* Re: [PATCH net-next] net: stmmac: set total length of the packet to be transmitted in TDES3
2017-04-10 18:33 [PATCH net-next] net: stmmac: set total length of the packet to be transmitted in TDES3 Niklas Cassel
@ 2017-04-11 5:40 ` Giuseppe CAVALLARO
2017-04-11 7:20 ` Alexandre Torgue
2017-04-11 8:48 ` Joao Pinto
` (2 subsequent siblings)
3 siblings, 1 reply; 7+ messages in thread
From: Giuseppe CAVALLARO @ 2017-04-11 5:40 UTC (permalink / raw)
To: Niklas Cassel, alexandre.torgue, Joao.Pinto
Cc: netdev, linux-kernel, Niklas Cassel
Hi Niklas
patch looks ok for me, Alex any feedback?
peppe
On 4/10/2017 8:33 PM, Niklas Cassel wrote:
> From: Niklas Cassel <niklas.cassel@axis.com>
>
> Field FL/TPL in register TDES3 is not correctly set on GMAC4.
> TX appears to be functional on GMAC 4.10a even if this field is not set,
> however, to avoid relying on undefined behavior, set the length in TDES3.
>
> The field has a different meaning depending on if the TSE bit in TDES3
> is set or not (TSO). However, regardless of the TSE bit, the field is
> not optional. The field is already set correctly when the TSE bit is set.
>
> Since there is no limit for the number of descriptors that can be
> used for a single packet, the field should be set to the sum of
> the buffers contained in:
> [<desc with First Descriptor bit set> ... <desc n> ...
> <desc with Last Descriptor bit set>], which should be equal to skb->len.
>
> Signed-off-by: Niklas Cassel <niklas.cassel@axis.com>
> ---
> drivers/net/ethernet/stmicro/stmmac/chain_mode.c | 6 +++---
> drivers/net/ethernet/stmicro/stmmac/common.h | 2 +-
> drivers/net/ethernet/stmicro/stmmac/dwmac4_descs.c | 3 ++-
> drivers/net/ethernet/stmicro/stmmac/enh_desc.c | 2 +-
> drivers/net/ethernet/stmicro/stmmac/norm_desc.c | 2 +-
> drivers/net/ethernet/stmicro/stmmac/ring_mode.c | 9 ++++++---
> drivers/net/ethernet/stmicro/stmmac/stmmac_main.c | 5 +++--
> 7 files changed, 17 insertions(+), 12 deletions(-)
>
> diff --git a/drivers/net/ethernet/stmicro/stmmac/chain_mode.c b/drivers/net/ethernet/stmicro/stmmac/chain_mode.c
> index 37881f81319e..e93c40b4631e 100644
> --- a/drivers/net/ethernet/stmicro/stmmac/chain_mode.c
> +++ b/drivers/net/ethernet/stmicro/stmmac/chain_mode.c
> @@ -52,7 +52,7 @@ static int stmmac_jumbo_frm(void *p, struct sk_buff *skb, int csum)
> tx_q->tx_skbuff_dma[entry].len = bmax;
> /* do not close the descriptor and do not set own bit */
> priv->hw->desc->prepare_tx_desc(desc, 1, bmax, csum, STMMAC_CHAIN_MODE,
> - 0, false);
> + 0, false, skb->len);
>
> while (len != 0) {
> tx_q->tx_skbuff[entry] = NULL;
> @@ -70,7 +70,7 @@ static int stmmac_jumbo_frm(void *p, struct sk_buff *skb, int csum)
> tx_q->tx_skbuff_dma[entry].len = bmax;
> priv->hw->desc->prepare_tx_desc(desc, 0, bmax, csum,
> STMMAC_CHAIN_MODE, 1,
> - false);
> + false, skb->len);
> len -= bmax;
> i++;
> } else {
> @@ -85,7 +85,7 @@ static int stmmac_jumbo_frm(void *p, struct sk_buff *skb, int csum)
> /* last descriptor can be set now */
> priv->hw->desc->prepare_tx_desc(desc, 0, len, csum,
> STMMAC_CHAIN_MODE, 1,
> - true);
> + true, skb->len);
> len = 0;
> }
> }
> diff --git a/drivers/net/ethernet/stmicro/stmmac/common.h b/drivers/net/ethernet/stmicro/stmmac/common.h
> index 90d28bcad880..b7ce3fbb5375 100644
> --- a/drivers/net/ethernet/stmicro/stmmac/common.h
> +++ b/drivers/net/ethernet/stmicro/stmmac/common.h
> @@ -373,7 +373,7 @@ struct stmmac_desc_ops {
> /* Invoked by the xmit function to prepare the tx descriptor */
> void (*prepare_tx_desc) (struct dma_desc *p, int is_fs, int len,
> bool csum_flag, int mode, bool tx_own,
> - bool ls);
> + bool ls, unsigned int tot_pkt_len);
> void (*prepare_tso_tx_desc)(struct dma_desc *p, int is_fs, int len1,
> int len2, bool tx_own, bool ls,
> unsigned int tcphdrlen,
> diff --git a/drivers/net/ethernet/stmicro/stmmac/dwmac4_descs.c b/drivers/net/ethernet/stmicro/stmmac/dwmac4_descs.c
> index 843ec69222ea..aa6476439aee 100644
> --- a/drivers/net/ethernet/stmicro/stmmac/dwmac4_descs.c
> +++ b/drivers/net/ethernet/stmicro/stmmac/dwmac4_descs.c
> @@ -304,12 +304,13 @@ static void dwmac4_rd_init_tx_desc(struct dma_desc *p, int mode, int end)
>
> static void dwmac4_rd_prepare_tx_desc(struct dma_desc *p, int is_fs, int len,
> bool csum_flag, int mode, bool tx_own,
> - bool ls)
