summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorhselasky <hselasky@FreeBSD.org>2017-05-27 07:47:52 +0000
committerLuiz Souza <luiz@netgate.com>2017-07-17 14:19:46 -0500
commite962be5686ab67404c9baa6de33086cc43673a82 (patch)
tree638d9a6d28267224e4c58401ac67a6bf5eb23a0d
parent963c98e7cd5f5ea2f9ae1304c1fb4b3cf6457d9a (diff)
downloadFreeBSD-src-e962be5686ab67404c9baa6de33086cc43673a82.zip
FreeBSD-src-e962be5686ab67404c9baa6de33086cc43673a82.tar.gz
MFC r318353:
Avoid use of contiguous memory allocations in busdma when possible. This patch improves the boundary checks in busdma to allow more cases using the regular page based kernel memory allocator. Especially in the case of having a non-zero boundary in the parent DMA tag. For example AMD64 based platforms set the PCI DMA tag boundary to PCI_DMA_BOUNDARY, 4GB, which before this patch caused contiguous memory allocations to be preferred when allocating more than PAGE_SIZE bytes. Even if the required alignment was less than PAGE_SIZE bytes. This patch also fixes the nsegments check for using kmem_alloc_attr() when the maximum segment size is less than PAGE_SIZE bytes. Updated some comments describing the code in question. Differential Revision: https://reviews.freebsd.org/D10645 Reviewed by: kib, jhb, gallatin, scottl Sponsored by: Mellanox Technologies (cherry picked from commit 6181e9a69342927c6b1d10bc6a0aff8386ac9e38)
-rw-r--r--sys/arm/arm/busdma_machdep-v4.c6
-rw-r--r--sys/arm/arm/busdma_machdep-v6.c10
-rw-r--r--sys/arm64/arm64/busdma_bounce.c29
-rw-r--r--sys/mips/mips/busdma_machdep.c10
-rw-r--r--sys/x86/x86/busdma_bounce.c31
5 files changed, 61 insertions, 25 deletions
diff --git a/sys/arm/arm/busdma_machdep-v4.c b/sys/arm/arm/busdma_machdep-v4.c
index f7e0e26..9e98b4a 100644
--- a/sys/arm/arm/busdma_machdep-v4.c
+++ b/sys/arm/arm/busdma_machdep-v4.c
@@ -736,8 +736,10 @@ bus_dmamem_alloc(bus_dma_tag_t dmat, void **vaddr, int flags,
if (bufzone != NULL && dmat->alignment <= bufzone->size &&
!_bus_dma_can_bounce(dmat->lowaddr, dmat->highaddr)) {
*vaddr = uma_zalloc(bufzone->umazone, mflags);
- } else if (dmat->nsegments >= btoc(dmat->maxsize) &&
- dmat->alignment <= PAGE_SIZE && dmat->boundary == 0) {
+ } else if (dmat->nsegments >=
+ howmany(dmat->maxsize, MIN(dmat->maxsegsz, PAGE_SIZE)) &&
+ dmat->alignment <= PAGE_SIZE &&
+ (dmat->boundary % PAGE_SIZE) == 0) {
*vaddr = (void *)kmem_alloc_attr(kernel_arena, dmat->maxsize,
mflags, 0, dmat->lowaddr, memattr);
} else {
diff --git a/sys/arm/arm/busdma_machdep-v6.c b/sys/arm/arm/busdma_machdep-v6.c
index 5bf4626..58e2012 100644
--- a/sys/arm/arm/busdma_machdep-v6.c
+++ b/sys/arm/arm/busdma_machdep-v6.c
@@ -779,7 +779,9 @@ bus_dmamem_alloc(bus_dma_tag_t dmat, void **vaddr, int flags,
* (the allocator aligns buffers to their size boundaries).
* - There's no need to handle lowaddr/highaddr exclusion zones.
* else allocate non-contiguous pages if...
- * - The page count that could get allocated doesn't exceed nsegments.
+ * - The page count that could get allocated doesn't exceed
+ * nsegments also when the maximum segment size is less
+ * than PAGE_SIZE.
* - The alignment constraint isn't larger than a page boundary.
* - There are no boundary-crossing constraints.
