BookS does not take single step interrupts on completion of rfi and
similar (rfid, hrfid, rfscv). This is not a completely clean way to
do it, but in general non-branch instructions that change NIP on
completion are excluded.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Improve the emulation accuracy of the single step and branch trace
interrupts for v2.07S. Set SRR1[33]=1, and set SIAR to completed
instruction address.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Single-step interrupts are suppressed if the nip is between 0x100 and
0xf00. This has been the case for a long time and it's not clear what
the intention is. Likely either an attempt to suppress trace interrupts
for instructions that cause an interrupt on completion, or a workaround
to prevent software tripping over itself single stepping its interrupt
handlers.
BookE interrupt vectors are set by IVOR registers, and BookS has AIL
modes and new interrupt types, so there are many interrupts including
the debug interrupt which can be outside this range. So any effect it
might have had does not cover most cases (including Linux on recent
BookS CPUs).
Remove this special case.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
[ clg : fixed typo in commit logs ]
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Linux sets these to control cache flush behaviour on Power9. Supervisor
and hypervisor are allowed to write, and reads are noops.
Add implementations to avoid noisy messages when booting Linux under the
pseries machine with guest_errors enabled.
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Joel Stanley <joel@jms.id.au>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Change radix model to always generate a storage interrupt when the R/C
bits are not set appropriately in a PTE instead of setting the bits
itself. According to the ISA both behaviors are valid, but in practice
this change more closely matches behavior observed on the POWER9 CPU.
From the POWER9 Processor User's Manual, Section 4.10.13.1: "When
performing Radix translation, the POWER9 hardware triggers the
appropriate interrupt ... for the mode and type of access whenever
Reference (R) and Change (C) bits require setting in either the guest or
host page-table entry (PTE)."
Signed-off-by: Shawn Anastasio <sanastasio@raptorengineering.com>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Changes the signature of the target-defined functions for
inserting/removing kvm hw breakpoints. The address and length arguments
are now of vaddr type, which both matches the type used internally in
accel/kvm/kvm-all.c and makes the api target-agnostic.
Signed-off-by: Anton Johansson <anjo@rev.ng>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20230807155706.9580-4-anjo@rev.ng>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
kvm_arch_get_default_type() returns the default KVM type. This hook is
particularly useful to derive a KVM type that is valid for "none"
machine model, which is used by libvirt to probe the availability of
KVM.
For MIPS, the existing mips_kvm_type() is reused. This function ensures
the availability of VZ which is mandatory to use KVM on the current
QEMU.
Cc: qemu-stable@nongnu.org
Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Message-id: 20230727073134.134102-2-akihiko.odaki@daynix.com
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
[PMM: added doc comment for new function]
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Until v2.07s, the VRMA page size (L||LP) was encoded in LPCR[VRMASD].
In v3.0 that moved to the partition table PS field.
The powernv machine can now run KVM HPT guests on POWER9/10 CPUs with
this fix and the patch to add ASDR.
Fixes: 3367c62f52 ("target/ppc: Support for POWER9 native hash")
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Message-ID: <20230730111842.39292-1-npiggin@gmail.com>
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
HDEC is defined to not wake from PM state. There is a check in the HDEC
timer to avoid setting the interrupt if we are in a PM state, but no
check on PM entry to lower HDEC if it already fired. This can cause a
HDECR wake up and QEMU abort with unsupported exception in Power Save
mode.
Fixes: 4b236b621b ("ppc: Initial HDEC support")
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Message-ID: <20230726182230.433945-4-npiggin@gmail.com>
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
The ASDR register was introduced in ISA v3.0. It has not been
implemented for HPT. With HPT, ASDR is the format of the slbmte RS
operand (containing VSID), which matches the ppc_slb_t field.
Fixes: 3367c62f52 ("target/ppc: Support for POWER9 native hash")
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Message-ID: <20230726182230.433945-2-npiggin@gmail.com>
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
The change to use translator_use_goto_tb went too far, as the
CF_SINGLE_STEP flag managed by the translator only handles
gdb single stepping and not the architectural single stepping
modeled in DisasContext.singlestep_enabled.
Fixes: 6e9cc373ec ("target/ppc: Use translator_use_goto_tb")
Resolves: https://gitlab.com/qemu-project/qemu/-/issues/1795
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
The Power ISA has the concept of sub-processors:
Hardware is allowed to sub-divide a multi-threaded processor into
"sub-processors" that appear to privileged programs as multi-threaded
processors with fewer threads.
POWER9 and POWER10 have two modes, either every thread is a
sub-processor or all threads appear as one multi-threaded processor. In
the user manuals these are known as "LPAR per thread" / "Thread LPAR",
and "LPAR per core" / "1 LPAR", respectively.
The practical difference is: in thread LPAR mode, non-hypervisor SPRs
are not shared between threads and msgsndp can not be used to message
siblings. In 1 LPAR mode, some SPRs are shared and msgsndp is usable.
Thrad LPAR allows multiple partitions to run concurrently on the same
core, and is a requirement for KVM to run on POWER9/10 (which does not
gang-schedule an LPAR on all threads of a core like POWER8 KVM).
Traditionally, SMT in PAPR environments including PowerVM and the
pseries QEMU machine with KVM acceleration behaves as in 1 LPAR mode.
In OPAL systems, Thread LPAR is used. When adding SMT to the powernv
machine, it is therefore preferable to emulate Thread LPAR.
To account for this difference between pseries and powernv, an LPAR mode
flag is added such that SPRs can be implemented as per-LPAR shared, and
that becomes either per-thread or per-core depending on the flag.
