Op 13-11-2024 om 13:18 schreef Jonas Rebmann:
Implement support for the word delay feature of i.MX51 (and onwards) via
the ECSPI interface.
Convert the requested delay to SPI cycles and account for an extra
inter-word delay inserted by the controller in addition to the requested
number of cycles, which was observed when testing this patch.
Disable dynamic burst when word delay is set. As the configurable delay
period in the controller is inserted after bursts, the burst length must
equal the word length.
Account for word delay in the transfer time estimation for
polling_limit_us.
Signed-off-by: Jonas Rebmann <jre@xxxxxxxxxxxxxx>
---
drivers/spi/spi-imx.c | 95 +++++++++++++++++++++++++++++++++++++++++++++------
1 file changed, 85 insertions(+), 10 deletions(-)
[...]
+static unsigned int spi_imx_transfer_estimate_time_us(struct spi_transfer *transfer)
+{
+ u64 result;
+
+ result = DIV_U64_ROUND_CLOSEST((u64)USEC_PER_SEC * transfer->len * BITS_PER_BYTE,
+ transfer->effective_speed_hz);
+ if (transfer->word_delay.value) {
+ unsigned int word_delay_us;
+ unsigned int words;
+
+ words = DIV_ROUND_UP(transfer->len * BITS_PER_BYTE, transfer->bits_per_word);
+ word_delay_us = DIV_ROUND_CLOSEST(spi_delay_to_ns(&transfer->word_delay, transfer),
+ NSEC_PER_USEC);
+ result += words * word_delay_us;
If the multiplication can overflow 32 bits to need to force a 64 bits
multiply.
result += (u64)words * word_delay_us;
But I'm wondering if `result` needs to be u64.
+ }
+
+ return min(result, U32_MAX);
Do you really expect this much? You're clipping to U32_MAX.
U32_MAX microsecs is already more than an hour.
+}
[...]
