<div dir="ltr"><br><div class="gmail_extra"><br><div class="gmail_quote">On Mon, Oct 24, 2016 at 1:36 PM, Friedman, Eli <span dir="ltr"><<a href="mailto:efriedma@codeaurora.org" target="_blank">efriedma@codeaurora.org</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">
  
    
  
  <div bgcolor="#FFFFFF"><span class="gmail-">
    <div class="gmail-m_-8905229054276185447moz-cite-prefix">On 10/24/2016 1:07 PM, Peter
      Collingbourne via llvm-dev wrote:<br>
    </div>
    <blockquote type="cite">
      <div dir="ltr">
        <div class="gmail_extra">
          <div class="gmail_quote">On Mon, Oct 10, 2016 at 8:12 PM,
            Peter Collingbourne <span dir="ltr"><<a href="mailto:peter@pcc.me.uk" target="_blank">peter@pcc.me.uk</a>></span> wrote:<br>
            <blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">
              <div dir="ltr">
                <div>The specific change I have in mind is to allow
                  !range metadata on GlobalObjects. This would<br>
                </div>
                <div>
                  <div>be similar to existing !range metadata, but it
                    would apply to the "address" of the attached
                    GlobalObject, rather than any value loaded from it.
                    Its presence on a GlobalObject would also imply that
                    the address of the GlobalObject is "fixed" at link
                    time.</div>
                </div>
              </div>
            </blockquote>
            <div> </div>
            <div>Going back to IR-level representation: here is an
              alternative representation based on a suggestion from Eli.</div>
            <div><br>
            </div>
            <div>Introduce a new type of GlobalValue called
              GlobalConstant. GlobalConstant would fit into the
              GlobalValue hierarchy like this:</div>
            <div>
              <ul>
                <li>GlobalValue<br>
                </li>
                <ul>
                  <li>GlobalConstant</li>
                  <li>GlobalPointer</li>
                  <ul>
                    <li>GlobalIndirectSymbol</li>
                    <ul>
                      <li>GlobalAlias</li>
                      <li>GlobalIFunc</li>
                    </ul>
                    <li>GlobalObject</li>
                    <ul>
                      <li>Function</li>
                      <li>GlobalVariable</li>
                    </ul>
                  </ul>
                </ul>
              </ul>
              <div>GlobalValue would no longer be assumed to be of
                pointer type. The getType() overload that takes a
                PointerType, as well as getValueType() would be moved
                down to GlobalPointer. (A nice side benefit of this is
                that it would help flush out cases where we are
                unnecessarily depending on global pointee types.)</div>
              <div><br>
              </div>
              <div>A GlobalConstant can either be a definition or a
                declaration. A definition would look like this:</div>
              <div><br>
              </div>
              <div>@foo = globalconst i32 42</div>
            </div>
          </div>
        </div>
      </div>
    </blockquote>
    <br></span>
    This is equivalent to writing "foo = 42" in assembly?</div></blockquote><div><br></div><div>Yes.</div><div> </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div bgcolor="#FFFFFF"><span class="gmail-"><br>
    <br>
    <blockquote type="cite">
      <div dir="ltr">
        <div class="gmail_extra">
          <div class="gmail_quote">
            <div>
              <div>while a declaration would look like this:</div>
              <div><br>
              </div>
              <div>@foo = external globalconst i32</div>
              <div><br>
              </div>
              <div>GlobalConstant could also hold a linkage and
                visibility. Looking at the other attributes that a
                GlobalValue can hold, many of them do not seem
                appropriate for GlobalConstant and could potentially be
                moved to GlobalPointer.</div>
              <div><br>
              </div>
              <div>Thoughts?</div>
              <br>
            </div>
          </div>
        </div>
      </div>
    </blockquote>
    <br></span>
    How do you plan to use this?  The concept makes sense, but I've
    never actually seen anyone use symbols this way.</div></blockquote><div><br></div><div>I plan to use this as part of the ThinLTO implementation of control flow integrity. See <a href="http://clang.llvm.org/docs/ControlFlowIntegrityDesign.html">http://clang.llvm.org/docs/ControlFlowIntegrityDesign.html</a> for a description of how the design currently works in regular LTO.</div><div><br></div><div>If you look at the asm snippets in that design doc, you will see a number of hardcoded constants -- these constants are calculated at LTO time based on whole program information. I want the intermediate object files to depend on the constants so that their values can be substituted in at link time. In ThinLTO, object files are cached, so if a value changes I want to avoid invalidating the cache entries that depend on that value.</div><div><br></div><div>Thanks,</div></div>-- <br><div class="gmail_signature"><div dir="ltr">-- <div>Peter</div></div></div>
</div></div>