[PATCH v4 0/2] swiotlb: allocate padding slots if necessary

[PATCH v4 0/2] swiotlb: allocate padding slots if necessary

[Petr Tesarik]

From: Petr Tesarik <petr.tesarik1@huawei-partners.com>

If the allocation alignment is bigger than IO_TLB_SIZE and min_align_mask
covers some bits in the original address between IO_TLB_SIZE and
alloc_align_mask, preserve these bits by allocating additional padding
slots before the actual swiotlb buffer.

Changes from v3
---------------
* Add a parameter to swiotlb_align_offset(), so it is not open-coded in
  swiotlb_tbl_map_single().
* Extend the description of swiotlb_align_offset() and convert it to the
  kernel-doc format.
* Use bit operations to calculate number of padding slots and the padding
  inside a slot. This matches all other such calculations in swiotlb.c.

Changes from v2
---------------
* Fix assignment of an uninitialized variable to pad_slots.
* Improve commit message wrt INVALID_PHYS_ADDR.

Changes from v1
---------------
* Rename padding to pad_slots.
* Set pad_slots only for the first allocated non-padding slot.
* Do not bother initializing orig_addr to INVALID_PHYS_ADDR.
* Change list and pad_slots to unsigned short to avoid growing
  struct io_tlb_slot on 32-bit targets.
* Add build-time check that list and pad_slots can hold the maximum
  allowed value of IO_TLB_SEGSIZE.


Petr Tesarik (2):
  swiotlb: extend buffer pre-padding to alloc_align_mask if necessary
  bug: introduce ASSERT_VAR_CAN_HOLD()

 include/linux/build_bug.h | 10 +++++++
 kernel/dma/swiotlb.c      | 61 ++++++++++++++++++++++++++++++---------
 2 files changed, 58 insertions(+), 13 deletions(-)

-- 
2.34.1

[PATCH v4 1/2] swiotlb: extend buffer pre-padding to alloc_align_mask if necessary

[Petr Tesarik]

From: Petr Tesarik <petr.tesarik1@huawei-partners.com>

Allow a buffer pre-padding of up to alloc_align_mask, even if it requires
allocating additional IO TLB slots.

If the allocation alignment is bigger than IO_TLB_SIZE and min_align_mask
covers any non-zero bits in the original address between IO_TLB_SIZE and
alloc_align_mask, these bits are not preserved in the swiotlb buffer
address.

To fix this case, increase the allocation size and use a larger offset
within the allocated buffer. As a result, extra padding slots may be
allocated before the mapping start address.

Leave orig_addr in these padding slots initialized to INVALID_PHYS_ADDR.
These slots do not correspond to any CPU buffer, so attempts to sync the
data should be ignored.

The padding slots should be automatically released when the buffer is
unmapped. However, swiotlb_tbl_unmap_single() takes only the address of the
DMA buffer slot, not the first padding slot. Save the number of padding
slots in struct io_tlb_slot and use it to adjust the slot index in
swiotlb_release_slots(), so all allocated slots are properly freed.

Fixes: 2fd4fa5d3fb5 ("swiotlb: Fix alignment checks when both allocation and DMA masks are present")
Link: https://lore.kernel.org/linux-iommu/20240311210507.217daf8b@meshulam.tesarici.cz/
Signed-off-by: Petr Tesarik <petr.tesarik1@huawei-partners.com>
---
 kernel/dma/swiotlb.c | 59 ++++++++++++++++++++++++++++++++++----------
 1 file changed, 46 insertions(+), 13 deletions(-)

diff --git a/kernel/dma/swiotlb.c b/kernel/dma/swiotlb.c
index 86fe172b5958..d7a8cb93ef2d 100644
--- a/kernel/dma/swiotlb.c
+++ b/kernel/dma/swiotlb.c
@@ -69,11 +69,14 @@
  * @alloc_size:	Size of the allocated buffer.
  * @list:	The free list describing the number of free entries available
  *		from each index.
+ * @pad_slots:	Number of preceding padding slots. Valid only in the first
+ *		allocated non-padding slot.
  */
 struct io_tlb_slot {
 	phys_addr_t orig_addr;
 	size_t alloc_size;
-	unsigned int list;
+	unsigned short list;
+	unsigned short pad_slots;
 };
 
