SG-5100 takes over 20 minutes to boot after eMMC failure

haraldinho

I purchased the Transcend 430S M.2 512 GB SATA III 3D NAND. Works perfectly except the boot issue. I am working with Netgate support to see if I can get the issue resolved. Will report back if I have a solution.

haraldinho

On advice of Netgate Support I did a clean reinstall with UFS (was on ZFS). The problem remains exactly the same. I feel the problem is the eMMC memory that is toast but still is getting poked by the 5100. I expected it to be disabled after installing my M.2 SSD, but it is clearly not.

Loads of these error message in the logs:

mmcsd0: Error indicated: 1 Timeout
mmcsd0: Error indicated: 1 Timeout
sdhci_pci0-slot0: Controller timeout
sdhci_pci0-slot0: ============== REGISTER DUMP ==============
sdhci_pci0-slot0: Sys addr: 0x05840000 | Version:  0x00001002
sdhci_pci0-slot0: Blk size: 0x00005200 | Blk cnt:  0x00000010
sdhci_pci0-slot0: Argument: 0x00000010 | Trn mode: 0x00000033
sdhci_pci0-slot0: Present:  0x1fff0206 | Host ctl: 0x00000025
sdhci_pci0-slot0: Power:    0x0000000b | Blk gap:  0x00000080
sdhci_pci0-slot0: Wake-up:  0x00000000 | Clock:    0x00000207
sdhci_pci0-slot0: Timeout:  0x0000000d | Int stat: 0x00000001
sdhci_pci0-slot0: Int enab: 0x01ff003b | Sig enab: 0x01ff003a
sdhci_pci0-slot0: AC12 err: 0x00000000 | Host ctl2:0x0000000c
sdhci_pci0-slot0: Caps:     0x546ec8b2 | Caps2:    0x80000007
sdhci_pci0-slot0: Max curr: 0x00000000 | ADMA err: 0x00000000
sdhci_pci0-slot0: ADMA addr:0x00000000 | Slot int: 0x00000000
sdhci_pci0-slot0: ===========================================
mmcsd0: Error indicated: 1 Timeout
mmcsd0: Error indicated: 1 Timeout
sdhci_pci0-slot0: Controller timeout
sdhci_pci0-slot0: ============== REGISTER DUMP ==============
sdhci_pci0-slot0: Sys addr: 0x05840000 | Version:  0x00001002
sdhci_pci0-slot0: Blk size: 0x00005200 | Blk cnt:  0x00000010
sdhci_pci0-slot0: Argument: 0x00000200 | Trn mode: 0x00000033
sdhci_pci0-slot0: Present:  0x1fff0206 | Host ctl: 0x00000025
sdhci_pci0-slot0: Power:    0x0000000b | Blk gap:  0x00000080
sdhci_pci0-slot0: Wake-up:  0x00000000 | Clock:    0x00000207
sdhci_pci0-slot0: Timeout:  0x0000000d | Int stat: 0x00000001
sdhci_pci0-slot0: Int enab: 0x01ff003b | Sig enab: 0x01ff003a
sdhci_pci0-slot0: AC12 err: 0x00000000 | Host ctl2:0x0000000c
sdhci_pci0-slot0: Caps:     0x546ec8b2 | Caps2:    0x80000007
sdhci_pci0-slot0: Max curr: 0x00000000 | ADMA err: 0x00000000
sdhci_pci0-slot0: ADMA addr:0x00000000 | Slot int: 0x00000000
sdhci_pci0-slot0: ===========================================
mmcsd0: Error indicated: 1 Timeout
mmcsd0: Error indicated: 1 Timeout
mmcsd0: Error indicated: 1 Timeout
mmcsd0: Error indicated: 1 Timeout
mmcsd0: Error indicated: 1 Timeout

Now I am surfing the internet to see if somebody succeeded in actually getting this memory disabled.

Harald

gabacho4

@haraldinho very unfortunate to hear indeed. So far my eMMC isn’t giving me any issues that I know of but, based on the data, it looks like I’m due to have them.l eventually. I have 2 M.2 ssd on their way to this and the remote location. Crazy that what was a 600 dollar device can be crippled by something like this. Not being able to restart a device and having to rely on someone to physically power it back on is nuts. Gonna have to start considering my options if a real fix cannot be found.

stephenw10

@haraldinho You can disable it completely in pfSense and prevent those errors.
Create the file /boot/loader.conf.local and add to it:

hint.mmcsd.0.disabled="1"

But that doesn't help with the slow boot in the BIOS.

Steve

gabacho4

Since these are Lanner devices, have they been contacted regarding BIOS updates or anything? Seems they might be the last real hope since it’s their hardware, maybe?

stephenw10

Yes, we are working with them on a BIOS update now. It isn't to address this specifically but hopefully could do so.

