| Message ID | 20160923193935.3726427-1-arnd@arndb.de (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
Hi Arnd On 09/23/2016 09:38 PM, Arnd Bergmann wrote: > Building this driver with a 64-bit dma_addr_t type results in > a compiler warning: > > drivers/tty/serial/stm32-usart.c: In function 'stm32_of_dma_rx_probe': > drivers/tty/serial/stm32-usart.c:746:20: error: cast from pointer to integer of different size [-Werror=pointer-to-int-cast] > drivers/tty/serial/stm32-usart.c: In function 'stm32_of_dma_tx_probe': > drivers/tty/serial/stm32-usart.c:818:20: error: cast from pointer to integer of different size [-Werror=pointer-to-int-cast] > > While the type conversion here is harmless, this hints at a different > problem: we pass an __iomem pointer into a DMA engine, which expects > a phys_addr_t. This happens to work because stm32 has no MMU and > ioremap() is an identity mapping here, but it's still an incorrect > API use. Using dma_addr_t is doubly wrong here, because that would > be the result of dma_map_single() rather than the physical address. > > Using the mapbase instead fixes multiple issues: > > - the warning is gone > - we don't go through ioremap in error > - the cast is gone, making it use the correct resource_size_t/phys_addr_t > type in the process. > > Fixes: 3489187204eb ("serial: stm32: adding dma support") > Signed-off-by: Arnd Bergmann <arnd@arndb.de> > --- > drivers/tty/serial/stm32-usart.c | 4 ++-- > 1 file changed, 2 insertions(+), 2 deletions(-) > > diff --git a/drivers/tty/serial/stm32-usart.c b/drivers/tty/serial/stm32-usart.c > index 4d3001b77e7e..2adb678a863b 100644 > --- a/drivers/tty/serial/stm32-usart.c > +++ b/drivers/tty/serial/stm32-usart.c > @@ -743,7 +743,7 @@ static int stm32_of_dma_rx_probe(struct stm32_port *stm32port, > > /* Configure DMA channel */ > memset(&config, 0, sizeof(config)); > - config.src_addr = (dma_addr_t)port->membase + ofs->rdr; > + config.src_addr = port->mapbase + ofs->rdr; > config.src_addr_width = DMA_SLAVE_BUSWIDTH_1_BYTE; > > ret = dmaengine_slave_config(stm32port->rx_ch, &config); > @@ -815,7 +815,7 @@ static int stm32_of_dma_tx_probe(struct stm32_port *stm32port, > > /* Configure DMA channel */ > memset(&config, 0, sizeof(config)); > - config.dst_addr = (dma_addr_t)port->membase + ofs->tdr; > + config.dst_addr = port->mapbase + ofs->tdr; > config.dst_addr_width = DMA_SLAVE_BUSWIDTH_1_BYTE; > > ret = dmaengine_slave_config(stm32port->tx_ch, &config); > Indeed, thanks for this finding. Reviewed-by: Gerald Baeza <gerald.baeza@st.com>
diff --git a/drivers/tty/serial/stm32-usart.c b/drivers/tty/serial/stm32-usart.c index 4d3001b77e7e..2adb678a863b 100644 --- a/drivers/tty/serial/stm32-usart.c +++ b/drivers/tty/serial/stm32-usart.c @@ -743,7 +743,7 @@ static int stm32_of_dma_rx_probe(struct stm32_port *stm32port, /* Configure DMA channel */ memset(&config, 0, sizeof(config)); - config.src_addr = (dma_addr_t)port->membase + ofs->rdr; + config.src_addr = port->mapbase + ofs->rdr; config.src_addr_width = DMA_SLAVE_BUSWIDTH_1_BYTE; ret = dmaengine_slave_config(stm32port->rx_ch, &config); @@ -815,7 +815,7 @@ static int stm32_of_dma_tx_probe(struct stm32_port *stm32port, /* Configure DMA channel */ memset(&config, 0, sizeof(config)); - config.dst_addr = (dma_addr_t)port->membase + ofs->tdr; + config.dst_addr = port->mapbase + ofs->tdr; config.dst_addr_width = DMA_SLAVE_BUSWIDTH_1_BYTE; ret = dmaengine_slave_config(stm32port->tx_ch, &config);
Building this driver with a 64-bit dma_addr_t type results in a compiler warning: drivers/tty/serial/stm32-usart.c: In function 'stm32_of_dma_rx_probe': drivers/tty/serial/stm32-usart.c:746:20: error: cast from pointer to integer of different size [-Werror=pointer-to-int-cast] drivers/tty/serial/stm32-usart.c: In function 'stm32_of_dma_tx_probe': drivers/tty/serial/stm32-usart.c:818:20: error: cast from pointer to integer of different size [-Werror=pointer-to-int-cast] While the type conversion here is harmless, this hints at a different problem: we pass an __iomem pointer into a DMA engine, which expects a phys_addr_t. This happens to work because stm32 has no MMU and ioremap() is an identity mapping here, but it's still an incorrect API use. Using dma_addr_t is doubly wrong here, because that would be the result of dma_map_single() rather than the physical address. Using the mapbase instead fixes multiple issues: - the warning is gone - we don't go through ioremap in error - the cast is gone, making it use the correct resource_size_t/phys_addr_t type in the process. Fixes: 3489187204eb ("serial: stm32: adding dma support") Signed-off-by: Arnd Bergmann <arnd@arndb.de> --- drivers/tty/serial/stm32-usart.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)