 static bool swiotlb_force_bounce;
@@ -287,6 +290,7 @@ static void swiotlb_init_io_tlb_pool(struct io_tlb_pool *mem, phys_addr_t start,
 					 mem->nslabs - i);
 		mem->slots[i].orig_addr = INVALID_PHYS_ADDR;
 		mem->slots[i].alloc_size = 0;
+		mem->slots[i].pad_slots = 0;
 	}
 
 	memset(vaddr, 0, bytes);
@@ -821,12 +825,30 @@ void swiotlb_dev_init(struct device *dev)
 #endif
 }
 
-/*
- * Return the offset into a iotlb slot required to keep the device happy.
+/**
+ * swiotlb_align_offset() - Get required offset into an IO TLB allocation.
+ * @dev:         Owning device.
+ * @align_mask:  Allocation alignment mask.
+ * @addr:        DMA address.
+ *
+ * Return the minimum offset from the start of an IO TLB allocation which is
+ * required for a given buffer address and allocation alignment to keep the
+ * device happy.
+ *
+ * First, the address bits covered by min_align_mask must be identical in the
+ * original address and the bounce buffer address. High bits are preserved by
+ * choosing a suitable IO TLB slot, but bits below IO_TLB_SHIFT require extra
+ * padding bytes before the bounce buffer.
+ *
+ * Second, @align_mask specifies which bits of the first allocated slot must
+ * be zero. This may require allocating additional padding slots, and then the
+ * offset (in bytes) from the first such padding slot is returned.
  */
-static unsigned int swiotlb_align_offset(struct device *dev, u64 addr)
+static unsigned int swiotlb_align_offset(struct device *dev,
+					 unsigned int align_mask, u64 addr)
 {
-	return addr & dma_get_min_align_mask(dev) & (IO_TLB_SIZE - 1);
+	return addr & dma_get_min_align_mask(dev) &
+		(align_mask | (IO_TLB_SIZE - 1));
 }
 
 /*
@@ -847,7 +869,7 @@ static void swiotlb_bounce(struct device *dev, phys_addr_t tlb_addr, size_t size
 		return;
 
 	tlb_offset = tlb_addr & (IO_TLB_SIZE - 1);
-	orig_addr_offset = swiotlb_align_offset(dev, orig_addr);
+	orig_addr_offset = swiotlb_align_offset(dev, 0, orig_addr);
 	if (tlb_offset < orig_addr_offset) {
 		dev_WARN_ONCE(dev, 1,
 			"Access before mapping start detected. orig offset %u, requested offset %u.\n",
@@ -1005,7 +1027,7 @@ static int swiotlb_search_pool_area(struct device *dev, struct io_tlb_pool *pool
 	unsigned long max_slots = get_max_slots(boundary_mask);
 	unsigned int iotlb_align_mask = dma_get_min_align_mask(dev);
 	unsigned int nslots = nr_slots(alloc_size), stride;
-	unsigned int offset = swiotlb_align_offset(dev, orig_addr);
+	unsigned int offset = swiotlb_align_offset(dev, 0, orig_addr);
 	unsigned int index, slots_checked, count = 0, i;
 	unsigned long flags;
 	unsigned int slot_base;
@@ -1328,11 +1350,12 @@ phys_addr_t swiotlb_tbl_map_single(struct device *dev, phys_addr_t orig_addr,
 		unsigned long attrs)
 {
 	struct io_tlb_mem *mem = dev->dma_io_tlb_mem;
-	unsigned int offset = swiotlb_align_offset(dev, orig_addr);
+	unsigned int offset;
 	struct io_tlb_pool *pool;
 	unsigned int i;
 	int index;
 	phys_addr_t tlb_addr;
+	unsigned short pad_slots;
 
 	if (!mem || !mem->nslabs) {
 		dev_warn_ratelimited(dev,
@@ -1349,6 +1372,7 @@ phys_addr_t swiotlb_tbl_map_single(struct device *dev, phys_addr_t orig_addr,
 		return (phys_addr_t)DMA_MAPPING_ERROR;
 	}
 
