/*
* Process packets associated with a workstream and protocol. For reasons of
* fairness, we process up to one complete netisr queue at a time, moving the
* queue to a stack-local queue for processing, but do not loop refreshing
* from the global queue. The caller is responsible for deciding whether to
* loop, and for setting the NWS_RUNNING flag. The passed workstream will be
* locked on entry and relocked before return, but will be released while
* processing. The number of packets processed is returned.
*/
static u_int
netisr_process_workstream_proto(struct netisr_workstream *nwsp, u_int proto)
{
struct netisr_work local_npw, *npwp;
u_int handled;
struct mbuf *m;
NETISR_LOCK_ASSERT();
NWS_LOCK_ASSERT(nwsp);
KASSERT(nwsp->nws_flags & NWS_RUNNING,
("%s(%u): not running", __func__, proto));
KASSERT(proto >= 0 && proto < NETISR_MAXPROT,
("%s(%u): invalid proto\n", __func__, proto));
npwp = &nwsp->nws_work[proto];
if (npwp->nw_len == 0)
return (0);
/*
* Move the global work queue to a thread-local work queue.
*
* Notice that this means the effective maximum length of the queue
* is actually twice that of the maximum queue length specified in
* the protocol registration call.
*/
handled = npwp->nw_len;
local_npw = *npwp;
npwp->nw_head = NULL;
npwp->nw_tail = NULL;
npwp->nw_len = 0;
nwsp->nws_pendingbits &= ~(1 << proto);
NWS_UNLOCK(nwsp);
while ((m = local_npw.nw_head) != NULL) {
local_npw.nw_head = m->m_hdr.mh_nextpkt;
m->m_hdr.mh_nextpkt = NULL;
if (local_npw.nw_head == NULL)
local_npw.nw_tail = NULL;
local_npw.nw_len--;
VNET_ASSERT(m->M_dat.MH.MH_pkthdr.rcvif != NULL,
("%s:%d rcvif == NULL: m=%p", __func__, __LINE__, m));
CURVNET_SET(m->M_dat.MH.MH_pkthdr.rcvif->if_vnet);
netisr_proto[proto].np_handler(m);
CURVNET_RESTORE();
}
KASSERT(local_npw.nw_len == 0,
("%s(%u): len %u", __func__, proto, local_npw.nw_len));
if (netisr_proto[proto].np_drainedcpu)
netisr_proto[proto].np_drainedcpu(nwsp->nws_cpu);
NWS_LOCK(nwsp);
npwp->nw_handled += handled;
return (handled);
}
/*
* Define the code needed before returning to user mode, for trap and
* syscall.
*/
void
userret(struct thread *td, struct trapframe *frame)
{
struct proc *p = td->td_proc;
CTR3(KTR_SYSC, "userret: thread %p (pid %d, %s)", td, p->p_pid,
td->td_name);
KASSERT((p->p_flag & P_WEXIT) == 0,
("Exiting process returns to usermode"));
#if 0
#ifdef DIAGNOSTIC
/* Check that we called signotify() enough. */
PROC_LOCK(p);
thread_lock(td);
if (SIGPENDING(td) && ((td->td_flags & TDF_NEEDSIGCHK) == 0 ||
(td->td_flags & TDF_ASTPENDING) == 0))
printf("failed to set signal flags properly for ast()\n");
thread_unlock(td);
PROC_UNLOCK(p);
#endif
#endif
#ifdef KTRACE
KTRUSERRET(td);
#endif
/*
* If this thread tickled GEOM, we need to wait for the giggling to
* stop before we return to userland
*/
if (td->td_pflags & TDP_GEOM)
g_waitidle();
/*
* Charge system time if profiling.
*/
if (p->p_flag & P_PROFIL)
addupc_task(td, TRAPF_PC(frame), td->td_pticks * psratio);
/*
* Let the scheduler adjust our priority etc.
