Message ID | 1395330368-19311-1-git-send-email-will.newton@linaro.org |
---|---|
State | Accepted |
Headers | show |
On 20/03/14 15:46, Will Newton wrote: > The current code subtracts the size of the output section containing > relplt from RELASZ. In some cases this will be the same output > section Hi Will, I guess you can only trigger this "same output section" issue when using "ld-aarch64/aarch64.ld", right? -- Jiong > as the dynamic relocs causing a value of zero to be output. > Calculating the size from input sections seems to make more sense. > > bfd/ChangeLog: > > 2014-03-20 Will Newton <will.newton@linaro.org> > > * elfnn-aarch64.c (elfNN_aarch64_finish_dynamic_sections): > Set value of DT_PLTRELSZ and DT_RELASZ based on the size > of input sections rather than output sections. > > ld/testsuite/ChangeLog: > > 2014-03-20 Will Newton <will.newton@linaro.org> > > * ld-aarch64/aarch64-elf.exp: Add relasz dump test. > * ld-aarch64/relasz.d: New file. > * ld-aarch64/relasz.s: Likewise. > --- > bfd/elfnn-aarch64.c | 4 ++-- > ld/testsuite/ld-aarch64/aarch64-elf.exp | 2 ++ > ld/testsuite/ld-aarch64/relasz.d | 18 ++++++++++++++++++ > ld/testsuite/ld-aarch64/relasz.s | 9 +++++++++ > 4 files changed, 31 insertions(+), 2 deletions(-) > create mode 100644 ld/testsuite/ld-aarch64/relasz.d > create mode 100644 ld/testsuite/ld-aarch64/relasz.s > > Changes in v2: > - Add testcase > > diff --git a/bfd/elfnn-aarch64.c b/bfd/elfnn-aarch64.c > index c2adcc9..638938d 100644 > --- a/bfd/elfnn-aarch64.c > +++ b/bfd/elfnn-aarch64.c > @@ -6993,7 +6993,7 @@ elfNN_aarch64_finish_dynamic_sections (bfd *output_bfd, > break; > > case DT_PLTRELSZ: > - s = htab->root.srelplt->output_section; > + s = htab->root.srelplt; > dyn.d_un.d_val = s->size; > break; > > @@ -7007,7 +7007,7 @@ elfNN_aarch64_finish_dynamic_sections (bfd *output_bfd, > about changing the DT_RELA entry. */ > if (htab->root.srelplt != NULL) > { > - s = htab->root.srelplt->output_section; > + s = htab->root.srelplt; > dyn.d_un.d_val -= s->size; > } > break; > diff --git a/ld/testsuite/ld-aarch64/aarch64-elf.exp b/ld/testsuite/ld-aarch64/aarch64-elf.exp > index 32cf73a..845ea20 100644 > --- a/ld/testsuite/ld-aarch64/aarch64-elf.exp > +++ b/ld/testsuite/ld-aarch64/aarch64-elf.exp > @@ -157,3 +157,5 @@ run_dump_test "ifunc-19b" > run_dump_test "ifunc-20" > run_dump_test "ifunc-21" > run_dump_test "ifunc-22" > + > +run_dump_test "relasz" > diff --git a/ld/testsuite/ld-aarch64/relasz.d b/ld/testsuite/ld-aarch64/relasz.d > new file mode 100644 > index 0000000..5cc5595 > --- /dev/null > +++ b/ld/testsuite/ld-aarch64/relasz.d > @@ -0,0 +1,18 @@ > +#source: relasz.s > +#ld: -shared -Taarch64.ld > +#readelf: -d > +# Check that the RELASZ section has the correct size even if we are > +# using a non-default linker script that merges .rela.dyn and .rela.plt > +# in the output. > + > +Dynamic section at offset 0x[0-9a-f]+ contains 9 entries: > + Tag Type Name/Value > + 0x0000000000000004 \(HASH\) 0x[0-9a-f]+ > + 0x0000000000000005 \(STRTAB\) 0x[0-9a-f]+ > + 0x0000000000000006 \(SYMTAB\) 0x[0-9a-f]+ > + 0x000000000000000a \(STRSZ\) [0-9]+ \(bytes\) > + 0x000000000000000b \(SYMENT\) [0-9]+ \(bytes\) > + 0x0000000000000007 \(RELA\) 0x[0-9a-f]+ > + 0x0000000000000008 \(RELASZ\) 24 \(bytes\) > + 0x0000000000000009 \(RELAENT\) 24 \(bytes\) > + 0x0000000000000000 \(NULL\) 0x0 > diff --git a/ld/testsuite/ld-aarch64/relasz.s b/ld/testsuite/ld-aarch64/relasz.s > new file mode 100644 > index 0000000..8c7f891 > --- /dev/null > +++ b/ld/testsuite/ld-aarch64/relasz.s > @@ -0,0 +1,9 @@ > + .text > + .global func > + .type func, %function > +func: > + adrp x0, :got:foo > + ldr x0, [x0, #:got_lo12:foo] > + ldr w0, [x0] > + ret > + .size func, .-func
On 20 March 2014 17:57, Jiong Wang <jiong.wang@arm.com> wrote: > On 20/03/14 15:46, Will Newton wrote: >> >> The current code subtracts the size of the output section containing >> relplt from RELASZ. In some cases this will be the same output >> section > > > Hi Will, > > I guess you can only trigger this "same output section" issue when using > "ld-aarch64/aarch64.ld", right? Yes, it requires a custom linker script. The problem arose building OSv for aarch64: https://github.com/cloudius-systems/osv You could say the linker script is incorrect not to lay out the dynamic reloc sections explicitly but it seems counter intuitive behaviour to set RELASZ to zero when it clearly isn't. A number of ports base the size on the size of the output sections: lm32 m32r nds32 sh-32 xtensa s390-64 x86_64 But the rest use the size of input sections. > > -- > Jiong > > >> as the dynamic relocs causing a value of zero to be output. >> Calculating the size from input sections seems to make more sense. >> >> bfd/ChangeLog: >> >> 2014-03-20 Will Newton <will.newton@linaro.org> >> >> * elfnn-aarch64.c (elfNN_aarch64_finish_dynamic_sections): >> Set value of DT_PLTRELSZ and DT_RELASZ based on the size >> of input sections rather than output sections. >> >> ld/testsuite/ChangeLog: >> >> 2014-03-20 Will Newton <will.newton@linaro.org> >> >> * ld-aarch64/aarch64-elf.exp: Add relasz dump test. >> * ld-aarch64/relasz.d: New file. >> * ld-aarch64/relasz.s: Likewise. >> --- >> bfd/elfnn-aarch64.c | 4 ++-- >> ld/testsuite/ld-aarch64/aarch64-elf.exp | 2 ++ >> ld/testsuite/ld-aarch64/relasz.d | 18 ++++++++++++++++++ >> ld/testsuite/ld-aarch64/relasz.s | 9 +++++++++ >> 4 files changed, 31 insertions(+), 2 deletions(-) >> create mode 100644 ld/testsuite/ld-aarch64/relasz.d >> create mode 100644 ld/testsuite/ld-aarch64/relasz.s >> >> Changes in v2: >> - Add testcase >> >> diff --git a/bfd/elfnn-aarch64.c b/bfd/elfnn-aarch64.c >> index c2adcc9..638938d 100644 >> --- a/bfd/elfnn-aarch64.c >> +++ b/bfd/elfnn-aarch64.c >> @@ -6993,7 +6993,7 @@ elfNN_aarch64_finish_dynamic_sections (bfd >> *output_bfd, >> break; >> case DT_PLTRELSZ: >> - s = htab->root.srelplt->output_section; >> + s = htab->root.srelplt; >> dyn.d_un.d_val = s->size; >> break; >> @@ -7007,7 +7007,7 @@ elfNN_aarch64_finish_dynamic_sections (bfd >> *output_bfd, >> about changing the DT_RELA entry. */ >> if (htab->root.srelplt != NULL) >> { >> - s = htab->root.srelplt->output_section; >> + s = htab->root.srelplt; >> dyn.d_un.d_val -= s->size; >> } >> break; >> diff --git a/ld/testsuite/ld-aarch64/aarch64-elf.exp >> b/ld/testsuite/ld-aarch64/aarch64-elf.exp >> index 32cf73a..845ea20 100644 >> --- a/ld/testsuite/ld-aarch64/aarch64-elf.exp >> +++ b/ld/testsuite/ld-aarch64/aarch64-elf.exp >> @@ -157,3 +157,5 @@ run_dump_test "ifunc-19b" >> run_dump_test "ifunc-20" >> run_dump_test "ifunc-21" >> run_dump_test "ifunc-22" >> + >> +run_dump_test "relasz" >> diff --git a/ld/testsuite/ld-aarch64/relasz.d >> b/ld/testsuite/ld-aarch64/relasz.d >> new file mode 100644 >> index 0000000..5cc5595 >> --- /dev/null >> +++ b/ld/testsuite/ld-aarch64/relasz.d >> @@ -0,0 +1,18 @@ >> +#source: relasz.s >> +#ld: -shared -Taarch64.ld >> +#readelf: -d >> +# Check that the RELASZ section has the correct size even if we are >> +# using a non-default linker script that merges .rela.dyn and .rela.plt >> +# in the output. >> + >> +Dynamic section at offset 0x[0-9a-f]+ contains 9 entries: >> + Tag Type Name/Value >> + 0x0000000000000004 \(HASH\) 0x[0-9a-f]+ >> + 0x0000000000000005 \(STRTAB\) 0x[0-9a-f]+ >> + 0x0000000000000006 \(SYMTAB\) 0x[0-9a-f]+ >> + 0x000000000000000a \(STRSZ\) [0-9]+ \(bytes\) >> + 0x000000000000000b \(SYMENT\) [0-9]+ \(bytes\) >> + 0x0000000000000007 \(RELA\) 0x[0-9a-f]+ >> + 0x0000000000000008 \(RELASZ\) 24 \(bytes\) >> + 0x0000000000000009 \(RELAENT\) 24 \(bytes\) >> + 0x0000000000000000 \(NULL\) 0x0 >> diff --git a/ld/testsuite/ld-aarch64/relasz.s >> b/ld/testsuite/ld-aarch64/relasz.s >> new file mode 100644 >> index 0000000..8c7f891 >> --- /dev/null >> +++ b/ld/testsuite/ld-aarch64/relasz.s >> @@ -0,0 +1,9 @@ >> + .text >> + .global func >> + .type func, %function >> +func: >> + adrp x0, :got:foo >> + ldr x0, [x0, #:got_lo12:foo] >> + ldr w0, [x0] >> + ret >> + .size func, .-func > > > -- > Jiong > >
On 24 March 2014 08:49, Will Newton <will.newton@linaro.org> wrote: > On 20 March 2014 17:57, Jiong Wang <jiong.wang@arm.com> wrote: >> On 20/03/14 15:46, Will Newton wrote: >>> >>> The current code subtracts the size of the output section containing >>> relplt from RELASZ. In some cases this will be the same output >>> section >> >> >> Hi Will, >> >> I guess you can only trigger this "same output section" issue when using >> "ld-aarch64/aarch64.ld", right? > > Yes, it requires a custom linker script. The problem arose building > OSv for aarch64: > > https://github.com/cloudius-systems/osv > > You could say the linker script is incorrect not to lay out the > dynamic reloc sections explicitly but it seems counter intuitive > behaviour to set RELASZ to zero when it clearly isn't. > > A number of ports base the size on the size of the output sections: > > lm32 > m32r > nds32 > sh-32 > xtensa > s390-64 > x86_64 which implies these ports are all broken in the same way ? or do they address this issue in a different way? Cheers /Marcus
On Mon, Mar 24, 2014 at 12:26:28PM +0000, Marcus Shawcroft wrote: > On 24 March 2014 08:49, Will Newton <will.newton@linaro.org> wrote: [Re DT_PLTRELSZ and DT_RELASZ] > > A number of ports base the size on the size of the output sections: > > > > lm32 > > m32r > > nds32 > > sh-32 > > xtensa > > s390-64 > > x86_64 > > which implies these ports are all broken in the same way ? Yes, I think so. The patch looks good to me.