+	offset = swiotlb_align_offset(dev, alloc_align_mask, orig_addr);
 	index = swiotlb_find_slots(dev, orig_addr,
 				   alloc_size + offset, alloc_align_mask, &pool);
 	if (index == -1) {
@@ -1364,6 +1388,10 @@ phys_addr_t swiotlb_tbl_map_single(struct device *dev, phys_addr_t orig_addr,
 	 * This is needed when we sync the memory.  Then we sync the buffer if
 	 * needed.
 	 */
+	pad_slots = offset >> IO_TLB_SHIFT;
+	offset &= (IO_TLB_SIZE - 1);
+	index += pad_slots;
+	pool->slots[index].pad_slots = pad_slots;
 	for (i = 0; i < nr_slots(alloc_size + offset); i++)
 		pool->slots[index + i].orig_addr = slot_addr(orig_addr, i);
 	tlb_addr = slot_addr(pool->start, index) + offset;
@@ -1384,13 +1412,17 @@ static void swiotlb_release_slots(struct device *dev, phys_addr_t tlb_addr)
 {
 	struct io_tlb_pool *mem = swiotlb_find_pool(dev, tlb_addr);
 	unsigned long flags;
-	unsigned int offset = swiotlb_align_offset(dev, tlb_addr);
-	int index = (tlb_addr - offset - mem->start) >> IO_TLB_SHIFT;
-	int nslots = nr_slots(mem->slots[index].alloc_size + offset);
-	int aindex = index / mem->area_nslabs;
-	struct io_tlb_area *area = &mem->areas[aindex];
+	unsigned int offset = swiotlb_align_offset(dev, 0, tlb_addr);
+	int index, nslots, aindex;
+	struct io_tlb_area *area;
 	int count, i;
 
+	index = (tlb_addr - offset - mem->start) >> IO_TLB_SHIFT;
+	index -= mem->slots[index].pad_slots;
+	nslots = nr_slots(mem->slots[index].alloc_size + offset);
+	aindex = index / mem->area_nslabs;
+	area = &mem->areas[aindex];
+
 	/*
 	 * Return the buffer to the free list by setting the corresponding
 	 * entries to indicate the number of contiguous entries available.
@@ -1413,6 +1445,7 @@ static void swiotlb_release_slots(struct device *dev, phys_addr_t tlb_addr)
 		mem->slots[i].list = ++count;
 		mem->slots[i].orig_addr = INVALID_PHYS_ADDR;
 		mem->slots[i].alloc_size = 0;
+		mem->slots[i].pad_slots = 0;
 	}
 
 	/*
-- 
2.34.1

RE: [PATCH v4 1/2] swiotlb: extend buffer pre-padding to alloc_align_mask if necessary

[Michael Kelley]

