linux-nvgpu/drivers/gpu/nvgpu/gk20a/dbg_gpu_gk20a.c

/*
 * Tegra GK20A GPU Debugger/Profiler Driver
 *
 * Copyright (c) 2013-2017, NVIDIA CORPORATION.  All rights reserved.
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice shall be included in
 * all copies or substantial portions of the Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
 * DEALINGS IN THE SOFTWARE.
 */

#include <nvgpu/kmem.h>
#include <nvgpu/log.h>
#include <nvgpu/vm.h>
#include <nvgpu/atomic.h>
#include <nvgpu/mm.h>

#include "gk20a.h"
#include "gk20a/platform_gk20a.h"
#include "gr_gk20a.h"
#include "dbg_gpu_gk20a.h"
#include "regops_gk20a.h"

#include <nvgpu/hw/gk20a/hw_therm_gk20a.h>
#include <nvgpu/hw/gk20a/hw_gr_gk20a.h>
#include <nvgpu/hw/gk20a/hw_perf_gk20a.h>

/*
 * API to get first channel from the list of all channels
 * bound to the debug session
 */
struct channel_gk20a *
nvgpu_dbg_gpu_get_session_channel(struct dbg_session_gk20a *dbg_s)
{
	struct dbg_session_channel_data *ch_data;
	struct channel_gk20a *ch;
	struct gk20a *g = dbg_s->g;

	nvgpu_mutex_acquire(&dbg_s->ch_list_lock);
	if (nvgpu_list_empty(&dbg_s->ch_list)) {
		nvgpu_mutex_release(&dbg_s->ch_list_lock);
		return NULL;
	}

	ch_data = nvgpu_list_first_entry(&dbg_s->ch_list,
				   dbg_session_channel_data,
				   ch_entry);
	ch = g->fifo.channel + ch_data->chid;

	nvgpu_mutex_release(&dbg_s->ch_list_lock);

	return ch;
}

void gk20a_dbg_gpu_post_events(struct channel_gk20a *ch)
{
	struct dbg_session_data *session_data;
	struct dbg_session_gk20a *dbg_s;

	gk20a_dbg(gpu_dbg_fn | gpu_dbg_gpu_dbg, "");

	/* guard against the session list being modified */
	nvgpu_mutex_acquire(&ch->dbg_s_lock);

	nvgpu_list_for_each_entry(session_data, &ch->dbg_s_list,
				dbg_session_data, dbg_s_entry) {
		dbg_s = session_data->dbg_s;
		if (dbg_s->dbg_events.events_enabled) {
			gk20a_dbg(gpu_dbg_gpu_dbg, "posting event on session id %d",
					dbg_s->id);
			gk20a_dbg(gpu_dbg_gpu_dbg, "%d events pending",
					dbg_s->dbg_events.num_pending_events);

			dbg_s->dbg_events.num_pending_events++;

			nvgpu_cond_broadcast_interruptible(&dbg_s->dbg_events.wait_queue);
		}
	}

	nvgpu_mutex_release(&ch->dbg_s_lock);
}

bool gk20a_dbg_gpu_broadcast_stop_trigger(struct channel_gk20a *ch)
{
	struct dbg_session_data *session_data;
	struct dbg_session_gk20a *dbg_s;
	bool broadcast = false;

	gk20a_dbg(gpu_dbg_fn | gpu_dbg_gpu_dbg | gpu_dbg_intr, "");

	/* guard against the session list being modified */
	nvgpu_mutex_acquire(&ch->dbg_s_lock);

	nvgpu_list_for_each_entry(session_data, &ch->dbg_s_list,
				dbg_session_data, dbg_s_entry) {
		dbg_s = session_data->dbg_s;
		if (dbg_s->broadcast_stop_trigger) {
			gk20a_dbg(gpu_dbg_gpu_dbg | gpu_dbg_fn | gpu_dbg_intr,
					"stop trigger broadcast enabled");
			broadcast = true;
			break;
		}
	}

	nvgpu_mutex_release(&ch->dbg_s_lock);

	return broadcast;
}

int gk20a_dbg_gpu_clear_broadcast_stop_trigger(struct channel_gk20a *ch)
{
	struct dbg_session_data *session_data;
	struct dbg_session_gk20a *dbg_s;

	gk20a_dbg(gpu_dbg_fn | gpu_dbg_gpu_dbg | gpu_dbg_intr, "");

