mbox series

[v2,00/29] target/riscv: Convert to decodetree

Message ID 20181020071451.27808-1-kbastian@mail.uni-paderborn.de (mailing list archive)
Headers show
Series target/riscv: Convert to decodetree | expand

Message

Bastian Koppelmann Oct. 20, 2018, 7:14 a.m. UTC
Hi,

this patchset converts the RISC-V decoder to decodetree in three major steps:

1) Convert 32-bit instructions to decodetree [Patch 1-14]:
    Many of the gen_* functions are called by the decode functions for 16-bit
    and 32-bit functions. If we move translation code from the gen_*
    functions to the generated trans_* functions of decode-tree, we get a lot of
    duplication. Therefore, we mostly generate calls to the old gen_* function
    which are properly replaced after step 2).

    Each of the trans_ functions are grouped into files corresponding to their
    ISA extension, e.g. addi which is in RV32I is translated in the file
    'trans_rvi.inc.c'.

2) Convert 16-bit instructions to decodetree [Patch 15-17]:
    All 16 bit instructions have a direct mapping to a 32 bit instruction. Thus,
    we convert the arguments in the 16 bit trans_ function to the arguments of
    the corresponding 32 bit instruction and call the 32 bit trans_ function.

3) Remove old manual decoding in gen_* function [Patch 17-28]:
    this move all manual translation code into the trans_* instructions of
    decode tree, such that we can remove the old decode_* functions.

the full tree can be found here:
https://github.com/bkoppelmann/qemu/tree/riscv-dt-v2

Cheers,
Bastian

v1->v2:
        - ex_shift_amount returns uint32_t
        - use ctx->env instead of current_cpu->env_ptr
        - fixed functionspacing
        - RISCV32 now returns false instead of raising an exception
        - shift translators now also use gen_arithm_imm()
        - simplified fence/fence_i as suggested by Richard
        - simplified gen_amo() with function pointers
        - rs2 @atom_ld is now decimal
        - use simplfied gen_amo() with function pointers
        - REQUIRE_FPU uses do {} while (0)
        - Add REQUIRE_FPU to arithm helpers
        - RISCV32 now returns false instead of raising an exception
        - Add REQUIRE_FPU to arithm helpers
        - Stack allocate arg_c_* structs
        - ex_rvc_register returns int
        - special case of trans_c_addi4spn() returns false
        - consistently return false for reserved cases instead of raising an
          exception
        - simplified trans_c_srli by Richard's suggestion
        - remove extract_cj() since its result isn't used
        - trans_branch -> gen_branch
        - trans_load -> gen_load
        - removed negative memop check
        - trans_store -> gen_store
        - removed negative memop check
        - trans_arith_imm -> gen_arith_imm
        - Add missing TARGET_RISC64 checks
        - Reimplement shift translators that were omited in [0004/0028]
        - trans_shift -> gen_shift
        - Add missing TARGET_RISCV64 conditions
        - trans_arith_w -> gen_arith_w
        - Add missing gen_exception_illegal
        - dropped 0028


Bastian Koppelmann (29):
  target/riscv: Move CPURISCVState pointer to DisasContext
  targer/riscv: Activate decodetree and implemnt LUI & AUIPC
  target/riscv: Convert RVXI branch insns to decodetree
  target/riscv: Convert RVXI load/store insns to decodetree
  target/riscv: Convert RVXI arithmetic insns to decodetree
  target/riscv: Convert RVXI fence insns to decodetree
  target/riscv: Convert RVXI csr insns to decodetree
  target/riscv: Convert RVXM insns to decodetree
  target/riscv: Convert RV32A insns to decodetree
  target/riscv: Convert RV64A insns to decodetree
  target/riscv: Convert RV32F insns to decodetree
  target/riscv: Convert RV64F insns to decodetree
  target/riscv: Convert RV32D insns to decodetree
  target/riscv: Convert RV64D insns to decodetree
  target/riscv: Convert RV priv insns to decodetree
  target/riscv: Convert quadrant 0 of RVXC insns to decodetree
  target/riscv: Convert quadrant 1 of RVXC insns to decodetree
  target/riscv: Convert quadrant 2 of RVXC insns to decodetree
  target/riscv: Remove gen_jalr()
  target/riscv: Remove manual decoding from gen_branch()
  target/riscv: Remove manual decoding from gen_load()
  target/riscv: Remove manual decoding from gen_store()
  target/riscv: Move gen_arith_imm() decoding into trans_* functions
  target/riscv: make ADD/SUB/OR/XOR/AND insn use arg lists
  target/riscv: Remove shift and slt insn manual decoding
  target/riscv: Remove manual decoding of RV32/64M insn
  target/riscv: Remove gen_system()
  target/riscv: Remove decode_RV32_64G()
  target/riscv: Rename trans_arith to gen_arith

