Here is the second RFC for handling ECC errors flagged in the stop command
after a multiblock transfer. It is still RFC because I could only test it by
inducing ECC errors in software (see patch for TMIO below). Shimoda-san, can
you try this series with the SD tester again? That would be very kind.
Other than that, I hope the patch descriptions and comments explain the single
steps. Looking forward for thoughts.
Kind regards,
Wolfram
Changes since RFC v1:
* rebased to mmc/next as of today
* reworded commit message for patch 1
* added tested-tag from Shimoda-san for patch 1
* added patch 2
Wolfram Sang (2):
mmc: core: check also R1 response for stop commands
mmc: core: for data errors, take response of stop cmd into account
drivers/mmc/core/block.c | 16 +++++++++++++---
1 file changed, 13 insertions(+), 3 deletions(-)
Patch to simulate ECC errors:
diff --git a/drivers/mmc/host/tmio_mmc_pio.c b/drivers/mmc/host/tmio_mmc_pio.c
index a2d92f10501bdd..9773c7e5e4d154 100644
--- a/drivers/mmc/host/tmio_mmc_pio.c
+++ b/drivers/mmc/host/tmio_mmc_pio.c
@@ -553,6 +553,8 @@ void tmio_mmc_do_data_irq(struct tmio_mmc_host *host)
}
if (stop) {
+ static unsigned int induce_cnt = 0;
+
if (stop->opcode != MMC_STOP_TRANSMISSION || stop->arg)
dev_err(&host->pdev->dev, "unsupported stop:
CMD%u,0x%x. We did CMD12,0\n",
stop->opcode, stop->arg);
@@ -560,6 +562,9 @@ void tmio_mmc_do_data_irq(struct tmio_mmc_host *host)
/* fill in response from auto CMD12 */
stop->resp[0] = sd_ctrl_read16_and_16_as_32(host, CTL_RESPONSE);
+ if (induce_cnt++ % 100 == 0)
+ stop->resp[0] |= R1_CARD_ECC_FAILED;
+
sd_ctrl_write16(host, CTL_STOP_INTERNAL_ACTION, 0);
}
--
2.11.0
