D_CAN RX buffer size improvements
Subsystems (3)
cover D_CAN RX buffer size improvements
Re-sending entire patchset due to missed cover letter, sorry.
This patchset introduces support for 64 D_CAN message objects with an option of unequal split between RX/TX.
The rationale behind this is that there are lots of frame loss on higher bus speeds. Below are test results from my custom Sitara AM3352 based board:
Sender: timeout 15m cangen can0 -g 0 -i x
Target: candump can0,0~0,#FFFFFFFF -td -c -d -e
* Without patches:
- 15 minute RX test, 500kbps
- 16 RX / 16 TX message objects
- 77 received frames lost out of 4649415
* With patches applied:
- 15 hours RX test, 500kbps
- 56 RX / 8 TX message objects
- 41 received frames lost out of 279303376
Please note, I do not have ability to test pure C_CAN, so it is left untested.
---
Andrejs Cainikovs (2):
can: c_can: support 64 message objects for D_CAN
can: c_can: configurable amount of D_CAN RX objects
drivers/net/can/c_can/Kconfig | 20 ++++++++++ drivers/net/can/c_can/c_can.c | 93 +++++++++++++++++++++++++++---------------- drivers/net/can/c_can/c_can.h | 20 +++++++--- 3 files changed, 94 insertions(+), 39 deletions(-)
---
2.11.0
PATCH 1/2 can: c_can: support 64 message objects for D_CAN
D_CAN supports up to 128 message objects, comparing to 32 on C_CAN. However, some CPUs with D_CAN controller have their own limits:
TI AM335x Sitara CPU, for example, supports max of 64 message objects.
This patch extends max D_CAN message objects up to 64.
--- drivers/net/can/c_can/Kconfig | 12 ++++++++++++ drivers/net/can/c_can/c_can.c | 42 ++++++++++++++++++++++-------------------- drivers/net/can/c_can/c_can.h | 20 ++++++++++++++++---- 3 files changed, 50 insertions(+), 24 deletions(-)
drivers/net/can/c_can/Kconfig +12 -0
@@ -20,4 +20,16 @@ config CAN_C_CAN_PCI ---help--- This driver adds support for the C_CAN/D_CAN chips connected to the PCI bus. + +config CAN_C_CAN_DCAN_64_MSG_OBJECTS + bool "Use 64 message objects for D_CAN" + default n + ---help--- + D_CAN supports up to 128 message objects, comparing to 32 on + C_CAN. However, some CPUs with D_CAN controller have their + own limits: TI AM335x Sitara CPU, for example, supports max + of 64 message objects. + Enabling this option extends max D_CAN message objects up to + 64. + endif
drivers/net/can/c_can/c_can.c +22 -20
72 lines not on this page · raw message
drivers/net/can/c_can/c_can.h +16 -4
@@ -23,9 +23,15 @@#define C_CAN_H /* message object split */ + +#ifdef CONFIG_CAN_C_CAN_DCAN_64_MSG_OBJECTS +#define C_CAN_NO_OF_OBJECTS 64 +#else #define C_CAN_NO_OF_OBJECTS 32 -#define C_CAN_MSG_OBJ_RX_NUM 16 -#define C_CAN_MSG_OBJ_TX_NUM 16 +#endif + +#define C_CAN_MSG_OBJ_TX_NUM (C_CAN_NO_OF_OBJECTS >> 1) +#define C_CAN_MSG_OBJ_RX_NUM (C_CAN_NO_OF_OBJECTS - C_CAN_MSG_OBJ_TX_NUM)#define C_CAN_MSG_OBJ_RX_FIRST 1#define C_CAN_MSG_OBJ_RX_LAST (C_CAN_MSG_OBJ_RX_FIRST + \
@@ -35,9 +41,11 @@#define C_CAN_MSG_OBJ_TX_LAST (C_CAN_MSG_OBJ_TX_FIRST + \ C_CAN_MSG_OBJ_TX_NUM - 1) -#define C_CAN_MSG_OBJ_RX_SPLIT 9 -#define C_CAN_MSG_RX_LOW_LAST (C_CAN_MSG_OBJ_RX_SPLIT - 1) +#ifdef CONFIG_CAN_C_CAN_DCAN_64_MSG_OBJECTS +#define RECEIVE_OBJECT_BITS 0xffffffff +#else #define RECEIVE_OBJECT_BITS 0x0000ffff +#endif enum reg {C_CAN_CTRL_REG =0 ,
