diff mbox series

[net-next,v2,1/1] net: stmmac: Add dma_wmb() barrier before setting OWN bit in set_rx_owner()

Message ID 20240821060307.46350-1-ende.tan@starfivetech.com (mailing list archive)
State Changes Requested
Delegated to: Netdev Maintainers
Headers show
Series [net-next,v2,1/1] net: stmmac: Add dma_wmb() barrier before setting OWN bit in set_rx_owner() | expand

Checks

Context Check Description
netdev/series_format success Single patches do not need cover letters
netdev/tree_selection success Clearly marked for net-next
netdev/ynl success Generated files up to date; no warnings/errors; no diff in generated;
netdev/fixes_present success Fixes tag not required for -next series
netdev/header_inline success No static functions without inline keyword in header files
netdev/build_32bit success Errors and warnings before: 16 this patch: 16
netdev/build_tools success No tools touched, skip
netdev/cc_maintainers success CCed 9 of 9 maintainers
netdev/build_clang success Errors and warnings before: 16 this patch: 16
netdev/verify_signedoff success Signed-off-by tag matches author and committer
netdev/deprecated_api success None detected
netdev/check_selftest success No net selftest shell script
netdev/verify_fixes success No Fixes tag
netdev/build_allmodconfig_warn success Errors and warnings before: 18 this patch: 18
netdev/checkpatch success total: 0 errors, 0 warnings, 0 checks, 55 lines checked
netdev/build_clang_rust success No Rust files in patch. Skipping build
netdev/kdoc success Errors and warnings before: 38 this patch: 38
netdev/source_inline success Was 0 now: 0
netdev/contest success net-next-2024-08-21--09-00 (tests: 711)

Commit Message

EnDe Tan Aug. 21, 2024, 6:03 a.m. UTC
From: Tan En De <ende.tan@starfivetech.com>

Currently, some set_rx_owner() callbacks set interrupt-on-completion bit
in addition to OWN bit, without inserting a dma_wmb() barrier in
between. This might cause missed interrupt if the DMA sees the OWN bit
before the interrupt-on-completion bit is set.

Thus, this patch adds dma_wmb() barrier right before setting OWN bit in
each of the callbacks. Now that the responsibility of calling dma_wmb()
is delegated to the callbacks, let's simplify main driver code by
removing dma_wmb() before stmmac_set_rx_owner().

Signed-off-by: Tan En De <ende.tan@starfivetech.com>
---
v2:
- Avoid introducing a new function just to set the interrupt-on-completion
  bit, as it is wasteful to do so.
- Delegate the responsibility of calling dma_wmb() from main driver code
  to set_rx_owner() callbacks (i.e. let callbacks to manage the low-level
  ordering/barrier rather than cluttering up the main driver code).
v1:
- https://patchwork.kernel.org/project/netdevbpf/patch/20240814092438.3129-1-ende.tan@starfivetech.com/
---
 drivers/net/ethernet/stmicro/stmmac/dwmac4_descs.c   | 5 ++++-
 drivers/net/ethernet/stmicro/stmmac/dwxgmac2_descs.c | 5 +++--
 drivers/net/ethernet/stmicro/stmmac/enh_desc.c       | 1 +
 drivers/net/ethernet/stmicro/stmmac/norm_desc.c      | 1 +
 drivers/net/ethernet/stmicro/stmmac/stmmac_main.c    | 2 --
 5 files changed, 9 insertions(+), 5 deletions(-)

