1 Title: Understanding the current GC, part III
2 Tags: en, d, dgc, understanding the current gc, druntime, gc, mark-sweep, allocation
4 In the `previous post`_ we focused on the ``Gcx`` object, the core of the GC
5 in druntime (and Phobos and Tango, they are all are based on the same
6 implementation). In this post we will focus on **allocation**, which a little
7 more complex than it should be in my opinion.
9 It was not an easy task to follow how allocation works. A ``GC.malloc()`` call
10 spawns into this function calls::
14 '---> GC.mallocNoSync(size, bits)
16 |---> Gcx.allocPage(bin_size)
18 | '---> Pool.allocPages(n_pages)
20 | '---> Pool.extendPages(n_pages)
22 | '---> os_mem_commit(addr, offset, size)
24 |---> Gcx.fullcollectshell()
26 |---> Gcx.newPool(n_pages)
28 | '---> Pool.initialize(n_pages)
30 | |---> os_mem_map(mem_size)
32 | '---> GCBits.alloc(bits_size)
34 '---> Gcx.bigAlloc(size)
36 |---> Pool.allocPages(n_pages)
39 |---> Gcx.fullcollectshell()
45 | |---> os_mem_decommit(addr, offset, size)
47 | |---> os_mem_map(addr, size)
51 '---> Gcx.newPool(n_pages)
54 Doesn't look so simple, ugh?
56 The map/commit differentiation of Windows doesn't exactly help simplicity. Note
57 that ``Pool.initialize()`` maps the memory (reserve the address space) while
58 ``Pool.allocPages()`` (through ``Pool.extendPages()``) commit the new memory
59 (ask the OS to actually reserve the virtual memory). I don't know how good is
60 this for Windows (or put in another way, how bad could it be for Windows if all
61 mapped memory gets immediately committed), but it adds a new layer of
62 complexity (that's not even needed in Posix OSs). The whole branch starting at
63 ``Gcx.allocPage(bin_size)`` would be gone if this distinction it's not made.
64 Besides this, it worsen Posix OSs performance, because there are some
65 non-trivial lookups to handle this non-existing non-committed pages, even when
66 the ``os_mem_commit()`` and ``os_mem_decommit()`` functions are NOP and can be
67 optimized out, the lookups are there.
69 .. admonition:: Mental Note
71 See if getting rid of the ``commit()``/``decommit()`` stuff improves
74 But well, let's forget about this issue for now and live with it. Here is
75 a summary of what all this functions do.
79 I recommend to give another read to the (updated) `previous posts of
80 this series`_, specially if you are not familiar with the ``Pool``
81 concept and implementation.
83 ``GC.malloc(size, bits)``
84 This is just a wrapper for multi-threaded code, it takes the ``GCLock`` if
85 necessary and calls ``GC.mallocNoSync(size, bits)``.
87 ``GC.mallocNoSync(size, bits)``
88 This function has 2 different algorithms for small objects (less than
89 a page of 4KiB) and another for big objects.
91 It does some common work for both cases, like logging and adding a sentinel
92 for debugging purposes (if those feature are enabled), finding the bin size
93 (``bin_size``) that better fits ``size`` (and cache the result as an
94 optimization for consecutive calls to malloc with the same size) and
95 setting the ``bits`` (``NO_SCAN``, ``NO_MOVE``, ``FINALIZE``) to the
98 Small objects (``bin_size < B_PAGE``)
99 Looks at the free list (``Gcx.bucket``) trying to find a page with the
100 minimum bin size that's equals or bigger than ``size``. If it can't
101 succeed, it calls ``Gcx.allocPage(bin_size)`` to find room in
102 uncommitted pages. If there still no room for the requested amount of
103 memory, it triggers a collection (``Gcx.fullcollectshell()``). If there
104 is still no luck, ``Gcx.newPage(1)`` is called to ask the OS for more
105 memory. Then it calls again ``Gcx.allocPage(bin_size)`` (remember the
106 new memory is just mmap'ped but not commit'ed) and if there is no room
107 in the free list still, an out of memory error is issued.
109 Big objects (``B_PAGE`` and ``B_PAGEPLUS``)
110 It simply calls ``Gcx.bigAlloc(size)`` and issue an out of memory error
111 if that call fails to get the requested memory.
