<table border="1" cellspacing="0" cellpadding="8">
    <tr>
        <th>Issue</th>
        <td>
            <a href=https://github.com/llvm/llvm-project/issues/107069>107069</a>
        </td>
    </tr>

    <tr>
        <th>Summary</th>
        <td>
            [LLD][RISCV] Why LLD places global_pointer only by looking at sdata_begin
        </td>
    </tr>

    <tr>
      <th>Labels</th>
      <td>
            lld
      </td>
    </tr>

    <tr>
      <th>Assignees</th>
      <td>
      </td>
    </tr>

    <tr>
      <th>Reporter</th>
      <td>
          LukeZhuang
      </td>
    </tr>
</table>

<pre>
    In RISCV, lld places `global_pointer` by looking for sdata section, and place it at `sdata_begin + 0x800`, otherwise place it at `elfHeader + 0x800` (the following code is from lld/ELF/Writer.cpp in <4befe65c>):
```
    // RISC-V's gp can address +/- 2 KiB, set it to .sdata + 0x800. This symbol
 // should only be defined in an executable. If .sdata does not exist, its
    // value/section does not matter but it has to be relative, so set its
    // st_shndx arbitrarily to 1 (ctx.out.elfHeader).
    if (config->emachine == EM_RISCV) {
      if (!config->shared) {
 OutputSection *sec = findSection(".sdata");
 addOptionalRegular("__global_pointer$", sec ? sec : ctx.out.elfHeader,
 0x800, STV_DEFAULT);
        // Set riscvGlobalPointer to be used by the optional global pointer
        // relaxation.
        if (config->relaxGP) {
          Symbol *s = symtab.find("__global_pointer$");
          if (s && s->isDefined())
 ctx.sym.riscvGlobalPointer = cast<Defined>(s);
        }
 }
    }
```
The issue is currently we have a program on our target machine that does not have .sdata section at all, thus I found it places global pointer at `0x10000 + 0x800`, which is pretty far away from the symbols we want to relax in .sbss section.

So I just wonder is there a reason that lld decides to place global pointer with this simple strategy? I saw GNU LD uses such way to calculate it in the default linker script:
```
__global_pointer$ = MIN(__SDATA_BEGIN__ + 0x800,
 MAX(__DATA_BEGIN__ + 0x800, __BSS_END__ - 0x800));
```
Can LLD have the same behaviour as GNU LD in this case? Thanks!
</pre>
<img width="1px" height="1px" alt="" src="http://email.email.llvm.org/o/eJx8Vl1v4jgU_TXm5apR6gCFBx5oA120TGc0dGZW-xLdxDeJpyZGtsPHv1_ZCaW0nUWoadP7ee65x0ZrZdUQzdjono3SAbau1ma2bl_o37rFphrkWpxmqwa-rzYPPxl_AKUE7BQWZIGN40rpHFW207JxZNg4hvwESusX2VRQagNWoEOwVDipG--PTe8P0gE6HyTYZDlVsgHG7yE-TuKYjWNvrl1N5iAtvXciVf5FKMhcuQDjE1cTlFopffBFFFoQSAul0VtfPOPLxXrJ-PKXkY5MVOx24PMmD8OcShqPCpYsGJ-yZM7ilMVzX0j3DX8CADDuAwRMbn4yfmeh2kGBDaAQhqz1JTG-vAEOf8t734Yl50t3GqIOkdeiI3iupQV72uZa9Sn6-LbWrRKgG3WCnEBQKRsSvlpsgI5UtA5zRRGsynNYoclCox3QUVrnM0tnPxS-R9US48t-LBevLTpHBvI2VFuj9RXnBIYUOrmn0Iruu_kY17rM1o04AppcOoNGqpOPcOvHUrhjpFsXvQ6O8Wl0CSHLYKSbUlY3LFnQFotaNgQsSVmSwuJL1nNwCuzu_uJ4dmX89uJtazQkrm2_tm7Xuk3fM-NzS4WPDqVsxObM0AnjvAOTcR54cPZHIb7uvBGq71S1Ck1nnWXvtoAPg6ufuk-w7J9z-ASChz54R2D-AJvnn1m6WM5_rJ-vsvefHukNOTDSFvvHkPpbl7mfVmtJ-D30i6D7iqGrEc41fhbUT_mI3j66_v_72QTDx2-fjMJ_NoHKAeAArz1tHeaRR_n_AfvY7Tm136gx42OwPr20abcKIdzUfzsvj689baNPkPGFFGgdSx7Ozn7NJ_ZTkO_S8ya-_vb29TtJeK69wtg26EzRGkONUyc4ENS4J0DYGV0Z3IJuQLcGHJqK_LJ1BHc1ussKBpfoSja94qFSnh2ubi2soNRtI_yK9kp8PdxeIePjbRzH8QdJPdSyqH2pO0POnaBEA3jAUyeRnjSdGFnfwQGbIFth5F56Iptbe66s50n3c6NhBb9b6-CgGy_M0vpoxiNgCK1uulb9CSKokIKCvHS6_q6Dg3Q1uCCMcrtTBNYZdFSd_DqtwOIBHp9-wDr1ZLdg26IG34LTUKAqWoUunBWyCQ0JKrFVDpRsXsiALYzcuT8p_Cf0DPT5snpifJJlm3T-PM_uF4-rpyx7g-7rLn-Z_xMM_2gHWXa_2WSLpzTL4Ob8evqWiu9qesAG1uu0Y0cYEW4JcqpxLz2l0J7xCB17HqIlD9Zzjc2LZfx2IGaJmCZTHNDs9o6PktE4SSaDejYe3yWC59NRnIxzPpzcCZqWfBhPxpOEx7kYyBmP-TCexkk8TSajSZRPeZFPBCZD5PGECzaMaYtSRUrtt5E21SCsw-w2vovH04HCnJQNdwzOwxHM_WXDzLz5Td5Wlg1jJa2zlwBOOhWuJet1ykYpG9136j9K4Vd9Clhccf88rP60vFxC0MGb-8WgNWpWO7ezfvpB9irp6jaPCr1lfOnz94-bndG_qXCML0M3lvFl39B-xv8LAAD__3Oz0ik">