thodg/cgminer

diff --git a/01-cgminer.rules b/01-cgminer.rules
index 972468c..733a774 100644
--- a/01-cgminer.rules
+++ b/01-cgminer.rules
@@ -30,3 +30,6 @@ ATTRS{idVendor}=="04d8", ATTRS{idProduct}=="f60a", SUBSYSTEMS=="usb", ACTION=="a
 
 #HashFast
 ATTRS{idVendor}=="297c", ATTRS{idProduct}=="0001", SUBSYSTEMS=="usb", ACTION=="add", MODE="0666", GROUP="plugdev", ENV{ID_MM_DEVICE_IGNORE}="1"
+
+# BXF
+ATTRS{idVendor}=="198c", ATTRS{idProduct}=="b1f1", SUBSYSTEMS=="usb", ACTION=="add", MODE="0666", GROUP="plugdev"
diff --git a/ASIC-README b/ASIC-README
index bf83045..a535c2b 100644
--- a/ASIC-README
+++ b/ASIC-README
@@ -3,8 +3,8 @@ SUPPORTED DEVICES
 Currently supported devices include the Avalon (including BitBurner and
 Klondike), the Butterfly Labs SC range of devices, the ASICMINER block
 erupters, the BF1 (bitfury) USB (red and blue) devices, KnCminer Mercury,
-Saturn and Jupiter devices, BlackArrow Bitfury devices and upcoming
-Hashfast devices.
+Saturn and Jupiter devices, BlackArrow Bitfury devices, Bi*fury USB devices
+and upcoming Hashfast devices.
 
 No COM ports on windows or TTY devices will be used by cgminer as it
 communicates directly with them via USB so it is normal for them to not exist or
@@ -32,7 +32,7 @@ ASICMINER block erupters will come up as AMU.
 ASICMINER devices need the --enable-icarus option when compiling cgminer.
 Also note that the AMU is managed by the Icarus driver which is detailed
 in the FPGA-README. Configuring them uses the same mechanism as outlined
-below for getting started with butterfly labs ASICs.
+below for getting started with USB ASICs.
 
 
 BlackArrow Bitfury devices
@@ -47,13 +47,14 @@ BITFURY devices
 
 Bitfury devices need the --enable-bitfury option when compiling cgminer.
 
-Currently only the BPMC BF1 devices AKA redfury/bluefury are supported and
-come up as BF1. There are no options available for them. Bitfury device are
-also set up as per the butterfly labs ASICs below.
+Currently the BPMC/BGMC BF1 devices AKA redfury/bluefury are supported and
+come up as BF1, along with the Bi*fury USB devices which come up as BXF.
+There are no options available for them. Bitfury USB devices are also set up
+as per the USB ASICs below.
 
 
-
-GETTING STARTED WITH BUTTERFLY LABS ASICS
+---
+GETTING STARTED WITH USB ASICS
 
 Unlike other software, cgminer uses direct USB communication instead of the
 ancient serial USB communication to be much faster, more reliable and use a
@@ -64,9 +65,13 @@ devices requires different drivers.
 WINDOWS:
 
 On windows, the direct USB support requires the installation of a WinUSB
-driver (NOT the ftdi_sio driver), and attach it to the Butterfly labs device.
-The easiest way to do this is to use the zadig utility which will install the
-drivers for you and then once you plug in your device you can choose the
+driver (NOT the ftdi_sio driver), and attach it to the chosen USB device.
+When configuring your device, plug it in and wait for windows to attempt to
+install a driver on its own. It may think it has succeeded or failed but wait
+for it to finish regardless. This is NOT the driver you want installed. At this
+point you need to associate your device with the WinUSB driver. The easiest
+way to do this is to use the zadig utility which you must right click on and
+run as administrator. Then once you plug in your device you can choose the
 "list all devices" from the "option" menu and you should be able to see the
 device as something like: "BitFORCE SHA256 SC". Choose the install or replace
 driver option and select WinUSB. You can either google for zadig or download
@@ -103,6 +108,18 @@ After this you can either manually restart udev and re-login, or more easily
 just reboot.
 
 
+OSX:
+
+On OSX, like Linux, no drivers need to be installed. However some devices
+like the bitfury USB sticks automatically load a driver thinking they're a
+modem and the driver needs to be unloaded for cgminer to work:
+
+sudo kextunload -b com.apple.driver.AppleUSBCDC
+sudo kextunload -b com.apple.driver.AppleUSBCDCACMData
+
+
+---
+
 ASIC SPECIFIC COMMANDS
 
 --avalon-auto       Adjust avalon overclock frequency dynamically for best hashrate
diff --git a/Makefile.am b/Makefile.am
index 05002bc..b791eba 100644
--- a/Makefile.am
+++ b/Makefile.am
@@ -69,7 +69,7 @@ cgminer_SOURCES += driver-bitforce.c
 endif
 
 if HAS_HASHFAST
-cgminer_SOURCES += driver-hashfast.c driver-hashfast.h hf_protocol.h
+cgminer_SOURCES += driver-hashfast.c driver-hashfast.h hf_protocol.h hf_protocol_be.h
 endif
 
 if HAS_BITFURY
diff --git a/NEWS b/NEWS
index a1209b7..f851e42 100644
--- a/NEWS
+++ b/NEWS
@@ -1,3 +1,129 @@
+Version 3.8.3 - 23rd November 2013
+
+- Set the bitfury device start times from when we first get valid work.
+- Fix stack corruption of zeroing too much in bf1 driver.
+- Make usb_detect return the cgpu associated with it to check if it succeeds to
+decide on whether to increment the device count or not.
+- Set tv work start time for bxf driver.
+- Age the bxf work items over 90 seconds, not the bf1 work items.
+- Zero the read buffer in _usb_read to avoid stale data and only use stack
+memory instead of using the bulkbuf since it is only used in _usb_read.
+- Leave room for temperatures above 100 degrees and pad consistently for bxf
+statline.
+- Drop json stratum auth failed message log level to verbose.
+- Change the processed value not the bufsiz in response to an end of message
+marker.
+- Don't lose data beyond the end of message in a usb read.
+- Silence irrelevant warning.
+- Only check strlen on end if end exists.
+- Simplify the end of message detection in _usb_read and allow it to return
+without doing another read if the message is already in the buffer.
+- Increase work ageing time to 90 seconds for bxf driver to account for firmware
+changes.
+- Use the age_queued_work function in the bitfury driver.
+- Provide a function to discard queued work based on age.
+- The json_val in api.c is a borrowed reference, not a new one so don't decref
+it.
+- Decrement json references in api.c to not leak memory.
+- line 2913 added urlencode
+- With reliable writes to the avalon there is no need for the sleep delays
+between writes.
+- There is no need to limit usb write transfers to maxpacketsize and it's
+harmful for large transfers on slow devices such as wrt routers.
+- Disable USB stats which were not meant to be enabled by default and add extra
+memory for a memory error when stats are enabled.
+- Set limit and count to integers to not overflow during failed hotplug attempts
+and then not trying again.
+- Update api example compilation instructions.
+
+
+Version 3.8.2 - 16th November 2013
+
+- Add more verbose documentation to the readme files for windows users.
+- Add more information on libusb failure to init telling users to check README
+file.
+- Add information on unloading cdc drivers on osx to README
+- Prevent a deadlock with use of restart_threads by spawning a thread to send
+the driver flush work messages.
+- Set priority of various threads if possible.
+- Add bxf data to api output.
+- Do not hold the mining thread lock in restart_threads when calling the driver
+flush work commands.
+- Send extra work regularly to the bxf device and parse the needwork command by
+sending the amount of work it requests.
+- Allow messages to have arbitrary offsets in the bxf parser in case we have
+lingering buffered data.
+- Send the maxroll command to the bxf driver and store the value to see if we
+need to update it.
+- Add sending of flush command to bxf on flush_work
+- Add flush and version commands to bxf start up, flush buffer and try to parse
+version response string.
+- Abstract out bxf recv message.
+- Add extra bxf commands to usbutils
+- Abstract out bxf send message to allow us to easily add extra commands.
+- Don't run device restart code if the device is not enabled.
+- Expand size of bitfury statline
+- Various driver fixes for bitfury devices, including a flag from when first
+valid work appears.
+- Look up work results in bxf driver from correct variable.
+- Correct incorrect error code in bxf driver for usb writes and add debugging.
+- Add bxf details to usbutils.
+- Implement a statline showing temperature for bxf
+- Add api data for bxf device, sharing the hashrate function with bf1.
+- Count no matching work as a hw error on bxf
+- Add BXF to udev rules.
+- Work id should be hexadecimal in bxf messages.
+- Add unrecognised string debugging to bxf driver.
+- Implement the main scanloop for bxf, trying to prevent it from ntime rolling
+work if the work protocol does not allow it.
+- Parse bxf work submits fully, submitting the results.
+- Provide a function for setting the work ntime.
+- Implement a skeleton parse bxf submit function.
+- Use the bxf read thread to set the device target and send its first work item.
+- Implement a bxf send work function and set update and restart functions to
+sending new work since that's the equivalent for that device.
+- Add temperature parsing to bxf driver
+- Create and destroy a basic bxf read thread.
+- Remove the buffer from bitfury info since it is only used on one pass in the
+bf1 device.
+- Add a rudimentary bxf detect one function.
+- Rename all bf1 specific functions in the bitfury driver, using a switch to
+choose correct function.
+- Rename bitfury_getinfo to bf1_getinfo since it's unique to bf1 devices.
+- Separate out the bf1 reset from bitfury reset.
+- Store the bitfury identity in the info struct.
+- BaB - updated tested OS comment
+- Uniquely identify the BF1 and BXF bitfury devices.
+- Remove the default libusb WinUsb pipe policies that don't suit us.
+- Only set the winusb pipe policy if it doesn't match our requirements instead
+of every transfer.
+- klondike - dont try to flush if not initialised
+- api.c trylock() add missing locklock
+- Use our new zero length packet support directly in windows.
+- Enable support for zero length packet on windows and auto clear pipe stalls.
+- util.c: Decreasing reference count on allocated JSON obects to prevent memory
+leak
+- api.c: Release apisock on error in api()
+- api.c: Release io_data->ptr when releasing io_data in io_free()
+- We can't connect to a GBT pool at all with fix protocol enabled.
+- Initialise the stgd lock mutex earlier to prevent dereferences when pool
+testing occurs before it.
+- Klondike support I2C USB layout also - as KLI
+- Return error codes in avalon_read() if they're not timeouts.
+- Break out of the avalon idle loop if we get a send error.
+- Set avalon ftdi latency to just less than the time it would take to fill the
+ftdi buffer at 115200 baud
+- Update example.conf
+- Only limit packetsize on usb out writes.
+- We must chop up every 64 bytes returned on an ftdi chip, not just the first 2
+bytes so revert to parsing the data internally in the avalon instead of using
+usbutils' simple ftdi parser.
+- Only retry 3 times in hfa_reset.
+- Only add_cgpu in hashfast driver once we have a real driver set up.
+- Clean up properly if hfa_detect_common fails in the hashfast driver.
+- --shares should be scaled to diff1 not absolute number of shares
+
+
 Version 3.8.1 - 11th November 2013
 
 - Revert "Send a zero length packet at the end of every usb transfer on windows
diff --git a/README b/README
index 3e19f78..cf2420e 100644
--- a/README
+++ b/README
@@ -235,7 +235,8 @@ See FGPA-README for more information regarding this.
 
 Cgminer should automatically find all of your Avalon ASIC, BFL ASIC, BitForce
 FPGAs, Icarus bitstream FPGAs, Klondike ASIC, ASICMINER usb block erupters,
-KnC ASICs, BaB ASICs, Hashfast ASICs and ModMiner FPGAs.
+KnC ASICs, BaB ASICs, Hashfast ASICs, ModMiner FPGAs, BPMC/BGMC BF1 USB ASICs
+and Bi*fury USB ASICs.
 
 ---
 
@@ -244,14 +245,24 @@ SETTING UP USB DEVICES
 WINDOWS:
 
 On windows, the direct USB support requires the installation of a WinUSB
-driver (NOT the ftdi_sio driver), and attach it to your devices.
-The easiest way to do this is to use the zadig utility which will install the
-drivers for you and then once you plug in your device you can choose the
+driver (NOT the ftdi_sio driver), and attach it to the chosen USB device.
+When configuring your device, plug it in and wait for windows to attempt to
+install a driver on its own. It may think it has succeeded or failed but wait
+for it to finish regardless. This is NOT the driver you want installed. At this
+point you need to associate your device with the WinUSB driver. The easiest
+way to do this is to use the zadig utility which you must right click on and
+run as administrator. Then once you plug in your device you can choose the
 "list all devices" from the "option" menu and you should be able to see the
 device as something like: "BitFORCE SHA256 SC". Choose the install or replace
 driver option and select WinUSB. You can either google for zadig or download
 it from the cgminer directory in the DOWNLOADS link above.
 
+When you first switch a device over to WinUSB with zadig and it shows that
+correctly on the left of the zadig window, but it still gives permission
+errors, you may need to unplug the USB miner and then plug it back in. Some
+users may need to reboot at this point.
+
+
 LINUX:
 
 On linux, the direct USB support requires no drivers at all. However due to
@@ -276,6 +287,19 @@ directory with the following command:
 After this you can either manually restart udev and re-login, or more easily
 just reboot.
 
