There is ecspi ERR009165 on i.mx6/7 soc family, which cause FIFO
transfer to be send twice in DMA mode. Please get more information from:
https://www.nxp.com/docs/en/errata/IMX6DQCE.pdf. The workaround is adding
new sdma ram script which works in XCH mode as PIO inside sdma instead
of SMC mode, meanwhile, 'TX_THRESHOLD' should be 0. The issue should be
exist on all legacy i.mx6/7 soc family before i.mx6ul.
NXP fix this design issue from i.mx6ul, so newer chips including i.mx6ul/
6ull/6sll do not need this workaroud anymore. All other i.mx6/7/8 chips
still need this workaroud. This patch set add new 'fsl,imx6ul-ecspi'
for ecspi driver and 'ecspi_fixed' in sdma driver to choose if need errata
or not.
The first two reverted patches should be the same issue, though, it
seems 'fixed' by changing to other shp script. Hope Sean or Sascha could
have the chance to test this patch set if could fix their issues.
Besides, enable sdma support for i.mx8mm/8mq and fix ecspi1 not work
on i.mx8mm because the event id is zero.
PS:
Please get sdma firmware from below linux-firmware and copy it to your
local rootfs /lib/firmware/imx/sdma.
https://git.kernel.org/pub/scm/linux/kernel/git/firmware/linux-firmware.git/tree/imx/sdma
v2:
1.Add commit log for reverted patches.
2.Add comment for 'ecspi_fixed' in sdma driver.
3.Add 'fsl,imx6sll-ecspi' compatible instead of 'fsl,imx6ul-ecspi'
rather than remove.
v3:
1.Confirm with design team make sure ERR009165 fixed on i.mx6ul/i.mx6ull
/i.mx6sll, not fixed on i.mx8m/8mm and other i.mx6/7 legacy chips.
Correct dts related dts patch in v2.
2.Clean eratta information in binding doc and new 'tx_glitch_fixed' flag
in spi-imx driver to state ERR009165 fixed or not.
3.Enlarge burst size to fifo size for tx since tx_wml set to 0 in the
errata workaroud, thus improve performance as possible.
v4:
1.Add Ack tag from Mark and Vinod
2.Remove checking 'event_id1' zero as 'event_id0'.
v5:
1.Add the last patch for compatible with the current uart driver which
using rom script, so both uart ram script and rom script supported
in latest firmware, by default uart rom script used. UART driver
will be broken without this patch.
v6:
1.Resend after rebase the latest next branch.
2.Remove below No.13~No.15 patches of v5 because they were mergered.
ARM: dts: imx6ul: add dma support on ecspi
ARM: dts: imx6sll: correct sdma compatible
arm64: defconfig: Enable SDMA on i.mx8mq/8mm
3.Revert "dmaengine: imx-sdma: fix context cache" since
'context_loaded' removed.
v7:
1.Put the last patch 13/13 'Revert "dmaengine: imx-sdma: fix context
cache"' to the ahead of 03/13 'Revert "dmaengine: imx-sdma: refine
to load context only once" so that no building waring during comes out
during bisect.
2.Address Sascha's comments, including eliminating any i.mx6sx in this
series, adding new 'is_imx6ul_ecspi()' instead imx in imx51 and taking
care SMC bit for PIO.
3.Add back missing 'Reviewed-by' tag on 08/15(v5):09/13(v7)
'spi: imx: add new i.mx6ul compatible name in binding doc'
v8:
1.remove 0003-Revert-dmaengine-imx-sdma-fix-context-cache.patch and merge
it into 04/13 of v7
2.add 0005-spi-imx-fallback-to-PIO-if-dma-setup-failure.patch for no any
ecspi function broken even if sdma firmware not updated.
3.merge 'tx.dst_maxburst' changes in the two continous patches into one
patch to avoid confusion.
4.fix typo 'duplicated'.
v9:
1. add "spi: imx: add dma_sync_sg_for_device after fallback from dma"
to fix the potential issue brought by commit bcd8e7761ec9("spi: imx:
fallback to PIO if dma setup failure") which is the only one patch
of v8 merged. Thanks Matthias for reporting:
https://lore.kernel.org/linux-arm-kernel/[email protected]/
2. remove 05/13 of v8 "spi: imx:fallback to PIO if dma setup failure"
since it's been merged.
Robin Gong (13):
spi: imx: add dma_sync_sg_for_device after fallback from dma
Revert "ARM: dts: imx6q: Use correct SDMA script for SPI5 core"
Revert "ARM: dts: imx6: Use correct SDMA script for SPI cores"
Revert "dmaengine: imx-sdma: refine to load context only once"
dmaengine: imx-sdma: remove duplicated sdma_load_context
dmaengine: imx-sdma: add mcu_2_ecspi script
spi: imx: fix ERR009165
spi: imx: remove ERR009165 workaround on i.mx6ul
spi: imx: add new i.mx6ul compatible name in binding doc
dmaengine: imx-sdma: remove ERR009165 on i.mx6ul
dma: imx-sdma: add i.mx6ul compatible name
dmaengine: imx-sdma: fix ecspi1 rx dma not work on i.mx8mm
dmaengine: imx-sdma: add uart rom script
.../devicetree/bindings/dma/fsl-imx-sdma.txt | 1 +
.../devicetree/bindings/spi/fsl-imx-cspi.txt | 1 +
arch/arm/boot/dts/imx6q.dtsi | 2 +-
arch/arm/boot/dts/imx6qdl.dtsi | 8 +--
drivers/dma/imx-sdma.c | 67 ++++++++++++--------
drivers/spi/spi-imx.c | 73 +++++++++++++++++++---
include/linux/platform_data/dma-imx-sdma.h | 8 ++-
7 files changed, 120 insertions(+), 40 deletions(-)
--
2.7.4
In case dma transfer failed and fallback to pio, tx_buf/rx_buf need to be
taken care cache since they have already been maintained by spi.c
Fixes: bcd8e7761ec9("spi: imx: fallback to PIO if dma setup failure")
Signed-off-by: Robin Gong <[email protected]>
Reported-by: Matthias Schiffer <[email protected]>
Link: https://lore.kernel.org/linux-arm-kernel/[email protected]/
---
drivers/spi/spi-imx.c | 12 ++++++++++++
1 file changed, 12 insertions(+)
diff --git a/drivers/spi/spi-imx.c b/drivers/spi/spi-imx.c
index b7a85e3..84aebee 100644
--- a/drivers/spi/spi-imx.c
+++ b/drivers/spi/spi-imx.c
@@ -1456,6 +1456,13 @@ static int spi_imx_pio_transfer(struct spi_device *spi,
return -ETIMEDOUT;
}
+ if (transfer->rx_sg.sgl) {
+ struct device *rx_dev = spi->controller->dma_rx->device->dev;
+
+ dma_sync_sg_for_device(rx_dev, transfer->rx_sg.sgl,
+ transfer->rx_sg.nents, DMA_TO_DEVICE);
+ }
+
return transfer->len;
}
@@ -1521,10 +1528,15 @@ static int spi_imx_transfer(struct spi_device *spi,
* firmware may not be updated as ERR009165 required.
*/
if (spi_imx->usedma) {
+ struct device *tx_dev = spi->controller->dma_tx->device->dev;
+
ret = spi_imx_dma_transfer(spi_imx, transfer);
if (ret != -EINVAL)
return ret;
+ dma_sync_sg_for_cpu(tx_dev, transfer->tx_sg.sgl,
+ transfer->tx_sg.nents, DMA_FROM_DEVICE);
+
spi_imx->devtype_data->disable_dma(spi_imx);
spi_imx->usedma = false;
--
2.7.4
This reverts commit ad0d92d7ba6aecbe2705907c38ff8d8be4da1e9c, because
in spi-imx case, burst length may be changed dynamically.
