summaryrefslogtreecommitdiff
path: root/Documentation/vm/hmm.rst
blob: 4e3e9362afeb10e642e732dea04f1850da9069fa (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
.. hmm:

=====================================
Heterogeneous Memory Management (HMM)
=====================================

Provide infrastructure and helpers to integrate non-conventional memory (device
memory like GPU on board memory) into regular kernel path, with the cornerstone
of this being specialized struct page for such memory (see sections 5 to 7 of
this document).

HMM also provides optional helpers for SVM (Share Virtual Memory), i.e.,
allowing a device to transparently access program addresses coherently with
the CPU meaning that any valid pointer on the CPU is also a valid pointer
for the device. This is becoming mandatory to simplify the use of advanced
heterogeneous computing where GPU, DSP, or FPGA are used to perform various
computations on behalf of a process.

This document is divided as follows: in the first section I expose the problems
related to using device specific memory allocators. In the second section, I
expose the hardware limitations that are inherent to many platforms. The third
section gives an overview of the HMM design. The fourth section explains how
CPU page-table mirroring works and the purpose of HMM in this context. The
fifth section deals with how device memory is represented inside the kernel.
Finally, the last section presents a new migration helper that allows
leveraging the device DMA engine.

.. contents:: :local:

Problems of using a device specific memory allocator
====================================================

Devices with a large amount of on board memory (several gigabytes) like GPUs
have historically managed their memory through dedicated driver specific APIs.
This creates a disconnect between memory allocated and managed by a device
driver and regular application memory (private anonymous, shared memory, or
regular file backed memory). From here on I will refer to this aspect as split
address space. I use shared address space to refer to the opposite situation:
i.e., one in which any application memory region can be used by a device
transparently.

Split address space happens because devices can only access memory allocated
through a device specific API. This implies that all memory objects in a program
are not equal from the device point of view which complicates large programs
that rely on a wide set of libraries.

Concretely, this means that code that wants to leverage devices like GPUs needs
to copy objects between generically allocated memory (malloc, mmap private, mmap
share) and memory allocated through the device driver API (this still ends up
with an mmap but of the device file).

For flat data sets (array, grid, image, ...) this isn't too hard to achieve but
for complex data sets (list, tree, ...) it's hard to get right. Duplicating a
complex data set needs to re-map all the pointer relations between each of its
elements. This is error prone and programs get harder to debug because of the
duplicate data set and addresses.

Split address space also means that libraries cannot transparently use data
they are getting from the core program or another library and thus each library
might have to duplicate its input data set using the device specific memory
allocator. Large projects suffer from this and waste resources because of the
various memory copies.

Duplicating each library API to accept as input or output memory allocated by
each device specific allocator is not a viable option. It would lead to a
combinatorial explosion in the library entry points.

Finally, with the advance of high level language constructs (in C++ but in
other languages too) it is now possible for the compiler to leverage GPUs and
other devices without programmer knowledge. Some compiler identified patterns
are only do-able with a shared address space. It is also more reasonable to use
a shared address space for all other patterns.


I/O bus, device memory characteristics
======================================

I/O buses cripple shared address spaces due to a few limitations. Most I/O
buses only allow basic memory access from device to main memory; even cache
coherency is often optional. Access to device memory from a CPU is even more
limited. More often than not, it is not cache coherent.

If we only consider the PCIE bus, then a device can access main memory (often
through an IOMMU) and be cache coherent with the CPUs. However, it only allows
a limited set of atomic operations from the device on main memory. This is worse
in the other direction: the CPU can only access a limited range of the device
memory and cannot perform atomic operations on it. Thus device memory cannot
be considered the same as regular memory from the kernel point of view.

Another crippling factor is the limited bandwidth (~32GBytes/s with PCIE 4.0
and 16 lanes). This is 33 times less than the fastest GPU memory (1 TBytes/s).
The final limitation is latency. Access to main memory from the device has an
order of magnitude higher latency than when the device accesses its own memory.

