[PATCH for-8.1 00/17] centralize CPU extensions logic

Daniel Henrique Barboza posted 17 patches 1 year, 1 month ago
target/riscv/cpu.c | 516 +++++++++++++++++++++++++--------------------
target/riscv/cpu.h |   9 +-
target/riscv/csr.c |  52 ++---
3 files changed, 323 insertions(+), 254 deletions(-)
[PATCH for-8.1 00/17] centralize CPU extensions logic
Posted by Daniel Henrique Barboza 1 year, 1 month ago
Hi,

During the review of a series that did some work in the RISCV_FEATURES*
enum, Liu Zhiwei commented on how we could centralize the all the
extension validation code and integrate it with write_misa() [1].

This does at least part of what was suggested. The idea is that, ATM, we
have too many places setting cpu->cfg and the validation logic is
scattered around (e.g. there are some contraints in write_misa() that
should be applicable elsewhere). This series is an attempt to centralize
things a bit.

The main accomplishments of this series are:

- the parent device riscv-cpu no longer sets any cpu->cfg attribute. This
  was forcing init() functions to disable extensions that it wouldn't
  use just because the parent device was enabling it;

- all validations are centered in validate_set_extensions(), with
  pontual exceptions in write_misa() that has exclusive contraints;

- set_misa() now writes cpu->cfg. No need to have one function to set
  env->misa_ext and then another to set cpu->cfg;

- register_cpu_props() now only exposes user-facing props;

- all validations from validate_set_extensions() are done at the start
  of the function. Validate first, set extensions after;

- RVE is now forbidden in all validations, not just in write_misa();

- RVG is now forbidden in write_misa();

- write_misa now uses set_misa() and validate_set_extensions().



[1] https://lists.gnu.org/archive/html/qemu-devel/2023-02/msg05092.html

Daniel Henrique Barboza (17):
  target/riscv/cpu.c: add riscv_cpu_validate_v()
  target/riscv/cpu.c: remove set_vext_version()
  target/riscv/cpu.c: remove set_priv_version()
  target/riscv: add PRIV_VERSION_LATEST macro
  target/riscv/cpu.c: add riscv_cpu_validate_priv_spec()
  target/riscv: move realize() validations to
    riscv_cpu_validate_set_extensions()
  target/riscv/cpu.c: remove cfg setup from riscv_cpu_init()
  target/riscv/cpu.c: avoid set_misa() in validate_set_extensions()
  target/riscv/cpu.c: set cpu config in set_misa()
  target/riscv/cpu.c: redesign register_cpu_props()
  target/riscv/cpu.c: move riscv_cpu_validate_v() up
  target/riscv: put env->misa_ext <-> cpu->cfg code into helpers
  target/riscv/cpu.c: split riscv_cpu_validate_priv_spec()
  target/riscv/cpu.c: do not allow RVE to be set
  target/riscv: add RVG
  target/riscv: do not allow RVG in write_misa()
  target/riscv: rework write_misa()

 target/riscv/cpu.c | 516 +++++++++++++++++++++++++--------------------
 target/riscv/cpu.h |   9 +-
 target/riscv/csr.c |  52 ++---
 3 files changed, 323 insertions(+), 254 deletions(-)

-- 
2.39.2
Re: [PATCH for-8.1 00/17] centralize CPU extensions logic
Posted by Daniel Henrique Barboza 1 year, 1 month ago
Just realized that the subject doesn't mention 'riscv' anywhere.

Yes, this is target/riscv specific. I'll make sure to mention that in the
future versions.

Thanks,


Daniel

On 3/8/23 17:19, Daniel Henrique Barboza wrote:
> Hi,
> 
> During the review of a series that did some work in the RISCV_FEATURES*
> enum, Liu Zhiwei commented on how we could centralize the all the
> extension validation code and integrate it with write_misa() [1].
> 
> This does at least part of what was suggested. The idea is that, ATM, we
> have too many places setting cpu->cfg and the validation logic is
> scattered around (e.g. there are some contraints in write_misa() that
> should be applicable elsewhere). This series is an attempt to centralize
> things a bit.
> 
> The main accomplishments of this series are:
> 
> - the parent device riscv-cpu no longer sets any cpu->cfg attribute. This
>    was forcing init() functions to disable extensions that it wouldn't
>    use just because the parent device was enabling it;
> 
> - all validations are centered in validate_set_extensions(), with
>    pontual exceptions in write_misa() that has exclusive contraints;
> 
> - set_misa() now writes cpu->cfg. No need to have one function to set
>    env->misa_ext and then another to set cpu->cfg;
> 
> - register_cpu_props() now only exposes user-facing props;
> 
> - all validations from validate_set_extensions() are done at the start
>    of the function. Validate first, set extensions after;
> 
> - RVE is now forbidden in all validations, not just in write_misa();
> 
> - RVG is now forbidden in write_misa();
> 
> - write_misa now uses set_misa() and validate_set_extensions().
> 
> 
> 
> [1] https://lists.gnu.org/archive/html/qemu-devel/2023-02/msg05092.html
> 
> Daniel Henrique Barboza (17):
>    target/riscv/cpu.c: add riscv_cpu_validate_v()
>    target/riscv/cpu.c: remove set_vext_version()
>    target/riscv/cpu.c: remove set_priv_version()
>    target/riscv: add PRIV_VERSION_LATEST macro
>    target/riscv/cpu.c: add riscv_cpu_validate_priv_spec()
>    target/riscv: move realize() validations to
>      riscv_cpu_validate_set_extensions()
>    target/riscv/cpu.c: remove cfg setup from riscv_cpu_init()
>    target/riscv/cpu.c: avoid set_misa() in validate_set_extensions()
>    target/riscv/cpu.c: set cpu config in set_misa()
>    target/riscv/cpu.c: redesign register_cpu_props()
>    target/riscv/cpu.c: move riscv_cpu_validate_v() up
>    target/riscv: put env->misa_ext <-> cpu->cfg code into helpers
>    target/riscv/cpu.c: split riscv_cpu_validate_priv_spec()
>    target/riscv/cpu.c: do not allow RVE to be set
>    target/riscv: add RVG
>    target/riscv: do not allow RVG in write_misa()
>    target/riscv: rework write_misa()
> 
>   target/riscv/cpu.c | 516 +++++++++++++++++++++++++--------------------
>   target/riscv/cpu.h |   9 +-
>   target/riscv/csr.c |  52 ++---
>   3 files changed, 323 insertions(+), 254 deletions(-)
>