Skip to contentSkip to conversation

net: macb: Fix rx buffer cache handling

PATCH Superseded Stefan Roese 3 participants 7 messages Aug 23, 2019 Network drivers

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 net: macb: Fix rx buffer cache handling

With commit c6d07bf440bc ("net/macb: increase RX buffer size for GEM") ethernet support does not work any more with d-cache enabled on the AT91SAM. The reason is, that MACB_RX_BUFFER_SIZE was changed from 4096 to 128 but this change was not refected in the rx_buffer flush and invalidate functions, as these also use this macro.

This patch now fixes this by calculating the rx buffer size correctly again in those functions. With this change, ethernet works again reliably on my AT91SAM board.

Cc Ramon Fried <[email protected]>
Cc Eugen Hristev <[email protected]>
Cc Anup Patel <[email protected]>
Cc Bin Meng <[email protected]>
Cc Joe Hershberger <[email protected]>
---
 drivers/net/macb.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)
drivers/net/macb.c +4 -2
index c99cf663a4..fca380d012 100644
@@ -296,13 +296,15 @@ static inline void macb_flush_ring_desc(struct macb_device *macb, bool rx)
 static inline void macb_flush_rx_buffer(struct macb_device *macb)
 {
 	flush_dcache_range(macb->rx_buffer_dma, macb->rx_buffer_dma +
-			   ALIGN(MACB_RX_BUFFER_SIZE, PKTALIGN));
+			   ALIGN(MACB_RX_BUFFER_SIZE * MACB_RX_RING_SIZE,
+				 PKTALIGN));
 }
 
 static inline void macb_invalidate_rx_buffer(struct macb_device *macb)
 {
 	invalidate_dcache_range(macb->rx_buffer_dma, macb->rx_buffer_dma +
-				ALIGN(MACB_RX_BUFFER_SIZE, PKTALIGN));
+				ALIGN(MACB_RX_BUFFER_SIZE * MACB_RX_RING_SIZE,
+				      PKTALIGN));
 }
 
 #if defined(CONFIG_CMD_NET)
Signature · 1 line
2.23.0
#1 lore raw

PATCH 1/1 net: macb: Fix rx buffer cache handling

Hi Stefan,

Quoted 26 lines · Stefan Roese

With commit c6d07bf440bc ("net/macb: increase RX buffer size for GEM") ethernet support does not work any more with d-cache enabled on the AT91SAM. The reason is, that MACB_RX_BUFFER_SIZE was changed from 4096 to 128 but this change was not refected in the rx_buffer flush and invalidate functions, as these also use this macro.

This patch now fixes this by calculating the rx buffer size correctly again in those functions. With this change, ethernet works again reliably on my AT91SAM board.

Cc Ramon Fried <[email protected]>
Cc Eugen Hristev <[email protected]>
Cc Anup Patel <[email protected]>
Cc Bin Meng <[email protected]>
Cc Joe Hershberger <[email protected]>
---
 drivers/net/macb.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)
drivers/net/macb.c +0 -0
index c99cf663a4..fca380d012 100644

drivers/net/macb.c static inline void macb_flush_ring_desc(struct macb_device *macb, bool rx) · line 300

Quoted 6 lines
drivers/net/macb.c +2 -1
@@ -296,13 +296,15 @@ static inline void macb_flush_ring_desc(struct macb_device *macb, bool rx)
 static inline void macb_flush_rx_buffer(struct macb_device *macb)
 {
        flush_dcache_range(macb->rx_buffer_dma, macb->rx_buffer_dma +
-                          ALIGN(MACB_RX_BUFFER_SIZE, PKTALIGN));
+                          ALIGN(MACB_RX_BUFFER_SIZE * MACB_RX_RING_SIZE,
+                                PKTALIGN));

This looks wrong to me.

