darkkirb/old-cross-binutils
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity
old-cross-binutils/sim/ppc/device.maybe

877 lines
24 KiB
Text
Raw Normal View History

New simulator changes from Andrew
1996-07-23 15:42:42 +00:00
/* This file is part of the program psim.
Copyright (C) 1994-1996, Andrew Cagney <cagney@highland.com.au>
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program; if not, write to the Free Software
Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
*/
#ifndef _DEVICE_H_
#define _DEVICE_H_
#ifndef INLINE_DEVICE
#define INLINE_DEVICE
#endif
/* declared in basics.h, this object is used everywhere */
/* typedef struct _device device; */
/* Device templates:
*** THIS SECTION DESCRIBES HOW A DEVICE HAS A STATIC AND DYNAMIC
COMPONENT ** on the device in the tree is dynamic. *****
A device node is created from its template. The only valid
operation on a template is to create a device node from it: */
INLINE_DEVICE\
(device *) device_template_create_device
(device *parent,
const char *name,
const char *unit_address,
const char *args);
/* The create is paramaterized by both the devices unit address (a
string that is converted into numeric form by the devices parent)
and optionally extra argument information.
The actual device node is constructed by a number of pieces provided
by the template function: */
typedef struct _device_callbacks device_callbacks;
INLINE_DEVICE\
(device *) device_create_from
(const char *name,
const device_unit *unit_address,
void *data,
const device_callbacks *callbacks,
device *parent);
/* OpenBoot discusses the creation of packages (devices). */
/* Devices:
As with OpenBoot, all nodes in the device tree are considered to be
devices. Each node then has associated with it a number of methods
and properties (duscussed later).
OpenBoot documentation refers to devices, device nodes, packages,
package instances, methods, static methods and properties. This
device implementation uses its own termonology. Where ever it
exists, the notes will indicate a correspondance between PSIM terms
and those found in OpenBoot.
device:
A device is the basic building block in this model. A device can
be further categorized into one of three classes - template, node
and instance.
device-node (aka device):
The device tree is constructed from device-nodes. Each node has
both local state (data), a relationship with the device nodes
around it and an address (unit-address) on the parents bus `bus' */
INLINE_DEVICE\
(device *) device_parent
(device *me);
INLINE_DEVICE\
(device *) device_sibling
(device *me);
INLINE_DEVICE\
(device *) device_child
(device *me);
INLINE_DEVICE\
(const char *) device_name
(device *me);
INLINE_DEVICE\
(const char *) device_path
(device *me);
INLINE_DEVICE\
(void *) device_data
(device *me);
INLINE_DEVICE\
(psim *) device_system
(device *me);
typedef struct _device_unit {
int nr_cells;
unsigned32 cells[4]; /* unused cells are zero */
} device_unit;
INLINE_DEVICE\
(const device_unit *) device_unit_address
(device *me);
/* Each device-node normally corresponds to a hardware component of
the system being modeled. Leaf nodes matching external devices and
intermediate nodes matching bridges and controllers.
Device nodes also support methods that are an abstraction of the
transactions that occure in real hardware. These operations
(io/dma read/writes and interrupts) are discussed separatly.
OpenBoot refers to device nodes by many names. The most common are
device, device node and package. */
/* Properties:
In IEEE1275 many of the the characteristics of a device are stored
in the device tree as properties. Each property consists of a name
and an associated (implicitly typed) value. A device will have a
list of properties attached to it. The user is able to manipulate
the list, adding and removing properties and set/modify the value
of each property.
PSIM's device tree follows this model but with the addition of
strongly typing each property's value. The simulator will detect
at run time, the incorrect use of a property.
In addition to the standard use of properties, Both PSIM and
individual devices will use properties to record simulation
configuration information. For instance, a disk device might store
in a string property called <<file>> the name of the file that
contains the disk image to use. */
/* The following are valid property types. The property `array' is a
for generic untyped data. */
typedef enum {
array_property,
boolean_property,
ihandle_property,
integer_property,
string_property,
} device_property_type;
typedef struct _device_property device_property;
struct _device_property {
device *owner;
const char *name;
device_property_type type;
unsigned sizeof_array;
const void *array;
const device_property *original;
object_disposition disposition;
};
/* iterate through the properties attached to a device */
INLINE_DEVICE\
(const device_property *) device_next_property
(const device_property *previous);
INLINE_DEVICE\
(const device_property *) device_find_property
(device *me,
const char *property); /* NULL for first property */
/* Manipulate the properties belonging to a given device.