Signed-off-by: Robin Gong <[email protected]>
Acked-by: Sascha Hauer <[email protected]>
---
drivers/dma/imx-sdma.c | 8 --------
1 file changed, 8 deletions(-)
diff --git a/drivers/dma/imx-sdma.c b/drivers/dma/imx-sdma.c
index 9177403..b1f61eb 100644
--- a/drivers/dma/imx-sdma.c
+++ b/drivers/dma/imx-sdma.c
@@ -377,7 +377,6 @@ struct sdma_channel {
unsigned long watermark_level;
u32 shp_addr, per_addr;
enum dma_status status;
- bool context_loaded;
struct imx_dma_data data;
struct work_struct terminate_worker;
};
@@ -984,9 +983,6 @@ static int sdma_load_context(struct sdma_channel *sdmac)
int ret;
unsigned long flags;
- if (sdmac->context_loaded)
- return 0;
-
if (sdmac->direction == DMA_DEV_TO_MEM)
load_address = sdmac->pc_from_device;
else if (sdmac->direction == DMA_DEV_TO_DEV)
@@ -1029,8 +1025,6 @@ static int sdma_load_context(struct sdma_channel *sdmac)
spin_unlock_irqrestore(&sdma->channel_0_lock, flags);
- sdmac->context_loaded = true;
-
return ret;
}
@@ -1069,7 +1063,6 @@ static void sdma_channel_terminate_work(struct work_struct *work)
vchan_get_all_descriptors(&sdmac->vc, &head);
spin_unlock_irqrestore(&sdmac->vc.lock, flags);
vchan_dma_desc_free_list(&sdmac->vc, &head);
- sdmac->context_loaded = false;
}
static int sdma_terminate_all(struct dma_chan *chan)
@@ -1338,7 +1331,6 @@ static void sdma_free_chan_resources(struct dma_chan *chan)
sdmac->event_id0 = 0;
sdmac->event_id1 = 0;
- sdmac->context_loaded = false;
sdma_set_channel_priority(sdmac, 0);
--
2.7.4
Since sdma_transfer_init() will do sdma_load_context before any
sdma transfer, no need once more in sdma_config_channel().
Signed-off-by: Robin Gong <[email protected]>
Acked-by: Vinod Koul <[email protected]>
---
drivers/dma/imx-sdma.c | 5 +----
1 file changed, 1 insertion(+), 4 deletions(-)
diff --git a/drivers/dma/imx-sdma.c b/drivers/dma/imx-sdma.c
index b1f61eb..4440ddb 100644
--- a/drivers/dma/imx-sdma.c
+++ b/drivers/dma/imx-sdma.c
@@ -1137,7 +1137,6 @@ static void sdma_set_watermarklevel_for_p2p(struct sdma_channel *sdmac)
static int sdma_config_channel(struct dma_chan *chan)
{
struct sdma_channel *sdmac = to_sdma_chan(chan);
- int ret;
sdma_disable_channel(chan);
@@ -1177,9 +1176,7 @@ static int sdma_config_channel(struct dma_chan *chan)
sdmac->watermark_level = 0; /* FIXME: M3_BASE_ADDRESS */
}
- ret = sdma_load_context(sdmac);
-
- return ret;
+ return 0;
}
static int sdma_set_channel_priority(struct sdma_channel *sdmac,
--
2.7.4
Add mcu_2_ecspi script to fix ecspi errata ERR009165.
Signed-off-by: Robin Gong <[email protected]>
Acked-by: Vinod Koul <[email protected]>
---
drivers/dma/imx-sdma.c | 3 +++
1 file changed, 3 insertions(+)
diff --git a/drivers/dma/imx-sdma.c b/drivers/dma/imx-sdma.c
index 4440ddb..db4132f 100644
--- a/drivers/dma/imx-sdma.c
+++ b/drivers/dma/imx-sdma.c
@@ -920,6 +920,9 @@ static void sdma_get_pc(struct sdma_channel *sdmac,
emi_2_per = sdma->script_addrs->mcu_2_ata_addr;
break;
case IMX_DMATYPE_CSPI:
+ per_2_emi = sdma->script_addrs->app_2_mcu_addr;
+ emi_2_per = sdma->script_addrs->mcu_2_ecspi_addr;
+ break;
case IMX_DMATYPE_EXT:
case IMX_DMATYPE_SSI:
case IMX_DMATYPE_SAI:
--
2.7.4
Change to XCH mode even in dma mode, please refer to the below
errata:
https://www.nxp.com/docs/en/errata/IMX6DQCE.pdf
Signed-off-by: Robin Gong <[email protected]>
Acked-by: Mark Brown <[email protected]>
---
drivers/spi/spi-imx.c | 10 +++-------
1 file changed, 3 insertions(+), 7 deletions(-)
diff --git a/drivers/spi/spi-imx.c b/drivers/spi/spi-imx.c
index 84aebee..d4fe5dd 100644
--- a/drivers/spi/spi-imx.c
+++ b/drivers/spi/spi-imx.c
@@ -591,8 +591,8 @@ static int mx51_ecspi_prepare_transfer(struct spi_imx_data *spi_imx,
ctrl |= mx51_ecspi_clkdiv(spi_imx, t->speed_hz, &clk);
spi_imx->spi_bus_clk = clk;
- if (spi_imx->usedma)
- ctrl |= MX51_ECSPI_CTRL_SMC;
+ /* ERR009165: work in XHC mode as PIO */
+ ctrl &= ~MX51_ECSPI_CTRL_SMC;
writel(ctrl, spi_imx->base + MX51_ECSPI_CTRL);
@@ -623,7 +623,7 @@ static void mx51_setup_wml(struct spi_imx_data *spi_imx)
* and enable DMA request.
*/
writel(MX51_ECSPI_DMA_RX_WML(spi_imx->wml - 1) |
- MX51_ECSPI_DMA_TX_WML(spi_imx->wml) |
+ MX51_ECSPI_DMA_TX_WML(0) |
MX51_ECSPI_DMA_RXT_WML(spi_imx->wml) |
MX51_ECSPI_DMA_TEDEN | MX51_ECSPI_DMA_RXDEN |
MX51_ECSPI_DMA_RXTDEN, spi_imx->base + MX51_ECSPI_DMA);
@@ -1273,10 +1273,6 @@ static int spi_imx_sdma_init(struct device *dev, struct spi_imx_data *spi_imx,
{
int ret;
- /* use pio mode for i.mx6dl chip TKT238285 */
- if (of_machine_is_compatible("fsl,imx6dl"))
- return 0;
-
spi_imx->wml = spi_imx->devtype_data->fifo_size / 2;
/* Prepare for TX DMA: */
--
2.7.4
ERR009165 fixed on i.mx6ul/6ull/6sll. All other i.mx6/7 and
i.mx8m/8mm still need this errata. Please refer to nxp official
errata document from https://www.nxp.com/ .
For removing workaround on those chips. Add new i.mx6ul type.
Signed-off-by: Robin Gong <[email protected]>
Acked-by: Mark Brown <[email protected]>
---
drivers/spi/spi-imx.c | 59 ++++++++++++++++++++++++++++++++++++++++++++++-----
1 file changed, 54 insertions(+), 5 deletions(-)
diff --git a/drivers/spi/spi-imx.c b/drivers/spi/spi-imx.c
index d4fe5dd..437ff1e 100644
--- a/drivers/spi/spi-imx.c
+++ b/drivers/spi/spi-imx.c
@@ -57,6 +57,7 @@ enum spi_imx_devtype {
IMX35_CSPI, /* CSPI on all i.mx except above */
IMX51_ECSPI, /* ECSPI on i.mx51 */
IMX53_ECSPI, /* ECSPI on i.mx53 and later */
+ IMX6UL_ECSPI, /* ERR009165 fix from i.mx6ul */
};
struct spi_imx_data;
@@ -76,6 +77,11 @@ struct spi_imx_devtype_data {
bool has_slavemode;
unsigned int fifo_size;
bool dynamic_burst;
+ /*
+ * ERR009165 fixed or not:
+ * https://www.nxp.com/docs/en/errata/IMX6DQCE.pdf
+ */
+ bool tx_glitch_fixed;
enum spi_imx_devtype devtype;
};
@@ -132,6 +138,11 @@ static inline int is_imx51_ecspi(struct spi_imx_data *d)
return d->devtype_data->devtype == IMX51_ECSPI;
}
+static inline int is_imx6ul_ecspi(struct spi_imx_data *d)
+{
+ return d->devtype_data->devtype == IMX6UL_ECSPI;
+}
+
static inline int is_imx53_ecspi(struct spi_imx_data *d)
{
return d->devtype_data->devtype == IMX53_ECSPI;
@@ -591,8 +602,14 @@ static int mx51_ecspi_prepare_transfer(struct spi_imx_data *spi_imx,
ctrl |= mx51_ecspi_clkdiv(spi_imx, t->speed_hz, &clk);
spi_imx->spi_bus_clk = clk;
- /* ERR009165: work in XHC mode as PIO */
- ctrl &= ~MX51_ECSPI_CTRL_SMC;
+ /*
+ * ERR009165: work in XHC mode instead of SMC as PIO on the chips
+ * before i.mx6ul.
