bfgminer/driver-alchemist.c

/*
 * Copyright 2015 John Stefanopoulos
 * Copyright 2014-2015 Luke Dashjr
 *
 * This program is free software; you can redistribute it and/or modify it
 * under the terms of the GNU General Public License as published by the Free
 * Software Foundation; either version 3 of the License, or (at your option)
 * any later version.  See COPYING for more details.
 */

// THIS DRIVER REQUIRES THE LATEST FIRMWARE FOR AlCHEMINERS DEVELOPED BY JSTEFANOP
// IT WILL NOT WORK WITH THE STOCK FACTORY FIRMWARE ON THE BOARDS
// PLEASE CONTACT JSTEFANOP AT MAC DOT COM OR JSTEFANOP ON LITECOINTALK DOT ORG FOR MORE INFO

#include "config.h"

#include <stdbool.h>
#include <stdint.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
#include <stdio.h>
#include <fcntl.h>

#include "deviceapi.h"
#include "logging.h"
#include "lowlevel.h"
#include "lowl-vcom.h"
#include "util.h"

static const uint8_t alchemist_max_chips = 0x20;
#define ALCHEMIST_DEFAULT_FREQUENCY  352
#define ALCHEMIST_MIN_CLOCK          200
#define ALCHEMIST_MAX_CLOCK          400
// Number of seconds full board of 1728 cores @ 352mhz takes to scan full range
#define ALCHEMIST_HASH_SPEED         134.0
#define ALCHEMIST_MAX_NONCE          0xffffffff
#define ALCHEMIST_READ_SIZE            9
#define alchemist_max_clusters_per_chip  6
#define alchemist_max_cores_per_cluster  9
static const uint8_t alchemist_g_head[] = {
	0xd4, 0x59, 0x2d, 0x01, 0x1d, 0x01, 0x8e, 0xa7, 0x4e, 0xbb, 0x17, 0xb8, 0x06, 0x6b, 0x2a, 0x75,
	0x83, 0x99, 0xd5, 0xf1, 0x9b, 0x5c, 0x60, 0x73, 0xd0, 0x9b, 0x50, 0x0d, 0x92, 0x59, 0x82, 0xad,
	0xc4, 0xb3, 0xed, 0xd3, 0x52, 0xef, 0xe1, 0x46, 0x67, 0xa8, 0xca, 0x9f, 0x27, 0x9f, 0x63, 0x30,
	0xcc, 0xbb, 0xb9, 0x10, 0x3b, 0x9e, 0x3a, 0x53, 0x50, 0x76, 0x50, 0x52, 0x08, 0x1d, 0xdb, 0xae,
	0x89, 0x8f, 0x1e, 0xf6, 0xb8, 0xc6, 0x4f, 0x3b, 0xce, 0xf7, 0x15, 0xf6,    0,    0,    0,    1,
	   0,    0,    0,    1, 0x8e, 0xa7,    0,    0,    0,    0,    0,    0,    0,    0,    0,    0,
	   0,    0,    0,    0,    0,    0,    0
};

BFG_REGISTER_DRIVER(alchemist_drv)

static const struct bfg_set_device_definition alchemist_set_device_funcs_probe[];

struct alchemist_chip {
	uint8_t chipid;
	uint8_t global_reg[8];
	uint16_t chip_mask[alchemist_max_clusters_per_chip];
	uint32_t clst_offset[alchemist_max_clusters_per_chip];
	unsigned active_cores;
	unsigned freq;
};

static
void alchemist_chip_init(struct alchemist_chip * const chip, const uint8_t chipid)
{
	*chip = (struct alchemist_chip){
		.chipid = chipid,
		.global_reg = {0, 4, 0x40, 0, 0, 0, 0, 1},
		.chip_mask = {0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000},
		.clst_offset = {0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000},
		.active_cores = 1728,
		.freq = ALCHEMIST_DEFAULT_FREQUENCY,
	};
}

static
void alchemist_reset_board(const char * const devpath)
{
	char gpio[8];
	int fd;
	char buf[50];
	
