path: root/man7/spufs.7

.\" Copyright (c) International Business Machines Corp., 2006
.\"
.\" SPDX-License-Identifier: GPL-2.0-or-later
.\"
.\" HISTORY:
.\" 2005-09-28, created by Arnd Bergmann <arndb@de.ibm.com>,
.\"   Mark Nutter <mnutter@us.ibm.com> and
.\"   Ulrich Weigand <Ulrich.Weigand@de.ibm.com>
.\" 2006-06-16, revised by Eduardo M. Fleury <efleury@br.ibm.com>
.\" 2007-07-10, quite a lot of polishing by mtk
.\" 2007-09-28, updates for newer kernels by Jeremy Kerr <jk@ozlabs.org>
.\"
.TH spufs 7 (date) "Linux man-pages (unreleased)"
.SH NAME
spufs \- SPU filesystem
.SH DESCRIPTION
The SPU filesystem is used on PowerPC machines that implement the
Cell Broadband Engine Architecture in order to access Synergistic
Processor Units (SPUs).
.P
The filesystem provides a name space similar to POSIX shared
memory or message queues.
Users that have write permissions
on the filesystem can use
.BR spu_create (2)
to establish SPU contexts under the
.B spufs
root directory.
.P
Every SPU context is represented by a directory containing
a predefined set of files.
These files can be
used for manipulating the state of the logical SPU.
Users can change permissions on the files, but can't
add or remove files.
.SS Mount options
.TP
.B uid=<uid>
Set the user owning the mount point; the default is 0 (root).
.TP
.B gid=<gid>
Set the group owning the mount point; the default is 0 (root).
.TP
.B mode=<mode>
Set the mode of the top-level directory in
.BR spufs ,
as an octal mode string.
The default is 0775.
.SS Files
The files in
.B spufs
mostly follow the standard behavior for regular system calls like
.BR read (2)
or
.BR write (2),
but often support only a subset of the operations
supported on regular filesystems.
This list details the supported
operations and the deviations from the standard behavior described
in the respective man pages.
.P
All files that support the
.BR read (2)
operation also support
.BR readv (2)
and all files that support the
.BR write (2)
operation also support
.BR writev (2).
All files support the
.BR access (2)
and
.BR stat (2)
family of operations, but for the latter call,
the only fields of the returned
.I stat
structure that contain reliable information are
.IR st_mode ,
.IR st_nlink ,
.IR st_uid ,
and
.IR st_gid .
.P
All files support the
.BR chmod (2)/\c
.BR fchmod (2)
and
.BR chown (2)/\c
.BR fchown (2)
operations, but will not be able to grant permissions that contradict
the possible operations (e.g., read access on the
.I wbox
file).
.P
The current set of files is:
.TP
.I /capabilities
Contains a comma-delimited string representing the capabilities of this
SPU context.
Possible capabilities are:
.RS
.TP
.B sched
This context may be scheduled.
.TP
.B step
This context can be run in single-step mode, for debugging.
.P
New capabilities flags may be added in the future.
.RE
.TP
.I /mem
the contents of the local storage memory of the SPU.
This can be accessed like a regular shared memory
file and contains both code and data in the address
space of the SPU.
The possible operations on an open
.I mem
file are:
.RS
.TP
.BR read (2)
.TQ
.BR pread (2)
.TQ
.BR write (2)
.TQ
.BR pwrite (2)
.TQ
.BR lseek (2)
These operate as usual, with the exception that
.BR lseek (2),
.BR write (2),
and
.BR pwrite (2)
are not supported beyond the end of the file.
The file size
is the size of the local storage of the SPU,
which is normally 256 kilobytes.
.TP
.BR mmap (2)
Mapping
.I mem
into the process address space provides access to the SPU local
storage within the process address space.
Only
.B MAP_SHARED
mappings are allowed.
.RE
.TP
.I /regs
Contains the saved general-purpose registers of the SPU context.