+
+OSX:
+
+On OSX, like Linux, no drivers need to be installed. However some devices
+like the bitfury USB sticks automatically load a driver thinking they're a
+modem and the driver needs to be unloaded for cgminer to work:
+
+sudo kextunload -b com.apple.driver.AppleUSBCDC
+sudo kextunload -b com.apple.driver.AppleUSBCDCACMData
+
+
+---
+
 Advanced USB options:
 
 The --usb option can restrict how many Avalon, BFL ASIC, BitForce FPGAs,
diff --git a/api-example.c b/api-example.c
index d14b6aa..26d5a0e 100644
--- a/api-example.c
+++ b/api-example.c
@@ -8,7 +8,7 @@
  */
 
 /* Compile:
- *   gcc api-example.c -Icompat/jansson -Icompat/libusb-1.0/libusb -o cgminer-api
+ *   gcc api-example.c -Icompat/jansson-2.5 -Icompat/libusb-1.0/libusb -o cgminer-api
  */
 
 #include "config.h"
diff --git a/api.c b/api.c
index c811c61..bbc8e17 100644
--- a/api.c
+++ b/api.c
@@ -1626,10 +1626,14 @@ uint64_t api_trylock(void *lock, const char *file, const char *func, const int l
 	LOCKINFO *info;
 	uint64_t id;
 
+	locklock();
+
 	info = findlock(lock, CGLOCK_UNKNOWN, file, func, linenum);
 	id = lock_id++;
 	addgettry(info, id, file, func, linenum, false);
 
+	lockunlock();
+
 	return id;
 }
 
@@ -4378,7 +4382,7 @@ void api(int api_thr_id)
 	struct sockaddr_in cli;
 	socklen_t clisiz;
 	char cmdbuf[100];
-	char *cmd;
+	char *cmd = NULL;
 	char *param;
 	bool addrok;
 	char group;
@@ -4586,6 +4590,7 @@ void api(int api_thr_id)
 								}
 							}
 						}
+						json_decref(json_config);
 					}
 				}
 
diff --git a/cgminer.c b/cgminer.c
index 3634234..65ca941 100644
--- a/cgminer.c
+++ b/cgminer.c
@@ -2484,7 +2484,7 @@ share_result(json_t *val, json_t *res, json_t *err, const struct work *work,
 				       hashshow, cgpu->drv->name, cgpu->device_id, resubmit ? "(resubmit)" : "", worktime);
 		}
 		sharelog("accept", work);
-		if (opt_shares && total_accepted >= opt_shares) {
+		if (opt_shares && total_diff_accepted >= opt_shares) {
 			applog(LOG_WARNING, "Successfully mined %d accepted shares as requested and exiting.", opt_shares);
 			kill_work();
 			return;
@@ -3529,6 +3529,17 @@ static void _copy_work(struct work *work, const struct work *base_work, int noff
 		work->coinbase = strdup(base_work->coinbase);
 }
 
+void set_work_ntime(struct work *work, int ntime)
+{
+	uint32_t *work_ntime = (uint32_t *)(work->data + 68);
+
+	*work_ntime = htobe32(ntime);
+	if (work->ntime) {
+		free(work->ntime);
+		work->ntime = bin2hex((unsigned char *)work_ntime, 4);
+	}
+}
+
 /* Generates a copy of an existing work struct, creating fresh heap allocations
  * for all dynamically allocated arrays within the struct. noffset is used for
  * when a driver has internally rolled the ntime, noffset is a relative value.
@@ -3851,11 +3862,13 @@ int restart_wait(struct thr_info *thr, unsigned int mstime)
 	
 static void flush_queue(struct cgpu_info *cgpu);
 
-static void restart_threads(void)
+static void *restart_thread(void __maybe_unused *arg)
 {
 	struct pool *cp = current_pool();
 	struct cgpu_info *cgpu;
-	int i;
+	int i, mt;
+
+	pthread_detach(pthread_self());
 
 	/* Artificially set the lagging flag to avoid pool not providing work
 	 * fast enough  messages after every long poll */
@@ -3865,15 +3878,19 @@ static void restart_threads(void)
 	discard_stale();
 
 	rd_lock(&mining_thr_lock);
-	for (i = 0; i < mining_threads; i++) {
+	mt = mining_threads;
+	rd_unlock(&mining_thr_lock);
+
+	for (i = 0; i < mt; i++) {
 		cgpu = mining_thr[i]->cgpu;
 		if (unlikely(!cgpu))
 			continue;
+		if (cgpu->deven != DEV_ENABLED)
+			continue;
 		mining_thr[i]->work_restart = true;
 		flush_queue(cgpu);
 		cgpu->drv->flush_work(cgpu);
 	}
-	rd_unlock(&mining_thr_lock);
 
 	mutex_lock(&restart_lock);
 	pthread_cond_broadcast(&restart_cond);
@@ -3885,6 +3902,17 @@ static void restart_threads(void)
 	 * early. */
 	cancel_usb_transfers();
 #endif
+	return NULL;
+}
+
+/* In order to prevent a deadlock via the various drv->flush_work
+ * implementations we send the restart messages via a separate thread. */
+static void restart_threads(void)
+{
+	pthread_t rthread;
+
+	if (unlikely(pthread_create(&rthread, NULL, restart_thread, NULL)))
+		quit(1, "Failed to create restart thread");
 }
 
 static void signal_work_update(void)
@@ -4436,6 +4464,30 @@ void zero_stats(void)
 	}
 }
 
+static void set_highprio(void)
+{
+#ifndef WIN32
+	int ret = nice(-10);
+
+	if (!ret)
+		applog(LOG_DEBUG, "Unable to set thread to high priority");
+#else
+	SetThreadPriority(GetCurrentThread(), THREAD_PRIORITY_HIGHEST);
+#endif
+}
+
+static void set_lowprio(void)
+{
+#ifndef WIN32
+	int ret = nice(10);
+
+	if (!ret)
+		applog(LOG_INFO, "Unable to set thread to low priority");
+#else
+	SetThreadPriority(GetCurrentThread(), THREAD_PRIORITY_LOWEST);
+#endif
+}
+
 #ifdef HAVE_CURSES
 static void display_pools(void)
 {
@@ -4841,6 +4893,7 @@ static void *api_thread(void *userdata)
 
 	RenameThread("api");
 
+	set_lowprio();
 	api(api_thr_id);
 
 	PTH(mythr) = 0L;
@@ -5503,7 +5556,7 @@ retry_stratum:
 	}
 
 	/* Probe for GBT support on first pass */
-	if (!pool->probed && !opt_fix_protocol) {
+	if (!pool->probed) {
 		applog(LOG_DEBUG, "Probing for GBT support");
 		val = json_rpc_call(curl, pool->rpc_url, pool->rpc_userpass,
 				    gbt_req, true, false, &rolltime, pool, false);
@@ -6330,6 +6383,30 @@ void __work_completed(struct cgpu_info *cgpu, struct work *work)
 	cgpu->queued_count--;
 	HASH_DEL(cgpu->queued_work, work);
 }
+
+/* This iterates over a queued hashlist finding work started more than secs
+ * seconds ago and discards the work as completed. The driver must set the
+ * work->tv_work_start value appropriately. Returns the number of items aged. */
+int age_queued_work(struct cgpu_info *cgpu, double secs)
+{
+	struct work *work, *tmp;
+	struct timeval tv_now;
+	int aged = 0;
+
+	cgtime(&tv_now);
+
+	wr_lock(&cgpu->qlock);
+	HASH_ITER(hh, cgpu->queued_work, work, tmp) {
+		if (tdiff(&tv_now, &work->tv_work_start) > secs) {
+			__work_completed(cgpu, work);
+			aged++;
+		}
+	}
+	wr_unlock(&cgpu->qlock);
+
+	return aged;
+}
+
 /* This function should be used by queued device drivers when they're sure
  * the work struct is no longer in use. */
 void work_completed(struct cgpu_info *cgpu, struct work *work)
@@ -6501,6 +6578,7 @@ void *miner_thread(void *userdata)
 	applog(LOG_DEBUG, "Waiting on sem in miner thread");
 	cgsem_wait(&mythr->sem);
 
+	set_highprio();
 	drv->hash_work(mythr);
 out:
 	drv->thread_shutdown(mythr);
@@ -6779,6 +6857,8 @@ static void *watchpool_thread(void __maybe_unused *userdata)
 
 	RenameThread("watchpool");
 
+	set_lowprio();
+
 	while (42) {
 		struct timeval now;
 		int i;
@@ -6862,6 +6942,7 @@ static void *watchdog_thread(void __maybe_unused *userdata)
 
 	RenameThread("watchdog");
 
+	set_lowprio();
 	memset(&zero_tv, 0, sizeof(struct timeval));
 	cgtime(&rotate_tv);
 
@@ -7112,9 +7193,9 @@ void print_summary(void)
 	}
 
 	if (opt_shares) {
-		applog(LOG_WARNING, "Mined %d accepted shares of %d requested\n", total_accepted, opt_shares);
-		if (opt_shares > total_accepted)
-			applog(LOG_WARNING, "WARNING - Mined only %d shares of %d requested.", total_accepted, opt_shares);
+		applog(LOG_WARNING, "Mined %.0f accepted shares of %d requested\n", total_diff_accepted, opt_shares);
+		if (opt_shares > total_diff_accepted)
+			applog(LOG_WARNING, "WARNING - Mined only %.0f shares of %d requested.", total_diff_accepted, opt_shares);
 	}
 	applog(LOG_WARNING, " ");
 
@@ -7649,6 +7730,8 @@ static void *hotplug_thread(void __maybe_unused *userdata)
 
 	RenameThread("hotplug");
 
+	set_lowprio();
+
 	hotplug_mode = true;
 
 	cgsleep_ms(5000);
@@ -7780,6 +7863,13 @@ int main(int argc, char *argv[])
 	if (unlikely(pthread_cond_init(&gws_cond, NULL)))
 		quit(1, "Failed to pthread_cond_init gws_cond");
 
+	/* Create a unique get work queue */
+	getq = tq_new();
+	if (!getq)
+		quit(1, "Failed to create getq");
+	/* We use the getq mutex as the staged lock */
+	stgd_lock = &getq->mutex;
+
 	initialise_usb();
 
 	snprintf(packagename, sizeof(packagename), "%s %s", PACKAGE, VERSION);
@@ -8006,13 +8096,6 @@ int main(int argc, char *argv[])
 			quit(1, "Failed to calloc mining_thr[%d]", i);
 	}
 
-	/* Create a unique get work queue */
-	getq = tq_new();
-	if (!getq)
-		quit(1, "Failed to create getq");
-	/* We use the getq mutex as the staged lock */
-	stgd_lock = &getq->mutex;
-
 	if (opt_benchmark)
 		goto begin_bench;
 
diff --git a/compat/libusb-1.0/libusb/os/windows_usb.c b/compat/libusb-1.0/libusb/os/windows_usb.c
index afc4dfc..ac359c4 100644
--- a/compat/libusb-1.0/libusb/os/windows_usb.c
+++ b/compat/libusb-1.0/libusb/os/windows_usb.c
@@ -1822,9 +1822,6 @@ static int windows_submit_transfer(struct usbi_transfer *itransfer)
 		return submit_control_transfer(itransfer);
 	case LIBUSB_TRANSFER_TYPE_BULK:
 	case LIBUSB_TRANSFER_TYPE_INTERRUPT:
-		if (IS_XFEROUT(transfer) &&
-		    transfer->flags & LIBUSB_TRANSFER_ADD_ZERO_PACKET)
-			return LIBUSB_ERROR_NOT_SUPPORTED;
 		return submit_bulk_transfer(itransfer);
 	case LIBUSB_TRANSFER_TYPE_ISOCHRONOUS:
 		return submit_iso_transfer(itransfer);
@@ -2391,20 +2388,8 @@ static int winusb_configure_endpoints(struct libusb_device_handle *dev_handle, i
 			PIPE_TRANSFER_TIMEOUT, sizeof(ULONG), &timeout)) {
 			usbi_dbg("failed to set PIPE_TRANSFER_TIMEOUT for control endpoint %02X", endpoint_address);
 		}
-		if (i == -1) continue;	// Other policies don't apply to control endpoint
-		policy = false;
-		if (!WinUsb_SetPipePolicy(winusb_handle, endpoint_address,
-			SHORT_PACKET_TERMINATE, sizeof(UCHAR), &policy)) {
-			usbi_dbg("failed to disable SHORT_PACKET_TERMINATE for endpoint %02X", endpoint_address);
-		}
-		if (!WinUsb_SetPipePolicy(winusb_handle, endpoint_address,
-			IGNORE_SHORT_PACKETS, sizeof(UCHAR), &policy)) {
-			usbi_dbg("failed to disable IGNORE_SHORT_PACKETS for endpoint %02X", endpoint_address);
-		}
-		if (!WinUsb_SetPipePolicy(winusb_handle, endpoint_address,
-			ALLOW_PARTIAL_READS, sizeof(UCHAR), &policy)) {
-			usbi_dbg("failed to disable ALLOW_PARTIAL_READS for endpoint %02X", endpoint_address);
-		}
+		if (i == -1)
+			continue;	// Other policies don't apply to control endpoint
 		policy = true;
 		if (!WinUsb_SetPipePolicy(winusb_handle, endpoint_address,
 			AUTO_CLEAR_STALL, sizeof(UCHAR), &policy)) {
@@ -2649,6 +2634,8 @@ static int winusb_submit_bulk_transfer(struct usbi_transfer *itransfer)
 	bool ret;
 	int current_interface;
 	struct winfd wfd;
+	ULONG ppolicy = sizeof(UCHAR);
+	UCHAR policy;
 