	// TODO: allow custom reset pin through --set for non-stock controller
	if (strcmp(devpath, "/dev/ttyO1") == 0)
		strcpy(gpio, "gpio117");
	else if (strcmp(devpath, "/dev/ttyO2") == 0)
		strcpy(gpio, "gpio110");
	else if (strcmp(devpath, "/dev/ttyO3") == 0)
		strcpy(gpio, "gpio111");
	else if (strcmp(devpath, "/dev/ttyO4") == 0)
		strcpy(gpio, "gpio112");
	else if (strcmp(devpath, "/dev/ttyUSB0") == 0)
		strcpy(gpio, "gpio113");
	else if (strcmp(devpath, "/dev/ttyUSB1") == 0)
		strcpy(gpio, "gpio114");
	else if (strcmp(devpath, "/dev/ttyUSB2") == 0)
		strcpy(gpio, "gpio115");
	else if (strcmp(devpath, "/dev/ttyUSB3") == 0)
		strcpy(gpio, "gpio116");
	else if (strcmp(devpath, "/dev/ttyAMA0") == 0)
		strcpy(gpio, "gpio25");
	else
		return;
	
	sprintf(buf, "/sys/class/gpio/%s/value", gpio);
	
	fd = open(buf, O_WRONLY);
	
	if (write(fd, "0", 1) != 1)
		applog(LOG_DEBUG, "%s: %s %s", alchemist_drv.dname, "GPIO write error", devpath);
	cgsleep_ms(100);
	if (write(fd, "1", 1) != 1)
		applog(LOG_DEBUG, "%s: %s %s", alchemist_drv.dname, "GPIO write error", devpath);
	
	close(fd);
}

static
void alchemist_set_diag_mode(struct alchemist_chip * const chip, bool diag_enable)
{
	if (diag_enable)
		chip->global_reg[1] |= 1;
	else
		chip->global_reg[1] &= ~1;
}

static
bool alchemist_write_global_reg(const int fd, const struct alchemist_chip * const chip)
{
	uint8_t buf[113];
	memset(&buf, 0, 102);
	memcpy(&buf[102], &chip->global_reg[0], 8);
	buf[110] = 0;
	buf[111] = 0xff;
	buf[112] = chip->chipid;
	
	//char output[(sizeof(chip->global_reg) * 2) + 1];
	//bin2hex(output, chip->global_reg, sizeof(chip->global_reg));
	//applog(LOG_DEBUG, "GLOBAL REG %s", output);
	
	if (write(fd, buf, sizeof(buf)) != sizeof(buf))
		return false;
	return true;
}

static
bool alchemist_write_cluster_reg(const int fd, const struct alchemist_chip * const chip, const uint16_t cores_active, const uint32_t offset, const uint8_t clstid)
{
	uint8_t buf[113];
	memset(&buf, 0, 104);
	pk_u16be(buf, 104, cores_active);
	pk_u32be(buf, 106, offset);
	buf[110] = clstid;
	buf[111] = 0xfe;
	buf[112] = chip->chipid;
	//applog(LOG_DEBUG, " %u: %u: %u : %u", buf[106], buf[107], buf[108], buf[109]);
	if (write(fd, buf, sizeof(buf)) != sizeof(buf))
		return false;
	return true;
}


static
bool alchemist_init_pll(const int fd, struct alchemist_chip * const chip)
{
	unsigned freq = chip->freq;
	uint8_t divider = (freq - 16)/16;
	
	divider <<= 1;
	
	uint8_t bytes1 = 0x60 | ((divider & 0xf0) >> 4);
	uint8_t bytes2 = 0x20 | ((divider & 0xf0) >> 4);
	uint8_t bytes3 = 0x00 | ((divider & 0x0f) << 4);
	
	pk_u16be(chip->global_reg, 2, 0x4000);
	chip->global_reg[1] |= 0xc;
	if (!alchemist_write_global_reg(fd, chip))
		return false;
	
	chip->global_reg[2] = bytes1;
	chip->global_reg[3] = bytes3;
	cgsleep_ms(20);
	if (!alchemist_write_global_reg(fd, chip))
		return false;
	
	chip->global_reg[2] = bytes2;
	chip->global_reg[1] &= ~8;
	cgsleep_ms(20);
	if (!alchemist_write_global_reg(fd, chip))
		return false;
	
	chip->global_reg[1] &= ~4;
	cgsleep_ms(20);
	if (!alchemist_write_global_reg(fd, chip))
		return false;
	
