From: Thomas Schaefer Date: Tue, 2 Jul 2019 11:43:13 +0000 Subject: [U-Boot] [PATCH 2/2] drivers/spi: fsl_qspi: improve timeout calculation In-Reply-To: References: <20190701153735.170-1-thomas.schaefer@kontron.com> <20190701153735.170-3-thomas.schaefer@kontron.com> Message-ID: <7982fe35cb0240b4a4a441fa34dd18f0@kontron.com> List-Id: MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: u-boot@lists.denx.de > Von: Jagan Teki > Gesendet: Dienstag, 2. Juli 2019 13:12 > > On Mon, Jul 1, 2019 at 9:07 PM Thomas Schaefer wrote: > > > > Use readl_poll_timeout instead of explicit calculation > > > > Signed-off-by: Thomas Schaefer > > --- > > drivers/spi/fsl_qspi.c | 17 ++++++----------- > > 1 file changed, 6 insertions(+), 11 deletions(-) > > > > diff --git a/drivers/spi/fsl_qspi.c b/drivers/spi/fsl_qspi.c index > > 2c5937509f..41abe1996f 100644 > > --- a/drivers/spi/fsl_qspi.c > > +++ b/drivers/spi/fsl_qspi.c > > @@ -10,6 +10,7 @@ > > #include > > #include > > #include > > +#include > > #include > > #include > > #include > > @@ -150,19 +151,13 @@ static void qspi_write32(u32 flags, u32 *addr, > > u32 val) static inline int is_controller_busy(const struct > > fsl_qspi_priv *priv) { > > u32 val; > > - const u32 mask = QSPI_SR_BUSY_MASK | QSPI_SR_AHB_ACC_MASK | > > - QSPI_SR_IP_ACC_MASK; > > - unsigned long timeout = timer_get_us() + 1000; > > + u32 mask = QSPI_SR_BUSY_MASK | QSPI_SR_AHB_ACC_MASK | > > + QSPI_SR_IP_ACC_MASK; > > > > - do { > > - val = qspi_read32(priv->flags, &priv->regs->sr); > > + if (priv->flags & QSPI_FLAG_REGMAP_ENDIAN_BIG) > > + mask = (u32)cpu_to_be32(mask); > > This look like a BE check, which was not there before isn't it? > Hi Jagan, The previous implementation was using the qspi_read32 function, that implemented this endianess checking. As the readl_poll_timeout macro finally ends with #define __arch_getl(a) (*(volatile unsigned int *)(a)) in arch/arm/include/asm/io.h. When swapping the 'mask' in case of BE, the result is the same as when using qspi_read32 before. However, I did not test this as the QSPI_FLAG_REGMAP_ENDIAN_BIG is not set on i.MX7 systems. Also I found that current linux kernel does this BE check (negated LE check) with the 'mask' variable in drivers/spi/spi-fsl-qspi.c if (!q->devtype_data->little_endian) mask = (u32)cpu_to_be32(mask); Best reards, Thomas