> + bool ls, unsigned int tot_pkt_len)
> {
> unsigned int tdes3 = le32_to_cpu(p->des3);
>
> p->des2 |= cpu_to_le32(len & TDES2_BUFFER1_SIZE_MASK);
>
> + tdes3 |= tot_pkt_len & TDES3_PACKET_SIZE_MASK;
> if (is_fs)
> tdes3 |= TDES3_FIRST_DESCRIPTOR;
> else
> diff --git a/drivers/net/ethernet/stmicro/stmmac/enh_desc.c b/drivers/net/ethernet/stmicro/stmmac/enh_desc.c
> index 323b59ec74a3..7546b3664113 100644
> --- a/drivers/net/ethernet/stmicro/stmmac/enh_desc.c
> +++ b/drivers/net/ethernet/stmicro/stmmac/enh_desc.c
> @@ -315,7 +315,7 @@ static void enh_desc_release_tx_desc(struct dma_desc *p, int mode)
>
> static void enh_desc_prepare_tx_desc(struct dma_desc *p, int is_fs, int len,
> bool csum_flag, int mode, bool tx_own,
> - bool ls)
> + bool ls, unsigned int tot_pkt_len)
> {
> unsigned int tdes0 = le32_to_cpu(p->des0);
>
> diff --git a/drivers/net/ethernet/stmicro/stmmac/norm_desc.c b/drivers/net/ethernet/stmicro/stmmac/norm_desc.c
> index efb818ebd55e..f817f8f36569 100644
> --- a/drivers/net/ethernet/stmicro/stmmac/norm_desc.c
> +++ b/drivers/net/ethernet/stmicro/stmmac/norm_desc.c
> @@ -191,7 +191,7 @@ static void ndesc_release_tx_desc(struct dma_desc *p, int mode)
>
> static void ndesc_prepare_tx_desc(struct dma_desc *p, int is_fs, int len,
> bool csum_flag, int mode, bool tx_own,
> - bool ls)
> + bool ls, unsigned int tot_pkt_len)
> {
> unsigned int tdes1 = le32_to_cpu(p->des1);
>
> diff --git a/drivers/net/ethernet/stmicro/stmmac/ring_mode.c b/drivers/net/ethernet/stmicro/stmmac/ring_mode.c
> index 31213e64513d..28e4b5d50ce6 100644
> --- a/drivers/net/ethernet/stmicro/stmmac/ring_mode.c
> +++ b/drivers/net/ethernet/stmicro/stmmac/ring_mode.c
> @@ -59,7 +59,8 @@ static int stmmac_jumbo_frm(void *p, struct sk_buff *skb, int csum)
>
> desc->des3 = cpu_to_le32(des2 + BUF_SIZE_4KiB);
> priv->hw->desc->prepare_tx_desc(desc, 1, bmax, csum,
> - STMMAC_RING_MODE, 0, false);
> + STMMAC_RING_MODE, 0,
> + false, skb->len);
> tx_q->tx_skbuff[entry] = NULL;
> entry = STMMAC_GET_ENTRY(entry, DMA_TX_SIZE);
>
> @@ -79,7 +80,8 @@ static int stmmac_jumbo_frm(void *p, struct sk_buff *skb, int csum)
>
> desc->des3 = cpu_to_le32(des2 + BUF_SIZE_4KiB);
> priv->hw->desc->prepare_tx_desc(desc, 0, len, csum,
> - STMMAC_RING_MODE, 1, true);
> + STMMAC_RING_MODE, 1,
> + true, skb->len);
> } else {
> des2 = dma_map_single(priv->device, skb->data,
> nopaged_len, DMA_TO_DEVICE);
> @@ -91,7 +93,8 @@ static int stmmac_jumbo_frm(void *p, struct sk_buff *skb, int csum)
> tx_q->tx_skbuff_dma[entry].is_jumbo = true;
> desc->des3 = cpu_to_le32(des2 + BUF_SIZE_4KiB);
> priv->hw->desc->prepare_tx_desc(desc, 1, nopaged_len, csum,
> - STMMAC_RING_MODE, 0, true);
> + STMMAC_RING_MODE, 0,
> + true, skb->len);
> }
>
> tx_q->cur_tx = entry;
> diff --git a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
> index a89f76b27ea2..5c1ebc70bebd 100644
> --- a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
> +++ b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
> @@ -3033,7 +3033,8 @@ static netdev_tx_t stmmac_xmit(struct sk_buff *skb, struct net_device *dev)
>
> /* Prepare the descriptor and set the own bit too */
> priv->hw->desc->prepare_tx_desc(desc, 0, len, csum_insertion,
> - priv->mode, 1, last_segment);
> + priv->mode, 1, last_segment,
> + skb->len);
> }
>
> entry = STMMAC_GET_ENTRY(entry, DMA_TX_SIZE);
> @@ -3116,7 +3117,7 @@ static netdev_tx_t stmmac_xmit(struct sk_buff *skb, struct net_device *dev)
> /* Prepare the first descriptor setting the OWN bit too */
> priv->hw->desc->prepare_tx_desc(first, 1, nopaged_len,
> csum_insertion, priv->mode, 1,
> - last_segment);
> + last_segment, skb->len);
>
> /* The own bit must be the latest setting done when prepare the
> * descriptor and then barrier is needed to make sure that
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [PATCH net-next] net: stmmac: set total length of the packet to be transmitted in TDES3
2017-04-11 5:40 ` Giuseppe CAVALLARO
@ 2017-04-11 7:20 ` Alexandre Torgue
0 siblings, 0 replies; 7+ messages in thread
From: Alexandre Torgue @ 2017-04-11 7:20 UTC (permalink / raw)
To: Giuseppe CAVALLARO, Niklas Cassel, Joao.Pinto
Cc: netdev, linux-kernel, Niklas Cassel
Hi,
On 04/11/2017 07:40 AM, Giuseppe CAVALLARO wrote:
> Hi Niklas
>
> patch looks ok for me, Alex any feedback?
