PPC64_OPT_LOCALENTRY is incompatible with tail calls
Alan Modra
amodra@gmail.com
Sat Sep 26 09:33:15 GMT 2020
The save of r2 in __glink_PLTresolve is the culprit. Remove it,
unless we know we need it for --plt-localentry. --plt-localentry
should not be used with power10 pc-relative code that makes tail
calls.
The patch also removes use of r2 as a scratch reg in the ELFv2
__glink_PLTresolve. Using r2 isn't a problem, this is just reducing
the number of scratch regs.
bfd/
* elf64-ppc.c (GLINK_PLTRESOLVE_SIZE): Depend on has_plt_localentry0.
(LD_R0_0R11, ADD_R11_R0_R11): Define.
(ppc64_elf_tls_setup): Disable params->plt_localentry0 when power10
code detected.
(ppc64_elf_size_stubs): Update __glink_PLTresolve eh_frame.
(ppc64_elf_build_stubs): Move r2 save to start of __glink_PLTresolve,
and only emit for has_plt_localentry0. Don't use r2 in the stub.
ld/
* testsuite/ld-powerpc/elfv2so.d,
* testsuite/ld-powerpc/notoc2.d,
* testsuite/ld-powerpc/tlsdesc.wf,
* testsuite/ld-powerpc/tlsdesc2.d,
* testsuite/ld-powerpc/tlsdesc2.wf,
* testsuite/ld-powerpc/tlsopt5.d,
* testsuite/ld-powerpc/tlsopt5.wf,
* testsuite/ld-powerpc/tlsopt6.d,
* testsuite/ld-powerpc/tlsopt6.wf: Update __glink_PLTresolve.
diff --git a/bfd/elf64-ppc.c b/bfd/elf64-ppc.c
index 320717cb45..4194802d48 100644
--- a/bfd/elf64-ppc.c
+++ b/bfd/elf64-ppc.c
@@ -211,9 +211,10 @@ static bfd_vma opd_entry_value
#define PLD_R12_PC 0x04100000e5800000ULL
#define PNOP 0x0700000000000000ULL
-/* __glink_PLTresolve stub instructions. We enter with the index in R0. */
+/* __glink_PLTresolve stub instructions. We enter with the index in
+ R0 for ELFv1, and the address of a glink branch in R12 for ELFv2. */
#define GLINK_PLTRESOLVE_SIZE(htab) \
- (8u + (htab->opd_abi ? 11 * 4 : 14 * 4))
+ (8u + (htab->opd_abi ? 11 * 4 : htab->has_plt_localentry0 ? 14 * 4 : 13 * 4))
/* 0: */
/* .quad plt0-1f */
/* __glink: */
@@ -229,11 +230,14 @@ static bfd_vma opd_entry_value
/* mtctr %12 */
/* ld %11,16(%11) */
/* bctr */
-#define MFLR_R0 0x7c0802a6 /* mflr %r0 */
-#define MTLR_R0 0x7c0803a6 /* mtlr %r0 */
-#define SUB_R12_R12_R11 0x7d8b6050 /* subf %r12,%r11,%r12 */
-#define ADDI_R0_R12 0x380c0000 /* addi %r0,%r12,0 */
-#define SRDI_R0_R0_2 0x7800f082 /* rldicl %r0,%r0,62,2 */
+
+#define MFLR_R0 0x7c0802a6 /* mflr %r0 */
+#define MTLR_R0 0x7c0803a6 /* mtlr %r0 */
+#define SUB_R12_R12_R11 0x7d8b6050 /* subf %r12,%r11,%r12 */
+#define ADDI_R0_R12 0x380c0000 /* addi %r0,%r12,0 */
+#define SRDI_R0_R0_2 0x7800f082 /* rldicl %r0,%r0,62,2 */
+#define LD_R0_0R11 0xe80b0000 /* ld %r0,0(%r11) */
+#define ADD_R11_R0_R11 0x7d605a14 /* add %r11,%r0,%r11 */
/* Pad with this. */
#define NOP 0x60000000
@@ -7736,6 +7740,19 @@ ppc64_elf_tls_setup (struct bfd_link_info *info)
--plt-localentry can cause trouble. */
