paweld/fpc
1
0
Fork 0
mirror of https://gitlab.com/freepascal.org/fpc/source.git synced 2025-04-06 15:47:52 +02:00
Code Issues Packages Projects Releases Wiki Activity

* generate jump tables into the same section as the code as otherwise we'll get bogus relocations (in case of clang.exe) or a future support for armasm64.exe will reject the relative symbols outright

Browse Source 
git-svn-id: trunk@44949 -
This commit is contained in:
svenbarth 2020-04-21 06:06:36 +00:00
parent b88107fa1d
commit 7f4b94e408
1 changed files with 29 additions and 16 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

45
compiler/aarch64/ncpuset.pas
View File

@ -42,7 +42,7 @@ implementation
uses
systems,
verbose,globals,constexp,
verbose,globals,constexp,cutils,
symconst,symdef,defutil,
paramgr,
cpuinfo,
@ -254,24 +254,37 @@ implementation
{ and finally jump }
current_asmdata.CurrAsmList.concat(taicpu.op_reg(A_BR,jumpreg));
{ generate jump table }
if not(target_info.system in systems_darwin) then
sectype:=sec_rodata
if target_info.system=system_aarch64_win64 then
begin
{ for Windows we need to make sure that the jump table is located in the
same section as the corresponding code as for one clang generates a
ABSOLUTE32 relocation that can not be handled correctly and armasm64
rejects the difference entries due to the symbols being located in
different sections }
sectype:=sec_code;
new_section(current_procinfo.aktlocaldata,sectype,lower(current_procinfo.procdef.mangledname),getprocalign);
end
else
begin
{ on Mac OS X, dead code stripping ("smart linking") happens based on
global symbols: every global/static symbol (symbols that do not
start with "L") marks the start of a new "subsection" that is
discarded by the linker if there are no references to this symbol.
This means that if you put the jump table in the rodata section, it
will become part of the block of data associated with the previous
non-L-label in the rodata section and stay or be thrown away
depending on whether that block of data is referenced. Therefore,
jump tables must be added in the code section and since aktlocaldata
is inserted right after the routine, it will become part of the
same subsection that contains the routine's code }
sectype:=sec_code;
if not(target_info.system in systems_darwin) then
sectype:=sec_rodata
else
begin
{ on Mac OS X, dead code stripping ("smart linking") happens based on
global symbols: every global/static symbol (symbols that do not
start with "L") marks the start of a new "subsection" that is
discarded by the linker if there are no references to this symbol.
This means that if you put the jump table in the rodata section, it
will become part of the block of data associated with the previous
non-L-label in the rodata section and stay or be thrown away
depending on whether that block of data is referenced. Therefore,
jump tables must be added in the code section and since aktlocaldata
is inserted right after the routine, it will become part of the
same subsection that contains the routine's code }
sectype:=sec_code;
end;
new_section(current_procinfo.aktlocaldata,sectype,current_procinfo.procdef.mangledname,4);
end;
new_section(current_procinfo.aktlocaldata,sectype,current_procinfo.procdef.mangledname,4);
if target_info.system in systems_darwin then
begin
{ additionally, these tables are now marked via ".data_region jt32"