Commit 0e4bc95d authored by Anton Blanchard's avatar Anton Blanchard Committed by Benjamin Herrenschmidt

powerpc/iommu: Reduce spinlock coverage in iommu_alloc and iommu_free

We currently hold the IOMMU spinlock around tce_build and tce_flush.
This causes our spinlock hold times to be much higher than required
and can impact multiqueue adapters.

This patch moves tce_build and tce_flush outside of the lock in
iommu_alloc, and tce_flush outside of the lock in iommu_free.

Some performance numbers were obtained with a Chelsio T3 adapter on
two POWER7 boxes, running a 100 session TCP round robin test.

Performance improved 32% with this patch applied.
Signed-off-by: default avatarAnton Blanchard <anton@samba.org>
Signed-off-by: default avatarBenjamin Herrenschmidt <benh@kernel.crashing.org>
parent c1703e85
...@@ -170,13 +170,11 @@ static dma_addr_t iommu_alloc(struct device *dev, struct iommu_table *tbl, ...@@ -170,13 +170,11 @@ static dma_addr_t iommu_alloc(struct device *dev, struct iommu_table *tbl,
int build_fail; int build_fail;
spin_lock_irqsave(&(tbl->it_lock), flags); spin_lock_irqsave(&(tbl->it_lock), flags);
entry = iommu_range_alloc(dev, tbl, npages, NULL, mask, align_order); entry = iommu_range_alloc(dev, tbl, npages, NULL, mask, align_order);
spin_unlock_irqrestore(&(tbl->it_lock), flags);
if (unlikely(entry == DMA_ERROR_CODE)) { if (unlikely(entry == DMA_ERROR_CODE))
spin_unlock_irqrestore(&(tbl->it_lock), flags);
return DMA_ERROR_CODE; return DMA_ERROR_CODE;
}
entry += tbl->it_offset; /* Offset into real TCE table */ entry += tbl->it_offset; /* Offset into real TCE table */
ret = entry << IOMMU_PAGE_SHIFT; /* Set the return dma address */ ret = entry << IOMMU_PAGE_SHIFT; /* Set the return dma address */
...@@ -192,9 +190,10 @@ static dma_addr_t iommu_alloc(struct device *dev, struct iommu_table *tbl, ...@@ -192,9 +190,10 @@ static dma_addr_t iommu_alloc(struct device *dev, struct iommu_table *tbl,
* not altered. * not altered.
*/ */
if (unlikely(build_fail)) { if (unlikely(build_fail)) {
spin_lock_irqsave(&(tbl->it_lock), flags);
__iommu_free(tbl, ret, npages); __iommu_free(tbl, ret, npages);
spin_unlock_irqrestore(&(tbl->it_lock), flags); spin_unlock_irqrestore(&(tbl->it_lock), flags);
return DMA_ERROR_CODE; return DMA_ERROR_CODE;
} }
...@@ -202,8 +201,6 @@ static dma_addr_t iommu_alloc(struct device *dev, struct iommu_table *tbl, ...@@ -202,8 +201,6 @@ static dma_addr_t iommu_alloc(struct device *dev, struct iommu_table *tbl,
if (ppc_md.tce_flush) if (ppc_md.tce_flush)
ppc_md.tce_flush(tbl); ppc_md.tce_flush(tbl);
spin_unlock_irqrestore(&(tbl->it_lock), flags);
/* Make sure updates are seen by hardware */ /* Make sure updates are seen by hardware */
mb(); mb();
...@@ -244,8 +241,8 @@ static void iommu_free(struct iommu_table *tbl, dma_addr_t dma_addr, ...@@ -244,8 +241,8 @@ static void iommu_free(struct iommu_table *tbl, dma_addr_t dma_addr,
unsigned long flags; unsigned long flags;
spin_lock_irqsave(&(tbl->it_lock), flags); spin_lock_irqsave(&(tbl->it_lock), flags);
__iommu_free(tbl, dma_addr, npages); __iommu_free(tbl, dma_addr, npages);
spin_unlock_irqrestore(&(tbl->it_lock), flags);
/* Make sure TLB cache is flushed if the HW needs it. We do /* Make sure TLB cache is flushed if the HW needs it. We do
* not do an mb() here on purpose, it is not needed on any of * not do an mb() here on purpose, it is not needed on any of
...@@ -253,8 +250,6 @@ static void iommu_free(struct iommu_table *tbl, dma_addr_t dma_addr, ...@@ -253,8 +250,6 @@ static void iommu_free(struct iommu_table *tbl, dma_addr_t dma_addr,
*/ */
if (ppc_md.tce_flush) if (ppc_md.tce_flush)
ppc_md.tce_flush(tbl); ppc_md.tce_flush(tbl);
spin_unlock_irqrestore(&(tbl->it_lock), flags);
} }
int iommu_map_sg(struct device *dev, struct iommu_table *tbl, int iommu_map_sg(struct device *dev, struct iommu_table *tbl,
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment