Commit 50a2c6e55f introduced a bug where QEMU would segfault on startup
when using KVM on ARM hosts, because kvm_arm_reset_cpu() accesses
cpu->cpreg_reset_values, which is not allocated before
kvm_arch_init_vcpu(). Fix this by not calling cpu_reset() until after
qemu_init_vcpu().
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Andreas Faerber <afaerber@suse.de>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
Message-id: 1401194263-13010-1-git-send-email-christoffer.dall@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Linux makes a habit of writing the same value to the SCTLR that it
already holds. In a sample boot of the kernel to a shell prompt
it wrote the SCTLR with the value it already held 325465 times,
and wrote different values just 3 times.
Skip flushing the TLB if the SCTLR value isn't actually being changed;
this speeds up my sample boot by 3-5%.
Reported-by: Laurent Desnogues <laurent.desnogues@gmail.com>
Reviewed-by: Laurent Desnogues <laurent.desnogues@gmail.com>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 1399560029-19007-1-git-send-email-peter.maydell@linaro.org
After commit 767adce2d, they are redundant. This way we don't assign them
except when needed. Once there, there were lots of cases where the ".fields"
indentation was wrong:
.fields = (VMStateField []) {
and
.fields = (VMStateField []) {
Change all the combinations to:
.fields = (VMStateField[]){
The biggest problem (apart from aesthetics) was that checkpatch complained
when we copy&pasted the code from one place to another.
Signed-off-by: Juan Quintela <quintela@redhat.com>
[PMM: fixed minor conflict, corrected commit message typos]
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Now that we have a CPU object with a reset method, it is better to
keep the KVM reset close to the CPU reset. Using qemu_register_reset
as we do now keeps them far apart.
With this patch, PPC no longer calls the kvm_arch_ function, so
it can get removed there. Other arches call it from their CPU
reset handler, and the function gets an ARMCPU/X86CPU/S390CPU.
Note that ARM- and s390-specific functions are called kvm_arm_*
and kvm_s390_*, while x86-specific functions are called kvm_arch_*.
That follows the convention used by the different architectures.
Changing that is the topic of a separate patch.
Reviewed-by: Gleb Natapov <gnatapov@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
As the macro verifies the value is positive, rename it
to make the function clearer.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Harmless typo as opc1 defaults to zero and opc2 gets
re-declared to its correct value.
Signed-off-by: Edgar E. Iglesias <edgar.iglesias@xilinx.com>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Message-id: 1398926097-28097-4-git-send-email-edgar.iglesias@gmail.com
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
For linked branches, updates to the link register happen
conceptually after the read of the branch target register.
Signed-off-by: Edgar E. Iglesias <edgar.iglesias@xilinx.com>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Cc: qemu-stable@nongnu.org
Message-id: 1398926097-28097-3-git-send-email-edgar.iglesias@gmail.com
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Edgar E. Iglesias <edgar.iglesias@xilinx.com>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Message-id: 1398926097-28097-2-git-send-email-edgar.iglesias@gmail.com
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Like was done for AArch32 for WFE, implement both WFE and YIELD as a
yield operation. This speeds up multi-core system emulation.
Signed-off-by: Rob Herring <rob.herring@linaro.org>
Message-id: 1397588401-20366-1-git-send-email-robherring2@gmail.com
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
XScale defines some implementation-specific coprocessor registers
for doing cache lockdown operations. Since QEMU doesn't model a
cache no proper implementation is possible, but NOP out the
registers so that guest code like u-boot that tries to use them
doesn't crash.
Reported-by: <prqek@centrum.cz>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
The test for the U bit was incorrectly inverted in the scalar case of SQXTUN.
This doesn't affect the vector case as the U bit is used to select XTN(2).
Reported-by: Hao Liu <hao.liu@arm.com>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Claudio Fontana <claudio.fontana@huawei.com>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Cc: qemu-stable@nongnu.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
The smlald (and probably smlsld) instruction was doing incorrect sign
extensions of the operands amongst 64bit result calculation. The
instruction psuedo-code is:
operand2 = if m_swap then ROR(R[m],16) else R[m];
product1 = SInt(R[n]<15:0>) * SInt(operand2<15:0>);
product2 = SInt(R[n]<31:16>) * SInt(operand2<31:16>);
result = product1 + product2 + SInt(R[dHi]:R[dLo]);
R[dHi] = result<63:32>;
R[dLo] = result<31:0>;
The result calculation should be done in 64 bit arithmetic, and hence
product1 and product2 should be sign extended to 64b before calculation.