 	CHECK_WINUSB_AVAILABLE;
 
@@ -2671,9 +2658,20 @@ static int winusb_submit_bulk_transfer(struct usbi_transfer *itransfer)
 	}
 
 	if (IS_XFERIN(transfer)) {
-		usbi_dbg("reading %d bytes", transfer->length);
+		WinUsb_GetPipePolicy(wfd.handle, transfer->endpoint, AUTO_CLEAR_STALL, &ppolicy, &policy);
+		if (!policy) {
+			policy = TRUE;
+			WinUsb_SetPipePolicy(wfd.handle, transfer->endpoint, AUTO_CLEAR_STALL, ppolicy, &policy);
+		}
 		ret = WinUsb_ReadPipe(wfd.handle, transfer->endpoint, transfer->buffer, transfer->length, NULL, wfd.overlapped);
 	} else {
+		if (transfer->flags & LIBUSB_TRANSFER_ADD_ZERO_PACKET) {
+			WinUsb_GetPipePolicy(wfd.handle, transfer->endpoint, SHORT_PACKET_TERMINATE, &ppolicy, &policy);
+			if (!policy) {
+				policy = TRUE;
+				WinUsb_SetPipePolicy(wfd.handle, transfer->endpoint, SHORT_PACKET_TERMINATE, ppolicy, &policy);
+			}
+		}
 		usbi_dbg("writing %d bytes", transfer->length);
 		ret = WinUsb_WritePipe(wfd.handle, transfer->endpoint, transfer->buffer, transfer->length, NULL, wfd.overlapped);
 	}
diff --git a/configure.ac b/configure.ac
index 4caa4e7..10b0eea 100644
--- a/configure.ac
+++ b/configure.ac
@@ -2,7 +2,7 @@
 ##--##--##--##--##--##--##--##--##--##--##--##--##--##--##--##--##
 m4_define([v_maj], [3])
 m4_define([v_min], [8])
-m4_define([v_mic], [1])
+m4_define([v_mic], [3])
 ##--##--##--##--##--##--##--##--##--##--##--##--##--##--##--##--##
 m4_define([v_ver], [v_maj.v_min.v_mic])
 m4_define([lt_rev], m4_eval(v_maj + v_min))
diff --git a/driver-avalon.c b/driver-avalon.c
index 8a892d8..aaabe55 100644
--- a/driver-avalon.c
+++ b/driver-avalon.c
@@ -142,12 +142,10 @@ static int avalon_write(struct cgpu_info *avalon, char *buf, ssize_t len, int ep
 	return AVA_SEND_OK;
 }
 
-static int avalon_send_task(const struct avalon_task *at, struct cgpu_info *avalon,
-			    struct avalon_info *info)
-
+static int avalon_send_task(const struct avalon_task *at, struct cgpu_info *avalon)
 {
 	uint8_t buf[AVALON_WRITE_SIZE + 4 * AVALON_DEFAULT_ASIC_NUM];
-	int delay, ret, i, ep = C_AVALON_TASK;
+	int ret, i, ep = C_AVALON_TASK;
 	uint32_t nonce_range;
 	size_t nr_len;
 
@@ -188,10 +186,6 @@ static int avalon_send_task(const struct avalon_task *at, struct cgpu_info *aval
 	tt |= ((buf[4] & 0x80) ? (1 << 0) : 0);
 	buf[4] = tt;
 #endif
-	delay = nr_len * 10 * 1000000;
-	delay = delay / info->baud;
-	delay += 4000;
-
 	if (at->reset) {
 		ep = C_AVALON_RESET;
 		nr_len = 1;
@@ -200,15 +194,8 @@ static int avalon_send_task(const struct avalon_task *at, struct cgpu_info *aval
 		applog(LOG_DEBUG, "Avalon: Sent(%u):", (unsigned int)nr_len);
 		hexdump(buf, nr_len);
 	}
-	/* Sleep from the last time we sent data */
-	cgsleep_us_r(&info->cgsent, info->send_delay);
-
-	cgsleep_prepare_r(&info->cgsent);
 	ret = avalon_write(avalon, (char *)buf, nr_len, ep);
 
-	applog(LOG_DEBUG, "Avalon: Sent: Buffer delay: %dus", info->send_delay);
-	info->send_delay = delay;
-
 	return ret;
 }
 
@@ -291,14 +278,31 @@ static inline bool avalon_cts(char c)
 
 static int avalon_read(struct cgpu_info *avalon, char *buf, size_t bufsize, int ep)
 {
-	int err, amount;
+	size_t total = 0, readsize = bufsize + 2;
+	char readbuf[AVALON_READBUF_SIZE];
+	int err, amount, ofs = 2, cp;
 
-	err = usb_read_once(avalon, buf, bufsize, &amount, ep);
+	err = usb_read_once(avalon, readbuf, readsize, &amount, ep);
 	applog(LOG_DEBUG, "%s%i: Get avalon read got err %d",
 	       avalon->drv->name, avalon->device_id, err);
-	if (unlikely(err && err != LIBUSB_ERROR_TIMEOUT))
-		amount = -1;
-	return amount;
+	if (err && err != LIBUSB_ERROR_TIMEOUT)
+		return err;
+
+	if (amount < 2)
+		goto out;
+
+	/* The first 2 of every 64 bytes are status on FTDIRL */
+	while (amount > 2) {
+		cp = amount - 2;
+		if (cp > 62)
+			cp = 62;
+		memcpy(&buf[total], &readbuf[ofs], cp);
+		total += cp;
+		amount -= cp + 2;
+		ofs += 64;
+	}
+out:
+	return total;
 }
 
 static int avalon_reset(struct cgpu_info *avalon, bool initial)
@@ -320,7 +324,7 @@ static int avalon_reset(struct cgpu_info *avalon, bool initial)
 			 AVALON_DEFAULT_FREQUENCY);
 
 	wait_avalon_ready(avalon);
-	ret = avalon_send_task(&at, avalon, info);
+	ret = avalon_send_task(&at, avalon);
 	if (unlikely(ret == AVA_SEND_ERROR))
 		return -1;
 
@@ -561,7 +565,8 @@ static void avalon_idle(struct cgpu_info *avalon, struct avalon_info *info)
 		avalon_init_task(&at, 0, 0, info->fan_pwm, info->timeout,
 				 info->asic_count, info->miner_count, 1, 1,
 				 info->frequency);
-		avalon_send_task(&at, avalon, info);
+		if (avalon_send_task(&at, avalon) == AVA_SEND_ERROR)
+			break;
 	}
 	applog(LOG_WARNING, "%s%i: Idling %d miners", avalon->drv->name, avalon->device_id, i);
 	wait_avalon_ready(avalon);
@@ -736,7 +741,7 @@ static void bitburner_get_version(struct cgpu_info *avalon)
 	}
 }
 
-static bool avalon_detect_one(libusb_device *dev, struct usb_find_devices *found)
+static struct cgpu_info *avalon_detect_one(libusb_device *dev, struct usb_find_devices *found)
 {
 	int baud, miner_count, asic_count, timeout, frequency;
 	int this_option_offset;
@@ -761,7 +766,9 @@ static bool avalon_detect_one(libusb_device *dev, struct usb_find_devices *found
 				 &asic_count, &timeout, &frequency,
 				 (usb_ident(avalon) == IDENT_BBF && opt_bitburner_fury_options != NULL) ? opt_bitburner_fury_options : opt_avalon_options);
 
-	avalon->usbdev->usb_type = USB_TYPE_FTDI;
+	/* Even though this is an FTDI type chip, we want to do the parsing
+	 * all ourselves so set it to std usb type */
+	avalon->usbdev->usb_type = USB_TYPE_STD;
 
 	/* We have a real Avalon! */
 	avalon_initialise(avalon);
@@ -844,7 +851,7 @@ static bool avalon_detect_one(libusb_device *dev, struct usb_find_devices *found
 		bitburner_get_version(avalon);
 	}
 
-	return true;
+	return avalon;
 
 unshin:
 
@@ -857,7 +864,7 @@ shin:
 
 	avalon = usb_free_cgpu(avalon);
 
-	return false;
+	return NULL;
 }
 
 static void avalon_detect(bool __maybe_unused hotplug)
@@ -1152,7 +1159,7 @@ static void *avalon_send_tasks(void *userdata)
 			}
 			mutex_unlock(&info->qlock);
 
-			ret = avalon_send_task(&at, avalon, info);
+			ret = avalon_send_task(&at, avalon);
 
 			if (unlikely(ret == AVA_SEND_ERROR)) {
 				/* Send errors are fatal */
diff --git a/driver-avalon.h b/driver-avalon.h
index 0a11161..5b6cf83 100644
--- a/driver-avalon.h
+++ b/driver-avalon.h
@@ -65,7 +65,8 @@
 
 #define AVALON_FTDI_READSIZE 510
 #define AVALON_READBUF_SIZE 8192
-#define AVALON_LATENCY 16 /* This is basically the ftdi default anyway */
+/* Set latency to just less than full 64 byte packet size at 115200 baud */
+#define AVALON_LATENCY 4
 
 struct avalon_task {
 	uint8_t reset		:1;
@@ -144,8 +145,6 @@ struct avalon_info {
 	pthread_mutex_t lock;
 	pthread_mutex_t qlock;
 	cgsem_t qsem;
-	cgtimer_t cgsent;
-	int send_delay;
 
 	int nonces;
 	int auto_queued;
diff --git a/driver-bab.c b/driver-bab.c
index 545e8bb..7db3df1 100644
--- a/driver-bab.c
+++ b/driver-bab.c
@@ -17,7 +17,8 @@
 #include "sha2.h"
 
 /*
- * Tested on RPi running Raspbian with BlackArrow BitFury V1 16 chip GPIO board
+ * Tested on RPi running both Raspbian and Arch
+ *  with BlackArrow BitFury V1 16 chip GPIO board
  */
 
 #ifndef LINUX
diff --git a/driver-bflsc.c b/driver-bflsc.c
index f3e45c4..5dafa9c 100644
--- a/driver-bflsc.c
+++ b/driver-bflsc.c
@@ -688,7 +688,7 @@ ne:
 	return ok;
 }
 
-static bool bflsc_detect_one(struct libusb_device *dev, struct usb_find_devices *found)
+static struct cgpu_info *bflsc_detect_one(struct libusb_device *dev, struct usb_find_devices *found)
 {
 	struct bflsc_info *sc_info = NULL;
 	char buf[BFLSC_BUFSIZ+1];
@@ -888,7 +888,7 @@ reinit:
 	mutex_init(&bflsc->device_mutex);
 	rwlock_init(&sc_info->stat_lock);
 
-	return true;
+	return bflsc;
 
 unshin:
 
@@ -906,7 +906,7 @@ shin:
 
 	bflsc = usb_free_cgpu(bflsc);
 
-	return false;
+	return NULL;
 }
 
 static void bflsc_detect(bool __maybe_unused hotplug)
diff --git a/driver-bitforce.c b/driver-bitforce.c
index 901ba08..d4f51c1 100644
--- a/driver-bitforce.c
+++ b/driver-bitforce.c
@@ -162,7 +162,7 @@ failed:
 		mutex_unlock(&bitforce->device_mutex);
 }
 
-static bool bitforce_detect_one(struct libusb_device *dev, struct usb_find_devices *found)
+static struct cgpu_info *bitforce_detect_one(struct libusb_device *dev, struct usb_find_devices *found)
 {
 	char buf[BITFORCE_BUFSIZ+1];
 	int err, amount;
@@ -270,7 +270,7 @@ reinit:
 
 	mutex_init(&bitforce->device_mutex);
 
-	return true;
+	return bitforce;
 
 unshin:
 
@@ -285,7 +285,7 @@ shin:
 
 	bitforce = usb_free_cgpu(bitforce);
 
-	return false;
+	return NULL;
 }
 
 static void bitforce_detect(bool __maybe_unused hotplug)
diff --git a/driver-bitfury.c b/driver-bitfury.c
index 3331d82..7181488 100644
--- a/driver-bitfury.c
+++ b/driver-bitfury.c
@@ -18,7 +18,7 @@
 #define BF1MSGSIZE 7
 #define BF1INFOSIZE 14
 
-static void bitfury_empty_buffer(struct cgpu_info *bitfury)
+static void bf1_empty_buffer(struct cgpu_info *bitfury)
 {
 	char buf[512];
 	int amount;
@@ -28,12 +28,12 @@ static void bitfury_empty_buffer(struct cgpu_info *bitfury)
 	} while (amount);
 }
 
-static int bitfury_open(struct cgpu_info *bitfury)
+static bool bf1_open(struct cgpu_info *bitfury)
 {
 	uint32_t buf[2];
 	int err;
 
-	bitfury_empty_buffer(bitfury);
+	bf1_empty_buffer(bitfury);
 	/* Magic sequence to reset device only really needed for windows but
 	 * harmless on linux. */
 	buf[0] = 0x80250000;
@@ -53,19 +53,33 @@ static int bitfury_open(struct cgpu_info *bitfury)
 	return (err == BF1MSGSIZE);
 }
 
-static void bitfury_close(struct cgpu_info *bitfury)
+static void bf1_close(struct cgpu_info *bitfury)
 {
-	bitfury_empty_buffer(bitfury);
+	bf1_empty_buffer(bitfury);
 }
 
