---
- branch: MAIN
  date: Thu May 26 07:24:55 UTC 2016
  files:
  - new: '1.94'
    old: '1.93'
    path: src/sys/arch/amd64/amd64/locore.S
    pathrev: src/sys/arch/amd64/amd64/locore.S@1.94
    type: modified
  - new: '1.125'
    old: '1.124'
    path: src/sys/arch/i386/i386/locore.S
    pathrev: src/sys/arch/i386/i386/locore.S@1.125
    type: modified
  id: 20160526T072455Z.2cb547cb1d633a333c5608da50fa53959e3b1fd1
  log: "There is an issue in the way the fillkpt macro sets up pages on both\namd64
    and i386.\n\nThe fillkpt loop is equivalent to the following:\n\n\tdo {\n\t\t/*
    fill in the slot */\n\t\t/* increment %ebx to the next slot */\n\t\t/* increment
    %eax to the next pa */\n\t} while (%ecx > 0)\n\nThe issue here is that if %ecx
    = 0 (i.e., the chunk we are trying to\nmap is zero-sized), there is still one
    entry created in the page table.\nThe kernel expects the va<->pa translation to
    be linear in low memory.\nIf there is a zero-sized chunk, the dead entry creates
    a +4096 offset in\nthe virtual space, with two consecutive entries that point
    to the same\nphysical address. In other words, the mappings are not linear anymore,\nwhich
    causes the kernel to die.\n\nBefore my recent changes, there were only two big
    chunks that were\nmapped, and neither of these could be zero-sized. Now, with
    multiple,\nfine-grained chunks, it is possible that the [SYMS]+[PRELOADED_MODULES]\nchunk
    could be zero-sized.\n\n[PRELOADED_MODULES] is almost never here, and [SYMS] is
    always here on\ndefault kernels. Except for floppies, where the bootloader does
    not load\n[SYMS].\n\nShould fix PR 51148.\n"
  module: src
  subject: 'CVS commit: src/sys/arch'
  unixtime: '1464247495'
  user: maxv