This file contains the 128-bit values of each register,
from register 0 to register 127, in order.
This allows the general-purpose registers to be
inspected for debugging.
.IP
Reading to or writing from this file requires that the context is
scheduled out, so use of this file is not recommended in normal
program operation.
.IP
The
.I regs
file is not present on contexts that have been created with the
.B SPU_CREATE_NOSCHED
flag.
.TP
.I /mbox
The first SPU-to-CPU communication mailbox.
This file is read-only and can be read in units of 4 bytes.
The file can be used only in nonblocking mode \- even
.BR poll (2)
cannot be used to block on this file.
The only possible operation on an open
.I mbox
file is:
.RS
.TP
.BR read (2)
If
.I count
is smaller than four,
.BR read (2)
returns \-1 and sets
.I errno
to
.BR EINVAL .
If there is no data available in the mailbox (i.e., the SPU has not
sent a mailbox message), the return value is set to \-1 and
.I errno
is set to
.BR EAGAIN .
When data
has been read successfully, four bytes are placed in
the data buffer and the value four is returned.
.RE
.TP
.I /ibox
The second SPU-to-CPU communication mailbox.
This file is similar to the first mailbox file, but can be read
in blocking I/O mode, thus calling
.BR read (2)
on an open
.I ibox
file will block until the SPU has written data to its interrupt mailbox
channel (unless the file has been opened with
.BR O_NONBLOCK ,
see below).
Also,
.BR poll (2)
and similar system calls can be used to monitor for the presence
of mailbox data.
.IP
The possible operations on an open
.I ibox
file are:
.RS
.TP
.BR read (2)
If
.I count
is smaller than four,
.BR read (2)
returns \-1 and sets
.I errno
to
.BR EINVAL .
If there is no data available in the mailbox and the file
descriptor has been opened with
.BR O_NONBLOCK ,
the return value is set to \-1 and
.I errno
is set to
.BR EAGAIN .
.IP
If there is no data available in the mailbox and the file
descriptor has been opened without
.BR O_NONBLOCK ,
the call will
block until the SPU writes to its interrupt mailbox channel.
When data has been read successfully, four bytes are placed in
the data buffer and the value four is returned.
.TP
.BR poll (2)
Poll on the
.I ibox
file returns
.I "(POLLIN | POLLRDNORM)"
whenever data is available for reading.
.RE
.TP
.I /wbox
The CPU-to-SPU communication mailbox.
It is write-only and can be written in units of four bytes.
If the mailbox is full,
.BR write (2)
will block, and
.BR poll (2)
can be used to block until the mailbox is available for writing again.
The possible operations on an open
.I wbox
file are:
.RS
.TP
.BR write (2)
If
.I count
is smaller than four,
.BR write (2)
returns \-1 and sets
.I errno
to
.BR EINVAL .
If there is no space available in the mailbox and the file
descriptor has been opened with
.BR O_NONBLOCK ,
the return
value is set to \-1 and
.I errno
is set to
.BR EAGAIN .
.IP
If there is no space available in the mailbox and the file
descriptor has been opened without
.BR O_NONBLOCK ,
the call will block until the SPU reads from its
PPE (PowerPC Processing Element)
mailbox channel.
When data has been written successfully,
the system call returns four as its function result.
.TP
.BR poll (2)
A poll on the
.I wbox
file returns
.I "(POLLOUT | POLLWRNORM)"
whenever space is available for writing.
.RE
.TP
.I /mbox_stat
.TQ
.I /ibox_stat
.TQ
.I /wbox_stat
These are read-only files that contain the length of the current
queue of each mailbox\[em]that is, how many words can be read from
.IR mbox " or " ibox
or how many words can be written to
.I wbox
without blocking.
The files can be read only in four-byte units and return
a big-endian binary integer number.
