From: Tejas Bhumkar tejas.arvind.bhumkar@amd.com
To reduce the CPU load in waiting for the OSPI internal SRAM to clear in indirect mode, it's better to use the CQSPI_REG_IRQSTATUS register to check for indirect operation to complete. Enabled interrupt for Indirect Complete and Transfer Watermark Breach interrupt status register bits and using readl_poll_timeout function to poll for Indirect Operation Complete bit gets set.
Here not enabling IRQ coming to GIC, only IRQ from IP itself is able to poll bits.
It is observed that the Indirect Operation Complete bit is getting set at an average time of 0.172 usec.
Signed-off-by: Tejas Bhumkar tejas.arvind.bhumkar@amd.com Link: https://lore.kernel.org/r/9cd21aabccc87850290009a4feb0db0fbcb46ace.170185366... State: waiting --- drivers/spi/cadence_qspi.h | 11 +++++++++++ drivers/spi/cadence_qspi_apb.c | 24 +++++++++++++++++++----- 2 files changed, 30 insertions(+), 5 deletions(-)
diff --git a/drivers/spi/cadence_qspi.h b/drivers/spi/cadence_qspi.h index 693474a287..a48d77a154 100644 --- a/drivers/spi/cadence_qspi.h +++ b/drivers/spi/cadence_qspi.h @@ -201,6 +201,17 @@ (((readl((reg_base) + CQSPI_REG_SDRAMLEVEL)) >> \ CQSPI_REG_SDRAMLEVEL_WR_LSB) & CQSPI_REG_SDRAMLEVEL_WR_MASK)
+/* Interrupt status bits */ +#define CQSPI_REG_IRQ_UNDERFLOW BIT(1) +#define CQSPI_REG_IRQ_IND_COMP BIT(2) +#define CQSPI_REG_IRQ_WATERMARK BIT(6) + +#define CQSPI_IRQ_MASK_WR (CQSPI_REG_IRQ_IND_COMP | \ + CQSPI_REG_IRQ_WATERMARK | \ + CQSPI_REG_IRQ_UNDERFLOW) + +#define CQSPI_IRQ_STATUS_MASK GENMASK(16, 0) + struct cadence_spi_plat { unsigned int max_hz; void *regbase; diff --git a/drivers/spi/cadence_qspi_apb.c b/drivers/spi/cadence_qspi_apb.c index 93ab2b5635..3c71a5effc 100644 --- a/drivers/spi/cadence_qspi_apb.c +++ b/drivers/spi/cadence_qspi_apb.c @@ -31,6 +31,7 @@ #include <linux/bitops.h> #include <linux/delay.h> #include <linux/errno.h> +#include <linux/iopoll.h> #include <wait_bit.h> #include <spi.h> #include <spi-mem.h> @@ -845,7 +846,7 @@ cadence_qspi_apb_indirect_write_execute(struct cadence_spi_priv *priv, const u8 *bb_txbuf = txbuf; void *bounce_buf = NULL; unsigned int write_bytes; - int ret; + int ret, cr;
/* * Use bounce buffer for non 32 bit aligned txbuf to avoid data @@ -859,9 +860,15 @@ cadence_qspi_apb_indirect_write_execute(struct cadence_spi_priv *priv, bb_txbuf = bounce_buf; }
- /* Configure the indirect read transfer bytes */ + /* Configure the indirect write transfer bytes */ writel(n_tx, priv->regbase + CQSPI_REG_INDIRECTWRBYTES);
+ /* Clear all interrupts */ + writel(CQSPI_IRQ_STATUS_MASK, priv->regbase + CQSPI_REG_IRQSTATUS); + + /* Enable interrupt for corresponding interrupt status register bit's */ + writel(CQSPI_IRQ_MASK_WR, priv->regbase + CQSPI_REG_IRQMASK); + /* Start the indirect write transfer */ writel(CQSPI_REG_INDIRECTWR_START, priv->regbase + CQSPI_REG_INDIRECTWR); @@ -880,9 +887,10 @@ cadence_qspi_apb_indirect_write_execute(struct cadence_spi_priv *priv, bb_txbuf + rounddown(write_bytes, 4), write_bytes % 4);
- ret = wait_for_bit_le32(priv->regbase + CQSPI_REG_SDRAMLEVEL, - CQSPI_REG_SDRAMLEVEL_WR_MASK << - CQSPI_REG_SDRAMLEVEL_WR_LSB, 0, 10, 0); + /* Wait up to Indirect Operation Complete bit to set */ + ret = readl_poll_timeout(priv->regbase + CQSPI_REG_IRQSTATUS, cr, + cr & CQSPI_REG_IRQ_IND_COMP, 10); + if (ret) { printf("Indirect write timed out (%i)\n", ret); goto failwr; @@ -900,6 +908,9 @@ cadence_qspi_apb_indirect_write_execute(struct cadence_spi_priv *priv, goto failwr; }
+ /* Disable interrupt. */ + writel(0, priv->regbase + CQSPI_REG_IRQMASK); + /* Clear indirect completion status */ writel(CQSPI_REG_INDIRECTWR_DONE, priv->regbase + CQSPI_REG_INDIRECTWR); @@ -917,6 +928,9 @@ cadence_qspi_apb_indirect_write_execute(struct cadence_spi_priv *priv, return 0;
failwr: + /* Disable interrupt. */ + writel(0, priv->regbase + CQSPI_REG_IRQMASK); + /* Cancel the indirect write */ writel(CQSPI_REG_INDIRECTWR_CANCEL, priv->regbase + CQSPI_REG_INDIRECTWR);