*/
sched_userret(td);
KASSERT(td->td_locks == 0,
("userret: Returning with %d locks held.", td->td_locks));
#ifdef VIMAGE
/* Unfortunately td_vnet_lpush needs VNET_DEBUG. */
VNET_ASSERT(curvnet == NULL,
("%s: Returning on td %p (pid %d, %s) with vnet %p set in %s",
__func__, td, p->p_pid, td->td_name, curvnet,
(td->td_vnet_lpush != NULL) ? td->td_vnet_lpush : "N/A"));
#endif
#ifdef XEN
PT_UPDATES_FLUSH();
#endif
}
/*
* Define the code needed before returning to user mode, for trap and
* syscall.
*/
void
userret(struct thread *td, struct trapframe *frame)
{
struct proc *p = td->td_proc;
CTR3(KTR_SYSC, "userret: thread %p (pid %d, %s)", td, p->p_pid,
td->td_name);
KASSERT((p->p_flag & P_WEXIT) == 0,
("Exiting process returns to usermode"));
#if 0
#ifdef DIAGNOSTIC
/* Check that we called signotify() enough. */
PROC_LOCK(p);
thread_lock(td);
if (SIGPENDING(td) && ((td->td_flags & TDF_NEEDSIGCHK) == 0 ||
(td->td_flags & TDF_ASTPENDING) == 0))
printf("failed to set signal flags properly for ast()\n");
thread_unlock(td);
PROC_UNLOCK(p);
#endif
#endif
#ifdef KTRACE
KTRUSERRET(td);
#endif
/*
* If this thread tickled GEOM, we need to wait for the giggling to
* stop before we return to userland
*/
if (td->td_pflags & TDP_GEOM)
g_waitidle();
/*
* Charge system time if profiling.
*/
if (p->p_flag & P_PROFIL)
addupc_task(td, TRAPF_PC(frame), td->td_pticks * psratio);
/*
* Let the scheduler adjust our priority etc.
*/
sched_userret(td);
#ifdef XEN
PT_UPDATES_FLUSH();
#endif
/*
* Check for misbehavior.
*
* In case there is a callchain tracing ongoing because of
* hwpmc(4), skip the scheduler pinning check.
* hwpmc(4) subsystem, infact, will collect callchain informations
* at ast() checkpoint, which is past userret().
*/
WITNESS_WARN(WARN_PANIC, NULL, "userret: returning");
KASSERT(td->td_critnest == 0,
("userret: Returning in a critical section"));
KASSERT(td->td_locks == 0,
("userret: Returning with %d locks held", td->td_locks));
KASSERT(td->td_rw_rlocks == 0,
("userret: Returning with %d rwlocks held in read mode",
td->td_rw_rlocks));
KASSERT((td->td_pflags & TDP_NOFAULTING) == 0,
("userret: Returning with pagefaults disabled"));
KASSERT(td->td_no_sleeping == 0,
("userret: Returning with sleep disabled"));
KASSERT(td->td_pinned == 0 || (td->td_pflags & TDP_CALLCHAIN) != 0,
("userret: Returning with with pinned thread"));
KASSERT(td->td_vp_reserv == 0,
("userret: Returning while holding vnode reservation"));
KASSERT((td->td_flags & TDF_SBDRY) == 0,
("userret: Returning with stop signals deferred"));
#ifdef VIMAGE
/* Unfortunately td_vnet_lpush needs VNET_DEBUG. */
VNET_ASSERT(curvnet == NULL,
("%s: Returning on td %p (pid %d, %s) with vnet %p set in %s",
__func__, td, p->p_pid, td->td_name, curvnet,
(td->td_vnet_lpush != NULL) ? td->td_vnet_lpush : "N/A"));
#endif
#ifdef RACCT
PROC_LOCK(p);
while (p->p_throttled == 1)
msleep(p->p_racct, &p->p_mtx, 0, "racct", 0);
PROC_UNLOCK(p);
#endif
}