* else allocate a block of contiguous pages because one or more of the
@@ -788,8 +790,10 @@ bus_dmamem_alloc(bus_dma_tag_t dmat, void **vaddr, int flags,
if (bufzone != NULL && dmat->alignment <= bufzone->size &&
!exclusion_bounce(dmat)) {
*vaddr = uma_zalloc(bufzone->umazone, mflags);
- } else if (dmat->nsegments >= btoc(dmat->maxsize) &&
- dmat->alignment <= PAGE_SIZE && dmat->boundary == 0) {
+ } else if (dmat->nsegments >=
+ howmany(dmat->maxsize, MIN(dmat->maxsegsz, PAGE_SIZE)) &&
+ dmat->alignment <= PAGE_SIZE &&
+ (dmat->boundary % PAGE_SIZE) == 0) {
*vaddr = (void *)kmem_alloc_attr(kernel_arena, dmat->maxsize,
mflags, 0, dmat->lowaddr, memattr);
} else {
diff --git a/sys/arm64/arm64/busdma_bounce.c b/sys/arm64/arm64/busdma_bounce.c
index 7039153..b33e956 100644
--- a/sys/arm64/arm64/busdma_bounce.c
+++ b/sys/arm64/arm64/busdma_bounce.c
@@ -454,22 +454,35 @@ bounce_bus_dmamem_alloc(bus_dma_tag_t dmat, void** vaddr, int flags,
(*mapp)->flags = DMAMAP_FROM_DMAMEM;
/*
- * XXX:
- * (dmat->alignment <= dmat->maxsize) is just a quick hack; the exact
- * alignment guarantees of malloc need to be nailed down, and the
- * code below should be rewritten to take that into account.
+ * Allocate the buffer from the malloc(9) allocator if...
+ * - It's small enough to fit into a single power of two sized bucket.
+ * - The alignment is less than or equal to the maximum size
+ * - The low address requirement is fulfilled.
+ * else allocate non-contiguous pages if...
+ * - The page count that could get allocated doesn't exceed
+ * nsegments also when the maximum segment size is less
+ * than PAGE_SIZE.
+ * - The alignment constraint isn't larger than a page boundary.
+ * - There are no boundary-crossing constraints.
+ * else allocate a block of contiguous pages because one or more of the
+ * constraints is something that only the contig allocator can fulfill.
*
- * In the meantime, we'll warn the user if malloc gets it wrong.
+ * NOTE: The (dmat->common.alignment <= dmat->maxsize) check
+ * below is just a quick hack. The exact alignment guarantees
+ * of malloc(9) need to be nailed down, and the code below
+ * should be rewritten to take that into account.
+ *
+ * In the meantime warn the user if malloc gets it wrong.
*/
if ((dmat->common.maxsize <= PAGE_SIZE) &&
(dmat->common.alignment <= dmat->common.maxsize) &&
dmat->common.lowaddr >= ptoa((vm_paddr_t)Maxmem) &&
attr == VM_MEMATTR_DEFAULT) {
*vaddr = malloc(dmat->common.maxsize, M_DEVBUF, mflags);
- } else if (dmat->common.nsegments >= btoc(dmat->common.maxsize) &&
+ } else if (dmat->common.nsegments >=
+ howmany(dmat->common.maxsize, MIN(dmat->common.maxsegsz, PAGE_SIZE)) &&
dmat->common.alignment <= PAGE_SIZE &&
- (dmat->common.boundary == 0 ||
- dmat->common.boundary >= dmat->common.lowaddr)) {
+ (dmat->common.boundary % PAGE_SIZE) == 0) {
/* Page-based multi-segment allocations allowed */
*vaddr = (void *)kmem_alloc_attr(kernel_arena,
dmat->common.maxsize, mflags, 0ul, dmat->common.lowaddr,
diff --git a/sys/mips/mips/busdma_machdep.c b/sys/mips/mips/busdma_machdep.c
index d9eb9d4..bd2cada 100644
--- a/sys/mips/mips/busdma_machdep.c
+++ b/sys/mips/mips/busdma_machdep.c
@@ -680,7 +680,9 @@ bus_dmamem_alloc(bus_dma_tag_t dmat, void** vaddrp, int flags,
* (the allocator aligns buffers to their size boundaries).
* - There's no need to handle lowaddr/highaddr exclusion zones.
