PXA3xx NAND controller support broken in 2.6.39

Hi,

Somewhere between 2.6.38 and 2.6.39, support for PXA3xx NAND controller was broken. I bisected it down, and the relevant commits appear to be

4332c11 mtd: pxa3xx_nand: clean the keep configure code
401e67e mtd: pxa3xx_nand: mtd scan id process could be defined by
driver itself
4eb2da8 mtd: pxa3xx_nand: unify prepare command
38caf7f mtd: pxa3xx_nand: discard wait_for_event,write_cmd,__readid function f8155a4 mtd: pxa3xx_nand: rework irq logic
e353a20 mtd: pxa3xx_nand: make scan procedure more clear
52d039f mtd: pxa3xx_nand: NULL dereference in pxa3xx_nand_probe 9d8b104 mtd: pxa3xx_nand: set oob length in the runtime
18c81b1 mtd: pxa3xx_nand: remove the flash info in driver structure 227a886 mtd: pxa3xx_nand: introduce default timing to reduce read id times c1f8247 mtd: pxa3xx_nand: condense the flash definition
99d3896 mtd: pxa3xx: fix build error when CONFIG_MTD_PARTITIONS is not defined

Unfortunately, this series does not consist of atomic commits and is hence not bisectable by itself, so I can't name a specific commit which breaks it.

On a raumfeld board (which has its board support code in mainline), the effect is:

[    1.803796] Wait time out!!!
[    2.003795] Wait time out!!!
[    2.203780] Wait time out!!!
[    2.403779] Wait time out!!!
[    2.406643] No NAND device found.
[    2.409939] pxa3xx-nand pxa3xx-nand: failed to scan nand
[    2.418850] UBI error: ubi_init: UBI error: cannot initialize UBI, error -19

With the code in 2.6.38, it works just fine. Any detailed debug output you want me to provide in order to fix it?

Daniel

When keep_config is set, the detection would goes different routine. That the driver would read out the setting which is set previously by bootloader. While most bootloader keep the irq mask as off, and current driver need all irq default open, keep_config behavior would lead to no irq at all.

---
 drivers/mtd/nand/pxa3xx_nand.c |   12 +++++++-----
 1 files changed, 7 insertions(+), 5 deletions(-)

index ff07012..9896aef 100644
@@ -813,7 +813,7 @@ static int pxa3xx_nand_detect_config(struct pxa3xx_nand_info *info)
 	info->page_size = ndcr & NDCR_PAGE_SZ ? 2048 : 512;
 	/* set info fields needed to read id */
 	info->read_id_bytes = (info->page_size == 2048) ? 4 : 2;
-	info->reg_ndcr = ndcr;
+	info->reg_ndcr = ndcr & ~NDCR_INT_MASK;
 	info->cmdset = &default_cmdset;
 
 	info->ndtr0cs0 = nand_readl(info, NDTR0CS0);

@@ -882,7 +882,7 @@ static int pxa3xx_nand_scan(struct mtd_info *mtd)
 	struct pxa3xx_nand_info *info = mtd->priv;
 	struct platform_device *pdev = info->pdev;
 	struct pxa3xx_nand_platform_data *pdata = pdev->dev.platform_data;
-	struct nand_flash_dev pxa3xx_flash_ids[2] = { {NULL,}, {NULL,} };
+	struct nand_flash_dev pxa3xx_flash_ids[2], *def = NULL;
 	const struct pxa3xx_nand_flash *f = NULL;
 	struct nand_chip *chip = mtd->priv;
 	uint32_t id = -1;

