1 Coherent Accelerator Interface (CXL)
2 ====================================
7 The coherent accelerator interface is designed to allow the
8 coherent connection of accelerators (FPGAs and other devices) to a
9 POWER system. These devices need to adhere to the Coherent
10 Accelerator Interface Architecture (CAIA).
12 IBM refers to this as the Coherent Accelerator Processor Interface
13 or CAPI. In the kernel it's referred to by the name CXL to avoid
14 confusion with the ISDN CAPI subsystem.
16 Coherent in this context means that the accelerator and CPUs can
17 both access system memory directly and with the same effective
25 +----------+ +---------+
31 +----------+ +---------+
35 +---+------+ PCIE +---------+
37 The POWER8 chip has a Coherently Attached Processor Proxy (CAPP)
38 unit which is part of the PCIe Host Bridge (PHB). This is managed
39 by Linux by calls into OPAL. Linux doesn't directly program the
42 The FPGA (or coherently attached device) consists of two parts.
43 The POWER Service Layer (PSL) and the Accelerator Function Unit
44 (AFU). The AFU is used to implement specific functionality behind
45 the PSL. The PSL, among other things, provides memory address
46 translation services to allow each AFU direct access to userspace
49 The AFU is the core part of the accelerator (eg. the compression,
50 crypto etc function). The kernel has no knowledge of the function
51 of the AFU. Only userspace interacts directly with the AFU.
53 The PSL provides the translation and interrupt services that the
54 AFU needs. This is what the kernel interacts with. For example, if
55 the AFU needs to read a particular effective address, it sends
56 that address to the PSL, the PSL then translates it, fetches the
57 data from memory and returns it to the AFU. If the PSL has a
58 translation miss, it interrupts the kernel and the kernel services
59 the fault. The context to which this fault is serviced is based on
60 who owns that acceleration function.
66 There are two programming modes supported by the AFU. Dedicated
67 and AFU directed. AFU may support one or both modes.
69 When using dedicated mode only one MMU context is supported. In
70 this mode, only one userspace process can use the accelerator at
73 When using AFU directed mode, up to 16K simultaneous contexts can
74 be supported. This means up to 16K simultaneous userspace
75 applications may use the accelerator (although specific AFUs may
76 support fewer). In this mode, the AFU sends a 16 bit context ID
77 with each of its requests. This tells the PSL which context is
78 associated with each operation. If the PSL can't translate an
79 operation, the ID can also be accessed by the kernel so it can
80 determine the userspace context associated with an operation.
86 A portion of the accelerator MMIO space can be directly mapped
87 from the AFU to userspace. Either the whole space can be mapped or
88 just a per context portion. The hardware is self describing, hence
89 the kernel can determine the offset and size of the per context
96 AFUs may generate interrupts that are destined for userspace. These
97 are received by the kernel as hardware interrupts and passed onto
98 userspace by a read syscall documented below.
100 Data storage faults and error interrupts are handled by the kernel
104 Work Element Descriptor (WED)
105 =============================
107 The WED is a 64-bit parameter passed to the AFU when a context is
108 started. Its format is up to the AFU hence the kernel has no
109 knowledge of what it represents. Typically it will be the
110 effective address of a work queue or status block where the AFU
111 and userspace can share control and status information.
119 For AFUs operating in AFU directed mode, two character device
120 files will be created. /dev/cxl/afu0.0m will correspond to a
121 master context and /dev/cxl/afu0.0s will correspond to a slave
122 context. Master contexts have access to the full MMIO space an
123 AFU provides. Slave contexts have access to only the per process
124 MMIO space an AFU provides.
126 For AFUs operating in dedicated process mode, the driver will
127 only create a single character device per AFU called
128 /dev/cxl/afu0.0d. This will have access to the entire MMIO space
129 that the AFU provides (like master contexts in AFU directed).
131 The types described below are defined in include/uapi/misc/cxl.h
133 The following file operations are supported on both slave and
140 Opens the device and allocates a file descriptor to be used with
143 A dedicated mode AFU only has one context and only allows the
144 device to be opened once.
146 An AFU directed mode AFU can have many contexts, the device can be
147 opened once for each context that is available.
149 When all available contexts are allocated the open call will fail
152 Note: IRQs need to be allocated for each context, which may limit
153 the number of contexts that can be created, and therefore
154 how many times the device can be opened. The POWER8 CAPP
155 supports 2040 IRQs and 3 are used by the kernel, so 2037 are
156 left. If 1 IRQ is needed per context, then only 2037
157 contexts can be allocated. If 4 IRQs are needed per context,
158 then only 2037/4 = 509 contexts can be allocated.
164 CXL_IOCTL_START_WORK:
165 Starts the AFU context and associates it with the current
166 process. Once this ioctl is successfully executed, all memory
167 mapped into this process is accessible to this AFU context
168 using the same effective addresses. No additional calls are
169 required to map/unmap memory. The AFU memory context will be
170 updated as userspace allocates and frees memory. This ioctl
171 returns once the AFU context is started.