Some platforms are developing new I/O buses or additions/modifications to PCIE
to address some of these limitations (OpenCAPI, CCIX). They mainly allow
two-way cache coherency between CPU and device and allow all atomic operations the
architecture supports. Sadly, not all platforms are following this trend and
some major architectures are left without hardware solutions to these problems.

So for shared address space to make sense, not only must we allow devices to
access any memory but we must also permit any memory to be migrated to device
memory while the device is using it (blocking CPU access while it happens).


Shared address space and migration
==================================

HMM intends to provide two main features. The first one is to share the address
space by duplicating the CPU page table in the device page table so the same
address points to the same physical memory for any valid main memory address in
the process address space.

To achieve this, HMM offers a set of helpers to populate the device page table
while keeping track of CPU page table updates. Device page table updates are
not as easy as CPU page table updates. To update the device page table, you must
allocate a buffer (or use a pool of pre-allocated buffers) and write GPU
specific commands in it to perform the update (unmap, cache invalidations, and
flush, ...). This cannot be done through common code for all devices. Hence
why HMM provides helpers to factor out everything that can be while leaving the
hardware specific details to the device driver.

The second mechanism HMM provides is a new kind of ZONE_DEVICE memory that
allows allocating a struct page for each page of device memory. Those pages
are special because the CPU cannot map them. However, they allow migrating
main memory to device memory using existing migration mechanisms and everything
looks like a page that is swapped out to disk from the CPU point of view. Using a
struct page gives the easiest and cleanest integration with existing mm
mechanisms. Here again, HMM only provides helpers, first to hotplug new ZONE_DEVICE
memory for the device memory and second to perform migration. Policy decisions
of what and when to migrate is left to the device driver.

Note that any CPU access to a device page triggers a page fault and a migration
back to main memory. For example, when a page backing a given CPU address A is
migrated from a main memory page to a device page, then any CPU access to
address A triggers a page fault and initiates a migration back to main memory.

With these two features, HMM not only allows a device to mirror process address
space and keeps both CPU and device page tables synchronized, but also
leverages device memory by migrating the part of the data set that is actively being
used by the device.


Address space mirroring implementation and API
==============================================

Address space mirroring's main objective is to allow duplication of a range of
CPU page table into a device page table; HMM helps keep both synchronized. A
device driver that wants to mirror a process address space must start with the
registration of a mmu_interval_notifier::

 int mmu_interval_notifier_insert(struct mmu_interval_notifier *interval_sub,
				  struct mm_struct *mm, unsigned long start,
				  unsigned long length,
				  const struct mmu_interval_notifier_ops *ops);

During the ops->invalidate() callback the device driver must perform the
update action to the range (mark range read only, or fully unmap, etc.). The
device must complete the update before the driver callback returns.

When the device driver wants to populate a range of virtual addresses, it can
use::

  long hmm_range_fault(struct hmm_range *range);

It will trigger a page fault on missing or read-only entries if write access is
requested (see below). Page faults use the generic mm page fault code path just
like a CPU page fault.

Both functions copy CPU page table entries into their pfns array argument. Each
entry in that array corresponds to an address in the virtual range. HMM
provides a set of flags to help the driver identify special CPU page table
entries.

Locking within the sync_cpu_device_pagetables() callback is the most important
aspect the driver must respect in order to keep things properly synchronized.
The usage pattern is::

 int driver_populate_range(...)
 {
      struct hmm_range range;
      ...

      range.notifier = &interval_sub;
      range.start = ...;
      range.end = ...;
      range.pfns = ...;
      range.flags = ...;
      range.values = ...;
      range.pfn_shift = ...;

      if (!mmget_not_zero(interval_sub->notifier.mm))
          return -EFAULT;

 again:
      range.notifier_seq = mmu_interval_read_begin(&interval_sub);
      down_read(&mm->mmap_sem);
      ret = hmm_range_fault(&range);
      if (ret) {
          up_read(&mm->mmap_sem);
          if (ret == -EBUSY)
                 goto again;
          return ret;
      }
      up_read(&mm->mmap_sem);

      take_lock(driver->update);
      if (mmu_interval_read_retry(&ni, range.notifier_seq) {
          release_lock(driver->update);
          goto again;
      }