Shouldn't it be:

         flush_dcache_range(macb->rx_buffer_dma, macb->rx_buffer_dma +
                          ALIGN(macb->rx_buffer_size, PKTALIGN));

drivers/net/macb.c static inline void macb_flush_ring_desc(struct macb_device *macb, bool rx)

Quoted 12 lines
drivers/net/macb.c +2 -2
 }
 
 static inline void macb_invalidate_rx_buffer(struct macb_device *macb)
 {
        invalidate_dcache_range(macb->rx_buffer_dma, macb->rx_buffer_dma +
-                               ALIGN(MACB_RX_BUFFER_SIZE, PKTALIGN));
+                               ALIGN(MACB_RX_BUFFER_SIZE * MACB_RX_RING_SIZE,
+                                     PKTALIGN));
 }
 
 #if defined(CONFIG_CMD_NET)
--

Regards,
Bin

#2 lore raw

PATCH 1/1 net: macb: Fix rx buffer cache handling

Quoted 39 lines · Bin Meng

Hi Stefan,

Quoted 26 lines · Stefan Roese

With commit c6d07bf440bc ("net/macb: increase RX buffer size for GEM") ethernet support does not work any more with d-cache enabled on the AT91SAM. The reason is, that MACB_RX_BUFFER_SIZE was changed from 4096 to 128 but this change was not refected in the rx_buffer flush and invalidate functions, as these also use this macro.

This patch now fixes this by calculating the rx buffer size correctly again in those functions. With this change, ethernet works again reliably on my AT91SAM board.

Cc Ramon Fried <[email protected]>
Cc Eugen Hristev <[email protected]>
Cc Anup Patel <[email protected]>
Cc Bin Meng <[email protected]>
Cc Joe Hershberger <[email protected]>
---
 drivers/net/macb.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)
drivers/net/macb.c +0 -0
index c99cf663a4..fca380d012 100644

drivers/net/macb.c static inline void macb_flush_ring_desc(struct macb_device *macb, bool rx) · line 300

Quoted 6 lines
drivers/net/macb.c +2 -1
@@ -296,13 +296,15 @@ static inline void macb_flush_ring_desc(struct macb_device *macb, bool rx)
 static inline void macb_flush_rx_buffer(struct macb_device *macb)
 {
        flush_dcache_range(macb->rx_buffer_dma, macb->rx_buffer_dma +
-                          ALIGN(MACB_RX_BUFFER_SIZE, PKTALIGN));
+                          ALIGN(MACB_RX_BUFFER_SIZE * MACB_RX_RING_SIZE,
+                                PKTALIGN));

This looks wrong to me.

Quoted 4 lines

Shouldn't it be:

         flush_dcache_range(macb->rx_buffer_dma, macb->rx_buffer_dma +
                          ALIGN(macb->rx_buffer_size, PKTALIGN));

Ah, for MACB, looks it should be "MACB_RX_BUFFER_SIZE * MACB_RX_RING_SIZE", but for GEM, I am not sure.

Should it be:

          flush_dcache_range(macb->rx_buffer_dma, macb->rx_buffer_dma +
                           ALIGN(macb->rx_buffer_size *

MACB_RX_RING_SIZE, PKTALIGN));

Quoted 11 lines
Quoted 11 lines
drivers/net/macb.c +2 -1
 }
 
 static inline void macb_invalidate_rx_buffer(struct macb_device *macb)
 {
        invalidate_dcache_range(macb->rx_buffer_dma, macb->rx_buffer_dma +
-                               ALIGN(MACB_RX_BUFFER_SIZE, PKTALIGN));
+                               ALIGN(MACB_RX_BUFFER_SIZE * MACB_RX_RING_SIZE,
+                                     PKTALIGN));
 }
 
Quoted 2 lines

drivers/net/macb.c static inline void macb_flush_ring_desc(struct macb_device *macb, bool rx)

Quoted 2 lines
drivers/net/macb.c +0 -1
 #if defined(CONFIG_CMD_NET)
--

Regards,
Bin

#3 lore raw

PATCH 1/1 net: macb: Fix rx buffer cache handling

Hi Bin,

Quoted 45 lines · Bin Meng
Quoted 39 lines · Bin Meng

Hi Stefan,

Quoted 26 lines · Stefan Roese

With commit c6d07bf440bc ("net/macb: increase RX buffer size for GEM") ethernet support does not work any more with d-cache enabled on the AT91SAM. The reason is, that MACB_RX_BUFFER_SIZE was changed from 4096 to 128 but this change was not refected in the rx_buffer flush and invalidate functions, as these also use this macro.

This patch now fixes this by calculating the rx buffer size correctly again in those functions. With this change, ethernet works again reliably on my AT91SAM board.

