On 5/7/25 13:28, Varadarajan Narayanan wrote:
Add a function to obtain the block device for SCSI.

Signed-off-by: Varadarajan Narayanan <quic_var...@quicinc.com>
---
v3: * s/scsi_get_blk/scsi_get_blk_by_uuid
     * s/partition_name/uuid
---
  drivers/scsi/scsi-uclass.c | 30 ++++++++++++++++++++++++++++++
  include/scsi.h             | 10 ++++++++++
  2 files changed, 40 insertions(+)

diff --git a/drivers/scsi/scsi-uclass.c b/drivers/scsi/scsi-uclass.c
index 1ee8236c05c..3eb6069649f 100644
--- a/drivers/scsi/scsi-uclass.c
+++ b/drivers/scsi/scsi-uclass.c
@@ -10,7 +10,9 @@
#define LOG_CATEGORY UCLASS_SCSI +#include <blk.h>
  #include <dm.h>
+#include <part.h>
  #include <scsi.h>
int scsi_exec(struct udevice *dev, struct scsi_cmd *pccb)
@@ -23,6 +25,34 @@ int scsi_exec(struct udevice *dev, struct scsi_cmd *pccb)
        return ops->exec(dev, pccb);
  }
+int scsi_get_blk_by_uuid(const char *uuid,
+                        struct blk_desc **blk_desc_ptr,
+                        struct disk_partition *part_info_ptr)
+{
+       static int is_scsi_scanned;
+       struct blk_desc *blk;
+       int i, ret;
+
+       if (!is_scsi_scanned) {
+               scsi_scan(false /* no verbose */);
+               is_scsi_scanned = 1;

At this point I think we should just unconditionally scsi_scan() like we do with MMC... Tom (others?) any thoughts on this? Anyhow this is probably fine.

One thing to be extremely wary of is that we don't call scsi_scan() after initialising EFI driver handles because that will cause big issues (ran into this one before), since scsi_scan() will unbind and rebind drivers if it was already called. I don't think that's a problem with the current ordering of things.

+       }
+
+       for (i = 0; i < blk_find_max_devnum(UCLASS_SCSI) + 1; i++) {
+               ret = blk_get_desc(UCLASS_SCSI, i, &blk);
+               if (ret)
+                       continue;

As before, please check for the end of the list and break here (-ENODEV is the magic value). Really it's bad practise to ignore ret, we should only ever continue if ret is -ENOENT, otherwise we should return the errorcode (and probably translate a -ENODEV to -ENOENT since that makes more sense here)

Kind regards,

+
+               ret = part_get_info_by_uuid(blk, uuid, part_info_ptr);
+               if (ret > 0) {
+                       *blk_desc_ptr = blk;
+                       return 0;
+               }
+       }
+
+       return -1;

        return -ENOENT;

Kind regards,> +}
+
  int scsi_bus_reset(struct udevice *dev)
  {
        struct scsi_ops *ops = scsi_get_ops(dev);
diff --git a/include/scsi.h b/include/scsi.h
index ab53b47b58f..8d6c5116419 100644
--- a/include/scsi.h
+++ b/include/scsi.h
@@ -351,6 +351,16 @@ int scsi_scan(bool verbose);
   */
  int scsi_scan_dev(struct udevice *dev, bool verbose);
+/**
+ * scsi_get_blk_by_uuid() - Provides SCSI partition information.
+ *
+ * @uuid:              UUID of the partition for fetching its info
+ * @blk_desc_ptr:      Provides the blk descriptor
+ * @part_info_ptr:     Provides partition info
+ */
+int scsi_get_blk_by_uuid(const char *uuid, struct blk_desc **blk_desc_ptr,
+                        struct disk_partition *part_info_ptr);
+
  #define SCSI_IDENTIFY                                 0xC0  /* not used */
/* Hardware errors */
--
Casey (she/they)

Reply via email to