On Fri, 2007-04-27 at 18:13 -0600, Alex Williamson wrote:
> We need to get to a CONFIG_IA64_GENERIC kernel working, and the first
> step to doing that seems to be fixing our usage of the i386 swiotlb and
> pci-dma-xen. The patch below paravirtualizes lib/swiotlb.c and rips out
> a lot of the code we had around for supporting the i386 versions. It
> should be a little more straight-forward to switch to the generic kernel
> flavor after this. This hasn't had much testing, so please don't run
> this on a machine with data you care about. I'm running it with
> swiotlb=force to try to exercise it, and it's holding up well. Comments
> welcome. Thanks,
FWIW, here's changes from the stock swiotlb for paravirtualization.
Thanks,
Alex
--- swiotlb.c 2006-09-19 21:42:06.000000000 -0600
+++ swiotlb.c 2007-04-27 17:43:14.000000000 -0600
@@ -32,11 +32,20 @@
#include <linux/init.h>
#include <linux/bootmem.h>
+#ifdef CONFIG_XEN
+#define VIRT_TO_PHYS(x) virt_to_bus(x)
+#define PHYS_TO_VIRT(x) bus_to_virt(x)
+#else
+#define VIRT_TO_PHYS(x) virt_to_phys(x)
+#define PHYS_TO_VIRT(x) phys_to_virt(x)
+#define range_straddles_page_boundary(x, y) (0)
+#endif
+
#define OFFSET(val,align) ((unsigned long) \
( (val) & ( (align) - 1)))
#define SG_ENT_VIRT_ADDRESS(sg) (page_address((sg)->page) +
(sg)->offset)
-#define SG_ENT_PHYS_ADDRESS(SG) virt_to_phys(SG_ENT_VIRT_ADDRESS(SG))
+#define SG_ENT_PHYS_ADDRESS(SG) VIRT_TO_PHYS(SG_ENT_VIRT_ADDRESS(SG))
/*
* Maximum allowable number of contiguous slabs to map,
@@ -139,6 +148,13 @@
io_tlb_nslabs = ALIGN(io_tlb_nslabs, IO_TLB_SEGSIZE);
}
+#ifdef CONFIG_XEN
+ if (is_running_on_xen()) {
+ /* Round up to a power of two */
+ while (io_tlb_nslabs & (io_tlb_nslabs - 1))
+ io_tlb_nslabs += io_tlb_nslabs & ~(io_tlb_nslabs - 1);
+ }
+#endif
/*
* Get IO TLB memory from the low pages
*/
@@ -147,6 +163,17 @@
panic("Cannot allocate SWIOTLB buffer");
io_tlb_end = io_tlb_start + io_tlb_nslabs * (1 << IO_TLB_SHIFT);
+#ifdef CONFIG_XEN
+ /* TODO: add logic for trying to get lower pages */
+ for (i = 0 ; i < io_tlb_nslabs ; i += IO_TLB_SEGSIZE) {
+ if (xen_create_contiguous_region(
+ (unsigned long)io_tlb_start +
+ (i << IO_TLB_SHIFT),
+ get_order(IO_TLB_SEGSIZE << IO_TLB_SHIFT), 32))
+ panic("Failed to setup Xen contiguous region");
+ }
+#endif
+
/*
* Allocate and initialize the free list array. This array is used
* to find contiguous free memory regions of size up to IO_TLB_SEGSIZE
@@ -162,6 +189,11 @@
* Get the overflow emergency buffer
*/
io_tlb_overflow_buffer = alloc_bootmem_low(io_tlb_overflow);
+#ifdef CONFIG_XEN
+ if (xen_create_contiguous_region((unsigned long)io_tlb_overflow_buffer,
+ get_order(io_tlb_overflow), 32))
+ panic("Failed to setup Xen contiguous region for overflow");
+#endif
printk(KERN_INFO "Placing software IO TLB between 0x%lx - 0x%lx\n",
virt_to_phys(io_tlb_start), virt_to_phys(io_tlb_end));
}
@@ -188,6 +220,13 @@
io_tlb_nslabs = ALIGN(io_tlb_nslabs, IO_TLB_SEGSIZE);
}
+#ifdef CONFIG_XEN
+ if (is_running_on_xen()) {
