Merge branch 'master' into libusbx

cgminer.c

@@ -3218,8 +3218,6 @@ static void __kill_work(void)
 	/* Release USB resources in case it's a restart
 	 * and not a QUIT */
 	if (!opt_scrypt) {
-		usb_polling = false;
-
 		applog(LOG_DEBUG, "Releasing all USB devices");
 		usb_cleanup();
 
@@ -4006,15 +4004,15 @@ static void set_blockdiff(const struct work *work)
 static bool test_work_current(struct work *work)
 {
 	bool ret = true;
-	char *hexstr;
+	char hexstr[20];
 
 	if (work->mandatory)
 		return ret;
 
 	/* Hack to work around dud work sneaking into test */
-	hexstr = bin2hex(work->data + 8, 18);
+	__bin2hex(hexstr, work->data + 8, 18);
 	if (!strncmp(hexstr, "000000000000000000000000000000000000", 36))
-		goto out_free;
+		return ret;
 
 	/* Search to see if this block exists yet and if not, consider it a
 	 * new block and set the current block details to this one */
@@ -4049,7 +4047,7 @@ static bool test_work_current(struct work *work)
 			applog(LOG_DEBUG, "Deleted block %d from database", deleted_block);
 		set_curblock(hexstr, work->data);
 		if (unlikely(new_blocks == 1))
-			goto out_free;
+			return ret;
 
 		work->work_block = ++work_block;
 
@@ -4072,8 +4070,6 @@ static bool test_work_current(struct work *work)
 		}
 	}
 	work->longpoll = false;
-out_free:
-	free(hexstr);
 	return ret;
 }
 
@@ -5467,8 +5463,8 @@ static void *stratum_sthread(void *userdata)
 		quit(1, "Failed to create stratum_q in stratum_sthread");
 
 	while (42) {
+		char noncehex[12], nonce2hex[20];
 		struct stratum_share *sshare;
-		char *noncehex, *nonce2hex;
 		uint32_t *hash32, nonce;
 		char s[1024], nonce2[8];
 		struct work *work;
@@ -5497,7 +5493,7 @@ static void *stratum_sthread(void *userdata)
 		/* This work item is freed in parse_stratum_response */
 		sshare->work = work;
 		nonce = *((uint32_t *)(work->data + 76));
-		noncehex = bin2hex((const unsigned char *)&nonce, 4);
+		__bin2hex(noncehex, (const unsigned char *)&nonce, 4);
 		memset(s, 0, 1024);
 
 		mutex_lock(&sshare_lock);
@@ -5508,15 +5504,11 @@ static void *stratum_sthread(void *userdata)
 		memset(nonce2, 0, 8);
 		/* We only use uint32_t sized nonce2 increments internally */
 		memcpy(nonce2, &work->nonce2, sizeof(uint32_t));
-		nonce2hex = bin2hex((const unsigned char *)nonce2, work->nonce2_len);
+		__bin2hex(nonce2hex, (const unsigned char *)nonce2, work->nonce2_len);
-		if (unlikely(!nonce2hex))
-			quit(1, "Failed to bin2hex nonce2 in stratum_thread");
 
 		snprintf(s, sizeof(s),
 			"{\"params\": [\"%s\", \"%s\", \"%s\", \"%s\", \"%s\"], \"id\": %d, \"method\": \"mining.submit\"}",
 			pool->rpc_user, work->job_id, nonce2hex, work->ntime, noncehex, sshare->id);
-		free(noncehex);
-		free(nonce2hex);
 
 		applog(LOG_INFO, "Submitting share %08lx to pool %d",
 					(long unsigned int)htole32(hash32[6]), pool->pool_no);
@@ -5974,7 +5966,7 @@ static void gen_stratum_work(struct pool *pool, struct work *work)
 	cgtime(&work->tv_staged);
 }
 
-static struct work *get_work(struct thr_info *thr, const int thr_id)
+struct work *get_work(struct thr_info *thr, const int thr_id)
 {
 	struct work *work = NULL;
 
@@ -6532,6 +6524,53 @@ void hash_queued_work(struct thr_info *mythr)
 	cgpu->deven = DEV_DISABLED;
 }
 