It sounds good for me to.
>
> peppe
>
> On 4/10/2017 8:33 PM, Niklas Cassel wrote:
>> From: Niklas Cassel <niklas.cassel@axis.com>
>>
>> Field FL/TPL in register TDES3 is not correctly set on GMAC4.
>> TX appears to be functional on GMAC 4.10a even if this field is not set,
>> however, to avoid relying on undefined behavior, set the length in TDES3.
>>
>> The field has a different meaning depending on if the TSE bit in TDES3
>> is set or not (TSO). However, regardless of the TSE bit, the field is
>> not optional. The field is already set correctly when the TSE bit is set.
>>
>> Since there is no limit for the number of descriptors that can be
>> used for a single packet, the field should be set to the sum of
>> the buffers contained in:
>> [<desc with First Descriptor bit set> ... <desc n> ...
>> <desc with Last Descriptor bit set>], which should be equal to skb->len.
>>
>> Signed-off-by: Niklas Cassel <niklas.cassel@axis.com>
>> ---
>> drivers/net/ethernet/stmicro/stmmac/chain_mode.c | 6 +++---
>> drivers/net/ethernet/stmicro/stmmac/common.h | 2 +-
>> drivers/net/ethernet/stmicro/stmmac/dwmac4_descs.c | 3 ++-
>> drivers/net/ethernet/stmicro/stmmac/enh_desc.c | 2 +-
>> drivers/net/ethernet/stmicro/stmmac/norm_desc.c | 2 +-
>> drivers/net/ethernet/stmicro/stmmac/ring_mode.c | 9 ++++++---
>> drivers/net/ethernet/stmicro/stmmac/stmmac_main.c | 5 +++--
>> 7 files changed, 17 insertions(+), 12 deletions(-)
>>
>> diff --git a/drivers/net/ethernet/stmicro/stmmac/chain_mode.c
>> b/drivers/net/ethernet/stmicro/stmmac/chain_mode.c
>> index 37881f81319e..e93c40b4631e 100644
>> --- a/drivers/net/ethernet/stmicro/stmmac/chain_mode.c
>> +++ b/drivers/net/ethernet/stmicro/stmmac/chain_mode.c
>> @@ -52,7 +52,7 @@ static int stmmac_jumbo_frm(void *p, struct sk_buff
>> *skb, int csum)
>> tx_q->tx_skbuff_dma[entry].len = bmax;
>> /* do not close the descriptor and do not set own bit */
>> priv->hw->desc->prepare_tx_desc(desc, 1, bmax, csum,
>> STMMAC_CHAIN_MODE,
>> - 0, false);
>> + 0, false, skb->len);
>> while (len != 0) {
>> tx_q->tx_skbuff[entry] = NULL;
>> @@ -70,7 +70,7 @@ static int stmmac_jumbo_frm(void *p, struct sk_buff
>> *skb, int csum)
>> tx_q->tx_skbuff_dma[entry].len = bmax;
>> priv->hw->desc->prepare_tx_desc(desc, 0, bmax, csum,
>> STMMAC_CHAIN_MODE, 1,
>> - false);
>> + false, skb->len);
>> len -= bmax;
>> i++;
>> } else {
>> @@ -85,7 +85,7 @@ static int stmmac_jumbo_frm(void *p, struct sk_buff
>> *skb, int csum)
>> /* last descriptor can be set now */
>> priv->hw->desc->prepare_tx_desc(desc, 0, len, csum,
>> STMMAC_CHAIN_MODE, 1,
>> - true);
>> + true, skb->len);
>> len = 0;
>> }
>> }
>> diff --git a/drivers/net/ethernet/stmicro/stmmac/common.h
>> b/drivers/net/ethernet/stmicro/stmmac/common.h
>> index 90d28bcad880..b7ce3fbb5375 100644
>> --- a/drivers/net/ethernet/stmicro/stmmac/common.h
>> +++ b/drivers/net/ethernet/stmicro/stmmac/common.h
>> @@ -373,7 +373,7 @@ struct stmmac_desc_ops {
>> /* Invoked by the xmit function to prepare the tx descriptor */
>> void (*prepare_tx_desc) (struct dma_desc *p, int is_fs, int len,
>> bool csum_flag, int mode, bool tx_own,
>> - bool ls);
>> + bool ls, unsigned int tot_pkt_len);
>> void (*prepare_tso_tx_desc)(struct dma_desc *p, int is_fs, int
>> len1,
>> int len2, bool tx_own, bool ls,
>> unsigned int tcphdrlen,
>> diff --git a/drivers/net/ethernet/stmicro/stmmac/dwmac4_descs.c
>> b/drivers/net/ethernet/stmicro/stmmac/dwmac4_descs.c
>> index 843ec69222ea..aa6476439aee 100644
>> --- a/drivers/net/ethernet/stmicro/stmmac/dwmac4_descs.c
>> +++ b/drivers/net/ethernet/stmicro/stmmac/dwmac4_descs.c
