https://sourceware.org/bugzilla/show_bug.cgi?id=34026
--- Comment #8 from Jens Remus <jremus at linux dot ibm.com> ---
Reviewing the code I come to the conclusion that Indu or I need to fix the
following in SFrame generation. Not sure what the correct approach is for the
individual cases though.
1)
.cfi_startproc simple
.cfi_def_cfa_register <register>
=> If SP/FP or flexible FDE, create a FRE and store CFA base register.
Otherwise ignore. Unsure what to do about the missing CFA offset. Initialize
to 0?
2)
.cfi_startproc simple
.cfi_def_cfa_offset <offset>
=> Create a FRE and store CFA offset. Unsure what to do about the missing CFA
base register. Initialize to 0?
3)
.cfi_startproc simple
.cfi_adjust_cfa_offset <offset>
=> Initially I thought to warn with ".cfi_adjust_cfa_offset without prior CFA
offset", but .eh_frame generation does handle this as if the CFA offset was 0
and emits "DW_CFA_def_cfa_offset: <offset>". Therefore this should be treated
likewise. Same as for case 2 I am unsure what to do about the missing CFA base
register. Initialize to 0? Both cases should be handled the same for
consistency.
4)
.cfi_startproc simple
.cfi_offset <register>, <offset>
=> Initially I thought to warn with ".cfi_offset without prior CFA definition",
but .eh_frame generation does emit "DW_CFA_offset: <register> at cfa+<offset>".
Not sure if that is valid .eh_frame, as there is no CFA definition. Given
.sframe generation does not handle .cfi_startproc simple; .cfi_remember_state,
it should be fine to warn?
5)
.cfi_startproc
.cfi_restore_state
=> .sframe generation can either silently return NOTREPRESENTED or warn with
".cfi_restore_state without prior .cfi_remember_state". The CFI directive
processing already errors with "CFI state restore without previous remember".
6)
.cfi_startproc simple
.cfi_restore <register>
=> .eh_frame emits "DW_CFA_restore: <register>. If FP/RA warn with "{FP|RA}
register <register> in .cfi_restore without prior SFrame FRE state"?
7)
.cfi_startproc simple
.cfi_undefined <register>
=> For SP/FP there is already a warning. For RA warn with "RA register
<register> in .cfi_undefined without prior SFrame FRE state"?
8)
.cfi_startproc simple
.cfi_negate_ra_state
=> Warn with ".cfi_negate_ra_state without prior SFrame FRE state"?
9)
.cfi_startproc simple
.cfi_escape with DW_CFA_def_cfa_expression
10)
.cfi_startproc simple
.cfi_escape with DW_CFA_expression
11)
.cfi_startproc simple
.cfi_escape with DW_CFA_GNU_args_size
--
You are receiving this mail because:
You are on the CC list for the bug.
