* bugreport: <https://bugs.libre-soc.org/show_bug.cgi?id=1132>
* (External) [qemu source code](https://github.com/qemu/qemu)
-* (External) [qemu cpu flags for x86*](https://www.linux-kvm.org/page/Tuning_KVM)
+* (External) [qemu cpu flags for x86](https://www.linux-kvm.org/page/Tuning_KVM)
## Reasoning:
-Libre-SOC builds upon the OpenPOWER ISA v3.0 but does not include many
-parts of the architecture typically found on production silicon, like
-AltiVec, VSX, and 128-bit hardware floating point. There are no commercially available chips that support OpenPOWER ISA v3.0 SFFS without VSX, AltiVec, and 128-bit hardware floating point.
+Libre-SOC builds upon the OpenPOWER ISA v3.0 but does not include
+many parts of the architecture typically found on production silicon,
+like AltiVec, VSX, and 128-bit hardware floating point. There are no
+commercially available chips that support OpenPOWER ISA v3.0 SFFS without
+VSX, AltiVec, and 128-bit hardware floating point.
-Because of this, we are stuck in a chicken and egg problem where we need the chip to test software, but we also need to test whether software is compliant with SFFS by using such a chip. This is where QEMU as an emulator can help us.
+Because of this, we are stuck in a chicken and egg problem where we need
+the chip to test software, but we also need to test whether software
+is compliant with SFFS by using such a chip. This is where QEMU as an
+emulator can help us.
-The goal is to add a CPU definition in QEMU that specifically emulates an OpenPOWER ISA v3.0 SFFS chip that we can use to run tests and full operating system images on.
+The goal is to add a CPU definition in QEMU that specifically emulates
+an OpenPOWER ISA v3.0 SFFS chip that we can use to run tests and full
+operating system images on.
## Progress and setbacks
### We have first tried using POWER7 which is available on QEMU
- POWER7 runs little endian without VSX, but is only ISA 2.06 compliant.
-* The assumption was that Any additions from ISA 3.0 would be emulated by the Linux kernel, that however did not work.
-* We tried booting [Gentoo SFFS](https://libre-soc.org/SFFS/gentoo_bootstrap/) with a kernel compiled for POWER9, POWER8, and POWER7, all seem to fail with a kernel panic due to missing instructions.
+* The assumption was that Any additions from ISA 3.0 would be emulated
+ by the Linux kernel, that however did not work.
+* We tried booting [[gentoo_bootstrap]] with a kernel compiled
+ for POWER9, POWER8, and POWER7, all seem to fail with a kernel panic
+ due to missing instructions.
* The POWER9 instructions were VSX from the v3.0 ISA
* likewise the POWER8 but also including some v2.08 instructions deprecated in v3.0
* POWER7 included some v2.07 instructions definitely deprecated in v3.0
-* We also tried booting it with a kernel built for all three POWER versions mentioned earlier and also disabling VSX and/or AltiVec, still fails with a kernel panic due to missing instructions.
+* We also tried booting it with a kernel built for all three POWER
+versions mentioned earlier and also disabling VSX and/or AltiVec, still
+fails with a kernel panic due to missing instructions.
### We have tried patching QEMU's definition for POWER9 to exclude VSX and AltiVec
- This did not work and needs considerable amounts of patching.
-* The relevant files are `tcg/ppc/tcg-target.h` and `target/ppc/{compat.c,cpu.c,cpu.h,cpu_init.c,cpu_models.c,cpu_models.h,cpu-qom.h,cpu-param.h}` none seemed to work when removing such features.
+* The relevant files are `tcg/ppc/tcg-target.h` and
+ `target/ppc/{compat.c,cpu.c,cpu.h,cpu_init.c,cpu_models.c,cpu_models.h,cpu-qom.h,cpu-param.h}`
+ none seemed to work when removing such features.
### We have researched in length if the CPU capabilities feature of QEMU is available on POWER
-* It is not available and needs to be added in, unlike x86* and some other architectures that QEMU supports.
-* However, it being a feature of QEMU to start means patching it in will be simpler than adding a fresh feature that was not available in the code before.
+* It is not available and needs to be added in, unlike x86 and some
+ other architectures that QEMU supports.
+* However, it being a feature of QEMU to start means patching it in
+ will be simpler than adding a fresh feature that was not available in
+ the code before.
## Conclusion as of 25 Aug 2023:
-This task will need a considerable amount of research and work
+
+This task will need a considerable amount of additional research and work