Skip to contentSkip to conversation

net: octeontx2: Make sure the buffer is 128 byte aligned

PATCH Applied Kevin Hao 5 participants 6 messages Jan 21, 2021 – Jan 23, 2021

Subsystems (3)
lore
Share & tools
lore

Kevin Hao

author
#0 lore raw

PATCH 1/1 net: octeontx2: Make sure the buffer is 128 byte aligned

The octeontx2 hardware needs the buffer to be 128 byte aligned. But in the current implementation of napi_alloc_frag(), it can't guarantee the return address is 128 byte aligned even the request size is a multiple of 128 bytes, so we have to request an extra 128 bytes and use the PTR_ALIGN() to make sure that the buffer is aligned correctly.

Fixes 7a36e4918e30 ("octeontx2-pf: Use the napi_alloc_frag() to alloc the pool buffers")
---
 drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.c +2 -1
index bdfa2e293531..5ddedc3b754d 100644
@@ -488,10 +488,11 @@ dma_addr_t __otx2_alloc_rbuf(struct otx2_nic *pfvf, struct otx2_pool *pool)
 	dma_addr_t iova;
 	u8 *buf;
 
-	buf = napi_alloc_frag(pool->rbsize);
+	buf = napi_alloc_frag(pool->rbsize + OTX2_ALIGN);
 	if (unlikely(!buf))
 		return -ENOMEM;
 
+	buf = PTR_ALIGN(buf, OTX2_ALIGN);
 	iova = dma_map_single_attrs(pfvf->dev, buf, pool->rbsize,
 				    DMA_FROM_DEVICE, DMA_ATTR_SKIP_CPU_SYNC);
 	if (unlikely(dma_mapping_error(pfvf->dev, iova))) {
Signature · 1 line
2.29.2
Tested-by Subbaraya Sundeep on v1 1/1
#1 lore raw

Hi Kevin,

Tested at my end and works fine. Thanks for the patch.

Sundeep

Quoted 34 lines · Kevin Hao

The octeontx2 hardware needs the buffer to be 128 byte aligned. But in the current implementation of napi_alloc_frag(), it can't guarantee the return address is 128 byte aligned even the request size is a multiple of 128 bytes, so we have to request an extra 128 bytes and use the PTR_ALIGN() to make sure that the buffer is aligned correctly.

Fixes 7a36e4918e30 ("octeontx2-pf: Use the napi_alloc_frag() to alloc the pool buffers")
---
 drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.c +2 -2
index bdfa2e293531..5ddedc3b754d 100644
@@ -488,10 +488,11 @@ dma_addr_t __otx2_alloc_rbuf(struct otx2_nic *pfvf, struct otx2_pool *pool)
        dma_addr_t iova;
        u8 *buf;
 
-       buf = napi_alloc_frag(pool->rbsize);
+       buf = napi_alloc_frag(pool->rbsize + OTX2_ALIGN);
        if (unlikely(!buf))
                return -ENOMEM;
 
+       buf = PTR_ALIGN(buf, OTX2_ALIGN);
        iova = dma_map_single_attrs(pfvf->dev, buf, pool->rbsize,
                                    DMA_FROM_DEVICE, DMA_ATTR_SKIP_CPU_SYNC);
        if (unlikely(dma_mapping_error(pfvf->dev, iova))) {
--

2.29.2

#2 lore raw
In reply to Kevin Hao

From: Kevin Hao

Quoted 34 lines

Sent: 21 January 2021 07:09

The octeontx2 hardware needs the buffer to be 128 byte aligned. But in the current implementation of napi_alloc_frag(), it can't guarantee the return address is 128 byte aligned even the request size is a multiple of 128 bytes, so we have to request an extra 128 bytes and use the PTR_ALIGN() to make sure that the buffer is aligned correctly.

Fixes 7a36e4918e30 ("octeontx2-pf: Use the napi_alloc_frag() to alloc the pool buffers")
---
 drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.c +2 -2
index bdfa2e293531..5ddedc3b754d 100644
@@ -488,10 +488,11 @@ dma_addr_t __otx2_alloc_rbuf(struct otx2_nic *pfvf, struct otx2_pool *pool)
 	dma_addr_t iova;
 	u8 *buf;
 
-	buf = napi_alloc_frag(pool->rbsize);
+	buf = napi_alloc_frag(pool->rbsize + OTX2_ALIGN);
 	if (unlikely(!buf))
 		return -ENOMEM;
 
+	buf = PTR_ALIGN(buf, OTX2_ALIGN);
 	iova = dma_map_single_attrs(pfvf->dev, buf, pool->rbsize,
 				    DMA_FROM_DEVICE, DMA_ATTR_SKIP_CPU_SYNC);
 	if (unlikely(dma_mapping_error(pfvf->dev, iova))) {
--
Quoted 1 line

2.29.2

Doesn't that break the 'free' code ?
Surely it needs the original pointer.

It isn't obvious that page_frag_free() it correct when the allocator is napi_alloc_frag() either.
I'd have thought it ought to be returned to the pool.