The only possible operation on an open
.I *box_stat
file is:
.RS
.TP
.BR read (2)
If
.I count
is smaller than four,
.BR read (2)
returns \-1 and sets
.I errno
to
.BR EINVAL .
Otherwise, a four-byte value is placed in the data buffer.
This value is the number of elements that can be read from (for
.I mbox_stat
and
.IR ibox_stat )
or written to (for
.IR wbox_stat )
the respective mailbox without blocking or returning an
.B EAGAIN
error.
.RE
.TP
.I /npc
.TQ
.I /decr
.TQ
.I /decr_status
.TQ
.I /spu_tag_mask
.TQ
.I /event_mask
.TQ
.I /event_status
.TQ
.I /srr0
.TQ
.I /lslr
Internal registers of the SPU.
These files contain an ASCII string
representing the hex value of the specified register.
Reads and writes on these
files (except for
.IR npc ,
see below) require that the SPU context be scheduled out,
so frequent access to
these files is not recommended for normal program operation.
.IP
The contents of these files are:
.RS
.TP 16
.I npc
Next Program Counter \- valid only when the SPU is in a stopped state.
.TP
.I decr
SPU Decrementer
.TP
.I decr_status
Decrementer Status
.TP
.I spu_tag_mask
MFC tag mask for SPU DMA
.TP
.I event_mask
Event mask for SPU interrupts
.TP
.I event_status
Number of SPU events pending (read-only)
.TP
.I srr0
Interrupt Return address register
.TP
.I lslr
Local Store Limit Register
.RE
.IP
The possible operations on these files are:
.RS
.TP
.BR read (2)
Reads the current register value.
If the register value is larger than the buffer passed to the
.BR read (2)
system call, subsequent reads will continue reading from the same
buffer, until the end of the buffer is reached.
.IP
When a complete string has been read, all subsequent read operations
will return zero bytes and a new file descriptor needs to be opened
to read a new value.
.TP
.BR write (2)
A
.BR write (2)
operation on the file sets the register to the
value given in the string.
The string is parsed from the beginning
until the first nonnumeric character or the end of the buffer.
Subsequent writes to the same file descriptor overwrite the
previous setting.
.IP
Except for the
.I npc
file, these files are not present on contexts that have been created with
the
.B SPU_CREATE_NOSCHED
flag.
.RE
.TP
.I /fpcr
This file provides access to the Floating Point Status and
Control Register (fcpr) as a binary, four-byte file.
The operations on the
.I fpcr
file are:
.RS
.TP
.BR read (2)
If
.I count
is smaller than four,
.BR read (2)
returns \-1 and sets
.I errno
to
.BR EINVAL .
Otherwise, a four-byte value is placed in the data buffer;
this is the current value of the
.I fpcr
register.
.TP
.BR write (2)
If
.I count
is smaller than four,
.BR write (2)
returns \-1 and sets
.I errno
to
.BR EINVAL .
Otherwise, a four-byte value is copied from the data buffer,
updating the value of the
.I fpcr
register.
.RE
.TP
.I /signal1
.TQ
.I /signal2
The files provide access to the two signal notification channels
of an SPU.
These are read-write files that operate on four-byte words.
Writing to one of these files triggers an interrupt on the SPU.
The value written to the signal files can
be read from the SPU through a channel read or from
host user space through the file.
After the value has been read by the SPU, it is reset to zero.
The possible operations on an open
.I signal1
or
.I signal2
file are:
.RS
.TP
.BR read (2)
If
.I count
is smaller than four,
.BR read (2)
returns \-1 and sets
.I errno
to
.BR EINVAL .
Otherwise, a four-byte value is placed in the data buffer;
this is the current value of the specified signal notification
register.
.TP
.BR write (2)
If
.I count
is smaller than four,
.BR write (2)
returns \-1 and sets
.I errno
to
.BR EINVAL .
Otherwise, a four-byte value is copied from the data buffer,
updating the value of the specified signal notification
register.