>> @@ -304,12 +304,13 @@ static void dwmac4_rd_init_tx_desc(struct
>> dma_desc *p, int mode, int end)
>> static void dwmac4_rd_prepare_tx_desc(struct dma_desc *p, int
>> is_fs, int len,
>> bool csum_flag, int mode, bool tx_own,
>> - bool ls)
>> + bool ls, unsigned int tot_pkt_len)
>> {
>> unsigned int tdes3 = le32_to_cpu(p->des3);
>> p->des2 |= cpu_to_le32(len & TDES2_BUFFER1_SIZE_MASK);
>> + tdes3 |= tot_pkt_len & TDES3_PACKET_SIZE_MASK;
>> if (is_fs)
>> tdes3 |= TDES3_FIRST_DESCRIPTOR;
>> else
>> diff --git a/drivers/net/ethernet/stmicro/stmmac/enh_desc.c
>> b/drivers/net/ethernet/stmicro/stmmac/enh_desc.c
>> index 323b59ec74a3..7546b3664113 100644
>> --- a/drivers/net/ethernet/stmicro/stmmac/enh_desc.c
>> +++ b/drivers/net/ethernet/stmicro/stmmac/enh_desc.c
>> @@ -315,7 +315,7 @@ static void enh_desc_release_tx_desc(struct
>> dma_desc *p, int mode)
>> static void enh_desc_prepare_tx_desc(struct dma_desc *p, int
>> is_fs, int len,
>> bool csum_flag, int mode, bool tx_own,
>> - bool ls)
>> + bool ls, unsigned int tot_pkt_len)
>> {
>> unsigned int tdes0 = le32_to_cpu(p->des0);
>> diff --git a/drivers/net/ethernet/stmicro/stmmac/norm_desc.c
>> b/drivers/net/ethernet/stmicro/stmmac/norm_desc.c
>> index efb818ebd55e..f817f8f36569 100644
>> --- a/drivers/net/ethernet/stmicro/stmmac/norm_desc.c
>> +++ b/drivers/net/ethernet/stmicro/stmmac/norm_desc.c
>> @@ -191,7 +191,7 @@ static void ndesc_release_tx_desc(struct dma_desc
>> *p, int mode)
>> static void ndesc_prepare_tx_desc(struct dma_desc *p, int is_fs,
>> int len,
>> bool csum_flag, int mode, bool tx_own,
>> - bool ls)
>> + bool ls, unsigned int tot_pkt_len)
>> {
>> unsigned int tdes1 = le32_to_cpu(p->des1);
>> diff --git a/drivers/net/ethernet/stmicro/stmmac/ring_mode.c
>> b/drivers/net/ethernet/stmicro/stmmac/ring_mode.c
>> index 31213e64513d..28e4b5d50ce6 100644
>> --- a/drivers/net/ethernet/stmicro/stmmac/ring_mode.c
>> +++ b/drivers/net/ethernet/stmicro/stmmac/ring_mode.c
>> @@ -59,7 +59,8 @@ static int stmmac_jumbo_frm(void *p, struct sk_buff
>> *skb, int csum)
>> desc->des3 = cpu_to_le32(des2 + BUF_SIZE_4KiB);
>> priv->hw->desc->prepare_tx_desc(desc, 1, bmax, csum,
>> - STMMAC_RING_MODE, 0, false);
>> + STMMAC_RING_MODE, 0,
>> + false, skb->len);
>> tx_q->tx_skbuff[entry] = NULL;
>> entry = STMMAC_GET_ENTRY(entry, DMA_TX_SIZE);
>> @@ -79,7 +80,8 @@ static int stmmac_jumbo_frm(void *p, struct
>> sk_buff *skb, int csum)
>> desc->des3 = cpu_to_le32(des2 + BUF_SIZE_4KiB);
>> priv->hw->desc->prepare_tx_desc(desc, 0, len, csum,
>> - STMMAC_RING_MODE, 1, true);
>> + STMMAC_RING_MODE, 1,
>> + true, skb->len);
>> } else {
>> des2 = dma_map_single(priv->device, skb->data,
>> nopaged_len, DMA_TO_DEVICE);
>> @@ -91,7 +93,8 @@ static int stmmac_jumbo_frm(void *p, struct sk_buff
>> *skb, int csum)
>> tx_q->tx_skbuff_dma[entry].is_jumbo = true;
>> desc->des3 = cpu_to_le32(des2 + BUF_SIZE_4KiB);
>> priv->hw->desc->prepare_tx_desc(desc, 1, nopaged_len, csum,
>> - STMMAC_RING_MODE, 0, true);
>> + STMMAC_RING_MODE, 0,
>> + true, skb->len);
>> }
>> tx_q->cur_tx = entry;
>> diff --git a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
>> b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
>> index a89f76b27ea2..5c1ebc70bebd 100644
>> --- a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
>> +++ b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
>> @@ -3033,7 +3033,8 @@ static netdev_tx_t stmmac_xmit(struct sk_buff
>> *skb, struct net_device *dev)
>> /* Prepare the descriptor and set the own bit too */
>> priv->hw->desc->prepare_tx_desc(desc, 0, len, csum_insertion,
>> - priv->mode, 1, last_segment);