if (htab->params->plt_localentry0 < 0)
htab->params->plt_localentry0 = 0;
+ if (htab->params->plt_localentry0 && htab->has_power10_relocs)
+ {
+ /* The issue is that __glink_PLTresolve saves r2, which is done
+ because glibc ld.so _dl_runtime_resolve restores r2 to support
+ a glibc plt call optimisation where global entry code is
+ skipped on calls that resolve to the same binary. The
+ __glink_PLTresolve save of r2 is incompatible with code
+ making tail calls, because the tail call might go via the
+ resolver and thus overwrite the proper saved r2. */
+ _bfd_error_handler (_("warning: --plt-localentry is incompatible with "
+ "power10 pc-relative code"));
+ htab->params->plt_localentry0 = 0;
+ }
if (htab->params->plt_localentry0
&& elf_link_hash_lookup (&htab->elf, "GLIBC_2.26",
FALSE, FALSE, FALSE) == NULL)
@@ -13874,11 +13891,11 @@ ppc64_elf_size_stubs (struct bfd_link_info *info)
/* Augmentation. */
p += 1;
- *p++ = DW_CFA_advance_loc + 1;
+ *p++ = DW_CFA_advance_loc + (htab->has_plt_localentry0 ? 3 : 2);
*p++ = DW_CFA_register;
*p++ = 65;
*p++ = htab->opd_abi ? 12 : 0;
- *p++ = DW_CFA_advance_loc + (htab->opd_abi ? 5 : 7);
+ *p++ = DW_CFA_advance_loc + (htab->opd_abi ? 4 : 2);
*p++ = DW_CFA_restore_extended;
*p++ = 65;
p += ((24 + align - 1) & -align) - 24;
@@ -14474,23 +14491,60 @@ ppc64_elf_build_stubs (struct bfd_link_info *info,
}
else
{
+ unsigned int insn;
+
+ /* 0:
+ . .quad plt0-1f # plt0 entry relative to 1:
+ #
+ # We get here with r12 initially @ a glink branch
+ # Load the address of _dl_runtime_resolve from plt0 and
+ # jump to it, with r0 set to the index of the PLT entry
+ # to be resolved and r11 the link map.
+ __glink_PLTresolve:
+ . std %r2,24(%r1) # optional
+ . mflr %r0
+ . bcl 20,31,1f
+ 1:
+ . mflr %r11
+ . mtlr %r0
+ . ld %r0,(0b-1b)(%r11)
+ . sub %r12,%r12,%r11
+ . add %r11,%r0,%r11
+ . addi %r0,%r12,1b-2f
+ . ld %r12,0(%r11)
+ . srdi %r0,%r0,2
+ . mtctr %r12
+ . ld %r11,8(%r11)
+ . bctr
+ 2:
+ . b __glink_PLTresolve
+ . ...
+ . b __glink_PLTresolve */
+
+ if (htab->has_plt_localentry0)
+ {
+ bfd_put_32 (htab->glink->owner, STD_R2_0R1 + 24, p);
+ p += 4;
+ }
bfd_put_32 (htab->glink->owner, MFLR_R0, p);
p += 4;
bfd_put_32 (htab->glink->owner, BCL_20_31, p);
p += 4;
bfd_put_32 (htab->glink->owner, MFLR_R11, p);
p += 4;
- bfd_put_32 (htab->glink->owner, STD_R2_0R1 + 24, p);
- p += 4;
- bfd_put_32 (htab->glink->owner, LD_R2_0R11 | (-16 & 0xfffc), p);
- p += 4;
bfd_put_32 (htab->glink->owner, MTLR_R0, p);
p += 4;
+ if (htab->has_plt_localentry0)
+ insn = LD_R0_0R11 | (-20 & 0xfffc);
+ else
+ insn = LD_R0_0R11 | (-16 & 0xfffc);
+ bfd_put_32 (htab->glink->owner, insn, p);
+ p += 4;
bfd_put_32 (htab->glink->owner, SUB_R12_R12_R11, p);
p += 4;
- bfd_put_32 (htab->glink->owner, ADD_R11_R2_R11, p);
