Re: Adding support for WC (write-combining) memory to bus_dma

On Thursday, July 12, 2012 1:51:20 pm John Baldwin wrote:
> On Thursday, July 12, 2012 12:37:13 pm Scott Long wrote:
> > Yup, this is a problem, and I like your fix; this kind of state is exactly what
> > belongs in the map.
> 
> Why don't I break out the fix first as a separate patch.

Here is a patch to just fix this.  I also mirrored the changes into powerpc
since it had copied the same code from x86 (albeit disabled).  If you feel
the malloc stats via M_DEVBUF are important, I can restore those (probably
by adding a contigmalloc_attr() wrapper).

Index: powerpc/powerpc/busdma_machdep.c
===================================================================
--- powerpc/powerpc/busdma_machdep.c	(revision 238365)
+++ powerpc/powerpc/busdma_machdep.c	(working copy)
_at__at_ -46,6 +46,8 _at__at_
 #include <sys/sysctl.h>
 
 #include <vm/vm.h>
+#include <vm/vm_extern.h>
+#include <vm/vm_kern.h>
 #include <vm/vm_page.h>
 #include <vm/vm_map.h>
 
_at__at_ -130,6 +132,7 _at__at_
 	bus_dmamap_callback_t *callback;
 	void		      *callback_arg;
 	STAILQ_ENTRY(bus_dmamap) links;
+	int		       contigalloc;
 };
 
 static STAILQ_HEAD(, bus_dmamap) bounce_map_waitinglist;
_at__at_ -489,6 +492,7 _at__at_
 bus_dmamem_alloc(bus_dma_tag_t dmat, void** vaddr, int flags,
 		 bus_dmamap_t *mapp)
 {
+	vm_memattr_t attr;
 	int mflags;
 
 	if (flags & BUS_DMA_NOWAIT)
_at__at_ -500,6 +504,12 _at__at_
 
 	if (flags & BUS_DMA_ZERO)
 		mflags |= M_ZERO;
+#ifdef NOTYET
+	if (flags & BUS_DMA_NOCACHE)
+		attr = VM_MEMATTR_UNCACHEABLE;
+	else
+#endif
+		attr = VM_MEMATTR_DEFAULT;
 
 	/* 
 	 * XXX:
_at__at_ -511,7 +521,8 _at__at_
 	 */
 	if ((dmat->maxsize <= PAGE_SIZE) &&
 	   (dmat->alignment < dmat->maxsize) &&
-	    dmat->lowaddr >= ptoa((vm_paddr_t)Maxmem)) {
+	    dmat->lowaddr >= ptoa((vm_paddr_t)Maxmem) &&
+	    attr == VM_MEMATTR_DEFAULT) {
 		*vaddr = malloc(dmat->maxsize, M_DEVBUF, mflags);
 	} else {
 		/*
_at__at_ -520,9 +531,10 _at__at_
 		 *     multi-seg allocations yet though.
 		 * XXX Certain AGP hardware does.
 		 */
-		*vaddr = contigmalloc(dmat->maxsize, M_DEVBUF, mflags,
-		    0ul, dmat->lowaddr, dmat->alignment? dmat->alignment : 1ul,
-		    dmat->boundary);
+		*vaddr = (void *)kmem_alloc_contig(kernel_map, dmat->maxsize,
+		    mflags, 0ul, dmat->lowaddr, dmat->alignment ?
+		    dmat->alignment : 1ul, dmat->boundary, attr);
+		(*mapp)->contigalloc = 1;
 	}
 	if (*vaddr == NULL) {
 		CTR4(KTR_BUSDMA, "%s: tag %p tag flags 0x%x error %d",
_at__at_ -531,11 +543,6 _at__at_
 	} else if (vtophys(*vaddr) & (dmat->alignment - 1)) {
 		printf("bus_dmamem_alloc failed to align memory properly.\n");
 	}
-#ifdef NOTYET
-	if (flags & BUS_DMA_NOCACHE)
-		pmap_change_attr((vm_offset_t)*vaddr, dmat->maxsize,
-		    VM_MEMATTR_UNCACHEABLE);
-#endif
 	CTR4(KTR_BUSDMA, "%s: tag %p tag flags 0x%x error %d",
 	    __func__, dmat, dmat->flags, 0);
 	return (0);
_at__at_ -548,18 +555,12 _at__at_
 void
 bus_dmamem_free(bus_dma_tag_t dmat, void *vaddr, bus_dmamap_t map)
 {
-	bus_dmamap_destroy(dmat, map);
 
-#ifdef NOTYET
-	pmap_change_attr((vm_offset_t)vaddr, dmat->maxsize, VM_MEMATTR_DEFAULT);
-#endif
-	if ((dmat->maxsize <= PAGE_SIZE) &&
-	   (dmat->alignment < dmat->maxsize) &&
-	    dmat->lowaddr >= ptoa((vm_paddr_t)Maxmem))
+	if (!map->contigalloc)
 		free(vaddr, M_DEVBUF);
-	else {
-		contigfree(vaddr, dmat->maxsize, M_DEVBUF);
-	}
+	else
+		kmem_free(kernel_map, (vm_offset_t)vaddr, dmat->maxsize);
+	bus_dmamap_destroy(dmat, map);
 	CTR3(KTR_BUSDMA, "%s: tag %p flags 0x%x", __func__, dmat, dmat->flags);
 }
 
Index: x86/x86/busdma_machdep.c
===================================================================
--- x86/x86/busdma_machdep.c	(revision 238365)
+++ x86/x86/busdma_machdep.c	(working copy)
_at__at_ -42,6 +42,8 _at__at_
 #include <sys/sysctl.h>
 
 #include <vm/vm.h>
+#include <vm/vm_extern.h>
+#include <vm/vm_kern.h>
 #include <vm/vm_page.h>
 #include <vm/vm_map.h>
 