>> + priv->mode, 1, last_segment,
>> + skb->len);
>> }
>> entry = STMMAC_GET_ENTRY(entry, DMA_TX_SIZE);
>> @@ -3116,7 +3117,7 @@ static netdev_tx_t stmmac_xmit(struct sk_buff
>> *skb, struct net_device *dev)
>> /* Prepare the first descriptor setting the OWN bit too */
>> priv->hw->desc->prepare_tx_desc(first, 1, nopaged_len,
>> csum_insertion, priv->mode, 1,
>> - last_segment);
>> + last_segment, skb->len);
>> /* The own bit must be the latest setting done when
>> prepare the
>> * descriptor and then barrier is needed to make sure that
>
>
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [PATCH net-next] net: stmmac: set total length of the packet to be transmitted in TDES3
2017-04-10 18:33 [PATCH net-next] net: stmmac: set total length of the packet to be transmitted in TDES3 Niklas Cassel
2017-04-11 5:40 ` Giuseppe CAVALLARO
@ 2017-04-11 8:48 ` Joao Pinto
2017-04-11 9:10 ` Niklas Cassel
2017-04-11 12:41 ` Giuseppe CAVALLARO
2017-04-13 16:40 ` David Miller
3 siblings, 1 reply; 7+ messages in thread
From: Joao Pinto @ 2017-04-11 8:48 UTC (permalink / raw)
To: Niklas Cassel, peppe.cavallaro, alexandre.torgue, Joao.Pinto
Cc: netdev, linux-kernel, Niklas Cassel
Hi Niklas,
Às 7:33 PM de 4/10/2017, Niklas Cassel escreveu:
> From: Niklas Cassel <niklas.cassel@axis.com>
>
> Field FL/TPL in register TDES3 is not correctly set on GMAC4.
> TX appears to be functional on GMAC 4.10a even if this field is not set,
> however, to avoid relying on undefined behavior, set the length in TDES3.
>
> The field has a different meaning depending on if the TSE bit in TDES3
> is set or not (TSO). However, regardless of the TSE bit, the field is
> not optional. The field is already set correctly when the TSE bit is set.
>
> Since there is no limit for the number of descriptors that can be
> used for a single packet, the field should be set to the sum of
> the buffers contained in:
> [<desc with First Descriptor bit set> ... <desc n> ...
> <desc with Last Descriptor bit set>], which should be equal to skb->len.
>
> Signed-off-by: Niklas Cassel <niklas.cassel@axis.com>
Sounds fine to me. Did you check for performance improvement? Thanks.
Joao
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [PATCH net-next] net: stmmac: set total length of the packet to be transmitted in TDES3
2017-04-11 8:48 ` Joao Pinto
@ 2017-04-11 9:10 ` Niklas Cassel
0 siblings, 0 replies; 7+ messages in thread
From: Niklas Cassel @ 2017-04-11 9:10 UTC (permalink / raw)
To: Joao Pinto, peppe.cavallaro, alexandre.torgue; +Cc: netdev, linux-kernel
On 04/11/2017 10:48 AM, Joao Pinto wrote:
>
> Hi Niklas,
>
> Às 7:33 PM de 4/10/2017, Niklas Cassel escreveu:
>> From: Niklas Cassel <niklas.cassel@axis.com>
>>
>> Field FL/TPL in register TDES3 is not correctly set on GMAC4.
>> TX appears to be functional on GMAC 4.10a even if this field is not set,
>> however, to avoid relying on undefined behavior, set the length in TDES3.
>>
>> The field has a different meaning depending on if the TSE bit in TDES3
>> is set or not (TSO). However, regardless of the TSE bit, the field is
>> not optional. The field is already set correctly when the TSE bit is set.
>>
>> Since there is no limit for the number of descriptors that can be
>> used for a single packet, the field should be set to the sum of
>> the buffers contained in:
>> [<desc with First Descriptor bit set> ... <desc n> ...
>> <desc with Last Descriptor bit set>], which should be equal to skb->len.
>>
>> Signed-off-by: Niklas Cassel <niklas.cassel@axis.com>
>
> Sounds fine to me. Did you check for performance improvement? Thanks.
