From a466942fd880d157cc0b5968805b2159f556fc20 Mon Sep 17 00:00:00 2001 From: Con Kolivas Date: Fri, 12 Aug 2011 20:20:42 +1000 Subject: [PATCH] When pinging a sick cpu, flush finish and then ping it in a separate thread in the hope it recovers, but without blocking code elsewhere. --- main.c | 38 +++++++++++++++++++++++++++++++++++++- 1 file changed, 37 insertions(+), 1 deletion(-) diff --git a/main.c b/main.c index 5d081aa3..7058fcf4 100644 --- a/main.c +++ b/main.c @@ -3415,10 +3415,46 @@ static void *reinit_gpu(void *userdata) return NULL; } + +static void *ping_gputhread(void *userdata) +{ + struct cgpu_info *cgpu = (struct cgpu_info *)userdata; + int gpu = cgpu->cpu_gpu; + struct thr_info *thr; + _clState *clState; + int thr_id; + + for (thr_id = 0; thr_id < gpu_threads; thr_id ++) { + if (dev_from_id(thr_id) != gpu) + continue; + + thr = &thr_info[thr_id]; + clState = clStates[thr_id]; + tq_push(thr->q, &ping); + applog(LOG_WARNING, "Attempting to flush command queue of thread %d", thr_id); + clFlush(clState->commandQueue); + clFinish(clState->commandQueue); + tq_push(thr->q, &ping); + } + + return NULL; +} + +static void ping_gpu(struct cgpu_info *cgpu) +{ + pthread_t ping_thread; + + if (unlikely(pthread_create(&ping_thread, NULL, ping_gputhread, (void *)cgpu))) + applog(LOG_ERR, "Failed to create ping thread"); +} #else static void *reinit_gpu(void *userdata) { } + +static void ping_gpu(struct cgpu_info *cgpu) +{ +} #endif static void reinit_device(struct cgpu_info *cgpu) @@ -3531,7 +3567,7 @@ static void *watchdog_thread(void *userdata) gpus[gpu].status = LIFE_SICK; applog(LOG_ERR, "Thread %d idle for more than 60 seconds, GPU %d declared SICK!", i, gpu); /* Sent it a ping, it might respond */ - tq_push(thr->q, &ping); + ping_gpu(thr->cgpu); } else if (now.tv_sec - thr->last.tv_sec > 300 && gpus[i].status == LIFE_SICK) { gpus[gpu].status = LIFE_DEAD; applog(LOG_ERR, "Thread %d idle for more than 5 minutes, GPU %d declared DEAD!", i, gpu);