+ bfd_put_32 (htab->glink->owner, ADD_R11_R0_R11, p);
p += 4;
- bfd_put_32 (htab->glink->owner, ADDI_R0_R12 | (-48 & 0xffff), p);
+ bfd_put_32 (htab->glink->owner, ADDI_R0_R12 | (-44 & 0xffff), p);
p += 4;
bfd_put_32 (htab->glink->owner, LD_R12_0R11, p);
p += 4;
diff --git a/ld/testsuite/ld-powerpc/elfv2so.d b/ld/testsuite/ld-powerpc/elfv2so.d
index 0162bd0880..4018f0536c 100644
--- a/ld/testsuite/ld-powerpc/elfv2so.d
+++ b/ld/testsuite/ld-powerpc/elfv2so.d
@@ -74,12 +74,11 @@ Disassembly of section \.text:
.*: (7c 08 02 a6|a6 02 08 7c) mflr r0
.*: (42 9f 00 05|05 00 9f 42) bcl .*
.*: (7d 68 02 a6|a6 02 68 7d) mflr r11
-.*: (18 00 41 f8|f8 41 00 18) std r2,24\(r1\)
-.*: (e8 4b ff f0|f0 ff 4b e8) ld r2,-16\(r11\)
.*: (7c 08 03 a6|a6 03 08 7c) mtlr r0
+.*: (e8 0b ff f0|f0 ff 0b e8) ld r0,-16\(r11\)
.*: (7d 8b 60 50|50 60 8b 7d) subf r12,r11,r12
-.*: (7d 62 5a 14|14 5a 62 7d) add r11,r2,r11
-.*: (38 0c ff d0|d0 ff 0c 38) addi r0,r12,-48
+.*: (7d 60 5a 14|14 5a 60 7d) add r11,r0,r11
+.*: (38 0c ff d4|d4 ff 0c 38) addi r0,r12,-44
.*: (e9 8b 00 00|00 00 8b e9) ld r12,0\(r11\)
.*: (78 00 f0 82|82 f0 00 78) rldicl r0,r0,62,2
.*: (7d 89 03 a6|a6 03 89 7d) mtctr r12
@@ -87,16 +86,16 @@ Disassembly of section \.text:
.*: (4e 80 04 20|20 04 80 4e) bctr
.* <f5@plt>:
-.*: (4b ff ff c8|c8 ff ff 4b) b .* <__glink_PLTresolve>
+.*: (4b ff ff cc|cc ff ff 4b) b .* <__glink_PLTresolve>
.* <f3@plt>:
-.*: (4b ff ff c4|c4 ff ff 4b) b .* <__glink_PLTresolve>
+.*: (4b ff ff c8|c8 ff ff 4b) b .* <__glink_PLTresolve>
.* <f2@plt>:
-.*: (4b ff ff c0|c0 ff ff 4b) b .* <__glink_PLTresolve>
+.*: (4b ff ff c4|c4 ff ff 4b) b .* <__glink_PLTresolve>
.* <f4@plt>:
-.*: (4b ff ff bc|bc ff ff 4b) b .* <__glink_PLTresolve>
+.*: (4b ff ff c0|c0 ff ff 4b) b .* <__glink_PLTresolve>
.* <f1@plt>:
-.*: (4b ff ff b8|b8 ff ff 4b) b .* <__glink_PLTresolve>
+.*: (4b ff ff bc|bc ff ff 4b) b .* <__glink_PLTresolve>
diff --git a/ld/testsuite/ld-powerpc/notoc2.d b/ld/testsuite/ld-powerpc/notoc2.d
index 1e519c0d1b..3448f8b37e 100644
--- a/ld/testsuite/ld-powerpc/notoc2.d
+++ b/ld/testsuite/ld-powerpc/notoc2.d
@@ -22,8 +22,8 @@ Disassembly of section \.text:
.*: (39 80 ff ff|ff ff 80 39)
.*: (06 10 00 00|00 00 10 06) pla r12,0
.*: (39 80 00 00|00 00 80 39)
-.*: (06 10 00 00|00 00 10 06) pla r3,92
-.*: (38 60 00 5c|5c 00 60 38)
+.*: (06 10 00 00|00 00 10 06) pla r3,88
+.*: (38 60 00 58|58 00 60 38)
.*: (4b ff ff 99|99 ff ff 4b) bl .* <.*\.plt_call\.puts>
.*: (60 00 00 00|00 00 00 60) nop
#pass
diff --git a/ld/testsuite/ld-powerpc/tlsdesc.wf b/ld/testsuite/ld-powerpc/tlsdesc.wf
index 09503fa005..e7d4522b50 100644
--- a/ld/testsuite/ld-powerpc/tlsdesc.wf
+++ b/ld/testsuite/ld-powerpc/tlsdesc.wf
@@ -38,9 +38,9 @@ Contents of the \.eh_frame section:
DW_CFA_nop
