Re: [PATCH v4 1/3] RISC-V: Minimal support of scalar crypto extension

[Kito Cheng <kito.cheng@sifive.com>]

Maybe hold a string for the canonical order? I think accepting those
extensions is kind of buggy.

My patch for this proposal:
https://sourceware.org/pipermail/binutils/2022-February/119827.html


On Mon, Feb 21, 2022 at 10:44 PM Tsukasa OI
<research_trasio@irq.a4lg.com> wrote:
>
>
>
> On 2022/02/21 23:16, Philipp Tomsich wrote:
> > On Mon, 21 Feb 2022 at 15:14, <jiawei@iscas.ac.cn> wrote:
> >>
> >>
> >>
> >>
> >> &gt; -----原始邮件-----
> >> &gt; 发件人: "Jan Beulich" <jbeulich@suse.com>
> >> &gt; 发送时间: 2022-02-21 21:24:11 (星期一)
> >> &gt; 收件人: jiawei <jiawei@iscas.ac.cn>
> >> &gt; 抄送: kito.cheng@sifive.com, nelson.chu@sifive.com, jimw@sifive.com, philipp.tomsich@vrull.eu, mjos@pqshield.com, ben.marshall@pqshield.com, cmuellner@ventanamicro.com, palmer@dabbelt.com, andrew@sifive.com, lazyparser@gmail.com, siyu@isrc.iscas.ac.cn, schwab@linux-m68k.org, binutils@sourceware.org
> >> &gt; 主题: Re: [PATCH v4 1/3] RISC-V: Minimal support of scalar crypto extension
> >> &gt;
> >> &gt; On 15.11.2021 04:03, jiawei wrote:
> >> &gt; &gt; Minimal support of scalar crypto extension, add "k" in riscv_supported_std_ext[] to make the   order check right with "zk" behind "zb". "zbk*" is sub-extension for k-ext, so it added behind "zbs" in riscv_supported_std_z_ext[].
> >> &gt; &gt; ---
> >> &gt; &gt;  bfd/elfxx-riscv.c | 28 ++++++++++++++++++++++++++++
> >> &gt; &gt;  1 file changed, 28 insertions(+)
> >> &gt; &gt;
> >> &gt; &gt; diff --git a/bfd/elfxx-riscv.c b/bfd/elfxx-riscv.c
> >> &gt; &gt; index 3ffbaad66dd..152fbe3d160 100644
> >> &gt; &gt; --- a/bfd/elfxx-riscv.c
> >> &gt; &gt; +++ b/bfd/elfxx-riscv.c
> >> &gt; &gt; @@ -1075,6 +1075,20 @@ static struct riscv_implicit_subset riscv_implicit_subsets[] =
> >> &gt; &gt;    {"q", "d",         check_implicit_always},
> >> &gt; &gt;    {"d", "f",         check_implicit_always},
> >> &gt; &gt;    {"f", "zicsr",     check_implicit_always},
> >> &gt; &gt; +  {"zk", "zkn",      check_implicit_always},
> >> &gt; &gt; +  {"zk", "zkr",      check_implicit_always},
> >> &gt; &gt; +  {"zk", "zkt",      check_implicit_always},
> >> &gt; &gt; +  {"zkn", "zbkb",    check_implicit_always},
> >> &gt; &gt; +  {"zkn", "zbkc",    check_implicit_always},
> >> &gt; &gt; +  {"zkn", "zbkx",    check_implicit_always},
> >> &gt; &gt; +  {"zkn", "zkne",    check_implicit_always},
> >> &gt; &gt; +  {"zkn", "zknd",    check_implicit_always},
> >> &gt; &gt; +  {"zkn", "zknh",    check_implicit_always},
> >> &gt; &gt; +  {"zks", "zbkb",    check_implicit_always},
> >> &gt; &gt; +  {"zks", "zbkc",    check_implicit_always},
> >> &gt; &gt; +  {"zks", "zbkx",    check_implicit_always},
> >> &gt; &gt; +  {"zks", "zksed",   check_implicit_always},
> >> &gt; &gt; +  {"zks", "zksh",    check_implicit_always},
> >> &gt; &gt;    {NULL, NULL, NULL}
> >> &gt; &gt;  };
> >> &gt; &gt;
> >> &gt; &gt; @@ -1127,6 +1141,7 @@ static struct riscv_supported_ext riscv_supported_std_ext[] =
