Cr50: I2CS: Clear IRQ at beginning of ISR

If the host sends a back to back I2CS write followed by a read of the
access register, then the read IRQ can be missed by FW because it was
clearing this bit at the end of the ISR. This would result in a
following write to have the incorrect number of bytes since the
address register byte from the read that wasn't processed gets left in
the HW write fifo.

BRANCH=none
BUG=chrome-os-partner:40397
TEST=manual
The issue was happening at the beginning of depthcharge. Without this
fix, I would see the Cr50 console message:
'data size mismatch for reg 0x0 rx 2, need 1'
After moving the IRQ clear could not reproduce this message. In
addition, the debug I2CS log showed that there was a read transaction
immediately following the write.

Change-Id: I9854dde6880a789e0acb2b1f6a06b43c73a5a2df
Signed-off-by: Scott <scollyer@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/382687
Commit-Ready: Scott Collyer <scollyer@chromium.org>
Tested-by: Scott Collyer <scollyer@chromium.org>
Reviewed-by: Randall Spangler <rspangler@chromium.org>
Reviewed-by: Vadim Bendebury <vbendeb@chromium.org>
This commit is contained in:
Scott
2016-09-08 18:43:10 -07:00
committed by chrome-bot
parent 017606c927
commit c5c14ef348

View File

@@ -119,6 +119,9 @@ DECLARE_HOOK(HOOK_INIT, i2cs_init, HOOK_PRIO_DEFAULT);
/* Process the 'end of a write cycle' interrupt. */
static void _i2cs_write_complete_int(void)
{
/* Reset the IRQ condition. */
GWRITE_FIELD(I2CS, INT_STATE, INTR_WRITE_COMPLETE, 1);
if (write_complete_handler_) {
uint16_t bytes_written;
uint16_t bytes_processed;
@@ -164,9 +167,6 @@ static void _i2cs_write_complete_int(void)
/* Invoke the callback to process the message. */
write_complete_handler_(i2cs_buffer, bytes_processed);
}
/* Reset the IRQ condition. */
GWRITE_FIELD(I2CS, INT_STATE, INTR_WRITE_COMPLETE, 1);
}
DECLARE_IRQ(GC_IRQNUM_I2CS0_INTR_WRITE_COMPLETE_INT,
_i2cs_write_complete_int, 1);