SET on the other hand will force the properties value. The
simulation is aborted if the property was present but of a
conflicting type.
FIND returns the specified properties value, aborting the
simulation if the property is missing. Code locating a property
should first check its type (using device_find_property above) and
then obtain its value using the below. */
INLINE_DEVICE\
(void) device_set_array_property
(device *me,
const char *property,
const void *array,
int sizeof_array);
INLINE_DEVICE\
(const device_property *) device_find_array_property
(device *me,
const char *property);
#if 0
INLINE_DEVICE\
(void) device_set_boolean_property
(device *me,
const char *property,
int bool);
#endif
INLINE_DEVICE\
(int) device_find_boolean_property
(device *me,
const char *property);
#if 0
INLINE_DEVICE\
(void) device_set_ihandle_property
(device *me,
const char *property,
device_instance *ihandle);
#endif
INLINE_DEVICE\
(device_instance *) device_find_ihandle_property
(device *me,
const char *property);
#if 0
INLINE_DEVICE\
(void) device_set_integer_property
(device *me,
const char *property,
signed_word integer);
#endif
INLINE_DEVICE\
(signed_word) device_find_integer_property
(device *me,
const char *property);
#if 0
INLINE_DEVICE\
(void) device_set_string_property
(device *me,
const char *property,
const char *string);
#endif
INLINE_DEVICE\
(const char *) device_find_string_property
(device *me,
const char *property);
/* Instances:
As with IEEE1275, a device can be opened, creating an instance.
Instances provide more abstract interfaces to the underlying
hardware. For example, the instance methods for a disk may include
code that is able to interpret file systems found on disks. Such
methods would there for allow the manipulation of files on the
disks file system. The operations would be implemented using the
basic block I/O model provided by the disk.
This model includes methods that faciliate the creation of device
instance and (should a given device support it) standard operations
on those instances. */
*** device-instance ***
Devices support an abstract I/O model. A unique I/O instance can be
created from a device node and then this instance used to perform
I/O that is independant of other instances. */
typedef struct _device_instance_callbacks device_instance_callbacks;
INLINE_DEVICE\
(device_instance *) device_create_instance_from
(device *me, /*OR*/ device_instance *parent,
void *data,
const char *path,
const char *args,
const device_instance_callbacks *callbacks);
INLINE_DEVICE\
(device_instance *) device_create_instance
(device *me,
const char *device_specifier);
INLINE_DEVICE\
(void) device_instance_delete
(device_instance *instance);
INLINE_DEVICE\
(int) device_instance_read
(device_instance *instance,
void *addr,
unsigned_word len);
INLINE_DEVICE\
(int) device_instance_write
(device_instance *instance,
const void *addr,
unsigned_word len);
INLINE_DEVICE\
(int) device_instance_seek
(device_instance *instance,
unsigned_word pos_hi,
unsigned_word pos_lo);
INLINE_DEVICE\
(unsigned_word) device_instance_claim
(device_instance *instance,
unsigned_word address,
unsigned_word length,
unsigned_word alignment);
INLINE_DEVICE\
(void) device_instance_release
(device_instance *instance,
unsigned_word address,
unsigned_word length);
INLINE_DEVICE\
(device *) device_instance_device
(device_instance *instance);
INLINE_DEVICE\
(const char *) device_instance_path
(device_instance *instance);
INLINE_DEVICE\
(void *) device_instance_data
(device_instance *instance);
/* A device instance can be marked (when created) as being permenant.
Such instances are assigned a reserved address and are *not*
deleted between simulation runs.
OpenBoot refers to a device instace as a package instance */
/* PIO:
*** DESCRIBE HERE WHAT A PIO OPERATION IS and how, broadly it is
modeled ****
During initialization, each device attaches its self to is parent
registering the address spaces that it is interested in:
a. The <<com>> device attaches its self to its parent <<phb>>
device at address <<0x3f8>> through to address <<0x3f8 + 16>>.
b. The <<phb>> has in turn attached its self to addresses
<<0xf0000000 .. 0xf0100000>>.
