arch/powerpc: Use dma_zalloc_coherent

PATCH Applied Sabyasachi Gupta 5 participants 7 messages Nov 5, 2018 – Nov 27, 2018

Subsystems (2)

PowerPC DMA mapping helpers

lore

Share & tools

lore

Sabyasachi Gupta

author

2018-11-05 03:28 UTC #0 lore raw

PATCH 1/1 arch/powerpc: Use dma_zalloc_coherent

Replaced dma_alloc_coherent + memset with dma_zalloc_coherent

Signed-off-by Sabyasachi Gupta <[email protected]>

---
 arch/powerpc/platforms/pasemi/dma_lib.c | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)

arch/powerpc/platforms/pasemi/dma_lib.c +1 -3

index 53384eb..d18d164 100644
@@ -255,15 +255,13 @@ int pasemi_dma_alloc_ring(struct pasemi_dmachan *chan, int ring_size)
 
 	chan->ring_size = ring_size;
 
-	chan->ring_virt = dma_alloc_coherent(&dma_pdev->dev,
+	chan->ring_virt = dma_zalloc_coherent(&dma_pdev->dev,
 					     ring_size * sizeof(u64),
 					     &chan->ring_dma, GFP_KERNEL);
 
 	if (!chan->ring_virt)
 		return -ENOMEM;
 
-	memset(chan->ring_virt, 0, ring_size * sizeof(u64));
-
 	return 0;
 }
 EXPORT_SYMBOL(pasemi_dma_alloc_ring);

Signature · 1 line

2.7.4

Sabyasachi Gupta

author

2018-11-15 18:00 UTC #1 lore raw

Quoted 4 lines · Sabyasachi Gupta

Replaced dma_alloc_coherent + memset with dma_zalloc_coherent
Signed-off-by Sabyasachi Gupta <[email protected]>

Any comment on this patch?

Quoted 28 lines

---
 arch/powerpc/platforms/pasemi/dma_lib.c | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)

arch/powerpc/platforms/pasemi/dma_lib.c +1 -4

index 53384eb..d18d164 100644
@@ -255,15 +255,13 @@ int pasemi_dma_alloc_ring(struct pasemi_dmachan *chan, int ring_size)
 
        chan->ring_size = ring_size;
 
-       chan->ring_virt = dma_alloc_coherent(&dma_pdev->dev,
+       chan->ring_virt = dma_zalloc_coherent(&dma_pdev->dev,
                                             ring_size * sizeof(u64),
                                             &chan->ring_dma, GFP_KERNEL);
 
        if (!chan->ring_virt)
                return -ENOMEM;
 
-       memset(chan->ring_virt, 0, ring_size * sizeof(u64));
-
        return 0;
 }
 EXPORT_SYMBOL(pasemi_dma_alloc_ring);
--

2.7.4

Joe Perches

2018-11-15 19:25 UTC #2 lore raw

Quoted 7 lines · Sabyasachi Gupta

Quoted 3 lines · Sabyasachi Gupta
Replaced dma_alloc_coherent + memset with dma_zalloc_coherent
Signed-off-by Sabyasachi Gupta <[email protected]>
Any comment on this patch?

It's obviously correct.

You might realign the arguments on the next lines to the open parenthesis.

Perhaps there should be new function calls added for symmetry to the other alloc functions for multiplication overflow protection.

Perhaps:

void *dma_alloc_array_coherent()
void *dma_calloc_coherent()

Something like

---
 include/linux/dma-mapping.h | 19 +++++++++++++++++++
 1 file changed, 19 insertions(+)

include/linux/dma-mapping.h +19 -0

index 15bd41447025..95bebf8883b1 100644
@@ -565,6 +565,25 @@ static inline void *dma_alloc_coherent(struct device *dev, size_t size,
 			(gfp & __GFP_NOWARN) ? DMA_ATTR_NO_WARN : 0);
 }
 
+static inline void *dma_alloc_array_coherent(struct device *dev,
+					     size_t n, size_t size,
+					     dma_addr_t *dma_handle, gfp_t gfp)
+{
+	size_t bytes;
+
+	if (unlikely(check_mul_overflow(n, size, &bytes)))
+		return NULL;
+	return dma_alloc_coherent(dev, bytes, dma_handle, gfp);
+}
+
+static inline void *dma_calloc_coherent(struct device *dev,
+					size_t n, size_t size,
+					dma_addr_t *dma_handle, gfp_t gfp)
+{
+	return dma_alloc_array_coherent(dev, n, size, dma_handle,
+					gfp | __GFP_ZERO);
+}
+
 static inline void dma_free_coherent(struct device *dev, size_t size,
 		void *cpu_addr, dma_addr_t dma_handle)
 {

