devdrivslides11_7

Download Report

Transcript devdrivslides11_7

I/O, Devices & Device Drivers
I/O subsystem
• It is the responsibility of the I/O subsystem
of an OS to schedule, manage, & control
the interactions between processes and
devices.
• In a single/process user system if a program
requests access to an I/O device it gains control
of that device immediately, and the I/O
subsystem manages the transfer of data
• In a multiprogramming environment, many
processes can be requesting service from the
same device. The I/O subsystem is then
responsible for:
– Determining if the device is available
• If it is the operation is initiated and handled by the I/O
subsystem
– If the device is unavailable
• The request is preserved in a device queue
• When the device becomes available, the next pending
operation is initiated.
What is a device driver
• A device driver is software that manages
communications with a specific I/O device
or type of I/O device.
• Device drivers are one of the most
frequently written type of OS software
• Implementing device drivers can be
complex because of the characteristics of
the individual device
DEVICE
CONTROL OPERATIONS DATA TRANSFER OPERATIONS
Disk
Initialize
Read block
Seek to Sector, Track, Cyl.
Write block
Seek Home Position
Rewind
Forward Space Record
Magnetic Tape
Backspace Record
Read Record
Write Record
Read Backwards
Printer
Initialize
Load or select font
Select paper tray
Print character
Load font
Initialize
Terminal
**
Read Character
Write Character
Mouse
Read position
Read button status
** Control operations are normally invoked by special characters in the data stream
Figure 9--1: I/O Operations for Typical Devices
Character Devices
• These devices transfer one character at a
time between the device and the
computer.
– Keyboards
– Printers
Block Storage Devices
• Usually magnetic storage devices:
magnetic tape and disks.
• These are devices that transfer large
groups of bytes called blocks.
• Once an operation is started, it proceeds
without further intervention by the OS.
Magnetic tapes are a continuous strip of storage medium which
contains a # of parallel tracks. 8 bits of data are written in parallel on 8
tracks with a 9th track used for a parity bit. In this form, data may be
written densely, maybe up to thousands of bytes per block. However
and inch or more may be required between blocks for control
information.
Usefulness of tape is limited by it serial nature.
track
Arm
sector
Circular
base of
support
material on
which a
magnetic
film is
deposited.
Processors view of I/O
I/O
CPU
Device
Interface
Driver
controller
Data
Status
Control
Device
Separate and
protected,
accessed
through
special
instructions
ie: Inport and
Outport
Memory mapped I/O
Implementation strategies for handling devices
• Simple I/O
– AKA Programmed I/O
– Used on small and medium sized computers
to handle character devices
• Block transfer
– Used on small and medium sized computers
to handle block devices
• I/O processors – used on large computers
Polling .vs. Interrupts
• Once an I/O operation is initiated, the OS
must tract the operation until it is
complete.
• There are two techniques that can be used
to monitor device activity: Polling or
Interrupts
Simple Polling algorithm
/* initialize pointer and index */
ix=0;
count = BUFLEN;
/* begin polling loop */
while ((count > 0) && (no device error)) {
/* wait while device is busy */
while (device is busy) {}
/* process character if no error */
if (no device error) {
write buffer[ix];
/* write next character */
count = count - 1;
/* decrement count */
ix = ix + 1;
/* increment index */
}
}
• Using interrupts for device management:
– Starting I/O
– Returning to the process or OS
– Device generates an interrupt signal
Interrupts
MAIN PROGRAM:
/* initialize index, count, and device flag */
ix = 0;
count = BUFLEN;
dev_flag = 0;
...
enable interrupts for device and initiate operation
...
/* loop until all characters are written */
while (dev_flag == 0) {}
...