0+4c 0+14 0+50 FDE cie=0+ pc=0+2f8\.\.0+32c
- DW_CFA_advance_loc: 4 to 0+2fc
+ DW_CFA_advance_loc: 8 to 0+300
DW_CFA_register: r65 in r12
- DW_CFA_advance_loc: 20 to 0+310
+ DW_CFA_advance_loc: 16 to 0+310
DW_CFA_restore_extended: r65
0+64 0+10 0+68 FDE cie=0+ pc=0+2e0\.\.0+2ec
diff --git a/ld/testsuite/ld-powerpc/tlsdesc2.d b/ld/testsuite/ld-powerpc/tlsdesc2.d
index 47aedbecb8..c271c949b0 100644
--- a/ld/testsuite/ld-powerpc/tlsdesc2.d
+++ b/ld/testsuite/ld-powerpc/tlsdesc2.d
@@ -53,12 +53,11 @@ Disassembly of section \.text:
.*: (7c 08 02 a6|a6 02 08 7c) mflr r0
.*: (42 9f 00 05|05 00 9f 42) bcl .*
.*: (7d 68 02 a6|a6 02 68 7d) mflr r11
-.*: (f8 41 00 18|18 00 41 f8) std r2,24\(r1\)
-.*: (e8 4b ff f0|f0 ff 4b e8) ld r2,-16\(r11\)
.*: (7c 08 03 a6|a6 03 08 7c) mtlr r0
+.*: (e8 0b ff f0|f0 ff 0b e8) ld r0,-16\(r11\)
.*: (7d 8b 60 50|50 60 8b 7d) subf r12,r11,r12
-.*: (7d 62 5a 14|14 5a 62 7d) add r11,r2,r11
-.*: (38 0c ff d0|d0 ff 0c 38) addi r0,r12,-48
+.*: (7d 60 5a 14|14 5a 60 7d) add r11,r0,r11
+.*: (38 0c ff d4|d4 ff 0c 38) addi r0,r12,-44
.*: (e9 8b 00 00|00 00 8b e9) ld r12,0\(r11\)
.*: (78 00 f0 82|82 f0 00 78) rldicl r0,r0,62,2
.*: (7d 89 03 a6|a6 03 89 7d) mtctr r12
@@ -66,4 +65,4 @@ Disassembly of section \.text:
.*: (4e 80 04 20|20 04 80 4e) bctr
.* <__tls_get_addr_opt@plt>:
-.*: (4b ff ff c8|c8 ff ff 4b) b .* <__glink_PLTresolve>
+.*: (4b ff ff cc|cc ff ff 4b) b .* <__glink_PLTresolve>
diff --git a/ld/testsuite/ld-powerpc/tlsdesc2.wf b/ld/testsuite/ld-powerpc/tlsdesc2.wf
index cb92c294b1..79a417ba06 100644
--- a/ld/testsuite/ld-powerpc/tlsdesc2.wf
+++ b/ld/testsuite/ld-powerpc/tlsdesc2.wf
@@ -37,10 +37,10 @@ Contents of the \.eh_frame section:
DW_CFA_nop
DW_CFA_nop
-0+4c 0+14 0+50 FDE cie=0+ pc=0+318\.\.0+354
- DW_CFA_advance_loc: 4 to 0+31c
+0+4c 0+14 0+50 FDE cie=0+ pc=0+318\.\.0+350
+ DW_CFA_advance_loc: 8 to 0+320
DW_CFA_register: r65 in r0
- DW_CFA_advance_loc: 28 to 0+338
+ DW_CFA_advance_loc: 8 to 0+328
DW_CFA_restore_extended: r65
0+64 0+10 0+68 FDE cie=0+ pc=0+300\.\.0+30c
diff --git a/ld/testsuite/ld-powerpc/tlsopt5.d b/ld/testsuite/ld-powerpc/tlsopt5.d
index 0fcb79821b..efd6debc55 100644
--- a/ld/testsuite/ld-powerpc/tlsopt5.d
+++ b/ld/testsuite/ld-powerpc/tlsopt5.d
@@ -49,12 +49,11 @@ Disassembly of section \.text:
.*: (a6 02 08 7c|7c 08 02 a6) mflr r0
.*: (05 00 9f 42|42 9f 00 05) bcl .*
.*: (a6 02 68 7d|7d 68 02 a6) mflr r11
-.*: (18 00 41 f8|f8 41 00 18) std r2,24\(r1\)
-.*: (f0 ff 4b e8|e8 4b ff f0) ld r2,-16\(r11\)
.*: (a6 03 08 7c|7c 08 03 a6) mtlr r0
+.*: (f0 ff 0b e8|e8 0b ff f0) ld r0,-16\(r11\)
.*: (50 60 8b 7d|7d 8b 60 50) subf r12,r11,r12
-.*: (14 5a 62 7d|7d 62 5a 14) add r11,r2,r11
-.*: (d0 ff 0c 38|38 0c ff d0) addi r0,r12,-48
+.*: (14 5a 60 7d|7d 60 5a 14) add r11,r0,r11
