[Bug gold/14566] New: wrongly aligned segments when -z max-page-size does not match -z common-page-size

[amodra at gmail dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=14566

             Bug #: 14566
           Summary: wrongly aligned segments when -z max-page-size does
                    not match -z common-page-size
           Product: binutils
           Version: 2.24 (HEAD)
            Status: NEW
          Severity: normal
          Priority: P2
         Component: gold
        AssignedTo: address@hidden
        ReportedBy: address@hidden
                CC: address@hidden
    Classification: Unclassified


Seen on powerpc with current default max-page-size of 64k and common-page-size
of 4k.  Testsuite results contain
============================
39 of 136 tests failed
See testsuite/test-suite.log
============================
These are all segmentation faults on trying to execute the very first
instruction at _start, due to page protection being set to RW rather than RE.

If I change the default common-page-size to 64k, then I get only 6 failures.
"Good" basic_test load segments for powerpc64 are
  LOAD           0x000000 0x0000000010000000 0x0000000010000000 0x001c78
0x001c78 R E 0x10000
  LOAD           0x001c78 0x0000000010011c78 0x0000000010011c78 0x000608
0x0006b0 RW  0x10000
Bad ones are
  LOAD           0x000000 0x0000000010000000 0x0000000010000000 0x001c78
0x001c78 R E 0x10000
  LOAD           0x002000 0x0000000010002000 0x0000000010002000 0x000608
0x0006b0 RW  0x10000

Note how vaddr for the bad second segment starts inside the abi page of the
first segment.  It should be at 0x100012000.

The problem appears to be code in layout.cc after this comment.

      // Now that we know the size of this segment, we may be able
      // to save a page in memory, at the cost of wasting some
      // file space, by instead aligning to the start of a new
      // page.  Here we use the real machine page size rather than
      // the ABI mandated page size.  If the segment has been
      // aligned so that the relro data ends at a page boundary,
      // we do not try to realign it.

-z common-page-size shouldn't ever result in a segment starting at a lower
address than what it would with common-page-size equal to max-page-size.
Normally, a RW segment following a RO segment starts on the next max-page-size
page at the same page offset as the RO segment finished.  That results in
minimum file size.  However it is sometimes advantageous to start the RW
segment at a higher common-page-size aligned address, trading a somewhat larger
file for a smaller number of actual in-memory pages.

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.