_at__at_ -131,7 +133,7 _at__at_
 
 static STAILQ_HEAD(, bus_dmamap) bounce_map_waitinglist;
 static STAILQ_HEAD(, bus_dmamap) bounce_map_callbacklist;
-static struct bus_dmamap nobounce_dmamap;
+static struct bus_dmamap nobounce_dmamap, contig_dmamap;
 
 static void init_bounce_pages(void *dummy);
 static int alloc_bounce_zone(bus_dma_tag_t dmat);
_at__at_ -465,7 +467,7 _at__at_
 int
 bus_dmamap_destroy(bus_dma_tag_t dmat, bus_dmamap_t map)
 {
-	if (map != NULL && map != &nobounce_dmamap) {
+	if (map != NULL && map != &nobounce_dmamap && map != &contig_dmamap) {
 		if (STAILQ_FIRST(&map->bpages) != NULL) {
 			CTR3(KTR_BUSDMA, "%s: tag %p error %d",
 			    __func__, dmat, EBUSY);
_at__at_ -490,6 +492,7 _at__at_
 bus_dmamem_alloc(bus_dma_tag_t dmat, void** vaddr, int flags,
 		 bus_dmamap_t *mapp)
 {
+	vm_memattr_t attr;
 	int mflags;
 
 	if (flags & BUS_DMA_NOWAIT)
_at__at_ -512,6 +515,10 _at__at_
 	}
 	if (flags & BUS_DMA_ZERO)
 		mflags |= M_ZERO;
+	if (flags & BUS_DMA_NOCACHE)
+		attr = VM_MEMATTR_UNCACHEABLE;
+	else
+		attr = VM_MEMATTR_DEFAULT;
 
 	/* 
 	 * XXX:
_at__at_ -523,7 +530,8 _at__at_
 	 */
 	if ((dmat->maxsize <= PAGE_SIZE) &&
 	   (dmat->alignment < dmat->maxsize) &&
-	    dmat->lowaddr >= ptoa((vm_paddr_t)Maxmem)) {
+	    dmat->lowaddr >= ptoa((vm_paddr_t)Maxmem) &&
+	    attr == VM_MEMATTR_DEFAULT) {
 		*vaddr = malloc(dmat->maxsize, M_DEVBUF, mflags);
 	} else {
 		/*
_at__at_ -532,9 +540,10 _at__at_
 		 *     multi-seg allocations yet though.
 		 * XXX Certain AGP hardware does.
 		 */
-		*vaddr = contigmalloc(dmat->maxsize, M_DEVBUF, mflags,
-		    0ul, dmat->lowaddr, dmat->alignment? dmat->alignment : 1ul,
-		    dmat->boundary);
+		*vaddr = (void *)kmem_alloc_contig(kernel_map, dmat->maxsize,
+		    mflags, 0ul, dmat->lowaddr, dmat->alignment ?
+		    dmat->alignment : 1ul, dmat->boundary, attr);
+		*mapp = &contig_dmamap;
 	}
 	if (*vaddr == NULL) {
 		CTR4(KTR_BUSDMA, "%s: tag %p tag flags 0x%x error %d",
_at__at_ -543,9 +552,6 _at__at_
 	} else if (vtophys(*vaddr) & (dmat->alignment - 1)) {
 		printf("bus_dmamem_alloc failed to align memory properly.\n");
 	}
-	if (flags & BUS_DMA_NOCACHE)
-		pmap_change_attr((vm_offset_t)*vaddr, dmat->maxsize,
-		    PAT_UNCACHEABLE);
 	CTR4(KTR_BUSDMA, "%s: tag %p tag flags 0x%x error %d",
 	    __func__, dmat, dmat->flags, 0);
 	return (0);
_at__at_ -560,18 +566,15 _at__at_
 {
 	/*
 	 * dmamem does not need to be bounced, so the map should be
-	 * NULL
+	 * NULL if malloc() was used and contig_dmamap if
+	 * contigmalloc() was used.
 	 */
-	if (map != NULL)
+	if (!(map == NULL || map == &contig_dmamap))
 		panic("bus_dmamem_free: Invalid map freed\n");
-	pmap_change_attr((vm_offset_t)vaddr, dmat->maxsize, PAT_WRITE_BACK);
-	if ((dmat->maxsize <= PAGE_SIZE) &&
-	   (dmat->alignment < dmat->maxsize) &&
-	    dmat->lowaddr >= ptoa((vm_paddr_t)Maxmem))
+	if (map == NULL)
 		free(vaddr, M_DEVBUF);
-	else {
-		contigfree(vaddr, dmat->maxsize, M_DEVBUF);
-	}
+	else
+		kmem_free(kernel_map, (vm_offset_t)vaddr, dmat->maxsize);
 	CTR3(KTR_BUSDMA, "%s: tag %p flags 0x%x", __func__, dmat, dmat->flags);
 }
 
_at__at_ -662,7 +665,7 _at__at_
 	vm_offset_t vaddr;
 	int seg, error;
 
-	if (map == NULL)
+	if (map == NULL || map == &contig_dmamap)
 		map = &nobounce_dmamap;
 
 	if ((dmat->flags & BUS_DMA_COULD_BOUNCE) != 0) {
_at__at_ -1139,7 +1142,7 _at__at_
 	struct bounce_page *bpage;
 
 	KASSERT(dmat->bounce_zone != NULL, ("no bounce zone in dma tag"));
-	KASSERT(map != NULL && map != &nobounce_dmamap,
+	KASSERT(map != NULL && map != &nobounce_dmamap && map != &contig_dmamap,
 	    ("add_bounce_page: bad map %p", map));
 
 	bz = dmat->bounce_zone;

-- 
John Baldwin