openwrt/target/linux/brcm2708/patches-3.10/0149-bcm2708_fb-use-IRQ-for-DMA-copies.patch

From ab41011a32fdc35352ee6b308f29b6954056d787 Mon Sep 17 00:00:00 2001
From: Luke Diamand <luked@broadcom.com>
Date: Wed, 1 Jan 2014 00:45:29 +0000
Subject: [PATCH 149/174] bcm2708_fb: use IRQ for DMA copies

The copyarea ioctl() uses DMA to speed things along. This
was busy-waiting for completion. This change supports using
an interrupt instead for larger transfers. For small
transfers, busy-waiting is still likely to be faster.

Signed-off-by: Luke Diamand <luke@diamand.org>
---
 arch/arm/mach-bcm2708/dma.c              |  8 ++++
 arch/arm/mach-bcm2708/include/mach/dma.h |  2 +
 drivers/video/bcm2708_fb.c               | 64 ++++++++++++++++++++++++++++++--
 3 files changed, 70 insertions(+), 4 deletions(-)

--- a/arch/arm/mach-bcm2708/dma.c
+++ b/arch/arm/mach-bcm2708/dma.c
@@ -83,6 +83,14 @@ extern void bcm_dma_wait_idle(void __iom
 
 EXPORT_SYMBOL_GPL(bcm_dma_start);
 
+extern bool bcm_dma_is_busy(void __iomem *dma_chan_base)
+{
+	dsb();
+
+	return readl(dma_chan_base + BCM2708_DMA_CS) & BCM2708_DMA_ACTIVE;
+}
+EXPORT_SYMBOL_GPL(bcm_dma_is_busy);
+
 /* Complete an ongoing DMA (assuming its results are to be ignored)
    Does nothing if there is no DMA in progress.
    This routine waits for the current AXI transfer to complete before
--- a/arch/arm/mach-bcm2708/include/mach/dma.h
+++ b/arch/arm/mach-bcm2708/include/mach/dma.h
@@ -64,11 +64,13 @@ struct bcm2708_dma_cb {
 	unsigned long next;
 	unsigned long pad[2];
 };
+struct scatterlist;
 
 extern int bcm_sg_suitable_for_dma(struct scatterlist *sg_ptr, int sg_len);
 extern void bcm_dma_start(void __iomem *dma_chan_base,
 			  dma_addr_t control_block);
 extern void bcm_dma_wait_idle(void __iomem *dma_chan_base);
+extern bool bcm_dma_is_busy(void __iomem *dma_chan_base);
 extern int /*rc*/ bcm_dma_abort(void __iomem *dma_chan_base);
 
 /* When listing features we can ask for when allocating DMA channels give
--- a/drivers/video/bcm2708_fb.c
+++ b/drivers/video/bcm2708_fb.c
@@ -21,6 +21,7 @@
 #include <linux/mm.h>
 #include <linux/fb.h>
 #include <linux/init.h>
+#include <linux/interrupt.h>
 #include <linux/ioport.h>
 #include <linux/list.h>
 #include <linux/platform_device.h>
@@ -48,6 +49,11 @@ static const char *bcm2708_name = "BCM27
 
 #define DRIVER_NAME "bcm2708_fb"
 
+static u32 dma_busy_wait_threshold = 1<<15;
+module_param(dma_busy_wait_threshold, int, 0644);
+MODULE_PARM_DESC(dma_busy_wait_threshold, "Busy-wait for DMA completion below this area");
+
+
 /* this data structure describes each frame buffer device we find */
 
 struct fbinfo_s {
@@ -77,6 +83,7 @@ struct bcm2708_fb {
 	void *cb_base;		/* DMA control blocks */
 	dma_addr_t cb_handle;
 	struct dentry *debugfs_dir;
+	wait_queue_head_t dma_waitq;
 	struct bcm2708_fb_stats stats;
 };
 
@@ -95,6 +102,10 @@ static int bcm2708_fb_debugfs_init(struc
 			"dma_copies",
 			offsetof(struct bcm2708_fb_stats, dma_copies)
 		},
+		{
+			"dma_irqs",
+			offsetof(struct bcm2708_fb_stats, dma_irqs)
+		},
 	};
 
 	fb->debugfs_dir = debugfs_create_dir(DRIVER_NAME, NULL);
@@ -400,6 +411,7 @@ static void bcm2708_fb_copyarea(struct f
 	int bytes_per_pixel = (info->var.bits_per_pixel + 7) >> 3;
 	/* Channel 0 supports larger bursts and is a bit faster */
 	int burst_size = (fb->dma_chan == 0) ? 8 : 2;
+	int pixels = region->width * region->height;
 
