Skip to content

Commit

Permalink
mmc: rtsx_pci: Fix long reads when clock is prescaled
Browse files Browse the repository at this point in the history
[ Upstream commit 3ac5e45 ]

For unexplained reasons, the prescaler register for this device needs to
be cleared (set to 1) while performing a data read or else the command
will hang. This does not appear to affect the real clock rate sent out
on the bus, so I assume it's purely to work around a hardware bug.

During normal operation, the prescaler is already set to 1, so nothing
needs to be done. However, in "initial mode" (which is used for sub-MHz
clock speeds, like the core sets while enumerating cards), it's set to
128 and so we need to reset it during data reads. We currently fail to
do this for long reads.

This has no functional affect on the driver's operation currently
written, as the MMC core always sets a clock above 1MHz before
attempting any long reads. However, the core could conceivably set any
clock speed at any time and the driver should still work, so I think
this fix is worthwhile.

I personally encountered this issue while performing data recovery on an
external chip. My connections had poor signal integrity, so I modified
the core code to reduce the clock speed. Without this change, I saw the
card enumerate but was unable to actually read any data.

Writes don't seem to work in the situation described above even with
this change (and even if the workaround is extended to encompass data
write commands). I was not able to find a way to get them working.

Signed-off-by: Thomas Hebb <tommyhebb@gmail.com>
Link: https://lore.kernel.org/r/2fef280d8409ab0100c26c6ac7050227defd098d.1627818365.git.tommyhebb@gmail.com
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
  • Loading branch information
tchebb authored and gregkh committed Sep 18, 2021
1 parent c0598b6 commit 4d3b34e
Showing 1 changed file with 23 additions and 13 deletions.
36 changes: 23 additions & 13 deletions drivers/mmc/host/rtsx_pci_sdmmc.c
Expand Up @@ -542,9 +542,22 @@ static int sd_write_long_data(struct realtek_pci_sdmmc *host,
return 0;
}

static inline void sd_enable_initial_mode(struct realtek_pci_sdmmc *host)
{
rtsx_pci_write_register(host->pcr, SD_CFG1,
SD_CLK_DIVIDE_MASK, SD_CLK_DIVIDE_128);
}

static inline void sd_disable_initial_mode(struct realtek_pci_sdmmc *host)
{
rtsx_pci_write_register(host->pcr, SD_CFG1,
SD_CLK_DIVIDE_MASK, SD_CLK_DIVIDE_0);
}

static int sd_rw_multi(struct realtek_pci_sdmmc *host, struct mmc_request *mrq)
{
struct mmc_data *data = mrq->data;
int err;

if (host->sg_count < 0) {
data->error = host->sg_count;
Expand All @@ -553,22 +566,19 @@ static int sd_rw_multi(struct realtek_pci_sdmmc *host, struct mmc_request *mrq)
return data->error;
}

if (data->flags & MMC_DATA_READ)
return sd_read_long_data(host, mrq);
if (data->flags & MMC_DATA_READ) {
if (host->initial_mode)
sd_disable_initial_mode(host);

return sd_write_long_data(host, mrq);
}
err = sd_read_long_data(host, mrq);

static inline void sd_enable_initial_mode(struct realtek_pci_sdmmc *host)
{
rtsx_pci_write_register(host->pcr, SD_CFG1,
SD_CLK_DIVIDE_MASK, SD_CLK_DIVIDE_128);
}
if (host->initial_mode)
sd_enable_initial_mode(host);

static inline void sd_disable_initial_mode(struct realtek_pci_sdmmc *host)
{
rtsx_pci_write_register(host->pcr, SD_CFG1,
SD_CLK_DIVIDE_MASK, SD_CLK_DIVIDE_0);
return err;
}

return sd_write_long_data(host, mrq);
}

static void sd_normal_rw(struct realtek_pci_sdmmc *host,
Expand Down

0 comments on commit 4d3b34e

Please sign in to comment.