Comments

Eric Dumazet Aug. 21, 2024, 7:57 a.m. UTC | #1
On Wed, Aug 21, 2024 at 8:03 AM <ende.tan@starfivetech.com> wrote:
>
> From: Tan En De <ende.tan@starfivetech.com>
>
> Currently, some set_rx_owner() callbacks set interrupt-on-completion bit
> in addition to OWN bit, without inserting a dma_wmb() barrier in
> between. This might cause missed interrupt if the DMA sees the OWN bit
> before the interrupt-on-completion bit is set.
>
> Thus, this patch adds dma_wmb() barrier right before setting OWN bit in
> each of the callbacks. Now that the responsibility of calling dma_wmb()
> is delegated to the callbacks, let's simplify main driver code by
> removing dma_wmb() before stmmac_set_rx_owner().
>
> Signed-off-by: Tan En De <ende.tan@starfivetech.com>
> ---
> v2:
> - Avoid introducing a new function just to set the interrupt-on-completion
>   bit, as it is wasteful to do so.
> - Delegate the responsibility of calling dma_wmb() from main driver code
>   to set_rx_owner() callbacks (i.e. let callbacks to manage the low-level
>   ordering/barrier rather than cluttering up the main driver code).
> v1:
> - https://patchwork.kernel.org/project/netdevbpf/patch/20240814092438.3129-1-ende.tan@starfivetech.com/
> ---
>  drivers/net/ethernet/stmicro/stmmac/dwmac4_descs.c   | 5 ++++-
>  drivers/net/ethernet/stmicro/stmmac/dwxgmac2_descs.c | 5 +++--
>  drivers/net/ethernet/stmicro/stmmac/enh_desc.c       | 1 +
>  drivers/net/ethernet/stmicro/stmmac/norm_desc.c      | 1 +
>  drivers/net/ethernet/stmicro/stmmac/stmmac_main.c    | 2 --
>  5 files changed, 9 insertions(+), 5 deletions(-)
>
> diff --git a/drivers/net/ethernet/stmicro/stmmac/dwmac4_descs.c b/drivers/net/ethernet/stmicro/stmmac/dwmac4_descs.c
> index 1c5802e0d7f4..95aea6ad485b 100644
> --- a/drivers/net/ethernet/stmicro/stmmac/dwmac4_descs.c
> +++ b/drivers/net/ethernet/stmicro/stmmac/dwmac4_descs.c
> @@ -186,10 +186,13 @@ static void dwmac4_set_tx_owner(struct dma_desc *p)
>
>  static void dwmac4_set_rx_owner(struct dma_desc *p, int disable_rx_ic)
>  {
> -       p->des3 |= cpu_to_le32(RDES3_OWN | RDES3_BUFFER1_VALID_ADDR);
> +       p->des3 |= cpu_to_le32(RDES3_BUFFER1_VALID_ADDR);
>
>         if (!disable_rx_ic)
>                 p->des3 |= cpu_to_le32(RDES3_INT_ON_COMPLETION_EN);

RDES3_INT_ON_COMPLETION_EN was possibly written after RDES3_OWN

This looks like a bug fix to me, this should target the net tree.

> +
> +       dma_wmb();
> +       p->des3 |= cpu_to_le32(RDES3_OWN);
>  }
>
>  static int dwmac4_get_tx_ls(struct dma_desc *p)
> diff --git a/drivers/net/ethernet/stmicro/stmmac/dwxgmac2_descs.c b/drivers/net/ethernet/stmicro/stmmac/dwxgmac2_descs.c
> index fc82862a612c..d76ae833c840 100644
> --- a/drivers/net/ethernet/stmicro/stmmac/dwxgmac2_descs.c
> +++ b/drivers/net/ethernet/stmicro/stmmac/dwxgmac2_descs.c
> @@ -56,10 +56,11 @@ static void dwxgmac2_set_tx_owner(struct dma_desc *p)
>
>  static void dwxgmac2_set_rx_owner(struct dma_desc *p, int disable_rx_ic)
>  {
> -       p->des3 |= cpu_to_le32(XGMAC_RDES3_OWN);
> -
>         if (!disable_rx_ic)
>                 p->des3 |= cpu_to_le32(XGMAC_RDES3_IOC);

Same kind of bug here.