-static void bitfury_identify(struct cgpu_info *bitfury)
+static void bf1_identify(struct cgpu_info *bitfury)
 {
 	int amount;
 
 	usb_write(bitfury, "L", 1, &amount, C_BF1_IDENTIFY);
 }
 
-static bool bitfury_getinfo(struct cgpu_info *bitfury, struct bitfury_info *info)
+static void bitfury_identify(struct cgpu_info *bitfury)
+{
+	struct bitfury_info *info = bitfury->device_data;
+
+	switch(info->ident) {
+		case IDENT_BF1:
+			bf1_identify(bitfury);
+			break;
+		case IDENT_BXF:
+		default:
+			break;
+	}
+}
+
+static bool bf1_getinfo(struct cgpu_info *bitfury, struct bitfury_info *info)
 {
 	int amount, err;
 	char buf[16];
@@ -93,11 +107,11 @@ static bool bitfury_getinfo(struct cgpu_info *bitfury, struct bitfury_info *info
 
 	applog(LOG_INFO, "%s %d: Getinfo returned version %d, product %s serial %08x", bitfury->drv->name,
 	       bitfury->device_id, info->version, info->product, info->serial);
-	bitfury_empty_buffer(bitfury);
+	bf1_empty_buffer(bitfury);
 	return true;
 }
 
-static bool bitfury_reset(struct cgpu_info *bitfury)
+static bool bf1_reset(struct cgpu_info *bitfury)
 {
 	int amount, err;
 	char buf[16];
@@ -122,64 +136,352 @@ static bool bitfury_reset(struct cgpu_info *bitfury)
 	}
 	applog(LOG_DEBUG, "%s %d: Getreset returned %s", bitfury->drv->name,
 	       bitfury->device_id, buf);
-	bitfury_empty_buffer(bitfury);
+	bf1_empty_buffer(bitfury);
 	return true;
 }
 
-static bool bitfury_detect_one(struct libusb_device *dev, struct usb_find_devices *found)
+static bool bxf_send_msg(struct cgpu_info *bitfury, char *buf, enum usb_cmds cmd)
 {
-	struct cgpu_info *bitfury;
-	struct bitfury_info *info;
+	int err, amount, len;
+
+	len = strlen(buf);
+	applog(LOG_DEBUG, "%s %d: Sending %s", bitfury->drv->name, bitfury->device_id, buf);
+	err = usb_write(bitfury, buf, len, &amount, cmd);
+	if (err || amount != len) {
+		applog(LOG_WARNING, "%s %d: Error %d sending %s sent %d of %d", bitfury->drv->name,
+		       bitfury->device_id, err, usb_cmdname(cmd), amount, len);
+		return false;
+	}
+	return true;
+}
 
-	bitfury = usb_alloc_cgpu(&bitfury_drv, 1);
+/* Returns the amount received only if we receive a full message, otherwise
+ * it returns the err value. */
+static int bxf_recv_msg(struct cgpu_info *bitfury, char *buf)
+{
+	int err, amount;
 
-	if (!usb_init(bitfury, dev, found))
-		goto out;
-	applog(LOG_INFO, "%s %d: Found at %s", bitfury->drv->name,
-	       bitfury->device_id, bitfury->device_path);
+	err = usb_read_nl(bitfury, buf, 512, &amount, C_BXF_READ);
+	if (amount)
+		applog(LOG_DEBUG, "%s %d: Received %s", bitfury->drv->name, bitfury->device_id, buf);
+	if (!err)
+		return amount;
+	return err;
+}
+
+/* Keep reading till the first timeout or error */
+static void bxf_clear_buffer(struct cgpu_info *bitfury)
+{
+	int err, retries = 0;
+	char buf[512];
+
+	do {
+		err = bxf_recv_msg(bitfury, buf);
+		usb_buffer_clear(bitfury);
+		if (err < 0)
+			break;
+	} while (retries++ < 10);
+}
+
+static bool bxf_send_flush(struct cgpu_info *bitfury)
+{
+	char buf[8];
+
+	sprintf(buf, "flush\n");
+	return bxf_send_msg(bitfury, buf, C_BXF_FLUSH);
+}
+
+static bool bxf_detect_one(struct cgpu_info *bitfury, struct bitfury_info *info)
+{
+	int err, retries = 0;
+	char buf[512];
+
+	if (!bxf_send_flush(bitfury))
+		return false;
+
+	bxf_clear_buffer(bitfury);
+
+	sprintf(buf, "version\n");
+	if (!bxf_send_msg(bitfury, buf, C_BXF_VERSION))
+		return false;
+
+	do {
+		err = bxf_recv_msg(bitfury, buf);
+		if (err < 0 && err != LIBUSB_ERROR_TIMEOUT)
+			return false;
+		if (err > 0 && !strncmp(buf, "version", 7)) {
+			sscanf(&buf[8], "%d.%d rev %d chips %d", &info->ver_major,
+			       &info->ver_minor, &info->hw_rev, &info->chips);
+			applog(LOG_INFO, "%s %d: Version %d.%d rev %d chips %d",
+			       bitfury->drv->name, bitfury->device_id, info->ver_major,
+			       info->ver_minor, info->hw_rev, info->chips);
+			break;
+		}
+		/* Keep parsing if the buffer is full without counting it as
+		 * a retry. */
+		if (usb_buffer_size(bitfury))
+			continue;
+	} while (retries++ < 10);
+
+	if (!add_cgpu(bitfury))
+		quit(1, "Failed to add_cgpu in bxf_detect_one");
+
+	update_usb_stats(bitfury);
+	applog(LOG_INFO, "%s %d: Successfully initialised %s",
+	       bitfury->drv->name, bitfury->device_id, bitfury->device_path);
 
-	info = calloc(sizeof(struct bitfury_info), 1);
-	if (!info)
-		quit(1, "Failed to calloc info in bitfury_detect_one");
-	bitfury->device_data = info;
-	/* This does not artificially raise hashrate, it simply allows the
-	 * hashrate to adapt quickly on starting. */
 	info->total_nonces = 1;
+	/* This unsets it to make sure it gets set on the first pass */
+	info->maxroll = -1;
 
-	if (!bitfury_open(bitfury))
+	return true;
+}
+
+static bool bf1_detect_one(struct cgpu_info *bitfury, struct bitfury_info *info)
+{
+	if (!bf1_open(bitfury))
 		goto out_close;
 
 	/* Send getinfo request */
-	if (!bitfury_getinfo(bitfury, info))
+	if (!bf1_getinfo(bitfury, info))
 		goto out_close;
 
 	/* Send reset request */
-	if (!bitfury_reset(bitfury))
+	if (!bf1_reset(bitfury))
 		goto out_close;
 
-	bitfury_identify(bitfury);
-	bitfury_empty_buffer(bitfury);
+	bf1_identify(bitfury);
+	bf1_empty_buffer(bitfury);
 
 	if (!add_cgpu(bitfury))
-		quit(1, "Failed to add_cgpu in bitfury_detect_one");
+		quit(1, "Failed to add_cgpu in bf1_detect_one");
 
 	update_usb_stats(bitfury);
 	applog(LOG_INFO, "%s %d: Successfully initialised %s",
 	       bitfury->drv->name, bitfury->device_id, bitfury->device_path);
+
+	/* This does not artificially raise hashrate, it simply allows the
+	 * hashrate to adapt quickly on starting. */
+	info->total_nonces = 1;
+
 	return true;
 out_close:
-	bitfury_close(bitfury);
-	usb_uninit(bitfury);
-out:
-	bitfury = usb_free_cgpu(bitfury);
+	bf1_close(bitfury);
 	return false;
 }
 
+static struct cgpu_info *bitfury_detect_one(struct libusb_device *dev, struct usb_find_devices *found)
+{
+	struct cgpu_info *bitfury;
+	struct bitfury_info *info;
+	enum sub_ident ident;
+	bool ret = false;
+
+	bitfury = usb_alloc_cgpu(&bitfury_drv, 1);
+
+	if (!usb_init(bitfury, dev, found))
+		goto out;
+	applog(LOG_INFO, "%s %d: Found at %s", bitfury->drv->name,
+	       bitfury->device_id, bitfury->device_path);
+
+	info = calloc(sizeof(struct bitfury_info), 1);
+	if (!info)
+		quit(1, "Failed to calloc info in bitfury_detect_one");
+	bitfury->device_data = info;
+	info->ident = ident = usb_ident(bitfury);
+	switch (ident) {
+		case IDENT_BF1:
+			ret = bf1_detect_one(bitfury, info);
+			break;
+		case IDENT_BXF:
+			ret = bxf_detect_one(bitfury, info);
+			break;
+		default:
+			applog(LOG_INFO, "%s %d: Unrecognised bitfury device",
+			       bitfury->drv->name, bitfury->device_id);
+			break;
+	}
+
+	if (!ret) {
+		free(info);
+		usb_uninit(bitfury);
+out:
+		bitfury = usb_free_cgpu(bitfury);
+	}
+	return bitfury;
+}
+
 static void bitfury_detect(bool __maybe_unused hotplug)
 {
 	usb_detect(&bitfury_drv, bitfury_detect_one);
 }
 
+static void parse_bxf_submit(struct cgpu_info *bitfury, struct bitfury_info *info, char *buf)
+{
+	struct work *match_work, *tmp, *work = NULL;
+	struct thr_info *thr = info->thr;
+	uint32_t nonce, timestamp;
+	int workid;
+
+	if (!sscanf(&buf[7], "%x %x %x", &nonce, &workid, &timestamp)) {
+		applog(LOG_WARNING, "%s %d: Failed to parse submit response",
+		       bitfury->drv->name, bitfury->device_id);
+		return;
+	}
+
+	applog(LOG_DEBUG, "%s %d: Parsed nonce %u workid %d timestamp %u",
+	       bitfury->drv->name, bitfury->device_id, nonce, workid, timestamp);
+
+	rd_lock(&bitfury->qlock);
+	HASH_ITER(hh, bitfury->queued_work, match_work, tmp) {
+		if (match_work->subid == workid) {
+			work = copy_work(match_work);
+			break;
+		}
+	}
+	rd_unlock(&bitfury->qlock);
+
+	if (!work) {
+		/* Discard first results from any previous run */
+		if (unlikely(!info->valid))
+			return;
+
+		applog(LOG_INFO, "%s %d: No matching work", bitfury->drv->name, bitfury->device_id);
+
+		mutex_lock(&info->lock);
+		info->no_matching_work++;
+		mutex_unlock(&info->lock);
+
+		inc_hw_errors(thr);
+		return;
+	}
+	/* Set the device start time from when we first get valid results */
+	if (unlikely(!info->valid)) {
+		info->valid = true;
+		cgtime(&bitfury->dev_start_tv);
+	}
+	set_work_ntime(work, timestamp);
+	if (submit_nonce(thr, work, nonce)) {
+		mutex_lock(&info->lock);
+		info->nonces++;
+		mutex_unlock(&info->lock);
+	}
+	free_work(work);
+}
+
+static void parse_bxf_temp(struct cgpu_info *bitfury, struct bitfury_info *info, char *buf)
+{
+	unsigned int temp;
+
+	if (!sscanf(&buf[5], "%u", &temp)) {
+		applog(LOG_INFO, "%s %d: Failed to parse temperature",
+		       bitfury->drv->name, bitfury->device_id);
+		return;
+	}
+	mutex_lock(&info->lock);
+	info->temperature = (double)temp / 10;
+	mutex_unlock(&info->lock);
+}
+
+static void bxf_update_work(struct cgpu_info *bitfury, struct bitfury_info *info);
+
+static void parse_bxf_needwork(struct cgpu_info *bitfury, struct bitfury_info *info,
+			       char *buf)
+{
+	int needed;
+
+	if (!sscanf(&buf[9], "%d", &needed)) {
+		applog(LOG_INFO, "%s %d: Failed to parse needwork",
+		       bitfury->drv->name, bitfury->device_id);
+		return;
+	}
+	while (needed-- > 0)
+		bxf_update_work(bitfury, info);
+}
+
+static void *bxf_get_results(void *userdata)
+{
+	struct cgpu_info *bitfury = userdata;
+	struct bitfury_info *info = bitfury->device_data;
+	char threadname[24], buf[512];
+
+	snprintf(threadname, 24, "bxf_recv/%d", bitfury->device_id);
+
+	/* We operate the device at lowest diff since it's not a lot of results
+	 * to process and gives us a better indicator of the nonce return rate
+	 * and hardware errors. */
+	sprintf(buf, "target ffffffff\n");
+	if (!bxf_send_msg(bitfury, buf, C_BXF_TARGET))
+		goto out;
+
+	/* Read thread sends the first work item to get the device started
+	 * since it will roll ntime and make work itself from there on. */
+	bxf_update_work(bitfury, info);
+	bxf_update_work(bitfury, info);
+
+	while (likely(!bitfury->shutdown)) {
+		char *msg;
+		int err;
+
+		if (unlikely(bitfury->usbinfo.nodev))
+			break;
+
+		err = bxf_recv_msg(bitfury, buf);
+		if (err < 0) {
+			if (err != LIBUSB_ERROR_TIMEOUT)
+				break;
+			continue;
+		}
+		if (!err)
+			continue;
+
+		msg = strstr(buf, "submit");
+		if (msg) {
+			parse_bxf_submit(bitfury, info, msg);
+			continue;
+		}
+		msg = strstr(buf, "temp");
+		if (msg) {
+			parse_bxf_temp(bitfury, info, msg);
+			continue;
+		}
+		msg = strstr(buf, "needwork");
+		if (msg) {
+			parse_bxf_needwork(bitfury, info, msg);
+			continue;
+		}
+		applog(LOG_DEBUG, "%s %d: Unrecognised string %s",
+		       bitfury->drv->name, bitfury->device_id, buf);
+	}
+out:
+	return NULL;
+}
+
+static bool bxf_prepare(struct cgpu_info *bitfury, struct bitfury_info *info)
+{
+	mutex_init(&info->lock);
+	if (pthread_create(&info->read_thr, NULL, bxf_get_results, (void *)bitfury))
+		quit(1, "Failed to create bxf read_thr");
+	return true;
+}
+
+static bool bitfury_prepare(struct thr_info *thr)
+{
+	struct cgpu_info *bitfury = thr->cgpu;
+	struct bitfury_info *info = bitfury->device_data;
+
+	info->thr = thr;
+
+	switch(info->ident) {
+		case IDENT_BXF:
+			return bxf_prepare(bitfury, info);
+			break;
+		case IDENT_BF1:
+		default:
+			return true;
+	}
+}
+
 static uint32_t decnonce(uint32_t in)
 {
 	uint32_t out;
@@ -220,16 +522,31 @@ static bool bitfury_checkresults(struct thr_info *thr, struct work *work, uint32
 	return false;
 }
 