* else allocate non-contiguous pages if...
- * - The page count that could get allocated doesn't exceed nsegments.
+ * - The page count that could get allocated doesn't exceed
+ * nsegments also when the maximum segment size is less
+ * than PAGE_SIZE.
* - The alignment constraint isn't larger than a page boundary.
* - There are no boundary-crossing constraints.
* else allocate a block of contiguous pages because one or more of the
@@ -689,8 +691,10 @@ bus_dmamem_alloc(bus_dma_tag_t dmat, void** vaddrp, int flags,
if (bufzone != NULL && dmat->alignment <= bufzone->size &&
!_bus_dma_can_bounce(dmat->lowaddr, dmat->highaddr)) {
vaddr = uma_zalloc(bufzone->umazone, mflags);
- } else if (dmat->nsegments >= btoc(dmat->maxsize) &&
- dmat->alignment <= PAGE_SIZE && dmat->boundary == 0) {
+ } else if (dmat->nsegments >=
+ howmany(dmat->maxsize, MIN(dmat->maxsegsz, PAGE_SIZE)) &&
+ dmat->alignment <= PAGE_SIZE &&
+ (dmat->boundary % PAGE_SIZE) == 0) {
vaddr = (void *)kmem_alloc_attr(kernel_arena, dmat->maxsize,
mflags, 0, dmat->lowaddr, memattr);
} else {
diff --git a/sys/x86/x86/busdma_bounce.c b/sys/x86/x86/busdma_bounce.c
index 78d04b9..bed0cdb 100644
--- a/sys/x86/x86/busdma_bounce.c
+++ b/sys/x86/x86/busdma_bounce.c
@@ -396,23 +396,36 @@ bounce_bus_dmamem_alloc(bus_dma_tag_t dmat, void** vaddr, int flags,
else
attr = VM_MEMATTR_DEFAULT;
- /*
- * XXX:
- * (dmat->alignment <= dmat->maxsize) is just a quick hack; the exact
- * alignment guarantees of malloc need to be nailed down, and the
- * code below should be rewritten to take that into account.
+ /*
+ * Allocate the buffer from the malloc(9) allocator if...
+ * - It's small enough to fit into a single power of two sized bucket.
+ * - The alignment is less than or equal to the maximum size
+ * - The low address requirement is fulfilled.
+ * else allocate non-contiguous pages if...
+ * - The page count that could get allocated doesn't exceed
+ * nsegments also when the maximum segment size is less
+ * than PAGE_SIZE.
+ * - The alignment constraint isn't larger than a page boundary.
+ * - There are no boundary-crossing constraints.
+ * else allocate a block of contiguous pages because one or more of the
+ * constraints is something that only the contig allocator can fulfill.
+ *
+ * NOTE: The (dmat->common.alignment <= dmat->maxsize) check
+ * below is just a quick hack. The exact alignment guarantees
+ * of malloc(9) need to be nailed down, and the code below
+ * should be rewritten to take that into account.
*
- * In the meantime, we'll warn the user if malloc gets it wrong.
+ * In the meantime warn the user if malloc gets it wrong.
*/
if ((dmat->common.maxsize <= PAGE_SIZE) &&
(dmat->common.alignment <= dmat->common.maxsize) &&
dmat->common.lowaddr >= ptoa((vm_paddr_t)Maxmem) &&
attr == VM_MEMATTR_DEFAULT) {
*vaddr = malloc(dmat->common.maxsize, M_DEVBUF, mflags);
- } else if (dmat->common.nsegments >= btoc(dmat->common.maxsize) &&
+ } else if (dmat->common.nsegments >=
+ howmany(dmat->common.maxsize, MIN(dmat->common.maxsegsz, PAGE_SIZE)) &&
dmat->common.alignment <= PAGE_SIZE &&
- (dmat->common.boundary == 0 ||
- dmat->common.boundary >= dmat->common.lowaddr)) {
+ (dmat->common.boundary % PAGE_SIZE) == 0) {
/* Page-based multi-segment allocations allowed */
*vaddr = (void *)kmem_alloc_attr(kernel_arena,
dmat->common.maxsize, mflags, 0ul, dmat->common.lowaddr,
OpenPOWER on IntegriCloud