xv6/README-PSU at master · LongLeonardoLe/xv6 · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116

Changes made to xv6 to facilitate projects at Portland State University.

1. Access to the ''ticks'' global has been made atomic. This is necessary
to avoid deadlock with regards to tickslock/ptable.lock. Since Intel
has atomic guarantees and GNU gcc has a suitable alignment syntax, we
are able to do away with the tickslock everywhere but in the sleep()
system call (sleep() in proc.c invoked from sys_sleep() in sysproc.c).
The following changes were required.

1a. The definition for ticks was made 32-bit aligned in trap.c.
// set alignment to 32-bit for ticks. See Intel® 64 and IA-32 Architectures
// Software Developer’s Manual, Vol 3A, 8.1.1 Guaranteed Atomic Operations.
uint ticks __attribute__ ((aligned (4)));

1b. The code for handling the timer interrupt was changed in trap.c. Note
that the code change from the original for the ticklock is no longer
necessary.

  switch(tf->trapno){
  case T_IRQ0 + IRQ_TIMER:
    if(cpu->id == 0){
      acquire(&tickslock);
      ticks++;
      release(&tickslock);    // NOTE: MarkM has reversed these two lines.
      wakeup(&ticks);         // wakeup() does not require the tickslock to be held
    }

changed to:

  switch(tf->trapno){
  case T_IRQ0 + IRQ_TIMER:
    if(cpu->id == 0){
      atom_inc((int *)&ticks);
      wakeup(&ticks);
    }

1c. The file x86.c had several routines added. The atom_inc() was added for
atomic increment for ticks variable.

// Routines added for CS333
// atom_inc() added to simplify handling of ticks global
static inline void
atom_inc(volatile int *num)
{
  asm volatile ( "lock incl %0" : "=m" (*num));
}

// hlt() added by Noah Zentzis, Fall 2016.
static inline void
hlt()
{
  asm volatile("hlt");
}

static inline void
lock_inc(uint* mem)
{
  asm volatile("lock incl %0" : "=m" (mem));
}

static inline void
lock_dec(uint* mem)
{
  asm volatile("lock decl %0" : "=m" (mem));
}

static inline void
lock_add(uint* mem, uint n)
{
  asm volatile("lock add %0, %1" : "=m" (mem) : "d" (n));
}

// end of CS333 added routines

1d. Removed tickslock acquisition/free in sys_uptime (sysproc.c) since
reading is now atomic.

2. A new feature was added to the scheduler to turn off for
the remainder of the time slice when no work is in the ready queue.
proc.c in routine scheduler(). The hlt() routine is in x86.c.

  int idle;  // for checking if processor is idle

...

    idle = 1;  // assume idle unless we schedule a process

...

      idle = 0;  // not idle this timeslice

...

    release(&ptable.lock);
    // if idle, wait for next interrupt
    if (idle) {
      sti();
      hlt();
    }

3. A new user command, halt, was created to shutdown QEMU.  This is
distinct from the hlt() function. See sys_halt() in sysproc.c.

4. All uses of tickslock (e.g., sys_sleep()) have been removed. This
necessitated surgery on sleep() in pro.c. Calling cleep with a NULL
lock is now legal but must be limited to sys_sleep(). CAUTION

5. A general cleanup of the Makefile to be less confusing and more
specific to PSU.

6. The array states[] in proc.c was moved out of procdump() to be file
global. This helps with implementing the ps command in one of the projects.

7. A test program for the date() system call is given. The test program, date.c,
won't compile until the system call is implemented.