Steve

gabacho4

That makes me happy. Like the others, I would have assumed that whatever internal BIOS check takes place would identify, or could be told, that the eMMC wasn’t installed/working and be pointed to the M.2 SSD. That was my intended mitigation, based on that assumption, some day in the future should the internal memory ever die. Here’s to hoping they can make it happen.

Is this issue that potentially afflicts EVERY Netgate router with internal memory and the M.2 expansion capability?

hayescompatible

@stephenw10 said in SG-5100 takes over 20 minutes to boot after eMMC failure:

Yes, we are working with them on a BIOS update now. It isn't to address this specifically but hopefully could do so.

Steve

That sounds promising, I'm hopeful this problem will be addressed.

Honestly, this is what's preventing me from getting either of the 4100 or 6100 BASE models—if those don't gracefully handle failures of the eMMC either, then it's pretty much a certainty that the only way to ensure long-term reliability is to spring for the MAX versions or install an SSD yourself… in which case, there should be more prominent warnings made about the effects of an eMMC failure, how to monitor its health, more warnings in pfSense when enabling options that could potentially shorten its lifespan, etc.

haraldinho

@stephenw10 Hi Steve, it is really good to hear that there is a BIOS update in the works for the 5100. I would really appreciate if this specific issue could be taken along in the update. It does not seem to be a big thing to do and would really uncripple the devices of people with a dead eMMC. @stephenw10 is this something you can take up with the firmware folks and report back here?

stephenw10

I can certainly look into it.

haraldinho

@stephenw10 Hi Stephen, I thought to give this little hack a try and hey presto! My boot issues are gone!! I can do a warm reboot now without any issues! Speed is also good!

So a happy camper here! The only drawback to this solution is that you have to SSH into your box to do it and it will be wiped with a clean install. So it would still be best to solve it in BIOS or with some setting in pfSense where you can disable eMMC on Netgate boxes suffering from the issue.

Thanks Stephen!

gabacho4

@haraldinho what is the hack to which you refer?

EDIT: do you mean creating the file that Stephen mentioned earlier? If so, that is a great interim fix while they hopefully get something more permanent and formal put in place.

haraldinho

@gabacho4 the forum indentation is a bit confusing indeed. I replied to the post from @stephenw10 with the 'hack' and it then shows that post as a reply to Stephens post, but also as a new post at the bottom of the whole topic, with just a small reference to the fact that it is a reply to an earlier message. Anyway, the life saving configuration change is this:

Create the file /boot/loader.conf.local and add to it:
hint.mmcsd.0.disabled="1"

It is as simple as that :-)

The only thing I am not sure about is what happens to this setting when you do a regular upgrade to a new version of pfsense. @stephenw10 does it get preserved?

hayescompatible

@haraldinho said in SG-5100 takes over 20 minutes to boot after eMMC failure:

@gabacho4 the forum indentation is a bit confusing indeed. I replied to the post from @stephenw10 with the 'hack' and it then shows that post as a reply to Stephens post, but also as a new post at the bottom of the whole topic, with just a small reference to the fact that it is a reply to an earlier message. Anyway, the life saving configuration change is this:

Create the file /boot/loader.conf.local and add to it:
hint.mmcsd.0.disabled="1"

It is as simple as that :-)

The only thing I am not sure about is what happens to this setting when you do a regular upgrade to a new version of pfsense. @stephenw10 does it get preserved?

I'll have to try that later when I'm able to reboot my 5100. Sounds promising.

gabacho4

Just to make sure I understand the hack right - it’s merely telling pfSense to ignore the first of the storage options being provided by the BIOS therefore causing pfSense to turn to the M.2 right away rather than trying the eMMC. Correct? If so, this seems like something Netgate could code in as part of the pfSense config and make a toggle for in the GUI (“ignore eMMC at boot”) and we’d all happily be on our way. If coded, it would be persistent across any installation/upgrade or config restoration. Could it be that simple?

haraldinho

@gabacho4 That's how I understand what this setting does indeed. And I guess this setting could be done in pfsense, however, I feel that it is a design flaw of these Lanner boxes and should be fixed in BIOS. But I would be happy with either ;-)

gabacho4

@haraldinho fair enough point. I wonder if ALL of their devices with eMMC suffer from this. If so, the coding with GUI would make sense unless they can get BIOS updates for all the devices. If the 5100 is alone in suffering from this, the you’re right; a final fix from Lanner should be secured. Ultimately I just need a device that I can rely on 100%. I’m pretty spooked right now considering my very remote router could become (if not already) very dangerous to reboot or update. Yikes!

hayescompatible

@haraldinho said in SG-5100 takes over 20 minutes to boot after eMMC failure:

@gabacho4 That's how I understand what this setting does indeed. And I guess this setting could be done in pfsense, however, I feel that it is a design flaw of these Lanner boxes and should be fixed in BIOS. But I would be happy with either ;-)

The boot loader hint didn't work for me but I didn't think it would. My 5100 still appears to hang after I issue a reboot from the webConfigurator. The power light stays lit (good) but so do the lights on the igb0 and igb1 ports. They do this on a cold boot too but after a few seconds, they clear and then the boot process resumes. Still seems to be a BIOS issue.