Cc Ramon Fried <[email protected]>
Cc Eugen Hristev <[email protected]>
Cc Anup Patel <[email protected]>
Cc Bin Meng <[email protected]>
Cc Joe Hershberger <[email protected]>
---
  drivers/net/macb.c | 6 ++++--
  1 file changed, 4 insertions(+), 2 deletions(-)
drivers/net/macb.c +0 -0
index c99cf663a4..fca380d012 100644

drivers/net/macb.c static inline void macb_flush_ring_desc(struct macb_device *macb, bool rx) · line 300

Quoted 6 lines
drivers/net/macb.c +2 -1
@@ -296,13 +296,15 @@ static inline void macb_flush_ring_desc(struct macb_device *macb, bool rx)
  static inline void macb_flush_rx_buffer(struct macb_device *macb)
  {
         flush_dcache_range(macb->rx_buffer_dma, macb->rx_buffer_dma +
-                          ALIGN(MACB_RX_BUFFER_SIZE, PKTALIGN));
+                          ALIGN(MACB_RX_BUFFER_SIZE * MACB_RX_RING_SIZE,
+                                PKTALIGN));

This looks wrong to me.

Quoted 4 lines

Shouldn't it be:

          flush_dcache_range(macb->rx_buffer_dma, macb->rx_buffer_dma +
                           ALIGN(macb->rx_buffer_size, PKTALIGN));
Quoted 8 lines

Ah, for MACB, looks it should be "MACB_RX_BUFFER_SIZE * MACB_RX_RING_SIZE", but for GEM, I am not sure.

Should it be:

           flush_dcache_range(macb->rx_buffer_dma, macb->rx_buffer_dma +
                            ALIGN(macb->rx_buffer_size *

MACB_RX_RING_SIZE, PKTALIGN));

Yes, this looks correct. v2 will follow soon.

Thanks,
Stefan

#4 lore raw

PATCH 1/1 net: macb: Fix rx buffer cache handling

Quoted 54 lines · Stefan Roese

Hi Bin,

Quoted 50 lines · Bin Meng
Quoted 39 lines · Bin Meng

Hi Stefan,

Quoted 26 lines · Stefan Roese

With commit c6d07bf440bc ("net/macb: increase RX buffer size for GEM") ethernet support does not work any more with d-cache enabled on the AT91SAM. The reason is, that MACB_RX_BUFFER_SIZE was changed from 4096 to 128 but this change was not refected in the rx_buffer flush and invalidate functions, as these also use this macro.

This patch now fixes this by calculating the rx buffer size correctly again in those functions. With this change, ethernet works again reliably on my AT91SAM board.

Cc Ramon Fried <[email protected]>
Cc Eugen Hristev <[email protected]>
Cc Anup Patel <[email protected]>
Cc Bin Meng <[email protected]>
Cc Joe Hershberger <[email protected]>
---
  drivers/net/macb.c | 6 ++++--
  1 file changed, 4 insertions(+), 2 deletions(-)
drivers/net/macb.c +0 -0
index c99cf663a4..fca380d012 100644

drivers/net/macb.c static inline void macb_flush_ring_desc(struct macb_device *macb, bool rx) · line 300

Quoted 6 lines
drivers/net/macb.c +2 -1
@@ -296,13 +296,15 @@ static inline void macb_flush_ring_desc(struct macb_device *macb, bool rx)
  static inline void macb_flush_rx_buffer(struct macb_device *macb)
  {
         flush_dcache_range(macb->rx_buffer_dma, macb->rx_buffer_dma +
-                          ALIGN(MACB_RX_BUFFER_SIZE, PKTALIGN));
+                          ALIGN(MACB_RX_BUFFER_SIZE * MACB_RX_RING_SIZE,
+                                PKTALIGN));

This looks wrong to me.

Quoted 4 lines

Shouldn't it be:

          flush_dcache_range(macb->rx_buffer_dma, macb->rx_buffer_dma +
                           ALIGN(macb->rx_buffer_size, PKTALIGN));

Ah, for MACB, looks it should be "MACB_RX_BUFFER_SIZE * MACB_RX_RING_SIZE", but for GEM, I am not sure.

Should it be:

Quoted 8 lines
Quoted 3 lines
           flush_dcache_range(macb->rx_buffer_dma, macb->rx_buffer_dma +
                            ALIGN(macb->rx_buffer_size *

MACB_RX_RING_SIZE, PKTALIGN));

Yes, this looks correct. v2 will follow soon.