+ */
+ if (spi_imx->usedma && spi_imx->devtype_data->tx_glitch_fixed)
+ ctrl |= MX51_ECSPI_CTRL_SMC;
+ else
+ ctrl &= ~MX51_ECSPI_CTRL_SMC;
writel(ctrl, spi_imx->base + MX51_ECSPI_CTRL);
@@ -618,12 +635,16 @@ static int mx51_ecspi_prepare_transfer(struct spi_imx_data *spi_imx,
static void mx51_setup_wml(struct spi_imx_data *spi_imx)
{
+ u32 tx_wml = 0;
+
+ if (spi_imx->devtype_data->tx_glitch_fixed)
+ tx_wml = spi_imx->wml;
/*
* Configure the DMA register: setup the watermark
* and enable DMA request.
*/
writel(MX51_ECSPI_DMA_RX_WML(spi_imx->wml - 1) |
- MX51_ECSPI_DMA_TX_WML(0) |
+ MX51_ECSPI_DMA_TX_WML(tx_wml) |
MX51_ECSPI_DMA_RXT_WML(spi_imx->wml) |
MX51_ECSPI_DMA_TEDEN | MX51_ECSPI_DMA_RXDEN |
MX51_ECSPI_DMA_RXTDEN, spi_imx->base + MX51_ECSPI_DMA);
@@ -1017,6 +1038,23 @@ static struct spi_imx_devtype_data imx53_ecspi_devtype_data = {
.devtype = IMX53_ECSPI,
};
+static struct spi_imx_devtype_data imx6ul_ecspi_devtype_data = {
+ .intctrl = mx51_ecspi_intctrl,
+ .prepare_message = mx51_ecspi_prepare_message,
+ .prepare_transfer = mx51_ecspi_prepare_transfer,
+ .trigger = mx51_ecspi_trigger,
+ .rx_available = mx51_ecspi_rx_available,
+ .reset = mx51_ecspi_reset,
+ .setup_wml = mx51_setup_wml,
+ .fifo_size = 64,
+ .has_dmamode = true,
+ .dynamic_burst = true,
+ .has_slavemode = true,
+ .tx_glitch_fixed = true,
+ .disable = mx51_ecspi_disable,
+ .devtype = IMX6UL_ECSPI,
+};
+
static const struct platform_device_id spi_imx_devtype[] = {
{
.name = "imx1-cspi",
@@ -1040,6 +1078,9 @@ static const struct platform_device_id spi_imx_devtype[] = {
.name = "imx53-ecspi",
.driver_data = (kernel_ulong_t) &imx53_ecspi_devtype_data,
}, {
+ .name = "imx6ul-ecspi",
+ .driver_data = (kernel_ulong_t) &imx6ul_ecspi_devtype_data,
+ }, {
/* sentinel */
}
};
@@ -1052,6 +1093,7 @@ static const struct of_device_id spi_imx_dt_ids[] = {
{ .compatible = "fsl,imx35-cspi", .data = &imx35_cspi_devtype_data, },
{ .compatible = "fsl,imx51-ecspi", .data = &imx51_ecspi_devtype_data, },
{ .compatible = "fsl,imx53-ecspi", .data = &imx53_ecspi_devtype_data, },
+ { .compatible = "fsl,imx6ul-ecspi", .data = &imx6ul_ecspi_devtype_data, },
{ /* sentinel */ }
};
MODULE_DEVICE_TABLE(of, spi_imx_dt_ids);
@@ -1179,7 +1221,14 @@ static int spi_imx_dma_configure(struct spi_master *master)
tx.direction = DMA_MEM_TO_DEV;
tx.dst_addr = spi_imx->base_phys + MXC_CSPITXDATA;
tx.dst_addr_width = buswidth;
- tx.dst_maxburst = spi_imx->wml;
+ /*
+ * For ERR009165 with TX_THRESHOLD=0 could enlarge burst size to fifo
+ * size to speed up fifo filling as possible.
+ */
+ if (spi_imx->devtype_data->tx_glitch_fixed)
+ tx.dst_maxburst = spi_imx->wml;
+ else
+ tx.dst_maxburst = spi_imx->devtype_data->fifo_size;
ret = dmaengine_slave_config(master->dma_tx, &tx);
if (ret) {
dev_err(spi_imx->dev, "TX dma configuration failed with %d\n", ret);
@@ -1690,7 +1739,7 @@ static int spi_imx_probe(struct platform_device *pdev)
spi_imx->bitbang.master->mode_bits = SPI_CPOL | SPI_CPHA | SPI_CS_HIGH \
| SPI_NO_CS;
if (is_imx35_cspi(spi_imx) || is_imx51_ecspi(spi_imx) ||
- is_imx53_ecspi(spi_imx))
+ is_imx53_ecspi(spi_imx) || is_imx6ul_ecspi(spi_imx))
spi_imx->bitbang.master->mode_bits |= SPI_LOOP | SPI_READY;
spi_imx->spi_drctl = spi_drctl;
--
2.7.4
ECSPI issue fixed from i.mx6ul at hardware level, no need
ERR009165 anymore on those chips such as i.mx8mq.
Signed-off-by: Robin Gong <[email protected]>
Acked-by: Vinod Koul <[email protected]>
---
drivers/dma/imx-sdma.c | 29 ++++++++++++++++++++++++++++-
1 file changed, 28 insertions(+), 1 deletion(-)
diff --git a/drivers/dma/imx-sdma.c b/drivers/dma/imx-sdma.c
index db4132f..be86ae8 100644
--- a/drivers/dma/imx-sdma.c
+++ b/drivers/dma/imx-sdma.c
@@ -419,6 +419,13 @@ struct sdma_driver_data {
int num_events;
struct sdma_script_start_addrs *script_addrs;
bool check_ratio;
+ /*
+ * ecspi ERR009165 fixed should be done in sdma script
+ * and it has been fixed in soc from i.mx6ul.
+ * please get more information from the below link:
+ * https://www.nxp.com/docs/en/errata/IMX6DQCE.pdf
+ */
+ bool ecspi_fixed;
};
struct sdma_engine {
@@ -539,6 +546,13 @@ static struct sdma_driver_data sdma_imx6q = {
.script_addrs = &sdma_script_imx6q,
};
+static struct sdma_driver_data sdma_imx6ul = {
+ .chnenbl0 = SDMA_CHNENBL0_IMX35,
+ .num_events = 48,
+ .script_addrs = &sdma_script_imx6q,
+ .ecspi_fixed = true,
+};
+
static struct sdma_script_start_addrs sdma_script_imx7d = {
.ap_2_ap_addr = 644,
.uart_2_mcu_addr = 819,
@@ -587,6 +601,9 @@ static const struct platform_device_id sdma_devtypes[] = {
.name = "imx7d-sdma",
.driver_data = (unsigned long)&sdma_imx7d,
}, {
+ .name = "imx6ul-sdma",
+ .driver_data = (unsigned long)&sdma_imx6ul,
+ }, {
.name = "imx8mq-sdma",
.driver_data = (unsigned long)&sdma_imx8mq,
}, {
@@ -603,6 +620,7 @@ static const struct of_device_id sdma_dt_ids[] = {
{ .compatible = "fsl,imx31-sdma", .data = &sdma_imx31, },
{ .compatible = "fsl,imx25-sdma", .data = &sdma_imx25, },
{ .compatible = "fsl,imx7d-sdma", .data = &sdma_imx7d, },
+ { .compatible = "fsl,imx6ul-sdma", .data = &sdma_imx6ul, },
{ .compatible = "fsl,imx8mq-sdma", .data = &sdma_imx8mq, },
{ /* sentinel */ }
};
@@ -1169,8 +1187,17 @@ static int sdma_config_channel(struct dma_chan *chan)
if (sdmac->peripheral_type == IMX_DMATYPE_ASRC_SP ||
sdmac->peripheral_type == IMX_DMATYPE_ASRC)
sdma_set_watermarklevel_for_p2p(sdmac);
- } else
+ } else {
+ /*
+ * ERR009165 fixed from i.mx6ul, no errata need,
+ * set bit31 to let sdma script skip the errata.