-static int64_t bitfury_scanwork(struct thr_info *thr)
+static int64_t bitfury_rate(struct bitfury_info *info)
 {
-	struct cgpu_info *bitfury = thr->cgpu;
-	struct bitfury_info *info = bitfury->device_data;
-	int amount, i, aged = 0, total = 0, ms_diff;
+	double nonce_rate;
+	int64_t ret = 0;
+
+	info->cycles++;
+	info->total_nonces += info->nonces;
+	info->saved_nonces += info->nonces;
+	info->nonces = 0;
+	nonce_rate = (double)info->total_nonces / (double)info->cycles;
+	if (info->saved_nonces >= nonce_rate) {
+		info->saved_nonces -= nonce_rate;
+		ret = (double)0xffffffff * nonce_rate;
+	}
+	return ret;
+}
+
+static int64_t bf1_scan(struct thr_info *thr, struct cgpu_info *bitfury,
+			struct bitfury_info *info)
+{
+	int amount, i, aged, total = 0, ms_diff;
+	char readbuf[512], buf[45];
 	struct work *work, *tmp;
 	struct timeval tv_now;
-	double nonce_rate;
 	int64_t ret = 0;
-	char buf[45];
 
 	work = get_queue_work(thr, bitfury, thr->id);
 	if (unlikely(thr->work_restart)) {
@@ -247,7 +564,7 @@ static int64_t bitfury_scanwork(struct thr_info *thr)
 	cgtime(&tv_now);
 	ms_diff = 600 - ms_tdiff(&tv_now, &info->tv_start);
 	if (ms_diff > 0) {
-		usb_read_timeout_cancellable(bitfury, info->buf, 512, &amount, ms_diff,
+		usb_read_timeout_cancellable(bitfury, readbuf, 512, &amount, ms_diff,
 					     C_BF1_GETRES);
 		total += amount;
 	}
@@ -259,11 +576,11 @@ static int64_t bitfury_scanwork(struct thr_info *thr)
 	/* If a work restart was sent, just empty the buffer. */
 	if (unlikely(ms_diff < 10 || thr->work_restart))
 		ms_diff = 10;
-	usb_read_once_timeout_cancellable(bitfury, info->buf + total, BF1MSGSIZE,
+	usb_read_once_timeout_cancellable(bitfury, readbuf + total, BF1MSGSIZE,
 					  &amount, ms_diff, C_BF1_GETRES);
 	total += amount;
 	while (amount) {
-		usb_read_once_timeout(bitfury, info->buf + total, 512, &amount, 10,
+		usb_read_once_timeout(bitfury, readbuf + total, 512 - total, &amount, 10,
 				      C_BF1_GETRES);
 		total += amount;
 	};
@@ -288,7 +605,7 @@ out:
 		uint32_t nonce;
 
 		/* Ignore state & switched data in results for now. */
-		memcpy(&nonce, info->buf + i + 3, 4);
+		memcpy(&nonce, readbuf + i + 3, 4);
 		nonce = decnonce(nonce);
 
 		rd_lock(&bitfury->qlock);
@@ -301,37 +618,53 @@ out:
 		}
 		rd_unlock(&bitfury->qlock);
 
-		if (!found)
-			inc_hw_errors(thr);
+		if (!found) {
+			if (likely(info->valid))
+				inc_hw_errors(thr);
+		} else if (unlikely(!info->valid)) {
+			info->valid = true;
+			cgtime(&bitfury->dev_start_tv);
+		}
 	}
 
 	cgtime(&tv_now);
 
 	/* This iterates over the hashlist finding work started more than 6
-	 * seconds ago which equates to leaving 5 past work items in the array
-	 * to look for results. */
-	wr_lock(&bitfury->qlock);
-	HASH_ITER(hh, bitfury->queued_work, work, tmp) {
-		if (tdiff(&tv_now, &work->tv_work_start) > 6.0) {
-			__work_completed(bitfury, work);
-			aged++;
-		}
-	}
-	wr_unlock(&bitfury->qlock);
-
+	 * seconds ago. */
+	aged = age_queued_work(bitfury, 6.0);
 	if (aged) {
 		applog(LOG_DEBUG, "%s %d: Aged %d work items", bitfury->drv->name,
 		       bitfury->device_id, aged);
 	}
 
-	info->cycles++;
-	info->total_nonces += info->nonces;
-	info->saved_nonces += info->nonces;
-	info->nonces = 0;
-	nonce_rate = (double)info->total_nonces / (double)info->cycles;
-	if (info->saved_nonces >= nonce_rate) {
-		info->saved_nonces -= nonce_rate;
-		ret = (double)0xffffffff * nonce_rate;
+	ret = bitfury_rate(info);
+
+	if (unlikely(bitfury->usbinfo.nodev)) {
+		applog(LOG_WARNING, "%s %d: Device disappeared, disabling thread",
+		       bitfury->drv->name, bitfury->device_id);
+		ret = -1;
+	}
+	return ret;
+}
+
+static int64_t bxf_scan(struct cgpu_info *bitfury, struct bitfury_info *info)
+{
+	int64_t ret;
+	int aged;
+
+	bxf_update_work(bitfury, info);
+	cgsleep_ms(600);
+
+	mutex_lock(&info->lock);
+	ret = bitfury_rate(info);
+	mutex_unlock(&info->lock);
+
+	/* Keep no more than the last 90 seconds worth of work items in the
+	 * hashlist */
+	aged = age_queued_work(bitfury, 90.0);
+	if (aged) {
+		applog(LOG_DEBUG, "%s %d: Aged %d work items", bitfury->drv->name,
+		       bitfury->device_id, aged);
 	}
 
 	if (unlikely(bitfury->usbinfo.nodev)) {
@@ -342,9 +675,89 @@ out:
 	return ret;
 }
 
-static struct api_data *bitfury_api_stats(struct cgpu_info *cgpu)
+static int64_t bitfury_scanwork(struct thr_info *thr)
+{
+	struct cgpu_info *bitfury = thr->cgpu;
+	struct bitfury_info *info = bitfury->device_data;
+
+	switch(info->ident) {
+		case IDENT_BF1:
+			return bf1_scan(thr, bitfury, info);
+			break;
+		case IDENT_BXF:
+			return bxf_scan(bitfury, info);
+			break;
+		default:
+			return 0;
+	}
+}
+
+static void bxf_send_maxroll(struct cgpu_info *bitfury, int maxroll)
+{
+	char buf[20];
+
+	sprintf(buf, "maxroll %d\n", maxroll);
+	bxf_send_msg(bitfury, buf, C_BXF_MAXROLL);
+}
+
+static bool bxf_send_work(struct cgpu_info *bitfury, struct work *work)
+{
+	char buf[512], hexwork[156];
+
+	__bin2hex(hexwork, work->data, 76);
+	sprintf(buf, "work %s %x\n", hexwork, work->subid);
+	return bxf_send_msg(bitfury, buf, C_BXF_WORK);
+}
+
+static void bxf_update_work(struct cgpu_info *bitfury, struct bitfury_info *info)
+{
+	struct thr_info *thr = info->thr;
+	struct work *work;
+
+	work = get_queue_work(thr, bitfury, thr->id);
+	if (work->drv_rolllimit != info->maxroll) {
+		info->maxroll = work->drv_rolllimit;
+		bxf_send_maxroll(bitfury, info->maxroll);
+	}
+
+	mutex_lock(&info->lock);
+	work->subid = ++info->work_id;
+	mutex_unlock(&info->lock);
+
+	cgtime(&work->tv_work_start);
+	bxf_send_work(bitfury, work);
+}
+
+static void bitfury_flush_work(struct cgpu_info *bitfury)
+{
+	struct bitfury_info *info = bitfury->device_data;
+
+	switch(info->ident) {
+		case IDENT_BXF:
+			bxf_send_flush(bitfury);
+			bxf_update_work(bitfury, info);
+			bxf_update_work(bitfury, info);
+		case IDENT_BF1:
+		default:
+			break;
+	}
+}
+
+static void bitfury_update_work(struct cgpu_info *bitfury)
+{
+	struct bitfury_info *info = bitfury->device_data;
+
+	switch(info->ident) {
+		case IDENT_BXF:
+			bxf_update_work(bitfury, info);
+		case IDENT_BF1:
+		default:
+			break;
+	}
+}
+
+static struct api_data *bf1_api_stats(struct bitfury_info *info)
 {
-	struct bitfury_info *info = cgpu->device_data;
 	struct api_data *root = NULL;
 	double nonce_rate;
 	char serial[16];
@@ -361,18 +774,97 @@ static struct api_data *bitfury_api_stats(struct cgpu_info *cgpu)
 	return root;
 }
 
-static void bitfury_init(struct cgpu_info  *bitfury)
+static struct api_data *bxf_api_stats(struct bitfury_info *info)
 {
-	bitfury_close(bitfury);
-	bitfury_open(bitfury);
-	bitfury_reset(bitfury);
+	struct api_data *root = NULL;
+	double nonce_rate;
+	char buf[32];
+
+	sprintf(buf, "%d.%d", info->ver_major, info->ver_minor);
+	root = api_add_string(root, "Version", buf, true);
+	root = api_add_int(root, "Revision", &info->hw_rev,  false);
+	root = api_add_int(root, "Chips", &info->chips, false);
+	nonce_rate = (double)info->total_nonces / (double)info->cycles;
+	root = api_add_double(root, "NonceRate", &nonce_rate, true);
+	root = api_add_int(root, "NoMatchingWork", &info->no_matching_work, false);
+
+	return root;
+}
+
+static struct api_data *bitfury_api_stats(struct cgpu_info *cgpu)
+{
+	struct bitfury_info *info = cgpu->device_data;
+
+	switch(info->ident) {
+		case IDENT_BF1:
+			return bf1_api_stats(info);
+			break;
+		case IDENT_BXF:
+			return bxf_api_stats(info);
+			break;
+		default:
+			break;
+	}
+	return NULL;
+}
+
+static void bitfury_get_statline_before(char *buf, size_t bufsiz, struct cgpu_info *cgpu)
+{
+	struct bitfury_info *info = cgpu->device_data;
+
+	switch(info->ident) {
+		case IDENT_BXF:
+			tailsprintf(buf, bufsiz, "%03.1fC         | ", info->temperature);
+			break;
+		case IDENT_BF1:
+		default:
+			tailsprintf(buf, bufsiz, "               | ");
+			break;
+	}
+}
+
+static void bf1_init(struct cgpu_info *bitfury)
+{
+	bf1_close(bitfury);
+	bf1_open(bitfury);
+	bf1_reset(bitfury);
+}
+
+static void bitfury_init(struct cgpu_info *bitfury)
+{
+	struct bitfury_info *info = bitfury->device_data;
+
+	switch(info->ident) {
+		case IDENT_BF1:
+			bf1_init(bitfury);
+			break;
+		case IDENT_BXF:
+		default:
+			break;
+	}
+}
+
+static void bxf_close(struct bitfury_info *info)
+{
+	pthread_join(info->read_thr, NULL);
+	mutex_destroy(&info->lock);
 }
 
 static void bitfury_shutdown(struct thr_info *thr)
 {
 	struct cgpu_info *bitfury = thr->cgpu;
+	struct bitfury_info *info = bitfury->device_data;
 
-	bitfury_close(bitfury);
+	switch(info->ident) {
+		case IDENT_BF1:
+			bf1_close(bitfury);
+			break;
+		case IDENT_BXF:
+			bxf_close(info);
+			break;
+		default:
+			break;
+	}
 }
 
