diff mbox

[v6,05/19] swiotlb-xen: support autotranslate guests

Message ID 1380298207-29151-5-git-send-email-stefano.stabellini@eu.citrix.com (mailing list archive)
State New, archived
Headers show

Commit Message

Stefano Stabellini Sept. 27, 2013, 4:09 p.m. UTC
Support autotranslate guests in swiotlb-xen by keeping track of the
phys-to-bus and bus-to-phys mappings of the swiotlb buffer
(xen_io_tlb_start-xen_io_tlb_end).

Use a simple direct access on a pre-allocated array for phys-to-bus
queries. Use a red-black tree for bus-to-phys queries.

Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Reviewed-by: David Vrabel <david.vrabel@citrix.com>
Reviewed-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>


Changes in v6:
- add a xen_dma_info entry to the rbtree in xen_swiotlb_alloc_coherent to keep
track of the new mapping. Free the entry in xen_swiotlb_free_coherent;
- rename xen_dma_seg to dma_info in xen_swiotlb_alloc/free_coherent to avoid
confusions.

Changes in v5:
- fix xen_dma_add_entry error path;
- remove the spin_lock: the red-black tree is not modified at run time.

Changes in v4:
- add err_out label in xen_dma_add_entry;
- remove INVALID_ADDRESS, use DMA_ERROR_CODE instead;
- code style fixes;
- add in-code comments regarding the usage of xen_dma_seg[0].dma_addr.

Changes in v3:
- many code style and name changes;
- improve error checks in xen_dma_add_entry.
---
 drivers/xen/swiotlb-xen.c |  177 +++++++++++++++++++++++++++++++++++++++++----
 1 files changed, 161 insertions(+), 16 deletions(-)
diff mbox

Patch

diff --git a/drivers/xen/swiotlb-xen.c b/drivers/xen/swiotlb-xen.c
index b72f31c..84aef43 100644
--- a/drivers/xen/swiotlb-xen.c
+++ b/drivers/xen/swiotlb-xen.c
@@ -38,32 +38,131 @@ 
 #include <linux/bootmem.h>
 #include <linux/dma-mapping.h>
 #include <linux/export.h>
+#include <linux/slab.h>
+#include <linux/rbtree.h>
 #include <xen/swiotlb-xen.h>
 #include <xen/page.h>
 #include <xen/xen-ops.h>
 #include <xen/hvc-console.h>
+#include <xen/features.h>
 /*
  * Used to do a quick range check in swiotlb_tbl_unmap_single and
  * swiotlb_tbl_sync_single_*, to see if the memory was in fact allocated by this
  * API.
  */
 
+#define NR_DMA_SEGS  ((xen_io_tlb_nslabs + IO_TLB_SEGSIZE - 1) / IO_TLB_SEGSIZE)
 static char *xen_io_tlb_start, *xen_io_tlb_end;
 static unsigned long xen_io_tlb_nslabs;
 /*
  * Quick lookup value of the bus address of the IOTLB.
  */
 