INTERRUPT HANDLER:
save registers
/* check for errors */
if (error) {
dev_flag = -1;
/* set flag to error code */
disable interrupts for device
restore registers
return from interrupt
}
Interrupts continued
/* check for completion */
if (count==0) { /* operation completed */
dev_flag = 1; /* set flag to completion code */
disable interrupts for device
restore registers
return from interrupt
}
/* no error, not complete; process character */
write buffer[ix];
/* write next character */
ix = ix + 1;
/* increment index */
count = count - 1; /* decrement count */
restore registers
return from interrupt
Buffering
• Buffering is a technique that can be used
to improve device as well as CPU
throughput
• Buffering is the use of temporary storage
areas in memory to store data that is read
from an input device before it is needed.
Process can also use buffering to store
data before it is sent to the output device
Buffering with character devices
Ring Buffer
Buffering with DMA devices and blocking
PROCESS
BUFFER1
CPU/PROCESS
Device
Driver
BUFFER2
READ
MORE
DATA
DISK1
CPU/PROCESS
BUFFER1
Read Last logical
record from buffer 1
BUFFER2
Start read from
device to fill buffer 2
DISK1
Software Caching
Data structures for Device
Management
• I/O operations that are in progress need to
be represented by some type of data
structures in every operating system
• This data structure can be called by
various names DCB, IOB, IOCB
• It represents a device, channel or
controller, and the activity of the device
Data structures for device management: I/OCB
Channel (port ) Number
Controller Address
Device Name
Device Address
Interrupt Vector Address
Address of Interrupt Handler
Device Type
Address of Open Procedure
Address of Close Procedure
Address of Start I/O Procedure
Address of Cancel I/O Procedure
Buffer Address
Buffer Length
Current Buffer Pointer
Current Data Count
Current I/O Operation
Address of PCB of Process which requested the Operation
Address of I/O Request Parameters
Address of ECB for Current Operation
Figure 10--15: Information Stored in an I/O Control Block & DCB
Device Driver Organization
• Every device driver & I/O supervisor must provide
services to support the basic activities for each device,
including:
–
–
–
–
–
Preparing for and starting I/O
Servicing interrupts
Completion of I/O
Cancellation of I/O
Error detection and recovery
• The I/O system call interface includes procedures to
service common operations such as:
– Open, close, read, write, and control operations
• Which when called for a specific device, invoke the
corresponding component of a device driver.
Device driver organization: structure and functions of device drivers
Figure 10--17: Structure and Functions of Device Drivers
Figure 10--18: Flow of Control in a Device Driver
Parameters
• Individual standards for system calls and device
drivers for a specific OS will determine how
parameters are passed to device drivers.
• Most systems create a parameter list, using a
register to point to the address of this list.
• Part of the responsibility of the I/O scheduler is
to access these parameters, validate them, and
make them available to later components of the
device driver.
Preparing for I/O
•
•
Preparation for I/O is the first activity of a device driver when invoked via one of the device system
calls.
I/O system call parameters must:
–
–
–
–
–
•
•
This component of a device driver validates device-specific parameters of an I/O request.
Appropriate error return codes must be provided to the user for any errors detected in the service
request parameters.
The preparation for I/O might include:
–
–
–
•
specify the type of operation requested (read, write, control),
the address of the data buffer to be used,
the size of the data area,
and other relevant information. (font, or paper tray)
Additionally, device-dependent parameters, such as track and sector numbers for disk operations, may also
be required.
temporary buffer allocation or initialization,
formatting of data,
placing information in the appropriate location(s) in some I/O control block that is accessible to the device
interrupt handler and the operating system.
Once the I/O request parameters are validated, the device status must be checked. If the device is
busy or not ready, the device driver must take an appropriate system dependent action.
–
–
If an error condition or problem such as "device not ready" is detected, the driver might take action causing
the application process to terminate, or it might simply provide a return code to be passed back to the
process, which then must deal with the problem in a suitable way depending on the application.
In a multiprogramming operating system, the response may include queuing the I/O request for a shared
device that is presently busy with another operation.
Starting I/O
DEVICE
REQUIRED I/O INFORMATION
Disk
Operation Code
Memory Transfer Address
Number of Bytes to Transfer
Track Address
Sector Address
Tape
Operation Code
Memory Transfer Address
Number of Bytes to Transfer
Printer
Terminal
Serial Device
Operation Code (read/write)
Character to Transfer
Timer
Time interval
Interrupt Servicing
•
•
The most complicated part of most I/O drivers or I/O supervisor modules is
the interrupt handler.