+ */
+ if (sdmac->peripheral_type == IMX_DMATYPE_CSPI &&
+ sdmac->direction == DMA_MEM_TO_DEV &&
+ sdmac->sdma->drvdata->ecspi_fixed)
+ __set_bit(31, &sdmac->watermark_level);
__set_bit(sdmac->event_id0, sdmac->event_mask);
+ }
/* Address */
sdmac->shp_addr = sdmac->per_address;
--
2.7.4
Add i.mx6ul compatible name in binding doc.
Signed-off-by: Robin Gong <[email protected]>
Reviewed-by: Rob Herring <[email protected]>
---
Documentation/devicetree/bindings/dma/fsl-imx-sdma.txt | 1 +
1 file changed, 1 insertion(+)
diff --git a/Documentation/devicetree/bindings/dma/fsl-imx-sdma.txt b/Documentation/devicetree/bindings/dma/fsl-imx-sdma.txt
index c9e9740..12c316f 100644
--- a/Documentation/devicetree/bindings/dma/fsl-imx-sdma.txt
+++ b/Documentation/devicetree/bindings/dma/fsl-imx-sdma.txt
@@ -9,6 +9,7 @@ Required properties:
"fsl,imx53-sdma"
"fsl,imx6q-sdma"
"fsl,imx7d-sdma"
+ "fsl,imx6ul-sdma"
"fsl,imx8mq-sdma"
"fsl,imx8mm-sdma"
"fsl,imx8mn-sdma"
--
2.7.4
Because the number of ecspi1 rx event on i.mx8mm is 0, the condition
check ignore such special case without dma channel enabled, which caused
ecspi1 rx works failed. Actually, no need to check event_id0/event_id1
and replace checking 'event_id1' with 'DMA_DEV_TO_DEV', so that configure
event_id1 only in case DEV_TO_DEV.
Signed-off-by: Robin Gong <[email protected]>
Acked-by: Vinod Koul <[email protected]>
---
drivers/dma/imx-sdma.c | 18 ++++++++----------
1 file changed, 8 insertions(+), 10 deletions(-)
diff --git a/drivers/dma/imx-sdma.c b/drivers/dma/imx-sdma.c
index be86ae8..320104f 100644
--- a/drivers/dma/imx-sdma.c
+++ b/drivers/dma/imx-sdma.c
@@ -1183,7 +1183,7 @@ static int sdma_config_channel(struct dma_chan *chan)
if ((sdmac->peripheral_type != IMX_DMATYPE_MEMORY) &&
(sdmac->peripheral_type != IMX_DMATYPE_DSP)) {
/* Handle multiple event channels differently */
- if (sdmac->event_id1) {
+ if (sdmac->direction == DMA_DEV_TO_DEV) {
if (sdmac->peripheral_type == IMX_DMATYPE_ASRC_SP ||
sdmac->peripheral_type == IMX_DMATYPE_ASRC)
sdma_set_watermarklevel_for_p2p(sdmac);
@@ -1351,9 +1351,9 @@ static void sdma_free_chan_resources(struct dma_chan *chan)
sdma_channel_synchronize(chan);
- if (sdmac->event_id0 >= 0)
- sdma_event_disable(sdmac, sdmac->event_id0);
- if (sdmac->event_id1)
+ sdma_event_disable(sdmac, sdmac->event_id0);
+
+ if (sdmac->direction == DMA_DEV_TO_DEV)
sdma_event_disable(sdmac, sdmac->event_id1);
sdmac->event_id0 = 0;
@@ -1651,13 +1651,11 @@ static int sdma_config(struct dma_chan *chan,
memcpy(&sdmac->slave_config, dmaengine_cfg, sizeof(*dmaengine_cfg));
/* Set ENBLn earlier to make sure dma request triggered after that */
- if (sdmac->event_id0 >= 0) {
- if (sdmac->event_id0 >= sdmac->sdma->drvdata->num_events)
- return -EINVAL;
- sdma_event_enable(sdmac, sdmac->event_id0);
- }
+ if (sdmac->event_id0 >= sdmac->sdma->drvdata->num_events)
+ return -EINVAL;
+ sdma_event_enable(sdmac, sdmac->event_id0);
- if (sdmac->event_id1) {
+ if (sdmac->direction == DMA_DEV_TO_DEV) {
if (sdmac->event_id1 >= sdmac->sdma->drvdata->num_events)
return -EINVAL;
sdma_event_enable(sdmac, sdmac->event_id1);
--
2.7.4
There are two ways for SDMA accessing SPBA devices: one is SDMA->AIPS
->SPBA(masterA port), another is SDMA->SPBA(masterC port). Please refer
to the 'Figure 58-1. i.MX 6Dual/6Quad SPBA connectivity' of i.mx6DQ
Reference Manual. SDMA provide the corresponding app_2_mcu/mcu_2_app and
shp_2_mcu/mcu_2_shp script for such two options. So both AIPS and SPBA
scripts should keep the same behaviour, the issue only caught in AIPS
script sounds not solide.
The issue is more likely as the ecspi errata
ERR009165(http://www.nxp.com/docs/en/errata/IMX6DQCE.pdf):
eCSPI: TXFIFO empty flag glitch can cause the current FIFO transfer to
be sent twice
So revert commit 'df07101e1c4a' firstly.
Signed-off-by: Robin Gong <[email protected]>
Acked-by: Sascha Hauer <[email protected]>
---
arch/arm/boot/dts/imx6q.dtsi | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/arch/arm/boot/dts/imx6q.dtsi b/arch/arm/boot/dts/imx6q.dtsi
index 78a4d64..afdd9eb 100644
--- a/arch/arm/boot/dts/imx6q.dtsi
+++ b/arch/arm/boot/dts/imx6q.dtsi
@@ -177,7 +177,7 @@
clocks = <&clks IMX6Q_CLK_ECSPI5>,
<&clks IMX6Q_CLK_ECSPI5>;
clock-names = "ipg", "per";
- dmas = <&sdma 11 8 1>, <&sdma 12 8 2>;
+ dmas = <&sdma 11 7 1>, <&sdma 12 7 2>;
dma-names = "rx", "tx";
status = "disabled";
};
--
2.7.4
There are two ways for SDMA accessing SPBA devices: one is SDMA->AIPS
->SPBA(masterA port), another is SDMA->SPBA(masterC port). Please refer
to the 'Figure 58-1. i.MX 6Dual/6Quad SPBA connectivity' of i.mx6DQ
Reference Manual. SDMA provide the corresponding app_2_mcu/mcu_2_app and
shp_2_mcu/mcu_2_shp script for such two options. So both AIPS and SPBA
scripts should keep the same behaviour, the issue only caught in AIPS
script sounds not solide.
The issue is more likely as the ecspi errata
ERR009165(http://www.nxp.com/docs/en/errata/IMX6DQCE.pdf):
eCSPI: TXFIFO empty flag glitch can cause the current FIFO transfer to
be sent twice
So revert commit 'dd4b487b32a3' firstly.
