lecture16-recap-questions-and-answers.pdf

CS5460/6460: Operating Systems
Lecture 16: Midterm recap, sample
questions
Anton Burtsev
February, 2014

Describe the x86 address translation pipeline
(draw figure), explain stages.

What is the linear address? What address is in
the registers, e.g., in %eax?

Logical and linear addresses
● Segment selector (16 bit) + offset (32 bit)

What segments do the following instructions
use? push, jump, mov

Describe the linear to physical address
translation with the paging mechanism (use
provided diagram, mark and explain the steps).

Page directory entry (PDE)
● 20 bit address of the page table
● Pages 4KB each, we need 1M to cover 4GB
● R/W – writes allowed?
● To a 4MB region controlled by this entry
● U/S – user/supervisor
● If 0 – user-mode access is not allowed
● A – accessed

Page table entry (PTE)
● 20 bit address of the 4KB page
● Pages 4KB each, we need 1M to cover 4GB
● R/W – writes allowed?
● To a 4KB page
● U/S – user/supervisor
● If 0 user-mode access is not allowed
● A – accessed
● D – dirty – software has written to this page

Describe the steps and data structures involved
into a user to kernel transition (draw diagrams)

What segment is specified in the interrupt
descriptor? Why?

● Interrupt gate disables
interrupts
● Clears the IF flag in
EFLAGS register
● Trap gate doesn't
● IF flag is unchanged

Which stack is used for execution of an
interrupt handler? How does hardware find it?

Why does xv6 uses 4MB pages for the first
page table during boot?

Describe organization of the memory allocator
in xv6?

Describe how a per-CPU variables can be
stored?

swtch in xv6 doesn’t explicitly save and restore
all ﬁelds of struct context. Why is it okay that
swtch doesn’t contain any code that saves
%eip?

Read copy update
● Goal: remove “cat” from the
list
● There might be some readers
of “cat”
● Idea: control the pointer
dereference
● Make it atomic

Read copy update (2)
● Remove “cat”
● Update the “boa” pointer
● All subsequent reader will get
“gnu” as boa->next

● Wait for all readers to finish
● synchronize_rcu()

● Readers finished
● Safe to deallocate “cat”

● New state of the list

Under what conditions RCU is a good idea?

In the following piece of code explain the use of
memory barriers?

Reference counting is a potential scalability
bottleneck, what can be done to improve it?

Reference counting is a potential scalability
bottleneck, what can be done to improve it?
● Sloppy counters

Why O(1) is really O(1)?
● Hint: analyze all operations and explain why
they are constant.

Alyssa runs xv6 on a machine with 8 processors and 8
processes. Each process calls sbrk (3451) continuously,
growing and shrinking its address space. Alyssa
measures the number of sbrks per second and notices
that 8 processes achieve the same total throughput as 1
process, even though each process runs on a different
processor. She proﬁles the xv6 kernel while running her
processes and notices that most execution time is spent
in kalloc (2838) and kfree (2815), though little is spent in
memset. Why is the throughput of 8 processes the same
as that of 1 process?

kalloc(void)
{
struct run *r;
if(kmem.use_lock)
acquire(&kmem.lock);
r = kmem.freelist;
if(r)
kmem.freelist = r− >next;
if(kmem.use_lock)
release(&kmem.lock);
return (char*)r;
}
kfree(char *v) {
struct run *r;
memset(v, 1, PGSIZE);
if(kmem.use_lock)
acquire(&kmem.lock);
r = (struct run*)v;
r− >next = kmem.freelist;
kmem.freelist = r;
if(kmem.use_lock)
release(&kmem.lock);
}

What can be done to improve performance?

Suppose you wanted to change the system call
interface in xv6 so that, instead of returning the
system call result in EAX, the kernel pushed the
result on to the user space stack. Fill in the
code below to implement this. For the purposes
of this question, you can assume that the user
stack pointer points to valid memory.

3374 void
3375 syscall(void)
3376 {
3377 int num;
3378
3379 num = proc− >tf− >eax;
3380 if(num > 0 && num < NELEM(syscalls) && syscalls[num]) {
3381 proc− >tf− >eax = syscalls[num]();
3382 } else {
3383 cprintf("%d %s: unknown sys call %dn",
3384 proc− >pid, proc− >name, num);
3385 proc− >tf− >eax = − 1;
3386 }
3387 }

3374 void
3375 syscall(void)
3376 {
3377 int num;
3378
3379 num = proc− >tf− >eax;
3380 if(num > 0 && num < NELEM(syscalls) && syscalls[num]) {
3381 // proc− >tf− >eax = syscalls[num]();
proc->tf->esp -= 4;
*(int*)ptoc->tf->esp = syscalls[num]();
3382 } else {
3383 cprintf("%d %s: unknown sys call %dn",
3384 proc− >pid, proc− >name, num);
3385 // proc− >tf− >eax = − 1;
proc->tf->esp -= 4;
*(int*)ptoc->tf->esp = -1;
3386 }
3387 }

1474 acquire(struct spinlock *lk)
1475 {
1476 pushcli();
1477 if(holding(lk))
1478 panic("acquire");
...
1483 while(xchg(&lk− >locked, 1) != 0)
1484 ;
...
1489 }
Why does acquire disable
interrupts?

1474 acquire(struct spinlock *lk)
1475 {
1476 pushcli();
1477 if(holding(lk))
1478 panic("acquire");
...
1483 while(xchg(&lk− >locked, 1) != 0)
1484 ;
...
1489 }
What would go wrong if you
replaced pushcli() with just cli(),
and popcli() with just sti()?

Explain why it would be awkward for xv6 to give
a process different data and stack segments
(i.e. have DS and SS refer to descriptors with
different BASE ﬁelds).

lecture16-recap-questions-and-answers.pdf

More Related Content

Similar to lecture16-recap-questions-and-answers.pdf (20)

Recently uploaded (20)

lecture16-recap-questions-and-answers.pdf