@@ -76,6 +84,8 @@ enum reg {C_CAN_NEWDAT2_REG ,C_CAN_INTPND1_REG ,C_CAN_INTPND2_REG , +C_CAN_INTPND3_REG , +C_CAN_INTPND4_REG ,C_CAN_MSGVAL1_REG ,C_CAN_MSGVAL2_REG , C_CAN_FUNCTION_REG,
@@ -137,6 +147,8 @@ static const u16 reg_map_d_can[] = { [C_CAN_NEWDAT2_REG] = 0x9E, [C_CAN_INTPND1_REG] = 0xB0, [C_CAN_INTPND2_REG] = 0xB2, + [C_CAN_INTPND3_REG] = 0xB4, + [C_CAN_INTPND4_REG] = 0xB6, [C_CAN_MSGVAL1_REG] = 0xC4, [C_CAN_MSGVAL2_REG] = 0xC6, [C_CAN_IF1_COMREQ_REG] = 0x100,
Signature · 1 line
2.11.0
PATCH 2/2 can: c_can: configurable amount of D_CAN RX objects
In reply to Andrejs CainikovsMake number of D_CAN RX message objects configurable. This will allow having bigger (or smaller) RX buffer instead of 50/50 split for RX/TX.
--- drivers/net/can/c_can/Kconfig | 8 ++++++ drivers/net/can/c_can/c_can.c | 64 +++++++++++++++++++++++++++++-------------- drivers/net/can/c_can/c_can.h | 16 +++++------ 3 files changed, 58 insertions(+), 30 deletions(-)
drivers/net/can/c_can/Kconfig +8 -0
@@ -32,4 +32,12 @@ config CAN_C_CAN_DCAN_64_MSG_OBJECTS Enabling this option extends max D_CAN message objects up to 64. +config CAN_C_CAN_DCAN_RX_MSG_OBJECTS + int "Specify amount of D_CAN RX message objects" + depends on CAN_C_CAN_DCAN_64_MSG_OBJECTS + default 32 + ---help--- + Use specific number of message objects for RX, instead of + 50/50 split between RX/TX. + endif
drivers/net/can/c_can/c_can.c +43 -21
123 lines not on this page · raw message
drivers/net/can/c_can/c_can.h +7 -9
@@ -26,12 +26,12 @@#ifdef CONFIG_CAN_C_CAN_DCAN_64_MSG_OBJECTS #define C_CAN_NO_OF_OBJECTS 64 +#define C_CAN_MSG_OBJ_RX_NUM CONFIG_CAN_C_CAN_DCAN_RX_MSG_OBJECTS#else #define C_CAN_NO_OF_OBJECTS 32 +#define C_CAN_MSG_OBJ_RX_NUM 16#endif - -#define C_CAN_MSG_OBJ_TX_NUM (C_CAN_NO_OF_OBJECTS >> 1) -#define C_CAN_MSG_OBJ_RX_NUM (C_CAN_NO_OF_OBJECTS - C_CAN_MSG_OBJ_TX_NUM) +#define C_CAN_MSG_OBJ_TX_NUM (C_CAN_NO_OF_OBJECTS - C_CAN_MSG_OBJ_RX_NUM)#define C_CAN_MSG_OBJ_RX_FIRST 1#define C_CAN_MSG_OBJ_RX_LAST (C_CAN_MSG_OBJ_RX_FIRST + \
@@ -41,12 +41,6 @@#define C_CAN_MSG_OBJ_TX_LAST (C_CAN_MSG_OBJ_TX_FIRST + \ C_CAN_MSG_OBJ_TX_NUM - 1) -#ifdef CONFIG_CAN_C_CAN_DCAN_64_MSG_OBJECTS -#define RECEIVE_OBJECT_BITS 0xffffffff -#else -#define RECEIVE_OBJECT_BITS 0x0000ffff -#endif -enum reg {C_CAN_CTRL_REG =0 ,C_CAN_CTRL_EX_REG ,
@@ -82,6 +76,8 @@ enum reg {C_CAN_TXRQST2_REG ,C_CAN_NEWDAT1_REG ,C_CAN_NEWDAT2_REG , +C_CAN_NEWDAT3_REG , +C_CAN_NEWDAT4_REG ,C_CAN_INTPND1_REG ,C_CAN_INTPND2_REG , C_CAN_INTPND3_REG,
@@ -145,6 +141,8 @@ static const u16 reg_map_d_can[] = { [C_CAN_TXRQST2_REG] = 0x8A, [C_CAN_NEWDAT1_REG] = 0x9C, [C_CAN_NEWDAT2_REG] = 0x9E, + [C_CAN_NEWDAT3_REG] = 0xA0, + [C_CAN_NEWDAT4_REG] = 0xA2, [C_CAN_INTPND1_REG] = 0xB0, [C_CAN_INTPND2_REG] = 0xB2, [C_CAN_INTPND3_REG] = 0xB4,
Signature · 1 line
2.11.0
Hi everyone,
This series was recently brought to my attention, in connection to a long-standing
packet drop issue that we had on a Sitara-based product.
