[PATCH 2/2 v3] dmaengine: mv_xor: Add support for IO (PCIe) src/dst areas

[sr@denx.de (Stefan Roese)]

To enable the access to a specific area, the MVEBU XOR controllers needs
to have this area enabled / mapped via an address window. Right now,
only the DRAM memory area is enabled via such memory windows. So
using this driver to DMA to / from a e.g. PCIe memory region is
currently not supported.

This patch now adds support for such PCIe / IO regions by checking
if the src / dst address is located in an IO memory area in contrast
to being located in DRAM. This is done by using the newly introduced
MBus function mvebu_mbus_get_io_win_info(). If the src / dst address
is located in such an IO area, a new address window is created in
the XOR DMA controller. Enabling the controller to access this area.

Signed-off-by: Stefan Roese <sr@denx.de>
Cc: Gregory CLEMENT <gregory.clement@free-electrons.com>
Cc: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Cc: Marcin Wojtas <mw@semihalf.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Vinod Koul <vinod.koul@intel.com>
---
v2 -> v3:
- Included separate patch to add a stub function for mvebu_mbus_get_io_win_info()
  to this patch series
- COMPILE_TEST added again, as the stub function now allows compilation on
  all platforms

v1 -> v2:
- Removed COMPILE_TEST from Kconfig entry as this results in compilation
  errors on non-Marvell platforms because of references to MBus driver
  functions. Since this driver will never be used on non-Marvell platforms
  this should be okay.

RFC -> v1:
- Use a cache for the already mapped areas for performance reasons, as
  the MMIO registers don't need to get queried any more if the area is
  already mapped now

 drivers/dma/mv_xor.c | 91 +++++++++++++++++++++++++++++++++++++++++++++++++++-
 drivers/dma/mv_xor.h |  7 ++++
 2 files changed, 97 insertions(+), 1 deletion(-)

diff --git a/drivers/dma/mv_xor.c b/drivers/dma/mv_xor.c
index f4c9f98..0162750 100644
--- a/drivers/dma/mv_xor.c
+++ b/drivers/dma/mv_xor.c
@@ -470,12 +470,86 @@ static int mv_xor_alloc_chan_resources(struct dma_chan *chan)
 	return mv_chan->slots_allocated ? : -ENOMEM;
 }
 
+/*
+ * Check if source or destination is an PCIe/IO address (non-SDRAM) and add
+ * a new MBus window if necessary. Use a cache for these check so that
+ * the MMIO mapped registers don't have to be accessed for this check
+ * to speed up this process.
+ */
+static int mv_xor_add_io_win(struct mv_xor_chan *mv_chan, u32 addr)
+{
+	struct mv_xor_device *xordev = mv_chan->xordev;
+	void __iomem *base = mv_chan->mmr_high_base;
+	u32 win_enable;
+	u32 size;
+	u8 target, attr;
+	int ret;
+	int i;
+
+	/*
+	 * Loop over the cached windows to check, if the requested area
+	 * is already mapped. If this the case, nothing needs to be done
+	 * and we can return.
+	 */
+	for (i = 0; i < WINDOW_COUNT; i++) {
+		if (addr >= xordev->win_start[i] &&
+		    addr <= xordev->win_end[i]) {
+			/* Window is already mapped */
+			return 0;
+		}
+	}
+
+	/*
+	 * The window is not mapped, so we need to create the new mapping
+	 */
+
+	/* If no IO window is found that addr has to be located in SDRAM */
+	ret = mvebu_mbus_get_io_win_info(addr, &size, &target, &attr);
+	if (ret < 0)
+		return 0;
+
+	/*
+	 * Mask the base addr 'addr' according to 'size' read back from the
+	 * MBus window. Otherwise we might end up with an address located
+	 * somewhere in the middle of this area here.
+	 */
+	size -= 1;
+	addr &= ~size;
+
+	/*
+	 * Reading one of both enabled register is enough, as they are always
+	 * programmed to the identical values
+	 */
+	win_enable = readl(base + WINDOW_BAR_ENABLE(0));
+
+	/* Set 'i' to the first free window to write the new values to */
+	i = ffs(~win_enable) - 1;
+	if (i >= WINDOW_COUNT)
+		return -ENOMEM;
+
+	writel((addr & 0xffff0000) | (attr << 8) | target,
+	       base + WINDOW_BASE(i));
+	writel(size & 0xffff0000, base + WINDOW_SIZE(i));
+
+	/* Fill the caching variables for later use */
+	xordev->win_start[i] = addr;
+	xordev->win_end[i] = addr + size;
+
+	win_enable |= (1 << i);
+	win_enable |= 3 << (16 + (2 * i));
+	writel(win_enable, base + WINDOW_BAR_ENABLE(0));
+	writel(win_enable, base + WINDOW_BAR_ENABLE(1));
+
+	return 0;
+}
+
 static struct dma_async_tx_descriptor *
 mv_xor_prep_dma_xor(struct dma_chan *chan, dma_addr_t dest, dma_addr_t *src,
 		    unsigned int src_cnt, size_t len, unsigned long flags)
 {
 	struct mv_xor_chan *mv_chan = to_mv_xor_chan(chan);
 	struct mv_xor_desc_slot *sw_desc;
+	int ret;
 