---

Quoted 1 line

arch/powerpc/platforms/pasemi/dma_lib.c
Quoted 1 line
arch/powerpc/platforms/pasemi/dma_lib.c +0 -0

[]

Quoted 17 lines

arch/powerpc/platforms/pasemi/dma_lib.c +1 -3

@@ -255,15 +255,13 @@ int pasemi_dma_alloc_ring(struct pasemi_dmachan *chan, int ring_size)
 
        chan->ring_size = ring_size;
 
-       chan->ring_virt = dma_alloc_coherent(&dma_pdev->dev,
+       chan->ring_virt = dma_zalloc_coherent(&dma_pdev->dev,
                                             ring_size * sizeof(u64),
                                             &chan->ring_dma, GFP_KERNEL);
 en
        if (!chan->ring_virt)
                return -ENOMEM;
 
-       memset(chan->ring_virt, 0, ring_size * sizeof(u64));
-
        return 0;
 }
 EXPORT_SYMBOL(pasemi_dma_alloc_ring);

Souptick Joarder

2018-11-17 07:06 UTC #3 lore raw

Hi Joe,

Quoted 46 lines · Joe Perches

Quoted 7 lines · Sabyasachi Gupta
Quoted 3 lines · Sabyasachi Gupta
Replaced dma_alloc_coherent + memset with dma_zalloc_coherent
Signed-off-by Sabyasachi Gupta <[email protected]>
Any comment on this patch?
It's obviously correct.
You might realign the arguments on the next lines to the open parenthesis.
Perhaps there should be new function calls added for symmetry to the other alloc functions for multiplication overflow protection.
Perhaps:
void *dma_alloc_array_coherent()
void *dma_calloc_coherent()
Something like
---
 include/linux/dma-mapping.h | 19 +++++++++++++++++++
 1 file changed, 19 insertions(+)
include/linux/dma-mapping.h +9 -0
index 15bd41447025..95bebf8883b1 100644
@@ -565,6 +565,25 @@ static inline void *dma_alloc_coherent(struct device *dev, size_t size,
                        (gfp & __GFP_NOWARN) ? DMA_ATTR_NO_WARN : 0);
 }
 
+static inline void *dma_alloc_array_coherent(struct device *dev,
+                                            size_t n, size_t size,
+                                            dma_addr_t *dma_handle, gfp_t gfp)
+{
+       size_t bytes;
+
+       if (unlikely(check_mul_overflow(n, size, &bytes)))
+               return NULL;
+       return dma_alloc_coherent(dev, bytes, dma_handle, gfp);

include/linux/dma-mapping.h static inline void *dma_alloc_coherent(struct device *dev, size_t size, · line 586

Quoted 10 lines

include/linux/dma-mapping.h +10 -0

+}
+
+static inline void *dma_calloc_coherent(struct device *dev,
+                                       size_t n, size_t size,
+                                       dma_addr_t *dma_handle, gfp_t gfp)
+{
+       return dma_alloc_array_coherent(dev, n, size, dma_handle,
+                                       gfp | __GFP_ZERO);
+}
+

If I understood correctly, you are talking about adding these 2 new inline functions. We can do that.

Can you please help to understand the consumers of these 2 new inline ?

Quoted 26 lines

include/linux/dma-mapping.h +0 -1

 static inline void dma_free_coherent(struct device *dev, size_t size,
                void *cpu_addr, dma_addr_t dma_handle)
 {
 
---

Quoted 1 line

arch/powerpc/platforms/pasemi/dma_lib.c
Quoted 1 line
arch/powerpc/platforms/pasemi/dma_lib.c +0 -0

[]

Quoted 17 lines

arch/powerpc/platforms/pasemi/dma_lib.c +1 -3

@@ -255,15 +255,13 @@ int pasemi_dma_alloc_ring(struct pasemi_dmachan *chan, int ring_size)
 
        chan->ring_size = ring_size;
 
-       chan->ring_virt = dma_alloc_coherent(&dma_pdev->dev,
+       chan->ring_virt = dma_zalloc_coherent(&dma_pdev->dev,
                                             ring_size * sizeof(u64),
                                             &chan->ring_dma, GFP_KERNEL);
 en
        if (!chan->ring_virt)
                return -ENOMEM;
 
-       memset(chan->ring_virt, 0, ring_size * sizeof(u64));
-
        return 0;
 }
 EXPORT_SYMBOL(pasemi_dma_alloc_ring);

Joe Perches

2018-11-17 08:24 UTC #4 lore raw

Quoted 1 line · Souptick Joarder

Hi Joe,

Hi back.

