[PATCH 07/10] swiotlb: refactor swiotlb_map_page

[hch@lst.de (Christoph Hellwig)]

On Thu, Oct 18, 2018 at 08:37:15PM -0400, Konrad Rzeszutek Wilk wrote:
> > > +	if (!dma_capable(dev, dma_addr, size) ||
> > > +	    swiotlb_force == SWIOTLB_FORCE) {
> > > +		trace_swiotlb_bounced(dev, dma_addr, size, swiotlb_force);
> > > +		dma_addr = swiotlb_bounce_page(dev, &phys, size, dir, attrs);
> > > +	}
> > 
> > FWIW I prefer the inverse condition and early return of the original code
> > here, which also then allows a tail-call to swiotlb_bounce_page() (and saves
> > a couple of lines), but it's no biggie.
> > 
> > Reviewed-by: Robin Murphy <robin.murphy@arm.com>
> 
> I agree with Robin - it certainly makes it easier to read.
> 
> With that small change:
> Reviewed-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>

So I did this edit, and in this patch it does indeed look much cleaner.
But in patch 9 we introduce the cache maintainance, and have to invert
the condition again if we don't want a goto mess:

---
>From e840ec23360788d54a8ebd2ebc7cd0f0ef8bdb01 Mon Sep 17 00:00:00 2001
From: Christoph Hellwig <hch@lst.de>
Date: Fri, 19 Oct 2018 08:51:53 +0200
Subject: swiotlb: add support for non-coherent DMA

Handle architectures that are not cache coherent directly in the main
swiotlb code by calling arch_sync_dma_for_{device,cpu} in all the right
places from the various dma_map/unmap/sync methods when the device is
non-coherent.

Because swiotlb now uses dma_direct_alloc for the coherent allocation
that side is already taken care of by the dma-direct code calling into
arch_dma_{alloc,free} for devices that are non-coherent.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 kernel/dma/swiotlb.c | 33 +++++++++++++++++++++++----------
 1 file changed, 23 insertions(+), 10 deletions(-)

diff --git a/kernel/dma/swiotlb.c b/kernel/dma/swiotlb.c
index 1a01b0ac0a5e..ebecaf255ea2 100644
--- a/kernel/dma/swiotlb.c
+++ b/kernel/dma/swiotlb.c
@@ -21,6 +21,7 @@
 
 #include <linux/cache.h>
 #include <linux/dma-direct.h>
+#include <linux/dma-noncoherent.h>
 #include <linux/mm.h>
 #include <linux/export.h>
 #include <linux/spinlock.h>
@@ -671,11 +672,17 @@ dma_addr_t swiotlb_map_page(struct device *dev, struct page *page,
 	 * we can safely return the device addr and not worry about bounce
 	 * buffering it.
 	 */
-	if (dma_capable(dev, dev_addr, size) && swiotlb_force != SWIOTLB_FORCE)
-		return dev_addr;
+	if (!dma_capable(dev, dev_addr, size) ||
+	    swiotlb_force == SWIOTLB_FORCE) {
+		trace_swiotlb_bounced(dev, dev_addr, size, swiotlb_force);
+		dev_addr = swiotlb_bounce_page(dev, &phys, size, dir, attrs);
+	}
+
+	if (!dev_is_dma_coherent(dev) &&
+	    (attrs & DMA_ATTR_SKIP_CPU_SYNC) == 0)
+		arch_sync_dma_for_device(dev, phys, size, dir);
 
-	trace_swiotlb_bounced(dev, dev_addr, size, swiotlb_force);
-	return swiotlb_bounce_page(dev, &phys, size, dir, attrs);
+	return dev_addr;
 }
 
 /*
@@ -694,6 +701,10 @@ void swiotlb_unmap_page(struct device *hwdev, dma_addr_t dev_addr,
 
 	BUG_ON(dir == DMA_NONE);
 
+	if (!dev_is_dma_coherent(hwdev) &&
+	    (attrs & DMA_ATTR_SKIP_CPU_SYNC) == 0)
+		arch_sync_dma_for_cpu(hwdev, paddr, size, dir);
+
 	if (is_swiotlb_buffer(paddr)) {
 		swiotlb_tbl_unmap_single(hwdev, paddr, size, dir, attrs);
 		return;
@@ -730,15 +741,17 @@ swiotlb_sync_single(struct device *hwdev, dma_addr_t dev_addr,
 
 	BUG_ON(dir == DMA_NONE);
 
-	if (is_swiotlb_buffer(paddr)) {
+	if (!dev_is_dma_coherent(hwdev) && target == SYNC_FOR_CPU)
+		arch_sync_dma_for_cpu(hwdev, paddr, size, dir);
+
+	if (is_swiotlb_buffer(paddr))
 		swiotlb_tbl_sync_single(hwdev, paddr, size, dir, target);
-		return;
-	}
 
-	if (dir != DMA_FROM_DEVICE)
-		return;
+	if (!dev_is_dma_coherent(hwdev) && target == SYNC_FOR_DEVICE)
+		arch_sync_dma_for_device(hwdev, paddr, size, dir);
 
-	dma_mark_clean(phys_to_virt(paddr), size);
+	if (!is_swiotlb_buffer(paddr) && dir == DMA_FROM_DEVICE)
+		dma_mark_clean(phys_to_virt(paddr), size);
 }
 
 void
-- 
2.19.1