[llvm] r190773 - Implement function prefix data as an IR feature.
rnk at google.com
Mon Sep 23 13:20:07 PDT 2013
There are applications (ubsan) where you don't know if the function will
have the magic prefix, so accessing a negative offset is dicey if the entry
point is on a page boundary.
OTOH reading magic tag bits is almost just as dicey.
Adding this offset can be done as separate feature, see the symbol_offset
On Mon, Sep 23, 2013 at 1:14 PM, Richard Mitton <richard at codersnotes.com>wrote:
> I have a small question about this -
> Why does the prefix data have to begin with a jump?
> Couldn't you just make the entry point label appear after the prefix data?
> And then use negative offsets to access it given the function's address? It
> seems like the jump adds a performance hit which isn't needed.
> .word 0x1234 // prefix data
> push rbp
> // etc
> Richard Mitton
> richard at codersnotes.com
> On 09/23/2013 01:07 PM, Peter Collingbourne wrote:
>> On Mon, Sep 23, 2013 at 10:01:18AM -0700, Chris Lattner wrote:
>>> On Sep 15, 2013, at 6:08 PM, Peter Collingbourne <peter at pcc.me.uk>
>>> Author: pcc
>>>> Date: Sun Sep 15 20:08:15 2013
>>>> New Revision: 190773
>>>> URL: http://llvm.org/viewvc/llvm-**project?rev=190773&view=rev<http://llvm.org/viewvc/llvm-project?rev=190773&view=rev>
>>>> Implement function prefix data as an IR feature.
>>>> Previous discussion:
>>>> Differential Revision: http://llvm-reviews.chandlerc.**com/D1191<http://llvm-reviews.chandlerc.com/D1191>
>>> Cool, thanks Peter.
>>> Can I entice you to write a blog post about this? :-)
>> Please ask me again in a couple of weeks. There are a number of
>> higher priority tasks that I have to take care of this week...
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the llvm-commits