| Message ID | 20180628204538.866-2-wsa+renesas@sang-engineering.com |
|---|---|
| State | Accepted |
| Headers | show |
| Series | [1/1] i2c: rcar: handle RXDMA HW behaviour on Gen3 | expand |
Hi Wolfram, On Thu, Jun 28, 2018 at 10:45 PM Wolfram Sang <wsa+renesas@sang-engineering.com> wrote: > On Gen3, we can only do RXDMA once per transfer reliably. For that, we > must reset the device, then we can have RXDMA once. This patch > implements this. When there is no reset controller or the reset fails, > RXDMA will be blocked completely. Otherwise, it will be disabled after > the first RXDMA transfer. Based on a commit from the BSP by Hiromitsu > Yamasaki, yet completely refactored to handle multiple read messages > within one transfer. > > Signed-off-by: Wolfram Sang <wsa+renesas@sang-engineering.com> Thanks for your patch! > --- a/drivers/i2c/busses/i2c-rcar.c > +++ b/drivers/i2c/busses/i2c-rcar.c > @@ -751,6 +780,16 @@ static int rcar_i2c_master_xfer(struct i2c_adapter *adap, > > pm_runtime_get_sync(dev); > > + /* Gen3 needs a reset before allowing RXDMA once */ > + if (priv->devtype == I2C_RCAR_GEN3) { So on R-Car Gen3 the device is always reset, even if no reads will be done? Or is that something you cannot check for at this point? > + priv->flags |= ID_P_NO_RXDMA; > + if (!IS_ERR(priv->rstc)) { > + ret = rcar_i2c_do_reset(priv); > + if (ret == 0) > + priv->flags &= ~ID_P_NO_RXDMA; > + } > + } > + > rcar_i2c_init(priv); > > ret = rcar_i2c_bus_barrier(priv); > @@ -921,6 +960,15 @@ static int rcar_i2c_probe(struct platform_device *pdev) > if (ret < 0) > goto out_pm_put; > > + if (priv->devtype == I2C_RCAR_GEN3) { > + priv->rstc = devm_reset_control_get_exclusive(&pdev->dev, NULL); > + if (!IS_ERR(priv->rstc)) { > + ret = reset_control_status(priv->rstc); Why this call and check? To check if .status() is implemented (it always is on R-Car Gen3, if CONFIG_RESET_CONTROLLER is enabled), and to avoid the timeout in rcar_i2c_do_reset() on every transfer? > + if (ret < 0) > + priv->rstc = ERR_PTR(-ENOTSUPP); > + } > + } Anyway: Reviewed-by: Geert Uytterhoeven <geert+renesas@glider.be> Gr{oetje,eeting}s, Geert
> > + /* Gen3 needs a reset before allowing RXDMA once */ > > + if (priv->devtype == I2C_RCAR_GEN3) { > > So on R-Car Gen3 the device is always reset, even if no reads will be done? > Or is that something you cannot check for at this point? You got a point. I copied the behaviour from the BSP to do this as early as possible. Still, it might be worthwhile to merge this into rcar_i2c_request_dma() and move the call to it to the front. Will check. > > + if (priv->devtype == I2C_RCAR_GEN3) { > > + priv->rstc = devm_reset_control_get_exclusive(&pdev->dev, NULL); > > + if (!IS_ERR(priv->rstc)) { > > + ret = reset_control_status(priv->rstc); > > Why this call and check? To check if .status() is implemented (it always is > on R-Car Gen3, if CONFIG_RESET_CONTROLLER is enabled), and to avoid the > timeout in rcar_i2c_do_reset() on every transfer? Yes. We can save all the hazzle if it is not implemented. And I didn't want to implicitly assume that it will be implemented anywhere forever. That could be a very subtle bug later. Thanks, Wolfram
On Fri, Jun 29, 2018 at 09:54:04AM +0200, Wolfram Sang wrote: > > > > + /* Gen3 needs a reset before allowing RXDMA once */ > > > + if (priv->devtype == I2C_RCAR_GEN3) { > > > > So on R-Car Gen3 the device is always reset, even if no reads will be done? > > Or is that something you cannot check for at this point? > > You got a point. I copied the behaviour from the BSP to do this as early > as possible. Still, it might be worthwhile to merge this into > rcar_i2c_request_dma() and move the call to it to the front. Will check. I will apply this patch as is because it is easier to backport to stable this way. You are still right, we can do better and not issue a reset any time. To get there, though, I want two kinds of refactoring beforehand. And I think it is best to do this on top of this patch and refactor it along. BSP backporters can then decide if they want the refactored version or not.
On Thu, Jun 28, 2018 at 10:45:38PM +0200, Wolfram Sang wrote: > On Gen3, we can only do RXDMA once per transfer reliably. For that, we > must reset the device, then we can have RXDMA once. This patch > implements this. When there is no reset controller or the reset fails, > RXDMA will be blocked completely. Otherwise, it will be disabled after > the first RXDMA transfer. Based on a commit from the BSP by Hiromitsu > Yamasaki, yet completely refactored to handle multiple read messages > within one transfer. > > Signed-off-by: Wolfram Sang <wsa+renesas@sang-engineering.com> Applied to for-next, thanks!
