1SPU_CREATE(2)              Linux Programmer's Manual             SPU_CREATE(2)
2
3
4

NAME

6       spu_create - create a new spu context
7

SYNOPSIS

9       #include <sys/types.h>
10       #include <sys/spu.h>
11
12       int spu_create(const char *pathname, int flags, mode_t mode,
13                      int neighbor_fd);
14
15       Note: There is no glibc wrapper for this system call; see NOTES.
16

DESCRIPTION

18       The spu_create() system call is used on PowerPC machines that implement
19       the Cell Broadband Engine Architecture in order to  access  Synergistic
20       Processor Units (SPUs).  It creates a new logical context for an SPU in
21       pathname and returns a file descriptor associated  with  it.   pathname
22       must  refer  to  a  nonexistent directory in the mount point of the SPU
23       filesystem (spufs).  If spu_create() is successful, a directory is cre‐
24       ated  at  pathname  and  it  is  populated  with the files described in
25       spufs(7).
26
27       When a context is created, the returned file  descriptor  can  only  be
28       passed  to  spu_run(2), used as the dirfd argument to the *at family of
29       system calls (e.g., openat(2)), or closed; other operations are not de‐
30       fined.   A  logical SPU context is destroyed (along with all files cre‐
31       ated within the context's pathname directory) once the  last  reference
32       to  the  context has gone; this usually occurs when the file descriptor
33       returned by spu_create() is closed.
34
35       The mode argument (minus any bits set in the process's umask(2)) speci‐
36       fies the permissions used for creating the new directory in spufs.  See
37       stat(2) for a full list of the possible mode values.
38
39       The neighbor_fd is used only when the SPU_CREATE_AFFINITY_SPU  flag  is
40       specified; see below.
41
42       The  flags argument can be zero or any bitwise OR-ed combination of the
43       following constants:
44
45       SPU_CREATE_EVENTS_ENABLED
46              Rather than using signals for  reporting  DMA  errors,  use  the
47              event argument to spu_run(2).
48
49       SPU_CREATE_GANG
50              Create  an SPU gang instead of a context.  (A gang is a group of
51              SPU contexts that are functionally related  to  each  other  and
52              which  share  common  scheduling parameters—priority and policy.
53              In the future, gang scheduling may be  implemented  causing  the
54              group to be switched in and out as a single unit.)
55
56              A new directory will be created at the location specified by the
57              pathname argument.  This gang may be used to hold other SPU con‐
58              texts, by providing a pathname that is within the gang directory
59              to further calls to spu_create().
60
61       SPU_CREATE_NOSCHED
62              Create a context that is not  affected  by  the  SPU  scheduler.
63              Once  the  context is run, it will not be scheduled out until it
64              is destroyed by the creating process.
65
66              Because the context cannot be removed from the SPU,  some  func‐
67              tionality  is  disabled for SPU_CREATE_NOSCHED contexts.  Only a
68              subset of the files will be available in this context  directory
69              in spufs.  Additionally, SPU_CREATE_NOSCHED contexts cannot dump
70              a core file when crashing.
71
72              Creating SPU_CREATE_NOSCHED contexts requires  the  CAP_SYS_NICE
73              capability.
74
75       SPU_CREATE_ISOLATE
76              Create an isolated SPU context.  Isolated contexts are protected
77              from some PPE (PowerPC Processing Element) operations,  such  as
78              access to the SPU local store and the NPC register.
79
80              Creating  SPU_CREATE_ISOLATE contexts also requires the SPU_CRE‐
81              ATE_NOSCHED flag.
82
83       SPU_CREATE_AFFINITY_SPU (since Linux 2.6.23)
84              Create a context with affinity to  another  SPU  context.   This
85              affinity  information  is  used  within the SPU scheduling algo‐
86              rithm.  Using this flag requires that a file  descriptor  refer‐
87              ring to the other SPU context be passed in the neighbor_fd argu‐
88              ment.
89
90       SPU_CREATE_AFFINITY_MEM (since Linux 2.6.23)
91              Create a context with affinity to system memory.  This  affinity
92              information is used within the SPU scheduling algorithm.
93

RETURN VALUE

95       On  success,  spu_create() returns a new file descriptor.  On error, -1
96       is returned, and errno is set to one of the error codes listed below.
97

ERRORS

99       EACCES The current user does not have  write  access  to  the  spufs(7)
100              mount point.
101
102       EEXIST An SPU context already exists at the given pathname.
103
104       EFAULT pathname  is not a valid string pointer in the calling process's
105              address space.
106
107       EINVAL pathname is not a directory in the spufs(7) mount point, or  in‐
108              valid flags have been provided.
109
110       ELOOP  Too many symbolic links were found while resolving pathname.
111
112       EMFILE The per-process limit on the number of open file descriptors has
113              been reached.
114
115       ENAMETOOLONG
116              pathname is too long.
117
118       ENFILE The system-wide limit on the total number of open files has been
119              reached.
120
121       ENODEV An  isolated  context  was  requested, but the hardware does not
122              support SPU isolation.
123
124       ENOENT Part of pathname could not be resolved.
125
126       ENOMEM The kernel could not allocate all resources required.
127
128       ENOSPC There are not enough SPU resources available  to  create  a  new
129              context  or  the  user-specific limit for the number of SPU con‐
130              texts has been reached.
131
132       ENOSYS The functionality is not provided by the current system, because
133              either the hardware does not provide SPUs or the spufs module is
134              not loaded.
135
136       ENOTDIR
137              A part of pathname is not a directory.
138
139       EPERM  The SPU_CREATE_NOSCHED flag has been given, but  the  user  does
140              not have the CAP_SYS_NICE capability.
141

FILES

143       pathname must point to a location beneath the mount point of spufs.  By
144       convention, it gets mounted in /spu.
145

VERSIONS

147       The spu_create() system call was added to Linux in kernel 2.6.16.
148

CONFORMING TO

150       This call is Linux-specific and implemented only on the PowerPC  archi‐
151       tecture.  Programs using this system call are not portable.
152

NOTES

154       Glibc  does  not  provide a wrapper for this system call; call it using
155       syscall(2).  Note however, that spu_create() is meant to be  used  from
156       libraries  that  implement a more abstract interface to SPUs, not to be
157       used  from  regular  applications.    See   ⟨http://www.bsc.es/projects
158       /deepcomputing/linuxoncell/⟩ for the recommended libraries.
159
160       Prior  to  the  addition  of  the SPU_CREATE_AFFINITY_SPU flag in Linux
161       2.6.23, the spu_create() system call took only three  arguments  (i.e.,
162       there was no neighbor_fd argument).
163

EXAMPLES

165       See spu_run(2) for an example of the use of spu_create()
166

SEE ALSO

168       close(2), spu_run(2), capabilities(7), spufs(7)
169

COLOPHON

171       This  page  is  part of release 5.10 of the Linux man-pages project.  A
172       description of the project, information about reporting bugs,  and  the
173       latest     version     of     this    page,    can    be    found    at
174       https://www.kernel.org/doc/man-pages/.
175
176
177
178Linux                             2020-12-21                     SPU_CREATE(2)
Impressum