thunderbolt: Add MSI-X support

Intel Thunderbolt controllers support up to 16 MSI-X vectors. Using
MSI-X is preferred over MSI or legacy interrupt and may bring additional
performance because there is no need to check the status registers which
interrupt was triggered.

While there we convert comments in structs tb_ring and tb_nhi to follow
kernel-doc format more closely.

This code is based on the work done by Amir Levy and Michael Jamet.

Signed-off-by: Michael Jamet <michael.jamet@intel.com>
Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com>
Reviewed-by: Yehezkel Bernat <yehezkel.bernat@intel.com>
Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Andreas Noever <andreas.noever@gmail.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

1 files changed, 143 insertions, 22 deletions

diff --git a/drivers/thunderbolt/nhi.c b/drivers/thunderbolt/nhi.c
index a8c20413dbda..ed75c49748f5 100644
--- a/drivers/thunderbolt/nhi.c
+++ b/drivers/thunderbolt/nhi.c
@@ -21,6 +21,12 @@
 
 #define RING_TYPE(ring) ((ring)->is_tx ? "TX ring" : "RX ring")
 
+/*
+ * Minimal number of vectors when we use MSI-X. Two for control channel
+ * Rx/Tx and the rest four are for cross domain DMA paths.
+ */
+#define MSIX_MIN_VECS		6
+#define MSIX_MAX_VECS		16
 
 static int ring_interrupt_index(struct tb_ring *ring)
 {
@@ -42,6 +48,37 @@ static void ring_interrupt_active(struct tb_ring *ring, bool active)
 	int bit = ring_interrupt_index(ring) & 31;
 	int mask = 1 << bit;
 	u32 old, new;
+
+	if (ring->irq > 0) {
+		u32 step, shift, ivr, misc;
+		void __iomem *ivr_base;
+		int index;
+
+		if (ring->is_tx)
+			index = ring->hop;
+		else
+			index = ring->hop + ring->nhi->hop_count;
+
+		/*
+		 * Ask the hardware to clear interrupt status bits automatically
+		 * since we already know which interrupt was triggered.
+		 */
+		misc = ioread32(ring->nhi->iobase + REG_DMA_MISC);
+		if (!(misc & REG_DMA_MISC_INT_AUTO_CLEAR)) {
+			misc |= REG_DMA_MISC_INT_AUTO_CLEAR;
+			iowrite32(misc, ring->nhi->iobase + REG_DMA_MISC);
+		}
+
+		ivr_base = ring->nhi->iobase + REG_INT_VEC_ALLOC_BASE;
+		step = index / REG_INT_VEC_ALLOC_REGS * REG_INT_VEC_ALLOC_BITS;
+		shift = index % REG_INT_VEC_ALLOC_REGS * REG_INT_VEC_ALLOC_BITS;
+		ivr = ioread32(ivr_base + step);
+		ivr &= ~(REG_INT_VEC_ALLOC_MASK << shift);
+		if (active)
+			ivr |= ring->vector << shift;
+		iowrite32(ivr, ivr_base + step);
+	}
+
 	old = ioread32(ring->nhi->iobase + reg);
 	if (active)
 		new = old | mask;
@@ -239,8 +276,50 @@ int __ring_enqueue(struct tb_ring *ring, struct ring_frame *frame)
 	return ret;
 }
 
+static irqreturn_t ring_msix(int irq, void *data)
+{
+	struct tb_ring *ring = data;
+
+	schedule_work(&ring->work);
+	return IRQ_HANDLED;
+}
+
+static int ring_request_msix(struct tb_ring *ring, bool no_suspend)
+{
+	struct tb_nhi *nhi = ring->nhi;
+	unsigned long irqflags;
+	int ret;
+
+	if (!nhi->pdev->msix_enabled)
+		return 0;
+
+	ret = ida_simple_get(&nhi->msix_ida, 0, MSIX_MAX_VECS, GFP_KERNEL);
+	if (ret < 0)
+		return ret;
+
+	ring->vector = ret;
+
+	ring->irq = pci_irq_vector(ring->nhi->pdev, ring->vector);
+	if (ring->irq < 0)
+		return ring->irq;
+
+	irqflags = no_suspend ? IRQF_NO_SUSPEND : 0;
+	return request_irq(ring->irq, ring_msix, irqflags, "thunderbolt", ring);
+}
+
+static void ring_release_msix(struct tb_ring *ring)
+{
+	if (ring->irq <= 0)
+		return;
+
+	free_irq(ring->irq, ring);
+	ida_simple_remove(&ring->nhi->msix_ida, ring->vector);
+	ring->vector = 0;
+	ring->irq = 0;
+}
+
 static struct tb_ring *ring_alloc(struct tb_nhi *nhi, u32 hop, int size,
-				  bool transmit)
+				  bool transmit, unsigned int flags)
 {
 	struct tb_ring *ring = NULL;
 	dev_info(&nhi->pdev->dev, "allocating %s ring %d of size %d\n",
@@ -271,9 +350,14 @@ static struct tb_ring *ring_alloc(struct tb_nhi *nhi, u32 hop, int size,
 	ring->hop = hop;
 	ring->is_tx = transmit;
 	ring->size = size;
+	ring->flags = flags;
 	ring->head = 0;
 	ring->tail = 0;
 	ring->running = false;
+
+	if (ring_request_msix(ring, flags & RING_FLAG_NO_SUSPEND))
+		goto err;
+
 	ring->descriptors = dma_alloc_coherent(&ring->nhi->pdev->dev,
 			size * sizeof(*ring->descriptors),
 			&ring->descriptors_dma, GFP_KERNEL | __GFP_ZERO);
@@ -295,14 +379,16 @@ err:
 	return NULL;
 }
 