Signed-off-by: Robin Gong <[email protected]>
Acked-by: Sascha Hauer <[email protected]>
---
arch/arm/boot/dts/imx6qdl.dtsi | 8 ++++----
1 file changed, 4 insertions(+), 4 deletions(-)
diff --git a/arch/arm/boot/dts/imx6qdl.dtsi b/arch/arm/boot/dts/imx6qdl.dtsi
index 32114cf..9479a90 100644
--- a/arch/arm/boot/dts/imx6qdl.dtsi
+++ b/arch/arm/boot/dts/imx6qdl.dtsi
@@ -338,7 +338,7 @@
clocks = <&clks IMX6QDL_CLK_ECSPI1>,
<&clks IMX6QDL_CLK_ECSPI1>;
clock-names = "ipg", "per";
- dmas = <&sdma 3 8 1>, <&sdma 4 8 2>;
+ dmas = <&sdma 3 7 1>, <&sdma 4 7 2>;
dma-names = "rx", "tx";
status = "disabled";
};
@@ -352,7 +352,7 @@
clocks = <&clks IMX6QDL_CLK_ECSPI2>,
<&clks IMX6QDL_CLK_ECSPI2>;
clock-names = "ipg", "per";
- dmas = <&sdma 5 8 1>, <&sdma 6 8 2>;
+ dmas = <&sdma 5 7 1>, <&sdma 6 7 2>;
dma-names = "rx", "tx";
status = "disabled";
};
@@ -366,7 +366,7 @@
clocks = <&clks IMX6QDL_CLK_ECSPI3>,
<&clks IMX6QDL_CLK_ECSPI3>;
clock-names = "ipg", "per";
- dmas = <&sdma 7 8 1>, <&sdma 8 8 2>;
+ dmas = <&sdma 7 7 1>, <&sdma 8 7 2>;
dma-names = "rx", "tx";
status = "disabled";
};
@@ -380,7 +380,7 @@
clocks = <&clks IMX6QDL_CLK_ECSPI4>,
<&clks IMX6QDL_CLK_ECSPI4>;
clock-names = "ipg", "per";
- dmas = <&sdma 9 8 1>, <&sdma 10 8 2>;
+ dmas = <&sdma 9 7 1>, <&sdma 10 7 2>;
dma-names = "rx", "tx";
status = "disabled";
};
--
2.7.4
ERR009165 fixed from i.mx6ul, add its compatible name in binding doc.
Signed-off-by: Robin Gong <[email protected]>
Acked-by: Mark Brown <[email protected]>
Reviewed-by: Rob Herring <[email protected]>
---
Documentation/devicetree/bindings/spi/fsl-imx-cspi.txt | 1 +
1 file changed, 1 insertion(+)
diff --git a/Documentation/devicetree/bindings/spi/fsl-imx-cspi.txt b/Documentation/devicetree/bindings/spi/fsl-imx-cspi.txt
index 33bc58f..0a529ba 100644
--- a/Documentation/devicetree/bindings/spi/fsl-imx-cspi.txt
+++ b/Documentation/devicetree/bindings/spi/fsl-imx-cspi.txt
@@ -10,6 +10,7 @@ Required properties:
- "fsl,imx35-cspi" for SPI compatible with the one integrated on i.MX35
- "fsl,imx51-ecspi" for SPI compatible with the one integrated on i.MX51
- "fsl,imx53-ecspi" for SPI compatible with the one integrated on i.MX53 and later Soc
+ - "fsl,imx6ul-ecspi" for SPI compatible with the one integrated on i.MX6UL and later Soc
- "fsl,imx8mq-ecspi" for SPI compatible with the one integrated on i.MX8MQ
- "fsl,imx8mm-ecspi" for SPI compatible with the one integrated on i.MX8MM
- "fsl,imx8mn-ecspi" for SPI compatible with the one integrated on i.MX8MN
--
2.7.4
For the compatibility of NXP internal legacy kernel before 4.19 which
is based on uart ram script and upstreaming kernel based on uart rom
script, add both uart ram/rom script in latest sdma firmware. By default
uart rom script used.
Besides, add two multi-fifo scripts for SAI/PDM on i.mx8m/8mm and add
back qspi script miss for v4(i.mx7d/8m/8mm family, but v3 is for i.mx6).
rom script:
uart_2_mcu_addr
uartsh_2_mcu_addr /* through spba bus */
am script:
uart_2_mcu_ram_addr
uartsh_2_mcu_ram_addr /* through spba bus */
Please get latest sdma firmware from the below and put them into the path
(/lib/firmware/imx/sdma/):
https://git.kernel.org/pub/scm/linux/kernel/git/firmware/linux-firmware.git
/tree/imx/sdma
Signed-off-by: Robin Gong <[email protected]>
Acked-by: Vinod Koul <[email protected]>
---
drivers/dma/imx-sdma.c | 4 ++--
include/linux/platform_data/dma-imx-sdma.h | 8 ++++++--
2 files changed, 8 insertions(+), 4 deletions(-)
diff --git a/drivers/dma/imx-sdma.c b/drivers/dma/imx-sdma.c
index 320104f..2ca7935 100644
--- a/drivers/dma/imx-sdma.c
+++ b/drivers/dma/imx-sdma.c
@@ -1718,8 +1718,8 @@ static void sdma_issue_pending(struct dma_chan *chan)
#define SDMA_SCRIPT_ADDRS_ARRAY_SIZE_V1 34
#define SDMA_SCRIPT_ADDRS_ARRAY_SIZE_V2 38
-#define SDMA_SCRIPT_ADDRS_ARRAY_SIZE_V3 41
-#define SDMA_SCRIPT_ADDRS_ARRAY_SIZE_V4 42
+#define SDMA_SCRIPT_ADDRS_ARRAY_SIZE_V3 45
+#define SDMA_SCRIPT_ADDRS_ARRAY_SIZE_V4 46
static void sdma_add_scripts(struct sdma_engine *sdma,
const struct sdma_script_start_addrs *addr)
diff --git a/include/linux/platform_data/dma-imx-sdma.h b/include/linux/platform_data/dma-imx-sdma.h
index 30e676b..e12d2e8 100644
--- a/include/linux/platform_data/dma-imx-sdma.h
+++ b/include/linux/platform_data/dma-imx-sdma.h
@@ -20,12 +20,12 @@ struct sdma_script_start_addrs {
s32 per_2_firi_addr;
s32 mcu_2_firi_addr;
s32 uart_2_per_addr;
- s32 uart_2_mcu_addr;
+ s32 uart_2_mcu_ram_addr;
s32 per_2_app_addr;
s32 mcu_2_app_addr;
s32 per_2_per_addr;
s32 uartsh_2_per_addr;
- s32 uartsh_2_mcu_addr;
+ s32 uartsh_2_mcu_ram_addr;
s32 per_2_shp_addr;
s32 mcu_2_shp_addr;
s32 ata_2_mcu_addr;
@@ -52,6 +52,10 @@ struct sdma_script_start_addrs {
s32 zcanfd_2_mcu_addr;
s32 zqspi_2_mcu_addr;
s32 mcu_2_ecspi_addr;
+ s32 mcu_2_sai_addr;
+ s32 sai_2_mcu_addr;
+ s32 uart_2_mcu_addr;
+ s32 uartsh_2_mcu_addr;
/* End of v3 array */
s32 mcu_2_zqspi_addr;
/* End of v4 array */
--
2.7.4
On Sun, 2020-06-07 at 07:21 +0800, Robin Gong wrote:
> There is ecspi ERR009165 on i.mx6/7 soc family, which cause FIFO
> transfer to be send twice in DMA mode. Please get more information
> from:
> https://www.nxp.com/docs/en/errata/IMX6DQCE.pdf. The workaround is
> adding
> new sdma ram script which works in XCH mode as PIO inside sdma
> instead
> of SMC mode, meanwhile, 'TX_THRESHOLD' should be 0. The issue should
> be
> exist on all legacy i.mx6/7 soc family before i.mx6ul.