 target/riscv/Makefile.objs                    |   17 +
 target/riscv/insn16.decode                    |  126 ++
 target/riscv/insn32.decode                    |  256 +++
 .../riscv/insn_trans/trans_privileged.inc.c   |  111 ++
 target/riscv/insn_trans/trans_rva.inc.c       |  244 +++
 target/riscv/insn_trans/trans_rvc.inc.c       |  337 ++++
 target/riscv/insn_trans/trans_rvd.inc.c       |  413 ++++
 target/riscv/insn_trans/trans_rvf.inc.c       |  402 ++++
 target/riscv/insn_trans/trans_rvi.inc.c       |  629 ++++++
 target/riscv/insn_trans/trans_rvm.inc.c       |  125 ++
 target/riscv/translate.c                      | 1756 ++---------------
 11 files changed, 2853 insertions(+), 1563 deletions(-)
 create mode 100644 target/riscv/insn16.decode
 create mode 100644 target/riscv/insn32.decode
 create mode 100644 target/riscv/insn_trans/trans_privileged.inc.c
 create mode 100644 target/riscv/insn_trans/trans_rva.inc.c
 create mode 100644 target/riscv/insn_trans/trans_rvc.inc.c
 create mode 100644 target/riscv/insn_trans/trans_rvd.inc.c
 create mode 100644 target/riscv/insn_trans/trans_rvf.inc.c
 create mode 100644 target/riscv/insn_trans/trans_rvi.inc.c
 create mode 100644 target/riscv/insn_trans/trans_rvm.inc.c

--
2.19.1

Comments

Palmer Dabbelt Oct. 24, 2018, 10:21 p.m. UTC | #1
On Sat, 20 Oct 2018 00:14:22 PDT (-0700), kbastian@mail.uni-paderborn.de wrote:
> Hi,
>
> this patchset converts the RISC-V decoder to decodetree in three major steps:
>
> 1) Convert 32-bit instructions to decodetree [Patch 1-14]:
>     Many of the gen_* functions are called by the decode functions for 16-bit
>     and 32-bit functions. If we move translation code from the gen_*
>     functions to the generated trans_* functions of decode-tree, we get a lot of
>     duplication. Therefore, we mostly generate calls to the old gen_* function
>     which are properly replaced after step 2).
>
>     Each of the trans_ functions are grouped into files corresponding to their
>     ISA extension, e.g. addi which is in RV32I is translated in the file
>     'trans_rvi.inc.c'.
>
> 2) Convert 16-bit instructions to decodetree [Patch 15-17]:
>     All 16 bit instructions have a direct mapping to a 32 bit instruction. Thus,
>     we convert the arguments in the 16 bit trans_ function to the arguments of
>     the corresponding 32 bit instruction and call the 32 bit trans_ function.
>
> 3) Remove old manual decoding in gen_* function [Patch 17-28]:
>     this move all manual translation code into the trans_* instructions of
>     decode tree, such that we can remove the old decode_* functions.
>
> the full tree can be found here:
> https://github.com/bkoppelmann/qemu/tree/riscv-dt-v2

Thanks!

I dropped this on top of master and it appears I'm getting a bunch of oops when 
trying to boot Linux.  They're fairly far into the boot process and may be a 
mistake on my end, I was just wondering if you'd booted Linux?

I'll go through the patches and review them.