	return true;
}

static
bool alchemist_send_golden(const int fd, const struct alchemist_chip * const chip, const bool diag, const void * const data, const void * const target_p)
{
	uint8_t buf[113];
	const uint8_t * const target = target_p;
	
	memcpy(buf, data, 80);
	if (target && !target[0x1f])
		memcpy(&buf[80], target, 0x20);
	else
	{
		memset(&buf[80], 0xff, 0x1f);
		buf[111] = 0;
	}
	buf[112] = chip->chipid;
	if (diag)
		buf[112] |= 0x80;
	if (write(fd, buf, sizeof(buf)) != sizeof(buf))
		return false;
	return true;
}

static
bool alchemist_send_work(const struct thr_info * const thr, struct work * const work)
{
	struct cgpu_info *device = thr->cgpu;
	uint8_t buf[113];
	uint8_t cmd[113];
	const uint8_t * const target = work->target;
	
	unsigned char swpdata[80];
	
	buf[0] = 0xff;
	
	memset(&buf[1], 0, 0x18);
	memcpy(&buf[25], &target[24], 0x8);
	//pk_u64be(buf, 25, 0x0000feff01000000);
	
	swap32tobe(swpdata, work->data, 80/4);
	memcpy(&buf[33], swpdata, 80);
	
	for (int i = 0; i<113; i++) {
		cmd[i] = buf[112 - i];
	}
	
	//char output[(sizeof(cmd) * 2) + 1];
	//bin2hex(output, cmd, sizeof(cmd));
	//applog(LOG_DEBUG, "OUTPUT %s", output);
	
	if (write(device->device_fd, cmd, sizeof(cmd)) != sizeof(cmd))
		return false;
	
	work->blk.nonce = ALCHEMIST_MAX_NONCE;
	
	return true;
}

static
bool alchemist_detect_one(const char * const devpath)
{
	struct alchemist_chip *chips = NULL;
	const int fd = serial_open(devpath, 115200, 1, true);
	if (fd < 0)
		return_via_applog(err, , LOG_DEBUG, "%s: %s %s", alchemist_drv.dname, "Failed to open", devpath);
	
	applog(LOG_DEBUG, "%s: %s %s", alchemist_drv.dname, "Successfully opened", devpath);
	
	alchemist_reset_board(devpath);
	
	// Init chips, setup PLL, and scan for good cores
	chips = malloc(alchemist_max_chips * sizeof(*chips));
	
	struct alchemist_chip * const dummy_chip = &chips[0];
	alchemist_chip_init(dummy_chip, 0);
	
	// pick up any user-defined settings passed in via --set
	drv_set_defaults(&alchemist_drv, alchemist_set_device_funcs_probe, dummy_chip, devpath, detectone_meta_info.serial, 1);
	
	unsigned freq = dummy_chip->freq;
	
	unsigned total_cores = 0;
	{
		uint8_t buf[9];
		for (unsigned i = 0; i < alchemist_max_chips; ++i)
		{
			struct alchemist_chip * const chip = &chips[i];
			alchemist_chip_init(chip, i);
			chip->freq = freq;
			alchemist_set_diag_mode(chip, true);
			if (!alchemist_init_pll(fd, chip))
				return_via_applog(err, , LOG_DEBUG, "%s: Failed to (%s) %s", alchemist_drv.dname, "init PLL", devpath);
			if (!alchemist_send_golden(fd, chip, true, alchemist_g_head, NULL))
				return_via_applog(err, , LOG_DEBUG, "%s: Failed to (%s) %s", alchemist_drv.dname, "send scan job", devpath);
			
			while (serial_read(fd, buf, 9) == 9)
			{
				const uint8_t chipid = buf[8];
				if (chipid >= alchemist_max_chips)
					applog(LOG_DEBUG, "%s: Bad %s id (%u) during scan of %s chip %u", alchemist_drv.dname, "chip", chipid, devpath, i);
				const uint8_t clsid = buf[7];
				if (clsid >= alchemist_max_clusters_per_chip)
					applog(LOG_DEBUG, "%s: Bad %s id (%u) during scan of %s chip %u", alchemist_drv.dname, "cluster", clsid, devpath, i);
				const uint8_t coreid = buf[6];
				if (coreid >= alchemist_max_cores_per_cluster)
					applog(LOG_DEBUG, "%s: Bad %s id (%u) during scan of %s chip %u", alchemist_drv.dname, "core", coreid, devpath, i);
				