> NXP fix this design issue from i.mx6ul, so newer chips including
> i.mx6ul/
> 6ull/6sll do not need this workaroud anymore. All other i.mx6/7/8
> chips
> still need this workaroud. This patch set add new 'fsl,imx6ul-ecspi'
> for ecspi driver and 'ecspi_fixed' in sdma driver to choose if need
> errata
> or not.
> The first two reverted patches should be the same issue, though, it
> seems 'fixed' by changing to other shp script. Hope Sean or Sascha
> could
> have the chance to test this patch set if could fix their issues.
> Besides, enable sdma support for i.mx8mm/8mq and fix ecspi1 not work
> on i.mx8mm because the event id is zero.
Tested-by: Matthias Schiffer <[email protected]>
>
> PS:
> Please get sdma firmware from below linux-firmware and copy it to
> your
> local rootfs /lib/firmware/imx/sdma.
>
https://git.kernel.org/pub/scm/linux/kernel/git/firmware/linux-firmware.git/tree/imx/sdma
>
> v2:
> 1.Add commit log for reverted patches.
> 2.Add comment for 'ecspi_fixed' in sdma driver.
> 3.Add 'fsl,imx6sll-ecspi' compatible instead of 'fsl,imx6ul-ecspi'
> rather than remove.
> v3:
> 1.Confirm with design team make sure ERR009165 fixed on
> i.mx6ul/i.mx6ull
> /i.mx6sll, not fixed on i.mx8m/8mm and other i.mx6/7 legacy
> chips.
> Correct dts related dts patch in v2.
> 2.Clean eratta information in binding doc and new 'tx_glitch_fixed'
> flag
> in spi-imx driver to state ERR009165 fixed or not.
> 3.Enlarge burst size to fifo size for tx since tx_wml set to 0 in
> the
> errata workaroud, thus improve performance as possible.
> v4:
> 1.Add Ack tag from Mark and Vinod
> 2.Remove checking 'event_id1' zero as 'event_id0'.
> v5:
> 1.Add the last patch for compatible with the current uart driver
> which
> using rom script, so both uart ram script and rom script
> supported
> in latest firmware, by default uart rom script used. UART driver
> will be broken without this patch.
> v6:
> 1.Resend after rebase the latest next branch.
> 2.Remove below No.13~No.15 patches of v5 because they were
> mergered.
> ARM: dts: imx6ul: add dma support on ecspi
> ARM: dts: imx6sll: correct sdma compatible
> arm64: defconfig: Enable SDMA on i.mx8mq/8mm
> 3.Revert "dmaengine: imx-sdma: fix context cache" since
> 'context_loaded' removed.
> v7:
> 1.Put the last patch 13/13 'Revert "dmaengine: imx-sdma: fix
> context
> cache"' to the ahead of 03/13 'Revert "dmaengine: imx-sdma:
> refine
> to load context only once" so that no building waring during
> comes out
> during bisect.
> 2.Address Sascha's comments, including eliminating any i.mx6sx in
> this
> series, adding new 'is_imx6ul_ecspi()' instead imx in imx51 and
> taking
> care SMC bit for PIO.
> 3.Add back missing 'Reviewed-by' tag on 08/15(v5):09/13(v7)
> 'spi: imx: add new i.mx6ul compatible name in binding doc'
> v8:
> 1.remove 0003-Revert-dmaengine-imx-sdma-fix-context-cache.patch and
> merge
> it into 04/13 of v7
> 2.add 0005-spi-imx-fallback-to-PIO-if-dma-setup-failure.patch for
> no any
> ecspi function broken even if sdma firmware not updated.
> 3.merge 'tx.dst_maxburst' changes in the two continous patches into
> one
> patch to avoid confusion.
> 4.fix typo 'duplicated'.
> v9:
> 1. add "spi: imx: add dma_sync_sg_for_device after fallback from
> dma"
> to fix the potential issue brought by commit bcd8e7761ec9("spi:
> imx:
> fallback to PIO if dma setup failure") which is the only one
> patch
> of v8 merged. Thanks Matthias for reporting:
>
> https://lore.kernel.org/linux-arm-kernel/[email protected]/
> 2. remove 05/13 of v8 "spi: imx:fallback to PIO if dma setup
> failure"
> since it's been merged.
>
> Robin Gong (13):
> spi: imx: add dma_sync_sg_for_device after fallback from dma
> Revert "ARM: dts: imx6q: Use correct SDMA script for SPI5 core"
> Revert "ARM: dts: imx6: Use correct SDMA script for SPI cores"
> Revert "dmaengine: imx-sdma: refine to load context only once"
> dmaengine: imx-sdma: remove duplicated sdma_load_context
> dmaengine: imx-sdma: add mcu_2_ecspi script
> spi: imx: fix ERR009165
> spi: imx: remove ERR009165 workaround on i.mx6ul
> spi: imx: add new i.mx6ul compatible name in binding doc
> dmaengine: imx-sdma: remove ERR009165 on i.mx6ul
> dma: imx-sdma: add i.mx6ul compatible name
> dmaengine: imx-sdma: fix ecspi1 rx dma not work on i.mx8mm
> dmaengine: imx-sdma: add uart rom script
>
> .../devicetree/bindings/dma/fsl-imx-sdma.txt | 1 +
> .../devicetree/bindings/spi/fsl-imx-cspi.txt | 1 +
> arch/arm/boot/dts/imx6q.dtsi | 2 +-
> arch/arm/boot/dts/imx6qdl.dtsi | 8 +--
> drivers/dma/imx-sdma.c | 67
> ++++++++++++--------
> drivers/spi/spi-imx.c | 73
> +++++++++++++++++++---
> include/linux/platform_data/dma-imx-sdma.h | 8 ++-
> 7 files changed, 120 insertions(+), 40 deletions(-)
>
On Sun, Jun 07, 2020 at 07:21:05AM +0800, Robin Gong wrote:
> In case dma transfer failed and fallback to pio, tx_buf/rx_buf need to be
> taken care cache since they have already been maintained by spi.c
Is this needed as part of this series? This looks like an independent
fix and it seems better to get this in independently.
> Fixes: bcd8e7761ec9("spi: imx: fallback to PIO if dma setup failure")
> Signed-off-by: Robin Gong <[email protected]>
> Reported-by: Matthias Schiffer <[email protected]>
> Link: https://lore.kernel.org/linux-arm-kernel/[email protected]/
The Link is usually to the patch on the list.
> --- a/drivers/spi/spi-imx.c
> +++ b/drivers/spi/spi-imx.c
> @@ -1456,6 +1456,13 @@ static int spi_imx_pio_transfer(struct spi_device *spi,
> return -ETIMEDOUT;
> }
>
> + if (transfer->rx_sg.sgl) {
> + struct device *rx_dev = spi->controller->dma_rx->device->dev;
> +
> + dma_sync_sg_for_device(rx_dev, transfer->rx_sg.sgl,
> + transfer->rx_sg.nents, DMA_TO_DEVICE);
> + }
> +
> return transfer->len;
> }
This is confusing - why are we DMA mapping to the device after doing a
PIO transfer?
On 2020/06/08 22:35 Mark Brown <[email protected]> wrote:
> On Sun, Jun 07, 2020 at 07:21:05AM +0800, Robin Gong wrote:
> > In case dma transfer failed and fallback to pio, tx_buf/rx_buf need to
> > be taken care cache since they have already been maintained by spi.c
>
> Is this needed as part of this series? This looks like an independent fix and it
> seems better to get this in independently.
But that's used to fix one patch [05/13]of the v8 patch set. To be honest, I'm also
not sure how to handle it so that I merged both into first v9....For now, I think you
are right, since 'fallback pio' patch could be independent this series. Will resend in
v10.
>
> > Fixes: bcd8e7761ec9("spi: imx: fallback to PIO if dma setup failure")
> > Signed-off-by: Robin Gong <[email protected]>
> > Reported-by: Matthias Schiffer <[email protected]>
> > Link:
> > https://lore.kernel.org/linux-arm-kernel/5d246dd81607bb6e5cb9af86ad4e5
> > [email protected]/
>
> The Link is usually to the patch on the list.
