ovpxa24xx: Try to resume the device with minimal state changes after bad DMA (EEH)

drivers/dahdi/opvxa24xx/base.c

@@ -947,6 +947,9 @@ static void a24xx_proslic_check_hook(struct a24xx *wc, int card)
 	int hook;
 	struct a24xx_dev *wc_dev = &wc->dev;
 
+	if (wc_dev->eeh_triggered)
+		return;
+
 	/* For some reason we have to debounce the
 	   hook detector.  */
 
@@ -1096,6 +1099,9 @@ DAHDI_IRQ_HANDLER(a24xx_interrupt)
 	if (!ints)
 		return IRQ_NONE;
 
+	if (wc_dev->eeh_triggered)
+		return IRQ_RETVAL(1);
+
 	__opvx_a24xx_set_irqstatus(wc_dev, ints);    // clear interrupt register.
 			
 	for (x=0;x<wc_dev->max_cards; x++) {
@@ -1143,6 +1149,9 @@ DAHDI_IRQ_HANDLER(a24xx_interrupt)
 		}
 	}
 
+	if (wc_dev->eeh_triggered)
+		return IRQ_RETVAL(1);
+
 	if (ints & (1<<16)) {       /* it is our interrupts */
 		wc_dev->intcount++;
 		switch(ms_per_irq){
@@ -1331,7 +1340,10 @@ DAHDI_IRQ_HANDLER(a24xx_interrupt)
 				}
 				break;	
 		}
-		
+
+		if (wc_dev->eeh_triggered)
+			return IRQ_RETVAL(1);
+
 		if (!(wc_dev->intcount % ( 10000/ms_per_irq ))) {
 			/* Accept an alarm once per 10 seconds */
 			for (x=0;x<wc_dev->max_cards;x++)
@@ -1341,6 +1353,11 @@ DAHDI_IRQ_HANDLER(a24xx_interrupt)
 					}
 				}
 		}
+
+
+		if (wc_dev->eeh_triggered)
+			return IRQ_RETVAL(1);
+
 		/**/
 		for(order=0;order < ms_per_irq;order++){
 				a24xx_receive(wc, order);
@@ -1371,6 +1388,9 @@ static int interrupt_onecard_handler(struct a24xx *wc)
 						return IRQ_NONE;
 		}
 
+		if (wc_dev->eeh_triggered)
+			return 0;
+
 		__opvx_a24xx_set_irqstatus(wc_dev, ints);    // clear interrupt register.
 
 		for (x=0;x<wc_dev->max_cards; x++) {
@@ -1417,6 +1437,9 @@ static int interrupt_onecard_handler(struct a24xx *wc)
 			}
 		}
 
+		if (wc_dev->eeh_triggered)
+			return 0;
+
 		if ( ((ints & (1<<16)) && wc_dev->master) || (!wc_dev->master)) {       /* it is our interrupts */
 				wc_dev->intcount++;
 				switch(ms_per_irq){
@@ -1605,7 +1628,11 @@ static int interrupt_onecard_handler(struct a24xx *wc)
 						}
 						break;	
 				}
-		
+
+
+				if (wc_dev->eeh_triggered)
+					return 0;
+
 				if (!(wc_dev->intcount % ( 10000/ms_per_irq ))) {
 						/* Accept an alarm once per 10 seconds */
 						for (x=0;x<wc_dev->max_cards;x++)
@@ -1615,6 +1642,10 @@ static int interrupt_onecard_handler(struct a24xx *wc)
 								}
 						}
 				}
+
+				if (wc_dev->eeh_triggered)
+					return 0;
+
 				/**/
 				for(order=0;order < ms_per_irq;order++){
 						dahdi_transmit(&wc->span);
@@ -1747,6 +1778,10 @@ static int a24xx_ioctl(struct dahdi_chan *chan, unsigned int cmd, unsigned long
 	
 	struct fxs *const fxs = &wc_dev->mod[chan->chanpos - 1].fxs;
 	
+
+	if (wc_dev->eeh_triggered)
+		return -EFAULT;
+
 	switch (cmd) {
 	case DAHDI_ONHOOKTRANSFER:
 		if (wc_dev->modtype[chan->chanpos - 1] != MOD_TYPE_FXS) {
@@ -2288,6 +2323,7 @@ static int __devinit a24xx_init_one(struct pci_dev *pdev, const struct pci_devic
 		wc_dev = &wc->dev;
 		wc_dev->ledstate = 0;
 		wc_dev->device_id = *ent;
+		wc_dev->eeh_triggered = 0;
 		
 		for (x=0; x < sizeof(wc->chans)/sizeof(wc->chans[0]); ++x) {
 			wc->chans[x] = &wc->_chans[x];
@@ -2718,18 +2754,12 @@ static void __exit a24xx_cleanup(void)
  * Return value:
  * PCI_ERS_RESULT_NEED_RESET or PCI_ERS_RESULT_DISCONNECT
  */
-static pci_ers_result_t a24xx_pci_error_detected(struct pci_dev *pdev,
+static pci_ers_result_t a24xx_pci_error_detected(struct pci_dev __maybe_unused *pdev,
 					       pci_channel_state_t state)
 {
-	struct a24xx *wc = pci_get_drvdata(pdev);
-	struct a24xx_dev *wc_dev = &wc->dev;
-
 	if (state == pci_channel_io_perm_failure)
 		return PCI_ERS_RESULT_DISCONNECT;
 
-	__opvx_a24xx_disable_interrupts(wc_dev);
-	__opvx_a24xx_stop_dma(wc_dev);
-
 	printk(KERN_WARNING "OpenVox: Non-correctable non-fatal error reported, unfreezing...\n");
 	return PCI_ERS_RESULT_CAN_RECOVER;
 }
@@ -2751,19 +2781,17 @@ static pci_ers_result_t a24xx_pci_mmio_enabled(struct pci_dev __maybe_unused *pd
  *
  */
 
-static void a24xx_pci_resume(struct pci_dev *pdev)
+static void a24xx_pci_resume(struct pci_dev __maybe_unused *pdev)
 {
 	struct a24xx *wc = pci_get_drvdata(pdev);
 	struct a24xx_dev *wc_dev = &wc->dev;
 
 	printk(KERN_WARNING "OpenVox: Resuming operations after thaw...\n");
 
-	__opvx_a24xx_enable_interrupts(wc_dev);
-
-	/* Initialize Write/Buffers to all blank data */
-	memset((void *)wc_dev->writechunk,0, ms_per_irq * DAHDI_MAX_CHUNKSIZE * MAX_NUM_CARDS * 2 * 2);
+	wc_dev->eeh_triggered = 0;
 
-	__opvx_a24xx_start_dma(wc_dev, wc_dev->writedma);
+	/* Try to pick up where the driver left off */
+	interrupt_onecard_handler(wc);
 }
 
 /**

drivers/dahdi/opvxa24xx/base.h

@@ -127,6 +127,7 @@ enum battery_state {
 struct a24xx_dev {
 	struct pci_dev *dev;
 	struct pci_device_id device_id;
+	int eeh_triggered;
 	char *variety;
 	struct dahdi_device *ddev;
 	unsigned char ios;

drivers/dahdi/opvxa24xx/private.c

@@ -131,6 +131,7 @@ unsigned int __opvx_a24xx_getcreg(struct a24xx_dev *wc, unsigned int offset, uns
 	if (unlikely((*p) == ~((unsigned int)0))) {
 		struct eeh_dev *eehdev = pci_dev_to_eeh_dev(wc->dev);
 		eeh_dev_check_failure(eehdev);
+		wc->eeh_triggered = 1;
 	}
 #endif