+/* This version of hash_work is for devices drivers that want to do their own
+ * work management entirely, usually by using get_work(). Note that get_work
+ * is a blocking function and will wait indefinitely if no work is available
+ * so this must be taken into consideration in the driver. */
+void hash_driver_work(struct thr_info *mythr)
+{
+	struct timeval tv_start = {0, 0}, tv_end;
+	struct cgpu_info *cgpu = mythr->cgpu;
+	struct device_drv *drv = cgpu->drv;
+	const int thr_id = mythr->id;
+	int64_t hashes_done = 0;
+
+	while (likely(!cgpu->shutdown)) {
+		struct timeval diff;
+		int64_t hashes;
+
+		mythr->work_restart = false;
+
+		hashes = drv->scanwork(mythr);
+
+		if (unlikely(hashes == -1 )) {
+			applog(LOG_ERR, "%s %d failure, disabling!", drv->name, cgpu->device_id);
+			cgpu->deven = DEV_DISABLED;
+			dev_error(cgpu, REASON_THREAD_ZERO_HASH);
+			mt_disable(mythr, thr_id, drv);
+		}
+
+		hashes_done += hashes;
+		cgtime(&tv_end);
+		timersub(&tv_end, &tv_start, &diff);
+		/* Update the hashmeter at most 5 times per second */
+		if ((hashes_done && (diff.tv_sec > 0 || diff.tv_usec > 200000)) ||
+		    diff.tv_sec >= opt_log_interval) {
+			hashmeter(thr_id, &diff, hashes_done);
+			hashes_done = 0;
+			copy_time(&tv_start, &tv_end);
+		}
+
+		if (unlikely(mythr->pause || cgpu->deven != DEV_ENABLED))
+			mt_disable(mythr, thr_id, drv);
+
+		if (unlikely(mythr->work_restart))
+			drv->flush_work(cgpu);
+	}
+	cgpu->deven = DEV_DISABLED;
+}
+
 void *miner_thread(void *userdata)
 {
 	struct thr_info *mythr = userdata;
@@ -7180,6 +7219,8 @@ static void clean_up(void)
 	clear_adl(nDevs);
 #endif
 #ifdef USE_USBUTILS
+	usb_polling = false;
+	pthread_join(usb_poll_thread, NULL);
         libusb_exit(NULL);
 #endif
 
@@ -7751,11 +7792,12 @@ static void probe_pools(void)
 #ifdef USE_USBUTILS
 static void *libusb_poll_thread(void __maybe_unused *arg)
 {
+	struct timeval tv_end = {0, 200000};
+
 	RenameThread("usbpoll");
 
-	pthread_detach(pthread_self());
 	while (usb_polling)
-		libusb_handle_events(NULL);
+		libusb_handle_events_timeout_completed(NULL, &tv_end, NULL);
 
 	return NULL;
 }

driver-bitfury.c

@@ -220,18 +220,24 @@ static bool bitfury_checkresults(struct thr_info *thr, struct work *work, uint32
 	return false;
 }
 
-static int64_t bitfury_scanhash(struct thr_info *thr, struct work *work,
+static int64_t bitfury_scanwork(struct thr_info *thr)
-				int64_t __maybe_unused max_nonce)
 {
 	struct cgpu_info *bitfury = thr->cgpu;
 	struct bitfury_info *info = bitfury->device_data;
 	struct timeval tv_now;
+	struct work *work;
 	double nonce_rate;
 	int64_t ret = 0;
 	int amount, i;
 	char buf[45];
 	int ms_diff;
 
+	work = get_work(thr, thr->id);
+	if (unlikely(thr->work_restart)) {
+		free_work(work);
+		return 0;
+	}
+
 	buf[0] = 'W';
 	memcpy(buf + 1, work->midstate, 32);
 	memcpy(buf + 33, work->data + 64, 12);
@@ -298,8 +304,7 @@ static int64_t bitfury_scanhash(struct thr_info *thr, struct work *work,
 cascade:
 	for (i = BF1ARRAY_SIZE; i > 0; i--)
 		info->prevwork[i] = info->prevwork[i - 1];
-	info->prevwork[0] = copy_work(work);
+	info->prevwork[0] = work;
-	work->blk.nonce = 0xffffffff;
 
 	info->cycles++;
 	info->total_nonces += info->nonces;
@@ -358,7 +363,8 @@ struct device_drv bitfury_drv = {
 	.dname = "bitfury",
 	.name = "BF1",
 	.drv_detect = bitfury_detect,
-	.scanhash = bitfury_scanhash,
+	.hash_work = &hash_driver_work,
+	.scanwork = bitfury_scanwork,
 	.get_api_stats = bitfury_api_stats,
 	.reinit_device = bitfury_init,
 	.thread_shutdown = bitfury_shutdown,

miner.h

@@ -952,6 +952,7 @@ extern json_t *json_rpc_call(CURL *curl, const char *url, const char *userpass,
 #endif
 extern const char *proxytype(proxytypes_t proxytype);
 extern char *get_proxy(char *url, struct pool *pool);
+extern void __bin2hex(char *s, const unsigned char *p, size_t len);
 extern char *bin2hex(const unsigned char *p, size_t len);
 extern bool hex2bin(unsigned char *p, const char *hexstr, size_t len);
 