 /* Currently hardcoded to BF1 devices */
@@ -381,9 +873,13 @@ struct device_drv bitfury_drv = {
 	.dname = "bitfury",
 	.name = "BF1",
 	.drv_detect = bitfury_detect,
+	.thread_prepare = bitfury_prepare,
 	.hash_work = &hash_driver_work,
 	.scanwork = bitfury_scanwork,
+	.flush_work = bitfury_flush_work,
+	.update_work = bitfury_update_work,
 	.get_api_stats = bitfury_api_stats,
+	.get_statline_before = bitfury_get_statline_before,
 	.reinit_device = bitfury_init,
 	.thread_shutdown = bitfury_shutdown,
 	.identify_device = bitfury_identify
diff --git a/driver-bitfury.h b/driver-bitfury.h
index 79921b8..851d3cc 100644
--- a/driver-bitfury.h
+++ b/driver-bitfury.h
@@ -15,15 +15,31 @@
 
 struct bitfury_info {
 	struct cgpu_info *base_cgpu;
-	uint8_t version;
-	char product[8];
-	uint32_t serial;
-	char buf[512];
+	struct thr_info *thr;
+	enum sub_ident ident;
 	int nonces;
 	int total_nonces;
 	double saved_nonces;
 	int cycles;
+	bool valid; /* Set on first valid data being found */
+
+	/* BF1 specific data */
+	uint8_t version;
+	char product[8];
+	uint32_t serial;
 	struct timeval tv_start;
+
+	/* BXF specific data */
+	pthread_mutex_t lock;
+	pthread_t read_thr;
+	double temperature;
+	int work_id; // Current work->subid
+	int no_matching_work;
+	int maxroll; // Last maxroll sent to device
+	int ver_major;
+	int ver_minor;
+	int hw_rev;
+	int chips;
 };
 
 #endif /* BITFURY_H */
diff --git a/driver-hashfast.c b/driver-hashfast.c
index 281d2ea..4f4a661 100644
--- a/driver-hashfast.c
+++ b/driver-hashfast.c
@@ -147,20 +147,26 @@ static bool hfa_send_packet(struct cgpu_info *hashfast, struct hf_header *h, int
 
 static bool hfa_get_header(struct cgpu_info *hashfast, struct hf_header *h, uint8_t *computed_crc)
 {
-	int amount, ret, orig_len, len, ofs = 0, reads = 0;
+	int amount, ret, orig_len, len, ofs = 0;
+	cgtimer_t ts_start;
 	char buf[512];
 	char *header;
 
+	orig_len = len = sizeof(*h);
+
 	/* Read for up to 200ms till we find the first occurrence of HF_PREAMBLE
 	 * though it should be the first byte unless we get woefully out of
 	 * sync. */
-	orig_len = len = sizeof(*h);
+	cgtimer_time(&ts_start);
 	do {
+		cgtimer_t ts_now, ts_diff;
 
-		if (++reads > 20)
+		cgtimer_time(&ts_now);
+		cgtimer_sub(&ts_now, &ts_start, &ts_diff);
+		if (cgtimer_to_ms(&ts_diff) > 200)
 			return false;
 
-		ret = usb_read_timeout(hashfast, buf + ofs, len, &amount, 10, C_HF_GETHEADER);
+		ret = usb_read(hashfast, buf + ofs, len, &amount, C_HF_GETHEADER);
 		if (unlikely(ret && ret != LIBUSB_ERROR_TIMEOUT))
 			return false;
 		ofs += amount;
@@ -206,11 +212,11 @@ static bool hfa_reset(struct cgpu_info *hashfast, struct hashfast_info *info)
 	struct hf_usb_init_header usb_init, *hu = &usb_init;
 	struct hf_usb_init_base *db;
         struct hf_usb_init_options *ho;
+	int retries = 0, i;
 	char buf[1024];
 	struct hf_header *h = (struct hf_header *)buf;
 	uint8_t hcrc;
 	bool ret;
-	int i;
 
         // XXX Following items need to be defaults with command-line overrides
 	info->hash_clock_rate = 550;	// Hash clock rate in Mhz
@@ -260,7 +266,9 @@ tryagain:
 		applog(LOG_WARNING, "HFA %d: OP_USB_INIT: Tossing packet, valid but unexpected type %d",
                         hashfast->device_id, h->operation_code);
 		hfa_get_data(hashfast, buf, h->data_length);
-		goto tryagain;
+		if (retries++ < 3)
+			goto tryagain;
+		return false;
 	}
 
 	applog(LOG_DEBUG, "HFA %d: Good reply to OP_USB_INIT", hashfast->device_id);
@@ -409,7 +417,7 @@ static bool hfa_initialise(struct cgpu_info *hashfast)
 	return (err == 7);
 }
 
-static bool hfa_detect_one_usb(libusb_device *dev, struct usb_find_devices *found)
+static struct cgpu_info *hfa_detect_one(libusb_device *dev, struct usb_find_devices *found)
 {
 	struct cgpu_info *hashfast;
 
@@ -419,19 +427,25 @@ static bool hfa_detect_one_usb(libusb_device *dev, struct usb_find_devices *foun
 
 	if (!usb_init(hashfast, dev, found)) {
 		hashfast = usb_free_cgpu(hashfast);
-		return false;
+		return NULL;
 	}
 
 	hashfast->usbdev->usb_type = USB_TYPE_STD;
 
 	if (!hfa_initialise(hashfast)) {
 		hashfast = usb_free_cgpu(hashfast);
-		return false;
+		return NULL;
 	}
 
-	add_cgpu(hashfast);
+	if (!hfa_detect_common(hashfast)) {
+		usb_uninit(hashfast);
+		hashfast = usb_free_cgpu(hashfast);
+		return NULL;
+	}
+	if (!add_cgpu(hashfast))
+		return NULL;
 
-	return hfa_detect_common(hashfast);
+	return hashfast;
 }
 
 static void hfa_detect(bool hotplug)
@@ -439,7 +453,7 @@ static void hfa_detect(bool hotplug)
 	/* Set up the CRC tables only once. */
 	if (!hotplug)
 		hfa_init_crc8();
-	usb_detect(&hashfast_drv, hfa_detect_one_usb);
+	usb_detect(&hashfast_drv, hfa_detect_one);
 }
 
 static bool hfa_get_packet(struct cgpu_info *hashfast, struct hf_header *h)
diff --git a/driver-icarus.c b/driver-icarus.c
index 7da4a99..af35124 100644
--- a/driver-icarus.c
+++ b/driver-icarus.c
@@ -786,7 +786,7 @@ static void get_options(int this_option_offset, struct cgpu_info *icarus, int *b
 	}
 }
 
-static bool icarus_detect_one(struct libusb_device *dev, struct usb_find_devices *found)
+static struct cgpu_info *icarus_detect_one(struct libusb_device *dev, struct usb_find_devices *found)
 {
 	int this_option_offset = ++option_offset;
 	struct ICARUS_INFO *info;
@@ -1010,7 +1010,7 @@ cmr2_retry:
 		}
 	}
 
-	return true;
+	return icarus;
 
 unshin:
 
@@ -1022,7 +1022,7 @@ shin:
 
 	icarus = usb_free_cgpu(icarus);
 
-	return false;
+	return NULL;
 }
 
 static void icarus_detect(bool __maybe_unused hotplug)
diff --git a/driver-klondike.c b/driver-klondike.c
index 2446202..5da915c 100644
--- a/driver-klondike.c
+++ b/driver-klondike.c
@@ -765,7 +765,7 @@ static void control_init(struct cgpu_info *klncgpu)
 			  klncgpu->drv->name, klncgpu->device_id, err);
 }
 
-static bool klondike_detect_one(struct libusb_device *dev, struct usb_find_devices *found)
+static struct cgpu_info *klondike_detect_one(struct libusb_device *dev, struct usb_find_devices *found)
 {
 	struct cgpu_info *klncgpu = usb_alloc_cgpu(&klondike_drv, 1);
 	struct klondike_info *klninfo = NULL;
@@ -823,7 +823,7 @@ static bool klondike_detect_one(struct libusb_device *dev, struct usb_find_devic
 				applog(LOG_DEBUG, "Klondike cgpu added");
 				rwlock_init(&klninfo->stat_lock);
 				cglock_init(&klninfo->klist_lock);
-				return true;
+				return klncgpu;
 			}
 		}
 		usb_uninit(klncgpu);
@@ -831,7 +831,7 @@ static bool klondike_detect_one(struct libusb_device *dev, struct usb_find_devic
 	free(klninfo->free);
 	free(klninfo);
 	free(klncgpu);
-	return false;
+	return NULL;
 }
 
 static void klondike_detect(bool __maybe_unused hotplug)
@@ -1129,26 +1129,28 @@ static void klondike_flush_work(struct cgpu_info *klncgpu)
 	KLINE kline;
 	int slaves, dev;
 
-	wr_lock(&(klninfo->stat_lock));
-	klninfo->block_seq++;
-	slaves = klninfo->status[0].kline.ws.slavecount;
-	wr_unlock(&(klninfo->stat_lock));
+	if (klninfo->initialised) {
+		wr_lock(&(klninfo->stat_lock));
+		klninfo->block_seq++;
+		slaves = klninfo->status[0].kline.ws.slavecount;
+		wr_unlock(&(klninfo->stat_lock));
 
-	applog(LOG_DEBUG, "%s%i: flushing work",
-			  klncgpu->drv->name, klncgpu->device_id);
-	zero_kline(&kline);
-	kline.hd.cmd = KLN_CMD_ABORT;
-	for (dev = 0; dev <= slaves; dev++) {
-		kline.hd.dev = dev;
-		kitem = SendCmdGetReply(klncgpu, &kline, KSENDHD(0));
-		if (kitem != NULL) {
-			wr_lock(&(klninfo->stat_lock));
-			memcpy((void *)&(klninfo->status[dev]),
-				kitem,
-				sizeof(klninfo->status[dev]));
-			klninfo->jobque[dev].flushed = true;
-			wr_unlock(&(klninfo->stat_lock));
-			kitem = release_kitem(klncgpu, kitem);
+		applog(LOG_DEBUG, "%s%i: flushing work",
+				  klncgpu->drv->name, klncgpu->device_id);
+		zero_kline(&kline);
+		kline.hd.cmd = KLN_CMD_ABORT;
+		for (dev = 0; dev <= slaves; dev++) {
+			kline.hd.dev = dev;
+			kitem = SendCmdGetReply(klncgpu, &kline, KSENDHD(0));
+			if (kitem != NULL) {
+				wr_lock(&(klninfo->stat_lock));
+				memcpy((void *)&(klninfo->status[dev]),
+					kitem,
+					sizeof(klninfo->status[dev]));
+				klninfo->jobque[dev].flushed = true;
+				wr_unlock(&(klninfo->stat_lock));
+				kitem = release_kitem(klncgpu, kitem);
+			}
 		}
 	}
 }
diff --git a/driver-modminer.c b/driver-modminer.c
index e10cbba..642d13d 100644
--- a/driver-modminer.c
+++ b/driver-modminer.c
@@ -109,7 +109,7 @@ static void do_ping(struct cgpu_info *modminer)
 		modminer->drv->name, modminer->fpgaid, amount, err);
 }
 
-static bool modminer_detect_one(struct libusb_device *dev, struct usb_find_devices *found)
+static struct cgpu_info *modminer_detect_one(struct libusb_device *dev, struct usb_find_devices *found)
 {
 	char buf[0x100+1];
 	char *devname = NULL;
@@ -217,7 +217,7 @@ static bool modminer_detect_one(struct libusb_device *dev, struct usb_find_devic
 
 	modminer = usb_free_cgpu(modminer);
 
-	return true;
+	return modminer;
 
 unshin:
 	if (!added)
@@ -232,9 +232,9 @@ shin:
 	modminer = usb_free_cgpu(modminer);
 
 	if (added)
-		return true;
+		return modminer;
 	else
-		return false;
+		return NULL;
 }
 
 static void modminer_detect(bool __maybe_unused hotplug)
diff --git a/example.conf b/example.conf
index ae3366e..2448ed5 100644
--- a/example.conf
+++ b/example.conf
@@ -17,24 +17,9 @@
 	}
 ],
 
-"intensity" : "d,9,9,9",
-"gpu-engine" : "0-985,0-950,0-960,0-1000",
-"gpu-fan" : "0-85,0-85,0-85,0-85",
-"gpu-memclock" : "860,825,835,875",
-"gpu-powertune" : "20,20,20,20",
-"temp-cutoff" : "95,95,95,95",
-"temp-overheat" : "85,85,85,85",
-"temp-target" : "75,75,75,75",
-
-"auto-fan" : true,
-"auto-gpu" : true,
-"expiry" : "120",
 "failover-only" : true,
-"gpu-threads" : "2",
-"log" : "5",
-"queue" : "1",
-"scan-time" : "60",
-"temp-hysteresis" : "3",
-
-"kernel-path" : "/usr/local/bin"
+"no-submit-stale" : true,
+"api-listen" : true,
+"api-port" : "4028",
+"api-allow" : "W:192.168.1.0/24,W:127.0.0.1"
 }
diff --git a/hf_protocol.h b/hf_protocol.h
index 387c284..28cc13b 100644
--- a/hf_protocol.h
+++ b/hf_protocol.h
@@ -98,7 +98,7 @@
 
