[PATCH v2 1/3] RISC-V: Minimal support of scalar crypto extension
陈嘉炜
jiawei@iscas.ac.cn
Wed Nov 3 08:33:08 GMT 2021
"Jan Beulich" <jbeulich@suse.com>写道:
> On 03.11.2021 02:54, jiawei wrote:
> > Fix subextension order set by Nelson's suggestion.
>
> Isn't this rather the revision log than the description of the
> change?
Hi Jan, here is the log of discussion:
>>On Tue, Nov 2, 2021 at 8:23 PM 陈嘉炜 <jiawei@iscas.ac.cn> wrote:
>> Hi Nelson,
>> Thank you for the check, if we undo with this patch, here comes another unexpected results same as you mentioned:
>> riscv64-unknown-elf-as -misa-spec=20191213 -march=rv64i_zbb_zknd empty.s -o empty.o
>> Assembler messages:
>> Error: rv64i_zicsr_zifencei_zbb_zknd: prefixed ISA extension `zknd' is not in expected order. It must come before `zbb'
>> I think that zb* should goes before zk*, so I change the order at first patch to make 'zb' before 'zk'. Sorry for that I missed zicsr,
>> zifencei and don't check it.
>> What should the right order if we have a string both contain 'zb', 'zk' even 'zp', 'zv' like -march=rv64i_zicsr_zifencei_zbb_zknd?
>> You can add the entry for k in the riscv_supported_std_ext, for example,
>> @@ -1127,6 +1127,7 @@ static struct riscv_supported_ext
>> riscv_supported_std_ext[] =
{"c", ISA_SPEC_CLASS_20190608, 2, 0, 0 },
{"c", ISA_SPEC_CLASS_2P2, 2, 0, 0 },
{"b", ISA_SPEC_CLASS_NONE, RISCV_UNKNOWN_VERSION,
RISCV_UNKNOWN_VERSION, 0 },
+ {"k", ISA_SPEC_CLASS_NONE, RISCV_UNKNOWN_VERSION,
RISCV_UNKNOWN_VERSION, 0 },
{"j", ISA_SPEC_CLASS_NONE, RISCV_UNKNOWN_VERSION,
RISCV_UNKNOWN_VERSION, 0 },
{"t", ISA_SPEC_CLASS_NONE, RISCV_UNKNOWN_VERSION,
RISCV_UNKNOWN_VERSION, 0 },
{"p", ISA_SPEC_CLASS_NONE, RISCV_UNKNOWN_VERSION,
RISCV_UNKNOWN_VERSION, 0 },
>>
>> The standard extensions must be added in canonical order in the
>> riscv_supported_std_ext table, since I used to use this table to build
>> riscv_ext_order. So you could find a proper place to add the entry
>> for k extension.
>> "Nelson Chu" <nelson.chu@sifive.com>写道:
>> > Hi jiawei,
>> >
>> > On Tue, Nov 2, 2021 at 5:44 PM jiawei <jiawei@iscas.ac.cn> wrote:
>> > >
>> > > ---
>> > > bfd/elfxx-riscv.c | 2 +-
>> > > 1 file changed, 1 insertion(+), 1 deletion(-)
>> > >
>> > > diff --git a/bfd/elfxx-riscv.c b/bfd/elfxx-riscv.c
>> > > index cdb4fa0996a..0d8fc755b5c 100644
>> > > --- a/bfd/elfxx-riscv.c
>> > > +++ b/bfd/elfxx-riscv.c
>> > > @@ -1333,7 +1333,7 @@ riscv_compare_subsets (const char *subset1, const char *subset2)
>> > > order1 = riscv_ext_order[(*++subset1 - 'a')];
>> > > order2 = riscv_ext_order[(*++subset2 - 'a')];
>> > > if (order1 != order2)
>> > > - return order1 - order2;
>> > > + return order2 - order1;
>> > > }
>> > > return strcasecmp (++subset1, ++subset2);
>> > > }
>> > > --
>> > > 2.25.1
>> > >
>> >
>> > After I apply this change, I get the unexpected results as follows,
>> >
>> > nelson@LAPTOP-QFSGI1F2:~/test$
>> > ~/binutils-dev/build-linux64-upstream/build-install/bin/riscv64-unknown-linux-gnu-as
>> > -misa-spec=20191213 -march=rv64i_zicsr_zifencei_zbb empty.s -o empty.o
>> > Assembler messages:
>> > Error: rv64i_zicsr_zifencei_zbb: prefixed ISA extension `zbb' is not
>> > in expected order. It must come before `zifencei'
>> >
>> > BTW, the expected result is as follows, before applying this patch,
>> >
>> > nelson@LAPTOP-QFSGI1F2:~/test$
>> > ~/binutils-dev/build-linux64-upstream/build-install/bin/riscv64-unknown-linux-gnu-as
>> > -misa-spec=20191213 -march=rv64i_zicsr_zifencei_zbb empty.s -o empty.o
>> > nelson@LAPTOP-QFSGI1F2:~/test$
>> > ~/binutils-dev/build-linux64-upstream/build-install/bin/riscv64-unknown-linux-gnu-readelf
>> > -A empty.o Attribute Section: riscv
>> > File Attributes
>> > Tag_RISCV_arch: "rv64i2p1_zicsr2p0_zifencei2p0_zbb1p0"
>> >
>> > Could I ask what issues or cases you want to resolve originally, so
>> > that you need to change this?
>> >
>> > Thanks
>> > Nelson
>> </jiawei@iscas.ac.cn>
> > @@ -1146,6 +1161,19 @@ static struct riscv_supported_ext riscv_supported_std_z_ext[] =
> > {"zba", ISA_SPEC_CLASS_DRAFT, 1, 0, 0 },
> > {"zbc", ISA_SPEC_CLASS_DRAFT, 1, 0, 0 },
> > {"zbs", ISA_SPEC_CLASS_DRAFT, 1, 0, 0 },
> > + {"zbkb", ISA_SPEC_CLASS_DRAFT, 1, 0, 0 },
>
> I seem to vaguely recall that sorting is supposed to be alphabetical,
> which the insertion point here still would violate.
>
> Jan
Since 'zbk*' is sub-extension of Scalar Cryptography extension(k extension), so I added 'zbk*' behind 'zb*', I'm not sure if we need to change it into alphabetical strictly.
</jiawei@iscas.ac.cn>
More information about the Binutils
mailing list