FWIW, even after adding the boot loader hint that @stephenw10 suggested, pfSense/FreeBSD still sees the eMMC (via dmesg):

sdhci_pci0: <Intel Denverton eMMC 5.0 Controller> mem 0xdff9a000-0xdff9afff,0xdff99000-0xdff99fff irq 16 at device 28.0 on pci0
mmc0: <MMC/SD bus> on sdhci_pci0

Ideally there'd be a way to disable that completely in the BIOS.

haraldinho

@hayescompatible Hmm, that's a bummer. I must say I also have that same line still in my logs after the hack. So there is no difference there between our setups.

NOTE
after you made the modification, you need to power cycle your box once more to get the line loaded and effectuated. After that power cycle, warm reboots from the web configurator work again for me.

Some things you can check:

If you disable "Quiet mode" in BIOS and do a boot, do you see a difference in your logs with and without the hack?

For me all the references to timeouts like the one below went away in the logs (next to the fact I can now issue a Warm Reboot from the WebConfigurator.

mmcsd0: Error indicated: 1 Timeout
mmcsd0: Error indicated: 1 Timeout
sdhci_pci0-slot0: Controller timeout
sdhci_pci0-slot0: ============== REGISTER DUMP ==============
sdhci_pci0-slot0: Sys addr: 0x06040000 | Version:  0x00001002
sdhci_pci0-slot0: Blk size: 0x00005200 | Blk cnt:  0x00000010
sdhci_pci0-slot0: Argument: 0x00000010 | Trn mode: 0x00000033
sdhci_pci0-slot0: Present:  0x1fff0206 | Host ctl: 0x00000025
sdhci_pci0-slot0: Power:    0x0000000b | Blk gap:  0x00000080
sdhci_pci0-slot0: Wake-up:  0x00000000 | Clock:    0x00000207
sdhci_pci0-slot0: Timeout:  0x0000000d | Int stat: 0x00000001
sdhci_pci0-slot0: Int enab: 0x01ff003b | Sig enab: 0x01ff003a
sdhci_pci0-slot0: AC12 err: 0x00000000 | Host ctl2:0x0000000c
sdhci_pci0-slot0: Caps:     0x546ec8b2 | Caps2:    0x80000007
sdhci_pci0-slot0: Max curr: 0x00000000 | ADMA err: 0x00000000
sdhci_pci0-slot0: ADMA addr:0x00000000 | Slot int: 0x00000000
sdhci_pci0-slot0: ===========================================

Do you see the last line from below snippet in your logs (not sure why the characters are all repeated, but that's not the issues here)? It indicates the file is loaded during boot.

Loading /boot/defaults/loader.conf
Loading /boot/defaults/loader.conf
Loading /boot/device.hints
Loading /boot/loader.conf
HLLooaaddiinngg  //bboooott//llooaaddeerr..ccoonnff..llooccaall

if you ssh into your box, and enter the below command, does it then show the exact text of the hack?

cat /boot/loader.conf.local

It should show:

hint.mmcsd.0.disabled="1"

check character set issues
Double check the quote symbols. I spent many hours debugging 'wrong quote symbols' issues in the past (" and ' and `). Also carefully check the other spellings. If you copy-pasted the line, try deleting the file and recreating it by typing it manually.

On my box, the file did not exist. I created it using vi with this commands:

cd /boot
sudo vi loader.conf.local

I typed the hack line manually, no copy-paste. Had to google how vi works, but that is not that complicated for this simple modification. Just get vi into insert mode by pressing i, type the line, hit escape to get vi into command mode and then enter :w to save the file and then again escape followed by :q to quit. Then do a reboot. But whom am I telling, perhaps you are a Linux guru :-D

stephenw10

@haraldinho said in SG-5100 takes over 20 minutes to boot after eMMC failure:

The only thing I am not sure about is what happens to this setting when you do a regular upgrade to a new version of pfsense. @stephenw10 does it get preserved?

Yes, loader.conf.local is preserved across a firmware upgrade.

@hayescompatible said in SG-5100 takes over 20 minutes to boot after eMMC failure:

FWIW, even after adding the boot loader hint that @stephenw10 suggested, pfSense/FreeBSD still sees the eMMC (via dmesg):
sdhci_pci0: <Intel Denverton eMMC 5.0 Controller> mem 0xdff9a000-0xdff9afff,0xdff99000-0xdff99fff irq 16 at device 28.0 on pci0
mmc0: <MMC/SD bus> on sdhci_pci0

That's the mmc bus. mmcsd0 is the first mmc device on that bus which is what has been disabled.

If there was an issue with the bus or the controller you could disable those instead. Disabling only the emmc storage device itself is the minimum you need to do to stop is trying to read it.

Steve