The signal notification register will either be replaced with
the input data or will be updated to the bitwise OR operation
of the old value and the input data, depending on the contents
of the
.I signal1_type
or
.I signal2_type
files respectively.
.RE
.TP
.I /signal1_type
.TQ
.I /signal2_type
These two files change the behavior of the
.I signal1
and
.I signal2
notification files.
They contain a numeric ASCII string which is read
as either "1" or "0".
In mode 0 (overwrite), the hardware replaces the contents
of the signal channel with the data that is written to it.
In mode 1 (logical OR), the hardware accumulates the bits
that are subsequently written to it.
The possible operations on an open
.I signal1_type
or
.I signal2_type
file are:
.RS
.TP
.BR read (2)
When the count supplied to the
.BR read (2)
call is shorter than the required length for the digit (plus a newline
character), subsequent reads from the same file descriptor will
complete the string.
When a complete string has been read, all subsequent read operations
will return zero bytes and a new file descriptor needs to be opened
to read the value again.
.TP
.BR write (2)
A
.BR write (2)
operation on the file sets the register to the
value given in the string.
The string is parsed from the beginning
until the first nonnumeric character or the end of the buffer.
Subsequent writes to the same file descriptor overwrite the
previous setting.
.RE
.TP
.I /mbox_info
.TQ
.I /ibox_info
.TQ
.I /wbox_info
.TQ
.I /dma_into
.TQ
.I /proxydma_info
Read-only files that contain the saved state of the SPU mailboxes and
DMA queues.
This allows the SPU status to be inspected, mainly for debugging.
The
.I mbox_info
and
.I ibox_info
files each contain the four-byte mailbox message that has been written
by the SPU.
If no message has been written to these mailboxes, then
contents of these files is undefined.
The
.IR mbox_stat ,
.IR ibox_stat ,
and
.I wbox_stat
files contain the available message count.
.IP
The
.I wbox_info
file contains an array of four-byte mailbox messages, which have been
sent to the SPU.
With current CBEA machines, the array is four items in
length, so up to 4 * 4 = 16 bytes can be read from this file.
If any mailbox queue entry is empty,
then the bytes read at the corresponding location are undefined.
.IP
The
.I dma_info
file contains the contents of the SPU MFC DMA queue, represented as the
following structure:
.IP
.in +4n
.EX
struct spu_dma_info {
    uint64_t         dma_info_type;
    uint64_t         dma_info_mask;
    uint64_t         dma_info_status;
    uint64_t         dma_info_stall_and_notify;
    uint64_t         dma_info_atomic_command_status;
    struct mfc_cq_sr dma_info_command_data[16];
};
.EE
.in
.IP
The last member of this data structure is the actual DMA queue,
containing 16 entries.
The
.I mfc_cq_sr
structure is defined as:
.IP
.in +4n
.EX
struct mfc_cq_sr {
    uint64_t mfc_cq_data0_RW;
    uint64_t mfc_cq_data1_RW;
    uint64_t mfc_cq_data2_RW;
    uint64_t mfc_cq_data3_RW;
};
.EE
.in
.IP
The
.I proxydma_info
file contains similar information, but describes the proxy DMA queue
(i.e., DMAs initiated by entities outside the SPU) instead.
The file is in the following format:
.IP
.in +4n
.EX
struct spu_proxydma_info {
    uint64_t         proxydma_info_type;
    uint64_t         proxydma_info_mask;
    uint64_t         proxydma_info_status;
    struct mfc_cq_sr proxydma_info_command_data[8];
};
.EE
.in
.IP
Accessing these files requires that the SPU context is scheduled out -
frequent use can be inefficient.
These files should not be used for normal program operation.
.IP
These files are not present on contexts that have been created with the
.B SPU_CREATE_NOSCHED
flag.
.TP
.I /cntl
This file provides access to the SPU Run Control and SPU status
registers, as an ASCII string.