From: Petr Tesarik <petrtesarik@huaweicloud.com> Sent: Monday, March 25, 2024 1:31 AM
> 
> Allow a buffer pre-padding of up to alloc_align_mask, even if it requires
> allocating additional IO TLB slots.
> 
> If the allocation alignment is bigger than IO_TLB_SIZE and min_align_mask
> covers any non-zero bits in the original address between IO_TLB_SIZE and
> alloc_align_mask, these bits are not preserved in the swiotlb buffer
> address.
> 
> To fix this case, increase the allocation size and use a larger offset
> within the allocated buffer. As a result, extra padding slots may be
> allocated before the mapping start address.
> 
> Leave orig_addr in these padding slots initialized to INVALID_PHYS_ADDR.
> These slots do not correspond to any CPU buffer, so attempts to sync the
> data should be ignored.
> 
> The padding slots should be automatically released when the buffer is
> unmapped. However, swiotlb_tbl_unmap_single() takes only the address of the
> DMA buffer slot, not the first padding slot. Save the number of padding
> slots in struct io_tlb_slot and use it to adjust the slot index in
> swiotlb_release_slots(), so all allocated slots are properly freed.
> 
> Fixes: 2fd4fa5d3fb5 ("swiotlb: Fix alignment checks when both allocation and
> DMA masks are present")
> Link: https://lore.kernel.org/linux-iommu/20240311210507.217daf8b@meshulam.tesarici.cz/
> Signed-off-by: Petr Tesarik <petr.tesarik1@huawei-partners.com>
> ---
>  kernel/dma/swiotlb.c | 59 ++++++++++++++++++++++++++++++++++-------
> ---
>  1 file changed, 46 insertions(+), 13 deletions(-)
> 
> diff --git a/kernel/dma/swiotlb.c b/kernel/dma/swiotlb.c
> index 86fe172b5958..d7a8cb93ef2d 100644
> --- a/kernel/dma/swiotlb.c
> +++ b/kernel/dma/swiotlb.c
> @@ -69,11 +69,14 @@
>   * @alloc_size:	Size of the allocated buffer.
>   * @list:	The free list describing the number of free entries available
>   *		from each index.
> + * @pad_slots:	Number of preceding padding slots. Valid only in the first
> + *		allocated non-padding slot.
>   */
>  struct io_tlb_slot {
>  	phys_addr_t orig_addr;
>  	size_t alloc_size;
> -	unsigned int list;
> +	unsigned short list;
> +	unsigned short pad_slots;
>  };
> 
>  static bool swiotlb_force_bounce;
> @@ -287,6 +290,7 @@ static void swiotlb_init_io_tlb_pool(struct io_tlb_pool *mem, phys_addr_t start,
>  					 mem->nslabs - i);
>  		mem->slots[i].orig_addr = INVALID_PHYS_ADDR;
>  		mem->slots[i].alloc_size = 0;
> +		mem->slots[i].pad_slots = 0;
>  	}
> 
>  	memset(vaddr, 0, bytes);
> @@ -821,12 +825,30 @@ void swiotlb_dev_init(struct device *dev)
>  #endif
>  }
> 
> -/*
> - * Return the offset into a iotlb slot required to keep the device happy.
> +/**
> + * swiotlb_align_offset() - Get required offset into an IO TLB allocation.
> + * @dev:         Owning device.
> + * @align_mask:  Allocation alignment mask.
> + * @addr:        DMA address.
> + *
> + * Return the minimum offset from the start of an IO TLB allocation which is
> + * required for a given buffer address and allocation alignment to keep the
> + * device happy.
> + *
> + * First, the address bits covered by min_align_mask must be identical in the
> + * original address and the bounce buffer address. High bits are preserved by
> + * choosing a suitable IO TLB slot, but bits below IO_TLB_SHIFT require extra
> + * padding bytes before the bounce buffer.
> + *
> + * Second, @align_mask specifies which bits of the first allocated slot must
> + * be zero. This may require allocating additional padding slots, and then the
> + * offset (in bytes) from the first such padding slot is returned.
>   */
> -static unsigned int swiotlb_align_offset(struct device *dev, u64 addr)
> +static unsigned int swiotlb_align_offset(struct device *dev,
> +					 unsigned int align_mask, u64 addr)
