@@ -142,11 +142,9 @@ static inline u32 rx_max(struct dw_spi *dws)
static void dw_writer(struct dw_spi *dws)
{
- u32 max;
+ u32 max = tx_max(dws);
u16 txw = 0;
- spin_lock(&dws->buf_lock);
- max = tx_max(dws);
while (max--) {
/* Set the tx word if the transfer's original "tx" is not null */
if (dws->tx_end - dws->len) {
@@ -158,16 +156,13 @@ static void dw_writer(struct dw_spi *dws)
dw_write_io_reg(dws, DW_SPI_DR, txw);
dws->tx += dws->n_bytes;
}
- spin_unlock(&dws->buf_lock);
}
static void dw_reader(struct dw_spi *dws)
{
- u32 max;
+ u32 max = rx_max(dws);
u16 rxw;
- spin_lock(&dws->buf_lock);
- max = rx_max(dws);
while (max--) {
rxw = dw_read_io_reg(dws, DW_SPI_DR);
/* Care rx only if the transfer's original "rx" is not null */
@@ -179,7 +174,6 @@ static void dw_reader(struct dw_spi *dws)
}
dws->rx += dws->n_bytes;
}
- spin_unlock(&dws->buf_lock);
}
static void int_error_stop(struct dw_spi *dws, const char *msg)
@@ -291,21 +285,18 @@ static int dw_spi_transfer_one(struct spi_controller *master,
{
struct dw_spi *dws = spi_controller_get_devdata(master);
struct chip_data *chip = spi_get_ctldata(spi);
- unsigned long flags;
u8 imask = 0;
u16 txlevel = 0;
u32 cr0;
int ret;
dws->dma_mapped = 0;
- spin_lock_irqsave(&dws->buf_lock, flags);
dws->n_bytes = DIV_ROUND_UP(transfer->bits_per_word, BITS_PER_BYTE);
dws->tx = (void *)transfer->tx_buf;
dws->tx_end = dws->tx + transfer->len;
dws->rx = transfer->rx_buf;
dws->rx_end = dws->rx + transfer->len;
dws->len = transfer->len;
- spin_unlock_irqrestore(&dws->buf_lock, flags);
/* Ensure dw->rx and dw->rx_end are visible */
smp_mb();
@@ -464,7 +455,6 @@ int dw_spi_add_host(struct device *dev, struct dw_spi *dws)
dws->master = master;
dws->type = SSI_MOTO_SPI;
dws->dma_addr = (dma_addr_t)(dws->paddr + DW_SPI_DR);
- spin_lock_init(&dws->buf_lock);
spi_controller_set_devdata(master, dws);
@@ -131,7 +131,6 @@ struct dw_spi {
size_t len;
void *tx;
void *tx_end;
- spinlock_t buf_lock;
void *rx;
void *rx_end;
int dma_mapped;
There is no point in having the commit 19b61392c5a8 ("spi: spi-dw: Add lock protect dw_spi rx/tx to prevent concurrent calls") applied. The commit author made an assumption that the problem with the rx data mismatch was due to the lack of the data protection. While most likely it was caused by the lack of the memory barrier. So having the commit bfda044533b2 ("spi: dw: use "smp_mb()" to avoid sending spi data error") applied would be enough to fix the problem. Indeed the spin unlock operation makes sure each memory operation issued before the release will be completed before it's completed. In other words it works as an implicit one way memory barrier. So having both smp_mb() and the spin_unlock_irqrestore() here is just redundant. One of them would be enough. It's better to leave the smp_mb() since the Tx/Rx buffers consistency is provided by the data transfer algorithm implementation: first we initialize the buffers pointers, then make sure the assignments are visible by the other CPUs by calling the smp_mb(), only after that enable the interrupt, which handler uses the buffers. Signed-off-by: Serge Semin <Sergey.Semin@baikalelectronics.ru> --- Folks. I have also a doubt whether the SMP memory barrier is required there because the normal IO-methods like readl/writel imply a full memory barrier. So any memory operation performed before them are supposed to be seen by devices and another CPUs [1]. So most likely there could have been a problem with those IOs implementation on the subject platform or a spurious interrupt could have been raised during the data initialization. What do you think? Am I missing something? [1] "LINUX KERNEL MEMORY BARRIERS", Documentation/memory-barriers.txt, Section "KERNEL I/O BARRIER EFFECTS" --- drivers/spi/spi-dw-core.c | 14 ++------------ drivers/spi/spi-dw.h | 1 - 2 files changed, 2 insertions(+), 13 deletions(-)