xref: /linux/Documentation/userspace-api/accelerators/ocxl.rst (revision 0526b56cbc3c489642bd6a5fe4b718dea7ef0ee8)
1========================================================
2OpenCAPI (Open Coherent Accelerator Processor Interface)
3========================================================
4
5OpenCAPI is an interface between processors and accelerators. It aims
6at being low-latency and high-bandwidth. The specification is
7developed by the `OpenCAPI Consortium <http://opencapi.org/>`_.
8
9It allows an accelerator (which could be an FPGA, ASICs, ...) to access
10the host memory coherently, using virtual addresses. An OpenCAPI
11device can also host its own memory, that can be accessed from the
12host.
13
14OpenCAPI is known in linux as 'ocxl', as the open, processor-agnostic
15evolution of 'cxl' (the driver for the IBM CAPI interface for
16powerpc), which was named that way to avoid confusion with the ISDN
17CAPI subsystem.
18
19
20High-level view
21===============
22
23OpenCAPI defines a Data Link Layer (DL) and Transaction Layer (TL), to
24be implemented on top of a physical link. Any processor or device
25implementing the DL and TL can start sharing memory.
26
27::
28
29  +-----------+                         +-------------+
30  |           |                         |             |
31  |           |                         | Accelerated |
32  | Processor |                         |  Function   |
33  |           |  +--------+             |    Unit     |  +--------+
34  |           |--| Memory |             |    (AFU)    |--| Memory |
35  |           |  +--------+             |             |  +--------+
36  +-----------+                         +-------------+
37       |                                       |
38  +-----------+                         +-------------+
39  |    TL     |                         |    TLX      |
40  +-----------+                         +-------------+
41       |                                       |
42  +-----------+                         +-------------+
43  |    DL     |                         |    DLX      |
44  +-----------+                         +-------------+
45       |                                       |
46       |                   PHY                 |
47       +---------------------------------------+
48
49
50
51Device discovery
52================
53
54OpenCAPI relies on a PCI-like configuration space, implemented on the
55device. So the host can discover AFUs by querying the config space.
56
57OpenCAPI devices in Linux are treated like PCI devices (with a few
58caveats). The firmware is expected to abstract the hardware as if it
59was a PCI link. A lot of the existing PCI infrastructure is reused:
60devices are scanned and BARs are assigned during the standard PCI
61enumeration. Commands like 'lspci' can therefore be used to see what
62devices are available.
63
64The configuration space defines the AFU(s) that can be found on the
65physical adapter, such as its name, how many memory contexts it can
66work with, the size of its MMIO areas, ...
67
68
69
70MMIO
71====
72
73OpenCAPI defines two MMIO areas for each AFU:
74
75* the global MMIO area, with registers pertinent to the whole AFU.
76* a per-process MMIO area, which has a fixed size for each context.
77
78
79
80AFU interrupts
81==============
82
83OpenCAPI includes the possibility for an AFU to send an interrupt to a
84host process. It is done through a 'intrp_req' defined in the
85Transaction Layer, specifying a 64-bit object handle which defines the
86interrupt.
87
88The driver allows a process to allocate an interrupt and obtain its
8964-bit object handle, that can be passed to the AFU.
90
91
92
93char devices
94============
95
96The driver creates one char device per AFU found on the physical
97device. A physical device may have multiple functions and each
98function can have multiple AFUs. At the time of this writing though,
99it has only been tested with devices exporting only one AFU.
100
101Char devices can be found in /dev/ocxl/ and are named as:
102/dev/ocxl/<AFU name>.<location>.<index>
103
104where <AFU name> is a max 20-character long name, as found in the
105config space of the AFU.
106<location> is added by the driver and can help distinguish devices
107when a system has more than one instance of the same OpenCAPI device.
108<index> is also to help distinguish AFUs in the unlikely case where a
109device carries multiple copies of the same AFU.
110
111
112
113Sysfs class
114===========
115
116An ocxl class is added for the devices representing the AFUs. See
117/sys/class/ocxl. The layout is described in
118Documentation/ABI/testing/sysfs-class-ocxl
119
120
121
122User API
123========
124
125open
126----
127
128Based on the AFU definition found in the config space, an AFU may
129support working with more than one memory context, in which case the
130associated char device may be opened multiple times by different
131processes.
132
133
134ioctl
135-----
136
137OCXL_IOCTL_ATTACH:
138
139  Attach the memory context of the calling process to the AFU so that
140  the AFU can access its memory.
141
142OCXL_IOCTL_IRQ_ALLOC:
143
144  Allocate an AFU interrupt and return an identifier.
145
146OCXL_IOCTL_IRQ_FREE:
147
148  Free a previously allocated AFU interrupt.
149
150OCXL_IOCTL_IRQ_SET_FD:
151
152  Associate an event fd to an AFU interrupt so that the user process
153  can be notified when the AFU sends an interrupt.
154
155OCXL_IOCTL_GET_METADATA:
156
157  Obtains configuration information from the card, such at the size of
158  MMIO areas, the AFU version, and the PASID for the current context.
159
160OCXL_IOCTL_ENABLE_P9_WAIT:
161
162  Allows the AFU to wake a userspace thread executing 'wait'. Returns
163  information to userspace to allow it to configure the AFU. Note that
164  this is only available on POWER9.
165
166OCXL_IOCTL_GET_FEATURES:
167
168  Reports on which CPU features that affect OpenCAPI are usable from
169  userspace.
170
171
172mmap
173----
174
175A process can mmap the per-process MMIO area for interactions with the
176AFU.
177