Reviewed-by: Joel Stanley <joel@jms.id.au>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Tested-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Message-ID: <20230705120631.27670-2-npiggin@gmail.com>
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Currently on PPC64 qemu always dumps the guest memory in
Big Endian (BE) format even though the guest running in Little Endian
(LE) mode. So crash tool fails to load the dump as illustrated below:
Log :
$ virsh dump DOMAIN --memory-only dump.file
Domain 'DOMAIN' dumped to dump.file
$ crash vmlinux dump.file
<snip>
crash 8.0.2-1.el9
WARNING: endian mismatch:
crash utility: little-endian
dump.file: big-endian
WARNING: machine type mismatch:
crash utility: PPC64
dump.file: (unknown)
crash: dump.file: not a supported file format
<snip>
This happens because cpu_get_dump_info() passes cpu->env->has_hv_mode
to function ppc_interrupts_little_endian(), the cpu->env->has_hv_mode
always set for powerNV even though the guest is not running in hv mode.
The hv mode should be taken from msr_mask MSR_HVB bit
(cpu->env.msr_mask & MSR_HVB). This patch fixes the issue by passing
MSR_HVB value to ppc_interrupts_little_endian() in order to determine
the guest endianness.
The crash tool also expects guest kernel endianness should match the
endianness of the dump.
The patch was tested on POWER9 box booted with Linux as host in
following cases:
Host-Endianess Qemu-Target-Machine Qemu-Generated-Guest
Memory-Dump-Format
BE powernv(OPAL/PowerNV) LE
BE powernv(OPAL/PowerNV) BE
LE powernv(OPAL/PowerNV) LE
LE powernv(OPAL/PowerNV) BE
LE pseries(OPAL/PowerNV/pSeries) KVMHV LE
LE pseries TCG LE
Fixes: 5609400a42 ("target/ppc: Set the correct endianness for powernv memory
dumps")
Signed-off-by: Narayana Murty N <nnmlinux@linux.ibm.com>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
Reviewed-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Vaibhav Jain <vaibhav@linux.ibm.com>
Message-ID: <20230623072506.34713-1-nnmlinux@linux.ibm.com>
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
TFMR is the Time Facility Management Register which is specific to
POWER CPUs, and used for the purpose of timebase management (generally
by firmware, not the OS).
Add helpers for the TFMR register, which will form part of the core
timebase facility model in future but for now behaviour is unchanged.
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Message-ID: <20230625120317.13877-3-npiggin@gmail.com>
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
POWER book4 (implementation-specific) SPRs are sometimes in their own
functions, but in other cases are mixed with architected SPRs. Do some
spring cleaning on these.
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Message-ID: <20230625120317.13877-2-npiggin@gmail.com>
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
The 'kvm_sw_tlb' and 'tlb_dirty' fields introduced in commit
93dd5e852c ("kvm: ppc: booke206: use MMU API") are specific
to KVM and shouldn't be accessed when it is not available.
Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
Message-Id: <20230624192645.13680-1-philmd@linaro.org>
Doorbells in SMT need to coordinate msgsnd/msgclr and DPDES access from
multiple threads that affect the same state.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
A relatively simple case to begin with, CTRL is a SMT shared register
where reads and writes need to synchronise against state changes by
other threads in the core.
Atomic serialisation operations are used to achieve this.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
TGC SMT emulation needs to know whether it is running with SMT siblings,
to be able to iterate over siblings in a core, and to serialise
threads to access per-core shared SPRs. Add infrastructure to do these
things.
For now the sibling iteration and serialisation are implemented in a
simple but inefficient way. SMT shared state and sibling access is not
too common, and SMT configurations are mainly useful to test system
code, so performance is not to critical.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
[ clg: fix build breakage with clang ]
Signed-off-by: Cédric Le Goater <clg@kaod.org>
The top bits of the LEV field of the sc instruction are to be treated as
as a reserved field rather than a reserved value, meaning LEV is
effectively the bottom bit. LEV=0xF should be treated as LEV=1 and be
a hypercall, for example.
This changes the instruction execution to just set lev from the low bit
of the field. Processors which don't support the LEV field will continue
to ignore it.
ISA v3.1 defines LEV to be 2 bits, in order to add the 'sc 2' ultracall
instruction. TCG does not support Ultravisor, so don't worry about
that bit.
Suggested-by: "Harsh Prateek Bora" <harshpb@linux.ibm.com>
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Reviewed-by: Harsh Prateek Bora <harshpb@linux.ibm.com>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
The CTRL register is able to write the bit in the RUN field, which gets
reflected into the TS field which is read-only and contains the state of
the RUN field for all threads in the core.
TCG does not implement SMT, so the correct implementation just requires
mirroring the RUN bit into the first bit of the TS field.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
System call interrupts in ISA v3.1 CPUs add a LEV indication in SRR1
that corresponds with the LEV field of the instruction that caused the
interrupt.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Reviewed-by: Harsh Prateek Bora <harshpb@linux.ibm.com>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
The hypervisor emulation assistance interrupt modifies HEIR to
contain the value of the instruction which caused the exception.
Only TCG raises HEAI interrupts so this can be made TCG-only.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
ISA v3.1 introduced prefix instructions. Among the changes, various
synchronous interrupts report whether they were caused by a prefix
instruction in (H)SRR1.
The case of instruction fetch that causes an HDSI due to access of a
process-scoped table faulting on the partition scoped translation is the
tricky one. As with ISIs and HISIs, this does not try to set the prefix
bit because there is no instruction image to be loaded. The HDSI needs
the originating access type to be passed through to the handler to
distinguish this from HDSIs that fault translating process scoped tables
originating from a load or store instruction (in that case the prefix
bit should be provided).
Reviewed-by: Fabiano Rosas <farosas@suse.de>
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
[ clg: checkpatch issues ]
Signed-off-by: Cédric Le Goater <clg@kaod.org>