      /* Use pfns array content to update device page table,
       * under the update lock */

      release_lock(driver->update);
      return 0;
 }

The driver->update lock is the same lock that the driver takes inside its
invalidate() callback. That lock must be held before calling
mmu_interval_read_retry() to avoid any race with a concurrent CPU page table
update.

Leverage default_flags and pfn_flags_mask
=========================================

The hmm_range struct has 2 fields, default_flags and pfn_flags_mask, that specify
fault or snapshot policy for the whole range instead of having to set them
for each entry in the pfns array.

For instance, if the device flags for range.flags are::

    range.flags[HMM_PFN_VALID] = (1 << 63);
    range.flags[HMM_PFN_WRITE] = (1 << 62);

and the device driver wants pages for a range with at least read permission,
it sets::

    range->default_flags = (1 << 63);
    range->pfn_flags_mask = 0;

and calls hmm_range_fault() as described above. This will fill fault all pages
in the range with at least read permission.

Now let's say the driver wants to do the same except for one page in the range for
which it wants to have write permission. Now driver set::

    range->default_flags = (1 << 63);
    range->pfn_flags_mask = (1 << 62);
    range->pfns[index_of_write] = (1 << 62);

With this, HMM will fault in all pages with at least read (i.e., valid) and for the
address == range->start + (index_of_write << PAGE_SHIFT) it will fault with
write permission i.e., if the CPU pte does not have write permission set then HMM
will call handle_mm_fault().

Note that HMM will populate the pfns array with write permission for any page
that is mapped with CPU write permission no matter what values are set
in default_flags or pfn_flags_mask.


Represent and manage device memory from core kernel point of view
=================================================================

Several different designs were tried to support device memory. The first one
used a device specific data structure to keep information about migrated memory
and HMM hooked itself in various places of mm code to handle any access to
addresses that were backed by device memory. It turns out that this ended up
replicating most of the fields of struct page and also needed many kernel code
paths to be updated to understand this new kind of memory.

Most kernel code paths never try to access the memory behind a page
but only care about struct page contents. Because of this, HMM switched to
directly using struct page for device memory which left most kernel code paths
unaware of the difference. We only need to make sure that no one ever tries to
map those pages from the CPU side.

Migration to and from device memory
===================================

Because the CPU cannot access device memory, migration must use the device DMA
engine to perform copy from and to device memory. For this we need to use
migrate_vma_setup(), migrate_vma_pages(), and migrate_vma_finalize() helpers.


Memory cgroup (memcg) and rss accounting
========================================

For now, device memory is accounted as any regular page in rss counters (either
anonymous if device page is used for anonymous, file if device page is used for
file backed page, or shmem if device page is used for shared memory). This is a
deliberate choice to keep existing applications, that might start using device
memory without knowing about it, running unimpacted.

A drawback is that the OOM killer might kill an application using a lot of
device memory and not a lot of regular system memory and thus not freeing much
system memory. We want to gather more real world experience on how applications
and system react under memory pressure in the presence of device memory before
deciding to account device memory differently.


Same decision was made for memory cgroup. Device memory pages are accounted
against same memory cgroup a regular page would be accounted to. This does
simplify migration to and from device memory. This also means that migration
back from device memory to regular memory cannot fail because it would
go above memory cgroup limit. We might revisit this choice latter on once we
get more experience in how device memory is used and its impact on memory
resource control.


Note that device memory can never be pinned by a device driver nor through GUP
and thus such memory is always free upon process exit. Or when last reference
is dropped in case of shared memory or file backed memory.