David

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK Registration No: 1397386 (Wales)

author
#3 lore raw
Quoted 30 lines · David Laight

From: Kevin Hao

Quoted 29 lines

Sent: 21 January 2021 07:09

The octeontx2 hardware needs the buffer to be 128 byte aligned. But in the current implementation of napi_alloc_frag(), it can't guarantee the return address is 128 byte aligned even the request size is a multiple of 128 bytes, so we have to request an extra 128 bytes and use the PTR_ALIGN() to make sure that the buffer is aligned correctly.

Fixes 7a36e4918e30 ("octeontx2-pf: Use the napi_alloc_frag() to alloc the pool buffers")
---
 drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.c +1 -1
index bdfa2e293531..5ddedc3b754d 100644
@@ -488,10 +488,11 @@ dma_addr_t __otx2_alloc_rbuf(struct otx2_nic *pfvf, struct otx2_pool *pool)
 	dma_addr_t iova;
 	u8 *buf;
 
-	buf = napi_alloc_frag(pool->rbsize);
+	buf = napi_alloc_frag(pool->rbsize + OTX2_ALIGN);
 	if (unlikely(!buf))
 		return -ENOMEM;
 
Quoted 9 lines
Quoted 6 lines
drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.c +1 -1
+	buf = PTR_ALIGN(buf, OTX2_ALIGN);
 	iova = dma_map_single_attrs(pfvf->dev, buf, pool->rbsize,
 				    DMA_FROM_DEVICE, DMA_ATTR_SKIP_CPU_SYNC);
 	if (unlikely(dma_mapping_error(pfvf->dev, iova))) {
--

2.29.2

Doesn't that break the 'free' code ?
Surely it needs the original pointer.

Why do we care about the original pointer? The free code should work with the mangling poiner. Did I miss something?

Quoted 4 lines

It isn't obvious that page_frag_free() it correct when the allocator is napi_alloc_frag() either.
I'd have thought it ought to be returned to the pool.

Sorry, I didn't get what you mean. Could you elaborate a bit more?

Thanks,
Kevin

Quoted 7 lines

David

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK Registration No: 1397386 (Wales)

#4 lore raw
Quoted 42 lines · Kevin Hao
Quoted 30 lines · David Laight

From: Kevin Hao

Quoted 29 lines

Sent: 21 January 2021 07:09

The octeontx2 hardware needs the buffer to be 128 byte aligned. But in the current implementation of napi_alloc_frag(), it can't guarantee the return address is 128 byte aligned even the request size is a multiple of 128 bytes, so we have to request an extra 128 bytes and use the PTR_ALIGN() to make sure that the buffer is aligned correctly.

Fixes 7a36e4918e30 ("octeontx2-pf: Use the napi_alloc_frag() to alloc the pool buffers")
---
 drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.c +1 -1
index bdfa2e293531..5ddedc3b754d 100644
@@ -488,10 +488,11 @@ dma_addr_t __otx2_alloc_rbuf(struct otx2_nic *pfvf, struct otx2_pool *pool)
    dma_addr_t iova;
    u8 *buf;
 
-   buf = napi_alloc_frag(pool->rbsize);
+   buf = napi_alloc_frag(pool->rbsize + OTX2_ALIGN);
    if (unlikely(!buf))
            return -ENOMEM;
 
Quoted 9 lines
Quoted 6 lines
drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.c +1 -1
+   buf = PTR_ALIGN(buf, OTX2_ALIGN);
    iova = dma_map_single_attrs(pfvf->dev, buf, pool->rbsize,
                                DMA_FROM_DEVICE, DMA_ATTR_SKIP_CPU_SYNC);
    if (unlikely(dma_mapping_error(pfvf->dev, iova))) {
--

2.29.2

Doesn't that break the 'free' code ?
Surely it needs the original pointer.

Quoted 4 lines

Why do we care about the original pointer? The free code should work with the mangling poiner. Did I miss something?

Quoted 1 line

I too agree, put_page(buf) or put_page(buf + OTX2_ALIGN) is same.

#5 lore raw
In reply to Kevin Hao
This patch was applied to netdev/net.git (refs/heads/master):

Hello:

This patch was applied to netdev/net.git (refs/heads/master):

Quoted 11 lines · On Thu, 21 Jan 2021 15:09:06 +0800 you wrote:

The octeontx2 hardware needs the buffer to be 128 byte aligned. But in the current implementation of napi_alloc_frag(), it can't guarantee the return address is 128 byte aligned even the request size is a multiple of 128 bytes, so we have to request an extra 128 bytes and use the PTR_ALIGN() to make sure that the buffer is aligned correctly.

Fixes 7a36e4918e30 ("octeontx2-pf: Use the napi_alloc_frag() to alloc the pool buffers")

[...]

Here is the summary with links:
- net: octeontx2: Make sure the buffer is 128 byte aligned

    https://git.kernel.org/netdev/net/c/db2805150a0f

You are awesome, thank you!
--
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html