				if (buf[0] != 0xd9 || buf[1] != 0xeb || buf[2] != 0x86 || buf[3] != 0x63) {
					//chips[i].chip_good[clsid][coreid] = false;
					applog(LOG_DEBUG, "%s: Bad %s at core (%u) during scan of %s chip %u cluster %u", alchemist_drv.dname, "nonce", coreid, devpath, i, clsid);
				} else {
					++total_cores;
					chips[i].chip_mask[clsid] |= (1 << coreid);
				}
			}
		}
	}
	
	applog(LOG_DEBUG, "%s: Identified %d cores on %s", alchemist_drv.dname, total_cores, devpath);
	if (!total_cores)
		goto err;
	
	alchemist_reset_board(devpath);
	
	// config nonce ranges per cluster based on core responses
	unsigned mutiple = ALCHEMIST_MAX_NONCE / total_cores;
	uint32_t n_offset = 0x00000000;
	
	for (unsigned i = 0; i < alchemist_max_chips; ++i)
	{
		struct alchemist_chip * const chip = &chips[i];
		
		chips[i].active_cores = total_cores;
		
		alchemist_set_diag_mode(chip, false);
		if (!alchemist_init_pll(fd, chip))
			return_via_applog(err, , LOG_DEBUG, "%s: Failed to (%s) %s", alchemist_drv.dname, "init PLL", devpath);
		
		cgsleep_ms(10);
		
		for (unsigned x = 0; x < alchemist_max_clusters_per_chip; ++x) {
			unsigned gc = 0;
			
			uint16_t core_mask = chips[i].chip_mask[x];
			chips[i].clst_offset[x] = n_offset;
			
			//applog(LOG_DEBUG, "OFFSET %u CHIP %u CLUSTER %u", n_offset, i, x);
			
			if (!alchemist_write_cluster_reg(fd, chip, core_mask, n_offset, x))
				return_via_applog(err, , LOG_DEBUG, "%s: Failed to (%s) %s", alchemist_drv.dname, "send config register", devpath);
			
			for (unsigned z = 0; z < 15; ++z) {
				if (core_mask & 0x0001)
					gc += 1;
				core_mask >>= 1;
			}
			
			n_offset += mutiple * gc;
		}
	}
	
	if (serial_claim_v(devpath, &alchemist_drv))
		goto err;
	
	//serial_close(fd);
	struct cgpu_info * const cgpu = malloc(sizeof(*cgpu));
	*cgpu = (struct cgpu_info){
		.drv = &alchemist_drv,
		.device_path = strdup(devpath),
		.deven = DEV_ENABLED,
		.procs = 1,
		.threads = 1,
		.device_data = chips,
	};
	// NOTE: Xcode's clang has a bug where it cannot find fields inside anonymous unions (more details in fpgautils)
	cgpu->device_fd = fd;
	
	return add_cgpu(cgpu);

err:
	if (fd >= 0)
		serial_close(fd);
	free(chips);
	return false;
}

/*
 * scanhash mining loop
 */

static
void alchemist_submit_nonce(struct thr_info * const thr, const uint8_t buf[9], struct work * const work)
{
	struct cgpu_info *device = thr->cgpu;
	struct alchemist_chip *chips = device->device_data;
	
	uint32_t nonce = *(uint32_t *)buf;
	nonce = bswap_32(nonce);
	
	submit_nonce(thr, work, nonce);
	