173 Takes a pointer to a struct cxl_ioctl_start_work:
175 struct cxl_ioctl_start_work {
177 __u64 work_element_descriptor;
179 __s16 num_interrupts;
189 Indicates which optional fields in the structure are
192 work_element_descriptor:
193 The Work Element Descriptor (WED) is a 64-bit argument
194 defined by the AFU. Typically this is an effective
195 address pointing to an AFU specific structure
196 describing what work to perform.
199 Authority Mask Register (AMR), same as the powerpc
200 AMR. This field is only used by the kernel when the
201 corresponding CXL_START_WORK_AMR value is specified in
202 flags. If not specified the kernel will use a default
206 Number of userspace interrupts to request. This field
207 is only used by the kernel when the corresponding
208 CXL_START_WORK_NUM_IRQS value is specified in flags.
209 If not specified the minimum number required by the
210 AFU will be allocated. The min and max number can be
214 For ABI padding and future extensions
216 CXL_IOCTL_GET_PROCESS_ELEMENT:
217 Get the current context id, also known as the process element.
218 The value is returned from the kernel as a __u32.
224 An AFU may have an MMIO space to facilitate communication with the
225 AFU. If it does, the MMIO space can be accessed via mmap. The size
226 and contents of this area are specific to the particular AFU. The
227 size can be discovered via sysfs.
229 In AFU directed mode, master contexts are allowed to map all of
230 the MMIO space and slave contexts are allowed to only map the per
231 process MMIO space associated with the context. In dedicated
232 process mode the entire MMIO space can always be mapped.
234 This mmap call must be done after the START_WORK ioctl.
236 Care should be taken when accessing MMIO space. Only 32 and 64-bit
237 accesses are supported by POWER8. Also, the AFU will be designed
238 with a specific endianness, so all MMIO accesses should consider
239 endianness (recommend endian(3) variants like: le64toh(),
240 be64toh() etc). These endian issues equally apply to shared memory
241 queues the WED may describe.
247 Reads events from the AFU. Blocks if no events are pending
248 (unless O_NONBLOCK is supplied). Returns -EIO in the case of an
249 unrecoverable error or if the card is removed.
251 read() will always return an integral number of events.
253 The buffer passed to read() must be at least 4K bytes.
255 The result of the read will be a buffer of one or more events,
256 each event is of type struct cxl_event, of varying size.
259 struct cxl_event_header header;
261 struct cxl_event_afu_interrupt irq;
262 struct cxl_event_data_storage fault;
263 struct cxl_event_afu_error afu_error;
267 The struct cxl_event_header is defined as:
269 struct cxl_event_header {
272 __u16 process_element;
277 This defines the type of event. The type determines how
278 the rest of the event is structured. These types are
279 described below and defined by enum cxl_event_type.
282 This is the size of the event in bytes including the
283 struct cxl_event_header. The start of the next event can
284 be found at this offset from the start of the current
288 Context ID of the event.
291 For future extensions and padding.
293 If the event type is CXL_EVENT_AFU_INTERRUPT then the event
294 structure is defined as:
296 struct cxl_event_afu_interrupt {
298 __u16 irq; /* Raised AFU interrupt number */
303 These flags indicate which optional fields are present
304 in this struct. Currently all fields are mandatory.
307 The IRQ number sent by the AFU.
310 For future extensions and padding.
312 If the event type is CXL_EVENT_DATA_STORAGE then the event
313 structure is defined as:
315 struct cxl_event_data_storage {
325 These flags indicate which optional fields are present in
326 this struct. Currently all fields are mandatory.
329 The address that the AFU unsuccessfully attempted to
330 access. Valid accesses will be handled transparently by the
331 kernel but invalid accesses will generate this event.
334 This field gives information on the type of fault. It is a
335 copy of the DSISR from the PSL hardware when the address
336 fault occurred. The form of the DSISR is as defined in the
340 For future extensions
342 If the event type is CXL_EVENT_AFU_ERROR then the event structure
345 struct cxl_event_afu_error {
353 These flags indicate which optional fields are present in
354 this struct. Currently all fields are Mandatory.
357 Error status from the AFU. Defined by the AFU.
360 For future extensions and padding
365 A cxl sysfs class is added under /sys/class/cxl to facilitate
366 enumeration and tuning of the accelerators. Its layout is
367 described in Documentation/ABI/testing/sysfs-class-cxl
372 The following udev rules could be used to create a symlink to the
373 most logical chardev to use in any programming mode (afuX.Yd for
374 dedicated, afuX.Ys for afu directed), since the API is virtually
377 SUBSYSTEM=="cxl", ATTRS{mode}=="dedicated_process", SYMLINK="cxl/%b"
378 SUBSYSTEM=="cxl", ATTRS{mode}=="afu_directed", \
379 KERNEL=="afu[0-9]*.[0-9]*s", SYMLINK="cxl/%b"