The MMU mapping flow is optimized for improve the performance, the
unmapping flow could also be optimized to follow same flow.
Signed-off-by: Bingbu Cao <bingbu.cao@xxxxxxxxx>
---
v3: separate patch which make the unmap() as void
v2: squash the ipu6_mmu_pgsize() removal into this patchset
---
drivers/media/pci/intel/ipu6/ipu6-mmu.c | 114 +++++++++++---------------------
1 file changed, 38 insertions(+), 76 deletions(-)
diff --git a/drivers/media/pci/intel/ipu6/ipu6-mmu.c b/drivers/media/pci/intel/ipu6/ipu6-mmu.c
index e957ccb4691d..a8e087188ebb 100644
--- a/drivers/media/pci/intel/ipu6/ipu6-mmu.c
+++ b/drivers/media/pci/intel/ipu6/ipu6-mmu.c
@@ -255,41 +255,49 @@ static u32 *alloc_l2_pt(struct ipu6_mmu_info *mmu_info)
static size_t l2_unmap(struct ipu6_mmu_info *mmu_info, unsigned long iova,
phys_addr_t dummy, size_t size)
{
- u32 l1_idx = iova >> ISP_L1PT_SHIFT;
- u32 iova_start = iova;
+ unsigned int l2_entries;
unsigned int l2_idx;
- size_t unmapped = 0;
unsigned long flags;
+ u32 l1_idx;
u32 *l2_pt;
- dev_dbg(mmu_info->dev, "unmapping l2 page table for l1 index %u (iova 0x%8.8lx)\n",
- l1_idx, iova);
-
spin_lock_irqsave(&mmu_info->lock, flags);
- if (mmu_info->l1_pt[l1_idx] == mmu_info->dummy_l2_pteval) {
- spin_unlock_irqrestore(&mmu_info->lock, flags);
- dev_err(mmu_info->dev,
- "unmap iova 0x%8.8lx l1 idx %u which was not mapped\n",
- iova, l1_idx);
- return 0;
- }
-
- for (l2_idx = (iova_start & ISP_L2PT_MASK) >> ISP_L2PT_SHIFT;
- (iova_start & ISP_L1PT_MASK) + (l2_idx << ISP_PAGE_SHIFT)
- < iova_start + size && l2_idx < ISP_L2PT_PTES; l2_idx++) {
- l2_pt = mmu_info->l2_pts[l1_idx];
+ for (l1_idx = iova >> ISP_L1PT_SHIFT;
+ size > 0 && l1_idx < ISP_L1PT_PTES; l1_idx++) {
dev_dbg(mmu_info->dev,
- "unmap l2 index %u with pteval 0x%10.10llx\n",
- l2_idx, TBL_PHYS_ADDR(l2_pt[l2_idx]));
- l2_pt[l2_idx] = mmu_info->dummy_page_pteval;
+ "unmapping l2 pgtable (l1 index %u (iova 0x%8.8lx))\n",
+ l1_idx, iova);
- clflush_cache_range((void *)&l2_pt[l2_idx],
- sizeof(l2_pt[l2_idx]));
- unmapped++;
+ if (mmu_info->l1_pt[l1_idx] == mmu_info->dummy_l2_pteval) {
+ dev_err(mmu_info->dev,
+ "unmap not mapped iova 0x%8.8lx l1 index %u\n",
+ iova, l1_idx);
+ continue;
+ }
+ l2_pt = mmu_info->l2_pts[l1_idx];
+
+ l2_entries = 0;
+ for (l2_idx = (iova & ISP_L2PT_MASK) >> ISP_L2PT_SHIFT;
+ size > 0 && l2_idx < ISP_L2PT_PTES; l2_idx++) {
+ dev_dbg(mmu_info->dev,
+ "unmap l2 index %u with pteval 0x%10.10llx\n",
+ l2_idx, TBL_PHYS_ADDR(l2_pt[l2_idx]));
+ l2_pt[l2_idx] = mmu_info->dummy_page_pteval;
+
+ iova += ISP_PAGE_SIZE;
+ size -= ISP_PAGE_SIZE;
+
+ l2_entries++;
+ }
+
+ WARN_ON_ONCE(!l2_entries);
+ clflush_cache_range(&l2_pt[l2_idx - l2_entries],
+ sizeof(l2_pt[0]) * l2_entries);
}
+
+ WARN_ON_ONCE(size);
spin_unlock_irqrestore(&mmu_info->lock, flags);
-
- return unmapped << ISP_PAGE_SHIFT;
+ return l2_entries << ISP_PAGE_SHIFT;
}
static int l2_map(struct ipu6_mmu_info *mmu_info, unsigned long iova,
@@ -637,40 +645,13 @@ phys_addr_t ipu6_mmu_iova_to_phys(struct ipu6_mmu_info *mmu_info,
return phy_addr;
}
-static size_t ipu6_mmu_pgsize(unsigned long pgsize_bitmap,
- unsigned long addr_merge, size_t size)
-{
- unsigned int pgsize_idx;
- size_t pgsize;
-
- /* Max page size that still fits into 'size' */
- pgsize_idx = __fls(size);
-
- if (likely(addr_merge)) {
- /* Max page size allowed by address */
- unsigned int align_pgsize_idx = __ffs(addr_merge);
-
- pgsize_idx = min(pgsize_idx, align_pgsize_idx);
- }
-
- pgsize = (1UL << (pgsize_idx + 1)) - 1;
- pgsize &= pgsize_bitmap;
-
- WARN_ON(!pgsize);
-
- /* pick the biggest page */
- pgsize_idx = __fls(pgsize);
- pgsize = 1UL << pgsize_idx;
-
- return pgsize;
-}
-
size_t ipu6_mmu_unmap(struct ipu6_mmu_info *mmu_info, unsigned long iova,
size_t size)
{
- size_t unmapped_page, unmapped = 0;
unsigned int min_pagesz;
+ dev_dbg(mmu_info->dev, "unmapping iova 0x%lx size 0x%zx\n", iova, size);
+
/* find out the minimum page size supported */
min_pagesz = 1 << __ffs(mmu_info->pgsize_bitmap);
@@ -682,29 +663,10 @@ size_t ipu6_mmu_unmap(struct ipu6_mmu_info *mmu_info, unsigned long iova,
if (!IS_ALIGNED(iova | size, min_pagesz)) {
dev_err(NULL, "unaligned: iova 0x%lx size 0x%zx min_pagesz 0x%x\n",
iova, size, min_pagesz);
- return -EINVAL;
+ return 0;
}
- /*
- * Keep iterating until we either unmap 'size' bytes (or more)
- * or we hit an area that isn't mapped.
- */
- while (unmapped < size) {
- size_t pgsize = ipu6_mmu_pgsize(mmu_info->pgsize_bitmap,
- iova, size - unmapped);
-
- unmapped_page = __ipu6_mmu_unmap(mmu_info, iova, pgsize);
- if (!unmapped_page)
- break;
-
- dev_dbg(mmu_info->dev, "unmapped: iova 0x%lx size 0x%zx\n",
- iova, unmapped_page);
-
- iova += unmapped_page;
- unmapped += unmapped_page;
- }
-
- return unmapped;
+ return __ipu6_mmu_unmap(mmu_info, iova, size);
}
int ipu6_mmu_map(struct ipu6_mmu_info *mmu_info, unsigned long iova,
--
2.7.4