+ /* Round up to a power of two */
+ while (io_tlb_nslabs & (io_tlb_nslabs - 1))
+ io_tlb_nslabs += io_tlb_nslabs & ~(io_tlb_nslabs - 1);
+ }
+#endif
/*
* Get IO TLB memory from the low pages
*/
@@ -213,6 +252,16 @@
io_tlb_end = io_tlb_start + io_tlb_nslabs * (1 << IO_TLB_SHIFT);
memset(io_tlb_start, 0, io_tlb_nslabs * (1 << IO_TLB_SHIFT));
+#ifdef CONFIG_XEN
+ /* TODO: add logic for trying to get lower pages */
+ for (i = 0 ; i < io_tlb_nslabs ; i += IO_TLB_SEGSIZE) {
+ if (xen_create_contiguous_region(
+ (unsigned long)io_tlb_start +
+ (i << IO_TLB_SHIFT),
+ get_order(IO_TLB_SEGSIZE << IO_TLB_SHIFT), 32))
+ panic("Failed to setup Xen contiguous region");
+ }
+#endif
/*
* Allocate and initialize the free list array. This array is used
* to find contiguous free memory regions of size up to IO_TLB_SEGSIZE
@@ -242,6 +291,11 @@
if (!io_tlb_overflow_buffer)
goto cleanup4;
+#ifdef CONFIG_XEN
+ if (xen_create_contiguous_region((unsigned long)io_tlb_overflow_buffer,
+ get_order(io_tlb_overflow), 32))
+ panic("Failed to setup Xen contiguous region for overflow");
+#endif
printk(KERN_INFO "Placing %ldMB software IO TLB between 0x%lx - "
"0x%lx\n", (io_tlb_nslabs * (1 << IO_TLB_SHIFT)) >> 20,
virt_to_phys(io_tlb_start), virt_to_phys(io_tlb_end));
@@ -445,7 +499,18 @@
flags |= GFP_DMA;
ret = (void *)__get_free_pages(flags, order);
- if (ret && address_needs_mapping(hwdev, virt_to_phys(ret))) {
+#ifdef CONFIG_XEN
+ if (ret && is_running_on_xen()) {
+ u64 mask = hwdev->coherent_dma_mask ? hwdev->coherent_dma_mask :
+ DMA_32BIT_MASK;
+ if (xen_create_contiguous_region((unsigned long)ret, order,
+ fls64(mask))) {
+ free_pages((unsigned long)ret, order);
+ ret = NULL;
+ }
+ }
+#endif
+ if (ret && address_needs_mapping(hwdev, VIRT_TO_PHYS(ret))) {
/*
* The allocated memory isn't reachable by the device.
* Fall back on swiotlb_map_single().
@@ -465,11 +530,11 @@
if (swiotlb_dma_mapping_error(handle))
return NULL;
- ret = phys_to_virt(handle);
+ ret = PHYS_TO_VIRT(handle);
}
memset(ret, 0, size);
- dev_addr = virt_to_phys(ret);
+ dev_addr = VIRT_TO_PHYS(ret);
/* Confirm address can be DMA'd by device */
if (address_needs_mapping(hwdev, dev_addr)) {
@@ -487,9 +552,14 @@
dma_addr_t dma_handle)
{
if (!(vaddr >= (void *)io_tlb_start
- && vaddr < (void *)io_tlb_end))
+ && vaddr < (void *)io_tlb_end)) {
+#ifdef CONFIG_XEN
+ if (is_running_on_xen())
+ xen_destroy_contiguous_region((unsigned long)vaddr,
+ get_order(size));
+#endif
free_pages((unsigned long) vaddr, get_order(size));
- else
+ } else
/* DMA_TO_DEVICE to avoid memcpy in unmap_single */
swiotlb_unmap_single (hwdev, dma_handle, size, DMA_TO_DEVICE);
}
@@ -525,7 +595,7 @@
dma_addr_t
swiotlb_map_single(struct device *hwdev, void *ptr, size_t size, int dir)
{
- unsigned long dev_addr = virt_to_phys(ptr);
+ unsigned long dev_addr = VIRT_TO_PHYS(ptr);
void *map;
BUG_ON(dir == DMA_NONE);
@@ -534,7 +604,8 @@
* we can safely return the device addr and not worry about bounce
* buffering it.
*/
- if (!address_needs_mapping(hwdev, dev_addr) && !swiotlb_force)
+ if (!range_straddles_page_boundary(ptr, size) &&
+ !address_needs_mapping(hwdev, dev_addr) && !swiotlb_force)
return dev_addr;
/*
@@ -546,7 +617,7 @@
map = io_tlb_overflow_buffer;
}
- dev_addr = virt_to_phys(map);
+ dev_addr = VIRT_TO_PHYS(map);
/*
* Ensure that the address returned is DMA'ble
@@ -588,7 +659,7 @@
swiotlb_unmap_single(struct device *hwdev, dma_addr_t dev_addr, size_t size,
int dir)
{
- char *dma_addr = phys_to_virt(dev_addr);
+ char *dma_addr = PHYS_TO_VIRT(dev_addr);
BUG_ON(dir == DMA_NONE);
if (dma_addr >= io_tlb_start && dma_addr < io_tlb_end)
@@ -611,7 +682,7 @@
swiotlb_sync_single(struct device *hwdev, dma_addr_t dev_addr,
size_t size, int dir, int target)
{
- char *dma_addr = phys_to_virt(dev_addr);
+ char *dma_addr = PHYS_TO_VIRT(dev_addr);
BUG_ON(dir == DMA_NONE);
if (dma_addr >= io_tlb_start && dma_addr < io_tlb_end)
@@ -642,7 +713,7 @@
unsigned long offset, size_t size,
int dir, int target)
{
- char *dma_addr = phys_to_virt(dev_addr) + offset;
+ char *dma_addr = PHYS_TO_VIRT(dev_addr) + offset;
BUG_ON(dir == DMA_NONE);
if (dma_addr >= io_tlb_start && dma_addr < io_tlb_end)
@@ -695,7 +766,7 @@
for (i = 0; i < nelems; i++, sg++) {
addr = SG_ENT_VIRT_ADDRESS(sg);
- dev_addr = virt_to_phys(addr);
+ dev_addr = VIRT_TO_PHYS(addr);
if (swiotlb_force || address_needs_mapping(hwdev, dev_addr)) {
void *map = map_single(hwdev, addr, sg->length, dir);
sg->dma_address = virt_to_bus(map);
@@ -728,7 +799,7 @@
for (i = 0; i < nelems; i++, sg++)
if (sg->dma_address != SG_ENT_PHYS_ADDRESS(sg))
- unmap_single(hwdev, (void *)
phys_to_virt(sg->dma_address), sg->dma_length, dir);
+ unmap_single(hwdev, (void *)
PHYS_TO_VIRT(sg->dma_address), sg->dma_length, dir);
else if (dir == DMA_FROM_DEVICE)
mark_clean(SG_ENT_VIRT_ADDRESS(sg), sg->dma_length);
}
@@ -771,7 +842,7 @@
int
swiotlb_dma_mapping_error(dma_addr_t dma_addr)
{
- return (dma_addr == virt_to_phys(io_tlb_overflow_buffer));
+ return (dma_addr == VIRT_TO_PHYS(io_tlb_overflow_buffer));
}
/*
@@ -783,7 +854,7 @@
int
swiotlb_dma_supported (struct device *hwdev, u64 mask)
{
- return (virt_to_phys (io_tlb_end) - 1) <= mask;
+ return (VIRT_TO_PHYS(io_tlb_end) - 1) <= mask;
}
EXPORT_SYMBOL(swiotlb_init);
_______________________________________________
Xen-ia64-devel mailing list
Xen-ia64-devel@lists.xensource.com
http://lists.xensource.com/xen-ia64-devel