The following operations are supported:
.RS
.TP
.BR read (2)
Reads from the
.I cntl
file will return an ASCII string with the hex
value of the SPU Status register.
.TP
.BR write (2)
Writes to the
.I cntl
file will set the context's SPU Run Control register.
.RE
.TP
.I /mfc
Provides access to the Memory Flow Controller of the SPU.
Reading from the file returns the contents of the
SPU's MFC Tag Status register, and
writing to the file initiates a DMA from the MFC.
The following operations are supported:
.RS
.TP
.BR write (2)
Writes to this file need to be in the format of a MFC DMA command,
defined as follows:
.IP
.in +4n
.EX
struct mfc_dma_command {
    int32_t  pad;    /* reserved */
    uint32_t lsa;    /* local storage address */
    uint64_t ea;     /* effective address */
    uint16_t size;   /* transfer size */
    uint16_t tag;    /* command tag */
    uint16_t class;  /* class ID */
    uint16_t cmd;    /* command opcode */
};
.EE
.in
.IP
Writes are required to be exactly
.I sizeof(struct mfc_dma_command)
bytes in size.
The command will be sent to the SPU's MFC proxy queue, and the
tag stored in the kernel (see below).
.TP
.BR read (2)
Reads the contents of the tag status register.
If the file is opened in blocking mode (i.e., without
.BR O_NONBLOCK ),
then the read will block until a
DMA tag (as performed by a previous write) is complete.
In nonblocking mode,
the MFC tag status register will be returned without waiting.
.TP
.BR poll (2)
Calling
.BR poll (2)
on the
.I mfc
file will block until a new DMA can be
started (by checking for
.BR POLLOUT )
or until a previously started DMA
(by checking for
.BR POLLIN )
has been completed.
.IP
.I /mss
Provides access to the MFC MultiSource Synchronization (MSS) facility.
By
.BR mmap (2)-ing
this file, processes can access the MSS area of the SPU.
.IP
The following operations are supported:
.TP
.BR mmap (2)
Mapping
.B mss
into the process address space gives access to the SPU MSS area
within the process address space.
Only
.B MAP_SHARED
mappings are allowed.
.RE
.TP
.I /psmap
Provides access to the whole problem-state mapping of the SPU.
Applications can use this area to interface to the SPU, rather than
writing to individual register files in
.BR spufs .
.IP
The following operations are supported:
.RS
.TP
.BR mmap (2)
Mapping
.B psmap
gives a process a direct map of the SPU problem state area.
Only
.B MAP_SHARED
mappings are supported.
.RE
.TP
.I /phys\-id
Read-only file containing the physical SPU number that the SPU context
is running on.
When the context is not running, this file contains the
string "\-1".
.IP
The physical SPU number is given by an ASCII hex string.
.TP
.I /object\-id
Allows applications to store (or retrieve) a single 64-bit ID into the
context.
This ID is later used by profiling tools to uniquely identify
the context.
.RS
.TP
.BR write (2)
By writing an ASCII hex value into this file, applications can set the
object ID of the SPU context.
Any previous value of the object ID is overwritten.
.TP
.BR read (2)
Reading this file gives an ASCII hex string representing the object ID
for this SPU context.
.RE
.SH EXAMPLES
To automatically
.BR mount (8)
the SPU filesystem when booting, at the location
.I /spu
chosen by the user, put this line into the
.BR fstab (5)
configuration file:
.EX
none /spu spufs gid=spu 0 0
.EE
.\" .SH AUTHORS
.\" Arnd Bergmann <arndb@de.ibm.com>, Mark Nutter <mnutter@us.ibm.com>,
.\" Ulrich Weigand <Ulrich.Weigand@de.ibm.com>, Jeremy Kerr <jk@ozlabs.org>
.SH SEE ALSO
.BR close (2),
.BR spu_create (2),
.BR spu_run (2),
.BR capabilities (7)
.P
.I The Cell Broadband Engine Architecture (CBEA) specification