On 20 March 2014 15:46, Will Newton <will.newton@linaro.org> wrote: > The current code subtracts the size of the output section containing > relplt from RELASZ. In some cases this will be the same output > section as the dynamic relocs causing a value of zero to be output. > Calculating the size from input sections seems to make more sense. > > bfd/ChangeLog: > > 2014-03-20 Will Newton <will.newton@linaro.org> > > * elfnn-aarch64.c (elfNN_aarch64_finish_dynamic_sections): > Set value of DT_PLTRELSZ and DT_RELASZ based on the size > of input sections rather than output sections. OK, go ahead, thanks Will. /Marcus
diff --git a/bfd/elfnn-aarch64.c b/bfd/elfnn-aarch64.c index c2adcc9..638938d 100644 --- a/bfd/elfnn-aarch64.c +++ b/bfd/elfnn-aarch64.c @@ -6993,7 +6993,7 @@ elfNN_aarch64_finish_dynamic_sections (bfd *output_bfd, break; case DT_PLTRELSZ: - s = htab->root.srelplt->output_section; + s = htab->root.srelplt; dyn.d_un.d_val = s->size; break; @@ -7007,7 +7007,7 @@ elfNN_aarch64_finish_dynamic_sections (bfd *output_bfd, about changing the DT_RELA entry. */ if (htab->root.srelplt != NULL) { - s = htab->root.srelplt->output_section; + s = htab->root.srelplt; dyn.d_un.d_val -= s->size; } break; diff --git a/ld/testsuite/ld-aarch64/aarch64-elf.exp b/ld/testsuite/ld-aarch64/aarch64-elf.exp index 32cf73a..845ea20 100644 --- a/ld/testsuite/ld-aarch64/aarch64-elf.exp +++ b/ld/testsuite/ld-aarch64/aarch64-elf.exp @@ -157,3 +157,5 @@ run_dump_test "ifunc-19b" run_dump_test "ifunc-20" run_dump_test "ifunc-21" run_dump_test "ifunc-22" + +run_dump_test "relasz" diff --git a/ld/testsuite/ld-aarch64/relasz.d b/ld/testsuite/ld-aarch64/relasz.d new file mode 100644 index 0000000..5cc5595 --- /dev/null +++ b/ld/testsuite/ld-aarch64/relasz.d @@ -0,0 +1,18 @@ +#source: relasz.s +#ld: -shared -Taarch64.ld +#readelf: -d +# Check that the RELASZ section has the correct size even if we are +# using a non-default linker script that merges .rela.dyn and .rela.plt +# in the output. + +Dynamic section at offset 0x[0-9a-f]+ contains 9 entries: + Tag Type Name/Value + 0x0000000000000004 \(HASH\) 0x[0-9a-f]+ + 0x0000000000000005 \(STRTAB\) 0x[0-9a-f]+ + 0x0000000000000006 \(SYMTAB\) 0x[0-9a-f]+ + 0x000000000000000a \(STRSZ\) [0-9]+ \(bytes\) + 0x000000000000000b \(SYMENT\) [0-9]+ \(bytes\) + 0x0000000000000007 \(RELA\) 0x[0-9a-f]+ + 0x0000000000000008 \(RELASZ\) 24 \(bytes\) + 0x0000000000000009 \(RELAENT\) 24 \(bytes\) + 0x0000000000000000 \(NULL\) 0x0 diff --git a/ld/testsuite/ld-aarch64/relasz.s b/ld/testsuite/ld-aarch64/relasz.s new file mode 100644 index 0000000..8c7f891 --- /dev/null +++ b/ld/testsuite/ld-aarch64/relasz.s @@ -0,0 +1,9 @@ + .text + .global func + .type func, %function +func: + adrp x0, :got:foo + ldr x0, [x0, #:got_lo12:foo] + ldr w0, [x0] + ret + .size func, .-func