This patch introduces improved error handling for the dma_set_mask and
ioremap calls in the lasi700_probe function within drivers/scsi/lasi700.c.
Previously, the function did not properly handle the potential failure of
these calls, which could lead to improper device initialization and
unpredictable behavior.
Although the error addressed by this patch may not occur in the current
environment, I still suggest implementing these error handling routines
if the function is not highly time-sensitive. As the environment evolves
or the code gets reused in different contexts, there's a possibility that
these errors might occur. Addressing them now can prevent potential
debugging efforts in the future, which could be quite resource-intensive.
Signed-off-by: Haoran Liu <[email protected]>
---
drivers/scsi/lasi700.c | 15 ++++++++++++++-
1 file changed, 14 insertions(+), 1 deletion(-)
diff --git a/drivers/scsi/lasi700.c b/drivers/scsi/lasi700.c
index 86fe19e0468d..8d482bee940d 100644
--- a/drivers/scsi/lasi700.c
+++ b/drivers/scsi/lasi700.c
@@ -87,6 +87,7 @@ lasi700_probe(struct parisc_device *dev)
unsigned long base = dev->hpa.start + LASI_SCSI_CORE_OFFSET;
struct NCR_700_Host_Parameters *hostdata;
struct Scsi_Host *host;
+ int err;
hostdata = kzalloc(sizeof(*hostdata), GFP_KERNEL);
if (!hostdata) {
@@ -95,8 +96,20 @@ lasi700_probe(struct parisc_device *dev)
}
hostdata->dev = &dev->dev;
- dma_set_mask(&dev->dev, DMA_BIT_MASK(32));
+ err = dma_set_mask(&dev->dev, DMA_BIT_MASK(32));
+ if (err) {
+ dev_err(&dev->dev, "Failed to set DMA mask: %d\n", err);
+ kfree(hostdata);
+ return err;
+ }
+
hostdata->base = ioremap(base, 0x100);
+ if (!hostdata->base) {
+ dev_err(&dev->dev, "ioremap failed\n");
+ kfree(hostdata);
+ return -ENOMEM;
+ }
+
hostdata->differential = 0;
if (dev->id.sversion == LASI_700_SVERSION) {
--
2.17.1
On Wed, 2023-11-29 at 06:52 -0800, Haoran Liu wrote:
> This patch introduces improved error handling for the dma_set_mask
> and ioremap calls in the lasi700_probe function within
> drivers/scsi/lasi700.c. Previously, the function did not properly
> handle the potential failure of these calls, which could lead to
> improper device initialization and unpredictable behavior.
I have to ask what the point of this is? There was once an
architecture that couldn't support 32 bit DMA masks (the altix), but
it's long gone. The structural assumption is that a 32 bit DMA mask
can always be set, so drivers with simple DMA requirements (32 or 64
bit) usually only check 64 and assume they can fallback to 32.
> Although the error addressed by this patch may not occur in the
> current environment, I still suggest implementing these error
> handling routines if the function is not highly time-sensitive. As
> the environment evolves or the code gets reused in different
> contexts, there's a possibility that these errors might occur.
> Addressing them now can prevent potential debugging efforts in the
> future, which could be quite resource-intensive.
This particular operation can never fail on the PA-RISC GSC device,
which is the only one this driver supports, so the driver currently has
no error handling because this is something that can't happen.
James