Hello, Am 14.03.19 um 15:04 schrieb Dan Murphy: > Hello > > On 3/14/19 2:26 AM, Wolfgang Grandegger wrote: >> Hello Dan, >> >> we are close... >> >> Am 13.03.19 um 17:26 schrieb Dan Murphy: >>> Create a m_can platform framework that peripherial >>> devices can register to and use common code and register sets. >>> The peripherial devices may provide read/write and configuration >>> support of the IP. >>> >>> Signed-off-by: Dan Murphy <dmurphy@xxxxxx> >>> --- >>> >>> v9 - Added back the CSR clearing as this is needed for TCAN device, fixed clean >>> function to clean index for version > 30, removed extra skb free and fixed work >>> handler to handle FIFO full for perpheral devices - https://lore.kernel.org/patchwork/patch/1050120/ >>> >>> v8 - Added a clean function for the tx_skb, cleaned up skb on BUS_OFF and on xmit >>> BUSY - https://lore.kernel.org/patchwork/patch/1047980/ >>> v7 - Fixed remaining new checkpatch issues, removed CSR setting, fixed tx hard >>> start function to return tx_busy, and renamed device callbacks - https://lore.kernel.org/patchwork/patch/1047220/ >>> v6 - Squashed platform patch to this patch for bissectablity, fixed coding style >>> issues, updated Kconfig help, placed mcan reg offsets back into c file, renamed >>> priv->skb to priv->tx_skb and cleared perp interrupts at ISR start - >>> Patch 1 comments - https://lore.kernel.org/patchwork/patch/1042446/ >>> Patch 2 comments - https://lore.kernel.org/patchwork/patch/1042442/ >>> >>> drivers/net/can/m_can/Kconfig | 13 +- >>> drivers/net/can/m_can/Makefile | 1 + >>> drivers/net/can/m_can/m_can.c | 733 +++++++++++++------------ >>> drivers/net/can/m_can/m_can.h | 110 ++++ >>> drivers/net/can/m_can/m_can_platform.c | 202 +++++++ >>> 5 files changed, 714 insertions(+), 345 deletions(-) >>> create mode 100644 drivers/net/can/m_can/m_can.h >>> create mode 100644 drivers/net/can/m_can/m_can_platform.c >>> >>> diff --git a/drivers/net/can/m_can/Kconfig b/drivers/net/can/m_can/Kconfig >>> index 04f20dd39007..f7119fd72df4 100644 >>> --- a/drivers/net/can/m_can/Kconfig >>> +++ b/drivers/net/can/m_can/Kconfig >>> @@ -1,5 +1,14 @@ >>> config CAN_M_CAN >>> + tristate "Bosch M_CAN support" >>> + ---help--- >>> + Say Y here if you want support for Bosch M_CAN controller framework. >>> + This is common support for devices that embed the Bosch M_CAN IP. >>> + >>> +config CAN_M_CAN_PLATFORM >>> + tristate "Bosch M_CAN support for io-mapped devices" >>> depends on HAS_IOMEM >>> - tristate "Bosch M_CAN devices" >>> + depends on CAN_M_CAN >>> ---help--- >>> - Say Y here if you want to support for Bosch M_CAN controller. >>> + Say Y here if you want support for IO Mapped Bosch M_CAN controller. >>> + This support is for devices that have the Bosch M_CAN controller >>> + IP embedded into the device and the IP is IO Mapped to the processor. >>> diff --git a/drivers/net/can/m_can/Makefile b/drivers/net/can/m_can/Makefile >>> index 8bbd7f24f5be..057bbcdb3c74 100644 >>> --- a/drivers/net/can/m_can/Makefile >>> +++ b/drivers/net/can/m_can/Makefile >>> @@ -3,3 +3,4 @@ >>> # >>> >>> obj-$(CONFIG_CAN_M_CAN) += m_can.o >>> +obj-$(CONFIG_CAN_M_CAN_PLATFORM) += m_can_platform.o >>> diff --git a/drivers/net/can/m_can/m_can.c b/drivers/net/can/m_can/m_can.c >>> index 9b449400376b..56336aefc567 100644 >>> --- a/drivers/net/can/m_can/m_can.c >>> +++ b/drivers/net/can/m_can/m_can.c ... snip ... >>> +static netdev_tx_t m_can_start_xmit(struct sk_buff *skb, >>> + struct net_device *dev) >>> +{ >>> + struct m_can_priv *priv = netdev_priv(dev); >>> + >>> + if (can_dropped_invalid_skb(dev, skb)) >>> + return NETDEV_TX_OK; >>> + >>> + if (priv->is_peripherial) { >>> + if (priv->tx_skb) { >>> + netdev_err(dev, "hard_xmit called while tx busy\n"); >>> + return NETDEV_TX_BUSY; >>> + } >>> + >>> + if (priv->can.state == CAN_STATE_BUS_OFF) { >>> + m_can_clean(dev); >>> + } else { >>> + priv->tx_skb = skb; >>> + queue_work(priv->tx_wq, &priv->tx_work); >> >> You removed "netif_stop_queue(cdev->net)" here. Does that work with just >> *one* "tx_skb"? I think you need an array then!? I'm afraid, this will >> require quite some changes and you should also measure the benefit of >> queuing outgoing messages first. For the time being, it would be fine >> for me to stop the queue here and add an appropriate comment. >> > > Yes I removed this in favor of allowing the work handler to stop the queue. > > If the version is < 30 then the queue is stopped immediately. If the version > 30 then the queue is > only stopped if FIFO is full. Yes, and the next call to m_can_start_xmit() will probably return with NETDEV_TX_BUSY. You should at least add a "netif_stop_queue(cdev->net)" before returning NETDEV_TX_BUSY, otherwice the upper layer will retry immediately resulting in high CPU load. Anyway, the purpose of this return code is to signal errors/bugs in the start/stop flow control: https://elixir.bootlin.com/linux/latest/source/Documentation/networking/netdevices.txt#L81 > > I did see the message "tcan4x5x spi1.0 can0: hard_xmit called while tx busy" > more with no gap and no polling in the cangen request. You usually should never see that message. > >> BTW, how big is the FIFO size on your system: >> > > This value is set to 1. > >> tx_fifo_size = mram_config_vals[7]; Then I'm surprised that you see "hard_xmit called while tx busy" at all. Are you sure that "cdev->version" is 32 on the tcan4x5? Wolfgang