During the execution of the simulation propper, the following then
occure:
1. After any virtual to physical translation, the processor
passes the address to be read (or written to the core device).
(eg address 0xf00003f8).
2. The core device then looks up the specified addresses in its
address to device map, determines that in this case the address
belongs to the phb and passes it down.
3. The <<phb>> in turn determines that the address belongs to the
serial port and passes to that device the request for an access
to location <<0x3f8>>.
@figure mio
*/
/* Device Hardware
This model assumes that the data paths of the system being modeled
have a tree topology. That is, one or more processors sit at the
top of a tree. That tree containing leaf nodes (real devices) and
branch nodes (bridges).
For instance, consider the tree:
/pci # PCI-HOST bridge
/pci/pci1000,1@1 # A pci controller
/pci/isa8086 # PCI-ISA bridge
/pci/isa8086/fdc@300 # floppy disk controller on ISA bus
A processor needing to access the device fdc@300 on the ISA bus
would do so using a data path that goes through the pci-host bridge
(pci)and the isa-pci bridge (isa8086) to finally reach the device
fdc@300. As the data transfer passes through each intermediate
bridging node that bridge device is able to (just like with real
hardware) manipulate either the address or data involved in the
transfer. */
INLINE_DEVICE\
(unsigned) device_io_read_buffer
(device *me,
void *dest,
int space,
unsigned_word addr,
unsigned nr_bytes,
cpu *processor,
unsigned_word cia);
INLINE_DEVICE\
(unsigned) device_io_write_buffer
(device *me,
const void *source,
int space,
unsigned_word addr,
unsigned nr_bytes,
cpu *processor,
unsigned_word cia);
/* To avoid the need for an intermediate (bridging) node to ask each
of its child devices in turn if an IO access is intended for them,
parent nodes maintain a table mapping addresses directly to
specific devices. When a device is `connected' to its bus it
attaches its self to its parent. */
/* Address access attributes */
typedef enum _access_type {
access_invalid = 0,
access_read = 1,
access_write = 2,
access_read_write = 3,
access_exec = 4,
access_read_exec = 5,
access_write_exec = 6,
access_read_write_exec = 7,
} access_type;
/* Address attachement types */
typedef enum _attach_type {
attach_invalid,
attach_raw_memory,
attach_callback,
/* ... */
} attach_type;
INLINE_DEVICE\
(void) device_attach_address
(device *me,
const char *name,
attach_type attach,
int space,
unsigned_word addr,
unsigned nr_bytes,
access_type access,
device *who); /*callback/default*/
INLINE_DEVICE\
(void) device_detach_address
(device *me,
const char *name,
attach_type attach,
int space,
unsigned_word addr,
unsigned nr_bytes,
access_type access,
device *who); /*callback/default*/
/* where the attached address space can be any of
callback - all accesses to that range of addresses are past on to
the attached child device. The callback addresses are ordered
according to the callback level (attach_callback, .. + 1, .. + 2,
...). Lower levels are searched first. This facilitates the
implementation of more unusual addressing schema such as
subtractive decoding (as seen on the PCI bus). Within a given
callback level addresses must not overlap.
memory - the specified address space contains RAM, the node that is
having the ram attached is responsible for allocating space for and
maintaining that space. The device initiating the attach will not
be notified of accesses to such an attachement.
The memory attachment is very important. By giving the parent node
the responsability (and freedom) of managing the RAM, that node is
able to implement memory spaces more efficiently. For instance it
could `cache' accesses or merge adjacent memory areas.
