diff --git a/drivers/gpu/nvgpu/Makefile.nvgpu b/drivers/gpu/nvgpu/Makefile.nvgpu index 658f8b7f4..983df2427 100644 --- a/drivers/gpu/nvgpu/Makefile.nvgpu +++ b/drivers/gpu/nvgpu/Makefile.nvgpu @@ -23,6 +23,7 @@ endif obj-$(CONFIG_GK20A) := nvgpu.o nvgpu-y := \ + common/linux/module.o \ common/linux/kmem.o \ common/linux/timers.o \ common/linux/ioctl.o \ @@ -34,13 +35,13 @@ nvgpu-y := \ common/linux/nvgpu_mem.o \ common/linux/dma.o \ common/linux/soc.o \ + common/linux/driver_common.o \ common/mm/nvgpu_allocator.o \ common/mm/bitmap_allocator.o \ common/mm/buddy_allocator.o \ common/mm/page_allocator.o \ common/mm/lockless_allocator.o \ common/pramin.o \ - common/nvgpu_common.o \ common/semaphore.o \ common/as.o \ common/rbtree.o \ @@ -105,7 +106,7 @@ nvgpu-y := \ nvgpu-$(CONFIG_TEGRA_GK20A) += tegra/linux/platform_gk20a_tegra.o nvgpu-$(CONFIG_SYNC) += gk20a/sync_gk20a.o -nvgpu-$(CONFIG_GK20A_PCI) += pci.o +nvgpu-$(CONFIG_GK20A_PCI) += common/linux/pci.o nvgpu-$(CONFIG_TEGRA_GR_VIRTUALIZATION) += \ gk20a/platform_vgpu_tegra.o \ diff --git a/drivers/gpu/nvgpu/common/nvgpu_common.c b/drivers/gpu/nvgpu/common/linux/driver_common.c similarity index 96% rename from drivers/gpu/nvgpu/common/nvgpu_common.c rename to drivers/gpu/nvgpu/common/linux/driver_common.c index 0c812d34b..5c96b4e8f 100644 --- a/drivers/gpu/nvgpu/common/nvgpu_common.c +++ b/drivers/gpu/nvgpu/common/linux/driver_common.c @@ -24,7 +24,7 @@ #include "gk20a/gk20a_scale.h" #include "gk20a/gk20a.h" -#include "gk20a/gr_gk20a.h" +#include "module.h" #define EMC3D_DEFAULT_RATIO 750 @@ -124,6 +124,20 @@ static void nvgpu_init_mm_vars(struct gk20a *g) nvgpu_mutex_init(&g->mm.priv_lock); } +static int gk20a_secure_page_alloc(struct device *dev) +{ + struct gk20a_platform *platform = dev_get_drvdata(dev); + int err = 0; + + if (platform->secure_page_alloc) { + err = platform->secure_page_alloc(dev); + if (!err) + platform->secure_alloc_ready = true; + } + + return err; +} + int nvgpu_probe(struct gk20a *g, const char *debugfs_symlink, const char *interface_name, diff --git a/drivers/gpu/nvgpu/common/linux/module.c b/drivers/gpu/nvgpu/common/linux/module.c new file mode 100644 index 000000000..2cbf996bd --- /dev/null +++ b/drivers/gpu/nvgpu/common/linux/module.c @@ -0,0 +1,1052 @@ +/* + * GK20A Graphics + * + * Copyright (c) 2011-2017, NVIDIA CORPORATION. All rights reserved. + * + * This program is free software; you can redistribute it and/or modify it + * under the terms and conditions of the GNU General Public License, + * version 2, as published by the Free Software Foundation. + * + * This program is distributed in the hope it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + * + * You should have received a copy of the GNU General Public License + * along with this program. If not, see . + */ + +#include +#include +#include +#include +#include +#include +#include +#include + +#include +#include +#include + +#include "gk20a/gk20a.h" +#include "vgpu/vgpu.h" +#include "gk20a/gk20a_scale.h" +#include "gk20a/ctxsw_trace_gk20a.h" +#include "pci.h" +#include "module.h" +#ifdef CONFIG_TEGRA_19x_GPU +#include "nvgpu_gpuid_t19x.h" +#endif + +#define CLASS_NAME "nvidia-gpu" +/* TODO: Change to e.g. "nvidia-gpu%s" once we have symlinks in place. */ + +#define GK20A_WAIT_FOR_IDLE_MS 2000 + +#define CREATE_TRACE_POINTS +#include + +void gk20a_busy_noresume(struct device *dev) +{ + pm_runtime_get_noresume(dev); +} + +int gk20a_busy(struct gk20a *g) +{ + int ret = 0; + struct device *dev; + + if (!g) + return -ENODEV; + + atomic_inc(&g->usage_count); + + down_read(&g->busy_lock); + + if (!gk20a_can_busy(g)) { + ret = -ENODEV; + atomic_dec(&g->usage_count); + goto fail; + } + + dev = g->dev; + + if (pm_runtime_enabled(dev)) { + ret = pm_runtime_get_sync(dev); + if (ret < 0) { + pm_runtime_put_noidle(dev); + atomic_dec(&g->usage_count); + goto fail; + } + } else { + if (!g->power_on) { + ret = gk20a_gpu_is_virtual(dev) ? + vgpu_pm_finalize_poweron(dev) + : gk20a_pm_finalize_poweron(dev); + if (ret) { + atomic_dec(&g->usage_count); + goto fail; + } + } + } + + gk20a_scale_notify_busy(dev); + +fail: + up_read(&g->busy_lock); + + return ret < 0 ? ret : 0; +} + +void gk20a_idle_nosuspend(struct device *dev) +{ + pm_runtime_put_noidle(dev); +} + +void gk20a_idle(struct gk20a *g) +{ + struct device *dev; + + atomic_dec(&g->usage_count); + down_read(&g->busy_lock); + + dev = g->dev; + + if (!(dev && gk20a_can_busy(g))) + goto fail; + + if (pm_runtime_enabled(dev)) { +#ifdef CONFIG_PM + if (atomic_read(&g->dev->power.usage_count) == 1) + gk20a_scale_notify_idle(dev); +#endif + + pm_runtime_mark_last_busy(dev); + pm_runtime_put_sync_autosuspend(dev); + + } else { + gk20a_scale_notify_idle(dev); + } +fail: + up_read(&g->busy_lock); +} + +int gk20a_pm_finalize_poweron(struct device *dev) +{ + struct gk20a *g = get_gk20a(dev); + struct gk20a_platform *platform = gk20a_get_platform(dev); + int err, nice_value; + + gk20a_dbg_fn(""); + + if (g->power_on) + return 0; + + trace_gk20a_finalize_poweron(dev_name(dev)); + + /* Increment platform power refcount */ + if (platform->busy) { + err = platform->busy(dev); + if (err < 0) { + nvgpu_err(g, "failed to poweron platform dependency"); + return err; + } + } + + err = gk20a_restore_registers(g); + if (err) + return err; + + nice_value = task_nice(current); + set_user_nice(current, -20); + + err = gk20a_finalize_poweron(g); + set_user_nice(current, nice_value); + if (err) + goto done; + + trace_gk20a_finalize_poweron_done(dev_name(dev)); + + enable_irq(g->irq_stall); + if (g->irq_stall != g->irq_nonstall) + enable_irq(g->irq_nonstall); + g->irqs_enabled = 1; + + gk20a_scale_resume(g->dev); + + if (platform->has_cde) + gk20a_init_cde_support(g); + +done: + if (err) + g->power_on = false; + + return err; +} + +static int gk20a_pm_prepare_poweroff(struct device *dev) +{ + struct gk20a *g = get_gk20a(dev); + int ret = 0; + struct gk20a_platform *platform = gk20a_get_platform(dev); + + gk20a_dbg_fn(""); + + nvgpu_mutex_acquire(&g->poweroff_lock); + + if (!g->power_on) + goto done; + + gk20a_scale_suspend(dev); + + ret = gk20a_prepare_poweroff(g); + if (ret) + goto error; + + /* + * After this point, gk20a interrupts should not get + * serviced. + */ + disable_irq(g->irq_stall); + if (g->irq_stall != g->irq_nonstall) + disable_irq(g->irq_nonstall); + + /* + * is_fmodel needs to be in gk20a struct for deferred teardown + */ + g->is_fmodel = platform->is_fmodel; + + /* Decrement platform power refcount */ + if (platform->idle) + platform->idle(dev); + + /* Stop CPU from accessing the GPU registers. */ + gk20a_lockout_registers(g); + + nvgpu_mutex_release(&g->poweroff_lock); + return 0; + +error: + gk20a_scale_resume(dev); +done: + nvgpu_mutex_release(&g->poweroff_lock); + + return ret; +} + +static struct of_device_id tegra_gk20a_of_match[] = { +#ifdef CONFIG_TEGRA_GK20A + { .compatible = "nvidia,tegra124-gk20a", + .data = &gk20a_tegra_platform }, + { .compatible = "nvidia,tegra210-gm20b", + .data = &gm20b_tegra_platform }, +#ifdef CONFIG_ARCH_TEGRA_18x_SOC + { .compatible = "nvidia,tegra186-gp10b", + .data = &gp10b_tegra_platform }, +#endif +#ifdef CONFIG_TEGRA_19x_GPU + { .compatible = TEGRA_19x_GPU_COMPAT_TEGRA, + .data = &t19x_gpu_tegra_platform }, +#endif +#ifdef CONFIG_TEGRA_GR_VIRTUALIZATION + { .compatible = "nvidia,tegra124-gk20a-vgpu", + .data = &vgpu_tegra_platform }, +#endif +#else + { .compatible = "nvidia,tegra124-gk20a", + .data = &gk20a_generic_platform }, + { .compatible = "nvidia,tegra210-gm20b", + .data = &gk20a_generic_platform }, +#ifdef CONFIG_ARCH_TEGRA_18x_SOC + { .compatible = TEGRA_18x_GPU_COMPAT_TEGRA, + .data = &gk20a_generic_platform }, +#endif + +#endif + { .compatible = "nvidia,generic-gk20a", + .data = &gk20a_generic_platform }, + { .compatible = "nvidia,generic-gm20b", + .data = &gk20a_generic_platform }, +#ifdef CONFIG_ARCH_TEGRA_18x_SOC + { .compatible = "nvidia,generic-gp10b", + .data = &gk20a_generic_platform }, +#endif + { }, +}; + +#ifdef CONFIG_PM +/** + * __gk20a_do_idle() - force the GPU to idle and railgate + * + * In success, this call MUST be balanced by caller with __gk20a_do_unidle() + * + * Acquires two locks : &g->busy_lock and &platform->railgate_lock + * In success, we hold these locks and return + * In failure, we release these locks and return + */ +int __gk20a_do_idle(struct device *dev, bool force_reset) +{ + struct gk20a *g = get_gk20a(dev); + struct gk20a_platform *platform = dev_get_drvdata(dev); + struct nvgpu_timeout timeout; + int ref_cnt; + int target_ref_cnt = 0; + bool is_railgated; + int err = 0; + + /* acquire busy lock to block other busy() calls */ + down_write(&g->busy_lock); + + /* acquire railgate lock to prevent unrailgate in midst of do_idle() */ + nvgpu_mutex_acquire(&platform->railgate_lock); + + /* check if it is already railgated ? */ + if (platform->is_railgated(dev)) + return 0; + + /* + * release railgate_lock, prevent suspend by incrementing usage counter, + * re-acquire railgate_lock + */ + nvgpu_mutex_release(&platform->railgate_lock); + pm_runtime_get_sync(dev); + + /* + * One refcount taken in this API + * If User disables rail gating, we take one more + * extra refcount + */ + if (platform->user_railgate_disabled) + target_ref_cnt = 2; + else + target_ref_cnt = 1; + nvgpu_mutex_acquire(&platform->railgate_lock); + + nvgpu_timeout_init(g, &timeout, GK20A_WAIT_FOR_IDLE_MS, + NVGPU_TIMER_CPU_TIMER); + + /* check and wait until GPU is idle (with a timeout) */ + do { + nvgpu_msleep(1); + ref_cnt = atomic_read(&dev->power.usage_count); + } while (ref_cnt != target_ref_cnt && !nvgpu_timeout_expired(&timeout)); + + if (ref_cnt != target_ref_cnt) { + nvgpu_err(g, "failed to idle - refcount %d != 1", + ref_cnt); + goto fail_drop_usage_count; + } + + /* check if global force_reset flag is set */ + force_reset |= platform->force_reset_in_do_idle; + + nvgpu_timeout_init(g, &timeout, GK20A_WAIT_FOR_IDLE_MS, + NVGPU_TIMER_CPU_TIMER); + + if (platform->can_railgate && !force_reset) { + /* + * Case 1 : GPU railgate is supported + * + * if GPU is now idle, we will have only one ref count, + * drop this ref which will rail gate the GPU + */ + pm_runtime_put_sync(dev); + + /* add sufficient delay to allow GPU to rail gate */ + nvgpu_msleep(platform->railgate_delay); + + /* check in loop if GPU is railgated or not */ + do { + nvgpu_msleep(1); + is_railgated = platform->is_railgated(dev); + } while (!is_railgated && !nvgpu_timeout_expired(&timeout)); + + if (is_railgated) { + return 0; + } else { + nvgpu_err(g, "failed to idle in timeout"); + goto fail_timeout; + } + } else { + /* + * Case 2 : GPU