>  {
> -	return addr & dma_get_min_align_mask(dev) & (IO_TLB_SIZE - 1);
> +	return addr & dma_get_min_align_mask(dev) &
> +		(align_mask | (IO_TLB_SIZE - 1));
>  }
> 
>  /*
> @@ -847,7 +869,7 @@ static void swiotlb_bounce(struct device *dev, phys_addr_t tlb_addr, size_t size
>  		return;
> 
>  	tlb_offset = tlb_addr & (IO_TLB_SIZE - 1);
> -	orig_addr_offset = swiotlb_align_offset(dev, orig_addr);
> +	orig_addr_offset = swiotlb_align_offset(dev, 0, orig_addr);
>  	if (tlb_offset < orig_addr_offset) {
>  		dev_WARN_ONCE(dev, 1,
>  			"Access before mapping start detected. orig offset %u, requested offset %u.\n",
> @@ -1005,7 +1027,7 @@ static int swiotlb_search_pool_area(struct device *dev, struct io_tlb_pool *pool
>  	unsigned long max_slots = get_max_slots(boundary_mask);
>  	unsigned int iotlb_align_mask = dma_get_min_align_mask(dev);
>  	unsigned int nslots = nr_slots(alloc_size), stride;
> -	unsigned int offset = swiotlb_align_offset(dev, orig_addr);
> +	unsigned int offset = swiotlb_align_offset(dev, 0, orig_addr);
>  	unsigned int index, slots_checked, count = 0, i;
>  	unsigned long flags;
>  	unsigned int slot_base;
> @@ -1328,11 +1350,12 @@ phys_addr_t swiotlb_tbl_map_single(struct device *dev, phys_addr_t orig_addr,
>  		unsigned long attrs)
>  {
>  	struct io_tlb_mem *mem = dev->dma_io_tlb_mem;
> -	unsigned int offset = swiotlb_align_offset(dev, orig_addr);
> +	unsigned int offset;
>  	struct io_tlb_pool *pool;
>  	unsigned int i;
>  	int index;
>  	phys_addr_t tlb_addr;
> +	unsigned short pad_slots;
> 
>  	if (!mem || !mem->nslabs) {
>  		dev_warn_ratelimited(dev,
> @@ -1349,6 +1372,7 @@ phys_addr_t swiotlb_tbl_map_single(struct device *dev, phys_addr_t orig_addr,
>  		return (phys_addr_t)DMA_MAPPING_ERROR;
>  	}
> 
> +	offset = swiotlb_align_offset(dev, alloc_align_mask, orig_addr);
>  	index = swiotlb_find_slots(dev, orig_addr,
>  				   alloc_size + offset, alloc_align_mask, &pool);
>  	if (index == -1) {
> @@ -1364,6 +1388,10 @@ phys_addr_t swiotlb_tbl_map_single(struct device *dev, phys_addr_t orig_addr,
>  	 * This is needed when we sync the memory.  Then we sync the buffer if
>  	 * needed.
>  	 */
> +	pad_slots = offset >> IO_TLB_SHIFT;
> +	offset &= (IO_TLB_SIZE - 1);
> +	index += pad_slots;
> +	pool->slots[index].pad_slots = pad_slots;
>  	for (i = 0; i < nr_slots(alloc_size + offset); i++)
>  		pool->slots[index + i].orig_addr = slot_addr(orig_addr, i);
>  	tlb_addr = slot_addr(pool->start, index) + offset;
> @@ -1384,13 +1412,17 @@ static void swiotlb_release_slots(struct device *dev, phys_addr_t tlb_addr)
>  {
>  	struct io_tlb_pool *mem = swiotlb_find_pool(dev, tlb_addr);
>  	unsigned long flags;
> -	unsigned int offset = swiotlb_align_offset(dev, tlb_addr);
> -	int index = (tlb_addr - offset - mem->start) >> IO_TLB_SHIFT;
> -	int nslots = nr_slots(mem->slots[index].alloc_size + offset);
> -	int aindex = index / mem->area_nslabs;
> -	struct io_tlb_area *area = &mem->areas[aindex];
> +	unsigned int offset = swiotlb_align_offset(dev, 0, tlb_addr);
> +	int index, nslots, aindex;
> +	struct io_tlb_area *area;
>  	int count, i;
> 
> +	index = (tlb_addr - offset - mem->start) >> IO_TLB_SHIFT;
> +	index -= mem->slots[index].pad_slots;
> +	nslots = nr_slots(mem->slots[index].alloc_size + offset);
> +	aindex = index / mem->area_nslabs;
> +	area = &mem->areas[aindex];
> +
>  	/*
>  	 * Return the buffer to the free list by setting the corresponding
>  	 * entries to indicate the number of contiguous entries available.
> @@ -1413,6 +1445,7 @@ static void swiotlb_release_slots(struct device *dev, phys_addr_t tlb_addr)
>  		mem->slots[i].list = ++count;
>  		mem->slots[i].orig_addr = INVALID_PHYS_ADDR;
>  		mem->slots[i].alloc_size = 0;
> +		mem->slots[i].pad_slots = 0;
>  	}
> 
>  	/*
> --
> 2.34.1

