[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[PATCH 00/22] exec/cpu: Call cpu_exec_realizefn() once in cpu_common_rea
From: |
Philippe Mathieu-Daudé |
Subject: |
[PATCH 00/22] exec/cpu: Call cpu_exec_realizefn() once in cpu_common_realize() |
Date: |
Mon, 18 Sep 2023 18:02:33 +0200 |
Hi,
TL;DR: This series factor duplicated common code in CPUs
DeviceRealize() handlers out, moving as a single call in
cpu_common_realize().
In an effort to have most of:
- CPU core code independant of accelerators
- CPU core code target agnostic
- CPU target code independant of accelerators
- and cpu_reset() called automatically without having to
rely on global QEMUResetHandler,
I'm working on the core CPU code, unfortunately touching
files in all targets.
I suppose them term "exec" used in various areas of QEMU
started from what we call today "accel[erators]" [*]. So
cpu_exec_realizefn() can be read as cpu_accel_realize(),
or "generic code where an accelerator realizes its internal
fields on an abstract (target independent) CPU".
This series moves a common pattern used in all target's
cpu_realize() handlers to the common cpu_exec_realizefn().
Some optional code is used to check CPU requested features
are compatible with the accelerator possibilities. We
extracted this code as CPUClass::verify_accel_features()
handler. Better name welcomed :)
Some targets were calling cpu_reset() *before*
cpu_common_realizefn(), we moved it *after* (since RESET
shouldn't happen before REALIZE). I still have to audit
each target to confirm there are no side effects.
Besides this cpu_reset() change, the rest should be
relatively trivial to review, still I'd like feedback
from the respective target maintainers for the "move HW
creation after vCPU one" patches.
Regards,
Phil.
Follow-up: Make cpu_reset() accel-agnostic and move it
to cpu_common_realize() (not trivial due to
KVM run_on_cpu() calls).
[*] If Paolo/Richard confirm, I might post series renaming
various APIs s/exec/accel/, because various headers
meaning aren't clear to me.
Philippe Mathieu-Daudé (21):
target/i386: Only realize existing APIC device
hw/intc/apic: Pass CPU using QOM link property
target/i386/kvm: Correct comment in kvm_cpu_realize()
exec/cpu: Never call cpu_reset() before cpu_realize()
exec/cpu: Call qemu_init_vcpu() once in cpu_common_realize()
exec/cpu: Call cpu_remove_sync() once in cpu_common_unrealize()
exec/cpu: RFC Destroy vCPU address spaces in cpu_common_unrealize()
target/arm: Create timers *after* accelerator vCPU is realized
target/hppa: Create timer *after* accelerator vCPU is realized
target/nios2: Create IRQs *after* accelerator vCPU is realized
target/mips: Create clock *after* accelerator vCPU is realized
target/xtensa: Create IRQs *after* accelerator vCPU is realized
target/sparc: Init CPU environment *after* accelerator vCPU is
realized
exec/cpu: Introduce CPUClass::verify_accel_features()
target/arm: Extract verify_accel_features() from cpu_realize()
target/i386: Extract verify_accel_features() from cpu_realize()
target/s390x: Call s390_cpu_realize_sysemu from s390_realize_cpu_model
target/s390x: Have s390_realize_cpu_model() return a boolean
target/s390x: Use s390_realize_cpu_model() as verify_accel_features()
exec/cpu: Have cpu_exec_realize() return a boolean
exec/cpu: Call cpu_exec_realizefn() once in cpu_common_realize()
xianglai li (1):
exec/cpu: Introduce the CPU address space destruction function
include/exec/cpu-common.h | 8 ++++
include/hw/core/cpu.h | 7 +++-
target/s390x/s390x-internal.h | 4 +-
cpu.c | 11 ++++-
hw/core/cpu-common.c | 15 +++++++
hw/intc/apic_common.c | 2 +
softmmu/physmem.c | 24 +++++++++++
target/alpha/cpu.c | 10 -----
target/arm/cpu.c | 53 ++++++++++++------------
target/avr/cpu.c | 10 +----
target/cris/cpu.c | 11 +----
target/hexagon/cpu.c | 11 +----
target/hppa/cpu.c | 20 ++-------
target/i386/cpu-sysemu.c | 20 ++++-----
target/i386/cpu.c | 77 ++++++++++++++++++-----------------
target/i386/kvm/kvm-cpu.c | 3 +-
target/loongarch/cpu.c | 11 +----
target/m68k/cpu.c | 11 +----
target/microblaze/cpu.c | 9 ----
target/mips/cpu.c | 27 ++++--------
target/nios2/cpu.c | 20 +++------
target/openrisc/cpu.c | 11 +----
target/ppc/cpu_init.c | 8 ----
target/riscv/cpu.c | 10 +----
target/rx/cpu.c | 11 +----
target/s390x/cpu-sysemu.c | 3 +-
target/s390x/cpu.c | 21 +---------
target/s390x/cpu_models.c | 16 +++++---
target/sh4/cpu.c | 11 +----
target/sparc/cpu.c | 10 -----
target/tricore/cpu.c | 10 +----
target/xtensa/cpu.c | 13 +-----
32 files changed, 189 insertions(+), 299 deletions(-)
--
2.41.0
- [PATCH 00/22] exec/cpu: Call cpu_exec_realizefn() once in cpu_common_realize(),
Philippe Mathieu-Daudé <=
- [PATCH 01/22] target/i386: Only realize existing APIC device, Philippe Mathieu-Daudé, 2023/09/18
- [PATCH 02/22] hw/intc/apic: Pass CPU using QOM link property, Philippe Mathieu-Daudé, 2023/09/18
- [PATCH 03/22] target/i386/kvm: Correct comment in kvm_cpu_realize(), Philippe Mathieu-Daudé, 2023/09/18
- [RFC PATCH 04/22] exec/cpu: Never call cpu_reset() before cpu_realize(), Philippe Mathieu-Daudé, 2023/09/18
- [PATCH 05/22] exec/cpu: Call qemu_init_vcpu() once in cpu_common_realize(), Philippe Mathieu-Daudé, 2023/09/18
- [PATCH 06/22] exec/cpu: Call cpu_remove_sync() once in cpu_common_unrealize(), Philippe Mathieu-Daudé, 2023/09/18
- [PATCH 07/22] exec/cpu: Introduce the CPU address space destruction function, Philippe Mathieu-Daudé, 2023/09/18
- [PATCH 10/22] target/hppa: Create timer *after* accelerator vCPU is realized, Philippe Mathieu-Daudé, 2023/09/18
- [PATCH 12/22] target/mips: Create clock *after* accelerator vCPU is realized, Philippe Mathieu-Daudé, 2023/09/18
- [PATCH 08/22] exec/cpu: RFC Destroy vCPU address spaces in cpu_common_unrealize(), Philippe Mathieu-Daudé, 2023/09/18