> +
> +       dma_wmb();
> +       p->des3 |= cpu_to_le32(XGMAC_RDES3_OWN);
>  }
Serge Semin Aug. 23, 2024, 11:39 a.m. UTC | #2
Hi Tan

On Wed, Aug 21, 2024 at 02:03:07PM +0800, ende.tan@starfivetech.com wrote:
> From: Tan En De <ende.tan@starfivetech.com>
> 
> Currently, some set_rx_owner() callbacks set interrupt-on-completion bit
> in addition to OWN bit, without inserting a dma_wmb() barrier in
> between. This might cause missed interrupt if the DMA sees the OWN bit
> before the interrupt-on-completion bit is set.
> 
> Thus, this patch adds dma_wmb() barrier right before setting OWN bit in
> each of the callbacks. Now that the responsibility of calling dma_wmb()
> is delegated to the callbacks, let's simplify main driver code by
> removing dma_wmb() before stmmac_set_rx_owner().
> 
> Signed-off-by: Tan En De <ende.tan@starfivetech.com>
> ---
> v2:
> - Avoid introducing a new function just to set the interrupt-on-completion
>   bit, as it is wasteful to do so.
> - Delegate the responsibility of calling dma_wmb() from main driver code
>   to set_rx_owner() callbacks (i.e. let callbacks to manage the low-level
>   ordering/barrier rather than cluttering up the main driver code).
> v1:
> - https://patchwork.kernel.org/project/netdevbpf/patch/20240814092438.3129-1-ende.tan@starfivetech.com/
> ---
>  drivers/net/ethernet/stmicro/stmmac/dwmac4_descs.c   | 5 ++++-
>  drivers/net/ethernet/stmicro/stmmac/dwxgmac2_descs.c | 5 +++--
>  drivers/net/ethernet/stmicro/stmmac/enh_desc.c       | 1 +
>  drivers/net/ethernet/stmicro/stmmac/norm_desc.c      | 1 +
>  drivers/net/ethernet/stmicro/stmmac/stmmac_main.c    | 2 --
>  5 files changed, 9 insertions(+), 5 deletions(-)
> 
> diff --git a/drivers/net/ethernet/stmicro/stmmac/dwmac4_descs.c b/drivers/net/ethernet/stmicro/stmmac/dwmac4_descs.c
> index 1c5802e0d7f4..95aea6ad485b 100644
> --- a/drivers/net/ethernet/stmicro/stmmac/dwmac4_descs.c
> +++ b/drivers/net/ethernet/stmicro/stmmac/dwmac4_descs.c
> @@ -186,10 +186,13 @@ static void dwmac4_set_tx_owner(struct dma_desc *p)
>  
>  static void dwmac4_set_rx_owner(struct dma_desc *p, int disable_rx_ic)
>  {

> -	p->des3 |= cpu_to_le32(RDES3_OWN | RDES3_BUFFER1_VALID_ADDR);
> +	p->des3 |= cpu_to_le32(RDES3_BUFFER1_VALID_ADDR);
>  
>  	if (!disable_rx_ic)
>  		p->des3 |= cpu_to_le32(RDES3_INT_ON_COMPLETION_EN);
> +
> +	dma_wmb();
> +	p->des3 |= cpu_to_le32(RDES3_OWN);
>  }
>  
>  static int dwmac4_get_tx_ls(struct dma_desc *p)
> diff --git a/drivers/net/ethernet/stmicro/stmmac/dwxgmac2_descs.c b/drivers/net/ethernet/stmicro/stmmac/dwxgmac2_descs.c
> index fc82862a612c..d76ae833c840 100644
> --- a/drivers/net/ethernet/stmicro/stmmac/dwxgmac2_descs.c
> +++ b/drivers/net/ethernet/stmicro/stmmac/dwxgmac2_descs.c
> @@ -56,10 +56,11 @@ static void dwxgmac2_set_tx_owner(struct dma_desc *p)
>  
>  static void dwxgmac2_set_rx_owner(struct dma_desc *p, int disable_rx_ic)
>  {
> -	p->des3 |= cpu_to_le32(XGMAC_RDES3_OWN);
> -
>  	if (!disable_rx_ic)
>  		p->des3 |= cpu_to_le32(XGMAC_RDES3_IOC);
> +
> +	dma_wmb();
> +	p->des3 |= cpu_to_le32(XGMAC_RDES3_OWN);

I am not against moving the barrier here but really I don't see a firm
reason of why you can't collect the flags in a local variable and
then flush it out to the DES3 field.