On Mon, Jul 23, 2018 at 07:55:22PM +0200, Wolfram Sang wrote: > On Thu, Jun 28, 2018 at 10:45:38PM +0200, Wolfram Sang wrote: > > On Gen3, we can only do RXDMA once per transfer reliably. For that, we > > must reset the device, then we can have RXDMA once. This patch > > implements this. When there is no reset controller or the reset fails, > > RXDMA will be blocked completely. Otherwise, it will be disabled after > > the first RXDMA transfer. Based on a commit from the BSP by Hiromitsu > > Yamasaki, yet completely refactored to handle multiple read messages > > within one transfer. > > > > Signed-off-by: Wolfram Sang <wsa+renesas@sang-engineering.com> > > Applied to for-next, thanks! Reverted it from for-next, added to for-current and added stable.
diff --git a/drivers/i2c/busses/i2c-rcar.c b/drivers/i2c/busses/i2c-rcar.c index c67c0b1a7fc9..b95900769a90 100644 --- a/drivers/i2c/busses/i2c-rcar.c +++ b/drivers/i2c/busses/i2c-rcar.c @@ -32,6 +32,7 @@ #include <linux/of_device.h> #include <linux/platform_device.h> #include <linux/pm_runtime.h> +#include <linux/reset.h> #include <linux/slab.h> /* register offsets */ @@ -111,8 +112,9 @@ #define ID_ARBLOST (1 << 3) #define ID_NACK (1 << 4) /* persistent flags */ +#define ID_P_NO_RXDMA (1 << 30) /* HW forbids RXDMA sometimes */ #define ID_P_PM_BLOCKED (1 << 31) -#define ID_P_MASK ID_P_PM_BLOCKED +#define ID_P_MASK (ID_P_PM_BLOCKED | ID_P_NO_RXDMA) enum rcar_i2c_type { I2C_RCAR_GEN1, @@ -141,6 +143,8 @@ struct rcar_i2c_priv { struct dma_chan *dma_rx; struct scatterlist sg; enum dma_data_direction dma_direction; + + struct reset_control *rstc; }; #define rcar_i2c_priv_to_dev(p) ((p)->adap.dev.parent) @@ -371,6 +375,11 @@ static void rcar_i2c_dma_unmap(struct rcar_i2c_priv *priv) dma_unmap_single(chan->device->dev, sg_dma_address(&priv->sg), sg_dma_len(&priv->sg), priv->dma_direction); + /* Gen3 can only do one RXDMA per transfer and we just completed it */ + if (priv->devtype == I2C_RCAR_GEN3 && + priv->dma_direction == DMA_FROM_DEVICE) + priv->flags |= ID_P_NO_RXDMA; + priv->dma_direction = DMA_NONE; } @@ -408,8 +417,9 @@ static void rcar_i2c_dma(struct rcar_i2c_priv *priv) unsigned char *buf; int len; - /* Do not use DMA if it's not available or for messages < 8 bytes */ - if (IS_ERR(chan) || msg->len < 8 || !(msg->flags & I2C_M_DMA_SAFE)) + /* Do various checks to see if DMA is feasible at all */ + if (IS_ERR(chan) || msg->len < 8 || !(msg->flags & I2C_M_DMA_SAFE) || + (read && priv->flags & ID_P_NO_RXDMA)) return; if (read) { @@ -740,6 +750,25 @@ static void rcar_i2c_release_dma(struct rcar_i2c_priv *priv) } } +/* I2C is a special case, we need to poll the status of a reset */ +static int rcar_i2c_do_reset(struct rcar_i2c_priv *priv) +{ + int i, ret; + + ret = reset_control_reset(priv->rstc); + if (ret) + return ret; + + for (i = 0; i < LOOP_TIMEOUT; i++) { + ret = reset_control_status(priv->rstc); + if (ret == 0) + return 0; + udelay(1); + } + + return -ETIMEDOUT; +} + static int rcar_i2c_master_xfer(struct i2c_adapter *adap, struct i2c_msg *msgs, int num) @@ -751,6 +780,16 @@ static int rcar_i2c_master_xfer(struct i2c_adapter *adap, pm_runtime_get_sync(dev); + /* Gen3 needs a reset before allowing RXDMA once */ + if (priv->devtype == I2C_RCAR_GEN3) { + priv->flags |= ID_P_NO_RXDMA; + if (!IS_ERR(priv->rstc)) { + ret = rcar_i2c_do_reset(priv); + if (ret == 0) + priv->flags &= ~ID_P_NO_RXDMA; + } + } + rcar_i2c_init(priv); ret = rcar_i2c_bus_barrier(priv); @@ -921,6 +960,15 @@ static int rcar_i2c_probe(struct platform_device *pdev) if (ret < 0) goto out_pm_put; + if (priv->devtype == I2C_RCAR_GEN3) { + priv->rstc = devm_reset_control_get_exclusive(&pdev->dev, NULL); + if (!IS_ERR(priv->rstc)) { + ret = reset_control_status(priv->rstc); + if (ret < 0) + priv->rstc = ERR_PTR(-ENOTSUPP); + } + } + /* Stay always active when multi-master to keep arbitration working */ if (of_property_read_bool(dev->of_node, "multi-master")) priv->flags |= ID_P_PM_BLOCKED;
On Gen3, we can only do RXDMA once per transfer reliably. For that, we must reset the device, then we can have RXDMA once. This patch implements this. When there is no reset controller or the reset fails, RXDMA will be blocked completely. Otherwise, it will be disabled after the first RXDMA transfer. Based on a commit from the BSP by Hiromitsu Yamasaki, yet completely refactored to handle multiple read messages within one transfer. Signed-off-by: Wolfram Sang <wsa+renesas@sang-engineering.com> --- Changes since RFC: * poll reset status bit after reset (special case for I2C IP core). * refactor I2C reset into seperate function * improved comments * use newer reset API (devm_reset_control_get_exclusive) drivers/i2c/busses/i2c-rcar.c | 54 ++++++++++++++++++++++++++++++++++++++++--- 1 file changed, 51 insertions(+), 3 deletions(-)