irq: MSI first, and poll the INTx lines that never arrive

PCI drivers took their interrupt from INTERRUPT_LINE. On real hardware
that is the 8259's number, not the IOAPIC pin the device is wired to
(that is in ACPI _PRT, which needs AML), so nothing ever arrived. xHCI
is entirely interrupt-driven, so the USB keyboard and a USB root would
have hung even on a clean boot. Under QEMU the two numbers agree,
which is why it never showed.

- pci_irq_attach()/pci_irq_detach(): MSI, then MSI-X if the driver
  allows it, then INTx. The handler is registered before MSI is
  enabled, so no edge is lost. xHCI, e1000, r8169 and rtl8139 use it.
- Lines 24-29 are MSI vectors. The LAPIC error vector moves from 19,
  which is a PCI pin on Intel chipsets, to 30. Nothing is placed at or
  above vector 64 (T_SYSCALL).
- A storming line is masked and its handlers are polled instead of
  being left to eat the CPU. A level line that has had a handler for 2 s
  and never fired is polled every tick by a kernel thread.

Tested in QEMU q35 with the root on a USB stick: xHCI and e1000e on
MSI; then MSI disabled and INTx deliberately misrouted, where xHCI got
zero interrupts and the root still mounted through the poller. i440fx
INTx and the rtl8139 hand-over are unchanged.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HH7xukt4aaiGj1nQEAoQf1

kernel/drivers/bus/xhci/xhci.c

@@ -1169,7 +1169,7 @@ xhci_dump(void)
             rd32(x->rt, XHCI_IR0 + IR_IMAN), rd32(x->rt, XHCI_IR0 + IR_IMOD),
             rd32(x->rt, XHCI_IR0 + IR_ERSTSZ),
             rd64(x->rt, XHCI_IR0 + IR_ERDP), x->ev.deq, x->ev.cycle);
-    cprintf("  pci irq %d\n", x->pci->irq);
+    cprintf("  pci irq %d (%s)\n", x->pci->irq_line, pci_irq_kind(x->pci));
 
     for(p = 1; p <= x->maxports; p++){
       uint32 sc = rd32(x->op, XHCI_PORTSC(p));
@@ -1328,17 +1328,26 @@ xhci_probe(struct pci_dev *d)
   if(x->ndevices == 0)
     cprintf("xhci: no devices attached\n");
 
-  // Probing is over: from here a bulk transfer has a process behind it and
-  // can sleep rather than spin.
-  x->can_sleep = 1;
 
   // Now, and only now, turn interrupts on. The order is: claim the line
   // (which unmasks it and sets level/active-low), clear anything the
   // controller already considers pending so that it drives INTx low, and then
   // let it interrupt. The next event is a genuine low-to-high transition,
   // which is the only thing an interrupt controller can act on.
-  if(d->irq != 0xFF){
-    irq_register(d->irq, xhci_irq, x, "xhci", 1);   // PCI: level, active low
+  //
+  // MSI or MSI-X when the controller has either, which every xHCI does. On
+  // real hardware that is not a refinement: the INTx line the config space
+  // names is the 8259's number, the IOAPIC pin the controller is really on
+  // is a different one, and an xHCI waiting on INTx waited for ever -- the
+  // USB keyboard dead and a USB root never mounted.
+  //
+  // Probing is over: from here a bulk transfer has a process behind it and
+  // can sleep rather than spin -- but only if something will wake it. With
+  // no interrupt at all, transfers go on spinning as they did during probe.
+  if(pci_irq_attach(d, xhci_irq, x, "xhci", PCI_IRQF_MSIX) < 0){
+    cprintf("xhci: no interrupt; transfers will be polled\n");
+  } else {
+    x->can_sleep = 1;
 
     wr32(x->op, XHCI_USBSTS, STS_EINT);
     wr32(x->rt, XHCI_IR0 + IR_IMAN, IMAN_IP | IMAN_IE);

kernel/drivers/net/e1000/e1000.c

@@ -359,19 +359,20 @@ e1000_probe(struct pci_dev *d)
   wr(e, E1000_RCTL, E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_SECRC |
                     E1000_RCTL_SZ_2048 | E1000_RCTL_LBM_NONE);
 