-static u64 start_dma_addr;
+struct xen_dma_info {
+	dma_addr_t dma_addr;
+	phys_addr_t phys_addr;
+	size_t size;
+	struct rb_node rbnode;
+};
+
+/*
+ * This array of struct xen_dma_info is indexed by physical addresses,
+ * starting from virt_to_phys(xen_io_tlb_start). Each entry maps
+ * (IO_TLB_SEGSIZE << IO_TLB_SHIFT) bytes, except the last one that is
+ * smaller. Getting the dma address corresponding to a given physical
+ * address can be done by direct access with the right index on the
+ * array.
+ */
+static struct xen_dma_info *xen_dma_seg;
+/* 
+ * This tree keeps track of bus address to physical address
+ * mappings.
+ */
+static struct rb_root bus_to_phys = RB_ROOT;
+
+static int xen_dma_add_entry(struct xen_dma_info *new)
+{
+	struct rb_node **link = &bus_to_phys.rb_node;
+	struct rb_node *parent = NULL;
+	struct xen_dma_info *entry;
+	int rc = 0;
+
+	while (*link) {
+		parent = *link;
+		entry = rb_entry(parent, struct xen_dma_info, rbnode);
+
+		if (new->dma_addr == entry->dma_addr)
+			goto err_out;
+		if (new->phys_addr == entry->phys_addr)
+			goto err_out;
+
+		if (new->dma_addr < entry->dma_addr)
+			link = &(*link)->rb_left;
+		else
+			link = &(*link)->rb_right;
+	}
+	rb_link_node(&new->rbnode, parent, link);
+	rb_insert_color(&new->rbnode, &bus_to_phys);
+	goto out;
+
+err_out:
+	rc = -EINVAL;
+	pr_warn("%s: cannot add phys=%pa -> dma=%pa: phys=%pa -> dma=%pa already exists\n",
+			__func__, &new->phys_addr, &new->dma_addr, &entry->phys_addr, &entry->dma_addr);
+out:
+	return rc;
+}
+
+static struct xen_dma_info *xen_get_dma_info_from_dma(dma_addr_t dma_addr)
+{
+	struct rb_node *n = bus_to_phys.rb_node;
+	struct xen_dma_info *entry;
+
+	while (n) {
+		entry = rb_entry(n, struct xen_dma_info, rbnode);
+		if (entry->dma_addr <= dma_addr &&
+				entry->dma_addr + entry->size > dma_addr) {
+			return entry;
+		}
+		if (dma_addr < entry->dma_addr)
+			n = n->rb_left;
+		else
+			n = n->rb_right;
+	}
+
+	return NULL;
+}
 
 static dma_addr_t xen_phys_to_bus(phys_addr_t paddr)
 {
-	return phys_to_machine(XPADDR(paddr)).maddr;
+	int nr_seg;
+	unsigned long offset;
+	char *vaddr;
+
+	if (!xen_feature(XENFEAT_auto_translated_physmap))
+		return phys_to_machine(XPADDR(paddr)).maddr;
+
+	vaddr = (char *)phys_to_virt(paddr);
+	if (vaddr >= xen_io_tlb_end || vaddr < xen_io_tlb_start)
+		return DMA_ERROR_CODE;
+
+	offset = vaddr - xen_io_tlb_start;
+	nr_seg = offset / (IO_TLB_SEGSIZE << IO_TLB_SHIFT);
+
+	return xen_dma_seg[nr_seg].dma_addr +
+		(paddr - xen_dma_seg[nr_seg].phys_addr);
 }
 
 static phys_addr_t xen_bus_to_phys(dma_addr_t baddr)
 {
-	return machine_to_phys(XMADDR(baddr)).paddr;
+	if (xen_feature(XENFEAT_auto_translated_physmap)) {
+		struct xen_dma_info *dma = xen_get_dma_info_from_dma(baddr);
+		if (dma == NULL)
+			return DMA_ERROR_CODE;
+		else
+			return dma->phys_addr + (baddr - dma->dma_addr);
+	} else
+		return machine_to_phys(XMADDR(baddr)).paddr;
 }
 
 static dma_addr_t xen_virt_to_bus(void *address)
@@ -107,6 +206,9 @@  static int is_xen_swiotlb_buffer(dma_addr_t dma_addr)
 	unsigned long pfn = mfn_to_local_pfn(mfn);
 	phys_addr_t paddr;
 
+	if (xen_feature(XENFEAT_auto_translated_physmap))
+		return 1;
+
 	/* If the address is outside our domain, it CAN
 	 * have the same virtual address as another address
 	 * in our domain. Therefore _only_ check address within our domain.
@@ -124,13 +226,12 @@  static int max_dma_bits = 32;
 static int
 xen_swiotlb_fixup(void *buf, size_t size, unsigned long nslabs)
 {
-	int i, rc;
+	int i, j, rc;
 	int dma_bits;
-	dma_addr_t dma_handle;
 
 	dma_bits = get_order(IO_TLB_SEGSIZE << IO_TLB_SHIFT) + PAGE_SHIFT;
 
