1SPU_RUN(2) Linux Programmer's Manual SPU_RUN(2)
2
3
4
6 spu_run - execute an SPU context
7
9 #include <sys/spu.h>
10
11 int spu_run(int fd, unsigned int *npc, unsigned int *event);
12
13 Note: There is no glibc wrapper for this system call; see NOTES.
14
16 The spu_run() system call is used on PowerPC machines that implement
17 the Cell Broadband Engine Architecture in order to access Synergistic
18 Processor Units (SPUs). The fd argument is a file descriptor returned
19 by spu_create(2) that refers to a specific SPU context. When the con‐
20 text gets scheduled to a physical SPU, it starts execution at the
21 instruction pointer passed in npc.
22
23 Execution of SPU code happens synchronously, meaning that spu_run()
24 blocks while the SPU is still running. If there is a need to execute
25 SPU code in parallel with other code on either the main CPU or other
26 SPUs, a new thread of execution must be created first (e.g., using
27 pthread_create(3)).
28
29 When spu_run() returns, the current value of the SPU program counter is
30 written to npc, so successive calls to spu_run() can use the same npc
31 pointer.
32
33 The event argument provides a buffer for an extended status code. If
34 the SPU context was created with the SPU_CREATE_EVENTS_ENABLED flag,
35 then this buffer is populated by the Linux kernel before spu_run()
36 returns.
37
38 The status code may be one (or more) of the following constants:
39
40 SPE_EVENT_DMA_ALIGNMENT
41 A DMA alignment error occurred.
42
43 SPE_EVENT_INVALID_DMA
44 An invalid MFC DMA command was attempted.
45
46 SPE_EVENT_SPE_DATA_STORAGE
47 A DMA storage error occurred.
48
49 SPE_EVENT_SPE_ERROR
50 An illegal instruction was executed.
51
52 NULL is a valid value for the event argument. In this case, the events
53 will not be reported to the calling process.
54
56 On success, spu_run() returns the value of the spu_status register. On
57 error, it returns -1 and sets errno to one of the error codes listed
58 below.
59
60 The spu_status register value is a bit mask of status codes and option‐
61 ally a 14-bit code returned from the stop-and-signal instruction on the
62 SPU. The bit masks for the status codes are:
63
64 0x02 SPU was stopped by a stop-and-signal instruction.
65
66 0x04 SPU was stopped by a halt instruction.
67
68 0x08 SPU is waiting for a channel.
69
70 0x10 SPU is in single-step mode.
71
72 0x20 SPU has tried to execute an invalid instruction.
73
74 0x40 SPU has tried to access an invalid channel.
75
76 0x3fff0000
77 The bits masked with this value contain the code returned from a
78 stop-and-signal instruction. These bits are valid only if the
79 0x02 bit is set.
80
81 If spu_run() has not returned an error, one or more bits among the
82 lower eight ones are always set.
83
85 EBADF fd is not a valid file descriptor.
86
87 EFAULT npc is not a valid pointer, or event is non-NULL and an invalid
88 pointer.
89
90 EINTR A signal occurred while spu_run() was in progress; see sig‐
91 nal(7). The npc value has been updated to the new program
92 counter value if necessary.
93
94 EINVAL fd is not a valid file descriptor returned from spu_create(2).
95
96 ENOMEM There was not enough memory available to handle a page fault
97 resulting from a Memory Flow Controller (MFC) direct memory
98 access.
99
100 ENOSYS The functionality is not provided by the current system, because
101 either the hardware does not provide SPUs or the spufs module is
102 not loaded.
103
105 The spu_run() system call was added to Linux in kernel 2.6.16.
106
108 This call is Linux-specific and implemented only by the PowerPC archi‐
109 tecture. Programs using this system call are not portable.
110
112 Glibc does not provide a wrapper for this system call; call it using
113 syscall(2). Note however, that spu_run() is meant to be used from
114 libraries that implement a more abstract interface to SPUs, not to be
115 used from regular applications. See ⟨http://www.bsc.es/projects
116 /deepcomputing/linuxoncell/⟩ for the recommended libraries.
117
119 The following is an example of running a simple, one-instruction SPU
120 program with the spu_run() system call.
121
122 #include <stdlib.h>
123 #include <stdint.h>
124 #include <unistd.h>
125 #include <stdio.h>
126 #include <sys/types.h>
127 #include <fcntl.h>
128
129 #define handle_error(msg) \
130 do { perror(msg); exit(EXIT_FAILURE); } while (0)
131
132 int main(void)
133 {
134 int context, fd, spu_status;
135 uint32_t instruction, npc;
136
137 context = spu_create("/spu/example-context", 0, 0755);
138 if (context == -1)
139 handle_error("spu_create");
140
141 /* write a 'stop 0x1234' instruction to the SPU's
142 * local store memory
143 */
144 instruction = 0x00001234;
145
146 fd = open("/spu/example-context/mem", O_RDWR);
147 if (fd == -1)
148 handle_error("open");
149 write(fd, &instruction, sizeof(instruction));
150
151 /* set npc to the starting instruction address of the
152 * SPU program. Since we wrote the instruction at the
153 * start of the mem file, the entry point will be 0x0
154 */
155 npc = 0;
156
157 spu_status = spu_run(context, &npc, NULL);
158 if (spu_status == -1)
159 handle_error("open");
160
161 /* we should see a status code of 0x1234002:
162 * 0x00000002 (spu was stopped due to stop-and-signal)
163 * | 0x12340000 (the stop-and-signal code)
164 */
165 printf("SPU Status: 0x%08x\n", spu_status);
166
167 exit(EXIT_SUCCESS);
168 }
169
171 close(2), spu_create(2), capabilities(7), spufs(7)
172
174 This page is part of release 4.16 of the Linux man-pages project. A
175 description of the project, information about reporting bugs, and the
176 latest version of this page, can be found at
177 https://www.kernel.org/doc/man-pages/.
178
179
180
181Linux 2017-09-15 SPU_RUN(2)