mlx4_core: Support ICM tables in coherent memory
Enable having ICM tables in coherent memory, and use coherent memory
for the dMPT table. This will allow writing MPT entries for MRs both
via the SW2HW_MPT command and also directly by the driver for FMR
remapping without needing to flush or worry about cacheline boundaries.
Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Michael S. Tsirkin <mst@dev.mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
diff --git a/drivers/net/mlx4/icm.c b/drivers/net/mlx4/icm.c
index b7a4aa8..250e248 100644
--- a/drivers/net/mlx4/icm.c
+++ b/drivers/net/mlx4/icm.c
@@ -34,6 +34,7 @@
#include <linux/init.h>
#include <linux/errno.h>
#include <linux/mm.h>
+#include <linux/scatterlist.h>
#include <linux/mlx4/cmd.h>
@@ -50,19 +51,41 @@
MLX4_TABLE_CHUNK_SIZE = 1 << 18
};
-void mlx4_free_icm(struct mlx4_dev *dev, struct mlx4_icm *icm)
+static void mlx4_free_icm_pages(struct mlx4_dev *dev, struct mlx4_icm_chunk *chunk)
{
- struct mlx4_icm_chunk *chunk, *tmp;
int i;
- list_for_each_entry_safe(chunk, tmp, &icm->chunk_list, list) {
- if (chunk->nsg > 0)
- pci_unmap_sg(dev->pdev, chunk->mem, chunk->npages,
- PCI_DMA_BIDIRECTIONAL);
+ if (chunk->nsg > 0)
+ pci_unmap_sg(dev->pdev, chunk->mem, chunk->npages,
+ PCI_DMA_BIDIRECTIONAL);
- for (i = 0; i < chunk->npages; ++i)
- __free_pages(chunk->mem[i].page,
- get_order(chunk->mem[i].length));
+ for (i = 0; i < chunk->npages; ++i)
+ __free_pages(chunk->mem[i].page,
+ get_order(chunk->mem[i].length));
+}
+
+static void mlx4_free_icm_coherent(struct mlx4_dev *dev, struct mlx4_icm_chunk *chunk)
+{
+ int i;
+
+ for (i = 0; i < chunk->npages; ++i)
+ dma_free_coherent(&dev->pdev->dev, chunk->mem[i].length,
+ lowmem_page_address(chunk->mem[i].page),
+ sg_dma_address(&chunk->mem[i]));
+}
+
+void mlx4_free_icm(struct mlx4_dev *dev, struct mlx4_icm *icm, int coherent)
+{
+ struct mlx4_icm_chunk *chunk, *tmp;
+
+ if (!icm)
+ return;
+
+ list_for_each_entry_safe(chunk, tmp, &icm->chunk_list, list) {
+ if (coherent)
+ mlx4_free_icm_coherent(dev, chunk);
+ else
+ mlx4_free_icm_pages(dev, chunk);
kfree(chunk);
}
@@ -70,16 +93,45 @@
kfree(icm);
}
+static int mlx4_alloc_icm_pages(struct scatterlist *mem, int order, gfp_t gfp_mask)
+{
+ mem->page = alloc_pages(gfp_mask, order);
+ if (!mem->page)
+ return -ENOMEM;
+
+ mem->length = PAGE_SIZE << order;
+ mem->offset = 0;
+ return 0;
+}
+
+static int mlx4_alloc_icm_coherent(struct device *dev, struct scatterlist *mem,
+ int order, gfp_t gfp_mask)
+{
+ void *buf = dma_alloc_coherent(dev, PAGE_SIZE << order,
+ &sg_dma_address(mem), gfp_mask);
+ if (!buf)
+ return -ENOMEM;
+
+ sg_set_buf(mem, buf, PAGE_SIZE << order);
+ BUG_ON(mem->offset);
+ sg_dma_len(mem) = PAGE_SIZE << order;
+ return 0;
+}
+
struct mlx4_icm *mlx4_alloc_icm(struct mlx4_dev *dev, int npages,
- gfp_t gfp_mask)
+ gfp_t gfp_mask, int coherent)
{
struct mlx4_icm *icm;
struct mlx4_icm_chunk *chunk = NULL;
int cur_order;
+ int ret;
+
+ /* We use sg_set_buf for coherent allocs, which assumes low memory */
+ BUG_ON(coherent && (gfp_mask & __GFP_HIGHMEM));
icm = kmalloc(sizeof *icm, gfp_mask & ~(__GFP_HIGHMEM | __GFP_NOWARN));
if (!icm)
- return icm;
+ return NULL;