-	i = 0;
+	i = j = 0;
 	do {
 		int slabs = min(nslabs - i, (unsigned long)IO_TLB_SEGSIZE);
 
@@ -138,12 +239,18 @@  xen_swiotlb_fixup(void *buf, size_t size, unsigned long nslabs)
 			rc = xen_create_contiguous_region(
 				(unsigned long)buf + (i << IO_TLB_SHIFT),
 				get_order(slabs << IO_TLB_SHIFT),
-				dma_bits, &dma_handle);
+				dma_bits, &xen_dma_seg[j].dma_addr);
 		} while (rc && dma_bits++ < max_dma_bits);
 		if (rc)
 			return rc;
 
+		xen_dma_seg[j].phys_addr = virt_to_phys(buf + (i << IO_TLB_SHIFT));
+		xen_dma_seg[j].size = slabs << IO_TLB_SHIFT;
+		rc = xen_dma_add_entry(&xen_dma_seg[j]);
+		if (rc != 0)
+			return rc;
 		i += slabs;
+		j++;
 	} while (i < nslabs);
 	return 0;
 }
@@ -193,9 +300,10 @@  retry:
 	/*
 	 * Get IO TLB memory from any location.
 	 */
-	if (early)
+	if (early) {
 		xen_io_tlb_start = alloc_bootmem_pages(PAGE_ALIGN(bytes));
-	else {
+		xen_dma_seg = alloc_bootmem(sizeof(struct xen_dma_info) * NR_DMA_SEGS);
+	} else {
 #define SLABS_PER_PAGE (1 << (PAGE_SHIFT - IO_TLB_SHIFT))
 #define IO_TLB_MIN_SLABS ((1<<20) >> IO_TLB_SHIFT)
 		while ((SLABS_PER_PAGE << order) > IO_TLB_MIN_SLABS) {
@@ -210,6 +318,8 @@  retry:
 			xen_io_tlb_nslabs = SLABS_PER_PAGE << order;
 			bytes = xen_io_tlb_nslabs << IO_TLB_SHIFT;
 		}
+		xen_dma_seg = kzalloc(sizeof(struct xen_dma_info) * NR_DMA_SEGS,
+				GFP_KERNEL);
 	}
 	if (!xen_io_tlb_start) {
 		m_ret = XEN_SWIOTLB_ENOMEM;
@@ -232,7 +342,6 @@  retry:
 		m_ret = XEN_SWIOTLB_EFIXUP;
 		goto error;
 	}
-	start_dma_addr = xen_virt_to_bus(xen_io_tlb_start);
 	if (early) {
 		if (swiotlb_init_with_tbl(xen_io_tlb_start, xen_io_tlb_nslabs,
 			 verbose))
@@ -267,6 +376,7 @@  xen_swiotlb_alloc_coherent(struct device *hwdev, size_t size,
 	unsigned long vstart;
 	phys_addr_t phys;
 	dma_addr_t dev_addr;
+	struct xen_dma_info *dma_info = NULL;
 
 	/*
 	* Ignore region specifiers - the kernel's ideas of
@@ -290,7 +400,8 @@  xen_swiotlb_alloc_coherent(struct device *hwdev, size_t size,
 