I haven't tested this personally, but I've been notified that this was backported
to the v4.19 kernel, and the packet drop was fixed.
It seems the series never managed to get upstreamed,
but I think this should be resurrected and merged (probably with after some cleanup/review).
Thanks,
Ezequiel
Quoted 38 lines · Andrejs Cainikovs
Re-sending entire patchset due to missed cover letter, sorry.
This patchset introduces support for 64
D_CANmessage objects with an option of unequal split between RX/TX.The rationale behind this is that there are lots of frame loss on higher bus speeds. Below are test results from my custom Sitara AM3352 based board:
Sender: timeout 15m cangen can0 -g 0 -i x
Target: candump can0,0~0,#FFFFFFFF -td -c -d -e* Without patches:
- 15 minute RX test, 500kbps
- 16 RX / 16 TX message objects
- 77 received frames lost out of 4649415* With patches applied:
- 15 hours RX test, 500kbps
- 56 RX / 8 TX message objects
- 41 received frames lost out of 279303376Please note, I do not have ability to test pure
C_CAN, so it is left untested.---
Andrejs Cainikovs (2):
can:c_can: support 64 message objects forD_CAN
can:c_can: configurable amount ofD_CANRX objectsdrivers/net/can/c_can/Kconfig | 20 ++++++++++ drivers/net/can/c_can/c_can.c | 93 +++++++++++++++++++++++++++---------------- drivers/net/can/c_can/c_can.h | 20 +++++++--- 3 files changed, 94 insertions(+), 39 deletions(-)---
2.11.0
Hi Ezequiel,
Sorry for a late reply. I'm the author of this patch set, and I will have a look at this after I obtain the hardware. I hope this is still relevant.
Best regards,
Andrejs.
Quoted 3 lines
Hi everyone,
This series was recently brought to my attention, in connection to a
long-standing packet drop issue that we had on a Sitara-based product.
Quoted 2 lines
I haven't tested this personally, but I've been notified that this
was backported to the v4.19 kernel, and the packet drop was fixed.
Quoted 2 lines
It seems the series never managed to get upstreamed, but I think this
should be resurrected and merged (probably with after some cleanup/review).
Quoted 3 lines
Thanks,
Ezequiel
Marc Kleine-Budde
maintainerQuoted 3 lines · Andrejs Cainikovs
Sorry for a late reply. I'm the author of this patch set, and I will have a look at this after I obtain the hardware. I hope this is still relevant.
Dario (Cc'ed) created a proper patch series to support 64 message objects. The series has been mainlined in:
https://git.kernel.org/linus/132f2d45fb2302a582aef617ea766f3fa52a084c
Marc
Signature · 4 lines
Pengutronix e.K. | Marc Kleine-Budde | Embedded Linux | https://www.pengutronix.de | Vertretung West/Dortmund | Phone: +49-231-2826-924 | Amtsgericht Hildesheim, HRA 2686 | Fax: +49-5121-206917-5555 |
(Adding Federico and Max)
Quoted 10 lines · Marc Kleine-Budde
Quoted 3 lines · Andrejs Cainikovs
Sorry for a late reply. I'm the author of this patch set, and I will have a look at this after I obtain the hardware. I hope this is still relevant.
Dario (Cc'ed) created a proper patch series to support 64 message objects. The series has been mainlined in:
https://git.kernel.org/linus/132f2d45fb2302a582aef617ea766f3fa52a084c
Ah, that's really great news.
Thanks a lot Marc and Dario.
Signature · 2 lines
Kindly, Ezequiel