>
> Cheers,
> Bastian
>
> v1->v2:
>         - ex_shift_amount returns uint32_t
>         - use ctx->env instead of current_cpu->env_ptr
>         - fixed functionspacing
>         - RISCV32 now returns false instead of raising an exception
>         - shift translators now also use gen_arithm_imm()
>         - simplified fence/fence_i as suggested by Richard
>         - simplified gen_amo() with function pointers
>         - rs2 @atom_ld is now decimal
>         - use simplfied gen_amo() with function pointers
>         - REQUIRE_FPU uses do {} while (0)
>         - Add REQUIRE_FPU to arithm helpers
>         - RISCV32 now returns false instead of raising an exception
>         - Add REQUIRE_FPU to arithm helpers
>         - Stack allocate arg_c_* structs
>         - ex_rvc_register returns int
>         - special case of trans_c_addi4spn() returns false
>         - consistently return false for reserved cases instead of raising an
>           exception
>         - simplified trans_c_srli by Richard's suggestion
>         - remove extract_cj() since its result isn't used
>         - trans_branch -> gen_branch
>         - trans_load -> gen_load
>         - removed negative memop check
>         - trans_store -> gen_store
>         - removed negative memop check
>         - trans_arith_imm -> gen_arith_imm
>         - Add missing TARGET_RISC64 checks
>         - Reimplement shift translators that were omited in [0004/0028]
>         - trans_shift -> gen_shift
>         - Add missing TARGET_RISCV64 conditions
>         - trans_arith_w -> gen_arith_w
>         - Add missing gen_exception_illegal
>         - dropped 0028
>
>
> Bastian Koppelmann (29):
>   target/riscv: Move CPURISCVState pointer to DisasContext
>   targer/riscv: Activate decodetree and implemnt LUI & AUIPC
>   target/riscv: Convert RVXI branch insns to decodetree
>   target/riscv: Convert RVXI load/store insns to decodetree
>   target/riscv: Convert RVXI arithmetic insns to decodetree
>   target/riscv: Convert RVXI fence insns to decodetree
>   target/riscv: Convert RVXI csr insns to decodetree
>   target/riscv: Convert RVXM insns to decodetree
>   target/riscv: Convert RV32A insns to decodetree
>   target/riscv: Convert RV64A insns to decodetree
>   target/riscv: Convert RV32F insns to decodetree
>   target/riscv: Convert RV64F insns to decodetree
>   target/riscv: Convert RV32D insns to decodetree
>   target/riscv: Convert RV64D insns to decodetree
>   target/riscv: Convert RV priv insns to decodetree
>   target/riscv: Convert quadrant 0 of RVXC insns to decodetree
>   target/riscv: Convert quadrant 1 of RVXC insns to decodetree
>   target/riscv: Convert quadrant 2 of RVXC insns to decodetree
>   target/riscv: Remove gen_jalr()
>   target/riscv: Remove manual decoding from gen_branch()
>   target/riscv: Remove manual decoding from gen_load()
>   target/riscv: Remove manual decoding from gen_store()
>   target/riscv: Move gen_arith_imm() decoding into trans_* functions
>   target/riscv: make ADD/SUB/OR/XOR/AND insn use arg lists
>   target/riscv: Remove shift and slt insn manual decoding
>   target/riscv: Remove manual decoding of RV32/64M insn
>   target/riscv: Remove gen_system()
>   target/riscv: Remove decode_RV32_64G()
>   target/riscv: Rename trans_arith to gen_arith
>
>  target/riscv/Makefile.objs                    |   17 +
>  target/riscv/insn16.decode                    |  126 ++
>  target/riscv/insn32.decode                    |  256 +++
>  .../riscv/insn_trans/trans_privileged.inc.c   |  111 ++
>  target/riscv/insn_trans/trans_rva.inc.c       |  244 +++
>  target/riscv/insn_trans/trans_rvc.inc.c       |  337 ++++
>  target/riscv/insn_trans/trans_rvd.inc.c       |  413 ++++
>  target/riscv/insn_trans/trans_rvf.inc.c       |  402 ++++
>  target/riscv/insn_trans/trans_rvi.inc.c       |  629 ++++++
>  target/riscv/insn_trans/trans_rvm.inc.c       |  125 ++
>  target/riscv/translate.c                      | 1756 ++---------------
>  11 files changed, 2853 insertions(+), 1563 deletions(-)
>  create mode 100644 target/riscv/insn16.decode
>  create mode 100644 target/riscv/insn32.decode
>  create mode 100644 target/riscv/insn_trans/trans_privileged.inc.c
>  create mode 100644 target/riscv/insn_trans/trans_rva.inc.c
>  create mode 100644 target/riscv/insn_trans/trans_rvc.inc.c
>  create mode 100644 target/riscv/insn_trans/trans_rvd.inc.c
>  create mode 100644 target/riscv/insn_trans/trans_rvf.inc.c
>  create mode 100644 target/riscv/insn_trans/trans_rvi.inc.c
>  create mode 100644 target/riscv/insn_trans/trans_rvm.inc.c
Bastian Koppelmann Oct. 26, 2018, 10:53 a.m. UTC | #2
On 10/25/18 12:21 AM, Palmer Dabbelt wrote:
> On Sat, 20 Oct 2018 00:14:22 PDT (-0700), 
> kbastian@mail.uni-paderborn.de wrote:
>> Hi,
>>
>> this patchset converts the RISC-V decoder to decodetree in three 
>> major steps:
>>
>> 1) Convert 32-bit instructions to decodetree [Patch 1-14]:
>>     Many of the gen_* functions are called by the decode functions 
>> for 16-bit
>>     and 32-bit functions. If we move translation code from the gen_*
>>     functions to the generated trans_* functions of decode-tree, we 
>> get a lot of
>>     duplication. Therefore, we mostly generate calls to the old gen_* 
>> function
>>     which are properly replaced after step 2).
>>
>>     Each of the trans_ functions are grouped into files corresponding 
>> to their
>>     ISA extension, e.g. addi which is in RV32I is translated in the file
>>     'trans_rvi.inc.c'.
>>
>> 2) Convert 16-bit instructions to decodetree [Patch 15-17]:
>>     All 16 bit instructions have a direct mapping to a 32 bit 
>> instruction. Thus,
>>     we convert the arguments in the 16 bit trans_ function to the 
>> arguments of
>>     the corresponding 32 bit instruction and call the 32 bit trans_ 
>> function.
>>
>> 3) Remove old manual decoding in gen_* function [Patch 17-28]:
>>     this move all manual translation code into the trans_* 
>> instructions of
>>     decode tree, such that we can remove the old decode_* functions.
>>
>> the full tree can be found here:
>> https://github.com/bkoppelmann/qemu/tree/riscv-dt-v2
>
> Thanks!
>
> I dropped this on top of master and it appears I'm getting a bunch of 
> oops when trying to boot Linux.  They're fairly far into the boot 
> process and may be a mistake on my end, I was just wondering if you'd 
> booted Linux?

