In some instances the rules governing alpha semaphores are
too strict to meet the needs of the kernel. Another class of semaphores
exist, known as beta semaphores.
 |
| |
|
 | NOTE: Unlike alpha semaphores, beta semaphores can be held
while a process sleeps. |
|
| |
|
Beta semaphores are created in the kernel by a call to b_initsema().
Beta semaphores have services similar to alpha semaphore services.
The following table describes the principal kernel interface routines
for beta semaphore operations.
Table 1-14 Interface
routines used for beta semaphore operations
| Routine | Purpose |
|---|
| b_initsema() | Create a beta semaphore,
add it to the hash table, link it to the global list of semaphores. |
| b_termsema() | Unlink beta semaphore from
hash chain, |
b_psema() | Acquire the semaphore and
possibly sleep if not available. Operative assertions: beta semaphore
is valid, no spinlocks are held, interrupts are disabled, not in
interrupt context. |
| b_cpsema() | Acquire semaphore and return
0 if available. If not available fail and return 1. Operative assertions:
Beta semaphore is valid, interrupts are enabledwhen not on the boot
path. |
| b_vsema() | Release the semaphore. Operative
assertions: beta semaphore is valid, interrupts are enabled, semaphore
is locked and allowed to unlock. |
| b_disowns_sema() | Returns true if current kthread
does not own the specified beta semaphore. |
Beta semaphores use a hash table to access the associated
spinlock and wait list information (linked list of kthreads).
A kthread
at the head of a semaphore's wait queue is allowed to be
awakened and yet miss the semaphore a maximum of BETA_MISS_LIMIT
times. Other executing code is allowed to acquire the semaphore
between the time the semaphore is unlocked by the V operation and
the time the awakened kthread
can execute and lock it. If the miss limit is reached, the semaphore
is passed to the waiting kthread.
The number of misses the kthread
at the head of the semaphore's wait queue has taken is
maintained in the kthread's
proc structure.
Each V operation on the semaphore will awaken the kthread
at the head of the wait queue and unlock the semaphore if the miss
limit has not been reached. If the miss limit is reached, the V
operation will awaken the kthread
at the end of the wait queue but will not unlock the semaphore,
preventing other code from acquiring the semaphore. The awaken kthread
notices that the semaphore ownership has been passed to it. This
is indicated by the miss count being equal to BETA_MISS_MAX.
Beta Semaphore Structures |
|
The beta semaphore itself contains only the lock and owner
information. Both beta semaphore and its hash table are defined
in sem_beta.h.
Beta Semaphore Type Definition |
|
The beta semaphore is defined as typedef b_sema_t
(also defined as vm_sema_t)
and consists of the three fields.
Table 1-15 elements
in struct b_sema
| Element | Purpose |
|---|
| b_lock | Lock state for the semaphore,
which may have one for the following values: 1 = SEM_LOCKED
(Semaphore is locked) 2 = SEM_WANT
(Semaphore is locked but a thread is waiting on it).
|
| b_order | Indicator of what order
the semaphore should be locking in. |
| *b_owner | Pointer to the kthread
structure of the thread of control claiming the semaphore. (This
is the only thread information in the beta semaphore structure.) |
Note, the b_lock
field is not a spinlock. The spinlock guarding the beta semaphore
is in the hash table.
Beta Semaphore Hash Table |
|
The address of the beta semaphore is indexed into the beta
semaphore hash table (bh_sema_t)
to obtain the spinlock and waiter information.
Table 1-16 elements
in struct bh_sema_t
| Element | Purpose |
|---|
| *beta_spinlock | Pointer to the spinlock
(type lock_t) |
| *fp, *bp | Pointers to the struct kthread
that comprise a wait list for the beta semaphore. The waiters are
linked together using the kthread.kt_wait_list
and kthread.kt_rwait_list fields
in the thread structure. |
| *link | Pointer to the link list of beta semaphores. |
![[Beta semaphore structures]](img/gfx9.gif)