// SPDX-License-Identifier: GPL-2.0-or-later
//
// A driver's periodic work. See ldls/kpi/kpi.h for the provenance rule.
//
// e1000 is the reason this exists, and the way it failed is worth keeping.
// The driver probed, opened, accepted frames from the stack, mapped them and
// wrote the transmit tail register -- and the wire stayed silent. Reading the
// chip's own registers said why:
//
//     TCTL=0x010000f8   PSP, collision threshold, retry-on-late-collision
//                       -- and bit 1, the enable, clear
//     RDT=0             no receive descriptors offered to the device
//
// Neither is set by `ndo_open`. e1000 enables its transmitter in
// `e1000_watchdog`, once it has seen the link come up, and refills the receive
// ring from the same place. The watchdog is a `delayed_work`, and LDLS was
// dropping every one on the floor: `queue_delayed_work_on` resolved to a stub
// that returned zero, which is the correct return value and does nothing.
//
// So a hosted driver needs somewhere for deferred work to actually run. There
// is no thread to run it on, which is the same constraint that made the NAPI
// poll a loop -- and the same answer: keep the function, and call it from the
// loop that was going to spin anyway.
//
// **A work item here runs on the main loop, not concurrently.** That is why
// there are no locks, and it is also the limitation: a driver whose work item
// blocks stops everything, including the transmit path it is probably waiting
// for.

#include "core/types.h"
#include "uapi/user.h"
#include "uapi/posix.h"
#include "kpi.h"

#define KPI_NWORK 16

static struct {
  void         *work;                  // struct work_struct *, our identity
  unsigned long due;                   // in jiffies
  int           armed;
} wq[KPI_NWORK];

static void
rdfn(void *work, void (**fn)(void *))
{
  unsigned char *p = (unsigned char *)work + kpi_l.work_func;
  int i;
  for (i = 0; i < 8; i++) ((unsigned char *)fn)[i] = p[i];
}

static int
slot_for(void *work)
{
  int i, free = -1;
  for (i = 0; i < KPI_NWORK; i++) {
    if (wq[i].armed && wq[i].work == work)
      return i;
    if (!wq[i].armed && free < 0)
      free = i;
  }
  return free;
}

static int
arm(void *work, unsigned long delay)
{
  int i;

  if (kpi_l.work_func == 0) {
    kpi_log("kpi: deferred work with no work_struct.func offset\n");
    return 0;
  }
  i = slot_for(work);
  if (i < 0) {
    kpi_log("kpi: too many deferred work items\n");
    return 0;
  }
  // Re-arming an item that is already queued moves its deadline, which is what
  // queue_delayed_work does *not* do -- it leaves the first deadline standing.
  // The difference only shows up as work running sooner than a driver asked,
  // and running a watchdog early is harmless in a way that running it late is
  // not.
  wq[i].work  = work;
  wq[i].due   = kpi_jiffies_now() + delay;
  wq[i].armed = 1;
  return 1;
}

// queue_work_on(cpu, wq, work)
int kpi_queue_work_on(int cpu, void *q, void *work)
{ (void)cpu; (void)q; return arm(work, 0); }

// queue_delayed_work_on(cpu, wq, dwork, delay)
//
// `struct delayed_work` begins with its `work_struct`, so the pointer is
// already the right one -- which is the only reason this does not need a
// second BTF offset.
int
kpi_queue_delayed_work_on(int cpu, void *q, void *dwork, unsigned long delay)
{ (void)cpu; (void)q; return arm(dwork, delay); }

int
kpi_cancel_work_sync(void *work)
{
  int i = slot_for(work);
  if (i >= 0 && wq[i].armed && wq[i].work == work) { wq[i].armed = 0; return 1; }
  return 0;
}

int kpi_cancel_delayed_work_sync(void *dwork) { return kpi_cancel_work_sync(dwork); }
int kpi_cancel_delayed_work(void *dwork)      { return kpi_cancel_work_sync(dwork); }
int kpi_flush_work(void *work)                { (void)work; return 0; }
void kpi_flush_workqueue(void *q)             { (void)q; }

// Run whatever is due. Returns how many ran, so the caller's loop can tell
// whether anything happened.
//
// The slot is disarmed *before* the call, because the overwhelming majority of
// periodic work in a driver ends by queueing itself again -- and disarming
// after would throw that away and stop the clock.
int
kpi_work_run(void)
{
  unsigned long now = kpi_jiffies_now();
  int i, n = 0;

  for (i = 0; i < KPI_NWORK; i++) {
    void (*fn)(void *) = 0;
    void *w;

    if (!wq[i].armed || (long)(now - wq[i].due) < 0)
      continue;
    w = wq[i].work;
    wq[i].armed = 0;
    rdfn(w, &fn);
    if (!fn)
      continue;
    fn(w);
    n++;
  }
  return n;
}