Quoted 60 lines · Joe Perches

Quoted 45 lines · Sabyasachi Gupta
Quoted 7 lines
Quoted 3 lines · Sabyasachi Gupta
Replaced dma_alloc_coherent + memset with dma_zalloc_coherent
Signed-off-by Sabyasachi Gupta <[email protected]>
Any comment on this patch?
It's obviously correct.
You might realign the arguments on the next lines to the open parenthesis.
Perhaps there should be new function calls added for symmetry to the other alloc functions for multiplication overflow protection.
Perhaps:
void *dma_alloc_array_coherent()
void *dma_calloc_coherent()
Something like
---
 include/linux/dma-mapping.h | 19 +++++++++++++++++++
 1 file changed, 19 insertions(+)
include/linux/dma-mapping.h +9 -0
index 15bd41447025..95bebf8883b1 100644
@@ -565,6 +565,25 @@ static inline void *dma_alloc_coherent(struct device *dev, size_t size,
                        (gfp & __GFP_NOWARN) ? DMA_ATTR_NO_WARN : 0);
 }
 
+static inline void *dma_alloc_array_coherent(struct device *dev,
+                                            size_t n, size_t size,
+                                            dma_addr_t *dma_handle, gfp_t gfp)
+{
+       size_t bytes;
+
+       if (unlikely(check_mul_overflow(n, size, &bytes)))
+               return NULL;
+       return dma_alloc_coherent(dev, bytes, dma_handle, gfp);
include/linux/dma-mapping.h static inline void *dma_alloc_coherent(struct device *dev, size_t size, · line 586
Quoted 10 lines
include/linux/dma-mapping.h +10 -0
+}
+
+static inline void *dma_calloc_coherent(struct device *dev,
+                                       size_t n, size_t size,
+                                       dma_addr_t *dma_handle, gfp_t gfp)
+{
+       return dma_alloc_array_coherent(dev, n, size, dma_handle,
+                                       gfp | __GFP_ZERO);
+}
+
If I understood correctly, you are talking about adding these 2 new inline functions. We can do that.

Quoted 1 line

Can you please help to understand the consumers of these 2 new inline ?

Any call to dma_alloc_coherent with a multiply
_might_overflow the multiplication. The
check_mul_overflow simply avoids the overflow.

Michael Ellerman

reviewer

2018-11-16 10:06 UTC #5 lore raw

In reply to Sabyasachi Gupta

Quoted 8 lines · Sabyasachi Gupta

Quoted 4 lines · Sabyasachi Gupta
Replaced dma_alloc_coherent + memset with dma_zalloc_coherent
Signed-off-by Sabyasachi Gupta <[email protected]>
Any comment on this patch?

Wait longer :)

I'm still chasing bugs in 4.20-rc2, I haven't started merging many patches for 4.21 yet.

Your patches are tracked in patchwork here:
https://patchwork.ozlabs.org/project/linuxppc-dev/list/?submitter=75253

If they're still in "new" state around rc5 then feel free to ping me again.

cheers

Quoted 24 lines

arch/powerpc/platforms/pasemi/dma_lib.c +1 -4

index 53384eb..d18d164 100644
@@ -255,15 +255,13 @@ int pasemi_dma_alloc_ring(struct pasemi_dmachan *chan, int ring_size)
 
        chan->ring_size = ring_size;
 
-       chan->ring_virt = dma_alloc_coherent(&dma_pdev->dev,
+       chan->ring_virt = dma_zalloc_coherent(&dma_pdev->dev,
                                             ring_size * sizeof(u64),
                                             &chan->ring_dma, GFP_KERNEL);
 
        if (!chan->ring_virt)
                return -ENOMEM;
 
-       memset(chan->ring_virt, 0, ring_size * sizeof(u64));
-
        return 0;
 }
 EXPORT_SYMBOL(pasemi_dma_alloc_ring);
--

2.7.4

Michael Ellerman

Applied

2018-11-27 09:22 UTC #6 lore raw

In reply to Sabyasachi Gupta

Quoted 3 lines · Sabyasachi Gupta

Replaced dma_alloc_coherent + memset with dma_zalloc_coherent
Signed-off-by Sabyasachi Gupta <[email protected]>

Applied to powerpc next, thanks.

https://git.kernel.org/powerpc/c/c3d6a64bd1e82ea486bff46b17d540

cheers