On 12/14/2012 01:37 PM, Mathias Adam wrote: > Hi Hauke, > > thanks for picking this up! > > Am Do, 13.12.2012, 20:45 schrieb Hauke Mehrtens: >> On 11/11/2012 09:38 PM, Mathias Adam wrote: >>> ping... >>> Am So, 14.10.2012, 21:13 schrieb Mathias Adam: >>>> This patch adds support for Huawei E970 wireless gateway devices. It > has been tested on an E970 labelled as T-Mobile web'n'walk Box IV. > E960/B970 should work too, from what I know it's basically the same > hardware. >>>> The device has a Broadcom BCM5354 SoC and a built-in 3G USB modem. For > reference, it has already been addressed in this open ticket: > <https://dev.openwrt.org/ticket/2711> >>>> The device has a hardware watchdog which needs GPIO-7 to be toggled at > least >>>> every 1-2 seconds. Therefore a timer is being setup early in the boot > process >>>> in order to take care of this (see arch/mips/bcm47xx/time.c). Tested > and works: 3G wan, wlan+LED, VLAN config, failsafe using reset >>>> button, image to be used for upgrade from OEM firmware's web interface >>>> Link to the wiki page I've created: >>>> <http://wiki.openwrt.org/toh/huawei/e970> >>>> Issue: >>>> * lzma-loader crashes, so gzipped kernel is used. Presumably due to > watchdog >>>> reset during kernel decompress. >>>> Signed-off-by: Mathias Adam <[email protected]> >>>> --- >>>> Changes from previous patch: >>>> * rebased to svn-rev. 33761 >>>> * b43 wireless now works out of box (no changes were necessary on my > part) >>>> * modifications according to comments by Hauke Mehrtens on 2012-04-25 > * header for factory update package now created using dd and echo, no > need >>>> for binary header template file anymore >> I just had a closer look at the patch. >> I do not like the way how this is added, but if you just have 2 seconds > time to call the gpio I do not know a better way. >> How long do you have to send the ping on the gpio after the boot? Do you > have a log for me without this patch, so that I can see if there is a > better position for this code? > > I'm not completely happy with this solution either. At least it's known to > be working stable so far (I have one of my E970 routers running for nearly > a year now without problems). > > I think it doesn't have to be that early in the boot process as it is now. > Two reasons I remember: > 1) when trying to get lzma-loader to work I wanted the GPIO toggle as > early as possible within the kernel. But as lzma-loader doesn't work yet > anyway I'd suggest to stay with gzipped kernel for the moment. > 2) AFAIR the original firmware does it in a similar way (I'll look for a > download link as you asked later) > > > Here's a log snippet from my notes without the watchdog reset: > > ... > serial8250.0: ttyS0 at MMIO 0xb8000300 (irq = 3) is a U6_16550A > serial8250.0: ttyS1 at MMIO 0xb8000400 (irq = 3) is a U6_16550A > bcm47xx_pflash: flash init: 0x1c000000 0x02000000 > Physically mapped flash: Found 1 x16 devices at 0x0 in 16-bit bank. > Manufacturer ID 0x000001 Chip ID 0x001a01 > Amd/Fujitsu Extended Query Table at 0x0040 > Amd/Fujitsu Extended Query version 1.3. > Physically mapped flash: Swapping erase regions for top-boot CFI table. > number of CFI chips: 1 > bcm47xx_pflash: Flash device: 0x2000000 at 0xbfc00000 > bcm47xx_part: bootloader size: 262144 > bcm47xx_part: Looking for dual image > bcm47xx_part: TRX offset : 0 > bcm47xx_part: Updating TRX offsets and length: > bcm47xx_part: old trx = [0x0000001c, 0x0012cc00, 0x00000000], > len=0x002e1000 crc32=0x729db7c4 > bcm47xx_part: new trx = [0x0000001c, 0x0012cc00, 0x00000000], > len=0x0012cc00 crc32=0x48f25a83 > > > CFE version 1.0.37 for BCM947XX (32bit,SP,LE) > Build Date: <C8><FD> 2<D4><C2> 13 14:50:54 CST 2008 > ([email protected]) > Copyright (C) 2000,2001,2002,2003 Broadcom Corporation. > > Initializing Arena > Initializing Devices. > Boot partition size = 262144(0x40000) > et0: Broadcom BCM47xx 10/100 Mbps Ethernet Controller > CFE reboot flag0 = 0x12345678, flag1 = 0x0,flag2 = > 0xffffffff,reboot_number = 0x1 > CPU type 0x29029: 240MHz > Total memory: 32768 KBytes > ... > > > After that I found the first 64kiB behind CFE being erased, so I guess the > watchdog reset happened during updating the TRX header. As the flash write > usually takes a rather long time, in a second test I manually modified TRX > headers (len and crc) before the first boot so that the kernel doesn't > have to update them: > > ... > bcm47xx_pflash: Flash device: 0x2000000 at 0xbfc00000 > bcm47xx_part: bootloader size: 262144 > bcm47xx_part: Looking for dual image > bcm47xx_part: TRX offset : 0 > 4 bcm47xx partitions found on MTD device Physically mapped flash > Creating 4 MTD partitions on "Physically mapped flash": > 0x000000000000-0x000000040000 : "cfe" > 0x000000040000-0x0000003f0000 : "linux" > 0x00000016cc00-0x0000003f0000 : "rootfs" > mtd: partition "rootfs" must either start or end on erase block boundary > or be smaller than an erase block -- forcing read-oy > mtd: partition "rootfs" set to be root filesystem > mtd: partition "rootfs_data" created automatically, ofs=2E0000, len=110000 > > 0x0000002e0000-0x0000003f0000 : "rootfs_data" > 0x0000003f0000-0x000000400000 : "nvram" > bcm47xx_sflash: error registering platform driver: -19 > b44: b44.c:v2.0 > b44 ssb0:0: eth0: Broadcom 44xx/47xx 10/100BaseT Ethernet > 00:90:4c:c0:85:59 > BCM47xx Watchdog Timer enabled (30 seconds, nowayout) > TCP westwood registered > NET: Registered protocol family 17 > 802.1Q VLAN Support v1.8 > VFS: Mounted root (squashfs filesystem) readonly on device 31:2. > Freeing unused kernel memory: 156k freed > > > CFE version 1.0.37 for BCM947XX (32bit,SP,LE) > Build Date: �� 2�� 13 14:50:54 CST 2008 > ([email protected]) > ... > > > > The "BCM47xx Watchdog Timer" module might be a good position, > unfortunately it's being loaded after the flash driver, so this would be > too late in case the TRX headers have to be updated... > However I remember I first had it somewhere else while developing last > year. I'll go find my notes and get back then. > > > Mathias >
After looking at your logs it does not seam to be so bad as I thought yesterday. using a gzip compressed kernel and not a lzma compressed one is not so nice but I do not know a better way. This should be a normal watchdog driver witch registers to a platform device. to get the platform device driver early enough it should be registered with arch_initcall() like it is done in drivers/watchdog/mpc8xxx_wdt.c. The platform device should be registered in bcm47xx_register_bus_complete in arch/mips/bcm47xx/setup.c, I hope this is done after the ssb bus is completely initialized. This watchdog driver would be independent from bcm47xx or ssb code and could also be used by other hardware platforms having the same problem. You could also change the code that this new watchdog driver replaces the ssb watchdog driver on these huawei devices as it is not needed any more. Hauke _______________________________________________ openwrt-devel mailing list [email protected] https://lists.openwrt.org/mailman/listinfo/openwrt-devel