@@ -1384,12 +1385,14 @@ extern void submit_tested_work(struct thr_info *thr, struct work *work);
 extern bool submit_nonce(struct thr_info *thr, struct work *work, uint32_t nonce);
 extern bool submit_noffset_nonce(struct thr_info *thr, struct work *work, uint32_t nonce,
 			  int noffset);
+extern struct work *get_work(struct thr_info *thr, const int thr_id);
 extern struct work *get_queued(struct cgpu_info *cgpu);
 extern struct work *__find_work_bymidstate(struct work *que, char *midstate, size_t midstatelen, char *data, int offset, size_t datalen);
 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 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);
 extern void hash_queued_work(struct thr_info *mythr);
 extern void _wlog(const char *str);
 extern void _wlogprint(const char *str);

ocl.c

@@ -342,6 +342,7 @@ _clState *initCl(unsigned int gpu, char *name, size_t nameSize)
 	/* Check for OpenCL >= 1.0 support, needed for global offset parameter usage. */
 	char * devoclver = malloc(1024);
 	const char * ocl10 = "OpenCL 1.0";
+	const char * ocl11 = "OpenCL 1.1";
 
 	status = clGetDeviceInfo(devices[gpu], CL_DEVICE_VERSION, 1024, (void *)devoclver, NULL);
 	if (status != CL_SUCCESS) {
@@ -349,8 +350,12 @@ _clState *initCl(unsigned int gpu, char *name, size_t nameSize)
 		return NULL;
 	}
 	find = strstr(devoclver, ocl10);
-	if (!find)
+	if (!find) {
 		clState->hasOpenCL11plus = true;
+		find = strstr(devoclver, ocl11);
+		if (!find)
+			clState->hasOpenCL12plus = true;
+	}
 
 	status = clGetDeviceInfo(devices[gpu], CL_DEVICE_PREFERRED_VECTOR_WIDTH_INT, sizeof(cl_uint), (void *)&preferred_vwidth, NULL);
 	if (status != CL_SUCCESS) {
@@ -618,7 +623,8 @@ build:
 	if (clState->hasBitAlign) {
 		strcat(CompilerOptions, " -D BITALIGN");
 		applog(LOG_DEBUG, "cl_amd_media_ops found, setting BITALIGN");
-		if (strstr(name, "Cedar") ||
+		if (!clState->hasOpenCL12plus &&
+		    (strstr(name, "Cedar") ||
 		     strstr(name, "Redwood") ||
 		     strstr(name, "Juniper") ||
 		     strstr(name, "Cypress" ) ||
@@ -630,7 +636,7 @@ build:
 		     strstr(name, "Antilles" ) ||
 		     strstr(name, "Wrestler" ) ||
 		     strstr(name, "Zacate" ) ||
-		    strstr(name, "WinterPark" ))
+		     strstr(name, "WinterPark" )))
 			patchbfi = true;
 	} else
 		applog(LOG_DEBUG, "cl_amd_media_ops not found, will not set BITALIGN");

ocl.h

@@ -27,6 +27,7 @@ typedef struct {
 #endif
 	bool hasBitAlign;
 	bool hasOpenCL11plus;
+	bool hasOpenCL12plus;
 	bool goffset;
 	cl_uint vwidth;
 	size_t max_work_size;

usbutils.c

@@ -2245,7 +2245,6 @@ static int callback_wait(struct usb_transfer *ut, int *transferred, unsigned int
 
 	/* No need to sort out mutexes here since they won't be reused */
 	*transferred = transfer->actual_length;
-	libusb_free_transfer(transfer);
 
 	return ret;
 }
@@ -2296,6 +2295,7 @@ usb_bulk_transfer(struct libusb_device_handle *dev_handle, int intinfo,
 	errn = errno;
 	if (!err)
 		err = callback_wait(&ut, transferred, timeout);
+	libusb_free_transfer(ut.transfer);
 
 	STATS_TIMEVAL(&tv_finish);
 	USB_STATS(cgpu, &tv_start, &tv_finish, err, mode, cmd, seq, timeout);
@@ -2712,12 +2712,9 @@ static int usb_control_transfer(libusb_device_handle *dev_handle, uint8_t bmRequ
 				unsigned char *buffer, uint16_t wLength, unsigned int timeout)
 {
 	struct usb_transfer ut;
+	unsigned char buf[70];
 	int err, transferred;
-	unsigned char *buf;
 
