the
FEAT_EVT Enhanced Virtualization Traps support.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
The Cortex-A55 is one of the newer armv8.2+ CPUs; in particular
it supports the Privileged Access Never (PAN) feature. Add
a model of this CPU, so you can use a CPU type on the virt
board that models a specific real hardware CPU, rather than
having to use the QEMU-specific "max" CPU type.
Signed-off-by: Timofey Kutergin <tkutergin@gmail.com>
Message-id: 20221121150819.2782817-1-tkutergin@gmail.com
[PMM: tweaked commit message]
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
FEAT_E0PD adds new bits E0PD0 and E0PD1 to TCR_EL1, which allow the
OS to forbid EL0 access to half of the address space. Since this is
an EL0-specific variation on the existing TCR_ELx.{EPD0,EPD1}, we can
implement it entirely in aa64_va_parameters().
This requires moving the existing regime_is_user() to internals.h
so that the code in helper.c can get at it.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 20221021160131.3531787-1-peter.maydell@linaro.org
cpu64.c has ended up in a slightly odd order -- it starts with the
initfns for most of the models-real-hardware CPUs; after that comes a
bunch of support code for SVE, SME, pauth and LPA2 properties. Then
come the initfns for the 'host' and 'max' CPU types, and then after
that one more models-real-hardware CPU initfn, for a64fx. (This
ordering is partly historical and partly required because a64fx needs
the SVE properties.)
Reorder the file into:
* CPU property support functions
* initfns for real hardware CPUs
* initfns for host and max
* class boilerplate
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
The architectural feature FEAT_ETS (Enhanced Translation
Synchronization) is a set of tightened guarantees about memory
ordering involving translation table walks:
* if memory access RW1 is ordered-before memory access RW2 then it
is also ordered-before any translation table walk generated by RW2
that generates a translation fault, address size fault or access
fault
* TLB maintenance on non-exec-permission translations is guaranteed
complete after a DSB (ie it does not need the context
synchronization event that you have to have if you don’t have
FEAT_ETS)
For QEMU’s implementation we don’t reorder translation table walk
accesses, and we guarantee to finish the TLB maintenance as soon as
the TLB op is done (the tlb_flush functions will complete at the end
of the TLB, and TLB ops always end the TB because they’re sysreg
writes).
So we’re already compliant and all we need to do is say so in the ID
registers for the 'max' CPU.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20220819110052.2942289-6-peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Starting with v7 of the debug architecture, there are three extra
ID registers that add information on top of that provided in
DBGDIDR. These are DBGDEVID, DBGDEVID1 and DBGDEVID2. In the
v7 debug architecture, DBGDEVID is optional, present only of
DBGDIDR.DEVID_imp is set. In v7.1 all three must be present.
Implement the missing registers. Note that we only need to set the
values in the ARMISARegisters struct for the CPUs Cortex-A7, A15,
A53, A57 and A72 (plus the 32-bit 'max' which uses the Cortex-A53
values): earlier CPUs didn't implement v7 of the architecture, and
our other 64-bit CPUs (Cortex-A76, Neoverse-N1 and A64fx) don't have
AArch32 support at EL1.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20220630194116.3438513-5-peter.maydell@linaro.org
Mirror the properties for SVE. The main difference is
that any arbitrary set of powers of 2 may be supported,
and not the stricter constraints that apply to SVE.
Include a property to control FEAT_SME_FA64, as failing
to restrict the runtime to the proper subset of insns
could be a major point for bugs.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 20220620175235.60881-18-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
The bitmap need only hold 15 bits; bitmap is over-complicated.
We can simplify operations quite a bit with plain logical ops.
The introduction of SVE_VQ_POW2_MAP eliminates the need for
looping in order to search for powers of two. Simply perform
the logical ops and use count leading or trailing zeros as
required to find the result.
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20220607203306.657998-12-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
The FEAT_DoubleFault extension adds the following:
* All external aborts on instruction fetches and translation table
walks for instruction fetches must be synchronous. For QEMU this
is already true.
* SCR_EL3 has a new bit NMEA which disables the masking of SError
interrupts by PSTATE.A when the SError interrupt is taken to EL3.
For QEMU we only need to make the bit writable, because we have no
sources of SError interrupts.
* SCR_EL3 has a new bit EASE which causes synchronous external
aborts taken to EL3 to be taken at the same entry point as SError.
(Note that this does not mean that they are SErrors for purposes
of PSTATE.A masking or that the syndrome register reports them as
SErrors: it just means that the vector offset is different.)
* The existing SCTLR_EL3.IESB has an effective value of 1 when
SCR_EL3.NMEA is 1. For QEMU this is a no-op because we don't need
different behaviour based on IESB (we don't need to do anything to
ensure that error exceptions are synchronized).