Are there non fatal oops in booting Linux? I only checked whether I 
could get a terminal on Fedora Linux.

Cheers,

Bastian
Palmer Dabbelt Oct. 27, 2018, 6:20 a.m. UTC | #3
On Fri, 26 Oct 2018 03:53:11 PDT (-0700), kbastian@mail.uni-paderborn.de wrote:
>
> On 10/25/18 12:21 AM, Palmer Dabbelt wrote:
>> On Sat, 20 Oct 2018 00:14:22 PDT (-0700),
>> kbastian@mail.uni-paderborn.de wrote:
>>> Hi,
>>>
>>> this patchset converts the RISC-V decoder to decodetree in three
>>> major steps:
>>>
>>> 1) Convert 32-bit instructions to decodetree [Patch 1-14]:
>>>     Many of the gen_* functions are called by the decode functions
>>> for 16-bit
>>>     and 32-bit functions. If we move translation code from the gen_*
>>>     functions to the generated trans_* functions of decode-tree, we
>>> get a lot of
>>>     duplication. Therefore, we mostly generate calls to the old gen_*
>>> function
>>>     which are properly replaced after step 2).
>>>
>>>     Each of the trans_ functions are grouped into files corresponding
>>> to their
>>>     ISA extension, e.g. addi which is in RV32I is translated in the file
>>>     'trans_rvi.inc.c'.
>>>
>>> 2) Convert 16-bit instructions to decodetree [Patch 15-17]:
>>>     All 16 bit instructions have a direct mapping to a 32 bit
>>> instruction. Thus,
>>>     we convert the arguments in the 16 bit trans_ function to the
>>> arguments of
>>>     the corresponding 32 bit instruction and call the 32 bit trans_
>>> function.
>>>
>>> 3) Remove old manual decoding in gen_* function [Patch 17-28]:
>>>     this move all manual translation code into the trans_*
>>> instructions of
>>>     decode tree, such that we can remove the old decode_* functions.
>>>
>>> the full tree can be found here:
>>> https://github.com/bkoppelmann/qemu/tree/riscv-dt-v2
>>
>> Thanks!
>>
>> I dropped this on top of master and it appears I'm getting a bunch of
>> oops when trying to boot Linux.  They're fairly far into the boot
>> process and may be a mistake on my end, I was just wondering if you'd
>> booted Linux?
>
> Are there non fatal oops in booting Linux? I only checked whether I
> could get a terminal on Fedora Linux.

Most of the stuff past early boot is, as if it's in a userspace process or 
kernel thread you can recover.  My system eventually made it to a shell, but 
there was a lot of carnage on the way in -- most system services blew up.

I tested the same kernel+rootfs on master and it booted cleanly.