-	buf = malloc(70);
-	if (unlikely(!buf))
-		quit(1, "Failed to malloc buf in usb_control_transfer");
 	init_usb_transfer(&ut);
 	mutex_lock(&ut.mutex);
 	libusb_fill_control_setup(buf, bmRequestType, bRequest, wValue,
@@ -2731,10 +2728,13 @@ static int usb_control_transfer(libusb_device_handle *dev_handle, uint8_t bmRequ
 		unsigned char *ofbuf = libusb_control_transfer_get_data(ut.transfer);
 
 		memcpy(buffer, ofbuf, transferred);
-		return transferred;
+		err = transferred;
+		goto out;
 	}
 	if ((err) == LIBUSB_TRANSFER_CANCELLED)
 		err = LIBUSB_ERROR_TIMEOUT;
+out:
+	libusb_free_transfer(ut.transfer);
 	return err;
 }
 
@@ -2812,21 +2812,22 @@ out_:
 	return err;
 }
 
+/* We use the write devlock for control transfers since some control transfers
+ * are rare but may be changing settings within the device causing problems
+ * if concurrent transfers are happening. Using the write lock serialises
+ * any transfers. */
 int _usb_transfer(struct cgpu_info *cgpu, uint8_t request_type, uint8_t bRequest, uint16_t wValue, uint16_t wIndex, uint32_t *data, int siz, unsigned int timeout, enum usb_cmds cmd)
 {
 	int pstate, err;
 
-	DEVRLOCK(cgpu, pstate);
+	DEVWLOCK(cgpu, pstate);
 
 	err = __usb_transfer(cgpu, request_type, bRequest, wValue, wIndex, data, siz, timeout, cmd);
 
-	if (NOCONTROLDEV(err)) {
+	if (NOCONTROLDEV(err))
-		cg_ruwlock(&cgpu->usbinfo.devlock);
 		release_cgpu(cgpu);
-		cg_dwlock(&cgpu->usbinfo.devlock);
-	}
 
-	DEVRUNLOCK(cgpu, pstate);
+	DEVWUNLOCK(cgpu, pstate);
 
 	return err;
 }
@@ -2840,7 +2841,7 @@ int _usb_transfer_read(struct cgpu_info *cgpu, uint8_t request_type, uint8_t bRe
 	unsigned char tbuf[64];
 	int err, pstate;
 
-	DEVRLOCK(cgpu, pstate);
+	DEVWLOCK(cgpu, pstate);
 
 	USBDEBUG("USB debug: _usb_transfer_read(%s (nodev=%s),type=%"PRIu8",req=%"PRIu8",value=%"PRIu16",index=%"PRIu16",bufsiz=%d,timeout=%u,cmd=%s)", cgpu->drv->name, bool_str(cgpu->usbinfo.nodev), request_type, bRequest, wValue, wIndex, bufsiz, timeout, usb_cmdname(cmd));
 
@@ -2896,13 +2897,10 @@ int _usb_transfer_read(struct cgpu_info *cgpu, uint8_t request_type, uint8_t bRe
 		       err, libusb_error_name(err));
 	}
 out_noerrmsg:
-	if (NOCONTROLDEV(err)) {
+	if (NOCONTROLDEV(err))
-		cg_ruwlock(&cgpu->usbinfo.devlock);
 		release_cgpu(cgpu);
-		cg_dwlock(&cgpu->usbinfo.devlock);
-	}
 
-	DEVRUNLOCK(cgpu, pstate);
+	DEVWUNLOCK(cgpu, pstate);
 
 	return err;
 }

util.c

@@ -584,6 +584,16 @@ char *get_proxy(char *url, struct pool *pool)
 	return url;
 }
 
+/* Adequate size s==len*2 + 1 must be alloced to use this variant */
+void __bin2hex(char *s, const unsigned char *p, size_t len)
+{
+	int i;
+
+	for (i = 0; i < (int)len; i++)
+		sprintf(s + (i * 2), "%02x", (unsigned int)p[i]);
+
+}
+
 /* Returns a malloced array string of a binary value of arbitrary length. The
  * array is rounded up to a 4 byte size to appease architectures that need
  * aligned array  sizes */
@@ -591,7 +601,6 @@ char *bin2hex(const unsigned char *p, size_t len)
 {
 	ssize_t slen;
 	char *s;
-	int i;
 
 	slen = len * 2 + 1;
 	if (slen % 4)
@@ -600,8 +609,7 @@ char *bin2hex(const unsigned char *p, size_t len)
 	if (unlikely(!s))
 		quithere(1, "Failed to calloc");
 
-	for (i = 0; i < (int)len; i++)
+	__bin2hex(s, p, len);
-		sprintf(s + (i * 2), "%02x", (unsigned int)p[i]);
 
 	return s;
 }