spufs - SPU filesystem
The SPU filesystem is used on PowerPC machines that implement the Cell Broadband Engine Architecture in order to access Synergistic Processor Units (SPUs).
The filesystem provides a name space similar to POSIX shared memory or message queues. Users that have write permissions on the filesystem can use spu_create(2) to establish SPU contexts under the spufs root directory.
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.
Set the user owning the mount point; the default is 0 (root).
Set the group owning the mount point; the default is 0 (root).
Set the mode of the top-level directory in spufs, as an octal mode string. The default is 0775.
The files in spufs mostly follow the standard behavior for regular system calls like read(2) or 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.
All files that support the read(2) operation also
support readv(2) and all files that support the
write(2) operation also support
writev(2). All files support the
access(2) and stat(2) family of
operations, but for the latter call, the only fields of the returned
stat
structure that contain reliable information are
st_mode
, st_nlink
, st_uid
, and
st_gid
.
All files support the chmod(2)/
fchmod(2) and chown(2)/
fchown(2) operations, but will not be able to grant
permissions that contradict the possible operations (e.g., read access
on the wbox
file).
The current set of files is:
/capabilities
Contains a comma-delimited string representing the capabilities of this SPU context. Possible capabilities are:
This context may be scheduled.
This context can be run in single-step mode, for debugging.
New capabilities flags may be added in the future.
/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
mem
file are:
These operate as usual, with the exception that lseek(2), write(2), and 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.
Mapping mem
into the process address space provides access
to the SPU local storage within the process address space. Only
MAP_SHARED mappings are allowed.
/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.
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.
The regs
file is not present on contexts that have been
created with the SPU_CREATE_NOSCHED flag.
/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 poll(2) cannot be used to block
on this file. The only possible operation on an open mbox
file
is:
If count
is smaller than four, read(2)
returns -1 and sets errno
to 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 errno
is
set to EAGAIN. When data has been read successfully,
four bytes are placed in the data buffer and the value four is
returned.
/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 read(2) on an open ibox
file will
block until the SPU has written data to its interrupt mailbox channel
(unless the file has been opened with O_NONBLOCK, see
below). Also, poll(2) and similar system calls can be
used to monitor for the presence of mailbox data.
The possible operations on an open ibox
file are:
If count
is smaller than four, read(2)
returns -1 and sets errno
to EINVAL. If there
is no data available in the mailbox and the file descriptor has been
opened with O_NONBLOCK, the return value is set to -1
and errno
is set to EAGAIN.
If there is no data available in the mailbox and the file descriptor has been opened without 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.
Poll on the ibox
file returns (POLLIN | POLLRDNORM)
whenever data is available for reading.
/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,
write(2) will block, and poll(2) can
be used to block until the mailbox is available for writing again. The
possible operations on an open wbox
file are:
If count
is smaller than four, write(2)
returns -1 and sets errno
to EINVAL. If there
is no space available in the mailbox and the file descriptor has been
opened with O_NONBLOCK, the return value is set to -1
and errno
is set to EAGAIN.
If there is no space available in the mailbox and the file descriptor has been opened without 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.
A poll on the wbox
file returns (POLLOUT |
POLLWRNORM) whenever space is available for writing.
/mbox_stat
/ibox_stat
/wbox_stat
These are read-only files that contain the length of the current
queue of each mailbox—that is, how many words can be read from
mbox
or ibox
or how many words can be written to
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 *box_stat
file is:
If count
is smaller than four, read(2)
returns -1 and sets errno
to 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 mbox_stat
and
ibox_stat
) or written to (for wbox_stat
) the
respective mailbox without blocking or returning an
EAGAIN error.
/npc
/decr
/decr_status
/spu_tag_mask
/event_mask
/event_status
/srr0
/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 npc
, see below) require that the SPU
context be scheduled out, so frequent access to these files is not
recommended for normal program operation.
The contents of these files are:
npc
Next Program Counter - valid only when the SPU is in a stopped state.
decr
SPU Decrementer
decr_status
Decrementer Status
spu_tag_mask
MFC tag mask for SPU DMA
event_mask
Event mask for SPU interrupts
event_status
Number of SPU events pending (read-only)
srr0
Interrupt Return address register
lslr
Local Store Limit Register
The possible operations on these files are:
Reads the current register value. If the register value is larger than the buffer passed to the read(2) system call, subsequent reads will continue reading from the same buffer, until the end of the buffer is reached.
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.
A 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.
Except for the npc
file, these files are not present on
contexts that have been created with the
SPU_CREATE_NOSCHED flag.
/fpcr
This file provides access to the Floating Point Status and Control
Register (fcpr) as a binary, four-byte file. The operations on the
fpcr
file are:
If count
is smaller than four, read(2)
returns -1 and sets errno
to EINVAL.
Otherwise, a four-byte value is placed in the data buffer; this is the
current value of the fpcr
register.
If count
is smaller than four, write(2)
returns -1 and sets errno
to EINVAL.
Otherwise, a four-byte value is copied from the data buffer, updating
the value of the fpcr
register.
/signal1
/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 signal1
or signal2
file are:
If count
is smaller than four, read(2)
returns -1 and sets errno
to EINVAL.
Otherwise, a four-byte value is placed in the data buffer; this is the
current value of the specified signal notification register.
If count
is smaller than four, write(2)
returns -1 and sets errno
to 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 signal1_type
or
signal2_type
files respectively.
/signal1_type
/signal2_type
These two files change the behavior of the signal1
and
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
signal1_type
or signal2_type
file are:
When the count supplied to the 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.
A 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.
/mbox_info
/ibox_info
/wbox_info
/dma_into
/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 mbox_info
and 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 mbox_stat
, ibox_stat
,
and wbox_stat
files contain the available message count.
The 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.
The dma_info
file contains the contents of the SPU MFC DMA
queue, represented as the following structure:
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];
};
The last member of this data structure is the actual DMA queue,
containing 16 entries. The mfc_cq_sr
structure is defined
as:
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;
};
The 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:
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];
};
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.
These files are not present on contexts that have been created with the SPU_CREATE_NOSCHED flag.
/cntl
This file provides access to the SPU Run Control and SPU status registers, as an ASCII string. The following operations are supported:
/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:
Writes to this file need to be in the format of a MFC DMA command, defined as follows:
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 */
};
Writes are required to be exactly 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).
Reads the contents of the tag status register. If the file is opened in blocking mode (i.e., without 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.
Calling poll(2) on the mfc
file will block
until a new DMA can be started (by checking for
POLLOUT) or until a previously started DMA (by checking
for POLLIN) has been completed.
/mss
Provides access to the MFC MultiSource Synchronization
(MSS) facility. By mmap(2)-ing this file, processes can
access the MSS area of the SPU.
The following operations are supported:
Mapping mss into the process address space gives access to the SPU MSS area within the process address space. Only MAP_SHARED mappings are allowed.
/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 spufs.
The following operations are supported:
Mapping psmap gives a process a direct map of the SPU problem state area. Only MAP_SHARED mappings are supported.
/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".
The physical SPU number is given by an ASCII hex string.
/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.
close(2), spu_create(2), spu_run(2), capabilities(7)
The Cell Broadband Engine Architecture (CBEA) specification