-  if (d->irq != 0xFF)
-    // level_low = 1: this is a PCI interrupt, which is level-triggered and
-    // active low. Registering it as edge-triggered makes it fire once and then
-    // wedge, because the device holds the line asserted and no further edge
-    // ever arrives. See the comment on irq_register in include/irq.h.
-    irq_register(d->irq, e1000_intr, e, "e1000", 1);
+  // MSI where the part has it -- the 82574 and the rest of the PCIe
+  // generation do -- and INTx otherwise. MSI-X is not offered: the 82574 in
+  // MSI-X mode routes its causes through IVAR, which this driver does not
+  // program, and its interrupts would go nowhere. If neither works the
+  // network thread still collects frames on its tick.
+  pci_irq_attach(d, e1000_intr, e, "e1000", 0);
   wr(e, E1000_IMS, E1000_ICR_RXT0 | E1000_ICR_RXDMT0 | E1000_ICR_LSC);
 
   e->ready = 1;
 
-  kinfo("e1000: %x:%x:%x:%x:%x:%x, link %s, irq %d\n",
+  kinfo("e1000: %x:%x:%x:%x:%x:%x, link %s, irq %d (%s)\n",
           e->mac[0], e->mac[1], e->mac[2], e->mac[3], e->mac[4], e->mac[5],
-          (rd(e, E1000_STATUS) & E1000_STATUS_LU) ? "up" : "down", d->irq);
+          (rd(e, E1000_STATUS) & E1000_STATUS_LU) ? "up" : "down",
+          d->irq_line, pci_irq_kind(d));
   return 0;
 }
 
@@ -458,8 +459,7 @@ e1000_release(struct pci_dev *d)
   wr(e, E1000_TCTL, 0);                    // stop transmitting
   rd(e, E1000_STATUS);                     // post the writes
 
-  if (d->irq != 0xFF)
-    irq_unregister(d->irq, e1000_intr, e);
+  pci_irq_detach(d, e1000_intr, e);
 
   // And take the interface out of the stack.
   //

kernel/drivers/net/r8169/r8169.c

@@ -468,9 +468,9 @@ r8169_probe(struct pci_dev *d)
   wr16(r, R_IMR, ISR_ROK | ISR_RER | ISR_TOK | ISR_TER |
                  ISR_RDU | ISR_LINKCHG);
 
-  if (d->irq != 0xFF && d->irq < 24)
-    irq_register(d->irq, r8169_intr, r, "r8169", 1);
-  else
+  // MSI first: every 8111 has it, and on the boards these are soldered to it
+  // is the only interrupt that is sure to arrive. See pci_irq_attach().
+  if (pci_irq_attach(d, r8169_intr, r, "r8169", 0) < 0)
     cprintf("r8169: no usable interrupt line; the link will be polled\n");
 
   r->ready = 1;
@@ -487,10 +487,10 @@ r8169_probe(struct pci_dev *d)
     uint32 xid = (rd32(r, R_TCR) >> 20) & 0xFCF;
     int    bmsr = mii_read(r, 1);
 
-    cprintf("r8169: xid %x, %x:%x:%x:%x:%x:%x, irq %d, %s (bmsr %x)\n",
+    cprintf("r8169: xid %x, %x:%x:%x:%x:%x:%x, irq %d (%s), %s (bmsr %x)\n",
             xid,
             r->mac[0], r->mac[1], r->mac[2], r->mac[3], r->mac[4], r->mac[5],
-            d->irq,
+            d->irq_line, pci_irq_kind(d),
             (rd8(r, R_PHYSTATUS) & PHYSTATUS_LINK) ? "link up" : "link down",
             bmsr);
   }

kernel/drivers/net/rtl8139/rtl8139.c

@@ -360,8 +360,7 @@ rtl8139_probe(struct pci_dev *d)
   wr16(r, R_IMR, ISR_ROK | ISR_RER | ISR_TOK | ISR_TER |
                  ISR_RXOVW | ISR_FOVW | ISR_LINKCHG);
 
-  if (d->irq != 0xFF && d->irq < 24)
-    irq_register(d->irq, rtl8139_intr, r, "rtl8139", 1);
+  pci_irq_attach(d, rtl8139_intr, r, "rtl8139", 0);   // no MSI on this part
 
   r->ready = 1;
   kinfo("rtl8139: %x:%x:%x:%x:%x:%x, %s, irq %d\n",
@@ -405,8 +404,7 @@ rtl8139_release(struct pci_dev *d)
   wr8(r, R_CR, 0);                // receiver and transmitter off
   release(&r->lock);
 
-  if (d->irq != 0xFF && d->irq < 24)
-    irq_unregister(d->irq, rtl8139_intr, r);
+  pci_irq_detach(d, rtl8139_intr, r);
 
   // Reset, and wait for it, so the chip has truly stopped before its buffers
   // are reused. The reset bit clears itself when the chip is done.