>
Hello Joao,
I could not see any difference with 1 TCP stream in iperf3
Without the patch:
# ./iperf3-armhf -c 192.168.0.1
Connecting to host 192.168.0.1, port 5201
[ 4] local 192.168.0.105 port 51878 connected to 192.168.0.1 port 5201
[ 23.486426] random: fast init done
[ ID] Interval Transfer Bandwidth Retr Cwnd
[ 4] 0.00-1.00 sec 61.7 MBytes 517 Mbits/sec 0 294 KBytes
[ 4] 1.00-2.00 sec 61.3 MBytes 515 Mbits/sec 0 294 KBytes
[ 26.043850] random: crng init done
[ 4] 2.00-3.00 sec 61.4 MBytes 515 Mbits/sec 0 294 KBytes
[ 4] 3.00-4.00 sec 61.0 MBytes 512 Mbits/sec 0 294 KBytes
[ 4] 4.00-5.00 sec 61.5 MBytes 516 Mbits/sec 0 294 KBytes
[ 4] 5.00-6.00 sec 61.2 MBytes 514 Mbits/sec 0 294 KBytes
[ 4] 6.00-7.00 sec 61.5 MBytes 516 Mbits/sec 0 294 KBytes
[ 4] 7.00-8.00 sec 61.4 MBytes 515 Mbits/sec 0 294 KBytes
[ 4] 8.00-9.00 sec 61.2 MBytes 514 Mbits/sec 0 294 KBytes
[ 4] 9.00-10.00 sec 61.3 MBytes 514 Mbits/sec 0 294 KBytes
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID] Interval Transfer Bandwidth Retr
[ 4] 0.00-10.00 sec 614 MBytes 515 Mbits/sec 0 sender
[ 4] 0.00-10.00 sec 613 MBytes 514 Mbits/sec receiver
With the patch:
# ./iperf3-armhf -c 192.168.0.1
Connecting to host 192.168.0.1, port 5201
[ 4] local 192.168.0.105 port 37696 connected to 192.168.0.1 port 5201
[ 85.266262] random: fast init done
[ ID] Interval Transfer Bandwidth Retr Cwnd
[ 4] 0.00-1.00 sec 62.1 MBytes 520 Mbits/sec 0 288 KBytes
[ 4] 1.00-2.00 sec 61.0 MBytes 512 Mbits/sec 0 288 KBytes
[ 87.782547] random: crng init done
[ 4] 2.00-3.00 sec 61.3 MBytes 515 Mbits/sec 0 300 KBytes
[ 4] 3.00-4.00 sec 61.1 MBytes 512 Mbits/sec 0 300 KBytes
[ 4] 4.00-5.00 sec 61.4 MBytes 515 Mbits/sec 0 300 KBytes
[ 4] 5.00-6.00 sec 61.1 MBytes 513 Mbits/sec 0 300 KBytes
[ 4] 6.00-7.00 sec 61.1 MBytes 513 Mbits/sec 0 300 KBytes
[ 4] 7.00-8.00 sec 61.6 MBytes 517 Mbits/sec 0 300 KBytes
[ 4] 8.00-9.00 sec 61.3 MBytes 515 Mbits/sec 0 300 KBytes
[ 4] 9.00-10.00 sec 61.6 MBytes 517 Mbits/sec 0 300 KBytes
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID] Interval Transfer Bandwidth Retr
[ 4] 0.00-10.00 sec 614 MBytes 515 Mbits/sec 0 sender
[ 4] 0.00-10.00 sec 613 MBytes 514 Mbits/sec receiver
Note that the results above are with TSO off.
With TSO on, I get 921 Mbps.
However, the patch should not affect the case where TSO is on.
Regards,
Niklas
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [PATCH net-next] net: stmmac: set total length of the packet to be transmitted in TDES3
2017-04-10 18:33 [PATCH net-next] net: stmmac: set total length of the packet to be transmitted in TDES3 Niklas Cassel
2017-04-11 5:40 ` Giuseppe CAVALLARO
2017-04-11 8:48 ` Joao Pinto
@ 2017-04-11 12:41 ` Giuseppe CAVALLARO
2017-04-13 16:40 ` David Miller
3 siblings, 0 replies; 7+ messages in thread
From: Giuseppe CAVALLARO @ 2017-04-11 12:41 UTC (permalink / raw)
To: Niklas Cassel, alexandre.torgue, Joao.Pinto
Cc: netdev, linux-kernel, Niklas Cassel
On 4/10/2017 8:33 PM, Niklas Cassel wrote:
> From: Niklas Cassel <niklas.cassel@axis.com>
>
> Field FL/TPL in register TDES3 is not correctly set on GMAC4.
> TX appears to be functional on GMAC 4.10a even if this field is not set,
> however, to avoid relying on undefined behavior, set the length in TDES3.
>
> The field has a different meaning depending on if the TSE bit in TDES3
> is set or not (TSO). However, regardless of the TSE bit, the field is
> not optional. The field is already set correctly when the TSE bit is set.
>
> Since there is no limit for the number of descriptors that can be
> used for a single packet, the field should be set to the sum of
> the buffers contained in:
> [<desc with First Descriptor bit set> ... <desc n> ...
> <desc with Last Descriptor bit set>], which should be equal to skb->len.