 	/* Fallback to cfb_copyarea() if we don't like something */
 	if (bytes_per_pixel > 4 ||
@@ -492,8 +504,20 @@ static void bcm2708_fb_copyarea(struct f
 	cb->next = 0;
 
 
-	bcm_dma_start(fb->dma_chan_base, fb->cb_handle);
-	bcm_dma_wait_idle(fb->dma_chan_base);
+	if (pixels < dma_busy_wait_threshold) {
+		bcm_dma_start(fb->dma_chan_base, fb->cb_handle);
+		bcm_dma_wait_idle(fb->dma_chan_base);
+	} else {
+		void __iomem *dma_chan = fb->dma_chan_base;
+		cb->info |= BCM2708_DMA_INT_EN;
+		bcm_dma_start(fb->dma_chan_base, fb->cb_handle);
+		while (bcm_dma_is_busy(dma_chan)) {
+			wait_event_interruptible(
+				fb->dma_waitq,
+				!bcm_dma_is_busy(dma_chan));
+		}
+		fb->stats.dma_irqs++;
+	}
 	fb->stats.dma_copies++;
 }
 
@@ -504,6 +528,24 @@ static void bcm2708_fb_imageblit(struct
 	cfb_imageblit(info, image);
 }
 
+static irqreturn_t bcm2708_fb_dma_irq(int irq, void *cxt)
+{
+	struct bcm2708_fb *fb = cxt;
+
+	/* FIXME: should read status register to check if this is
+	 * actually interrupting us or not, in case this interrupt
+	 * ever becomes shared amongst several DMA channels
+	 *
+	 * readl(dma_chan_base + BCM2708_DMA_CS) & BCM2708_DMA_IRQ;
+	 */
+
+	/* acknowledge the interrupt */
+	writel(BCM2708_DMA_INT, fb->dma_chan_base + BCM2708_DMA_CS);
+
+	wake_up(&fb->dma_waitq);
+	return IRQ_HANDLED;
+}
+
 static struct fb_ops bcm2708_fb_ops = {
 	.owner = THIS_MODULE,
 	.fb_check_var = bcm2708_fb_check_var,
@@ -568,6 +610,7 @@ static int bcm2708_fb_register(struct bc
 	fb->fb.monspecs.dclkmax = 100000000;
 
 	bcm2708_fb_set_bitfields(&fb->fb.var);
+	init_waitqueue_head(&fb->dma_waitq);
 
 	/*
 	 * Allocate colourmap.
@@ -593,14 +636,15 @@ static int bcm2708_fb_probe(struct platf
 	struct bcm2708_fb *fb;
 	int ret;
 
-	fb = kmalloc(sizeof(struct bcm2708_fb), GFP_KERNEL);
+	fb = kzalloc(sizeof(struct bcm2708_fb), GFP_KERNEL);
 	if (!fb) {
 		dev_err(&dev->dev,
 			"could not allocate new bcm2708_fb struct\n");
 		ret = -ENOMEM;
 		goto free_region;
 	}
-	memset(fb, 0, sizeof(struct bcm2708_fb));
+
+	bcm2708_fb_debugfs_init(fb);
 
 
 	bcm2708_fb_debugfs_init(fb);
@@ -624,6 +668,14 @@ static int bcm2708_fb_probe(struct platf
 	}
 	fb->dma_chan = ret;
 
+	ret = request_irq(fb->dma_irq, bcm2708_fb_dma_irq,
+			  0, "bcm2708_fb dma", fb);
+	if (ret) {
+		pr_err("%s: failed to request DMA irq\n", __func__);
+		goto free_dma_chan;
+	}
+
+
 	pr_info("BCM2708FB: allocated DMA channel %d @ %p\n",
 	       fb->dma_chan, fb->dma_chan_base);
 
@@ -635,6 +687,8 @@ static int bcm2708_fb_probe(struct platf
 		goto out;
 	}
 
+free_dma_chan:
+	bcm_dma_chan_free(fb->dma_chan);
 free_cb:
 	dma_free_writecombine(&dev->dev, SZ_64K, fb->cb_base, fb->cb_handle);
 free_fb:
@@ -662,6 +716,8 @@ static int bcm2708_fb_remove(struct plat
 			  fb->dma);
 	bcm2708_fb_debugfs_deinit(fb);
 
+	free_irq(fb->dma_irq, fb);
+
 	kfree(fb);
 
 	return 0;