Getting back to your discussion with Andrew:
https://lore.kernel.org/netdev/06297829-0bf7-4a06-baaf-e32c39888947@lunn.ch/
you said:

> I didn't use local variable because I worry about CPU out-of-order execution. 
> For example,
> ```
> local_var = (INT_ON_COMPLETION | OWN)
> des3 |= local_var
> ```
> CPU optimization might result in this
> ```
> des3 |= INT_ON_COMPLETION
> des3 |= OWN
> ```
> or worst, out of order like this
> ```
> des3 |= OWN
> des3 |= INT_ON_COMPLETION
> ```

Why do you think the CPU would split up the pre-initialized local
variable write into the two-staged write?

Anyway Andrew is right about the descriptors memory nature. It's a
coherent memory to which the access is expensive and should be
minimized as much as possible.

-Serge(y)

>  }
>  
>  static int dwxgmac2_get_tx_ls(struct dma_desc *p)
> diff --git a/drivers/net/ethernet/stmicro/stmmac/enh_desc.c b/drivers/net/ethernet/stmicro/stmmac/enh_desc.c
> index 937b7a0466fc..9219fe69ea44 100644
> --- a/drivers/net/ethernet/stmicro/stmmac/enh_desc.c
> +++ b/drivers/net/ethernet/stmicro/stmmac/enh_desc.c
> @@ -289,6 +289,7 @@ static void enh_desc_set_tx_owner(struct dma_desc *p)
>  
>  static void enh_desc_set_rx_owner(struct dma_desc *p, int disable_rx_ic)
>  {
> +	dma_wmb();
>  	p->des0 |= cpu_to_le32(RDES0_OWN);
>  }
>  
> diff --git a/drivers/net/ethernet/stmicro/stmmac/norm_desc.c b/drivers/net/ethernet/stmicro/stmmac/norm_desc.c
> index 68a7cfcb1d8f..d0b703a3346f 100644
> --- a/drivers/net/ethernet/stmicro/stmmac/norm_desc.c
> +++ b/drivers/net/ethernet/stmicro/stmmac/norm_desc.c
> @@ -155,6 +155,7 @@ static void ndesc_set_tx_owner(struct dma_desc *p)
>  
>  static void ndesc_set_rx_owner(struct dma_desc *p, int disable_rx_ic)
>  {
> +	dma_wmb();
>  	p->des0 |= cpu_to_le32(RDES0_OWN);
>  }
>  
> diff --git a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
> index d9fca8d1227c..859a2c4c9e5c 100644
> --- a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
> +++ b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
> @@ -4848,7 +4848,6 @@ static inline void stmmac_rx_refill(struct stmmac_priv *priv, u32 queue)
>  		if (!priv->use_riwt)
>  			use_rx_wd = false;
>  
> -		dma_wmb();
>  		stmmac_set_rx_owner(priv, p, use_rx_wd);
>  
>  		entry = STMMAC_GET_ENTRY(entry, priv->dma_conf.dma_rx_size);
> @@ -5205,7 +5204,6 @@ static bool stmmac_rx_refill_zc(struct stmmac_priv *priv, u32 queue, u32 budget)
>  		if (!priv->use_riwt)
>  			use_rx_wd = false;
>  
> -		dma_wmb();
>  		stmmac_set_rx_owner(priv, rx_desc, use_rx_wd);
>  
>  		entry = STMMAC_GET_ENTRY(entry, priv->dma_conf.dma_rx_size);
> -- 
> 2.34.1
> 
>
diff mbox series

