Upload
theresa-yolanda
View
47
Download
0
Embed Size (px)
DESCRIPTION
Interrupts in Protected-Mode. Writing a protected-mode interrupt-service routine for the timer-tick interrupt. Rationale. Usefulness of a general-purpose computer is dependent on its ability to interact with various peripheral devices attached to it (e.g., keyboard, display, disk-drives, etc.) - PowerPoint PPT Presentation
Citation preview
Interrupts in Protected-Mode
Writing a protected-mode interrupt-service routine for the
timer-tick interrupt
Rationale
• Usefulness of a general-purpose computer is dependent on its ability to interact with various peripheral devices attached to it (e.g., keyboard, display, disk-drives, etc.)
• Devices require a prompt response from the cpu when various events occur, even when the cpu is busy running a program
• The x86 interrupt-mechanism provides this
New requirements
• Unlike real-mode, where all code executes with full privileges (i.e., ring 0), protected-mode code usually is executed with some privilege restrictions
• Normally these restrictions prevent direct control of the peripheral devices
• Thus, when responding to an interrupt in protected-mode, a ring-transition (and its accompanying stack-switch) are needed
Interrupt-Gate Descriptors
start-offset[ 31..16 ]
code-segment selector start-offset[ 15..0 ]
gatetype
P 0DPL
Legend:
P=present (1=yes, 0=no) DPL=Descriptor Privilege-Level (0,1,2,3)code-selector (specifies memory-segment containing procedure code)start-offset (specifies the procedure’s entry-point within its code-segment)gate-types: 0x6 = 16bit Interrupt-Gate, 0x7 = 16-bit Trap-Gate
0xE = 32bit Interrupt-Gate, 0xF = 32-bit Trap-Gate
Trap-Gate vs. Interrupt-Gate
• The only distinction between a Trap-Gate and an Interrupt-Gate is in whether or not the CPU will automatically clear the IF-bit (Interrupt-Flag in EFLAGS register) as part of its response to an interrupt-request
• This is needed in cases where an Interrupt Service Routine executes outside ring0, so could not execute ‘cli’ or ‘sti’ instructions
16bit-Gate vs. 32bit-Gate
• The CPU constructs different stackframes for the 16-bit versus the 32-bit gate-types
ring0 stack
ring0 stack
IP
CS
FLAGS
SP
SS
EIP
EFLAGS
ESP
SS:SP SS:ESP
16-bits 32-bits
CS
SS
= always pushed = pushed if privilege-level changed
Return-from-Interrupt
• The programmer who writes an Interrupt Service Routine must know whether the Gate was 16-bit or 32-bit, in order to use the correct ‘interrupt-return’ instruction
• In a code-segment whose default-bit is 0 (i.e., USE16), the ‘iret’ instruction performs the correct return-actions for a 16-bit Gate
• Use ‘iretl’ for returning with a 32-bit Gate
Interrupt Descriptor Table
• The Gate-Descriptors for device interrupts form an array (called the IDT) and reside in a special system memory-segment
• The CPU will locate the IDT by referring to the value in its IDTR register (48-bits)
• A pair of special instructions exists which allow reading and writing this register:
sidt mem ; store IDTR into a memory-operand
lidt mem ; load IDTR from a memory-operand
Format of register IDTR
base_address[ 31..0 ] segment-limit[ 15..0 ]
0151647
16-bits32-bits
The instruction ‘lidt’ is privileged (can only be executed in ring 0),but the instruction ‘sidt’ is unprivileged (it can execute in any ring)
These features are analogous to the instructions ‘sgdt’ and ‘lgdt’used to store or to load GDTR (Global Descriptor Table Register)
Register relationships
GDTR
IDTR
Interrupt-gate
code-descriptor
Interrupt DescriptorTable (256 entries) ISR
code-segment
Global Descriptor Table
INT ID
Two Interrupt-Controllers
x86CPU
MasterPICSlave
PIC
INTR
Programmable Interval-TimerKeyboard controller
Each PIC has a Mask Register
IRQ7
IRQ6
IRQ5
IRQ4
IRQ3
IRQ2
IRQ1
IRQ0
Master PICInterrupt-mask(I/O-port 0x21)
IRQ15
IRQ14
IRQ13
IRQ12
IRQ11
IRQ10
IRQ9
IRQ8
Slave PICInterrupt-mask(I/O-port 0xA1)
If a mask-bit is 1, the corresponding device-interrupts are masked;If a mask-bit is 0, the corresponding device-interrupts are unmasked
Demo-program: ‘pmtimer.s’
• Let’s create a ‘protected-mode’ program that will handle the timer-tick interrupts
• Its ISR (Interrupt Service Routine) is very similar to the real-mode interrupt-handler
• It increments a 32-bit counter (at 40:6C), and it sets a flag (at 40:70) at midnight
• It decrements an 8-bit counter (at 40:40), and turns off diskette motors when zero
• It sends EOI-notification to Master PIC
Two ‘threads’ in our demo
Build descriptor-tablesEnter protected-mode (most interrupts masked)
Do for ten seconds:{ Read tick_countShow tick_count }
Leave protected-mode(most interrupts unmasked)
Exit to our ‘loader’
Increment tick_count(maybe set midnight_flag)
Decrement motor_count(maybe set motor_status)
Issue EOI command
‘main’ program-thread
Interrupt Service Routine
ROM-BIOS DATA (threads share access)
tick_count motor_status
motor_countmidnight_flag
read
write
Defining INT-8 Gate-Descriptor
0x0000
sel_CS isrPIT
0x8600
16-bit Interrupt-Gate
selector for ring0 code-segment
Our label for the interrupt-handler’s entry-point
P=1DPL=0 S=0type=6
Key Steps in the Demo
• Initialize the Descriptor-Table(s)
• Enter Protected-Mode with IF=0
• Load GDTR, IDTR and segment-selectors
• Mask all device-interrupts except timer
• Set IF=1 to enable unmasked interrupts
• Continuously show tick-count (for 10secs)
• Reset IF=0 to disable interrupts (for exit)
Question
• Why, in our earlier ‘tickdemo.s’ program, did we NOT need to set the PICs’ mask-registers so that only the timer-interrupts were left ‘unmasked’?
• (Your answer to this thought-question will offer an insight into your present grasp of the ‘real-mode’ versus ‘protected-mode’ system-component interrelationships)
Can’t use ‘anchor’ for this one
• We encounter a problem if we attempt to execute our timer-tick demo-programs on any of the anchor-cluster’s workstations
• These demos (‘tickdemo’ and ‘pmtimer’) need to be executed on a machine in our classroom (or in the CS 5th-floor Lab), so we can watch the display-screen directly, rather than relying on ‘telnet’ (via pyramid)
In-class exercise #1
• Try modifying the ‘pmtimer.s’ program so that it uses a 32-bit Interrupt-Gate for the timer interrupt (gate-type 0xE) instead of the 16-bit Interrupt-Gate (gate-type 0x6)
• NOTE: You will need to adjust the ‘iret’ opcode in your Interrupt Service Routine to accommodate the resulting changes in the layout of your ISR’s stackframe.
In-class Exercise #2
• Apply your knowledge of privilege-rings to modify this demo so that its main routine (i.e., ‘exec_timer_tick_demo’) executes at privilege-level 3 (instead of at ring 0)
• You will need to add more descriptors to the Global Descriptor Table (for code and stack at ring3), and also a Call-Gate when returning to ring0 from ring3 (for quitting)