cgminer uses doubles for difficulty so will be unaffected

Heads up to everyone: I will be travelling from today for just over a week, and while I'll be checking in on the forums, I will be relatively unavailable and unable to do any significant debugging, fixes or updates. Everyone behave while I'm away, mkay?

Thanks for that.

This actually looks like a bug with the stratum reconnect code in cgminer since it's happening shortly after btcguild is issuing a reconnect (i.e. change address request), and is coincidental that it's happening for you with the nanofury. Unfortunately I don't have the time to debug and fix this at the moment so a workaround would be to use the redirection address you see in your logs directly instead of the standard btcguild pool address (or a different pool). I've just pushed some generic changes to git which might help this but I don't imagine it will be enough.

No I don't.

They're real crashes. Could be somehow related to being on non-pc (which is the only thing I tested on).

Both of mine perform quite differently:
Alas I'm travelling from tomorrow so I doubt I'll be able to do any debugging any time soon.

Looks like a real bug. Can you try a debug build following the directions here please:
http://ck.kolivas.org/apps/cgminer/debug/README-debug

Don't forget I always create drivers for cgminer that report hashrate based on real shares generated, not some arbitrary amount of hashes done (that doesn't translate into meaningful hashrate). I can happily make you a build that shows you making 2.5TH if you like. Make sure you check your pool to see how many effective shares you're submitting with your other drivers... People don't like the truth.

What did you build it on? I've only tried it on pc linux/windows so far.

Fixed in git, thanks.

Yes that's common as the failure is at the usb hub level, not at the device level. cgminer tries to reset devices that stop responding but what really needs to be reset is the hub. I guess I could potentially look at the parent device and try to send it a reset but then that would be the wrong thing to do if the device was at fault and not the hub. The errors we get back at the software level are not entirely helpful to distinguish the two.

Mining on a pool via the GBT protocol (and virtually everyone mines on stratum) only means that your mining software gets the summary of all the transactions that are included when you're mining. This is also why stratum was more popular because it uses a lot less bandwidth. However, as a miner, are you personally monitoring all those transactions and watching exactly what they're doing and making sure they're not doing a 51% attack? Do you have the tools and know how to investigate them in real time? Also if you're not actually mining on ghash.io then you're not even looking at what transactions they're including.

For it to protect you, you'd have to be mining there with the gbt protocol and investigate the transactions in real time, notice the transaction differences from what's going on in the bitcoin network at large, and then somehow inform the world that it was going on and then manage to convince the other 5PH of miners to stop mining there.

GBT in itself as it's currently implemented and used gives you precisely zero protection.

Sure you can run any USB device that cgminer supports off any device that you can build cgminer for (even with windows). I ran the avalon for 3 months off my regular PC rig's USB when I broke my tp link router. Open it up, pull the usb cable out and plug in a (printer type) usb cable that goes to your pc. It's actually more reliable that way too since the tplink is so woefully low power.

Will consider when I get back from holiday thanks. It's a one line change if you wish to alter it for your build in driver-bitfury.c line 417. When I almost burnt myself on one, I did think whoever called the devices I received "ice" was seriously smoking something.

No there is not, as I have seen it is virtually always best to leave it at 54, so I have added no command line option for now.

Posted the first version of cgminer (3.10.0) that supports these devices.

New version: 3.10.0, 9th January 2014

Mainly new drivers.

Human readable changelog:

- Minion driver courtesy of Kano. (More info about this from him hopefully).
- Nanofury driver. These are set up the same as every other USB device is on cgminer. Tested on both windows and linux (sorry no osx to test). Note the hashrate is once again based on only valid shares so may appear lower than other software using this device. No HW errors are currently counted (though they're most definitely there in abundance due to bitfury design). This is a driver based on all the other ones out there with a completely rewritten model to suit how cgminer drivers work.
- Hashfast driver fixes (no I still don't have one).
- Fixed BXF devices slowing down over time.


Full changelog:

