Home | History | Annotate | Download | only in docs
      1 ========================
      2 Segmented Stacks in LLVM
      3 ========================
      4 
      5 .. contents::
      6    :local:
      7 
      8 Introduction
      9 ============
     10 
     11 Segmented stack allows stack space to be allocated incrementally than as a
     12 monolithic chunk (of some worst case size) at thread initialization. This is
     13 done by allocating stack blocks (henceforth called *stacklets*) and linking them
     14 into a doubly linked list. The function prologue is responsible for checking if
     15 the current stacklet has enough space for the function to execute; and if not,
     16 call into the libgcc runtime to allocate more stack space. Segmented stacks are
     17 enabled with the ``"split-stack"`` attribute on LLVM functions.
     18 
     19 The runtime functionality is `already there in libgcc
     20 <http://gcc.gnu.org/wiki/SplitStacks>`_.
     21 
     22 Implementation Details
     23 ======================
     24 
     25 .. _allocating stacklets:
     26 
     27 Allocating Stacklets
     28 --------------------
     29 
     30 As mentioned above, the function prologue checks if the current stacklet has
     31 enough space. The current approach is to use a slot in the TCB to store the
     32 current stack limit (minus the amount of space needed to allocate a new block) -
     33 this slot's offset is again dictated by ``libgcc``. The generated
     34 assembly looks like this on x86-64:
     35 
     36 .. code-block:: nasm
     37 
     38     leaq     -8(%rsp), %r10
     39     cmpq     %fs:112,  %r10
     40     jg       .LBB0_2
     41 
     42     # More stack space needs to be allocated
     43     movabsq  $8, %r10   # The amount of space needed
     44     movabsq  $0, %r11   # The total size of arguments passed on stack
     45     callq    __morestack
     46     ret                 # The reason for this extra return is explained below
     47   .LBB0_2:
     48     # Usual prologue continues here
     49 
     50 The size of function arguments on the stack needs to be passed to
     51 ``__morestack`` (this function is implemented in ``libgcc``) since that number
     52 of bytes has to be copied from the previous stacklet to the current one. This is
     53 so that SP (and FP) relative addressing of function arguments work as expected.
     54 
     55 The unusual ``ret`` is needed to have the function which made a call to
     56 ``__morestack`` return correctly. ``__morestack``, instead of returning, calls
     57 into ``.LBB0_2``. This is possible since both, the size of the ``ret``
     58 instruction and the PC of call to ``__morestack`` are known. When the function
     59 body returns, control is transferred back to ``__morestack``. ``__morestack``
     60 then de-allocates the new stacklet, restores the correct SP value, and does a
     61 second return, which returns control to the correct caller.
     62 
     63 Variable Sized Allocas
     64 ----------------------
     65 
     66 The section on `allocating stacklets`_ automatically assumes that every stack
     67 frame will be of fixed size. However, LLVM allows the use of the ``llvm.alloca``
     68 intrinsic to allocate dynamically sized blocks of memory on the stack. When
     69 faced with such a variable-sized alloca, code is generated to:
     70 
     71 * Check if the current stacklet has enough space. If yes, just bump the SP, like
     72   in the normal case.
     73 * If not, generate a call to ``libgcc``, which allocates the memory from the
     74   heap.
     75 
     76 The memory allocated from the heap is linked into a list in the current
     77 stacklet, and freed along with the same. This prevents a memory leak.
     78