[PATCH AUTOSEL 6.16-6.12] scsi: lpfc: Ensure HBA_SETUP flag is used only for SLI4 in dev_loss_tmo_callbk

[Sasha Levin <sashal@kernel.org>]

From: Justin Tee <justin.tee@broadcom.com>

[ Upstream commit 1cced5779e7a3ff7ec025fc47c76a7bd3bb38877 ]

For SLI3, the HBA_SETUP flag is never set so the lpfc_dev_loss_tmo_callbk
always early returns.  Add a phba->sli_rev check for SLI4 mode so that
the SLI3 path can flow through the original dev_loss_tmo worker thread
design to lpfc_dev_loss_tmo_handler instead of early return.

Signed-off-by: Justin Tee <justin.tee@broadcom.com>
Link: https://lore.kernel.org/r/20250618192138.124116-9-justintee8345@gmail.com
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
---

LLM Generated explanations, may be completely bogus:

Now I have enough context. Let me analyze the commit thoroughly.

**Backport Status: YES**

This commit fixes a significant bug in the lpfc SCSI driver that causes
SLI3 mode HBAs to completely bypass dev_loss_tmo handling, which is a
critical issue for device loss timeout processing.

## Analysis:

### 1. **Bug Description:**
The commit fixes a bug where `lpfc_dev_loss_tmo_callbk()` always returns
early for SLI3 HBAs because the `HBA_SETUP` flag is never set for SLI3
mode. The code at line 186 checks:
```c
if (test_bit(FC_UNLOADING, &vport->load_flag) ||
    !test_bit(HBA_SETUP, &phba->hba_flag))
```

### 2. **Root Cause:**
- The `HBA_SETUP` flag is only set in `lpfc_sli4_hba_setup()` function
  (drivers/scsi/lpfc/lpfc_sli.c:8850)
- For SLI3, the initialization function is `lpfc_sli_hba_setup()` which
  never sets this flag
- This check was incorrectly added in commit 2be1d4f11944c (July 2024)
  without considering SLI3 mode

### 3. **Impact:**
- **SLI3 HBAs completely lose dev_loss_tmo functionality**: The early
  return prevents the worker thread event from being scheduled (lines
  277-289), which means `lpfc_dev_loss_tmo_handler()` is never called
- **Device removal handling is broken**: Without proper dev_loss_tmo
  processing, devices that disappear from the fabric won't be properly
  cleaned up
- **Memory leaks and stale references**: Node references won't be
  properly released, potentially causing memory leaks

### 4. **The Fix:**
The fix adds a proper SLI revision check:
```c
if (test_bit(FC_UNLOADING, &vport->load_flag) ||
    (phba->sli_rev == LPFC_SLI_REV4 &&
    !test_bit(HBA_SETUP, &phba->hba_flag)))
```
This ensures the HBA_SETUP check only applies to SLI4, allowing SLI3 to
flow through to the original worker thread design.

### 5. **Stable Backport Criteria:**
✓ **Fixes a real bug**: Completely broken dev_loss_tmo for SLI3 HBAs
✓ **Small and contained**: Single-line logic fix with minimal risk
✓ **No architectural changes**: Simply corrects a logic error
✓ **Critical functionality**: dev_loss_tmo is essential for proper SCSI
device management
✓ **Regression fix**: This fixes a regression introduced by commit
2be1d4f11944c

### 6. **Risk Assessment:**
- **Very low risk**: The change only adds a SLI revision check before an
  existing condition
- **Well-tested code path**: Restores SLI3 to use the original, well-
  tested dev_loss_tmo path
- **No new functionality**: Simply fixes broken existing functionality

This is a textbook example of a stable backport candidate - it fixes a
serious regression that completely breaks essential functionality for
SLI3 HBAs with a minimal, safe change.

 drivers/scsi/lpfc/lpfc_hbadisc.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/scsi/lpfc/lpfc_hbadisc.c b/drivers/scsi/lpfc/lpfc_hbadisc.c
index b88e54a7e65c..3962f07c9140 100644
--- a/drivers/scsi/lpfc/lpfc_hbadisc.c
+++ b/drivers/scsi/lpfc/lpfc_hbadisc.c
@@ -183,7 +183,8 @@ lpfc_dev_loss_tmo_callbk(struct fc_rport *rport)
 
 	/* Don't schedule a worker thread event if the vport is going down. */
 	if (test_bit(FC_UNLOADING, &vport->load_flag) ||
-	    !test_bit(HBA_SETUP, &phba->hba_flag)) {
+	    (phba->sli_rev == LPFC_SLI_REV4 &&
+	    !test_bit(HBA_SETUP, &phba->hba_flag))) {
 
 		spin_lock_irqsave(&ndlp->lock, iflags);
 		ndlp->rport = NULL;
-- 
2.39.5