Patch

diff --git a/drivers/net/ethernet/stmicro/stmmac/dwmac4_descs.c b/drivers/net/ethernet/stmicro/stmmac/dwmac4_descs.c
index 1c5802e0d7f4..95aea6ad485b 100644
--- a/drivers/net/ethernet/stmicro/stmmac/dwmac4_descs.c
+++ b/drivers/net/ethernet/stmicro/stmmac/dwmac4_descs.c
@@ -186,10 +186,13 @@  static void dwmac4_set_tx_owner(struct dma_desc *p)
 
 static void dwmac4_set_rx_owner(struct dma_desc *p, int disable_rx_ic)
 {
-	p->des3 |= cpu_to_le32(RDES3_OWN | RDES3_BUFFER1_VALID_ADDR);
+	p->des3 |= cpu_to_le32(RDES3_BUFFER1_VALID_ADDR);
 
 	if (!disable_rx_ic)
 		p->des3 |= cpu_to_le32(RDES3_INT_ON_COMPLETION_EN);
+
+	dma_wmb();
+	p->des3 |= cpu_to_le32(RDES3_OWN);
 }
 
 static int dwmac4_get_tx_ls(struct dma_desc *p)
diff --git a/drivers/net/ethernet/stmicro/stmmac/dwxgmac2_descs.c b/drivers/net/ethernet/stmicro/stmmac/dwxgmac2_descs.c
index fc82862a612c..d76ae833c840 100644
--- a/drivers/net/ethernet/stmicro/stmmac/dwxgmac2_descs.c
+++ b/drivers/net/ethernet/stmicro/stmmac/dwxgmac2_descs.c
@@ -56,10 +56,11 @@  static void dwxgmac2_set_tx_owner(struct dma_desc *p)
 
 static void dwxgmac2_set_rx_owner(struct dma_desc *p, int disable_rx_ic)
 {
-	p->des3 |= cpu_to_le32(XGMAC_RDES3_OWN);
-
 	if (!disable_rx_ic)
 		p->des3 |= cpu_to_le32(XGMAC_RDES3_IOC);
+
+	dma_wmb();
+	p->des3 |= cpu_to_le32(XGMAC_RDES3_OWN);
 }
 
 static int dwxgmac2_get_tx_ls(struct dma_desc *p)
diff --git a/drivers/net/ethernet/stmicro/stmmac/enh_desc.c b/drivers/net/ethernet/stmicro/stmmac/enh_desc.c
index 937b7a0466fc..9219fe69ea44 100644
--- a/drivers/net/ethernet/stmicro/stmmac/enh_desc.c
+++ b/drivers/net/ethernet/stmicro/stmmac/enh_desc.c
@@ -289,6 +289,7 @@  static void enh_desc_set_tx_owner(struct dma_desc *p)
 
 static void enh_desc_set_rx_owner(struct dma_desc *p, int disable_rx_ic)
 {
+	dma_wmb();
 	p->des0 |= cpu_to_le32(RDES0_OWN);
 }
 
diff --git a/drivers/net/ethernet/stmicro/stmmac/norm_desc.c b/drivers/net/ethernet/stmicro/stmmac/norm_desc.c
index 68a7cfcb1d8f..d0b703a3346f 100644
--- a/drivers/net/ethernet/stmicro/stmmac/norm_desc.c
+++ b/drivers/net/ethernet/stmicro/stmmac/norm_desc.c
@@ -155,6 +155,7 @@  static void ndesc_set_tx_owner(struct dma_desc *p)
 
 static void ndesc_set_rx_owner(struct dma_desc *p, int disable_rx_ic)
 {
+	dma_wmb();
 	p->des0 |= cpu_to_le32(RDES0_OWN);
 }
 
diff --git a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
index d9fca8d1227c..859a2c4c9e5c 100644
--- a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
+++ b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
@@ -4848,7 +4848,6 @@  static inline void stmmac_rx_refill(struct stmmac_priv *priv, u32 queue)
 		if (!priv->use_riwt)
 			use_rx_wd = false;
 
-		dma_wmb();
 		stmmac_set_rx_owner(priv, p, use_rx_wd);
 
 		entry = STMMAC_GET_ENTRY(entry, priv->dma_conf.dma_rx_size);
@@ -5205,7 +5204,6 @@  static bool stmmac_rx_refill_zc(struct stmmac_priv *priv, u32 queue, u32 budget)
 		if (!priv->use_riwt)
 			use_rx_wd = false;
 
-		dma_wmb();
 		stmmac_set_rx_owner(priv, rx_desc, use_rx_wd);
 
 		entry = STMMAC_GET_ENTRY(entry, priv->dma_conf.dma_rx_size);