u-boot.git/drivers/scsi/scsi.c, branch main

scsi: Enable FUA bit in WRITE10 commands

2026-04-27T19:44:06Z

Enable the Force Unit Access (FUA) bit in SCSI WRITE10 commands to ensure writes bypass the device's volatile cache and go directly to non-volatile storage. This provides write-through behavior that prevents data loss during board resets. Reviewed-by: Varadarajan Narayanan Signed-off-by: Aswin Murugan Signed-off-by: Balaji Selvanathan Reviewed-by: Simon Glass

Revert "scsi: sync cache on write"

2026-04-27T19:44:06Z

While the SYNC_CACHE approach prevents data loss during board resets, it causes system hangs during large sparse image flashing operations. This will be replaced with a FUA (Force Unit Access) approach in future commit, which achieves the same data integrity goal more efficiently. This reverts commit ffe4e6ab42d2534302be825e49a2d085acf80f30. Signed-off-by: Balaji Selvanathan Reviewed-by: Simon Glass

scsi: Adjust SCSI inquiry command data length

2026-03-16T16:39:26Z

Per the SCSI SPC-4 specification, the standard inquiry data length should not be less than 36 bytes. The current implementation uses 512 bytes, which causes detection failures on some UFS devices (e.g., Longsys) that do not expect a transfer length exceeding the standard inquiry size. Align the default standard inquiry length with the Linux kernel's implementation (see drivers/scsi/scsi_scan.c), which uses 36 bytes as the default. Devices requiring vendor-specific inquiry lengths should be handled through quirk settings in the future. Signed-off-by: ht.lin Signed-off-by: Macpaul Lin Reviewed-by: Marek Vasut

drivers: scsi: fix double decrement of block count in 64-bit LBA path

2025-12-22T22:19:39Z

The scsi_read function incorrectly decrements the block count twice when handling large disks that trigger the CONFIG_SYS_64BIT_LBA code path (reads beyond block 268,435,455). The variable 'blks' was being decremented both inside the 64-bit LBA block and after the successful scsi_exec() call, causing incorrect block count tracking and data abort errors on large capacity disks. Signed-off-by: Balaji Selvanathan Tested-by: Tony Dinh

scsi: Fix the name string memory leak during scsi scan

2025-11-11T17:54:48Z

There is a memory leak during the scsi scan process due to the strdup'ed name string is never freed. Actually it is unnecessary to pass a strdup'ed name string to blk_create_devicef() as we can use the name string on the stack directly. Signed-off-by: Bin Meng Reviewed-by: Heinrich Schuchardt

drivers: scsi: fix inaccurate block count reporting in scsi operations

2025-08-07T17:14:23Z

The 'blks' variable in scsi_read/write/erase functions is updated regardless of pass/fail of the scsi operation . If the scsi operation fails, 'blkcnt' is updated using an incorrect value of 'blks'. This wrong 'blkcnt' is returned to the caller and it assumes all blocks were processed correctly. Fix this by updating the 'blks' variable only if the scsi operation succeeds. Signed-off-by: Balaji Selvanathan Signed-off-by: Varadarajan Narayanan Reviewed-by: Neil Armstrong

scsi: Make static functions consistent using lbaint_t

2025-07-14T21:16:43Z

The static helper functions are inconsistent in their use of their third parameter which is used to pass a block count. Keep consistency by always using lbaint_t here. This will fix an issue where two left shifts were overflowing the variable type in use. This issue found by Smatch Signed-off-by: Andrew Goodbody Reviewed-by: Tom Rini

drivers: scsi: Add 'erase' support

2025-04-11T02:55:53Z

UFS devices uses the block and scsi frameworks. Enable UFS erase support by adding erase support to SCSI. Signed-off-by: Varadarajan Narayanan

scsi: sync cache on write

2025-04-10T20:21:41Z

We don't have a mechanism to safely shutdown block devices prior to a baord reset or driver removal. Prevent data loss by synchronizing the SCSI cache after every write. In particular this solves the issue of capsule updates looping on some devices because the board resets immediately after deleting the capsule file and this write wouldn't be flushed in time. This may impact NAND wear, but should be negligible given the usecases for disk write in U-Boot. Signed-off-by: Caleb Connolly Reviewed-by: Neil Armstrong

scsi: fix typo in setup_read_ext()

2025-04-10T20:21:41Z

This clears the 6th byte of cmd twice rather than setting the 9th byte to 0. Fix it. The only other command that sets the 9th byte is the 64-bit read, so this likely never caused issues in practise. Signed-off-by: Caleb Connolly Reviewed-by: Neil Armstrong