	// hashrate calc
	
	const uint8_t chipid = buf[8];
	const uint8_t clstid = buf[7];
	uint32_t range = chips[chipid].clst_offset[clstid];
	uint32_t mutiple = ALCHEMIST_MAX_NONCE / chips[chipid].active_cores;
	
	double diff_mutiple = .5/work->work_difficulty;
	
	for (unsigned x = 0; x < alchemist_max_cores_per_cluster; ++x) {
		if (nonce > range && nonce < (range + mutiple)) {
			uint64_t hashes = (nonce - range) * chips[chipid].active_cores * diff_mutiple;
			
			if (hashes > ALCHEMIST_MAX_NONCE)
				hashes = 1;
			
			hashes_done2(thr, hashes, NULL);
		}
		
		range += mutiple;
	}
}

// send work to the device
static
int64_t alchemist_scanhash(struct thr_info *thr, struct work *work, int64_t __maybe_unused max_nonce)
{
	struct cgpu_info *device = thr->cgpu;
	int fd = device->device_fd;
	struct alchemist_chip *chips = device->device_data;
	struct timeval start_tv, nonce_range_tv;
	
	// amount of time it takes this device to scan a nonce range:
	uint32_t nonce_full_range_sec = ALCHEMIST_HASH_SPEED * 352.0 / ALCHEMIST_DEFAULT_FREQUENCY * 1728.0 / chips[0].active_cores;
	// timer to break out of scanning should we close in on an entire nonce range
	// should break out before the range is scanned, so we are doing 95% of the range
	uint64_t nonce_near_range_usec = (nonce_full_range_sec * 1000000. * 0.95);
	timer_set_delay_from_now(&nonce_range_tv, nonce_near_range_usec);
	
	// start the job
	timer_set_now(&start_tv);
	
	if (!alchemist_send_work(thr, work)) {
		applog(LOG_DEBUG, "Failed to start job");
		dev_error(device, REASON_DEV_COMMS_ERROR);
	}
	
	uint8_t buf[9];
	int read = 0;
	bool range_nearly_scanned = false;
	
	while (!thr->work_restart                                              // true when new work is available (miner.c)
	    && ((read = serial_read(fd, buf, 9)) >= 0)                         // only check for failure - allow 0 bytes
	    && !(range_nearly_scanned = timer_passed(&nonce_range_tv, NULL)))  // true when we've nearly scanned a nonce range
	{
		if (read == 0)
			continue;
		
		if (read == 9) {
			alchemist_submit_nonce(thr, buf, work);
		}
		else
			applog(LOG_ERR, "%"PRIpreprv": Unrecognized response", device->proc_repr);
	}
	
	if (read == -1)
	{
		applog(LOG_ERR, "%s: Failed to read result", device->dev_repr);
		dev_error(device, REASON_DEV_COMMS_ERROR);
	}
	
	return 0;
}

/*
 * setup & shutdown
 */

static
bool alchemist_lowl_probe(const struct lowlevel_device_info * const info)
{
	return vcom_lowl_probe_wrapper(info, alchemist_detect_one);
}

static
void alchemist_thread_shutdown(struct thr_info *thr)
{
	struct cgpu_info *device = thr->cgpu;
	
	alchemist_reset_board(device->device_path);
	
	serial_close(device->device_fd);
}

/*
 * specify settings / options via RPC or command line
 */

// support for --set
// must be set before probing the device

// for setting clock and chips during probe / detect

static
const char *alchemist_set_clock(struct cgpu_info * const device, const char * const option, const char * const setting, char * const replybuf, enum bfg_set_device_replytype * const success)
{
	struct alchemist_chip * const chip = device->device_data;
	int val = atoi(setting);
	
	if (val < ALCHEMIST_MIN_CLOCK || val > ALCHEMIST_MAX_CLOCK || (val%16)) {
		sprintf(replybuf, "invalid clock: '%s' valid range %d-%d and a mutiple of 16",
		        setting, ALCHEMIST_MIN_CLOCK, ALCHEMIST_MAX_CLOCK);
		return replybuf;
	} else
		chip->freq = val;
	
	return NULL;
}

static
const struct bfg_set_device_definition alchemist_set_device_funcs_probe[] = {
	{ "clock", alchemist_set_clock, NULL },
	{ NULL },
};

struct device_drv alchemist_drv = {
	.dname = "alchemist",
	.name = "ALC",
	.drv_min_nonce_diff = common_scrypt_min_nonce_diff,
	// detect device
	.lowl_probe = alchemist_lowl_probe,
	// specify mining type - scanhash
	.minerloop = minerloop_scanhash,
	
	// scanhash mining hooks
	.scanhash = alchemist_scanhash,
	
	// teardown device
	.thread_shutdown = alchemist_thread_shutdown,
};