icm->refcount = 0;
INIT_LIST_HEAD(&icm->chunk_list);
@@ -101,12 +153,20 @@
while (1 << cur_order > npages)
--cur_order;
- chunk->mem[chunk->npages].page = alloc_pages(gfp_mask, cur_order);
- if (chunk->mem[chunk->npages].page) {
- chunk->mem[chunk->npages].length = PAGE_SIZE << cur_order;
- chunk->mem[chunk->npages].offset = 0;
+ if (coherent)
+ ret = mlx4_alloc_icm_coherent(&dev->pdev->dev,
+ &chunk->mem[chunk->npages],
+ cur_order, gfp_mask);
+ else
+ ret = mlx4_alloc_icm_pages(&chunk->mem[chunk->npages],
+ cur_order, gfp_mask);
- if (++chunk->npages == MLX4_ICM_CHUNK_LEN) {
+ if (!ret) {
+ ++chunk->npages;
+
+ if (coherent)
+ ++chunk->nsg;
+ else if (chunk->npages == MLX4_ICM_CHUNK_LEN) {
chunk->nsg = pci_map_sg(dev->pdev, chunk->mem,
chunk->npages,
PCI_DMA_BIDIRECTIONAL);
@@ -125,7 +185,7 @@
}
}
- if (chunk) {
+ if (!coherent && chunk) {
chunk->nsg = pci_map_sg(dev->pdev, chunk->mem,
chunk->npages,
PCI_DMA_BIDIRECTIONAL);
@@ -137,7 +197,7 @@
return icm;
fail:
- mlx4_free_icm(dev, icm);
+ mlx4_free_icm(dev, icm, coherent);
return NULL;
}
@@ -202,7 +262,7 @@
table->icm[i] = mlx4_alloc_icm(dev, MLX4_TABLE_CHUNK_SIZE >> PAGE_SHIFT,
(table->lowmem ? GFP_KERNEL : GFP_HIGHUSER) |
- __GFP_NOWARN);
+ __GFP_NOWARN, table->coherent);
if (!table->icm[i]) {
ret = -ENOMEM;
goto out;
@@ -210,7 +270,7 @@
if (mlx4_MAP_ICM(dev, table->icm[i], table->virt +
(u64) i * MLX4_TABLE_CHUNK_SIZE)) {
- mlx4_free_icm(dev, table->icm[i]);
+ mlx4_free_icm(dev, table->icm[i], table->coherent);
table->icm[i] = NULL;
ret = -ENOMEM;
goto out;
@@ -234,7 +294,7 @@
if (--table->icm[i]->refcount == 0) {
mlx4_UNMAP_ICM(dev, table->virt + i * MLX4_TABLE_CHUNK_SIZE,
MLX4_TABLE_CHUNK_SIZE / MLX4_ICM_PAGE_SIZE);
- mlx4_free_icm(dev, table->icm[i]);
+ mlx4_free_icm(dev, table->icm[i], table->coherent);
table->icm[i] = NULL;
}
@@ -309,7 +369,7 @@
int mlx4_init_icm_table(struct mlx4_dev *dev, struct mlx4_icm_table *table,
u64 virt, int obj_size, int nobj, int reserved,
- int use_lowmem)
+ int use_lowmem, int use_coherent)
{
int obj_per_chunk;
int num_icm;
@@ -327,6 +387,7 @@
table->num_obj = nobj;
table->obj_size = obj_size;
table->lowmem = use_lowmem;
+ table->coherent = use_coherent;
mutex_init(&table->mutex);
for (i = 0; i * MLX4_TABLE_CHUNK_SIZE < reserved * obj_size; ++i) {
@@ -336,11 +397,11 @@
table->icm[i] = mlx4_alloc_icm(dev, chunk_size >> PAGE_SHIFT,
(use_lowmem ? GFP_KERNEL : GFP_HIGHUSER) |
- __GFP_NOWARN);
+ __GFP_NOWARN, use_coherent);
if (!table->icm[i])
goto err;
if (mlx4_MAP_ICM(dev, table->icm[i], virt + i * MLX4_TABLE_CHUNK_SIZE)) {
- mlx4_free_icm(dev, table->icm[i]);
+ mlx4_free_icm(dev, table->icm[i], use_coherent);
table->icm[i] = NULL;
goto err;
}
@@ -359,7 +420,7 @@
if (table->icm[i]) {
mlx4_UNMAP_ICM(dev, virt + i * MLX4_TABLE_CHUNK_SIZE,
MLX4_TABLE_CHUNK_SIZE / MLX4_ICM_PAGE_SIZE);
- mlx4_free_icm(dev, table->icm[i]);
+ mlx4_free_icm(dev, table->icm[i], use_coherent);
}
return -ENOMEM;
@@ -373,7 +434,7 @@
if (table->icm[i]) {
mlx4_UNMAP_ICM(dev, table->virt + i * MLX4_TABLE_CHUNK_SIZE,
MLX4_TABLE_CHUNK_SIZE / MLX4_ICM_PAGE_SIZE);
- mlx4_free_icm(dev, table->icm[i]);
+ mlx4_free_icm(dev, table->icm[i], table->coherent);
}
kfree(table->icm);