	/* guard against the session list being modified */
	nvgpu_mutex_acquire(&ch->dbg_s_lock);

	nvgpu_list_for_each_entry(session_data, &ch->dbg_s_list,
				dbg_session_data, dbg_s_entry) {
		dbg_s = session_data->dbg_s;
		if (dbg_s->broadcast_stop_trigger) {
			gk20a_dbg(gpu_dbg_gpu_dbg | gpu_dbg_fn | gpu_dbg_intr,
					"stop trigger broadcast disabled");
			dbg_s->broadcast_stop_trigger = false;
		}
	}

	nvgpu_mutex_release(&ch->dbg_s_lock);

	return 0;
}

int dbg_set_powergate(struct dbg_session_gk20a *dbg_s, u32  powermode)
{
	int err = 0;
	struct gk20a *g = dbg_s->g;

	 /* This function must be called with g->dbg_sessions_lock held */

	gk20a_dbg(gpu_dbg_fn|gpu_dbg_gpu_dbg, "%s powergate mode = %d",
		   g->name, powermode);

	switch (powermode) {
	case NVGPU_DBG_GPU_POWERGATE_MODE_DISABLE:
		/* save off current powergate, clk state.
		 * set gpu module's can_powergate = 0.
		 * set gpu module's clk to max.
		 * while *a* debug session is active there will be no power or
		 * clocking state changes allowed from mainline code (but they
		 * should be saved).
		 */
		/* Allow powergate disable if the current dbg_session doesn't
		 * call a powergate disable ioctl and the global
		 * powergating_disabled_refcount is zero
		 */

		if ((dbg_s->is_pg_disabled == false) &&
		    (g->dbg_powergating_disabled_refcount++ == 0)) {

			gk20a_dbg(gpu_dbg_gpu_dbg | gpu_dbg_fn, "module busy");
			err = gk20a_busy(g);
			if (err)
				return err;

			/*do elpg disable before clock gating */
			nvgpu_pmu_pg_global_enable(g, false);

			if (g->ops.clock_gating.slcg_gr_load_gating_prod)
				g->ops.clock_gating.slcg_gr_load_gating_prod(g,
						false);
			if (g->ops.clock_gating.slcg_perf_load_gating_prod)
				g->ops.clock_gating.slcg_perf_load_gating_prod(g,
						false);
			if (g->ops.clock_gating.slcg_ltc_load_gating_prod)
				g->ops.clock_gating.slcg_ltc_load_gating_prod(g,
						false);

			gr_gk20a_init_cg_mode(g, BLCG_MODE, BLCG_RUN);
			g->elcg_enabled = false;
			gr_gk20a_init_cg_mode(g, ELCG_MODE, ELCG_RUN);

		}

		dbg_s->is_pg_disabled = true;
		break;

	case NVGPU_DBG_GPU_POWERGATE_MODE_ENABLE:
		/* restore (can) powergate, clk state */
		/* release pending exceptions to fault/be handled as usual */
		/*TBD: ordering of these? */

		/* Re-enabling powergate as no other sessions want
		 * powergate disabled and the current dbg-sessions had
		 * requested the powergate disable through ioctl
		*/
		if (dbg_s->is_pg_disabled &&
		    --g->dbg_powergating_disabled_refcount == 0) {

			g->elcg_enabled = true;
			gr_gk20a_init_cg_mode(g, ELCG_MODE, ELCG_AUTO);
			gr_gk20a_init_cg_mode(g, BLCG_MODE, BLCG_AUTO);

			if (g->ops.clock_gating.slcg_ltc_load_gating_prod)
				g->ops.clock_gating.slcg_ltc_load_gating_prod(g,
						g->slcg_enabled);
			if (g->ops.clock_gating.slcg_perf_load_gating_prod)
				g->ops.clock_gating.slcg_perf_load_gating_prod(g,
						g->slcg_enabled);
			if (g->ops.clock_gating.slcg_gr_load_gating_prod)
				g->ops.clock_gating.slcg_gr_load_gating_prod(g,
						g->slcg_enabled);

			nvgpu_pmu_pg_global_enable(g, true);

			gk20a_dbg(gpu_dbg_gpu_dbg | gpu_dbg_fn, "module idle");
			gk20a_idle(g);
		}

		dbg_s->is_pg_disabled = false;
		break;