>
> Signed-off-by: Niklas Cassel <niklas.cassel@axis.com>
Acked-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
> ---
> drivers/net/ethernet/stmicro/stmmac/chain_mode.c | 6 +++---
> drivers/net/ethernet/stmicro/stmmac/common.h | 2 +-
> drivers/net/ethernet/stmicro/stmmac/dwmac4_descs.c | 3 ++-
> drivers/net/ethernet/stmicro/stmmac/enh_desc.c | 2 +-
> drivers/net/ethernet/stmicro/stmmac/norm_desc.c | 2 +-
> drivers/net/ethernet/stmicro/stmmac/ring_mode.c | 9 ++++++---
> drivers/net/ethernet/stmicro/stmmac/stmmac_main.c | 5 +++--
> 7 files changed, 17 insertions(+), 12 deletions(-)
>
> diff --git a/drivers/net/ethernet/stmicro/stmmac/chain_mode.c b/drivers/net/ethernet/stmicro/stmmac/chain_mode.c
> index 37881f81319e..e93c40b4631e 100644
> --- a/drivers/net/ethernet/stmicro/stmmac/chain_mode.c
> +++ b/drivers/net/ethernet/stmicro/stmmac/chain_mode.c
> @@ -52,7 +52,7 @@ static int stmmac_jumbo_frm(void *p, struct sk_buff *skb, int csum)
> tx_q->tx_skbuff_dma[entry].len = bmax;
> /* do not close the descriptor and do not set own bit */
> priv->hw->desc->prepare_tx_desc(desc, 1, bmax, csum, STMMAC_CHAIN_MODE,
> - 0, false);
> + 0, false, skb->len);
>
> while (len != 0) {
> tx_q->tx_skbuff[entry] = NULL;
> @@ -70,7 +70,7 @@ static int stmmac_jumbo_frm(void *p, struct sk_buff *skb, int csum)
> tx_q->tx_skbuff_dma[entry].len = bmax;
> priv->hw->desc->prepare_tx_desc(desc, 0, bmax, csum,
> STMMAC_CHAIN_MODE, 1,
> - false);
> + false, skb->len);
> len -= bmax;
> i++;
> } else {
> @@ -85,7 +85,7 @@ static int stmmac_jumbo_frm(void *p, struct sk_buff *skb, int csum)
> /* last descriptor can be set now */
> priv->hw->desc->prepare_tx_desc(desc, 0, len, csum,
> STMMAC_CHAIN_MODE, 1,
> - true);
> + true, skb->len);
> len = 0;
> }
> }
> diff --git a/drivers/net/ethernet/stmicro/stmmac/common.h b/drivers/net/ethernet/stmicro/stmmac/common.h
> index 90d28bcad880..b7ce3fbb5375 100644
> --- a/drivers/net/ethernet/stmicro/stmmac/common.h
> +++ b/drivers/net/ethernet/stmicro/stmmac/common.h
> @@ -373,7 +373,7 @@ struct stmmac_desc_ops {
> /* Invoked by the xmit function to prepare the tx descriptor */
> void (*prepare_tx_desc) (struct dma_desc *p, int is_fs, int len,
> bool csum_flag, int mode, bool tx_own,
> - bool ls);
> + bool ls, unsigned int tot_pkt_len);
> void (*prepare_tso_tx_desc)(struct dma_desc *p, int is_fs, int len1,
> int len2, bool tx_own, bool ls,
> unsigned int tcphdrlen,
> diff --git a/drivers/net/ethernet/stmicro/stmmac/dwmac4_descs.c b/drivers/net/ethernet/stmicro/stmmac/dwmac4_descs.c
> index 843ec69222ea..aa6476439aee 100644
> --- a/drivers/net/ethernet/stmicro/stmmac/dwmac4_descs.c
> +++ b/drivers/net/ethernet/stmicro/stmmac/dwmac4_descs.c
> @@ -304,12 +304,13 @@ static void dwmac4_rd_init_tx_desc(struct dma_desc *p, int mode, int end)
>
> static void dwmac4_rd_prepare_tx_desc(struct dma_desc *p, int is_fs, int len,
> bool csum_flag, int mode, bool tx_own,
> - bool ls)
> + bool ls, unsigned int tot_pkt_len)
> {
> unsigned int tdes3 = le32_to_cpu(p->des3);
>
> p->des2 |= cpu_to_le32(len & TDES2_BUFFER1_SIZE_MASK);
>
> + tdes3 |= tot_pkt_len & TDES3_PACKET_SIZE_MASK;
> if (is_fs)
> tdes3 |= TDES3_FIRST_DESCRIPTOR;
> else
> diff --git a/drivers/net/ethernet/stmicro/stmmac/enh_desc.c b/drivers/net/ethernet/stmicro/stmmac/enh_desc.c
> index 323b59ec74a3..7546b3664113 100644
> --- a/drivers/net/ethernet/stmicro/stmmac/enh_desc.c
> +++ b/drivers/net/ethernet/stmicro/stmmac/enh_desc.c
> @@ -315,7 +315,7 @@ static void enh_desc_release_tx_desc(struct dma_desc *p, int mode)
>
> static void enh_desc_prepare_tx_desc(struct dma_desc *p, int is_fs, int len,
> bool csum_flag, int mode, bool tx_own,
> - bool ls)
> + bool ls, unsigned int tot_pkt_len)
> {
> unsigned int tdes0 = le32_to_cpu(p->des0);
>
> diff --git a/drivers/net/ethernet/stmicro/stmmac/norm_desc.c b/drivers/net/ethernet/stmicro/stmmac/norm_desc.c
> index efb818ebd55e..f817f8f36569 100644
> --- a/drivers/net/ethernet/stmicro/stmmac/norm_desc.c
> +++ b/drivers/net/ethernet/stmicro/stmmac/norm_desc.c
> @@ -191,7 +191,7 @@ static void ndesc_release_tx_desc(struct dma_desc *p, int mode)
>