@@ -942,8 +942,10 @@ static int pxa3xx_nand_scan(struct mtd_info *mtd)
 	pxa3xx_flash_ids[0].erasesize = f->page_size * f->page_per_block;
 	if (f->flash_width == 16)
 		pxa3xx_flash_ids[0].options = NAND_BUSWIDTH_16;
+	pxa3xx_flash_ids[1].name = NULL;
+	def = pxa3xx_flash_ids;
 KEEP_CONFIG:
-	if (nand_scan_ident(mtd, 1, pxa3xx_flash_ids))
+	if (nand_scan_ident(mtd, 1, def))
 		return -ENODEV;
 	/* calculate addressing information */
 	info->col_addr_cycles = (mtd->writesize >= 2048) ? 2 : 1;

@@ -954,9 +956,9 @@ KEEP_CONFIG:
 		info->row_addr_cycles = 2;
 	mtd->name = mtd_names[0];
 	chip->ecc.mode = NAND_ECC_HW;
-	chip->ecc.size = f->page_size;
+	chip->ecc.size = info->page_size;
 
-	chip->options = (f->flash_width == 16) ? NAND_BUSWIDTH_16 : 0;
+	chip->options = (info->reg_ndcr & NDCR_DWIDTH_M) ? NAND_BUSWIDTH_16 : 0;
 	chip->options |= NAND_NO_AUTOINCR;
 	chip->options |= NAND_NO_READRDY;
 

1.7.0.4

Also cc to MTD maillist and stable kernel.

<formletter>

This is not the correct way to submit patches for inclusion in the stable kernel tree. Please read Documentation/stable_kernel_rules.txt for how to do this properly.

</formletter>

Jup.

Lei - when you collect the patches for this driver to send them out, make sure every one gets a "Cc: stable at kernel.org" line under the S-o-b, so it can be picked automatically later on.

Just to be sure, I count four patches now:

- 1 from you regarding keep_config chip detection and DMA support
(which is *not* the one this thread refers to, you sent an amended version to me privately, and it might make sense to split this patch again)
- 1 from me for the blank page ECC issue
- 2 from Alex Lin (one for removal of unused variable, one plugs a memory leak)

Thanks,
Daniel

Please, re-send the patch to the MTD mailing list and add the prober -stable CC tag.

Best Regards,
Artem Bityutskiy (????? ????????)

Hi Lei,

Thank for the quick response. Your patch fixes the detection, but read operations won't succeed now:

[    1.602733] NAND device: Manufacturer ID: 0x20, Chip ID: 0xa1 (ST
Micro NAND 128MiB 1,8V 8-bit)
[    1.611426] Scanning device for bad blocks
[    1.870669] Creating 4 MTD partitions on "pxa3xx_nand-0":
[    1.876099] 0x000000000000-0x0000000a0000 : "Bootloader"
[    1.885350] 0x0000000a0000-0x0000000c0000 : "BootloaderEnvironment"
[    1.895608] 0x0000000c0000-0x000000120000 : "BootloaderSplashScreen"
[    1.905996] 0x000000120000-0x000008000000 : "UBI"
[    1.916999] UBI: attaching mtd3 to ubi0
[    1.920835] UBI: physical eraseblock size:   131072 bytes (128 KiB)
[    1.927165] UBI: logical eraseblock size:    126976 bytes
[    1.932553] UBI: smallest flash I/O unit:    2048
[    1.937227] UBI: VID header offset:          2048 (aligned 2048)
[    1.943215] UBI: data offset:                4096
[    1.950080] UBI error: ubi_io_read: error -74 (ECC error) while
reading 2048 bytes from PEB 0:2048, read 2048 bytes
[    2.041030] UBI error: ubi_io_read: error -74 (ECC error) while
reading 126976 bytes from PEB 0:4096, read 126976 bytes
[    2.053445] UBI error: ubi_io_read: error -74 (ECC error) while
reading 2048 bytes from PEB 1:2048, read 2048 bytes
[    2.144350] UBI error: ubi_io_read: error -74 (ECC error) while

reading 126976 bytes from PEB 1:4096, read 126976 bytes

(continues forever)

The same NAND content is readable just fine with 2.6.38. Is there a similar bug for the timing parameters maybe?

Daniel