On Mon, 27 Aug 2018 14:25:44 +0300
Adrian Hunter [off-list ref] wrote:

On 10/08/18 21:08, Aapo Vienamo wrote:

quoted

Add a quirk to disable card clock when the tuning command is sent.

This has to be done to prevent the SDHCI controller from hanging on
Tegra210. Without the quirk enabled there appears to be around 10%
chance that the tuning sequence will fail and time out due to the
controller locking up.

Signed-off-by: Aapo Vienamo <redacted>
---
 drivers/mmc/host/sdhci.c | 15 +++++++++++++++
 drivers/mmc/host/sdhci.h |  2 ++
 2 files changed, 17 insertions(+)

diff --git a/drivers/mmc/host/sdhci.c b/drivers/mmc/host/sdhci.c
index 04dc443..166b16f 100644
--- a/drivers/mmc/host/sdhci.c
+++ b/drivers/mmc/host/sdhci.c

@@ -2175,6 +2175,7 @@ static void sdhci_send_tuning(struct sdhci_host *host, u32 opcode)
 	struct mmc_request mrq = {};
 	unsigned long flags;
 	u32 b = host->sdma_boundary;
+	u16 clk;
 
 	spin_lock_irqsave(&host->lock, flags);
 

@@ -2183,6 +2184,13 @@ static void sdhci_send_tuning(struct sdhci_host *host, u32 opcode)
 	cmd.mrq = &mrq;
 
 	mrq.cmd = &cmd;
+
+	if (host->quirks2 & SDHCI_QUIRK2_TUNE_DIS_CARD_CLK) {
+		clk = sdhci_readw(host, SDHCI_CLOCK_CONTROL);
+		clk &= ~SDHCI_CLOCK_CARD_EN;
+		sdhci_writew(host, clk, SDHCI_CLOCK_CONTROL);  

Rather than using a quirk, could you use the sdhci I/O accessors to disable
the clock before the tuning comment is written, udelay(1), and then enable
it again?

This was the way it was implemented in the downstream kernel. However,
doing it in the IO accessor when a tuning command is sent seems to work
too.

 -Aapo