lists.openwall.net   lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  linux-hardening  linux-cve-announce  PHC 
Open Source and information security mailing list archives
 
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-Id: <20240308-cdns-qspi-mbly-v1-7-a503856dd205@bootlin.com>
Date: Fri, 08 Mar 2024 18:18:22 +0100
From: Théo Lebrun <theo.lebrun@...tlin.com>
To: Mark Brown <broonie@...nel.org>, Rob Herring <robh+dt@...nel.org>, 
 Krzysztof Kozlowski <krzysztof.kozlowski+dt@...aro.org>, 
 Conor Dooley <conor+dt@...nel.org>, Vaishnav Achath <vaishnav.a@...com>, 
 Thomas Bogendoerfer <tsbogend@...ha.franken.de>
Cc: linux-spi@...r.kernel.org, devicetree@...r.kernel.org, 
 linux-kernel@...r.kernel.org, linux-mips@...r.kernel.org, 
 Vladimir Kondratiev <vladimir.kondratiev@...ileye.com>, 
 Gregory CLEMENT <gregory.clement@...tlin.com>, 
 Thomas Petazzoni <thomas.petazzoni@...tlin.com>, 
 Tawfik Bayouk <tawfik.bayouk@...ileye.com>, 
 Théo Lebrun <theo.lebrun@...tlin.com>
Subject: [PATCH 07/11] spi: cadence-qspi: add no-IRQ mode to indirect reads

Support reads through polling, without any IRQ. The main reason is
performance; profiling shows that the first IRQ comes quickly on our
specific hardware. Once this IRQ arrives, we poll until all data is
retrieved. Avoid initial sleep to reduce IRQ count.

Hide this behavior behind a quirk flag.

This is confirmed through micro-benchmarks, but also end-to-end
performance tests. Mobileye EyeQ5, octal flash, reading 235M on a UBIFS
filesystem:
 - No optimizations, ~10.34s, ~22.7 MB/s, 199230 IRQs
 - CQSPI_SLOW_SRAM,  ~10.34s, ~22.7 MB/s,  70284 IRQs
 - CQSPI_RD_NO_IRQ,   ~9.37s, ~25.1 MB/s,    521 IRQs

Signed-off-by: Théo Lebrun <theo.lebrun@...tlin.com>
---
 drivers/spi/spi-cadence-quadspi.c | 13 +++++++++----
 1 file changed, 9 insertions(+), 4 deletions(-)

diff --git a/drivers/spi/spi-cadence-quadspi.c b/drivers/spi/spi-cadence-quadspi.c
index 0fc452bab0ee..973ea4edbe3a 100644
--- a/drivers/spi/spi-cadence-quadspi.c
+++ b/drivers/spi/spi-cadence-quadspi.c
@@ -43,6 +43,7 @@ static_assert(CQSPI_MAX_CHIPSELECT <= SPI_CS_CNT_MAX);
 #define CQSPI_SLOW_SRAM		BIT(4)
 #define CQSPI_NEEDS_APB_AHB_HAZARD_WAR	BIT(5)
 #define CQSPI_DETECT_FIFO_DEPTH		BIT(6)
+#define CQSPI_RD_NO_IRQ			BIT(7)
 
 /* Capabilities */
 #define CQSPI_SUPPORTS_OCTAL		BIT(0)
@@ -703,6 +704,7 @@ static int cqspi_indirect_read_execute(struct cqspi_flash_pdata *f_pdata,
 				       const size_t n_rx)
 {
 	struct cqspi_st *cqspi = f_pdata->cqspi;
+	bool use_irq = !(cqspi->ddata && cqspi->ddata->quirks & CQSPI_RD_NO_IRQ);
 	struct device *dev = &cqspi->pdev->dev;
 	void __iomem *reg_base = cqspi->iobase;
 	void __iomem *ahb_base = cqspi->ahb_base;
@@ -726,17 +728,20 @@ static int cqspi_indirect_read_execute(struct cqspi_flash_pdata *f_pdata,
 	 * all the read interrupts disabled for max performance.
 	 */
 
-	if (!cqspi->slow_sram)
+	if (use_irq && cqspi->slow_sram)
+		writel(CQSPI_REG_IRQ_WATERMARK, reg_base + CQSPI_REG_IRQMASK);
+	else if (use_irq)
 		writel(CQSPI_IRQ_MASK_RD, reg_base + CQSPI_REG_IRQMASK);
 	else
-		writel(CQSPI_REG_IRQ_WATERMARK, reg_base + CQSPI_REG_IRQMASK);
+		writel(0, reg_base + CQSPI_REG_IRQMASK);
 
 	reinit_completion(&cqspi->transfer_complete);
 	writel(CQSPI_REG_INDIRECTRD_START_MASK,
 	       reg_base + CQSPI_REG_INDIRECTRD);
 
 	while (remaining > 0) {
-		if (!wait_for_completion_timeout(&cqspi->transfer_complete,
+		if (use_irq &&
+		    !wait_for_completion_timeout(&cqspi->transfer_complete,
 						 msecs_to_jiffies(CQSPI_READ_TIMEOUT_MS)))
 			ret = -ETIMEDOUT;
 
@@ -778,7 +783,7 @@ static int cqspi_indirect_read_execute(struct cqspi_flash_pdata *f_pdata,
 			bytes_to_read = cqspi_get_rd_sram_level(cqspi);
 		}
 
-		if (remaining > 0) {
+		if (use_irq && remaining > 0) {
 			reinit_completion(&cqspi->transfer_complete);
 			if (cqspi->slow_sram)
 				writel(CQSPI_REG_IRQ_WATERMARK, reg_base + CQSPI_REG_IRQMASK);

-- 
2.44.0


Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