Okay, will remove it.
>
> > --- a/drivers/spi/spi-imx.c
> > +++ b/drivers/spi/spi-imx.c
> > @@ -1456,6 +1456,13 @@ static int spi_imx_pio_transfer(struct spi_device
> *spi,
> > return -ETIMEDOUT;
> > }
> >
> > + if (transfer->rx_sg.sgl) {
> > + struct device *rx_dev = spi->controller->dma_rx->device->dev;
> > +
> > + dma_sync_sg_for_device(rx_dev, transfer->rx_sg.sgl,
> > + transfer->rx_sg.nents, DMA_TO_DEVICE);
> > + }
> > +
> > return transfer->len;
> > }
>
> This is confusing - why are we DMA mapping to the device after doing a PIO
> transfer?
'transfer->rx_sg.sgl' condition check that's the case fallback PIO after DMA transfer
failed. But the spi core still think the buffer should be in 'device' while spi driver
touch it by PIO(CPU), so sync it back to device to ensure all received data flush to DDR.
On Mon, Jun 08, 2020 at 03:08:45PM +0000, Robin Gong wrote:
> > > + if (transfer->rx_sg.sgl) {
> > > + struct device *rx_dev = spi->controller->dma_rx->device->dev;
> > > +
> > > + dma_sync_sg_for_device(rx_dev, transfer->rx_sg.sgl,
> > > + transfer->rx_sg.nents, DMA_TO_DEVICE);
> > > + }
> > > +
> > This is confusing - why are we DMA mapping to the device after doing a PIO
> > transfer?
> 'transfer->rx_sg.sgl' condition check that's the case fallback PIO after DMA transfer
> failed. But the spi core still think the buffer should be in 'device' while spi driver
> touch it by PIO(CPU), so sync it back to device to ensure all received data flush to DDR.
So we sync it back to the device so that we can then do another sync to
CPU? TBH I'm a bit surprised that there's a requirement that we
explicitly undo a sync and that a redundant double sync in the same
direction might be an issue but I've not had a need to care so I'm
perfectly prepared to believe there is.
At the very least this needs a comment.
On 2020-06-08 16:31, Mark Brown wrote:
> On Mon, Jun 08, 2020 at 03:08:45PM +0000, Robin Gong wrote:
>
>>>> + if (transfer->rx_sg.sgl) {
>>>> + struct device *rx_dev = spi->controller->dma_rx->device->dev;
>>>> +
>>>> + dma_sync_sg_for_device(rx_dev, transfer->rx_sg.sgl,
>>>> + transfer->rx_sg.nents, DMA_TO_DEVICE);
>>>> + }
>>>> +
>
>>> This is confusing - why are we DMA mapping to the device after doing a PIO
>>> transfer?
>
>> 'transfer->rx_sg.sgl' condition check that's the case fallback PIO after DMA transfer
>> failed. But the spi core still think the buffer should be in 'device' while spi driver
>> touch it by PIO(CPU), so sync it back to device to ensure all received data flush to DDR.
>
> So we sync it back to the device so that we can then do another sync to
> CPU? TBH I'm a bit surprised that there's a requirement that we
> explicitly undo a sync and that a redundant double sync in the same
> direction might be an issue but I've not had a need to care so I'm
> perfectly prepared to believe there is.
>
> At the very least this needs a comment.
Yeah, something's off here - at the very least, syncing with
DMA_TO_DEVICE on the Rx buffer that was mapped with DMA_FROM_DEVICE is
clearly wrong. CONFIG_DMA_API_DEBUG should scream about that.
If the device has written to the buffer at all since dma_map_sg() was
called then you do need a dma_sync_sg_for_cpu() call before touching it
from a CPU fallback path, but if nobody's going to touch it from that
point until it's unmapped then there's no point syncing it again. The
my_card_interrupt_handler() example in DMA-API_HOWTO.txt demonstrates this.
Robin.
On 2020/06/08 23:32 Mark Brown <[email protected]> wrote:
> On Mon, Jun 08, 2020 at 03:08:45PM +0000, Robin Gong wrote:
>
> > > > + if (transfer->rx_sg.sgl) {
> > > > + struct device *rx_dev = spi->controller->dma_rx->device->dev;
> > > > +
> > > > + dma_sync_sg_for_device(rx_dev, transfer->rx_sg.sgl,
> > > > + transfer->rx_sg.nents, DMA_TO_DEVICE);
> > > > + }
> > > > +
>
> > > This is confusing - why are we DMA mapping to the device after doing
> > > a PIO transfer?
>
> > 'transfer->rx_sg.sgl' condition check that's the case fallback PIO
> > after DMA transfer failed. But the spi core still think the buffer
> > should be in 'device' while spi driver touch it by PIO(CPU), so sync it back to
> device to ensure all received data flush to DDR.
>
> So we sync it back to the device so that we can then do another sync to CPU?
Yes, spi.c will sync to CPU again in spi_unmap_buf() after transfer done finally.
Otherwise, the fresh received data by CPU in this fallback case may be invalidated
by spi.c, which led to the data corrupt on Matthias's side.
> TBH I'm a bit surprised that there's a requirement that we explicitly undo a
> sync and that a redundant double sync in the same direction might be an issue
Considering DMA transfer may be failed(for example, sdma firmware may not be
updated as ERR009165 depends on), we'd better fallback to PIO to ensure no any
function break here. Thus should clean fresh rx data from cache into external memory
as real 'device' received by DMA. Understood a bit confusing here, but that way could
be avoided by any code changing in spi.c. Or make some code changes in spi.c to cancel
spi_unmap_buf() in such fallback case?
> but I've not had a need to care so I'm perfectly prepared to believe there is.
>
> At the very least this needs a comment.
Okay, I'll add comment here in next.
On 2020/06/09 0:44 Robin Murphy <[email protected]> wrote:
> On 2020-06-08 16:31, Mark Brown wrote:
> > On Mon, Jun 08, 2020 at 03:08:45PM +0000, Robin Gong wrote:
> >
> >>>> + if (transfer->rx_sg.sgl) {
> >>>> + struct device *rx_dev = spi->controller->dma_rx->device->dev;
> >>>> +
> >>>> + dma_sync_sg_for_device(rx_dev, transfer->rx_sg.sgl,
> >>>> + transfer->rx_sg.nents, DMA_TO_DEVICE);
> >>>> + }
> >>>> +
> >
> >>> This is confusing - why are we DMA mapping to the device after doing
> >>> a PIO transfer?
> >
> >> 'transfer->rx_sg.sgl' condition check that's the case fallback PIO
> >> after DMA transfer failed. But the spi core still think the buffer
> >> should be in 'device' while spi driver touch it by PIO(CPU), so sync it back to
> device to ensure all received data flush to DDR.
> >
> > So we sync it back to the device so that we can then do another sync
> > to CPU? TBH I'm a bit surprised that there's a requirement that we
> > explicitly undo a sync and that a redundant double sync in the same
> > direction might be an issue but I've not had a need to care so I'm
> > perfectly prepared to believe there is.
> >
> > At the very least this needs a comment.
>
> Yeah, something's off here - at the very least, syncing with DMA_TO_DEVICE on
> the Rx buffer that was mapped with DMA_FROM_DEVICE is clearly wrong.
> CONFIG_DMA_API_DEBUG should scream about that.
>
> If the device has written to the buffer at all since dma_map_sg() was called
> then you do need a dma_sync_sg_for_cpu() call before touching it from a CPU
> fallback path, but if nobody's going to touch it from that point until it's
> unmapped then there's no point syncing it again. The
> my_card_interrupt_handler() example in DMA-API_HOWTO.txt demonstrates
> this.
Thanks for you post, but sorry, that's not spi-imx case now, because the rx data in device memory is not truly updated from 'device'/DMA, but from PIO, so that dma_sync_sg_for_cpu with DMA_FROM_DEVICE can't be used, otherwise the fresh data in cache will be invalidated.
But you're right, kernel warning comes out if CONFIG_DMA_API_DEBUG enabled...