-struct tb_ring *ring_alloc_tx(struct tb_nhi *nhi, int hop, int size)
+struct tb_ring *ring_alloc_tx(struct tb_nhi *nhi, int hop, int size,
+			      unsigned int flags)
 {
-	return ring_alloc(nhi, hop, size, true);
+	return ring_alloc(nhi, hop, size, true, flags);
 }
 
-struct tb_ring *ring_alloc_rx(struct tb_nhi *nhi, int hop, int size)
+struct tb_ring *ring_alloc_rx(struct tb_nhi *nhi, int hop, int size,
+			      unsigned int flags)
 {
-	return ring_alloc(nhi, hop, size, false);
+	return ring_alloc(nhi, hop, size, false, flags);
 }
 
 /**
@@ -413,6 +499,8 @@ void ring_free(struct tb_ring *ring)
 			 RING_TYPE(ring), ring->hop);
 	}
 
+	ring_release_msix(ring);
+
 	dma_free_coherent(&ring->nhi->pdev->dev,
 			  ring->size * sizeof(*ring->descriptors),
 			  ring->descriptors, ring->descriptors_dma);
@@ -428,9 +516,9 @@ void ring_free(struct tb_ring *ring)
 
 	mutex_unlock(&ring->nhi->lock);
 	/**
-	 * ring->work can no longer be scheduled (it is scheduled only by
-	 * nhi_interrupt_work and ring_stop). Wait for it to finish before
-	 * freeing the ring.
+	 * ring->work can no longer be scheduled (it is scheduled only
+	 * by nhi_interrupt_work, ring_stop and ring_msix). Wait for it
+	 * to finish before freeing the ring.
 	 */
 	flush_work(&ring->work);
 	mutex_destroy(&ring->lock);
@@ -528,9 +616,52 @@ static void nhi_shutdown(struct tb_nhi *nhi)
 	 * We have to release the irq before calling flush_work. Otherwise an
 	 * already executing IRQ handler could call schedule_work again.
 	 */
-	devm_free_irq(&nhi->pdev->dev, nhi->pdev->irq, nhi);
-	flush_work(&nhi->interrupt_work);
+	if (!nhi->pdev->msix_enabled) {
+		devm_free_irq(&nhi->pdev->dev, nhi->pdev->irq, nhi);
+		flush_work(&nhi->interrupt_work);
+	}
 	mutex_destroy(&nhi->lock);
+	ida_destroy(&nhi->msix_ida);
+}
+
+static int nhi_init_msi(struct tb_nhi *nhi)
+{
+	struct pci_dev *pdev = nhi->pdev;
+	int res, irq, nvec;
+
+	/* In case someone left them on. */
+	nhi_disable_interrupts(nhi);
+
+	ida_init(&nhi->msix_ida);
+
+	/*
+	 * The NHI has 16 MSI-X vectors or a single MSI. We first try to
+	 * get all MSI-X vectors and if we succeed, each ring will have
+	 * one MSI-X. If for some reason that does not work out, we
+	 * fallback to a single MSI.
+	 */
+	nvec = pci_alloc_irq_vectors(pdev, MSIX_MIN_VECS, MSIX_MAX_VECS,
+				     PCI_IRQ_MSIX);
+	if (nvec < 0) {
+		nvec = pci_alloc_irq_vectors(pdev, 1, 1, PCI_IRQ_MSI);
+		if (nvec < 0)
+			return nvec;
+
+		INIT_WORK(&nhi->interrupt_work, nhi_interrupt_work);
+
+		irq = pci_irq_vector(nhi->pdev, 0);
+		if (irq < 0)
+			return irq;
+
+		res = devm_request_irq(&pdev->dev, irq, nhi_msi,
+				       IRQF_NO_SUSPEND, "thunderbolt", nhi);
+		if (res) {
+			dev_err(&pdev->dev, "request_irq failed, aborting\n");
+			return res;
+		}
+	}
+
+	return 0;
 }
 
 static int nhi_probe(struct pci_dev *pdev, const struct pci_device_id *id)
@@ -545,12 +676,6 @@ static int nhi_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 		return res;
 	}
 
-	res = pci_enable_msi(pdev);
-	if (res) {
-		dev_err(&pdev->dev, "cannot enable MSI, aborting\n");
-		return res;
-	}
-
 	res = pcim_iomap_regions(pdev, 1 << 0, "thunderbolt");
 	if (res) {
 		dev_err(&pdev->dev, "cannot obtain PCI resources, aborting\n");
@@ -568,7 +693,6 @@ static int nhi_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 	if (nhi->hop_count != 12 && nhi->hop_count != 32)
 		dev_warn(&pdev->dev, "unexpected hop count: %d\n",
 			 nhi->hop_count);
-	INIT_WORK(&nhi->interrupt_work, nhi_interrupt_work);
 
 	nhi->tx_rings = devm_kcalloc(&pdev->dev, nhi->hop_count,
 				     sizeof(*nhi->tx_rings), GFP_KERNEL);
@@ -577,12 +701,9 @@ static int nhi_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 	if (!nhi->tx_rings || !nhi->rx_rings)
 		return -ENOMEM;
 
-	nhi_disable_interrupts(nhi); /* In case someone left them on. */
-	res = devm_request_irq(&pdev->dev, pdev->irq, nhi_msi,
-			       IRQF_NO_SUSPEND, /* must work during _noirq */
-			       "thunderbolt", nhi);
+	res = nhi_init_msi(nhi);
 	if (res) {
-		dev_err(&pdev->dev, "request_irq failed, aborting\n");
+		dev_err(&pdev->dev, "cannot enable MSI, aborting\n");
 		return res;
 	}