[PATCH v1 0/2] Improve eMMC performance for the StarFive JHB100 SoC

[PATCH v1 0/2] Improve eMMC performance for the StarFive JHB100 SoC

[Changhuang Liang]

This series implements a generic interface to configure CIT based on
time (ns). Different platforms can improve MMC performance by
configuring their respective CIT. This series has been tested on the
JHB100 SoC.

before:
dd if=/dev/mmcblk0 of=/dev/null bs=1M count=800
838860800 bytes (839 MB, 800 MiB) copied, 9.38511 s, 89.4 MB/s
dd if=/dev/zero of=/dev/mmcblk0 bs=1M count=800
838860800 bytes (839 MB, 800 MiB) copied, 12.2127 s, 68.7 MB/s

after:
dd if=/dev/mmcblk0 of=/dev/null bs=1M count=800
838860800 bytes (839 MB, 800 MiB) copied, 3.19472 s, 263 MB/s
dd if=/dev/zero of=/dev/mmcblk0 bs=1M count=800
838860800 bytes (839 MB, 800 MiB) copied, 10.7587 s, 78.0 MB/s

This series depends on series:
https://lore.kernel.org/all/20260930103258.13375-1-changhuang.liang@starfivetech.com/

Changhuang Liang (2):
  mmc: cqhci: Add support for setting command idle time in nanoseconds
  mmc: sdhci-of-dwcmshc: Set CQHCI command idle time for StarFive JHB100

 drivers/mmc/host/cqhci-core.c       | 60 +++++++++++++++++++++++++++++
 drivers/mmc/host/cqhci.h            |  6 +++
 drivers/mmc/host/sdhci-of-dwcmshc.c | 22 +++++++++++
 3 files changed, 88 insertions(+)

--
2.25.1

[PATCH v1 1/2] mmc: cqhci: Add support for setting command idle time in nanoseconds

[Changhuang Liang]

Add a helper that derives the internal timer clock frequency (ITCF) of
the CQE from the ITCFMUL and ITCFVAL fields of the CQHCI capabilities
register, and export cqhci_set_cmd_idle_time() to let host drivers
program the Send Status Command Idle Timer (CIT) in CQHCI_SSC1 using a
value in nanoseconds.

The CIT defines how long the bus has to be idle, in internal timer clock
cycles, before the CQE issues CMD13 to poll the Queue Status Register.
Its reset value is controller specific and can be far from optimal, so
host drivers may need to tune it for their platform. Doing the
conversion in the core avoids every host driver re-implementing the
ITCFMUL/ITCFVAL decoding.

The requested time is rounded up to whole clock cycles so that the
programmed idle time is never shorter than requested. The helper returns
-EOPNOTSUPP when the controller does not report a usable ITCF and
-ERANGE when the result does not fit the 16-bit CIT field, so callers
can report the failure instead of silently running with the reset
default.

Signed-off-by: Changhuang Liang <changhuang.liang@starfivetech.com>
---
 drivers/mmc/host/cqhci-core.c | 60 +++++++++++++++++++++++++++++++++++
 drivers/mmc/host/cqhci.h      |  6 ++++
 2 files changed, 66 insertions(+)

diff --git a/drivers/mmc/host/cqhci-core.c b/drivers/mmc/host/cqhci-core.c
index a466cd88b043..aff8d6619ea9 100644
--- a/drivers/mmc/host/cqhci-core.c
+++ b/drivers/mmc/host/cqhci-core.c
@@ -12,6 +12,7 @@
 #include <linux/scatterlist.h>
 #include <linux/platform_device.h>
 #include <linux/ktime.h>
+#include <linux/units.h>
 
 #include <linux/mmc/mmc.h>
 #include <linux/mmc/host.h>
@@ -1163,6 +1164,63 @@ struct cqhci_host *cqhci_pltfm_init(struct platform_device *pdev)
 }
 EXPORT_SYMBOL(cqhci_pltfm_init);
 