 	if (unlikely(len < MV_XOR_MIN_BYTE_COUNT))
 		return NULL;
@@ -486,6 +560,11 @@ mv_xor_prep_dma_xor(struct dma_chan *chan, dma_addr_t dest, dma_addr_t *src,
 		"%s src_cnt: %d len: %zu dest %pad flags: %ld\n",
 		__func__, src_cnt, len, &dest, flags);
 
+	/* Check if a new window needs to get added for 'dest' */
+	ret = mv_xor_add_io_win(mv_chan, dest);
+	if (ret)
+		return NULL;
+
 	sw_desc = mv_chan_alloc_slot(mv_chan);
 	if (sw_desc) {
 		sw_desc->type = DMA_XOR;
@@ -493,8 +572,13 @@ mv_xor_prep_dma_xor(struct dma_chan *chan, dma_addr_t dest, dma_addr_t *src,
 		mv_desc_init(sw_desc, dest, len, flags);
 		if (mv_chan->op_in_desc == XOR_MODE_IN_DESC)
 			mv_desc_set_mode(sw_desc);
-		while (src_cnt--)
+		while (src_cnt--) {
+			/* Check if a new window needs to get added for 'src' */
+			ret = mv_xor_add_io_win(mv_chan, src[src_cnt]);
+			if (ret)
+				return NULL;
 			mv_desc_set_src_addr(sw_desc, src_cnt, src[src_cnt]);
+		}
 	}
 
 	dev_dbg(mv_chan_to_devp(mv_chan),
@@ -959,6 +1043,7 @@ mv_xor_channel_add(struct mv_xor_device *xordev,
 		mv_chan->op_in_desc = XOR_MODE_IN_DESC;
 
 	dma_dev = &mv_chan->dmadev;
+	mv_chan->xordev = xordev;
 
 	/*
 	 * These source and destination dummy buffers are used to implement
@@ -1086,6 +1171,10 @@ mv_xor_conf_mbus_windows(struct mv_xor_device *xordev,
 		       dram->mbus_dram_target_id, base + WINDOW_BASE(i));
 		writel((cs->size - 1) & 0xffff0000, base + WINDOW_SIZE(i));
 
+		/* Fill the caching variables for later use */
+		xordev->win_start[i] = cs->base;
+		xordev->win_end[i] = cs->base + cs->size - 1;
+
 		win_enable |= (1 << i);
 		win_enable |= 3 << (16 + (2 * i));
 	}
diff --git a/drivers/dma/mv_xor.h b/drivers/dma/mv_xor.h
index bf56e08..88eeab2 100644
--- a/drivers/dma/mv_xor.h
+++ b/drivers/dma/mv_xor.h
@@ -80,12 +80,17 @@
 #define WINDOW_BAR_ENABLE(chan)	(0x40 + ((chan) << 2))
 #define WINDOW_OVERRIDE_CTRL(chan)	(0xA0 + ((chan) << 2))
 
+#define WINDOW_COUNT		8
+
 struct mv_xor_device {
 	void __iomem	     *xor_base;
 	void __iomem	     *xor_high_base;
 	struct clk	     *clk;
 	struct mv_xor_chan   *channels[MV_XOR_MAX_CHANNELS];
 	int		     xor_type;
+
+	u32                  win_start[WINDOW_COUNT];
+	u32                  win_end[WINDOW_COUNT];
 };
 
 /**
@@ -127,6 +132,8 @@ struct mv_xor_chan {
 	char			dummy_dst[MV_XOR_MIN_BYTE_COUNT];
 	dma_addr_t		dummy_src_addr, dummy_dst_addr;
 	u32                     saved_config_reg, saved_int_mask_reg;
+
+	struct mv_xor_device	*xordev;
 };
 
 /**
-- 
2.9.3