	default:
		nvgpu_err(g,
			   "unrecognized dbg gpu powergate mode: 0x%x",
			   powermode);
		err = -ENOTTY;
		break;
	}

	gk20a_dbg(gpu_dbg_fn|gpu_dbg_gpu_dbg, "%s powergate mode = %d done",
		   g->name, powermode);
	return err;
}

bool nvgpu_check_and_set_global_reservation(
				struct dbg_session_gk20a *dbg_s,
				struct dbg_profiler_object_data *prof_obj)
{
	struct gk20a *g = dbg_s->g;

	if (g->profiler_reservation_count == 0) {
		g->global_profiler_reservation_held = true;
		g->profiler_reservation_count = 1;
		dbg_s->has_profiler_reservation = true;
		prof_obj->has_reservation = true;
		return true;
	}
	return false;
}

bool nvgpu_check_and_set_context_reservation(
				struct dbg_session_gk20a *dbg_s,
				struct dbg_profiler_object_data *prof_obj)
{
	struct gk20a *g = dbg_s->g;

	/* Assumes that we've already checked that no global reservation
	 * is in effect.
	 */
	g->profiler_reservation_count++;
	dbg_s->has_profiler_reservation = true;
	prof_obj->has_reservation = true;
	return true;
}

void nvgpu_release_profiler_reservation(struct dbg_session_gk20a *dbg_s,
				struct dbg_profiler_object_data *prof_obj)
{
	struct gk20a *g = dbg_s->g;

	g->profiler_reservation_count--;
	if (g->profiler_reservation_count < 0)
		nvgpu_err(g, "Negative reservation count!");
	dbg_s->has_profiler_reservation = false;
	prof_obj->has_reservation = false;
	if (prof_obj->ch == NULL)
		g->global_profiler_reservation_held = false;
}

int gk20a_perfbuf_enable_locked(struct gk20a *g, u64 offset, u32 size)
{
	struct mm_gk20a *mm = &g->mm;
	u32 virt_addr_lo;
	u32 virt_addr_hi;
	u32 inst_pa_page;
	int err;

	err = gk20a_busy(g);
	if (err) {
		nvgpu_err(g, "failed to poweron");
		return err;
	}

	err = g->ops.mm.alloc_inst_block(g, &mm->perfbuf.inst_block);
	if (err)
		return err;

	g->ops.mm.init_inst_block(&mm->perfbuf.inst_block, mm->perfbuf.vm, 0);

	virt_addr_lo = u64_lo32(offset);
	virt_addr_hi = u64_hi32(offset);

	/* address and size are aligned to 32 bytes, the lowest bits read back
	 * as zeros */
	gk20a_writel(g, perf_pmasys_outbase_r(), virt_addr_lo);
	gk20a_writel(g, perf_pmasys_outbaseupper_r(),
			perf_pmasys_outbaseupper_ptr_f(virt_addr_hi));
	gk20a_writel(g, perf_pmasys_outsize_r(), size);

	/* this field is aligned to 4K */
	inst_pa_page = nvgpu_inst_block_addr(g,	&mm->perfbuf.inst_block) >> 12;

	/* A write to MEM_BLOCK triggers the block bind operation. MEM_BLOCK
	 * should be written last */
	gk20a_writel(g, perf_pmasys_mem_block_r(),
			perf_pmasys_mem_block_base_f(inst_pa_page) |
			perf_pmasys_mem_block_valid_true_f() |
			perf_pmasys_mem_block_target_lfb_f());

	gk20a_idle(g);
	return 0;
}

/* must be called with dbg_sessions_lock held */
int gk20a_perfbuf_disable_locked(struct gk20a *g)
{
	int err = gk20a_busy(g);
	if (err) {
		nvgpu_err(g, "failed to poweron");
		return err;
	}

	gk20a_writel(g, perf_pmasys_outbase_r(), 0);
	gk20a_writel(g, perf_pmasys_outbaseupper_r(),
			perf_pmasys_outbaseupper_ptr_f(0));
	gk20a_writel(g, perf_pmasys_outsize_r(), 0);

	gk20a_writel(g, perf_pmasys_mem_block_r(),
			perf_pmasys_mem_block_base_f(0) |
			perf_pmasys_mem_block_valid_false_f() |
			perf_pmasys_mem_block_target_f(0));

	gk20a_idle(g);

	return 0;
}