+static u64 cqhci_itcfmul_to_hz(u32 itcfmul)
+{
+	/* CQHCI_CAP.ITCFMUL: 0 = 1 kHz, 1 = 10 kHz, ..., 4 = 10 MHz, others reserved */
+	if (itcfmul > 4)
+		return 0;
+
+	return HZ_PER_KHZ * int_pow(10, itcfmul);
+}
+
+static void cqhci_itcf_init(struct cqhci_host *cq_host)
+{
+	u32 cap = cqhci_readl(cq_host, CQHCI_CAP);
+	u64 itcf;
+
+	itcf = cqhci_itcfmul_to_hz(CQHCI_ITCFMUL(cap)) * CQHCI_ITCFVAL(cap);
+	if (!itcf || itcf > U32_MAX) {
+		pr_debug("%s: cqhci: internal timer clock frequency not available (CAP 0x%08x)\n",
+			 mmc_hostname(cq_host->mmc), cap);
+		return;
+	}
+
+	cq_host->itcf_hz = itcf;
+}
+
+/**
+ * cqhci_set_cmd_idle_time - program the Send Status Command Idle Timer (CIT)
+ * @cq_host: CQHCI host
+ * @ns: idle time in nanoseconds
+ *
+ * Set the number of internal timer clock cycles the CQE waits on an idle bus
+ * before issuing CMD13 to poll the Queue Status Register. The cycle count is
+ * rounded up so that the resulting idle time is never shorter than @ns.
+ *
+ * Return: 0 on success, -EOPNOTSUPP if the controller does not report its
+ * internal timer clock frequency, or -ERANGE if @ns does not fit the CIT field.
+ */
+int cqhci_set_cmd_idle_time(struct cqhci_host *cq_host, u32 ns)
+{
+	u32 ssc1;
+	u64 cit;
+
+	if (!cq_host->itcf_hz)
+		return -EOPNOTSUPP;
+
+	cit = DIV_ROUND_UP_ULL((u64)cq_host->itcf_hz * ns, NSEC_PER_SEC);
+	if (!cit || cit > FIELD_MAX(CQHCI_SSC1_CIT_MASK))
+		return -ERANGE;
+
+	ssc1 = cqhci_readl(cq_host, CQHCI_SSC1);
+	ssc1 &= ~CQHCI_SSC1_CIT_MASK;
+	ssc1 |= FIELD_PREP(CQHCI_SSC1_CIT_MASK, cit);
+	cqhci_writel(cq_host, ssc1, CQHCI_SSC1);
+
+	return 0;
+}
+EXPORT_SYMBOL(cqhci_set_cmd_idle_time);
+
 static unsigned int cqhci_ver_major(struct cqhci_host *cq_host)
 {
 	return CQHCI_VER_MAJOR(cqhci_readl(cq_host, CQHCI_VER));
@@ -1207,6 +1265,8 @@ int cqhci_init(struct cqhci_host *cq_host, struct mmc_host *mmc,
 		goto out_err;
 	}
 
+	cqhci_itcf_init(cq_host);
+
 	spin_lock_init(&cq_host->lock);
 
 	init_completion(&cq_host->halt_comp);
diff --git a/drivers/mmc/host/cqhci.h b/drivers/mmc/host/cqhci.h
index 8fbbc48c3f85..07eb753f9586 100644
--- a/drivers/mmc/host/cqhci.h
+++ b/drivers/mmc/host/cqhci.h
@@ -26,6 +26,8 @@
 #define CQHCI_CAP_CS			0x10000000 /* Crypto Support */
 #define CQHCI_CAP_ITCFMUL		GENMASK(15, 12)
 #define CQHCI_ITCFMUL(x)		FIELD_GET(CQHCI_CAP_ITCFMUL, (x))
+#define CQHCI_CAP_ITCFVAL		GENMASK(9, 0)
+#define CQHCI_ITCFVAL(x)		FIELD_GET(CQHCI_CAP_ITCFVAL, (x))
 
 /* configuration */
 #define CQHCI_CFG			0x08
@@ -234,6 +236,9 @@ struct cqhci_host {
 	int num_slots;
 	int qcnt;
 
+	/* Internal Timer Clock Frequency (Hz), 0 if not reported by CQHCI_CAP */
+	u32 itcf_hz;
+
 	u32 dcmd_slot;
 	u32 caps;
 #define CQHCI_TASK_DESC_SZ_128		0x1
@@ -318,6 +323,7 @@ struct platform_device;
 irqreturn_t cqhci_irq(struct mmc_host *mmc, int cmd_error, int data_error);
 int cqhci_init(struct cqhci_host *cq_host, struct mmc_host *mmc, bool dma64);
 struct cqhci_host *cqhci_pltfm_init(struct platform_device *pdev);
+int cqhci_set_cmd_idle_time(struct cqhci_host *cq_host, u32 ns);
 int cqhci_deactivate(struct mmc_host *mmc);
 void cqhci_set_tran_desc(u8 *desc, dma_addr_t addr, int len, bool end, bool dma64);
 static inline int cqhci_suspend(struct mmc_host *mmc)
-- 
2.25.1