railgate is not supported or we explicitly + * do not want to depend on runtime PM + * + * if GPU is now idle, call prepare_poweroff() to save the + * state and then do explicit railgate + * + * __gk20a_do_unidle() needs to unrailgate, call + * finalize_poweron(), and then call pm_runtime_put_sync() + * to balance the GPU usage counter + */ + + /* Save the GPU state */ + err = gk20a_pm_prepare_poweroff(dev); + if (err) + goto fail_drop_usage_count; + + /* railgate GPU */ + platform->railgate(dev); + + nvgpu_udelay(10); + + g->forced_reset = true; + return 0; + } + +fail_drop_usage_count: + pm_runtime_put_noidle(dev); +fail_timeout: + nvgpu_mutex_release(&platform->railgate_lock); + up_write(&g->busy_lock); + return -EBUSY; +} + +/** + * gk20a_do_idle() - wrap up for __gk20a_do_idle() to be called + * from outside of GPU driver + * + * In success, this call MUST be balanced by caller with gk20a_do_unidle() + */ +int gk20a_do_idle(void) +{ + struct device_node *node = + of_find_matching_node(NULL, tegra_gk20a_of_match); + struct platform_device *pdev = of_find_device_by_node(node); + + int ret = __gk20a_do_idle(&pdev->dev, true); + + of_node_put(node); + + return ret; +} + +/** + * __gk20a_do_unidle() - unblock all the tasks blocked by __gk20a_do_idle() + */ +int __gk20a_do_unidle(struct device *dev) +{ + struct gk20a *g = get_gk20a(dev); + struct gk20a_platform *platform = dev_get_drvdata(dev); + int err; + + if (g->forced_reset) { + /* + * If we did a forced-reset/railgate + * then unrailgate the GPU here first + */ + platform->unrailgate(dev); + + /* restore the GPU state */ + err = gk20a_pm_finalize_poweron(dev); + if (err) + return err; + + /* balance GPU usage counter */ + pm_runtime_put_sync(dev); + + g->forced_reset = false; + } + + /* release the lock and open up all other busy() calls */ + nvgpu_mutex_release(&platform->railgate_lock); + up_write(&g->busy_lock); + + return 0; +} + +/** + * gk20a_do_unidle() - wrap up for __gk20a_do_unidle() + */ +int gk20a_do_unidle(void) +{ + struct device_node *node = + of_find_matching_node(NULL, tegra_gk20a_of_match); + struct platform_device *pdev = of_find_device_by_node(node); + + int ret = __gk20a_do_unidle(&pdev->dev); + + of_node_put(node); + + return ret; +} +#endif + +static void __iomem *gk20a_ioremap_resource(struct platform_device *dev, int i, + struct resource **out) +{ + struct resource *r = platform_get_resource(dev, IORESOURCE_MEM, i); + + if (!r) + return NULL; + if (out) + *out = r; + return devm_ioremap_resource(&dev->dev, r); +} + +static irqreturn_t gk20a_intr_isr_stall(int irq, void *dev_id) +{ + struct gk20a *g = dev_id; + + return g->ops.mc.isr_stall(g); +} + +static irqreturn_t gk20a_intr_isr_nonstall(int irq, void *dev_id) +{ + struct gk20a *g = dev_id; + + return g->ops.mc.isr_nonstall(g); +} + +static irqreturn_t gk20a_intr_thread_stall(int irq, void *dev_id) +{ + struct gk20a *g = dev_id; + + return g->ops.mc.isr_thread_stall(g); +} + +void gk20a_remove_support(struct gk20a *g) +{ +#ifdef CONFIG_TEGRA_COMMON + tegra_unregister_idle_unidle(); +#endif + nvgpu_kfree(g, g->dbg_regops_tmp_buf); + + if (g->pmu.remove_support) + g->pmu.remove_support(&g->pmu); + + if (g->gr.remove_support) + g->gr.remove_support(&g->gr); + + if (g->mm.remove_ce_support) + g->mm.remove_ce_support(&g->mm); + + if (g->fifo.remove_support) + g->fifo.remove_support(&g->fifo); + + if (g->mm.remove_support) + g->mm.remove_support(&g->mm); + + if (g->sim.remove_support) + g->sim.remove_support(&g->sim); + + /* free mappings to registers, etc */ + + if (g->regs) { + iounmap(g->regs); + g->regs = NULL; + } + if (g->bar1) { + iounmap(g->bar1); + g->bar1 = NULL; + } +} + +static int gk20a_init_support(struct platform_device *dev) +{ + int err = 0; + struct gk20a *g = get_gk20a(&dev->dev); + +#ifdef CONFIG_TEGRA_COMMON + tegra_register_idle_unidle(gk20a_do_idle, gk20a_do_unidle); +#endif + + g->regs = gk20a_ioremap_resource(dev, GK20A_BAR0_IORESOURCE_MEM, + &g->reg_mem); + if (IS_ERR(g->regs)) { + nvgpu_err(g, "failed to remap gk20a registers"); + err = PTR_ERR(g->regs); + goto fail; + } + + g->bar1 = gk20a_ioremap_resource(dev, GK20A_BAR1_IORESOURCE_MEM, + &g->bar1_mem); + if (IS_ERR(g->bar1)) { + nvgpu_err(g, "failed to remap gk20a bar1"); + err = PTR_ERR(g->bar1); + goto fail; + } + + if (nvgpu_platform_is_simulation(g)) { + g->sim.g = g; + g->sim.regs = gk20a_ioremap_resource(dev, + GK20A_SIM_IORESOURCE_MEM, + &g->sim.reg_mem); + if (IS_ERR(g->sim.regs)) { + nvgpu_err(g, "failed to remap gk20a sim regs"); + err = PTR_ERR(g->sim.regs); + goto fail; + } + + err = gk20a_init_sim_support(dev); + if (err) + goto fail; + } + + return 0; + +fail: + return err; +} + +static int gk20a_pm_railgate(struct device *dev) +{ + struct gk20a_platform *platform = dev_get_drvdata(dev); + int ret = 0; +#ifdef CONFIG_DEBUG_FS + struct gk20a *g = get_gk20a(dev); + + g->pstats.last_rail_gate_start = jiffies; + + if (g->pstats.railgating_cycle_count >= 1) + g->pstats.total_rail_ungate_time_ms = + g->pstats.total_rail_ungate_time_ms + + jiffies_to_msecs(g->pstats.last_rail_gate_start - + g->pstats.last_rail_ungate_complete); +#endif + + if (platform->railgate) + ret = platform->railgate(dev); + +#ifdef CONFIG_DEBUG_FS + g->pstats.last_rail_gate_complete = jiffies; +#endif + + return ret; +} + +static