On 2020/06/09 Robin Murphy <[email protected]> wrote:
> On 2020-06-09 06:21, Robin Gong wrote:
> > On 2020/06/09 0:44 Robin Murphy <[email protected]> wrote:
> >> On 2020-06-08 16:31, Mark Brown wrote:
> >>> On Mon, Jun 08, 2020 at 03:08:45PM +0000, Robin Gong wrote:
> >>>
> >>>>>> + if (transfer->rx_sg.sgl) {
> >>>>>> + struct device *rx_dev = spi->controller->dma_rx->device->dev;
> >>>>>> +
> >>>>>> + dma_sync_sg_for_device(rx_dev, transfer->rx_sg.sgl,
> >>>>>> + transfer->rx_sg.nents, DMA_TO_DEVICE);
> >>>>>> + }
> >>>>>> +
> >>>
> >>>>> This is confusing - why are we DMA mapping to the device after
> >>>>> doing a PIO transfer?
> >>>
> >>>> 'transfer->rx_sg.sgl' condition check that's the case fallback PIO
> >>>> after DMA transfer failed. But the spi core still think the buffer
> >>>> should be in 'device' while spi driver touch it by PIO(CPU), so
> >>>> sync it back to
> >> device to ensure all received data flush to DDR.
> >>>
> >>> So we sync it back to the device so that we can then do another sync
> >>> to CPU? TBH I'm a bit surprised that there's a requirement that we
> >>> explicitly undo a sync and that a redundant double sync in the same
> >>> direction might be an issue but I've not had a need to care so I'm
> >>> perfectly prepared to believe there is.
> >>>
> >>> At the very least this needs a comment.
> >>
> >> Yeah, something's off here - at the very least, syncing with
> >> DMA_TO_DEVICE on the Rx buffer that was mapped with
> DMA_FROM_DEVICE is clearly wrong.
> >> CONFIG_DMA_API_DEBUG should scream about that.
> >>
> >> If the device has written to the buffer at all since dma_map_sg() was
> >> called then you do need a dma_sync_sg_for_cpu() call before touching
> >> it from a CPU fallback path, but if nobody's going to touch it from
> >> that point until it's unmapped then there's no point syncing it
> >> again. The
> >> my_card_interrupt_handler() example in DMA-API_HOWTO.txt
> demonstrates
> >> this.
> > Thanks for you post, but sorry, that's not spi-imx case now, because the rx
> data in device memory is not truly updated from 'device'/DMA, but from PIO,
> so that dma_sync_sg_for_cpu with DMA_FROM_DEVICE can't be used,
> otherwise the fresh data in cache will be invalidated.
> > But you're right, kernel warning comes out if CONFIG_DMA_API_DEBUG
> enabled...
>
> Ah, I think I understand what's going on now. That's... really ugly :(
Yeah...The only reason is to avoid touch any spi core code...I'm trying to implement fallback at spi core so that can spi_unmap_buf directly if dma transfer error and no need
such dma_sync_* in spi client driver. Not sure if Mark could accept it. Thanks for your below great thoughts :)
>
> Looking at the SPI core code, I think a better way to handle this would be to
> have your fallback path call spi_unmap_buf() directly (or perform the same
> actions, if exporting that to drivers is unacceptable), then make sure
> ->can_dma() returns false after that such that spi_unmap_msg() won't try to
> unmap it again. That's a lot more reasonable than trying to fake up a
> DMA_TO_DEVICE transfer in the middle of a DMA_FROM_DEVICE operation on
> the same buffer.
>
> Alternatively, is it feasible to initiate a dummy DMA request during probe, such
> that you can detect the failure condition and give up on the DMA channel early,
> and not have to deal with it during a real SPI transfer?
>
> Robin.
On 2020-06-09 06:21, Robin Gong wrote:
> On 2020/06/09 0:44 Robin Murphy <[email protected]> wrote:
>> On 2020-06-08 16:31, Mark Brown wrote:
>>> On Mon, Jun 08, 2020 at 03:08:45PM +0000, Robin Gong wrote:
>>>
>>>>>> + if (transfer->rx_sg.sgl) {
>>>>>> + struct device *rx_dev = spi->controller->dma_rx->device->dev;
>>>>>> +
>>>>>> + dma_sync_sg_for_device(rx_dev, transfer->rx_sg.sgl,
>>>>>> + transfer->rx_sg.nents, DMA_TO_DEVICE);
>>>>>> + }
>>>>>> +
>>>
>>>>> This is confusing - why are we DMA mapping to the device after doing
>>>>> a PIO transfer?
>>>
>>>> 'transfer->rx_sg.sgl' condition check that's the case fallback PIO
>>>> after DMA transfer failed. But the spi core still think the buffer
>>>> should be in 'device' while spi driver touch it by PIO(CPU), so sync it back to
>> device to ensure all received data flush to DDR.
>>>
>>> So we sync it back to the device so that we can then do another sync
>>> to CPU? TBH I'm a bit surprised that there's a requirement that we
>>> explicitly undo a sync and that a redundant double sync in the same
>>> direction might be an issue but I've not had a need to care so I'm
>>> perfectly prepared to believe there is.
>>>
>>> At the very least this needs a comment.
>>
>> Yeah, something's off here - at the very least, syncing with DMA_TO_DEVICE on
>> the Rx buffer that was mapped with DMA_FROM_DEVICE is clearly wrong.
>> CONFIG_DMA_API_DEBUG should scream about that.
>>
>> If the device has written to the buffer at all since dma_map_sg() was called
>> then you do need a dma_sync_sg_for_cpu() call before touching it from a CPU
>> fallback path, but if nobody's going to touch it from that point until it's
>> unmapped then there's no point syncing it again. The
>> my_card_interrupt_handler() example in DMA-API_HOWTO.txt demonstrates
>> this.
> Thanks for you post, but sorry, that's not spi-imx case now, because the rx data in device memory is not truly updated from 'device'/DMA, but from PIO, so that dma_sync_sg_for_cpu with DMA_FROM_DEVICE can't be used, otherwise the fresh data in cache will be invalidated.
> But you're right, kernel warning comes out if CONFIG_DMA_API_DEBUG enabled...
Ah, I think I understand what's going on now. That's... really ugly :(
Looking at the SPI core code, I think a better way to handle this would
be to have your fallback path call spi_unmap_buf() directly (or perform
the same actions, if exporting that to drivers is unacceptable), then
make sure ->can_dma() returns false after that such that spi_unmap_msg()
won't try to unmap it again. That's a lot more reasonable than trying to
fake up a DMA_TO_DEVICE transfer in the middle of a DMA_FROM_DEVICE
operation on the same buffer.
Alternatively, is it feasible to initiate a dummy DMA request during
probe, such that you can detect the failure condition and give up on the
DMA channel early, and not have to deal with it during a real SPI transfer?
Robin.
On Tue, Jun 09, 2020 at 11:00:33AM +0100, Robin Murphy wrote:
> Ah, I think I understand what's going on now. That's... really ugly :(
> Looking at the SPI core code, I think a better way to handle this would be
> to have your fallback path call spi_unmap_buf() directly (or perform the
> same actions, if exporting that to drivers is unacceptable), then make sure
> ->can_dma() returns false after that such that spi_unmap_msg() won't try to
> unmap it again. That's a lot more reasonable than trying to fake up a
> DMA_TO_DEVICE transfer in the middle of a DMA_FROM_DEVICE operation on the
> same buffer.
Ideally the driver would be checking in can_dma() if the DMA controller
is able to perform transactions rather than letting things run as far as
trying to actually do the transfer, that's a whole lot cleaner and more
manageable than running into an error doing the transfer. I'm surprised
that there's no DMA API way to figure this out TBH.
We'll also need some handling for this changing at runtime, we're not
expecting this to be dynamic at all - we're expecting it to be a static
property of the controller/transfer combination, we didn't contemplate
this varying randomly at runtime. Instead of rechecking can_dma() we
ought to have a flag saying if we did the mapping (which the bodge Robin
suggests above could clear).