The current implementation was adding product1 and product2 together
then sign-extending the intermediate result leading to false negatives.
E.G. if product1 = product2 = 0x4000000, their sum = 0x80000000, which
will be incorrectly interpreted as -ve on sign extension.
We fix by doing the 64b extensions on both product1 and product2 before
any addition/subtraction happens.
We also fix where we were possibly incorrectly setting the Q saturation
flag for SMLSLD, which the ARM ARM specifically says is not set.
Reported-by: Christina Smith <christina.smith@xilinx.com>
Signed-off-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 2cddb6f5a15be4ab8d2160f3499d128ae93d304d.1397704570.git.peter.crosthwaite@xilinx.com
Cc: qemu-stable@nongnu.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Clean up useless 'break' statement after 'return' statement.
Signed-off-by: Chen Gang <gang.chen.5i5j@gmail.com>
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
For system mode, we may have a 64 bit CPU which is currently executing
in AArch32 state; if we're dumping CPU state to the logs we should
therefore show the correct state for the current execution state,
rather than hardwiring it based on the type of the CPU. For consistency
with how we handle translation, we leave the 32 bit dump function
as the default, and have it hand off control to the 64 bit dump code
if we're in AArch64 mode.
Reported-by: Rob Herring <rob.herring@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
The AArch64 implementation of the set_pc method needs to be updated to
handle the possibility that the CPU is in AArch32 mode; otherwise there
are weird crashes when doing interprocessing in system emulation mode
when an interrupt occurs and we fail to resynchronize the 32-bit PC
with the TB we need to execute next.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
The Cortex-A15's CBAR register is actually read-only (unlike that
of the Cortex-A9). Correct our model to match the hardware.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
The Cortex-A57, like most of the other ARM cores, has a CBAR
register which defines the base address of the per-CPU
peripherals. However it has a 64-bit view as well as a
32-bit view; expand the QOM reset-cbar property from UINT32
to UINT64 so this can be specified, and implement the
32-bit and 64-bit views of a 64-bit CBAR.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Implement a subset of the Cortex-A57's implementation defined system
registers. We provide RAZ/WI or reads-as-constant/writes-ignored
implementations of the various control and syndrome reigsters.
We do not implement registers which provide direct access to and
manipulation of the L1 cache, since QEMU doesn't implement caches.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Acked-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Implement the AArch64 RVBAR register, which indicates the reset
address. Since the reset address is implementation defined and
usually configurable by setting config signals in hardware, we
also provide a QOM property so it can be set at board level if
necessary.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Implement the AArch64 address translation operations.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Implement the auxiliary fault status registers AFSR0_EL1 and
AFSR1_EL1. These are present on v7 and later, and have IMPDEF
behaviour; we choose to RAZ/WI for all cores.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Many of the reginfo definitions in cp_reginfo[] use CP_ANY wildcards.
This is for a combination of reasons:
* early ARM implementations really did underdecode
* earlier versions of QEMU underdecoded and we can't tighten
this up because we don't know if guests really require this or not
* implementation convenience
For ARMv8 the architecture has tightened things up and system and
coprocessor registers are always specifically decoded. We take
advantage of this opportunity for a clean break by restricting
our CP_ANY wildcarded reginfo to pre-v8 CPUs, and providing
specifically decoded versions where necessary for v8 CPUs.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Acked-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
In ARMv8 the 32 bit coprocessor ID register space is tidied up to
remove the wildcarded aliases of the MIDR and the RAZ behaviour
for the unassigned space where crm = 3..7. Make sure we don't
expose thes wildcards for v8 cores. This means we need to have
a specific implementation for REVIDR, an IMPDEF register which
may be the same as the MIDR (and which we always implement as such).
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
The AArch64 usermode 'any' CPU type was accidentally specified
with the ARM_FEATURE_THUMB2EE bit set. This is incorrect since
ARMv8 removes Thumb2EE completely. Since we never implemented
Thumb2EE anyway having the feature bit set was fairly harmless
for user-mode, but the correct thing is to not set it at all.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Implement the ISR_EL1 register. This is actually present in
ARMv7 as well but was previously unimplemented. It is a
read-only register that indicates whether interrupts are
currently pending.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Implement the AArch64 view of the ACTLR (auxiliary control
register). Note that QEMU internally tends to call this
AUXCR for historical reasons.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Implement AArch64 view of the CONTEXTIDR register.
