[PATCH AUTOSEL 6.19-6.6] spi: spi-mem: Limit octal DTR constraints to octal DTR situations

[Sasha Levin <sashal@kernel.org>]

From: Miquel Raynal <miquel.raynal@bootlin.com>

[ Upstream commit 8618271887ca10ac5108fe7e1d82ba8f1b152cf9 ]

In this helper, any operation with a single DTR cycle (like 1S-1S-8D) is
considered requiring a duplicated command opcode. This is wrong as this
constraint only applies to octal DTR operations (8D-8D-8D).

Narrow the application of this constraint to the concerned bus
interface.

Note: none of the possible XD-XD-XD pattern, with X being one of {1, 2,
4} would benefit from this check either as there is only in octal DTR
mode that a single clock edge would be enough to transmit the full
opcode.

Make sure the constraint of expecting two bytes for the command is
applied to the relevant bus interface.

Reviewed-by: Tudor Ambarus <tudor.ambarus@linaro.org>
Signed-off-by: Miquel Raynal <miquel.raynal@bootlin.com>
Link: https://patch.msgid.link/20260109-winbond-v6-17-rc1-oddr-v2-3-1fff6a2ddb80@bootlin.com
Signed-off-by: Mark Brown <broonie@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
---

LLM Generated explanations, may be completely bogus:

The core of the fix — moving the `op->cmd.nbytes != 2` check inside a
tighter condition — is self-contained. It doesn't depend on the `swap16`
feature (that's a separate, pre-existing check). The new even-byte
checks for addr/dummy/data in 8D-8D-8D mode are a tightening of
validation that could also be backported.

However, the presence of the `swap16` check in the surrounding context
means the patch won't apply cleanly to older stable trees that lack
commit `030ace430afcf` (from 6.12 cycle). The fix would need minor
adaptation for older stable trees, but the logic change itself is
independent.

### 7. Assessment Summary

**What the bug is:** `spi_mem_default_supports_op()` incorrectly rejects
valid non-8D-8D-8D DTR operations (like 1S-1S-8D) by requiring
`cmd.nbytes == 2` for ALL DTR operations, not just 8D-8D-8D ones.

**Impact:** Real functional bug — any SPI memory device using mixed DTR
modes (where only the data phase is DTR but not the command phase) would
have its operations rejected. This prevents legitimate device
operations.

**Scope of the fix:** The change is confined to a single function in one
file. It narrows a condition check and adds additional validation only
for the specific 8D-8D-8D case.

**Risk:** Low. The fix is logically correct — it restricts a previously
over-broad check. It can't introduce new failures for 8D-8D-8D mode
(those checks are preserved and even strengthened), and it unblocks
mixed-DTR modes that were incorrectly blocked.

**Who benefits:** Users with SPI memory devices that use non-8D-8D-8D
DTR modes (e.g., some flash chips using 1S-1S-8D for certain
operations).

**Concerns:**
- Part of a larger patch series (v2-3), but the fix to
  `spi_mem_default_supports_op()` is self-contained
- May need minor context adjustment for older stable trees (due to
  `swap16` context)
- The even-byte-count validation is technically a new constraint for
  8D-8D-8D mode, but it enforces existing hardware requirements that
  were previously unchecked

This is a clear correctness fix for a real bug that has existed since
kernel 5.18. It's small, contained, well-reviewed (Reviewed-by from
Tudor Ambarus, a key MTD maintainer), and fixes incorrect rejection of
valid SPI operations.

**YES**

 drivers/spi/spi-mem.c | 15 +++++++++++++--
 1 file changed, 13 insertions(+), 2 deletions(-)

diff --git a/drivers/spi/spi-mem.c b/drivers/spi/spi-mem.c
index c8b2add2640e5..6c7921469b90b 100644
--- a/drivers/spi/spi-mem.c
+++ b/drivers/spi/spi-mem.c
@@ -178,8 +178,19 @@ bool spi_mem_default_supports_op(struct spi_mem *mem,
 		if (op->data.swap16 && !spi_mem_controller_is_capable(ctlr, swap16))
 			return false;
 
-		if (op->cmd.nbytes != 2)
-			return false;
+		/* Extra 8D-8D-8D limitations */
+		if (op->cmd.dtr && op->cmd.buswidth == 8) {
+			if (op->cmd.nbytes != 2)
+				return false;
+
+			if ((op->addr.nbytes % 2) ||
+			    (op->dummy.nbytes % 2) ||
+			    (op->data.nbytes % 2)) {
+				dev_err(&ctlr->dev,
+					"Even byte numbers not allowed in octal DTR operations\n");
+				return false;
+			}
+		}
 	} else {
 		if (op->cmd.nbytes != 1)
 			return false;
-- 
2.51.0