 	phys = virt_to_phys(ret);
 	dev_addr = xen_phys_to_bus(phys);
-	if (((dev_addr + size - 1 <= dma_mask)) &&
+	if (!xen_feature(XENFEAT_auto_translated_physmap) &&
+	    ((dev_addr + size - 1 <= dma_mask)) &&
 	    !range_straddles_page_boundary(phys, size))
 		*dma_handle = dev_addr;
 	else {
@@ -299,6 +410,22 @@  xen_swiotlb_alloc_coherent(struct device *hwdev, size_t size,
 			free_pages(vstart, order);
 			return NULL;
 		}
+
+		dma_info = kzalloc(sizeof(struct xen_dma_info), GFP_KERNEL);
+		if (!dma_info) {
+			pr_warn("cannot allocate xen_dma_info\n");
+			xen_destroy_contiguous_region(phys, order);
+			return NULL;
+		}
+		dma_info->phys_addr = phys;
+		dma_info->size = size;
+		dma_info->dma_addr = *dma_handle;
+		if (xen_dma_add_entry(dma_info)) {
+			pr_warn("cannot add new entry to bus_to_phys\n");
+			xen_destroy_contiguous_region(phys, order);
+			kfree(dma_info);
+			return NULL;
+		}
 	}
 	memset(ret, 0, size);
 	return ret;
@@ -312,6 +439,7 @@  xen_swiotlb_free_coherent(struct device *hwdev, size_t size, void *vaddr,
 	int order = get_order(size);
 	phys_addr_t phys;
 	u64 dma_mask = DMA_BIT_MASK(32);
+	struct xen_dma_info *dma_info = NULL;
 
 	if (dma_release_from_coherent(hwdev, order, vaddr))
 		return;
@@ -321,9 +449,14 @@  xen_swiotlb_free_coherent(struct device *hwdev, size_t size, void *vaddr,
 
 	phys = virt_to_phys(vaddr);
 
-	if (((dev_addr + size - 1 > dma_mask)) ||
-	    range_straddles_page_boundary(phys, size))
+	if (xen_feature(XENFEAT_auto_translated_physmap) ||
+		(((dev_addr + size - 1 > dma_mask)) ||
+		 range_straddles_page_boundary(phys, size))) {
 		xen_destroy_contiguous_region((unsigned long)vaddr, order);
+		dma_info = xen_get_dma_info_from_dma(dev_addr);
+		rb_erase(&dma_info->rbnode, &bus_to_phys);
+		kfree(dma_info);
+	}
 
 	free_pages((unsigned long)vaddr, order);
 }
@@ -351,14 +484,19 @@  dma_addr_t xen_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) &&
+	if (!xen_feature(XENFEAT_auto_translated_physmap) &&
+	    dma_capable(dev, dev_addr, size) &&
 	    !range_straddles_page_boundary(phys, size) && !swiotlb_force)
 		return dev_addr;
 
 	/*
 	 * Oh well, have to allocate and map a bounce buffer.
+	 * Pass the dma_addr of the first slab in the iotlb buffer as
+	 * argument so that swiotlb_tbl_map_single is free to allocate
+	 * the bounce buffer anywhere appropriate in io_tlb_start -
+	 * io_tlb_end.
 	 */
-	map = swiotlb_tbl_map_single(dev, start_dma_addr, phys, size, dir);
+	map = swiotlb_tbl_map_single(dev, xen_dma_seg[0].dma_addr, phys, size, dir);
 	if (map == SWIOTLB_MAP_ERROR)
 		return DMA_ERROR_CODE;
 
@@ -494,10 +632,17 @@  xen_swiotlb_map_sg_attrs(struct device *hwdev, struct scatterlist *sgl,
 		dma_addr_t dev_addr = xen_phys_to_bus(paddr);
 
 		if (swiotlb_force ||
+		    xen_feature(XENFEAT_auto_translated_physmap) ||
 		    !dma_capable(hwdev, dev_addr, sg->length) ||
 		    range_straddles_page_boundary(paddr, sg->length)) {
+			/*
+			 * Pass the dma_addr of the first slab in the iotlb buffer as
+			 * argument so that swiotlb_tbl_map_single is free to allocate
+			 * the bounce buffer anywhere appropriate in io_tlb_start -
+			 * io_tlb_end.
+			 */
 			phys_addr_t map = swiotlb_tbl_map_single(hwdev,
-								 start_dma_addr,
+								 xen_dma_seg[0].dma_addr,
 								 sg_phys(sg),
 								 sg->length,
 								 dir);