We tighten up the condition when we flush the TLB on a CONTEXTIDR
write to avoid needlessly flushing the TLB every time on a 64
bit system (and also on a 32 bit system using LPAE, as a bonus).
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
All the AArch32 ID registers are visible from AArch64
(in addition to the AArch64-specific ID_AA64* registers).
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
For ARMv8 there are two changes to the MVFR media feature registers:
* there is a new MVFR2 which is accessible from 32 bit code
* 64 bit code accesses these via the usual sysreg instructions
rather than with a floating-point specific instruction
Implement this.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Implement exception handling for AArch64 EL1. Exceptions from AArch64 or
AArch32 EL0 are supported.
Signed-off-by: Rob Herring <rob.herring@linaro.org>
[PMM: fixed minor style nits; updated to match changes in
previous patches; added some of the simpler cases of
illegal-exception-return support]
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Move arm_log_exception() into internals.h so we can use it from
helper-a64.c for the AArch64 exception entry code.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Implement the AArch64 SPSR_EL1. For compatibility with how KVM
handles SPSRs and with the architectural mapping between AArch32
and AArch64, we put this in the banked_spsr[] array in the slot
that is used for SVC in AArch32. This means we need to extend the
array from uint32_t to uint64_t, which requires some reworking
of the 32 bit KVM save/restore code.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Acked-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Implement handling for the AArch64 SP_EL0 system register.
This holds the EL0 stack pointer, and is only accessible when
it's not being used as the stack pointer, ie when we're in EL1
and EL1 is using its own stack pointer. We also provide a
definition of the SP_EL1 register; this isn't guest visible
as a system register for an implementation like QEMU which
doesn't provide EL2 or EL3; however it is useful for ensuring
the underlying state is migrated.
We need to update the state fields in the CPU state whenever
we switch stack pointers; this happens when we take an exception
and also when SPSEL is used to change the bit in PSTATE which
indicates which stack pointer EL1 should use.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Add the AArch64 ELR_EL1 register.
Note that this does not live in env->cp15: for KVM migration
compatibility we need to migrate it separately rather than
as part of the system registers, because the KVM-to-userspace
interface puts it in the struct kvm_regs rather than making
them visible via the ONE_REG ioctls.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Implement AArch64 views of ESR_EL1 and FAR_EL1, and make the 32 bit
DFSR, DFAR, IFAR share state with them as architecturally specified.
The IFSR doesn't share state with any AArch64 register visible at EL1,
so just rename the state field without widening it to 64 bits.
Signed-off-by: Rob Herring <rob.herring@linaro.org>
[PMM: Minor tweaks; fix some bugs involving inconsistencies between
use of offsetof() or offsetoflow32() and struct field width]
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
The ARM946 model currently uses the c5_data and c5_insn fields in the CPU
state struct to store the contents of its access permission registers.
This is confusing and a good source of bugs because for all the MMU-based
CPUs those fields are fault status and fault address registers, which
behave completely differently; they just happen to use the same cpreg
encoding. Split them out to use their own fields instead.
These registers are only present in PMSAv5 MPU systems (of which the
ARM946 is our only current example); PMSAv6 and PMSAv7 (which we have
no implementations of) handle access permissions differently. We name
the new state fields accordingly.
Note that this change fixes a bug where a data abort or prefetch abort
on the ARM946 would accidentally corrupt the access permission registers
because the interrupt handling code assumed the c5_data and c5_insn
fields were always fault status registers.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Implement the DC ZVA instruction, which clears a block of memory.
The fast path obtains a pointer to the underlying RAM via the TCG TLB
data structure so we can do a direct memset(), with fallback to a
simple byte-store loop in the slow path.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Acked-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Suppress the ID_AA64DFR0_EL1 PMUVer field, even if the CPU specific
value claims that it exists. QEMU doesn't currently implement it,
and not advertising it prevents the guest from trying to use it
and getting UNDEFs on unimplemented registers.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
---
This is arguably a hack, but otherwise Linux tries to prod
half a dozen PMU sysregs.
Add support for v8 page table walks. This supports stage 1 translations
for 4KB, 16KB and 64KB page sizes starting with 0 or 1 level.
Signed-off-by: Rob Herring <rob.herring@linaro.org>
[PMM: fix style nits, fold in 16/64K page support patch, use
arm_el_is_aa64() to decide whether to do 64 bit page table walk]
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
The current A32/T32 decoder bases its "is VFP/Neon enabled?" check
on the FPSCR.EN bit. This is correct if EL1 is AArch32, but for
an AArch64 EL1 the logic is different: it must act as if FPSCR.EN
is always set. Instead, trapping must happen according to CPACR
bits for cp10/cp11; these cover all of FP/Neon, including the
FPSCR/FPSID/MVFR register accesses which FPSCR.EN does not affect.