Thanks,
Stefan

Thanks Stefan,
Can you fix your commit message to show: Fixes:...

You can use the following bash alias to generate the line: (pass c6d07bf440bc as argument)
alias gfixes='git --no-pager show -s --abbrev-commit --abbrev=12 --pretty=format:"Fixes: %h (\"%s\")%n"'

#5 lore raw

PATCH 1/1 net: macb: Fix rx buffer cache handling

Quoted 11 lines
Quoted 10 lines
Quoted 5 lines

Should it be:

           flush_dcache_range(macb->rx_buffer_dma, macb->rx_buffer_dma +
                            ALIGN(macb->rx_buffer_size *

MACB_RX_RING_SIZE, PKTALIGN));

Yes, this looks correct. v2 will follow soon.

Thanks,
Stefan

The entire cache handling is broken in this driver, I will fix it in future patches.
Thanks, Ramon.

#6 lore raw

PATCH 1/1 net: macb: Fix rx buffer cache handling

In reply to Bin Meng
Quoted 46 lines · Bin Meng
Quoted 39 lines · Bin Meng

Hi Stefan,

Quoted 26 lines · Stefan Roese

With commit c6d07bf440bc ("net/macb: increase RX buffer size for GEM") ethernet support does not work any more with d-cache enabled on the AT91SAM. The reason is, that MACB_RX_BUFFER_SIZE was changed from 4096 to 128 but this change was not refected in the rx_buffer flush and invalidate functions, as these also use this macro.

This patch now fixes this by calculating the rx buffer size correctly again in those functions. With this change, ethernet works again reliably on my AT91SAM board.

Cc Ramon Fried <[email protected]>
Cc Eugen Hristev <[email protected]>
Cc Anup Patel <[email protected]>
Cc Bin Meng <[email protected]>
Cc Joe Hershberger <[email protected]>
---
 drivers/net/macb.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)
drivers/net/macb.c +0 -0
index c99cf663a4..fca380d012 100644

drivers/net/macb.c static inline void macb_flush_ring_desc(struct macb_device *macb, bool rx) · line 300

Quoted 6 lines
drivers/net/macb.c +2 -1
@@ -296,13 +296,15 @@ static inline void macb_flush_ring_desc(struct macb_device *macb, bool rx)
 static inline void macb_flush_rx_buffer(struct macb_device *macb)
 {
        flush_dcache_range(macb->rx_buffer_dma, macb->rx_buffer_dma +
-                          ALIGN(MACB_RX_BUFFER_SIZE, PKTALIGN));
+                          ALIGN(MACB_RX_BUFFER_SIZE * MACB_RX_RING_SIZE,
+                                PKTALIGN));

This looks wrong to me.

Quoted 4 lines

Shouldn't it be:

         flush_dcache_range(macb->rx_buffer_dma, macb->rx_buffer_dma +
                          ALIGN(macb->rx_buffer_size, PKTALIGN));
Quoted 9 lines

Ah, for MACB, looks it should be "MACB_RX_BUFFER_SIZE * MACB_RX_RING_SIZE", but for GEM, I am not sure.

Should it be:

          flush_dcache_range(macb->rx_buffer_dma, macb->rx_buffer_dma +
                           ALIGN(macb->rx_buffer_size *

MACB_RX_RING_SIZE, PKTALIGN));

Good catch !, I'll issue a fix soon.
Thanks,
Ramon

Quoted 16 lines
Quoted 11 lines
Quoted 11 lines
drivers/net/macb.c +2 -1
 }
 
 static inline void macb_invalidate_rx_buffer(struct macb_device *macb)
 {
        invalidate_dcache_range(macb->rx_buffer_dma, macb->rx_buffer_dma +
-                               ALIGN(MACB_RX_BUFFER_SIZE, PKTALIGN));
+                               ALIGN(MACB_RX_BUFFER_SIZE * MACB_RX_RING_SIZE,
+                                     PKTALIGN));
 }
 
Quoted 2 lines

drivers/net/macb.c static inline void macb_flush_ring_desc(struct macb_device *macb, bool rx)

Quoted 2 lines
drivers/net/macb.c +0 -1
 #if defined(CONFIG_CMD_NET)
--

Regards,
Bin