In addition to I/O and DMA, devices interact with the rest of the
system via interrupts. Interrupts are discussed separatly. */
/* DMA:
*** DESCRIBE HERE WHAT A DMA OPERATION IS AND HOW IT IS MODELED,
include an interation of an access being reflected back down ***
*/
/* Conversly, the device pci1000,1@1 my need to perform a dma transfer
into the cpu/memory core. Just as I/O moves towards the leaves,
dma transfers move towards the core via the initiating devices
parent nodes. The root device (special) converts the DMA transfer
into reads/writes to memory */
INLINE_DEVICE\
(unsigned) device_dma_read_buffer
(device *me,
void *dest,
int space,
unsigned_word addr,
unsigned nr_bytes);
INLINE_DEVICE\
(unsigned) device_dma_write_buffer
(device *me,
const void *source,
int space,
unsigned_word addr,
unsigned nr_bytes,
int violate_read_only_section);
/* Interrupts:
*** DESCRIBE HERE THE INTERRUPT NETWORK ***
PSIM models interrupts and their wiring as a directed graph of
connections between interrupt sources and destinations. The source
and destination are both a tupple consisting of a port number and
device. Both multiple destinations attached to a single source and
multiple sources attached to a single destination are allowed.
When a device drives an interrupt port with multiple destinations a
broadcast of that interrupt event (message to all destinations)
occures. Each of those destination (device/port) are able to
further propogate the interrupt until it reaches its ultimate
destination.
Normally an interrupt source would be a model of a real device
(such as a keyboard) while an interrupt destination would be an
interrupt controller. The facility that allows an interrupt to be
delivered to multiple devices and to be propogated from device to
device was designed to support the requirements specified by
OpenPIC (ISA interrupts go to both OpenPIC and 8259), CHRP (8259
connected to OpenPIC) and hardware designs such as PCI-PCI
bridges. */
/* Interrupting a processor
The cpu object provides methods for delivering external interrupts
to a given processor.
The problem of synchronizing external interrupt delivery with the
execution of the cpu is handled internally by the processor object. */
/* Interrupt Source
A device drives its interrupt line using the call: */
INLINE_DEVICE\
(void) device_interrupt_event
(device *me,
int my_port,
int value,
cpu *processor,
unsigned_word cia);
/* This interrupt event will then be propogated to any attached
interrupt destinations.
Any interpretation of PORT and VALUE is model dependant. However
as guidelines the following are recommended: PCI interrupts a-d
correspond to lines 0-3; level sensative interrupts be requested
with a value of one and withdrawn with a value of 0; edge sensative
interrupts always have a value of 1, the event its self is treated
as the interrupt.
Interrupt Destinations
Attached to each interrupt line of a device can be zero or more
desitinations. These destinations consist of a device/port pair.
A destination is attached/detached to a device line using the
attach and detach calls. */
INLINE_DEVICE\
(void) device_interrupt_attach
(device *me,
int my_port,
device *dest,
int dest_port,
object_disposition disposition);
INLINE_DEVICE\
(void) device_interrupt_detach
(device *me,
int my_port,
device *dest,
int dest_port);
/* DESTINATION is attached (detached) to LINE of the device ME
Interrupt conversion
Users refer to interrupt port numbers symbolically. For instance a
device may refer to its `INT' signal which is internally
represented by port 3.
To convert to/from the symbolic and internal representation of a
port name/number. The following functions are available. */
INLINE_DEVICE\
(int) device_interrupt_decode
(device *me,
const char *symbolic_name);
INLINE_DEVICE\
(int) device_interrupt_encode
(device *me,
int port_number,
char *buf,
int sizeof_buf);
/* Initialization:
In PSIM, the device tree is created and then initialized in stages.
When using devices it is important to be clear what initialization
the simulator assumes is being performed during each of these
stages.
Firstly, each device is created in isolation (using the create from
template method). Only after it has been created will a device be
inserted into the tree ready for initialization.
Once the tree is created, it is initialized as follows:
1. All properties (apart from those containing instances)
are (re)initialized
2. Any interrupts addeded as part of the simulation run
are removed.
4. The initialize address method of each device (in top
down order) is called. At this stage the device
is expected to:
o Clear address maps and delete allocated memory
associated with the devices children.
o (Re)attach its own addresses to its parent device.
o Ensure that it is otherwize sufficiently
initialized such that it is ready for a
device instance create call.
5. All properties containing an instance of
a device are (re)initialized
6. The initialize data method for each device is called (in
top down) order. At this stage the device is expected to:
o Perform any needed data transfers. Such
transfers would include the initialization
of memory created during the address initialization
stage using DMA.