int gk20a_pm_unrailgate(struct device *dev) +{ + struct gk20a_platform *platform = dev_get_drvdata(dev); + int ret = 0; + struct gk20a *g = get_gk20a(dev); + +#ifdef CONFIG_DEBUG_FS + g->pstats.last_rail_ungate_start = jiffies; + if (g->pstats.railgating_cycle_count >= 1) + g->pstats.total_rail_gate_time_ms = + g->pstats.total_rail_gate_time_ms + + jiffies_to_msecs(g->pstats.last_rail_ungate_start - + g->pstats.last_rail_gate_complete); + + g->pstats.railgating_cycle_count++; +#endif + + trace_gk20a_pm_unrailgate(dev_name(dev)); + + if (platform->unrailgate) { + nvgpu_mutex_acquire(&platform->railgate_lock); + ret = platform->unrailgate(dev); + nvgpu_mutex_release(&platform->railgate_lock); + } + +#ifdef CONFIG_DEBUG_FS + g->pstats.last_rail_ungate_complete = jiffies; +#endif + + return ret; +} + +static void gk20a_pm_shutdown(struct platform_device *pdev) +{ + struct gk20a_platform *platform = platform_get_drvdata(pdev); + struct gk20a *g = platform->g; + int err; + + nvgpu_info(g, "shutting down"); + + /* vgpu has nothing to clean up currently */ + if (gk20a_gpu_is_virtual(&pdev->dev)) + return; + + gk20a_driver_start_unload(g); + + /* If GPU is already railgated, + * just prevent more requests, and return */ + if (platform->is_railgated && platform->is_railgated(&pdev->dev)) { + __pm_runtime_disable(&pdev->dev, false); + nvgpu_info(g, "already railgated, shut down complete"); + return; + } + + /* Prevent more requests by disabling Runtime PM */ + __pm_runtime_disable(&pdev->dev, false); + + err = gk20a_wait_for_idle(&pdev->dev); + if (err) { + nvgpu_err(g, "failed to idle GPU, err=%d", err); + goto finish; + } + + err = gk20a_fifo_disable_all_engine_activity(g, true); + if (err) { + nvgpu_err(g, "failed to disable engine activity, err=%d", + err); + goto finish; + } + + err = gk20a_fifo_wait_engine_idle(g); + if (err) { + nvgpu_err(g, "failed to idle engines, err=%d", + err); + goto finish; + } + + if (gk20a_gpu_is_virtual(&pdev->dev)) + err = vgpu_pm_prepare_poweroff(&pdev->dev); + else + err = gk20a_pm_prepare_poweroff(&pdev->dev); + if (err) { + nvgpu_err(g, "failed to prepare for poweroff, err=%d", + err); + goto finish; + } + + err = gk20a_pm_railgate(&pdev->dev); + if (err) + nvgpu_err(g, "failed to railgate, err=%d", err); + +finish: + nvgpu_info(g, "shut down complete"); +} + +#ifdef CONFIG_PM +static int gk20a_pm_runtime_resume(struct device *dev) +{ + int err = 0; + + err = gk20a_pm_unrailgate(dev); + if (err) + goto fail; + + err = gk20a_pm_finalize_poweron(dev); + if (err) + goto fail_poweron; + + return 0; + +fail_poweron: + gk20a_pm_railgate(dev); +fail: + return err; +} + +static int gk20a_pm_runtime_suspend(struct device *dev) +{ + int err = 0; + + err = gk20a_pm_prepare_poweroff(dev); + if (err) + goto fail; + + err = gk20a_pm_railgate(dev); + if (err) + goto fail_railgate; + + return 0; + +fail_railgate: + gk20a_pm_finalize_poweron(dev); +fail: + pm_runtime_mark_last_busy(dev); + return err; +} + +static int gk20a_pm_suspend(struct device *dev) +{ + struct gk20a_platform *platform = dev_get_drvdata(dev); + struct gk20a *g = get_gk20a(dev); + int ret = 0; + + if (platform->user_railgate_disabled) + gk20a_idle_nosuspend(dev); + + if (atomic_read(&dev->power.usage_count) > 1) { + ret = -EBUSY; + goto fail; + } + + if (!g->power_on) + return 0; + + ret = gk20a_pm_runtime_suspend(dev); + if (ret) + goto fail; + + if (platform->suspend) + platform->suspend(dev); + + g->suspended = true; + + return 0; + +fail: + if (platform->user_railgate_disabled) + gk20a_busy_noresume(dev); + + return ret; +} + +static int gk20a_pm_resume(struct device *dev) +{ + struct gk20a *g = get_gk20a(dev); + struct gk20a_platform *platform = dev_get_drvdata(dev); + int ret = 0; + + if (platform->user_railgate_disabled) + gk20a_busy_noresume(dev); + + if (!g->suspended) + return 0; + + ret = gk20a_pm_runtime_resume(dev); + + g->suspended = false; + + return ret; +} + +static const struct dev_pm_ops gk20a_pm_ops = { + .runtime_resume = gk20a_pm_runtime_resume, + .runtime_suspend = gk20a_pm_runtime_suspend, + .resume = gk20a_pm_resume, + .suspend = gk20a_pm_suspend, +}; +#endif + +int gk20a_pm_init(struct device *dev) +{ + struct gk20a_platform *platform = dev_get_drvdata(dev); + int err = 0; + + gk20a_dbg_fn(""); + + /* Initialise pm runtime */ + if (platform->railgate_delay) { + pm_runtime_set_autosuspend_delay(dev, + platform->railgate_delay); + pm_runtime_use_autosuspend(dev); + } + + if (platform->can_railgate) { + pm_runtime_enable(dev); + if (!pm_runtime_enabled(dev)) + gk20a_pm_unrailgate(dev); + else + gk20a_pm_railgate(dev); + } else { + __pm_runtime_disable(dev, false); + gk20a_pm_unrailgate(dev); + } + + return err; +} + +static inline void set_gk20a(struct platform_device *pdev, struct gk20a *gk20a) +{ + gk20a_get_platform(&pdev->dev)->g = gk20a; +} + +static int gk20a_probe(struct platform_device *dev) +{ + struct gk20a *gk20a; + int err; + struct gk20a_platform *platform = NULL; + + if (dev->dev.of_node) { + const struct of_device_id *match; + + match = of_match_device(tegra_gk20a_of_match, &dev->dev); + if (match) + platform = (struct gk20a_platform *)match->data; + } else + platform = (struct gk20a_platform *)dev->dev.platform_data; + + if (!platform) { + dev_err(&dev->dev, "no platform data\n"); + return -ENODATA; + } + + gk20a_dbg_fn(""); + + platform_set_drvdata(dev, platform); + + if (gk20a_gpu_is_virtual(&dev->dev)) + return vgpu_probe(dev); + + gk20a = kzalloc(sizeof(struct gk20a), GFP_KERNEL); + if (!gk20a) { + dev_err(&dev->dev, "couldn't allocate gk20a support"); + return -ENOMEM; + } + + set_gk20a(dev, gk20a); + gk20a->dev = &dev->dev; + + if (nvgpu_platform_is_simulation(gk20a)) + platform->is_fmodel = true; + + nvgpu_kmem_init(gk20a); + + gk20a->irq_stall = platform_get_irq(dev, 0); + gk20a->irq_nonstall = platform_get_irq(dev, 1); + if (gk20a->irq_stall < 0 || gk20a->irq_nonstall < 0) + return -ENXIO; + + err = devm_request_threaded_irq(&dev->dev, + gk20a->irq_stall, + gk20a_intr_isr_stall, + gk20a_intr_thread_stall, + 0, "gk20a_stall", gk20a); + if (err) { + dev_err(&dev->dev, + "failed to request stall intr irq @ %d\n", + gk20a->irq_stall); + return err; + } + err = devm_request_irq(&dev->dev, + gk20a->irq_nonstall, + gk20a_intr_isr_nonstall, + 0, "gk20a_nonstall", gk20a); + if (err) { + dev_err(&dev->dev, + "failed to request non-stall intr irq @ %d\n", + gk20a->irq_nonstall); + return err; + } + disable_irq(gk20a->irq_stall); + if (gk20a->irq_stall != gk20a->irq_nonstall) + disable_irq(gk20a->irq_nonstall); + + err = gk20a_init_support(dev); + if (err) + return err; + +#ifdef CONFIG_RESET_CONTROLLER + platform->reset_control = devm_reset_control_get(&dev->dev, NULL); + if (IS_ERR(platform->reset_control)) + platform->reset_control = NULL; +#endif + + err = nvgpu_probe(gk20a, "gpu.0", INTERFACE_NAME, &nvgpu_class); + if (err) + return err; + + err = gk20a_pm_init(&dev->dev); + if (err) { + dev_err(&dev->dev, "pm init failed"); + return err; + } + + gk20a->mm.has_physical_mode = !nvgpu_is_hypervisor_mode(gk20a); + + return 0; +} + +static int __exit gk20a_remove(struct platform_device *pdev) +{ + struct device *dev = &pdev->dev; + struct gk20a *g = get_gk20a(dev); + struct gk20a_platform *platform = gk20a_get_platform(dev); + + gk20a_dbg_fn(""); + + if (gk20a_gpu_is_virtual(dev)) + return vgpu_remove(pdev); + + if (platform->has_cde) + gk20a_cde_destroy(g); + + gk20a_ctxsw_trace_cleanup(g); + + gk20a_sched_ctrl_cleanup(g); + + if (IS_ENABLED(CONFIG_GK20A_DEVFREQ)) + gk20a_scale_exit(dev); + + if (g->remove_support) + g->remove_support(g); + + gk20a_ce_destroy(g); + +#ifdef CONFIG_ARCH_TEGRA_18x_SOC + nvgpu_clk_arb_cleanup_arbiter(g); +#endif + + gk20a_user_deinit(dev, &nvgpu_class); + + debugfs_remove_recursive(platform->debugfs); + debugfs_remove_recursive(platform->debugfs_alias); + + gk20a_remove_sysfs(dev); + + if (platform->secure_buffer.destroy) + platform->secure_buffer.destroy(dev, + &platform->secure_buffer); + + if (pm_runtime_enabled(dev)) + pm_runtime_disable(dev); + + if (platform->remove) + platform->remove(dev); + + set_gk20a(pdev, NULL); + gk20a_put(g); + + gk20a_dbg_fn("removed"); + + return 0; +} + +static struct platform_driver gk20a_driver = { + .probe = gk20a_probe, + .remove = __exit_p(gk20a_remove), + .shutdown = gk20a_pm_shutdown, + .driver = { + .owner = THIS_MODULE, + .name = "gk20a", +#if LINUX_VERSION_CODE >= KERNEL_VERSION(3,18,0) + .probe_type = PROBE_PREFER_ASYNCHRONOUS, +#endif +#ifdef CONFIG_OF + .of_match_table = tegra_gk20a_of_match, +#endif +#ifdef CONFIG_PM + .pm = &gk20a_pm_ops, +#endif + .suppress_bind_attrs = true, + } +}; + +struct class nvgpu_class = { + .owner = THIS_MODULE, + .name = CLASS_NAME, +}; + +static int __init gk20a_init(void) +{ + + int ret; + + ret = class_register(&nvgpu_class); + if (ret) + return ret; + + ret = nvgpu_pci_init(); + if (ret) + return ret; + + return platform_driver_register(&gk20a_driver); +} + +static void __exit gk20a_exit(void) +{ + nvgpu_pci_exit(); + platform_driver_unregister(&gk20a_driver); + class_unregister(&nvgpu_class); +} + +MODULE_LICENSE("GPL v2"); +module_init(gk20a_init); +module_exit(gk20a_exit); diff --git a/drivers/gpu/nvgpu/common/linux/module.h b/drivers/gpu/nvgpu/common/linux/module.h new file mode 100644 index 000000000..45fa2f5c4 --- /dev/null +++ b/drivers/gpu/nvgpu/common/linux/module.h @@ -0,0 +1,22 @@ +/* + * Copyright (c) 2011-2017, NVIDIA CORPORATION. All rights reserved. + * + * This program is free software; you can redistribute it and/or modify it + * under the terms and conditions of the GNU General Public License, + * version 2, as published by the Free Software Foundation. + * + * This program is distributed in the hope it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + */ +#ifndef __NVGPU_COMMON_LINUX_MODULE_H__ +#define __NVGPU_COMMON_LINUX_MODULE_H__ + +struct gk20a; +struct device; + +int gk20a_pm_finalize_poweron(struct device *dev); +void gk20a_remove_support(struct gk20a *g); + +#endif diff --git a/drivers/gpu/nvgpu/pci.c b/drivers/gpu/nvgpu/common/linux/pci.c similarity index 99% rename from drivers/gpu/nvgpu/pci.c rename to drivers/gpu/nvgpu/common/linux/pci.c index fb54ae18a..f90b3a6ee 100644 --- a/drivers/gpu/nvgpu/pci.c +++ b/drivers/gpu/nvgpu/common/linux/pci.c @@ -24,6 +24,7 @@ #include "gk20a/gk20a.h" #include "gk20a/platform_gk20a.h" #include "clk/clk.h" +#include "module.h" #include "pci.h" diff --git a/drivers/gpu/nvgpu/pci.h b/drivers/gpu/nvgpu/common/linux/pci.h similarity index 100% rename from drivers/gpu/nvgpu/pci.h rename to drivers/gpu/nvgpu/common/linux/pci.h diff --git