When build libhugetlbfs tests with gold linker for i386, the below error occurs:
i586-oe-linux-gcc -m32 -march=i586 -Wl,-O1 -Wl,--hash-style=gnu -Wl,--as-needed --sysroot=/home/jenkins/oe/world/shr-core/tmp-glibc/sysroots/qemux86 -I.. -O2 -Wall -g -o obj32/linkhuge_rw.o -c linkhuge_rw.c | i586-oe-linux-gcc -m32 -march=i586 -Wl,-O1 -Wl,--hash-style=gnu -Wl,--as-needed --sysroot=/home/jenkins/oe/world/shr-core/tmp-glibc/sysroots/qemux86 -B./obj32 -Wl,-O1 -Wl,--hash-style=gnu -Wl,--as-needed -Wl,-z,noexecstack -ldl -L../obj32 -o obj32/linkhuge_rw -Wl,--no-as-needed -lpthread -ldl -lhugetlbfs_privutils -Wl,--hugetlbfs-align obj32/linkhuge_rw.o obj32/testutils.o | i586-oe-linux-ld: internal error in do_write, at /home/jenkins/oe/world/shr-core/tmp-glibc/work/x86_64-oe-linux/binutils-cross-i586/2.25-r0/git/gold/output.cc:464 | collect2: error: ld returned 1 exit status
But, it works well with GNU linker. --hugetlbfs-align flag passes "-zcommon-page-size=$SLICE_SIZE -zmax-page-size=$SLICE_SIZE", that are supported by gold linker too. But, it looks gold linker deal with them in a different way from gnu linker for i586.
The readelf shows the below result with GNU linker:
Elf file type is EXEC (Executable file) Entry point 0x8048fbd There are 8 program headers, starting at offset 52
Program Headers: Type Offset VirtAddr PhysAddr FileSiz MemSiz Flg Align PHDR 0x000034 0x08000034 0x08000034 0x00100 0x00100 R E 0x4 INTERP 0x048134 0x08048134 0x08048134 0x00013 0x00013 R 0x1 [Requesting program interpreter: /lib/ld-linux.so.2] LOAD 0x000000 0x08000000 0x08000000 0x5a5bc 0x5a5bc R E 0x400000 LOAD 0x05a5bc 0x0845a5bc 0x0845a5bc 0x1028c 0x202cc RW 0x400000 DYNAMIC 0x05a5d0 0x0845a5d0 0x0845a5d0 0x000e8 0x000e8 RW 0x4 NOTE 0x048148 0x08048148 0x08048148 0x00044 0x00044 R 0x4 GNU_EH_FRAME 0x059e5c 0x08059e5c 0x08059e5c 0x0009c 0x0009c R 0x4 GNU_STACK 0x000000 0x00000000 0x00000000 0x00000 0x00000 RW 0x10
"--relax" linker option doesn't solve this problem. Forced textsegment alignment to 0x08000000 with gold linker, the build will pass and readelf shows the same result with GNU linker:
Elf file type is EXEC (Executable file) Entry point 0x8048fbd There are 8 program headers, starting at offset 52
Program Headers: Type Offset VirtAddr PhysAddr FileSiz MemSiz Flg Align PHDR 0x000034 0x08000034 0x08000034 0x00100 0x00100 R E 0x4 INTERP 0x048134 0x08048134 0x08048134 0x00013 0x00013 R 0x1 [Requesting program interpreter: /lib/ld-linux.so.2] LOAD 0x000000 0x08000000 0x08000000 0x5a5bc 0x5a5bc R E 0x400000 LOAD 0x05a5bc 0x0845a5bc 0x0845a5bc 0x1028c 0x202cc RW 0x400000 DYNAMIC 0x05a5d0 0x0845a5d0 0x0845a5d0 0x000e8 0x000e8 RW 0x4 NOTE 0x048148 0x08048148 0x08048148 0x00044 0x00044 R 0x4 GNU_EH_FRAME 0x059e5c 0x08059e5c 0x08059e5c 0x0009c 0x0009c R 0x4 GNU_STACK 0x000000 0x00000000 0x00000000 0x00000 0x00000 RW 0x10
The fix just have impact on hugelink_rw test case, which needs --hugetlbfs-align flag.
Signed-off-by: Yang Shi yang.shi@linaro.org --- ld.hugetlbfs | 1 + 1 file changed, 1 insertion(+)
diff --git a/ld.hugetlbfs b/ld.hugetlbfs index 4417442..32bc6fb 100755 --- a/ld.hugetlbfs +++ b/ld.hugetlbfs @@ -98,6 +98,7 @@ if [ "$HTLB_ALIGN" == "slice" ]; then # otherwise it will be NULL. case "$EMU" in armelf*_linux_eabi) HTLBOPTS="$HTLBOPTS -Ttext-segment=$SLICE_SIZE" ;; + elf_i386) HTLBOPTS="$HTLBOPTS -Ttext-segment=0x08000000" ;; esac fi