Much of the device control is embedded in the interrupt handler, which is
given control asynchronously when the device needs service.
– This could involve processing the next character for a device in the case of a
non-DMA device,
– starting a second or third phase of an I/O operation for DMA devices (for
example, a read after a seek to sector on some disks).
•
Information needed by the interrupt handler must be provided by the main
portion of the device driver. Such information is usually contained in the
control blocks that represent the current I/O operation, just as PCBs
represent process execution to the dispatcher.
– Device interrupt handlers must save all registers and hardware status of the
interrupted program and restoring it before returning to the interrupted program.
– Certainly the most complex part of interrupt handler service is error recovery
– Error recovery routines can be either resident or dynamically loaded when
needed, depending on operating system design and requirements.
Possible errors for devices
DEVICE
POSSIBLE ERRORS
Disk
Invalid Track, Sector
Wrong Density
Power Unsafe
Data error
Tape
Data error
End of tape
Printer
Paper out
Paper jam
Off Line
Sample Error Handler
ERROR HANDLER:
/* initialize counter and flag */
loop = 0;
tape_error = TRUE;
/* repeat entire process up to ten times */
while ((loop < 10) && tape_error) {
/* backspace and reread up to nine times */
count = 0;
while ((count < 9) && tape_error) {
backspace record
read record again
if (read is ok) tape_error = FALSE;
else count = count + 1;
}
/* if error still present, backup further */
if (tape_error) {
/* backspace ten records */
count = 0;
while (count < 10) {
backspace record
count = count + 1;
}
/* skip forward nine records */
count = 0;
while (count < 9) {
forward space record
count = count + 1;
}
/* try reading again */
read record
if (read is ok) tape_error = FALSE;
loop = loop + 1; /* prepare for retry */
} /* repeat if necessary */
I/O completion
• Once the requested operation is finished,
some cleanup must still occur
– Setting the status of the process requesting
the operation
– Clearing the device busy status, and disabling
interrupts
– Search a device queue, for the next operation
Scheduling I/O
process1
I/O
event
3
process2
I/O
event
2
Device or I/O queue
process3
I/O
event
1
Multiple I/O queues
Figure 10--23: A Single I/O Queue
Figure 10--24: Multiple I/O Queues
PCB 0
Running Process
Priority 60
Ready Q
PCB 8
IO_Init Q
PCB1
Printer
Write
Priority 40
PCB 2
PCB 9
Disk Read
PCB 3
Priority 32
IO_Active Q
DISK_IOCB
PCB6 Address
ECB Address
PRINTER_IOCB
Disk Write
PCB7 Address
PCB 4
ECB Address
Printer
Write
PCB 5
Disk Read
Modified figure 10-25
Running Process
Ready Q
PCB 0
IO_Init Q
PCB1
Printer
Write
IO_Active Q
DISK_IOCB
Priority 60
PCB 6
PCB 3
Disk Write
Priority 42
PCB 8
PCB 4
Printer
Write
PCB2 Address
ECB Address
PRINTER_IOCB
PCB7 Address
ECB Address
Priority 40
PCB 5
Disk Read
PCB 9
Priority32
Modified Figure 1--27
•
IO_COMPLETE:
/* Cleanup after disk operation,
switch waiting process to ready state */
if (operation is complete) {
Save context of current process in PCB
Insert PCB in ReadyQ
Set Disk ECB for disk I/O just completed
Move PCB address in Disk IOCB to ReadyQ
Set Disk IOCB to idle state
}
/* setup next disk operation, if any */
Search IO_WaitQ for another disk request
if (disk request found) {
Move PCB address of next disk I/O request to Disk IOCB
Move DIsk ECB address to Disk IOCB
Start requested disk I/O operation
}
Invoke dispatcher to dispatch next process
•
Figure 10--26: Handling an I/O Request Complete Interrupt