[PATCH v1 2/2] mmc: sdhci-of-dwcmshc: Set CQHCI command idle time for StarFive JHB100

[Changhuang Liang]

Add a cqhci_host_ops for the StarFive JHB100 platform that sets the
CQHCI command idle time (CIT) to 320 ns during pre_enable.

Setting the command idle time improves eMMC performance on this
platform. Before/after dd results:

before:
dd if=/dev/mmcblk0 of=/dev/null bs=1M count=800
838860800 bytes (839 MB, 800 MiB) copied, 9.38511 s, 89.4 MB/s
dd if=/dev/zero of=/dev/mmcblk0 bs=1M count=800
838860800 bytes (839 MB, 800 MiB) copied, 12.2127 s, 68.7 MB/s

after:
dd if=/dev/mmcblk0 of=/dev/null bs=1M count=800
838860800 bytes (839 MB, 800 MiB) copied, 3.19472 s, 263 MB/s
dd if=/dev/zero of=/dev/mmcblk0 bs=1M count=800
838860800 bytes (839 MB, 800 MiB) copied, 10.7587 s, 78.0 MB/s

Signed-off-by: Changhuang Liang <changhuang.liang@starfivetech.com>
---
 drivers/mmc/host/sdhci-of-dwcmshc.c | 22 ++++++++++++++++++++++
 1 file changed, 22 insertions(+)

diff --git a/drivers/mmc/host/sdhci-of-dwcmshc.c b/drivers/mmc/host/sdhci-of-dwcmshc.c
index 249055f14162..36628b44f78b 100644
--- a/drivers/mmc/host/sdhci-of-dwcmshc.c
+++ b/drivers/mmc/host/sdhci-of-dwcmshc.c
@@ -283,6 +283,8 @@
 #define PHY_DELAY_CODE_SD		0x55
 #define PHY_DELAY_CODE_SDIO		0x29
 
+/* CIT for the JHB100 SoC */
+#define JHB100_CQHCI_CIT_NS	320
 struct rk35xx_priv {
 	struct reset_control *reset;
 	u8 txclk_tapnum;
@@ -2340,6 +2342,25 @@ static const struct sdhci_ops starfive_jhb100_ops = {
 	.platform_execute_tuning = th1520_execute_tuning,
 };
 
+static void dwcmshc_starfive_jhb100_pre_enable(struct mmc_host *mmc)
+{
+	struct cqhci_host *cq_host = mmc->cqe_private;
+	int ret;
+
+	ret = cqhci_set_cmd_idle_time(cq_host, JHB100_CQHCI_CIT_NS);
+	if (ret)
+		dev_warn_once(mmc_dev(mmc),
+			      "failed to set CQE command idle time: %d\n", ret);
+}
+
+static const struct cqhci_host_ops starfive_jhb100_cqhci_ops = {
+	.enable			= dwcmshc_sdhci_cqe_enable,
+	.disable		= sdhci_cqe_disable,
+	.pre_enable		= dwcmshc_starfive_jhb100_pre_enable,
+	.dumpregs		= dwcmshc_cqhci_dumpregs,
+	.set_tran_desc		= dwcmshc_set_tran_desc,
+};
+
 static int dwcmshc_starfive_jhb100_init(struct device *dev, struct sdhci_host *host,
 					struct dwcmshc_priv *dwc_priv)
 {
@@ -2412,6 +2433,7 @@ static const struct dwcmshc_pltfm_data sdhci_dwcmshc_jhb100_pdata = {
 		.quirks = SDHCI_QUIRK_CAP_CLOCK_BASE_BROKEN,
 		.quirks2 = SDHCI_QUIRK2_PRESET_VALUE_BROKEN,
 	},
+	.cqhci_host_ops = &starfive_jhb100_cqhci_ops,
 	.init = dwcmshc_starfive_jhb100_init,
 	.deinit = dwcmshc_starfive_jhb100_deinit,
 };
-- 
2.25.1