+.*: (d4 ff 0c 38|38 0c ff d4) addi r0,r12,-44
.*: (00 00 8b e9|e9 8b 00 00) ld r12,0\(r11\)
.*: (82 f0 00 78|78 00 f0 82) rldicl r0,r0,62,2
.*: (a6 03 89 7d|7d 89 03 a6) mtctr r12
@@ -62,7 +61,7 @@ Disassembly of section \.text:
.*: (20 04 80 4e|4e 80 04 20) bctr
.* <__tls_get_addr_opt@plt>:
-.* (c8 ff ff 4b|4b ff ff c8) b .*
+.* (cc ff ff 4b|4b ff ff cc) b .*
.* <aaaaa@plt>:
-.*: (c4 ff ff 4b|4b ff ff c4) b .*
+.*: (c8 ff ff 4b|4b ff ff c8) b .*
diff --git a/ld/testsuite/ld-powerpc/tlsopt5.wf b/ld/testsuite/ld-powerpc/tlsopt5.wf
index f0453610e0..84bd94ed72 100644
--- a/ld/testsuite/ld-powerpc/tlsopt5.wf
+++ b/ld/testsuite/ld-powerpc/tlsopt5.wf
@@ -16,9 +16,9 @@ Contents of the \.eh_frame section:
DW_CFA_restore_extended: r65
0+2c 0+14 0+30 FDE cie=0+ pc=.*
- DW_CFA_advance_loc: 4 to .*
+ DW_CFA_advance_loc: 8 to .*
DW_CFA_register: r65 in r0
- DW_CFA_advance_loc: 28 to .*
+ DW_CFA_advance_loc: 8 to .*
DW_CFA_restore_extended: r65
0+44 0+10 0+48 FDE cie=0+ pc=.*
diff --git a/ld/testsuite/ld-powerpc/tlsopt6.d b/ld/testsuite/ld-powerpc/tlsopt6.d
index 4ca64092c1..15def719cb 100644
--- a/ld/testsuite/ld-powerpc/tlsopt6.d
+++ b/ld/testsuite/ld-powerpc/tlsopt6.d
@@ -67,12 +67,11 @@ Disassembly of section \.text:
.*: (a6 02 08 7c|7c 08 02 a6) mflr r0
.*: (05 00 9f 42|42 9f 00 05) bcl .*
.*: (a6 02 68 7d|7d 68 02 a6) mflr r11
-.*: (18 00 41 f8|f8 41 00 18) std r2,24\(r1\)
-.*: (f0 ff 4b e8|e8 4b ff f0) ld r2,-16\(r11\)
.*: (a6 03 08 7c|7c 08 03 a6) mtlr r0
+.*: (f0 ff 0b e8|e8 0b ff f0) ld r0,-16\(r11\)
.*: (50 60 8b 7d|7d 8b 60 50) subf r12,r11,r12
-.*: (14 5a 62 7d|7d 62 5a 14) add r11,r2,r11
-.*: (d0 ff 0c 38|38 0c ff d0) addi r0,r12,-48
+.*: (14 5a 60 7d|7d 60 5a 14) add r11,r0,r11
+.*: (d4 ff 0c 38|38 0c ff d4) addi r0,r12,-44
.*: (00 00 8b e9|e9 8b 00 00) ld r12,0\(r11\)
.*: (82 f0 00 78|78 00 f0 82) rldicl r0,r0,62,2
.*: (a6 03 89 7d|7d 89 03 a6) mtctr r12
@@ -80,7 +79,7 @@ Disassembly of section \.text:
.*: (20 04 80 4e|4e 80 04 20) bctr
.* <__tls_get_addr_opt@plt>:
-.* (c8 ff ff 4b|4b ff ff c8) b .*
+.* (cc ff ff 4b|4b ff ff cc) b .*
.* <aaaaa@plt>:
-.*: (c4 ff ff 4b|4b ff ff c4) b .*
+.*: (c8 ff ff 4b|4b ff ff c8) b .*
diff --git a/ld/testsuite/ld-powerpc/tlsopt6.wf b/ld/testsuite/ld-powerpc/tlsopt6.wf
index abb414a633..c2b9616352 100644
--- a/ld/testsuite/ld-powerpc/tlsopt6.wf
+++ b/ld/testsuite/ld-powerpc/tlsopt6.wf
@@ -38,9 +38,9 @@ Contents of the \.eh_frame section:
DW_CFA_nop
0+4c 0+14 0+50 FDE cie=0+ pc=.*
- DW_CFA_advance_loc: 4 to .*
+ DW_CFA_advance_loc: 8 to .*
DW_CFA_register: r65 in r0
- DW_CFA_advance_loc: 28 to .*
+ DW_CFA_advance_loc: 8 to .*
DW_CFA_restore_extended: r65
0+64 0+10 0+68 FDE cie=0+ pc=.*
--
Alan Modra
Australia Development Lab, IBM
More information about the Binutils
mailing list