Re: [PATCH v2] mtd: rawnand: Ensure the nand chip supports cached reads

[Miquel Raynal]

Hi Martin,

miquel.raynal@xxxxxxxxxxx wrote on Tue, 26 Sep 2023 13:27:25 +0200:

> Hi Martin,
> 
> + Bean and Domenico, there is a question for you below.
> 
> martin@xxxxxxxxxx wrote on Mon, 25 Sep 2023 13:01:06 +0200:
> 
> > Hi Rouven,
> > 
> > On Fri, 2023-09-22 at 16:17 +0200, Rouven Czerwinski wrote:  
> > > Both the JEDEC and ONFI specification say that read cache sequential
> > > support is an optional command. This means that we not only need to
> > > check whether the individual controller supports the command, we also
> > > need to check the parameter pages for both ONFI and JEDEC NAND
> > > flashes
> > > before enabling sequential cache reads.
> > > 
> > > This fixes support for NAND flashes which don't support enabling
> > > cache
> > > reads, i.e. Samsung K9F4G08U0F or Toshiba TC58NVG0S3HTA00.
> > > 
> > > Sequential cache reads are now only available for ONFI and JEDEC
> > > devices, if individual vendors implement this, it needs to be enabled
> > > per vendor.
> > > 
> > > Tested on i.MX6Q with a Samsung NAND flash chip that doesn't support
> > > sequential reads.
> > > 
> > > Fixes: 003fe4b9545b ("mtd: rawnand: Support for sequential cache
> > > reads")
> > > Cc: stable@xxxxxxxxxxxxxxx
> > > Signed-off-by: Rouven Czerwinski <r.czerwinski@xxxxxxxxxxxxxx>    
> > 
> > Thanks for this. It works as expected for my Toshiba chip, obviously
> > because it doesn't use ONFI or JEDEC.
> > 
> > Unfortunately, my Micron chip does use ONFI, and it sets the cached-
> > read-supported bit. It then fails when reading afterwords:

I might have over reacted regarding my findings in Micron's datasheet,
I need to know if you use the on-die ECC engine or if you use the one
on the controller. In the former case the failure is expected. In the
latter case, it's not.

Thanks,
Miquèl

> > kernel: ONFI_OPT_CMD_READ_CACHE # debug added by me
> > kernel: nand: device found, Manufacturer ID: 0x2c, Chip ID: 0xdc
> > kernel: nand: Micron MT29F4G08ABAFAWP
> > kernel: nand: 512 MiB, SLC, erase size: 256 KiB, page size: 4096, OOB
> > size: 256
> > kernel: nand: continued read supported # debug added by me
> > kernel: Bad block table found at page 131008, version 0x01
> > kernel: Bad block table found at page 130944, version 0x01
> > kernel: 2 fixed-partitions partitions found on MTD device gpmi-nand
> > kernel: Creating 2 MTD partitions on "gpmi-nand":
> > kernel: 0x000000000000-0x000000800000 : "boot"
> > kernel: 0x000000800000-0x000020000000 : "ubi"
> > kernel: gpmi-nand 1806000.nand-controller: driver registered.
> > 
> > ...
> > 
> > kernel: ubi0: default fastmap pool size: 100
> > kernel: ubi0: default fastmap WL pool size: 50
> > kernel: ubi0: attaching mtd1
> > kernel: ubi0: scanning is finished
> > kernel: ubi0: attached mtd1 (name "ubi", size 504 MiB)
> > kernel: ubi0: PEB size: 262144 bytes (256 KiB), LEB size: 253952 bytes
> > kernel: ubi0: min./max. I/O unit sizes: 4096/4096, sub-page size 4096
> > kernel: ubi0: VID header offset: 4096 (aligned 4096), data offset: 8192
> > kernel: ubi0: good PEBs: 2012, bad PEBs: 4, corrupted PEBs: 0
> > kernel: ubi0: user volume: 9, internal volumes: 1, max. volumes count:
> > 128
> > kernel: ubi0: max/mean erase counter: 4/2, WL threshold: 4096, image
> > sequence number: 1431497221
> > kernel: ubi0: available PEBs: 12, total reserved PEBs: 2000, PEBs
> > reserved for bad PEB handling: 36
> > kernel: block ubiblock0_4: created from ubi0:4(rootfs.a)
> > kernel: ubi0: background thread "ubi_bgt0d" started, PID 36
> > kernel: block ubiblock0_6: created from ubi0:6(appfs.a)
> > kernel: block ubiblock0_7: created from ubi0:7(appfs.b)
> > 
> > ...
> > 
> > kernel: SQUASHFS error: Unable to read directory block [4b6d15c:ed1]
> > kernel: SQUASHFS error: Unable to read directory block [4b6f15e:125]
> > kernel: SQUASHFS error: Unable to read directory block [4b6d15c:1dae]
> > kernel: SQUASHFS error: Unable to read directory block [4b6d15c:ed1]
> > (d-sysctl)[55]: systemd-sysctl.service: Failed to set up credentials:
> > Protocol error
> > kernel: SQUASHFS error: Unable to read directory block [4b73162:14f0]
> > kernel: SQUASHFS error: Unable to read directory block [4b6f15e:838]
> > systemd[1]: Starting Create Static Device Nodes in /dev...
> > kernel: SQUASHFS error: Unable to read directory block [4b6d15c:ed1]
> > kernel: SQUASHFS error: Unable to read directory block [4b6d15c:ed1]
> > kernel: SQUASHFS error: Unable to read directory block [4b6f15e:838]
> > kernel: SQUASHFS error: Unable to read directory block [4b6d15c:1dae]
> > kernel: SQUASHFS error: Unable to read directory block [4b6f15e:125]
> > 
> > I've briefly tried adding some error info the the squashfs error
> > messages, but it looks like it's getting bad data. I.e. one failure a
> > sanity check of `dir_count`:
> > 
> > if (dir_count > SQUASHFS_DIR_COUNT)
> > 	goto data_error;
> > 
> > It fails with `dir_count` being 1952803684 ...
> > 
> > So is this a case of wrong/bad timings?
> > 
> > Miquel:
> > I can tell from the code, that the READCACHESEQ operations are followed
> > by NAND_OP_WAIT_RDY(tR_max, tRR_min). From the Micron datasheet[0], it
> > should be NAND_OP_WAIT_RDY(tRCBSY_max, tRR_min), where tRCBSY is
> > defined to be between 3 and 25 µs.  
> 
> I found a place in the ONFI spec states taht tRCBSY_max should be
> between 3 and tR_max, so indeed we should be fine on that regard.
> 
> However, I asked myself whether we could have issues when crossing
> boundaries. Block boundaries should be fine, however your device does
> not support crossing plane boundaries, as bit 4 ("read cache
> supported") of byte 114 ("Multi-plane operation attributes") in the
> memory organization block of the parameter page is not set (the value
> of the byte should be 0x0E if I get it right.
> 
> Anyway, our main issue here does not seem related to the boundaries. It
> does not seem to be explicitly marked anywhere else but on the front
> page:
> 	Advanced command set
> 	– Program page cache mode (4)
> 	– Read page cache mode (4)
> 	– Two-plane commands (4)
> 
> 	(4) These commands supported only with ECC disabled.
> 
> Read page cache mode without ECC makes the feature pretty useless IMHO.
> 
> Bean, Domenico, how do we know which devices allow ECC correction
> during sequential page reads and which don't? Is there a (vendor?) bit
> somewhere in the parameter page for that? Do we have any way to know
> besides a list of devices allowing that? If so, can you provide one
> with a few IDs? 
> 
> Thanks,
> Miquèl