[PATCH] mtd: rawnand: samsung: fix pages_per_eraseblock calculation

[Shuai Liu]

In samsung_nand_decode_id(), the division by memorg->pagesize binds
tighter than the left shift due to C operator precedence, so the
expression is evaluated as:

    (128 * 1024) << (shift / pagesize)

The shift amount is at most 7 while pagesize is at least 2048, so the
integer division always yields 0 and pages_per_eraseblock ends up
being 131072 regardless of the actual chip geometry.

nanddev_init() later computes mtd->erasesize as
pagesize * pages_per_eraseblock, overwriting the correct value set in
this function with an astronomic value (e.g. 0x20000000, 512 MiB, for
a chip with 4096-byte pages).  MTD then rejects partitions not
starting on such an eraseblock boundary and forces them read-only,
breaking the rootfs on affected boards.

Add parentheses around the shift expression so that the eraseblock
size is computed before being divided by the page size.

Reported-by: kevinshq <kevinshq at 163.com>
Closes: https://bugzilla.kernel.org/show_bug.cgi?id=222140
Fixes: 629a442cad5f ("mtd: rawnand: Fill memorg during detection")
Signed-off-by: Shuai Liu <847807867 at qq.com>
---
 drivers/mtd/nand/raw/nand_samsung.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/drivers/mtd/nand/raw/nand_samsung.c b/drivers/mtd/nand/raw/nand_samsung.c
index 0be6b7563..765c0a7d8 100644
--- a/drivers/mtd/nand/raw/nand_samsung.c
+++ b/drivers/mtd/nand/raw/nand_samsung.c
@@ -63,9 +63,9 @@ static void samsung_nand_decode_id(struct nand_chip *chip)
 
 		/* Get blocksize */
 		extid >>= 2;
-		memorg->pages_per_eraseblock = (128 * 1024) <<
-					       (((extid >> 1) & 0x04) |
-						(extid & 0x03)) /
+		memorg->pages_per_eraseblock = ((128 * 1024) <<
+						(((extid >> 1) & 0x04) |
+						 (extid & 0x03))) /
 					       memorg->pagesize;
 		mtd->erasesize = (128 * 1024) <<
 				 (((extid >> 1) & 0x04) | (extid & 0x03));
-- 
2.43.0