So for QEMU the things we need to change are:
* Make SCR_EL3.{NMEA,EASE} writable
* When taking a synchronous external abort at EL3, adjust the
vector entry point if SCR_EL3.EASE is set
* Advertise the feature in the ID registers
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20220531151431.949322-1-peter.maydell@linaro.org
The architectural feature RASv1p1 introduces the following new
features:
* new registers ERXPFGCDN_EL1, ERXPFGCTL_EL1 and ERXPFGF_EL1
* new bits in the fine-grained trap registers that control traps
for these new registers
* new trap bits HCR_EL2.FIEN and SCR_EL3.FIEN that control traps
for ERXPFGCDN_EL1, ERXPFGCTL_EL1, ERXPFGP_EL1
* a larger number of the ERXMISC<n>_EL1 registers
* the format of ERR<n>STATUS registers changes
The architecture permits that if ERRIDR_EL1.NUM is 0 (as it is for
QEMU) then all these new registers may UNDEF, and the HCR_EL2.FIEN
and SCR_EL3.FIEN bits may be RES0. We don't have any ERR<n>STATUS
registers (again, because ERRIDR_EL1.NUM is 0). QEMU does not yet
implement the fine-grained-trap extension. So there is nothing we
need to implement to be compliant with the feature spec. Make the
'max' CPU report the feature in its ID registers, and document it.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20220531114258.855804-1-peter.maydell@linaro.org
Currently we give all the v7-and-up CPUs a PMU with 4 counters. This
means that we don't provide the 6 counters that are required by the
Arm BSA (Base System Architecture) specification if the CPU supports
the Virtualization extensions.
Instead of having a single PMCR_NUM_COUNTERS, make each CPU type
specify the PMCR reset value (obtained from the appropriate TRM), and
use the 'N' field of that value to define the number of counters
provided.
This means that we now supply 6 counters instead of 4 for:
Cortex-A9, Cortex-A15, Cortex-A53, Cortex-A57, Cortex-A72,
Cortex-A76, Neoverse-N1, '-cpu max'
This CPU goes from 4 to 8 counters:
A64FX
These CPUs remain with 4 counters:
Cortex-A7, Cortex-A8
This CPU goes down from 4 to 3 counters:
Cortex-R5
Note that because we now use the PMCR reset value of the specific
implementation, we no longer set the LC bit out of reset. This has
an UNKNOWN value out of reset for all cores with any AArch32 support,
so guest software should be setting it anyway if it wants it.
This change was originally landed in commit f7fb73b8cd (during
the 6.0 release cycle) but was then reverted by commit
21c2dd77a6 before that release because it did not work with KVM.
This version fixes that by creating the scratch vCPU in
kvm_arm_get_host_cpu_features() with the KVM_ARM_VCPU_PMU_V3 feature
if KVM supports it, and then only asking KVM for the PMCR_EL0 value
if the vCPU has a PMU.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
[PMM: Added the correct value for a64fx]
Message-id: 20220513122852.4063586-1-peter.maydell@linaro.org
Make the GICv3 set its number of bits of physical priority from the
implementation-specific value provided in the CPU state struct, in
the same way we already do for virtual priority bits. Because this
would be a migration compatibility break, we provide a property
force-8-bit-prio which is enabled for 7.0 and earlier versioned board
models to retain the legacy "always use 8 bits" behaviour.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20220512151457.3899052-6-peter.maydell@linaro.org
Message-id: 20220506162129.2896966-5-peter.maydell@linaro.org
The Armv8.4 feature FEAT_IDST specifies that exceptions generated by
read accesses to the feature ID space should report a syndrome code
of 0x18 (EC_SYSTEMREGISTERTRAP) rather than 0x00 (EC_UNCATEGORIZED).
The feature ID space is defined to be:
op0 == 3, op1 == {0,1,3}, CRn == 0, CRm == {0-7}, op2 == {0-7}
In our implementation we might return the EC_UNCATEGORIZED syndrome
value for a system register access in four cases:
* no reginfo struct in the hashtable
* cp_access_ok() fails (ie ri->access doesn't permit the access)
* ri->accessfn returns CP_ACCESS_TRAP_UNCATEGORIZED at runtime
* ri->type includes ARM_CP_RAISES_EXC, and the readfn raises
an UNDEF exception at runtime
We have very few regdefs that set ARM_CP_RAISES_EXC, and none of
them are in the feature ID space. (In the unlikely event that any
are added in future they would need to take care of setting the
correct syndrome themselves.) This patch deals with the other
three cases, and enables FEAT_IDST for AArch64 -cpu max.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20220509155457.3560724-1-peter.maydell@linaro.org
