[FS#1376] regression: reading from 32 MB flash mtd fails with kernel 4.9

[LEDE Bugs]

A new Flyspray task has been opened.  Details are below. 

User who did this - Christian Beier (bk138) 

Attached to Project - OpenWrt/LEDE Project
Summary - regression: reading from 32 MB flash mtd fails with kernel 4.9
Task Type - Bug Report
Category - Kernel
Status - Unconfirmed
Assigned To - 
Operating System - All
Severity - Medium
Priority - Very Low
Reported Version - Trunk
Due in Version - Undecided
Due Date - Undecided
Details - 
Hi OpenWRT/LEDE people,

I have a device I am currently adding to OpenWRT (https://wiki.openwrt.org/inbox/engenius/engenius_emr3000v2) that boots just fine with Linux 4.4 but fails under Linux 4.9 with


[    0.951402] m25p80 spi0.0: found s25fl256s1, expected m25p80
[    0.957203] m25p80 spi0.0: s25fl256s1 (32768 Kbytes)
[    0.962266] 7 cmdlinepart partitions found on MTD device spi0.0
[    0.968277] Creating 7 MTD partitions on "spi0.0":
[    0.973141] 0x000000000000-0x000000040000 : "u-boot"
[    0.979849] 0x000000040000-0x000000050000 : "u-boot-env"
[    0.986630] 0x000000050000-0x000000060000 : "art"
[    0.992712] 0x000000060000-0x0000001c0000 : "kernel"
[    0.999136] 0x0000001c0000-0x000001fb0000 : "rootfs"
[    1.005551] mtd: device 4 (rootfs) set to be root filesystem
[    1.011311] mtdsplit: error occured while reading from "rootfs"
[    1.017348] 0x000001fb0000-0x000002000000 : "config"
[    1.023717] 0x000000060000-0x000001fb0000 : "firmware"


As you can see, this uses a 32MB flash chip "s25fl256s1". After weeks of digging I finally found the root of the issue which is this change: https://github.com/torvalds/linux/commit/08922f644878c9163ada8df3ef9def89be1d5e90. This uses a mmap read request that "Certain SPI controllers may provide", as the author states. Apparently the s25fl256s1 does not, at least not for the whole 32MB (spi_flash_read() works if I manually set the s25fl256s1's size to 16MB in spi-nor.c)

To determine if the chip supports this mmap read, the spi_flash_read_supported(spi) function gets called (https://github.com/torvalds/linux/blob/master/include/linux/spi/spi.h#L1218) which in turn calls a spi->controller->flash_read_supported(spi) callback which for OpenWRT/LEDE got wired up by https://git.openwrt.org/?p=openwrt/openwrt.git;a=commit;h=e0d58157d8e2ec3da0a1fe62a65aa756c5514871. Unfortunately, this commit only disabled the mmap read for multi-chip setups, the issue I kept running into is not taken care of. I guess something needs to be changed in ath79_spi_flash_read_supported() in spi-ath79.c, but I am stuck at this point, hence this bug report.

BTW ath79_spi_read_flash_data() which is wired to master->spi_flash_read in spi-ath79.c returns -EOPNOTSUPP for address widths > 3, which might be the reason I am running into this with the flash size set to its native 32MB, but not with 16MB...

   

 

More information can be found at the following URL:
https://bugs.lede-project.org/index.php?do=details&task_id=1376