Em Tue, Oct 11, 2022 at 08:33:34AM -0700, Yonghong Song escreveu:
>
>
> On 10/11/22 6:45 AM, Arnaldo Carvalho de Melo wrote:
> > Em Mon, Oct 10, 2022 at 10:57:24PM -0700, Yonghong Song escreveu:
> > > On 10/10/22 1:19 PM, Arnaldo Carvalho de Melo wrote:
> > > > > +++ b/btf_encoder.c
> > > > > @@ -593,6 +593,19 @@ static int32_t btf_encoder__add_func_param(struct btf_encoder *encoder, const ch
> > > > > }
> > > > > }
> > > > > +static int32_t btf_encoder__tag_type(struct btf_encoder *encoder, uint32_t type_id_off, uint32_t tag_type)
> > > > > +{
> > > > > + if (tag_type == 0)
> > > > > + return 0;
> > > > > +
> > > > > + if (encoder->cu->unspecified_type.tag && tag_type == encoder->cu->unspecified_type.type) {
> > >
> > > Why we need tag_type == encoder->cu->unspecified_type.type?
> > > Should we unconditionally convert an unspecified_type to btf type id 0?
> >
> > right, that is what is being done on that line, see,
> > DW_TAG_unspecified_type isn't some constant, what we have as the type in
> > functions in DWARF is the ID for that DW_TAG_unspecified_type for that
> > specific CU, see:
> > <0><c>: Abbrev Number: 1 (DW_TAG_compile_unit)
> > <d> DW_AT_stmt_list : 0
> > <11> DW_AT_low_pc : 0
> > <19> DW_AT_high_pc : 19
> > <1a> DW_AT_name : (indirect string, offset: 0): arch/x86/entry/entry.S
> > <1e> DW_AT_comp_dir : (indirect string, offset: 0x17): /root/git/linux
> > <22> DW_AT_producer : (indirect string, offset: 0x27): GNU AS 2.39.50
> > <26> DW_AT_language : 32769 (MIPS assembler)
> > <1><28>: Abbrev Number: 2 (DW_TAG_subprogram)
> > <29> DW_AT_name : (indirect string, offset: 0x36): entry_ibpb
> > <2d> DW_AT_external : 1
> > <2d> DW_AT_type : <0x37>
> > <2e> DW_AT_low_pc : 0
> > <36> DW_AT_high_pc : 19
> > <1><37>: Abbrev Number: 3 (DW_TAG_unspecified_type)
> > <1><38>: Abbrev Number: 0
> > Look at the info for entry_ibpb, there is no DW_TAG_unspecified_type
> > there, what it has is 0x37 as its type. We have to go see what that type
> > is, it could be DW_TAG_pointer_type with a DW_AT_type pointing to the
> > real type (or a series of references via DW_TAG_const_type,
> > DW_TAG_volatile_type, etc).
> > So in the DWARF loader we record if we have a DW_TAG_unspecified_type in
> > the current CU and what is its "small id", which won't be 0x37, as we
> > use an array for types, to compress the ID space.
> > Then, at the btf_encoding phase we look at functions type, and if it
> > matches the DW_TAG_unspecified_type type for that specific CU, then we
> > convert it to void.
> Thanks for explanation. Now I understand.
> encoder->cu->unspecified_type.tag indicates that at cu level, we have an
> unspecified type.
> encoder->cu->unspecified_type.type records the unspecified_type small_id. I
> assume this should be okay since compiler did dedup
> within the cu so we should have only one unspecified type per cu.
Yes, that is my assumption as well, that a CU will have just one
DW_TAG_unspecified_type, lemme confirm this by finding some other ASM
CU with more than one "function".
> tag_type == encoder->cu->unspecified_type.type ensures the
> current btf type should expect the unspecified type before it
> can be considered as an unspecified type.
> Acked-by: Yonghong Song <yhs@xxxxxx>
Thanks!
- Arnaldo