> static void ndesc_prepare_tx_desc(struct dma_desc *p, int is_fs, int len,
> bool csum_flag, int mode, bool tx_own,
> - bool ls)
> + bool ls, unsigned int tot_pkt_len)
> {
> unsigned int tdes1 = le32_to_cpu(p->des1);
>
> diff --git a/drivers/net/ethernet/stmicro/stmmac/ring_mode.c b/drivers/net/ethernet/stmicro/stmmac/ring_mode.c
> index 31213e64513d..28e4b5d50ce6 100644
> --- a/drivers/net/ethernet/stmicro/stmmac/ring_mode.c
> +++ b/drivers/net/ethernet/stmicro/stmmac/ring_mode.c
> @@ -59,7 +59,8 @@ static int stmmac_jumbo_frm(void *p, struct sk_buff *skb, int csum)
>
> desc->des3 = cpu_to_le32(des2 + BUF_SIZE_4KiB);
> priv->hw->desc->prepare_tx_desc(desc, 1, bmax, csum,
> - STMMAC_RING_MODE, 0, false);
> + STMMAC_RING_MODE, 0,
> + false, skb->len);
> tx_q->tx_skbuff[entry] = NULL;
> entry = STMMAC_GET_ENTRY(entry, DMA_TX_SIZE);
>
> @@ -79,7 +80,8 @@ static int stmmac_jumbo_frm(void *p, struct sk_buff *skb, int csum)
>
> desc->des3 = cpu_to_le32(des2 + BUF_SIZE_4KiB);
> priv->hw->desc->prepare_tx_desc(desc, 0, len, csum,
> - STMMAC_RING_MODE, 1, true);
> + STMMAC_RING_MODE, 1,
> + true, skb->len);
> } else {
> des2 = dma_map_single(priv->device, skb->data,
> nopaged_len, DMA_TO_DEVICE);
> @@ -91,7 +93,8 @@ static int stmmac_jumbo_frm(void *p, struct sk_buff *skb, int csum)
> tx_q->tx_skbuff_dma[entry].is_jumbo = true;
> desc->des3 = cpu_to_le32(des2 + BUF_SIZE_4KiB);
> priv->hw->desc->prepare_tx_desc(desc, 1, nopaged_len, csum,
> - STMMAC_RING_MODE, 0, true);
> + STMMAC_RING_MODE, 0,
> + true, skb->len);
> }
>
> tx_q->cur_tx = entry;
> diff --git a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
> index a89f76b27ea2..5c1ebc70bebd 100644
> --- a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
> +++ b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
> @@ -3033,7 +3033,8 @@ static netdev_tx_t stmmac_xmit(struct sk_buff *skb, struct net_device *dev)
>
> /* Prepare the descriptor and set the own bit too */
> priv->hw->desc->prepare_tx_desc(desc, 0, len, csum_insertion,
> - priv->mode, 1, last_segment);
> + priv->mode, 1, last_segment,
> + skb->len);
> }
>
> entry = STMMAC_GET_ENTRY(entry, DMA_TX_SIZE);
> @@ -3116,7 +3117,7 @@ static netdev_tx_t stmmac_xmit(struct sk_buff *skb, struct net_device *dev)
> /* Prepare the first descriptor setting the OWN bit too */
> priv->hw->desc->prepare_tx_desc(first, 1, nopaged_len,
> csum_insertion, priv->mode, 1,
> - last_segment);
> + last_segment, skb->len);
>
> /* The own bit must be the latest setting done when prepare the
> * descriptor and then barrier is needed to make sure that
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [PATCH net-next] net: stmmac: set total length of the packet to be transmitted in TDES3
2017-04-10 18:33 [PATCH net-next] net: stmmac: set total length of the packet to be transmitted in TDES3 Niklas Cassel
` (2 preceding siblings ...)
2017-04-11 12:41 ` Giuseppe CAVALLARO
@ 2017-04-13 16:40 ` David Miller
3 siblings, 0 replies; 7+ messages in thread
From: David Miller @ 2017-04-13 16:40 UTC (permalink / raw)
To: niklas.cassel
Cc: peppe.cavallaro, alexandre.torgue, Joao.Pinto, netdev,
linux-kernel, niklass
From: Niklas Cassel <niklas.cassel@axis.com>
Date: Mon, 10 Apr 2017 20:33:29 +0200
> From: Niklas Cassel <niklas.cassel@axis.com>
>
> Field FL/TPL in register TDES3 is not correctly set on GMAC4.
> TX appears to be functional on GMAC 4.10a even if this field is not set,
> however, to avoid relying on undefined behavior, set the length in TDES3.
>
> The field has a different meaning depending on if the TSE bit in TDES3
> is set or not (TSO). However, regardless of the TSE bit, the field is
> not optional. The field is already set correctly when the TSE bit is set.
>
> Since there is no limit for the number of descriptors that can be
> used for a single packet, the field should be set to the sum of
> the buffers contained in:
> [<desc with First Descriptor bit set> ... <desc n> ...
> <desc with Last Descriptor bit set>], which should be equal to skb->len.
>
> Signed-off-by: Niklas Cassel <niklas.cassel@axis.com>
Applied, thanks.
^ permalink raw reply [flat|nested] 7+ messages in thread