spi: tegra210-quad: Add runtime autosuspend support

Using Tegra234, it was found that it takes about 10us to disable clocks
and 20us to enable clocks, adding about 30us overhead per operation.
For a 4MB firmware update with 16,384 page programs, this results in
~491ms total overhead (12% impact).

With Tegra234, flash operations were observed to occur in bursts with
50-200μs gaps between page programs. Testing on Tegra234 with various
operation patterns shows 500ms delay provides optimal balance, and for
longer operations (>500ms), the overhead is negligible. Therefore,
update the driver to use pm-runtime autosuspend with the default timeout
of 500ms to reduce the clock gating overhead during consecutive QSPI
transfers.

Signed-off-by: Vishwaroop A <va@nvidia.com>
Link: https://patch.msgid.link/20260224092452.1482800-1-va@nvidia.com
Signed-off-by: Mark Brown <broonie@kernel.org>

diff --git a/drivers/spi/spi-tegra210-quad.c b/drivers/spi/spi-tegra210-quad.c
index 7ea5aa993596f6be9bb2cb8cdf4f7c1066442e9f..7fac695b514219e422c6fcb924376a10943a371f 100644 (file)
--- a/drivers/spi/spi-tegra210-quad.c
+++ b/drivers/spi/spi-tegra210-quad.c
@@ -1000,7 +1000,8 @@ static int tegra_qspi_setup(struct spi_device *spi)
 
        spin_unlock_irqrestore(&tqspi->lock, flags);
 
-       pm_runtime_put(tqspi->dev);
+       pm_runtime_mark_last_busy(tqspi->dev);
+       pm_runtime_put_autosuspend(tqspi->dev);
 
        return 0;
 }
@@ -1765,6 +1766,14 @@ static int tegra_qspi_probe(struct platform_device *pdev)
        init_completion(&tqspi->rx_dma_complete);
        init_completion(&tqspi->xfer_completion);
 
+       /*
+        * Set autosuspend delay to 500ms. Testing shows this value eliminates
+        * suspend/resume overhead during burst operations while allowing quick
+        * suspension during idle. For longer operations, the overhead is negligible.
+        */
+       pm_runtime_set_autosuspend_delay(&pdev->dev, 500);
+       pm_runtime_use_autosuspend(&pdev->dev);
+
        pm_runtime_enable(&pdev->dev);
        ret = pm_runtime_resume_and_get(&pdev->dev);
        if (ret < 0) {
@@ -1781,7 +1790,8 @@ static int tegra_qspi_probe(struct platform_device *pdev)
        tqspi->spi_cs_timing2 = tegra_qspi_readl(tqspi, QSPI_CS_TIMING2);
        tqspi->def_command2_reg = tegra_qspi_readl(tqspi, QSPI_COMMAND2);
 
-       pm_runtime_put(&pdev->dev);
+       pm_runtime_mark_last_busy(&pdev->dev);
+       pm_runtime_put_autosuspend(&pdev->dev);
 
        ret = request_threaded_irq(tqspi->irq, NULL,
                                   tegra_qspi_isr_thread, IRQF_ONESHOT,
@@ -1802,6 +1812,7 @@ static int tegra_qspi_probe(struct platform_device *pdev)
 exit_free_irq:
        free_irq(qspi_irq, tqspi);
 exit_pm_disable:
+       pm_runtime_dont_use_autosuspend(&pdev->dev);
        pm_runtime_force_suspend(&pdev->dev);
        tegra_qspi_deinit_dma(tqspi);
        return ret;
@@ -1814,6 +1825,7 @@ static void tegra_qspi_remove(struct platform_device *pdev)
 
        spi_unregister_controller(host);
        free_irq(tqspi->irq, tqspi);
+       pm_runtime_dont_use_autosuspend(&pdev->dev);
        pm_runtime_force_suspend(&pdev->dev);
        tegra_qspi_deinit_dma(tqspi);
 }
@@ -1839,7 +1851,8 @@ static int __maybe_unused tegra_qspi_resume(struct device *dev)
 
        tegra_qspi_writel(tqspi, tqspi->command1_reg, QSPI_COMMAND1);
        tegra_qspi_writel(tqspi, tqspi->def_command2_reg, QSPI_COMMAND2);
-       pm_runtime_put(dev);
+       pm_runtime_mark_last_busy(dev);
+       pm_runtime_put_autosuspend(dev);
 
        return spi_controller_resume(host);
 }