在 2023/3/18 上午12:26, Mark Brown 写道:
On Fri, Mar 17, 2023 at 04:29:50PM +0800, Yinbo Zhu wrote:
+static int loongson_spi_update_state(struct loongson_spi *loongson_spi,
+ struct spi_device *spi, struct spi_transfer *t)
+{
...
+ loongson_spi->hz = hz;
+ loongson_spi->spcr = div_tmp & 3;
+ loongson_spi->sper = (div_tmp >> 2) & 3;
+ val = loongson_spi_read_reg(loongson_spi, LOONGSON_SPI_SPCR_REG);
This is writing to general chip registers, apparently not per chip
select ones.
The loongson_spi_update_state was only be called in setup or transfer_one, and I will also
add a spin lock in tranfser_one.
+static int loongson_spi_setup(struct spi_device *spi)
+{
+ struct loongson_spi *loongson_spi;
+ spin_lock(&loongson_spi->lock);
+ loongson_spi_update_state(loongson_spi, spi, NULL);
As IIRC I mentioned last time setup() might be called while other
transfers are happening and therefore shouldn't affect parallel
operations on other devices.
I think add spin_lock in transfer_one interface that should be to fix
this issue, Do you think so?
loongson_spi_transfer_one(struct spi_controller *ctrl, struct spi_dev
{
struct loongson_spi *loongson_spi =
spi_master_get_devdata(spi->master);
+ spin_lock(&loongson_spi->lock);
loongson_spi_update_state(loongson_spi, spi, xfer);
+ spin_unlock(&loongson_spi->lock);
+static const struct of_device_id loongson_spi_id_table[] = {
+ { .compatible = "loongson,ls2k-spi", },
+ { }
+};
+MODULE_DEVICE_TABLE(of, loongson_spi_id_table);
+
+static struct platform_driver loongson_spi_driver = {
+ .probe = loongson_spi_platform_probe,
+ .driver = {
+ .name = "loongson-spi",
+ .owner = THIS_MODULE,
+ .bus = &platform_bus_type,
+ .pm = &loongson_spi_dev_pm_ops,
+ .of_match_table = loongson_spi_id_table,
+ },
+};
+
+#ifdef CONFIG_PCI
+static int loongson_spi_pci_register(struct pci_dev *pdev,
+ const struct pci_device_id *ent)
Again as I said last time the two buses should probably be separate
modules.
Otherwise this looks fine.
okay, I will do it.
