R_RISCV_ALIGN

[mamijaz]

When i compile a sample code bellow with fresh build riscv32-unknown-elf-gcc on Ubuntu 16.04

int main(){ int a = 10; int b = 20; (int)0x07FC = a+b;

while(1){}

return 0;

}

i got the results as

add.o: file format elf32-littleriscv Disassembly of section .text: 00000000 : 0: 00 01 00 # R_RISCV_ALIGN 00000003

: 3: fe010113 addi sp,sp,-32 7: 00812e23 sw s0,28(sp) b: 02010413 addi s0,sp,32 f: 00a00793 li a5,10 13: fef42623 sw a5,-20(s0) 17: 01400793 li a5,20 1b: fef42423 sw a5,-24(s0) 1f: 7fc00793 li a5,2044 23: fec42683 lw a3,-20(s0) 27: fe842703 lw a4,-24(s0) 2b: 00e68733 add a4,a3,a4 2f: 00e7a023 sw a4,0(a5) 00000033 <.L2>: 33: 0000006f j 66 <.L2+0x33>

but why there is a "R_RISCV_ALIGN" , is this a new feature, if then how can i remove it, will it not generate a misalignment exception. few weeks ago i was getting something like

add.o: file format elf32-littleriscv Disassembly of section .text: 00000000

: 0: fe010113 addi sp,sp,-32 4: 00812e23 sw s0,28(sp) 8: 02010413 addi s0,sp,32 c: 00a00793 li a5,10 10: fef42623 sw a5,-20(s0) 14: 01400793 li a5,20 18: fef42423 sw a5,-24(s0) 1c: 7fc00793 li a5,2044 20: fec42683 lw a3,-20(s0) 24: fe842703 lw a4,-24(s0) 28: 00e68733 add a4,a3,a4 2c: 00e7a023 sw a4,0(a5) 00000030 <.L2>: 30: 0000006f j 30 <.L2>

[palmer-dabbelt]

FYI: This is eliminated by linker relaxation, there's a blog post about how it works. https://www.sifive.com/blog/2017/08/28/all-aboard-part-3-linker-relaxation-in-riscv-toolchain/

[kito-cheng]

Hi @palmer-dabbelt does it mean align will insert 3 byte by default now? but I remember the conclusion about this issue is it should work even linker does not support relax with code shrink? https://github.com/riscv/riscv-binutils-gdb/pull/88

cc: @pkmx

[palmer-dabbelt]

This is a regression, fixing it is on my TODO list. Hopefully it'll be done by the end of next week.

On Sep 14, 2017 7:01 PM, "Kito Cheng" notifications@github.com wrote:

Hi @palmer-dabbelt https://github.com/palmer-dabbelt does it mean align will insert 3 byte by default now? but I remember the conclusion about this issue is it should work even linker not support relax with code shrink? riscv/riscv-binutils-gdb#88 https://github.com/riscv/riscv-binutils-gdb/pull/88

cc: @PkmX https://github.com/pkmx

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/riscv/riscv-tools/issues/132#issuecomment-329656894, or mute the thread https://github.com/notifications/unsubscribe-auth/ABUaXgvHa4kEnjVPlliubo8Dmc3nVlzRks5sidpsgaJpZM4PXCcI .

[kito-cheng]

Oh, thanks, I just worry about lld can't support that in this moment :)

[palmer-dabbelt]

Yep. It's super awkward to break their stuff right after we told them it was fine and it would work... sorry!

[kito-cheng]

BTW, we've implemented similar method but always use nop16 to stuff that, it will work even linker does not shrinking the code.

[palmer-dabbelt]

That's actually what I did upstream, this just leaked into a binary release. The problem is that approach still emits C instructions on non-C targets.

Can I see your patch?

On Sep 15, 2017 2:28 AM, "Kito Cheng" notifications@github.com wrote:

BTW,we've implemented similar method but always use nop16 to stuff that, it will work even linker does not shrinking the code.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/riscv/riscv-tools/issues/132#issuecomment-329730785, or mute the thread https://github.com/notifications/unsubscribe-auth/ABUaXpXE50dAJk0m2zz7a7PBhGNkfjgAks5sikMogaJpZM4PXCcI .

[kito-cheng]

For non-C target, I thought it should not a problem since all instructions is 32 bits? Hummm, but the problem that you want to resolve is when mixing C and non-C code, I'll ask my colleague next week.

Palmer Dabbelt notifications@github.com於 2017年9月15日 週五，下午10:09寫道：

That's actually what I did upstream, this just leaked into a binary release. The problem is that approach still emits C instructions on non-C targets.

Can I see your patch?

On Sep 15, 2017 2:28 AM, "Kito Cheng" notifications@github.com wrote:

BTW,we've implemented similar method but always use nop16 to stuff that, it will work even linker does not shrinking the code.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <https://github.com/riscv/riscv-tools/issues/132#issuecomment-329730785 , or mute the thread < https://github.com/notifications/unsubscribe-auth/ABUaXpXE50dAJk0m2zz7a7PBhGNkfjgAks5sikMogaJpZM4PXCcI

.

— You are receiving this because you commented.

Reply to this email directly, view it on GitHub https://github.com/riscv/riscv-tools/issues/132#issuecomment-329793401, or mute the thread https://github.com/notifications/unsubscribe-auth/ACmNcSDVFgXkNi0IVgzQtVXcV1r06fQLks5sioT_gaJpZM4PXCcI .