Add support for CPACR checks (which are also required for ARMv7,
but were unimplemented because Linux happens not to use them)
and make sure they generate exceptions with the correct syndrome.
We actually return incorrect syndrome information for cases
where FP is disabled but the specific instruction bit pattern
is unallocated: strictly these should be the Uncategorized
exception, not a "SIMD disabled" exception. This should be
mostly harmless, and the structure of the A32/T32 VFP/Neon
decoder makes it painful to put the 'FP disabled?' checks in
the right places.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Because unallocated encodings generate different exception syndrome
information from traps due to FP being disabled, we can't do a single
"is fp access disabled" check at a high level in the decode tree.
To help in catching bugs where the access check was forgotten in some
code path, we set this flag when the access check is done, and assert
that it is set at the point where we actually touch the FP regs.
This requires us to pass the DisasContext to the vec_reg_offset
and fp_reg_offset functions.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
For the A64 instruction set, the only FP/Neon disable trap
is the CPACR FPEN bits, which may indicate "enabled", "disabled"
or "disabled for EL0". Add a bit to the AArch64 tb flags indicating
whether FP/Neon access is currently enabled and make the decoder
emit code to raise exceptions on use of FP/Neon insns if it is not.
We use a new flag in DisasContext rather than borrowing the
existing vfp_enabled flag because the A32/T32 decoder is going
to need both.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Acked-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
---
I'm aware this is a rather hard to review patch; sorry.
I have done an exhaustive check that we have fp access checks
in all code paths with the aid of the assertions added in the
next patch plus the code-coverage hack patch I posted to the
list earlier.
This patch is correct as of
09e037354 target-arm: A64: Add saturating accumulate ops (USQADD/SUQADD)
which was the last of the Neon insns to be added, so assuming
no refactoring of the code it should be fine.
Set up the required syndrome information when we detect an MMU fault.
Signed-off-by: Rob Herring <rob.herring@linaro.org>
[PMM: split out from exception handling patch, tweaked to bring
in line with how we create other kinds of syndrome information]
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Add new helpers exception_with_syndrome (for generating an exception
with syndrome information) and exception_uncategorized (for generating
an exception with "Unknown or Uncategorized Reason", which have a syndrome
register value of zero), and use them to generate the correct syndrome
information for exceptions which are raised directly from generated code.
This patch includes moving the A32/T32 gen_exception_insn functions
further up in the source file; they will be needed for "VFP/Neon disabled"
exception generation later.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
For exceptions taken to AArch64, if a coprocessor/system register
access fails due to a trap or enable bit then the syndrome information
must include details of the failing instruction (crn/crm/opc1/opc2
fields, etc). Make the decoder construct the syndrome information
at translate time so it can be passed at runtime to the access-check
helper function and used as required.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
For AArch32 exceptions, the only information provided about
the cause of an exception is the individual exception type (data
abort, undef, etc), which we store in cs->exception_index. For
AArch64, the CPU provides much more detail about the cause of
the exception, which can be found in the syndrome register.
Create a set of fields in CPUARMState which must be filled in
whenever an exception is raised, so that exception entry can
correctly fill in the syndrome register for the guest.
This includes the information which in AArch32 appears in
the DFAR and IFAR (fault address registers) and the DFSR
and IFSR (fault status registers) for data aborts and
prefetch aborts, since if we end up taking the MMU fault
to AArch64 rather than AArch32 this will need to end up
in different system registers.
This patch does a refactoring which moves the setting of the
AArch32 DFAR/DFSR/IFAR/IFSR from the point where the exception
is raised to the point where it is taken. (This is no change
for cores with an MMU, retains the existing clearly incorrect
behaviour for ARM946 of trashing the MP access permissions
registers which share the c5_data and c5_insn state fields,
and has no effect for v7M because we don't implement its
MPU fault status or address registers.)
As a side effect of the cleanup we fix a bug in the AArch64
linux-user mode code where we were passing a 64 bit fault
address through the 32 bit c6_data/c6_insn fields: it now
goes via the always-64-bit exception.vaddress.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Implement the DAIF system register which is a view of the
DAIF bits in PSTATE. To avoid needing a readfn, we widen
the daif field in CPUARMState to uint64_t.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