- Set the mcp2210 transfer setting only when it changes.
- Buffer sizes in nanofury device data are unnecessarily large.
- Only perform spi reset on init, not with each transaction.
- Remove spi_detect_bitfury at nanofury startup and fix incorrect refresh time.
- Use a simple serialised work model for nanofury
- Use bitfury_checkresults to avoid hashing results twice in nanofury.
- Export bitfury_checkresults in libbitfury
- Pass extra parameters for later use in libbitfury_sendHashData
- Avoid double handling bswap of the nonce value in nanofury
- Avoid unnecessary rehashing in nanofury nonce checking.
- Remove the unused portions of atrvec in the nanofury driver
- Age work in nf1_scan to avoid risk of losing a work item and leaking memory.
- bitfury_work_to_payload is double handling the data unnecessarily
- Default bitrate on nanofury should be 200kHz
- localvec should be only 80 bytes not 80 words
- Wrong init value for nanofury
- Remove unused rehash values from nanofury driver.
- Only update info work in nanofury driver when it's empty.
- Fill the appropriate type of usb transfer when we know if it's an interrupt
transfer instead of a bulk one.
- Use the internal knowledge of the usb epinfo to determine whether we should be
doing an interrupt instead of a bulk transfer, and do not send a ZLP if so, and
limit read transfer to expected size automatically.
- Avoid bin2hex memleak when we start getting nanofury nonces
- Set atrvec only once and use a local array for each device's work.
- Cancel any spi transfers on nf1 close
- Add bitfury detection loop to nanofury startup
- Move spi init code to libbitfury
- Remove inappropriate extra config reg in nanofury setup.
- Status 0x30 should never happen with spi transfers.
- Fix spi transfer data size transmission mistakes.
- Minor correctness change in spi_add_data
- spi_txrx should always send and receive the same size message
- Random libbitfury changes.
- Set value of gpio pins to low on closing nanofury.
- Fix more init sequence for nanofury.
- Add basic initialisation for nf1 devices
- Add basic nf1_scan function.
- Basic import of libbitfury functions from nanofury branch
- Import functions from nanofury fork for libbitfury
- Meter out spi sends to only 2 bytes at a time, offsetting according to how
much data returns.
- Use the usb read limit function for mcp2210 reads.
- Provide a way for usb reads to just read the size asked for with a limit bool.
- Get pin value after an nf1 spi reset.
- Make sure what we send in the buffer doesn't change during spi reset for
nanofury
- Remove all standalone gpio setting change functions in mcp2210 and just use
the one global setting function.
- Set gpio values in the one function with all values for nanofury.
- Provide a helper function for setting all mcp2210 gpio settings.
- Add a helper function for getting all mcp2210 gpio settings.
- Set all pin designations and directions in one call for nanofury and don't
bother storing their values in the info struct.
- Provide helper functions for setting all pins and dirs on mcp2210
- Set all nanofury pin designations in one call
- Provide a helper function for setting all pin designations on mcp2210
- Store the spi settings in a struct for nanofury devices.
- Check the received status in mcp2210 spi transfers and repeat a zero byte send
if it's in progress.
- Set the bytes per spi transfer prior to each mcp2210 transfer.
- Separate out the send and receive functions for mcp2210 and check response
value in return.
- Check that mcp2210 spi settings have taken and check the value of the pin
during nanofury setup.
- Don't set GPIO pin designations after initial setting in nanofury since the
direction and values will be changed.
- Provide an mcp 2210 set gpio input helper function that sets a pin to gpio and
input.
- Move the set gpio output function to a generic mcp2210 version from nanofury
which also sets the pin to gpio.
- Implement a nanofury txrx with a larger buffer and cycling over data too large
to send.
- Implement magic spi reset sequence for nanofury.
- Add more spi magic to the nanofury init sequence.
- Add lots of magic spi initialisation to nanofury.
- Export reused components of bitfury management into a libbitfury and use for
bab and bitfury drivers.
- More init sequence for nanofury and implement a close function that sets all
pins to input.
- Reword offset header handling in hfa_get_header
- Sanity check in hfa_get_header
- Add more checks in hashfast driver for lost devices.
- Change spimode and send more data in nanofury setup.
- Add basic setup  comms to nanofury.
- Implement an mcp2210 spi transfer function.
- Set the initial spi settings for nanofury driver.
- Provide a helper function for gettings mcp2210 spi settings.
- Implement an mcp2210 set spi transfer settings function.
- Cancel any SPI transfers in progress in nanofury after initial setup.
- Implement an mcp2210 spi cancel function.
- Return only binary values for mcp2210 GPIO values.
- Set GPIO LED and power to high in nanofury driver.
- Implement initial part of nanofury init sequence for GPIO pin settings and add
output debugging of set values.
- Add helper functions for getting and setting mcp2210 gpio pin designations.
- Don't return an error in usb read if we've managed to get the whole read
length we've asked for.
- Use correct endpoint order for nanofury devices and read with a short timeout
on return loop from send_recv.
- Add mcp2210 helper functions for getting and setting one GPIO pin val and
direction.
- Create a generic gpio pin struct and add helpers for mcp get pin val and dirs.
- Check the receive msg of a send/receive cycle on mcp2210 matches the send
message.
- Add a set of usb commands to the usbutils defines for mcp2210 comms, and use
the same command name for send and receive.
- Create a generic mcp2210 send_rcv function.
- Include mcp header for bitfury and fix extra params in macro.
- Add basic SPI comms defines for mcp2210 and build rules for bitfury.
- Minion set some core defaults similar to final requirements
- minion compile warnings
- move driver-minion.c to main directory
- Minion with ioctl() stats, settings to attempt to emulate 21TH/s
- minion driver with results interrupt working
- tested working driver-minion.c without interrupts
- Working driver-minion.c v0.1
- driver-minion.c compilable untested
- minion driver - incomplete
- Add minion driver into cgminer
- Add basic device detection and updated udev rules for nanofury devices.
- Remove GPU from share logging example.
- Don't keep resetting BXF clockspeed to default.
- If no pools are active on startup wait 60s before trying to reconnect since we
likely have the wrong credentials rather than all the pools being out.
- Discard bad crc packets for hashfast driver instead of trying to process them.
- Update documentation for modified avalon options syntax and document relevant
55nm details.
- Modify the auto tuning sequence to work with the 50MHz changes required to
work with 55nm Avalon.
- 55nm avalon requires the delays between writes reinstated for stability.
- Use an equation instead of a lookup table to set the frequency for 55nm avalon
allowing arbitrary values to be used.
- Make the result return rate low detection on avalon less trigger happy.
- Always send the bxf device a clockspeed after parsing the temperature in case
the device has changed the clockspeed itself without notification.
- Fix BXF being inappropriately dependent on drillbit.

Seems to have gone away whatever it was, thanks.

Load on the US servers seems very high, total pool hashrate is way down and stats are reporting 0 for me despite my rigs merrily mining away on us1. Issues?

There was a minor packaging error in the 3.9.0 source tarball which would make bitfury code not fully compile support for bi*fury devices unless drillbit was enabled as well, so I've uploaded updated tarballs with only that one change, they will be tagged 3.9.0-1. There is no other change in this code so you do not need to update if everything's working for you.

I have been working on writing a nanofury driver (using direct USB communications as per our USB device driver model) but it is proving rather challenging so it appears that it will be some time before nanofury support is in mainline cgminer at this stage.