Hi Anssi,
On 2017/2/13 20:06, Anssi Hannula wrote:
sdhci_arasan_get_timeout_clock() divides the frequency it has with (1 <<
(13 + divisor)).
However, the divisor is not some Arasan-specific value, but instead is
just the Data Timeout Counter Value from the SDHCI Timeout Control
Register.
The TRM looks really mess, but yes, it's just the same of
SDHCI Timeout Control Register.
As the max_busy_timeout is in ms, so the timeout_clk should be in KHz,
thus the .get_timeout_clock should provide it in KHz.
So the sdhci-cadence.c return it in KHz, but sdhci-bcm-kona return it in Hz!
Now per you changes:
(1) if the unit is KHz, the SDHCI core just take it.
(2) if the unit is MHz, you code return it in Mhz and the SDHCI core
will agin multiple 1000 there..to make it as KHz.
The above looks correct to me , but other variant hosts do it in the
wrong way. Did I miss something, Adrian?
Applying it here like this is wrong as the sdhci driver already takes
that value into account when calculating timeouts, and in fact it *sets*
that register value based on how long a timeout is wanted.
Additionally, sdhci core interprets the .get_timeout_clock callback
return value as if it were read from hardware registers, i.e. the unit
should be kHz or MHz depending on SDHCI_TIMEOUT_CLK_UNIT capability bit.
This bit is set at least on the tested Zynq-7000 SoC.
With the tested hardware (SDHCI_TIMEOUT_CLK_UNIT set) this results in
too high a timeout clock rate being reported, causing the core to use
longer-than-needed timeouts. Additionally, on a partitioned MMC
(therefore having erase_group_def bit set) mmc_calc_max_discard()
disables discard support as it looks like controller does not support
the long timeouts needed for that.
Do not apply the extra divisor and return the timeout clock in the
expected unit.
Tested with a Zynq-7000 SoC and a partitioned Toshiba THGBMAG5A1JBAWR
eMMC card.
Signed-off-by: Anssi Hannula <anssi.hannula@xxxxxxxxxx>
Fixes: e3ec3a3d11ad ("mmc: arasan: Add driver for Arasan SDHCI")
Cc: <stable@xxxxxxxxxxxxxxx>
---
The .get_timeout_clock situation seems rather messy altogether, but I
wasn't sure which direction to take it to and it is probably best to
have that separate from this bugfix, anyway, so I kept the scope
limited.
drivers/mmc/host/sdhci-of-arasan.c | 14 +++++---------
1 file changed, 5 insertions(+), 9 deletions(-)
diff --git a/drivers/mmc/host/sdhci-of-arasan.c b/drivers/mmc/host/sdhci-of-arasan.c
index 410a55b..1cfd7f9 100644
--- a/drivers/mmc/host/sdhci-of-arasan.c
+++ b/drivers/mmc/host/sdhci-of-arasan.c
@@ -28,13 +28,9 @@
#include "sdhci-pltfm.h"
#include <linux/of.h>
-#define SDHCI_ARASAN_CLK_CTRL_OFFSET 0x2c
#define SDHCI_ARASAN_VENDOR_REGISTER 0x78
#define VENDOR_ENHANCED_STROBE BIT(0)
-#define CLK_CTRL_TIMEOUT_SHIFT 16
-#define CLK_CTRL_TIMEOUT_MASK (0xf << CLK_CTRL_TIMEOUT_SHIFT)
-#define CLK_CTRL_TIMEOUT_MIN_EXP 13
#define PHY_CLK_TOO_SLOW_HZ 400000
@@ -163,15 +159,15 @@ static int sdhci_arasan_syscon_write(struct sdhci_host *host,
static unsigned int sdhci_arasan_get_timeout_clock(struct sdhci_host *host)
{
- u32 div;
unsigned long freq;
struct sdhci_pltfm_host *pltfm_host = sdhci_priv(host);
- div = readl(host->ioaddr + SDHCI_ARASAN_CLK_CTRL_OFFSET);
- div = (div & CLK_CTRL_TIMEOUT_MASK) >> CLK_CTRL_TIMEOUT_SHIFT;
+ /* SDHCI timeout clock is in kHz */
+ freq = DIV_ROUND_UP(clk_get_rate(pltfm_host->clk), 1000);
- freq = clk_get_rate(pltfm_host->clk);
- freq /= 1 << (CLK_CTRL_TIMEOUT_MIN_EXP + div);
+ /* or in MHz */
+ if (host->caps & SDHCI_TIMEOUT_CLK_UNIT)
+ freq = DIV_ROUND_UP(freq, 1000);
return freq;
}