Did the same testing as on v3, covering both x86/x64 and arm64.
All looks good.

Tested-by: Michael Kelley <mhklinux@outlook.com>
Reviewed-by: Michael Kelley <mhklinux@outlook.com>

Re: [PATCH v4 1/2] swiotlb: extend buffer pre-padding to alloc_align_mask if necessary

[Christoph Hellwig]

Thanks,

applied to the dma-mapping tree for Linux 6.9-rc.

[PATCH v4 2/2] bug: introduce ASSERT_VAR_CAN_HOLD()

[Petr Tesarik]

From: Petr Tesarik <petr.tesarik1@huawei-partners.com>

Introduce an ASSERT_VAR_CAN_HOLD() macro to check at build time that a
variable can hold the given value.

Use this macro in swiotlb to make sure that the list and pad_slots fields
of struct io_tlb_slot are big enough to hold the maximum possible value of
IO_TLB_SEGSIZE.

Signed-off-by: Petr Tesarik <petr.tesarik1@huawei-partners.com>
---
 include/linux/build_bug.h | 10 ++++++++++
 kernel/dma/swiotlb.c      |  2 ++
 2 files changed, 12 insertions(+)

diff --git a/include/linux/build_bug.h b/include/linux/build_bug.h
index 3aa3640f8c18..6e2486508af0 100644
--- a/include/linux/build_bug.h
+++ b/include/linux/build_bug.h
@@ -86,4 +86,14 @@
 		"Offset of " #field " in " #type " has changed.")
 
 
+/*
+ * Compile time check that a variable can hold the given value
+ */
+#define ASSERT_VAR_CAN_HOLD(var, value) ({		\
+	typeof(value) __val = (value);			\
+	typeof(var) __tmp = __val;			\
+	BUILD_BUG_ON_MSG(__tmp != __val,		\
+		#var " cannot hold " #value ".");	\
+})
+
 #endif	/* _LINUX_BUILD_BUG_H */
diff --git a/kernel/dma/swiotlb.c b/kernel/dma/swiotlb.c
index d7a8cb93ef2d..a34ea2c35446 100644
--- a/kernel/dma/swiotlb.c
+++ b/kernel/dma/swiotlb.c
@@ -285,6 +285,8 @@ static void swiotlb_init_io_tlb_pool(struct io_tlb_pool *mem, phys_addr_t start,
 		mem->areas[i].used = 0;
 	}
 
+	ASSERT_VAR_CAN_HOLD(mem->slots[0].list, IO_TLB_SEGSIZE);
+	ASSERT_VAR_CAN_HOLD(mem->slots[0].pad_slots, IO_TLB_SEGSIZE);
 	for (i = 0; i < mem->nslabs; i++) {
 		mem->slots[i].list = min(IO_TLB_SEGSIZE - io_tlb_offset(i),
 					 mem->nslabs - i);
-- 
2.34.1

Re: [PATCH v4 2/2] bug: introduce ASSERT_VAR_CAN_HOLD()

[Christoph Hellwig]

On Mon, Mar 25, 2024 at 09:31:05AM +0100, Petr Tesarik wrote:
> From: Petr Tesarik <petr.tesarik1@huawei-partners.com>
> 
> Introduce an ASSERT_VAR_CAN_HOLD() macro to check at build time that a
> variable can hold the given value.

This really should be run past whoever maintains build_bug.h and
also cc linux-kernel.  Please also split the addition and the user
in swiotlb into separate patches.

I plan to pick up patch 1 for now, and wait for the discussion on
this addition.

Re: [PATCH v4 2/2] bug: introduce ASSERT_VAR_CAN_HOLD()

[Petr Tesařík]

On Tue, 26 Mar 2024 08:56:06 +0100
Christoph Hellwig <hch@lst.de> wrote:

> On Mon, Mar 25, 2024 at 09:31:05AM +0100, Petr Tesarik wrote:
> > From: Petr Tesarik <petr.tesarik1@huawei-partners.com>
> > 
> > Introduce an ASSERT_VAR_CAN_HOLD() macro to check at build time that a
> > variable can hold the given value.  
> 
> This really should be run past whoever maintains build_bug.h and
> also cc linux-kernel.  Please also split the addition and the user
> in swiotlb into separate patches.
> 
> I plan to pick up patch 1 for now, and wait for the discussion on
> this addition.

Fair enough. The use in swiotlb depends on patch 1 from this series,
but you're right, I can send the macro addition as a separate patch to
get some feedback.

Thank you!
Petr T