> >> &gt; &gt;    {"c",              ISA_SPEC_CLASS_20190608,        2, 0, 0 },
> >> &gt; &gt;    {"c",              ISA_SPEC_CLASS_2P2,             2, 0, 0 },
> >> &gt; &gt;    {"b",              ISA_SPEC_CLASS_NONE, RISCV_UNKNOWN_VERSION, RISCV_UNKNOWN_VERSION, 0 },
> >> &gt; &gt; +  {"k",              ISA_SPEC_CLASS_NONE, RISCV_UNKNOWN_VERSION, RISCV_UNKNOWN_VERSION, 0 },
> >> &gt;
> >> &gt; May I ask what purpose this addition serves? Without its use enabling
> >> &gt; smaller scope extensions implicitly, I find it at best unhelpful that
> >> &gt; ".option arch, +k" is accepted without having any effect.
> >> &gt;
> >> &gt; Jan
> >>
> >> It's same like bitmanip extension, add this will make k extension as a subextension
> >> and set it's canonical order as ISA spec defined.
> >
> > Let me point out that this is not the same as with the
> > bit-manipulation family of extensions.
> > Zb[abcs] are all standalone extensions, and there is no Zb.
> >
> > --Phil.
>
> I agree.
>
> But we cannot remove these now for a good reason.
>
> This is because riscv_supported_std_ext is used to determine canonical
> order of multi-letter extensions (starting with 'Z').  See
> riscv_init_ext_order function (bfd/elfxx-riscv.c) for details.
>
> Maybe we should make a marker (flag?) for "canonical order is determined
> by this but not an actual extension".  This will apply to...
>
> - "L"
> - "B"
> - "K"
> - "J"
> - "T"
> - "P"
> - "N"
>
> Tsukasa
>
> >
> >>
> >> &gt;
> >> &gt; &gt; @@ -1146,6 +1161,19 @@ static struct riscv_supported_ext riscv_supported_std_z_ext[] =
> >> &gt; &gt;    {"zba",            ISA_SPEC_CLASS_DRAFT,           1, 0,  0 },
> >> &gt; &gt;    {"zbc",            ISA_SPEC_CLASS_DRAFT,           1, 0,  0 },
> >> &gt; &gt;    {"zbs",               ISA_SPEC_CLASS_DRAFT,           1, 0,  0 },
> >> &gt; &gt; +  {"zbkb",           ISA_SPEC_CLASS_DRAFT,           1, 0,  0 },
> >> &gt; &gt; +  {"zbkc",           ISA_SPEC_CLASS_DRAFT,           1, 0,  0 },
> >> &gt; &gt; +  {"zbkx",           ISA_SPEC_CLASS_DRAFT,           1, 0,  0 },
> >> &gt; &gt; +  {"zk",             ISA_SPEC_CLASS_DRAFT,           1, 0,  0 },
> >> &gt; &gt; +  {"zkn",            ISA_SPEC_CLASS_DRAFT,           1, 0,  0 },
> >> &gt; &gt; +  {"zknd",           ISA_SPEC_CLASS_DRAFT,           1, 0,  0 },
> >> &gt; &gt; +  {"zkne",           ISA_SPEC_CLASS_DRAFT,           1, 0,  0 },
> >> &gt; &gt; +  {"zknh",           ISA_SPEC_CLASS_DRAFT,           1, 0,  0 },
> >> &gt; &gt; +  {"zkr",            ISA_SPEC_CLASS_DRAFT,           1, 0,  0 },
> >> &gt; &gt; +  {"zks",            ISA_SPEC_CLASS_DRAFT,           1, 0,  0 },
> >> &gt; &gt; +  {"zksed",  ISA_SPEC_CLASS_DRAFT,           1, 0,  0 },
> >> &gt; &gt; +  {"zksh",           ISA_SPEC_CLASS_DRAFT,           1, 0,  0 },
> >> &gt; &gt; +  {"zkt",            ISA_SPEC_CLASS_DRAFT,           1, 0,  0 },
> >> &gt; &gt;    {NULL, 0, 0, 0, 0}
> >> &gt; &gt;  };
> >> &gt; &gt;
> >> </jiawei@iscas.ac.cn></jbeulich@suse.com>
> >