113 ``Gcx.allocPage(bin_size)``
114 This function linearly search the ``pooltable`` for a ``Pool`` with an
115 allocable page (i.e. a page already mapped by not yet committed). This is
116 done through a call to ``Pool.allocPages(1)``. If a page is found, its bin
117 size is set to ``bin_size`` via the ``Pool``'s ``pagetable``, and all the
118 bins of that page are linked to the free list (``Gcx.bucket``).
120 ``Pool.allocPages(n_pages)``
121 Search for ``n_pages`` consecutive free pages (``B_FREE``) in the committed
122 pages (pages in the ``pagetable`` with index up to ``ncommited``). If
123 they're not found, ``Pool.extendPages(n_pages)`` is called to commit some
124 more mapped pages to fulfill the request.
126 ``Pool.extendPages(n_pages)``
127 Commit ``n_pages`` already mapped pages (calling ``os_mem_commit()``),
128 setting them as free (``B_FREE``) and updating the ``ncommited`` attribute.
129 If there are not that many uncommitted pages, it returns an error.
131 ``Gcx.newPool(n_pages)``
132 This function adds a new ``Pool`` to the ``pooltable``. It first adjusts
133 the ``n_pages`` variable using various rules (for example, it duplicates
134 the current allocated memory until 8MiB are allocated and then allocates
135 8MiB pools always, unless more memory is requested in the first place, of
136 course). Then a new ``Pool`` is created with the adjusted ``n_pages`` value
137 and it's initialized calling to ``Pool.initialize(n_pages)``, the
138 ``pooltable`` is resized to fit the new number of pools (``npools``) and
139 sorted using ``Pool.opCmp()`` (which uses the ``baseAddr`` to compare).
140 Finally the ``minAddr`` and ``maxAddr`` attributes are updated.
142 ``Pool.initialize(n_pages)``
143 Initializes all the ``Pool`` attributes, mapping the requested number of
144 pages (``n_pages``) using ``os_mem_map()``. All the bit sets (``mark``,
145 ``scan``, ``freebits``, ``noscan``) are allocated (using
146 ``GCBits.alloc()``) to ``n_pages * PAGESIZE / 16`` bits and the
147 ``pagetable`` too, setting all bins to ``B_UNCOMMITTED`` and
150 ``Gcx.bigAlloc(size)``
151 This is the weirdest function by far. There are very strange things, but
152 I'll try to explain what I understand from it (what I think it's trying to
155 It first make a simple lookup in the ``pooltable`` for ``n_pages``
156 consecutive pages in any existing ``Pool`` (calling
157 ``Pool.allocPages(n_pages)`` as in ``Gcx.allocPage()``). If this fails, it
158 runs a ``fullcollectshell()`` (if not ``disabled``) then calls to
159 ``minimize()`` (to prevent bloat) and then create a new pool (calling
160 ``newPool()`` followed by ``Pool.allocPages()``). If all that fails, it
161 returns an error. If something succeed, the bin size for the first page is
162 set to ``B_PAGE`` and the remaining pages are set to ``B_PAGEPLUS`` (if
163 any). If there is any unused memory at the end, it's initialized to
164 0 (to prevent false positives when scanning I guess).
166 The weird thing about this, is that a lot of lookups into the ``pooltable``
167 are done in certain condition, but I think they are not needed because
168 there are no changes that can make new room.
170 I don't know if this is legacy code that never got updated and have a lot
171 of useless lookups or if I'm getting something wrong. **Help is welcome!**
173 There is not much to say about ``os_mem_xxx()``, ``Gcx.minimize()`` and
174 ``Gcx.fullcollectshell()`` functions, they were briefly described in the
175 `previous posts of this series`_. ``Pool.Dtor()`` just undo what was done in
176 ``Pool.initialize()``.
178 A final word about the free list (``Gcx.bucket``). It's just a simple linked
179 list. It uses the first ``size_t`` bytes of the free bin to point to the next
180 free bin (there's always room for a pointer in a bin because their minimum size
181 is 16 bytes). A simple structure is used to *easy* this::
187 Then, the memory cell is casted to this structure to use the next
191 gcx.bucket[bin] = (cast(List*) p).next
193 I really have my doubts if this is even a little less cryptic than::
196 gcx.bucket[bin] = *(cast(void**) p)
198 But what the hell, this is no really important =)
201 .. _`previous post`: http://proj.llucax.com.ar/blog/dgc/blog/post/526122be
202 .. _`previous posts of this series`: http://proj.llucax.com.ar/blog/dgc/blog/tag/understanding%20the%20current%20gc
204 .. vim: set et sw=4 sts=4 :