 // Structure definitions, LE platforms
 
-#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+#if __BYTE_ORDER == __BIG_ENDIAN
 #include "hf_protocol_be.h"
 #else
 // Generic header
diff --git a/miner.h b/miner.h
index ef05dad..685a4fb 100644
--- a/miner.h
+++ b/miner.h
@@ -1444,6 +1444,7 @@ extern struct work *__find_work_bymidstate(struct work *que, char *midstate, siz
 extern struct work *find_queued_work_bymidstate(struct cgpu_info *cgpu, char *midstate, size_t midstatelen, char *data, int offset, size_t datalen);
 extern struct work *clone_queued_work_bymidstate(struct cgpu_info *cgpu, char *midstate, size_t midstatelen, char *data, int offset, size_t datalen);
 extern void __work_completed(struct cgpu_info *cgpu, struct work *work);
+extern int age_queued_work(struct cgpu_info *cgpu, double secs);
 extern void work_completed(struct cgpu_info *cgpu, struct work *work);
 extern struct work *take_queued_work_bymidstate(struct cgpu_info *cgpu, char *midstate, size_t midstatelen, char *data, int offset, size_t datalen);
 extern void hash_driver_work(struct thr_info *mythr);
@@ -1475,6 +1476,7 @@ extern void adl(void);
 extern void app_restart(void);
 extern void clean_work(struct work *work);
 extern void free_work(struct work *work);
+extern void set_work_ntime(struct work *work, int ntime);
 extern struct work *copy_work_noffset(struct work *base_work, int noffset);
 #define copy_work(work_in) copy_work_noffset(work_in, 0)
 extern struct thr_info *get_thread(int thr_id);
diff --git a/miner.php b/miner.php
index fd42271..ff2e457 100644
--- a/miner.php
+++ b/miner.php
@@ -2910,7 +2910,7 @@ function display()
 
  if ($allowcustompages === true)
  {
-	$pg = trim(getparam('pg', true));
+	$pg = urlencode(trim(getparam('pg', true)));
 	if ($pagesonly === true)
 	{
 		if ($pg !== null && $pg !== '')
diff --git a/usbutils.c b/usbutils.c
index 3b62545..b3459d2 100644
--- a/usbutils.c
+++ b/usbutils.c
@@ -139,6 +139,20 @@ static struct usb_intinfo bfu_ints[] = {
 	USB_EPS(1,  bfu1_epinfos),
 	USB_EPS(0,  bfu0_epinfos)
 };
+
+static struct usb_epinfo bxf0_epinfos[] = {
+	{ LIBUSB_TRANSFER_TYPE_INTERRUPT,	8,	EPI(1), 0, 0 }
+};
+
+static struct usb_epinfo bxf1_epinfos[] = {
+	{ LIBUSB_TRANSFER_TYPE_BULK,	64,	EPI(2), 0, 0 },
+	{ LIBUSB_TRANSFER_TYPE_BULK,	64,	EPO(2), 0, 0 }
+};
+
+static struct usb_intinfo bxf_ints[] = {
+	USB_EPS(1,  bxf1_epinfos),
+	USB_EPS(0,  bxf0_epinfos)
+};
 #endif
 
 #ifdef USE_HASHFAST
@@ -191,6 +205,20 @@ static struct usb_epinfo kln_epinfos[] = {
 static struct usb_intinfo kln_ints[] = {
 	USB_EPS(0, kln_epinfos)
 };
+
+static struct usb_epinfo kli0_epinfos[] = {
+	{ LIBUSB_TRANSFER_TYPE_INTERRUPT, 8,	EPI(1), 0, 0 }
+};
+
+static struct usb_epinfo kli1_epinfos[] = {
+	{ LIBUSB_TRANSFER_TYPE_BULK,	64,	EPI(2), 0, 0 },
+	{ LIBUSB_TRANSFER_TYPE_BULK,	64,	EPO(2), 0, 0 }
+};
+
+static struct usb_intinfo kli_ints[] = {
+	USB_EPS(1, kli1_epinfos),
+	USB_EPS(0, kli0_epinfos)
+};
 #endif
 
 #ifdef USE_ICARUS
@@ -299,7 +327,7 @@ static struct usb_find_devices find_dev[] = {
 	{
 		.drv = DRIVER_bitfury,
 		.name = "BF1",
-		.ident = IDENT_BFU,
+		.ident = IDENT_BF1,
 		.idVendor = 0x03eb,
 		.idProduct = 0x204b,
 		.config = 1,
@@ -309,6 +337,19 @@ static struct usb_find_devices find_dev[] = {
 		.iProduct = "Bitfury BF1",
 		INTINFO(bfu_ints)
 	},
+	{
+		.drv = DRIVER_bitfury,
+		.name = "BXF",
+		.ident = IDENT_BXF,
+		.idVendor = 0x198c,
+		.idProduct = 0xb1f1,
+		.config = 1,
+		.timeout = BITFURY_TIMEOUT_MS,
+		.latency = LATENCY_UNUSED,
+		.iManufacturer = "c-scape",
+		.iProduct = "bi?fury",
+		INTINFO(bxf_ints)
+	},
 #endif
 #ifdef USE_MODMINER
 	{
@@ -383,6 +424,16 @@ static struct usb_find_devices find_dev[] = {
 		.timeout = KLONDIKE_TIMEOUT_MS,
 		.latency = 10,
 		INTINFO(kln_ints) },
+	{
+		.drv = DRIVER_klondike,
+		.name = "KLI",
+		.ident = IDENT_KLN,
+		.idVendor = 0x04D8,
+		.idProduct = 0xF60A,
+		.config = 1,
+		.timeout = KLONDIKE_TIMEOUT_MS,
+		.latency = 10,
+		INTINFO(kli_ints) },
 #endif
 #ifdef USE_ICARUS
 	{
@@ -488,8 +539,8 @@ static const char *nodatareturned = "no data returned ";
 
 // For device limits by driver
 static struct driver_count {
-	uint32_t count;
-	uint32_t limit;
+	int count;
+	int limit;
 } drv_count[DRIVER_MAX];
 
 // For device limits by list of bus/dev
@@ -548,7 +599,7 @@ struct resource_reply *res_reply_head = NULL;
 #define MODE_BULK_WRITE (1 << 3)
 
 // Set this to 0 to remove stats processing
-#define DO_USB_STATS 1
+#define DO_USB_STATS 0
 
 static bool stats_initialised = false;
 
@@ -1359,6 +1410,7 @@ void usb_uninit(struct cgpu_info *cgpu)
 static void release_cgpu(struct cgpu_info *cgpu)
 {
 	struct cg_usb_device *cgusb = cgpu->usbdev;
+	bool initted = cgpu->usbinfo.initialised;
 	struct cgpu_info *lookcgpu;
 	int i;
 
@@ -1369,9 +1421,11 @@ static void release_cgpu(struct cgpu_info *cgpu)
 	applog(LOG_DEBUG, "USB release %s%i",
 			cgpu->drv->name, cgpu->device_id);
 
-	zombie_devs++;
-	total_count--;
-	drv_count[cgpu->drv->drv_id].count--;
+	if (initted) {
+		zombie_devs++;
+		total_count--;
+		drv_count[cgpu->drv->drv_id].count--;
+	}
 
 	cgpu->usbinfo.nodev = true;
 	cgpu->usbinfo.nodev_count++;
@@ -1381,8 +1435,10 @@ static void release_cgpu(struct cgpu_info *cgpu)
 	for (i = 0; i < total_devices; i++) {
 		lookcgpu = get_devices(i);
 		if (lookcgpu != cgpu && lookcgpu->usbdev == cgusb) {
-			total_count--;
-			drv_count[lookcgpu->drv->drv_id].count--;
+			if (initted) {
+				total_count--;
+				drv_count[lookcgpu->drv->drv_id].count--;
+			}
 
 			lookcgpu->usbinfo.nodev = true;
 			lookcgpu->usbinfo.nodev_count++;
@@ -1539,6 +1595,7 @@ static int _usb_init(struct cgpu_info *cgpu, struct libusb_device *dev, struct u
 					"USB init, open device failed, err %d, "
 					"you don't have privilege to access %s",
 					err, devstr);
+				applog(LOG_ERR, "See README file included for help");
 				break;
 #ifdef WIN32
 			// Windows specific message
@@ -1547,6 +1604,7 @@ static int _usb_init(struct cgpu_info *cgpu, struct libusb_device *dev, struct u
 					"USB init, open device failed, err %d, "
 					"you need to install a WinUSB driver for %s",
 					err, devstr);
+				applog(LOG_ERR, "See README.txt file included for help");
 				break;
 #endif
 			default:
@@ -1925,11 +1983,12 @@ static struct usb_find_devices *usb_check(__maybe_unused struct device_drv *drv,
 	return NULL;
 }
 
-void usb_detect(struct device_drv *drv, bool (*device_detect)(struct libusb_device *, struct usb_find_devices *))
+void usb_detect(struct device_drv *drv, struct cgpu_info *(*device_detect)(struct libusb_device *, struct usb_find_devices *))
 {
 	libusb_device **list;
 	ssize_t count, i;
 	struct usb_find_devices *found;
+	struct cgpu_info *cgpu;
 
 	applog(LOG_DEBUG, "USB scan devices: checking for %s devices", drv->name);
 
@@ -1974,9 +2033,11 @@ void usb_detect(struct device_drv *drv, bool (*device_detect)(struct libusb_devi
 			if (is_in_use(list[i]) || cgminer_usb_lock(drv, list[i]) == false)
 				free(found);
 			else {
-				if (!device_detect(list[i], found))
+				cgpu = device_detect(list[i], found);
+				if (!cgpu)
 					cgminer_usb_unlock(drv, list[i]);
 				else {
+					cgpu->usbinfo.initialised = true;
 					total_count++;
 					drv_count[drv->drv_id].count++;
 				}
@@ -2125,7 +2186,7 @@ static void newstats(struct cgpu_info *cgpu)
 
 	usb_stats[next_stat].name = cgpu->drv->name;
 	usb_stats[next_stat].device_id = -1;
-	usb_stats[next_stat].details = calloc(1, sizeof(struct cg_usb_stats_details) * C_MAX * 2);
+	usb_stats[next_stat].details = calloc(1, sizeof(struct cg_usb_stats_details) * C_MAX * 2 + 8);
 	if (unlikely(!usb_stats[next_stat].details))
 		quit(1, "USB failed to calloc details for %d", next_stat+1);
 
@@ -2234,37 +2295,6 @@ static void rejected_inc(struct cgpu_info *cgpu, uint32_t mode)
 }
 #endif
 
-static char *find_end(unsigned char *buf, unsigned char *ptr, int ptrlen, int tot, char *end, int endlen, bool first)
-{
-	unsigned char *search;
-
-	if (endlen > tot)
-		return NULL;
-
-	// If end is only 1 char - do a faster search
-	if (endlen == 1) {
-		if (first)
-			search = buf;
-		else
-			search = ptr;
-
-		return strchr((char *)search, *end);
-	} else {
-		if (first)
-			search = buf;
-		else {
-			// must allow end to have been chopped in 2
-			if ((tot - ptrlen) >= (endlen - 1))
-				search = ptr - (endlen - 1);
-			else
-				search = ptr - (tot - ptrlen);
-		}
-
-		return strstr((char *)search, end);
-	}
-}
-
-#define USB_MAX_READ 8192
 #define USB_RETRY_MAX 5
 
 struct usb_transfer {
@@ -2419,15 +2449,11 @@ usb_bulk_transfer(struct libusb_device_handle *dev_handle, int intinfo,
 	struct usb_transfer ut;
 	unsigned char endpoint;
 	int err, errn;
-	/* End of transfer and zero length packet required */
-	bool eot = false, zlp = false;
 #if DO_USB_STATS
 	struct timeval tv_start, tv_finish;
 #endif
 	unsigned char buf[512];
 #ifdef WIN32
-	int dummy;
-
 	/* On windows the callback_timeout is a safety mechanism only. */
 	bulk_timeout = timeout;
 	callback_timeout += WIN_CALLBACK_EXTRA;
@@ -2440,11 +2466,6 @@ usb_bulk_transfer(struct libusb_device_handle *dev_handle, int intinfo,
 	usb_epinfo = &(cgpu->usbdev->found->intinfos[intinfo].epinfos[epinfo]);
 	endpoint = usb_epinfo->ep;
 
-	if (length > usb_epinfo->wMaxPacketSize)
-		length = usb_epinfo->wMaxPacketSize;
-	else if (length == usb_epinfo->wMaxPacketSize)
-		eot = true;
-
 	/* Avoid any async transfers during shutdown to allow the polling
 	 * thread to be shut down after all existing transfers are complete */
 	if (unlikely(cgpu->shutdown))
@@ -2454,15 +2475,7 @@ usb_bulk_transfer(struct libusb_device_handle *dev_handle, int intinfo,
 
 	if ((endpoint & LIBUSB_ENDPOINT_DIR_MASK) == LIBUSB_ENDPOINT_OUT) {
 		memcpy(buf, data, length);
-		/* If this is the last packet in a transfer and is the length
-		 * of the wMaxPacketSize then we need to send a zero length
-		 * packet to let the device know it's the end of the message.*/
-		if (eot)
-			zlp = true;
-#ifndef WIN32
-		/* Windows doesn't support this flag so we emulate it below */
 		ut.transfer->flags |= LIBUSB_TRANSFER_ADD_ZERO_PACKET;
-#endif
 	}
 
