On systems where some controllers get a dynamic ID assigned and some have
a fixed number (e.g. from ACPI tables), the current implementation might
run into an IDR collision: in case of a fixed bus number is gotten by a
driver (but not marked busy in IDR tree) and a driver with dynamic bus
number gets the same ID and predictably fails.
Fix this by means of checking-in fixed IDsin IDR as far as dynamic ones
at the moment of the controller registration.
Fixes: 9b61e302210e (spi: Pick spi bus number from Linux idr or spi alias)
Signed-off-by: Kirill Kapranov <[email protected]>
---
drivers/spi/spi.c | 9 +++++++++
1 file changed, 9 insertions(+)
diff --git a/drivers/spi/spi.c b/drivers/spi/spi.c
index ec395a6baf9c..97e303185d5b 100644
--- a/drivers/spi/spi.c
+++ b/drivers/spi/spi.c
@@ -2170,6 +2170,15 @@ int spi_register_controller(struct spi_controller *ctlr)
if (WARN(id < 0, "couldn't get idr"))
return id;
ctlr->bus_num = id;
+ } else {
+ /* devices with a fixed bus num must check-in with the num */
+ mutex_lock(&board_lock);
+ id = idr_alloc(&spi_master_idr, ctlr, ctlr->bus_num,
+ ctlr->bus_num + 1, GFP_KERNEL);
+ mutex_unlock(&board_lock);
+ if (WARN(id < 0, "couldn't get idr"))
+ return id == -ENOSPC ? -EBUSY : id;
+ ctlr->bus_num = id;
}
INIT_LIST_HEAD(&ctlr->queue);
spin_lock_init(&ctlr->queue_lock);
--
2.11.0
On Mon, Aug 13, 2018 at 07:48:10PM +0300, Kirill Kapranov wrote:
> On systems where some controllers get a dynamic ID assigned and some have
> a fixed number (e.g. from ACPI tables), the current implementation might
> run into an IDR collision: in case of a fixed bus number is gotten by a
> driver (but not marked busy in IDR tree) and a driver with dynamic bus
> number gets the same ID and predictably fails.
Is this something that's actually happened for you?
> + } else {
> + /* devices with a fixed bus num must check-in with the num */
> + mutex_lock(&board_lock);
The indentation here is weird, the comment isn't aligned with the code
it's commenting on.