*/
INLINE_DEVICE\
(void) device_tree_init
(device *root,
psim *system);
/* IOCTL:
Very simply, a catch all for any thing that turns up that until now
either hasn't been thought of or doesn't justify an extra function. */
EXTERN_DEVICE\
(int) device_ioctl
(device *me,
cpu *processor,
unsigned_word cia,
...);
/* External communcation:
Devices interface to the external environment */
/* device_error() reports the problem to the console and aborts the
simulation. The error message is prefixed with the name of the
reporting device. */
EXTERN_DEVICE\
(void volatile) device_error
(device *me,
const char *fmt,
...) __attribute__ ((format (printf, 2, 3)));
/* Tree utilities:
In addition to the standard method of creating a device from a
device template, the following sortcuts can be used.
Create a device or property from a textual representation */
EXTERN_DEVICE\
(device *) device_tree_add_parsed
(device *current,
const char *fmt,
...) __attribute__ ((format (printf, 2, 3)));
/* where FMT,... once formatted (using vsprintf) is used to locate and
create either a device or property. Its syntax is almost identical
to that used in OpenBoot documentation - the only extension is in
allowing properties and their values to be specified vis:
"/pci/pci1000,1@1/disk@0,0"
Path:
The path to a device or property can either be absolute (leading
`/') or relative (leading `.' or `..'). Relative paths start from
the CURRENT node. The new current node is returned as the result.
In addition, a path may start with a leading alias (resolved by
looking in /aliases).
Device name:
<name> "@" <unit> [ ":" <args> ]
Where <name> is the name of the template device, <unit> is a
textual specification of the devices unit address (that is
converted into a numeric form by the devices parent) and <args> are
optional additional information to be passed to the device-template
when it creates the device.
Properties:
Properties are specified in a similar way to devices except that
the last element on the path (which would have been the device) is
the property name. This path is then followed by the property
value. Unlike OpenBoot, the property values in the device tree are
strongly typed.
String property:
<property-name> " " <text>
<property-name> " " "\"" <text>
Boolean property:
<property-name> " " [ "true" | "false" ]
Integer property or integer array property:
<property-name> " " <number> { <number> }
Phandle property:
<property-name> " " "&" <path-to-device>
Ihandle property:
<property-name> " " "*" <path-to-device-to-open>
Duplicate existing property:
<property-name> " " "!" <path-to-original-property>
In addition to properties, the wiring of interrupts can be
specified:
Attach interrupt <line> of <device> to <controller>:
<device> " " ">" <my-port> <dest-port> <dest-device>
Once created, a device tree can be traversed in various orders: */
typedef void (device_tree_traverse_function)
(device *device,
void *data);
INLINE_DEVICE\
(void) device_tree_traverse
(device *root,
device_tree_traverse_function *prefix,
device_tree_traverse_function *postfix,
void *data);
/* Or dumped out in a format that can be read back in using
device_add_parsed() */
INLINE_DEVICE\
(void) device_tree_print_device
(device *device,
void *ignore_data_argument);
/* Individual nodes can be located using */
INLINE_DEVICE\
(device *) device_tree_find_device
(device *root,
const char *path);
/* And the current list of devices can be listed */
INLINE_DEVICE\
(void) device_usage
(int verbose);
/* ihandles and phandles:
Both device nodes and device instances, in OpenBoot firmware have
an external representation (phandles and ihandles) and these values
are both stored in the device tree in property nodes and passed
between the client program and the simulator during emulation
calls.
To limit the potential risk associated with trusing `data' from the
client program, the following mapping operators `safely' convert
between the two representations: */
INLINE_DEVICE\
(device *) external_to_device
(device *tree_member,
unsigned32 phandle);
INLINE_DEVICE\
(unsigned32) device_to_external
(device *me);
INLINE_DEVICE\
(device_instance *) external_to_device_instance
(device *tree_member,
unsigned32 ihandle);
INLINE_DEVICE\
(unsigned32) device_instance_to_external
(device_instance *me);
#endif /* _DEVICE_H_ */
Reference in a new issue Copy permalink