 	USBDEBUG("USB debug: @usb_bulk_transfer(%s (nodev=%s),intinfo=%d,epinfo=%d,data=%p,length=%d,timeout=%u,mode=%d,cmd=%s,seq=%d) endpoint=%d", cgpu->drv->name, bool_str(cgpu->usbinfo.nodev), intinfo, epinfo, data, length, timeout, mode, usb_cmdname(cmd), seq, (int)endpoint);
@@ -2505,35 +2518,33 @@ usb_bulk_transfer(struct libusb_device_handle *dev_handle, int intinfo,
 	}
 	if ((endpoint & LIBUSB_ENDPOINT_DIR_MASK) == LIBUSB_ENDPOINT_IN)
 		memcpy(data, buf, *transferred);
-	else if (zlp) {
-#ifdef WIN32
-		/* Send a zero length packet */
-		libusb_bulk_transfer(dev_handle, endpoint, NULL, 0, &dummy, 100);
-#endif
-	}
 
 	return err;
 }
 
 int _usb_read(struct cgpu_info *cgpu, int intinfo, int epinfo, char *buf, size_t bufsiz, int *processed, int timeout, const char *end, enum usb_cmds cmd, bool readonce, bool cancellable)
 {
-	struct cg_usb_device *usbdev;
-	bool ftdi;
+	unsigned char *ptr, usbbuf[USB_READ_BUFSIZE];
 	struct timeval read_start, tv_finish;
-	unsigned int initial_timeout;
 	int bufleft, err, got, tot, pstate;
-	bool first = true;
-	bool dobuffer;
-	char *search;
-	int endlen;
-	unsigned char *ptr, *usbbuf = cgpu->usbinfo.bulkbuf;
 	const size_t usbbufread = 512; /* Always read full size */
+	struct cg_usb_device *usbdev;
+	unsigned int initial_timeout;
+	bool first = true;
+	int endlen = 0;
+	char *eom = NULL;
 	double done;
+	bool ftdi;
+
+	memset(usbbuf, 0, USB_READ_BUFSIZE);
+	memset(buf, 0, bufsiz);
+
+	if (end)
+		endlen = strlen(end);
 
 	DEVRLOCK(cgpu, pstate);
 
 	if (cgpu->usbinfo.nodev) {
-		*buf = '\0';
 		*processed = 0;
 		USB_REJECT(cgpu, MODE_BULK_READ);
 
@@ -2552,75 +2563,6 @@ int _usb_read(struct cgpu_info *cgpu, int intinfo, int epinfo, char *buf, size_t
 	if (timeout == DEVTIMEOUT)
 		timeout = usbdev->found->timeout;
 
-	if (end == NULL) {
-		tot = usbdev->bufamt;
-		bufleft = bufsiz - tot;
-		if (tot)
-			memcpy(usbbuf, usbdev->buffer, tot);
-		ptr = usbbuf + tot;
-		usbdev->bufamt = 0;
-
-		err = LIBUSB_SUCCESS;
-		initial_timeout = timeout;
-		cgtime(&read_start);
-		while (bufleft > 0) {
-			got = 0;
-
-			err = usb_bulk_transfer(usbdev->handle, intinfo, epinfo,
-						ptr, usbbufread, &got, timeout,
-						cgpu, MODE_BULK_READ, cmd, first ? SEQ0 : SEQ1,
-						cancellable);
-			cgtime(&tv_finish);
-			ptr[got] = '\0';
-
-			USBDEBUG("USB debug: @_usb_read(%s (nodev=%s)) first=%s err=%d%s got=%d ptr='%s' usbbufread=%d", cgpu->drv->name, bool_str(cgpu->usbinfo.nodev), bool_str(first), err, isnodev(err), got, (char *)str_text((char *)ptr), (int)usbbufread);
-
-			IOERR_CHECK(cgpu, err);
-
-			if (ftdi) {
-				// first 2 bytes returned are an FTDI status
-				if (got > 2) {
-					got -= 2;
-					memmove(ptr, ptr+2, got+1);
-				} else {
-					got = 0;
-					*ptr = '\0';
-				}
-			}
-
-			tot += got;
-
-			if (err || readonce)
-				break;
-
-			ptr += got;
-			bufleft -= got;
-
-			first = false;
-
-			done = tdiff(&tv_finish, &read_start);
-			// N.B. this is: return last err with whatever size has already been read
-			timeout = initial_timeout - (done * 1000);
-			if (timeout <= 0)
-				break;
-		}
-
-		// N.B. usbdev->buffer was emptied before the while() loop
-		if (tot > (int)bufsiz) {
-			usbdev->bufamt = tot - bufsiz;
-			memcpy(usbdev->buffer, usbbuf + bufsiz, usbdev->bufamt);
-			tot -= usbdev->bufamt;
-			usbbuf[tot] = '\0';
-			applog(LOG_DEBUG, "USB: %s%i read1 buffering %d extra bytes",
-			       cgpu->drv->name, cgpu->device_id, usbdev->bufamt);
-		}
-
-		*processed = tot;
-		memcpy((char *)buf, (const char *)usbbuf, (tot < (int)bufsiz) ? tot + 1 : (int)bufsiz);
-
-		goto out_unlock;
-	}
-
 	tot = usbdev->bufamt;
 	bufleft = bufsiz - tot;
 	if (tot)
@@ -2628,13 +2570,13 @@ int _usb_read(struct cgpu_info *cgpu, int intinfo, int epinfo, char *buf, size_t
 	ptr = usbbuf + tot;
 	usbdev->bufamt = 0;
 
-	endlen = strlen(end);
 	err = LIBUSB_SUCCESS;
+	if (end != NULL)
+		eom = strstr((const char *)usbbuf, end);
+
 	initial_timeout = timeout;
 	cgtime(&read_start);
-
-	while (bufleft > 0) {
-		got = 0;
+	while (bufleft > 0 && !eom) {
 		err = usb_bulk_transfer(usbdev->handle, intinfo, epinfo,
 					ptr, usbbufread, &got, timeout,
 					cgpu, MODE_BULK_READ, cmd, first ? SEQ0 : SEQ1,
@@ -2658,59 +2600,48 @@ int _usb_read(struct cgpu_info *cgpu, int intinfo, int epinfo, char *buf, size_t
 		}
 
 		tot += got;
+		if (end != NULL)
+			eom = strstr((const char *)usbbuf, end);
 
 		if (err || readonce)
 			break;
 
-		if (find_end(usbbuf, ptr, got, tot, (char *)end, endlen, first))
-			break;
-
 		ptr += got;
 		bufleft -= got;
 
 		first = false;
 
 		done = tdiff(&tv_finish, &read_start);
-		// N.B. this is: return LIBUSB_SUCCESS with whatever size has already been read
+		// N.B. this is: return last err with whatever size has already been read
 		timeout = initial_timeout - (done * 1000);
 		if (timeout <= 0)
 			break;
 	}
 
-	dobuffer = false;
+	/* If we found the end of message marker, just use that data and
+	 * return success. */
+	if (eom) {
+		size_t eomlen = (void *)eom - (void *)usbbuf + endlen;
 
-	if ((search = find_end(usbbuf, usbbuf, tot, tot, (char *)end, endlen, true))) {
-		// end finishes after bufsiz
-		if ((search + endlen - (char *)usbbuf) > (int)bufsiz) {
-			usbdev->bufamt = tot - bufsiz;
-			dobuffer = true;
-		} else {
-			// extra data after end
-			if (*(search + endlen)) {
-				usbdev->bufamt = tot - (search + endlen - (char *)usbbuf);
-				dobuffer = true;
-			}
-		}
-	} else {
-		// no end, but still bigger than bufsiz
-		if (tot > (int)bufsiz) {
-			usbdev->bufamt = tot - bufsiz;
-			dobuffer = true;
+		if (eomlen < bufsiz) {
+			bufsiz = eomlen;
+			err = LIBUSB_SUCCESS;
 		}
 	}
 
-	if (dobuffer) {
+	// N.B. usbdev->buffer was emptied before the while() loop
+	if (tot > (int)bufsiz) {
+		usbdev->bufamt = tot - bufsiz;
+		memcpy(usbdev->buffer, usbbuf + bufsiz, usbdev->bufamt);
 		tot -= usbdev->bufamt;
-		memcpy(usbdev->buffer, usbbuf + tot, usbdev->bufamt);
 		usbbuf[tot] = '\0';
-		applog(LOG_DEBUG, "USB: %s%i read2 buffering %d extra bytes",
+		applog(LOG_DEBUG, "USB: %s%i read1 buffering %d extra bytes",
 			cgpu->drv->name, cgpu->device_id, usbdev->bufamt);
 	}
 
 	*processed = tot;
 	memcpy((char *)buf, (const char *)usbbuf, (tot < (int)bufsiz) ? tot + 1 : (int)bufsiz);
 
-out_unlock:
 	if (err && err != LIBUSB_ERROR_TIMEOUT) {
 		applog(LOG_WARNING, "%s %i %s usb read err:(%d) %s", cgpu->drv->name, cgpu->device_id, usb_cmdname(cmd),
 		       err, libusb_error_name(err));
diff --git a/usbutils.h b/usbutils.h
index 9a22247..035f83e 100644
--- a/usbutils.h
+++ b/usbutils.h
@@ -134,24 +134,25 @@ struct usb_intinfo {
 
 enum sub_ident {
 	IDENT_UNK = 0,
+	IDENT_AMU,
+	IDENT_AVA,
 	IDENT_BAJ,
 	IDENT_BAL,
-	IDENT_BAS,
 	IDENT_BAM,
+	IDENT_BAS,
+	IDENT_BBF,
+	IDENT_BF1,
 	IDENT_BFL,
-	IDENT_BFU,
-	IDENT_MMQ,
-	IDENT_AVA,
+	IDENT_BLT,
 	IDENT_BTB,
+	IDENT_BXF,
+	IDENT_CMR1,
+	IDENT_CMR2,
 	IDENT_HFA,
-	IDENT_BBF,
-	IDENT_KLN,
 	IDENT_ICA,
-	IDENT_AMU,
-	IDENT_BLT,
+	IDENT_KLN,
 	IDENT_LLT,
-	IDENT_CMR1,
-	IDENT_CMR2,
+	IDENT_MMQ,
 	IDENT_ZTX
 };
 
@@ -182,6 +183,12 @@ enum usb_types {
 };
 
 #define USB_MAX_READ 8192
+/*
+ * We add 4: 1 for null, 2 for FTDI status and 1 to round to 4 bytes
+ * If a single device ever has multiple end points then it will need
+ * multiple of these
+ */
+#define USB_READ_BUFSIZE (USB_MAX_READ + 4)
 
 struct cg_usb_device {
 	struct usb_find_devices *found;
@@ -223,6 +230,7 @@ struct cg_usb_info {
 	uint8_t device_address;
 	int usbstat;
 	bool nodev;
+	bool initialised;
 	int nodev_count;
 	struct timeval last_nodev;
 	uint32_t ioerr_count;
@@ -250,13 +258,6 @@ struct cg_usb_info {
 	uint64_t write_delay_count;
 	double total_write_delay;
 
-	/*
-	 * We add 4: 1 for null, 2 for FTDI status and 1 to round to 4 bytes
-	 * If a single device ever has multiple end points then it will need
-	 * multiple of these
-	 */
-	unsigned char bulkbuf[USB_MAX_READ+4];
-
 	uint64_t tmo_count;
 	struct cg_usb_tmo usb_tmo[USB_TMOS];
 };
@@ -342,6 +343,12 @@ struct cg_usb_info {
 	USB_ADD_COMMAND(C_BF1_FLUSH, "BF1Flush") \
 	USB_ADD_COMMAND(C_BF1_IFLUSH, "BF1InterruptFlush") \
 	USB_ADD_COMMAND(C_BF1_IDENTIFY, "BF1Identify") \
+	USB_ADD_COMMAND(C_BXF_READ, "BXFRead") \
+	USB_ADD_COMMAND(C_BXF_WORK, "BXFWork") \
+	USB_ADD_COMMAND(C_BXF_TARGET, "BXFTarget") \
+	USB_ADD_COMMAND(C_BXF_VERSION, "BXFVersion") \
+	USB_ADD_COMMAND(C_BXF_MAXROLL, "BXFMaxRoll") \
+	USB_ADD_COMMAND(C_BXF_FLUSH, "BXFFlush") \
 	USB_ADD_COMMAND(C_HF_RESET, "HFReset") \
 	USB_ADD_COMMAND(C_HF_PLL_CONFIG, "HFPLLConfig") \
 	USB_ADD_COMMAND(C_HF_ADDRESS, "HFAddress") \
@@ -382,7 +389,7 @@ struct cgpu_info *usb_alloc_cgpu(struct device_drv *drv, int threads);
 struct cgpu_info *usb_free_cgpu(struct cgpu_info *cgpu);
 void usb_uninit(struct cgpu_info *cgpu);
 bool usb_init(struct cgpu_info *cgpu, struct libusb_device *dev, struct usb_find_devices *found);
-void usb_detect(struct device_drv *drv, bool (*device_detect)(struct libusb_device *, struct usb_find_devices *));
+void usb_detect(struct device_drv *drv, struct cgpu_info *(*device_detect)(struct libusb_device *, struct usb_find_devices *));
 struct api_data *api_usb_stats(int *count);
 void update_usb_stats(struct cgpu_info *cgpu);
 int _usb_read(struct cgpu_info *cgpu, int intinfo, int epinfo, char *buf, size_t bufsiz, int *processed, int timeout, const char *end, enum usb_cmds cmd, bool readonce, bool cancellable);
diff --git a/util.c b/util.c
index 4e3d502..0e9670c 100644
--- a/util.c
+++ b/util.c
@@ -1839,7 +1839,7 @@ bool auth_stratum(struct pool *pool)
 			ss = json_dumps(err_val, JSON_INDENT(3));
 		else
 			ss = strdup("(unknown reason)");
-		applog(LOG_WARNING, "pool %d JSON stratum auth failed: %s", pool->pool_no, ss);
+		applog(LOG_INFO, "pool %d JSON stratum auth failed: %s", pool->pool_no, ss);
 		free(ss);
 
 		goto out;