a/drivers/gpu/nvgpu/gk20a/debug_gk20a.c b/drivers/gpu/nvgpu/gk20a/debug_gk20a.c index 85b24f2ea..1a9ffe774 100644 --- a/drivers/gpu/nvgpu/gk20a/debug_gk20a.c +++ b/drivers/gpu/nvgpu/gk20a/debug_gk20a.c @@ -228,6 +228,74 @@ void gk20a_init_debug_ops(struct gpu_ops *gops) gops->debug.show_dump = gk20a_debug_show_dump; } +#ifdef CONFIG_DEBUG_FS +static int railgate_residency_show(struct seq_file *s, void *data) +{ + struct device *dev = s->private; + struct gk20a_platform *platform = dev_get_drvdata(dev); + struct gk20a *g = get_gk20a(dev); + unsigned long time_since_last_state_transition_ms; + unsigned long total_rail_gate_time_ms; + unsigned long total_rail_ungate_time_ms; + + if (platform->is_railgated(dev)) { + time_since_last_state_transition_ms = + jiffies_to_msecs(jiffies - + g->pstats.last_rail_gate_complete); + total_rail_ungate_time_ms = g->pstats.total_rail_ungate_time_ms; + total_rail_gate_time_ms = + g->pstats.total_rail_gate_time_ms + + time_since_last_state_transition_ms; + } else { + time_since_last_state_transition_ms = + jiffies_to_msecs(jiffies - + g->pstats.last_rail_ungate_complete); + total_rail_gate_time_ms = g->pstats.total_rail_gate_time_ms; + total_rail_ungate_time_ms = + g->pstats.total_rail_ungate_time_ms + + time_since_last_state_transition_ms; + } + + seq_printf(s, "Time with Rails Gated: %lu ms\n" + "Time with Rails UnGated: %lu ms\n" + "Total railgating cycles: %lu\n", + total_rail_gate_time_ms, + total_rail_ungate_time_ms, + g->pstats.railgating_cycle_count - 1); + return 0; + +} + +static int railgate_residency_open(struct inode *inode, struct file *file) +{ + return single_open(file, railgate_residency_show, inode->i_private); +} + +static const struct file_operations railgate_residency_fops = { + .open = railgate_residency_open, + .read = seq_read, + .llseek = seq_lseek, + .release = single_release, +}; + +int gk20a_railgating_debugfs_init(struct device *dev) +{ + struct dentry *d; + struct gk20a_platform *platform = dev_get_drvdata(dev); + + if (!platform->can_railgate) + return 0; + + d = debugfs_create_file( + "railgate_residency", S_IRUGO|S_IWUSR, platform->debugfs, dev, + &railgate_residency_fops); + if (!d) + return -ENOMEM; + + return 0; +} +#endif + void gk20a_debug_init(struct device *dev, const char *debugfs_symlink) { struct gk20a_platform *platform = dev_get_drvdata(dev); diff --git a/drivers/gpu/nvgpu/gk20a/gk20a.c b/drivers/gpu/nvgpu/gk20a/gk20a.c index 2a80157d8..b3f4e5fe7 100644 --- a/drivers/gpu/nvgpu/gk20a/gk20a.c +++ b/drivers/gpu/nvgpu/gk20a/gk20a.c @@ -16,25 +16,7 @@ * along with this program. If not, see . */ -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include #include -#include -#include #include #include @@ -42,26 +24,22 @@ #include #include +#include + #include "gk20a.h" -#include "debug_gk20a.h" #include "channel_sync_gk20a.h" -#include "gk20a_scale.h" #include "ctxsw_trace_gk20a.h" #include "dbg_gpu_gk20a.h" #include "mc_gk20a.h" #include "hal.h" #include "vgpu/vgpu.h" -#include "pci.h" #include "bus_gk20a.h" #ifdef CONFIG_ARCH_TEGRA_18x_SOC #include "pstate/pstate.h" #endif -#define CREATE_TRACE_POINTS -#include - #ifdef CONFIG_TEGRA_19x_GPU #include "nvgpu_gpuid_t19x.h" #endif @@ -70,93 +48,6 @@ #include #include - -#ifdef CONFIG_ARM64 -#define __cpuc_flush_dcache_area __flush_dcache_area -#endif - -#define CLASS_NAME "nvidia-gpu" -/* TODO: Change to e.g. "nvidia-gpu%s" once we have symlinks in place. */ - -#define GK20A_NUM_CDEVS 7 - -#define GK20A_WAIT_FOR_IDLE_MS 2000 - -static int gk20a_pm_prepare_poweroff(struct device *dev); - -#ifdef CONFIG_DEBUG_FS -static int railgate_residency_show(struct seq_file *s, void *data) -{ - struct device *dev = s->private; - struct gk20a_platform *platform = dev_get_drvdata(dev); - struct gk20a *g = get_gk20a(dev); - unsigned long time_since_last_state_transition_ms; - unsigned long total_rail_gate_time_ms; - unsigned long total_rail_ungate_time_ms; - - if (platform->is_railgated(dev)) { - time_since_last_state_transition_ms = - jiffies_to_msecs(jiffies - - g->pstats.last_rail_gate_complete); - total_rail_ungate_time_ms = g->pstats.total_rail_ungate_time_ms; - total_rail_gate_time_ms = - g->pstats.total_rail_gate_time_ms + - time_since_last_state_transition_ms; - } else { - time_since_last_state_transition_ms = - jiffies_to_msecs(jiffies - - g->pstats.last_rail_ungate_complete); - total_rail_gate_time_ms = g->pstats.total_rail_gate_time_ms; - total_rail_ungate_time_ms = - g->pstats.total_rail_ungate_time_ms + - time_since_last_state_transition_ms; - } - - seq_printf(s, "Time with Rails Gated: %lu ms\n" - "Time with Rails UnGated: %lu ms\n" - "Total railgating cycles: %lu\n", - total_rail_gate_time_ms, - total_rail_ungate_time_ms, - g->pstats.railgating_cycle_count - 1); - return 0; - -} - -static int railgate_residency_open(struct inode *inode, struct file *file) -{ - return single_open(file, railgate_residency_show, inode->i_private); -} - -static const struct file_operations railgate_residency_fops = { - .open = railgate_residency_open, - .read = seq_read, - .llseek = seq_lseek, - .release = single_release, -}; - -int gk20a_railgating_debugfs_init(struct device *dev) -{ - struct dentry *d; - struct gk20a_platform *platform = dev_get_drvdata(dev); - - if (!platform->can_railgate) - return 0; - - d = debugfs_create_file( - "railgate_residency", S_IRUGO|S_IWUSR, platform->debugfs, dev, - &railgate_residency_fops); - if (!d) - return -ENOMEM; - - return 0; -} -#endif - -static inline void set_gk20a(struct platform_device *pdev, struct gk20a *gk20a) -{ - gk20a_get_platform(&pdev->dev)->g = gk20a; -} - void __nvgpu_check_gpu_state(struct gk20a *g) { u32 boot_0 = g->ops.mc.boot_0(g, NULL, NULL, NULL); @@ -199,181 +90,6 @@ void __gk20a_warn_on_no_regs(void) WARN_ONCE(1, "Attempted access to GPU regs after unmapping!"); } -void __iomem *gk20a_ioremap_resource(struct platform_device *dev, int i, - struct resource **out) -{ - struct resource *r = platform_get_resource(dev, IORESOURCE_MEM, i); - if (!r) - return NULL; - if (out) - *out = r; - return devm_ioremap_resource(&dev->dev, r); -} - -static irqreturn_t gk20a_intr_isr_stall(int irq, void *dev_id) -{ - struct gk20a *g = dev_id; - - return g->ops.mc.isr_stall(g); -} - -static irqreturn_t gk20a_intr_isr_nonstall(int irq, void *dev_id) -{ - struct gk20a *g = dev_id; - - return g->ops.mc.isr_nonstall(g); -} - -static irqreturn_t gk20a_intr_thread_stall(int irq, void *dev_id) -{ - struct gk20a *g = dev_id; - return g->ops.mc.isr_thread_stall(g); -} - -void gk20a_remove_support(struct gk20a *g) -{ -#ifdef CONFIG_TEGRA_COMMON - tegra_unregister_idle_unidle(); -#endif - if (g->dbg_regops_tmp_buf) - nvgpu_kfree(g, g->dbg_regops_tmp_buf); - - if (g->pmu.remove_support) - g->pmu.remove_support(&g->pmu); - - if (g->gr.remove_support) - g->gr.remove_support(&g->gr); - - if (g->mm.remove_ce_support) - g->mm.remove_ce_support(&g->mm); - - if (g->fifo.remove_support) - g->fifo.remove_support(&g->fifo); - - if (g->mm.remove_support) - g->mm.remove_support(&g->mm); - - if (g->sim.remove_support) - g->sim.remove_support(&g->sim); - - /* free mappings to registers, etc */ - - if (g->regs) { - iounmap(g->regs); - g->regs = NULL; - } - if (g->bar1) { - iounmap(g->bar1); - g->bar1 = NULL; - } -} - -static int gk20a_init_support(struct platform_device *dev) -{ - int err = 0; - struct gk20a *g = get_gk20a(&dev->dev); - -#ifdef CONFIG_TEGRA_COMMON - tegra_register_idle_unidle(gk20a_do_idle, gk20a_do_unidle); -#endif - - g->regs = gk20a_ioremap_resource(dev, GK20A_BAR0_IORESOURCE_MEM, - &g->reg_mem); - if (IS_ERR(g->regs)) { - nvgpu_err(g, "failed to remap gk20a registers\n"); - err = PTR_ERR(g->regs); - goto fail; - } - - g->bar1 = gk20a_ioremap_resource(dev, GK20A_BAR1_IORESOURCE_MEM, - &g->bar1_mem); - if (IS_ERR(g->bar1)) { - nvgpu_err(g, "failed to remap gk20a bar1\n"); - err = PTR_ERR(g->bar1); - goto fail; - } - - if (nvgpu_platform_is_simulation(g)) { - err = gk20a_init_sim_support(dev); - if (err) - goto fail; - } - - return 0; - - fail: - return err; -} - -static int gk20a_pm_prepare_poweroff(struct device *dev) -{ - struct gk20a *g = get_gk20a(dev); - int ret = 0; - struct gk20a_platform *platform = gk20a_get_platform(dev); - - gk20a_dbg_fn(""); - - nvgpu_mutex_acquire(&g->poweroff_lock); - - if (!g->power_on) - goto done; - - if (gk20a_fifo_is_engine_busy(g)) { - nvgpu_mutex_release(&g->poweroff_lock); - return -EBUSY; - } - gk20a_scale_suspend(dev); - - /* cancel any pending cde work */ - gk20a_cde_suspend(g); - - gk20a_ce_suspend(g); - - ret = gk20a_channel_suspend(g); - if (ret) - goto done; - - /* disable elpg before gr or fifo suspend */ - if (g->ops.pmu.is_pmu_supported(g)) - ret |= gk20a_pmu_destroy(g); - /* - * After this point, gk20a interrupts should not get - * serviced. - */ - disable_irq(g->irq_stall); - if (g->irq_stall != g->irq_nonstall) - disable_irq(g->irq_nonstall); - - ret |= gk20a_gr_suspend(g); - ret |= gk20a_mm_suspend(g); - ret |= gk20a_fifo_suspend(g); - - if (g->ops.pmu.mclk_deinit) - g->ops.pmu.mclk_deinit(g); - - /* Disable GPCPLL */ - if (g->ops.clk.suspend_clk_support) - ret |= g->ops.clk.suspend_clk_support(g); - -#ifdef CONFIG_ARCH_TEGRA_18x_SOC - if (g->ops.pmupstate) - gk20a_deinit_pstate_support(g); -#endif - g->power_on = false; - - /* Decrement platform power refcount */ - if (platform->idle) - platform->idle(dev); - - /* Stop CPU from accessing the GPU registers. */ - gk20a_lockout_registers(g); - -done: - nvgpu_mutex_release(&g->poweroff_lock); - - return ret; -} - static int gk20a_detect_chip(struct gk20a *g) { struct nvgpu_gpu_characteristics *gpu = &g->gpu_characteristics; @@ -392,36 +108,58 @@ static int gk20a_detect_chip(struct gk20a *g) return gpu_init_hal(g); } -int gk20a_pm_finalize_poweron(struct device *dev) +int gk20a_prepare_poweroff(struct gk20a *g) { - struct gk20a *g = get_gk20a(dev); - struct gk20a_platform *platform = gk20a_get_platform(dev); - int err, nice_value; + int ret = 0; + + gk20a_dbg_fn(""); + + if (gk20a_fifo_is_engine_busy(g)) + return -EBUSY; + + /* cancel any pending cde work */ + gk20a_cde_suspend(g); + + gk20a_ce_suspend(g); + + ret = gk20a_channel_suspend(g); + if (ret) + return ret; + + /* disable elpg before gr or fifo suspend */ + if (g->ops.pmu.is_pmu_supported(g)) + ret |= gk20a_pmu_destroy(g); + + ret |= gk20a_gr_suspend(g); + ret |= gk20a_mm_suspend(g); + ret |= gk20a_fifo_suspend(g); + + if (g->ops.pmu.mclk_deinit) + g->ops.pmu.mclk_deinit(g); + + /* Disable GPCPLL */ + if (g->ops.clk.suspend_clk_support) + ret |= g->ops.clk.suspend_clk_support(g); + +#ifdef CONFIG_ARCH_TEGRA_18x_SOC + if (g->ops.pmupstate) + gk20a_deinit_pstate_support(g); +#endif + g->power_on = false; + + return ret; +} + +int gk20a_finalize_poweron(struct gk20a *g) +{ + struct gk20a_platform *platform = gk20a_get_platform(g->dev); + int err; gk20a_dbg_fn(""); if (g->power_on) return 0; - trace_gk20a_finalize_poweron(g->name); - - /* Increment platform power refcount */ - if (platform->busy) { - err = platform->busy(dev); - if (err < 0) { - nvgpu_err(g, "%s: failed to poweron platform dependency\n", - __func__); - goto done; - } - } - - err = gk20a_restore_registers(g); - if (err) - return err; - - nice_value = task_nice(current); - set_user_nice(current, -20); - g->power_on = true; err = gk20a_detect_chip(g); @@ -586,31 +324,18 @@ int gk20a_pm_finalize_poweron(struct device *dev) g->ops.fb.set_debug_mode(g, g->mmu_debug_ctrl); gk20a_channel_resume(g); - set_user_nice(current, nice_value); - - gk20a_scale_resume(dev); - - trace_gk20a_finalize_poweron_done(g->name); - - if (platform->has_cde) - gk20a_init_cde_support(g); gk20a_init_ce_support(g); gk20a_init_mm_ce_context(g); - enable_irq(g->irq_stall); - if (g->irq_stall != g->irq_nonstall) - enable_irq(g->irq_nonstall); - g->irqs_enabled = 1; - if (g->ops.xve.available_speeds) { u32 speed; if (platform->disable_aspm && g->ops.xve.disable_aspm) g->ops.xve.disable_aspm(g); - g->ops.xve.sw_init(dev); + g->ops.xve.sw_init(g->dev); g->ops.xve.available_speeds(g, &speed); /* Set to max speed */ @@ -629,515 +354,15 @@ done: return err; } -static struct of_device_id tegra_gk20a_of_match[] = { -#ifdef CONFIG_TEGRA_GK20A - { .compatible = "nvidia,tegra124-gk20a", - .data = &gk20a_tegra_platform }, - { .compatible = "nvidia,tegra210-gm20b", - .data = &gm20b_tegra_platform }, -#ifdef CONFIG_ARCH_TEGRA_18x_SOC - { .compatible = "nvidia,tegra186-gp10b", - .data = &gp10b_tegra_platform }, -#endif -#ifdef CONFIG_TEGRA_19x_GPU - { .compatible = TEGRA_19x_GPU_COMPAT_TEGRA, - .data = &t19x_gpu_tegra_platform }, -#endif -#ifdef CONFIG_TEGRA_GR_VIRTUALIZATION - { .compatible = "nvidia,tegra124-gk20a-vgpu", - .data = &vgpu_tegra_platform }, -#endif -#else - { .compatible = "nvidia,tegra124-gk20a", - .data = &gk20a_generic_platform }, - { .compatible = "nvidia,tegra210-gm20b", - .data = &gk20a_generic_platform }, -#ifdef CONFIG_ARCH_TEGRA_18x_SOC - { .compatible = TEGRA_18x_GPU_COMPAT_TEGRA, - .data = &gk20a_generic_platform }, -#endif - -#endif - { .compatible = "nvidia,generic-gk20a", - .data = &gk20a_generic_platform }, - { .compatible = "nvidia,generic-gm20b", - .data = &gk20a_generic_platform }, -#ifdef CONFIG_ARCH_TEGRA_18x_SOC - { .compatible = "nvidia,generic-gp10b", - .data = &gk20a_generic_platform }, -#endif - { }, -}; - -static int gk20a_pm_railgate(struct device *dev) +/* + * Check if the device can go busy. Basically if the driver is currently + * in the process of dying then do not let new places make the driver busy. + */ +int gk20a_can_busy(struct gk20a *g) { - struct gk20a_platform *platform = dev_get_drvdata(dev); - int ret = 0; -#ifdef CONFIG_DEBUG_FS - struct gk20a *g = get_gk20a(dev); - - g->pstats.last_rail_gate_start = jiffies; - - if (g->pstats.railgating_cycle_count >= 1) - g->pstats.total_rail_ungate_time_ms = - g->pstats.total_rail_ungate_time_ms + - jiffies_to_msecs(g->pstats.last_rail_gate_start - - g->pstats.last_rail_ungate_complete); -#endif - - if (platform->railgate) - ret = platform->railgate(dev); - -#ifdef CONFIG_DEBUG_FS - g->pstats.last_rail_gate_complete = jiffies; -#endif - - return ret; -} - -static int gk20a_pm_unrailgate(struct device *dev) -{ - struct gk20a_platform *platform = dev_get_drvdata(dev); - int ret = 0; - struct gk20a *g = get_gk20a(dev); - -#ifdef CONFIG_DEBUG_FS - g->pstats.last_rail_ungate_start = jiffies; - if (g->pstats.railgating_cycle_count >= 1) - g->pstats.total_rail_gate_time_ms = - g->pstats.total_rail_gate_time_ms + - jiffies_to_msecs(g->pstats.last_rail_ungate_start - - g->pstats.last_rail_gate_complete); - - g->pstats.railgating_cycle_count++; -#endif - - trace_gk20a_pm_unrailgate(g->name); - - if (platform->unrailgate) { - nvgpu_mutex_acquire(&platform->railgate_lock); - ret = platform->unrailgate(dev); - nvgpu_mutex_release(&platform->railgate_lock); - } - -#ifdef CONFIG_DEBUG_FS - g->pstats.last_rail_ungate_complete = jiffies; -#endif - - return ret; -} - -static void gk20a_pm_shutdown(struct platform_device *pdev) -{ - struct gk20a_platform *platform = platform_get_drvdata(pdev); - struct gk20a *g = platform->g; - int err; - - nvgpu_info(g, "shutting down"); - - /* vgpu has nothing to clean up currently */ - if (gk20a_gpu_is_virtual(&pdev->dev)) - return; - - gk20a_driver_start_unload(g); - - /* If GPU is already railgated, - * just prevent more requests, and return */ - if (platform->is_railgated && platform->is_railgated(&pdev->dev)) { - __pm_runtime_disable(&pdev->dev, false); - nvgpu_info(g, "already railgated, shut down complete"); - return; - } - - /* Prevent more requests by disabling Runtime PM */ - __pm_runtime_disable(&pdev->dev, false); - - err = gk20a_wait_for_idle(&pdev->dev); - if (err) { - nvgpu_err(g, "failed to idle GPU, err=%d", err); - goto finish; - } - - err = gk20a_fifo_disable_all_engine_activity(g, true); - if (err) { - nvgpu_err(g, "failed to disable engine activity, err=%d", - err); - goto finish; - } - - err = gk20a_fifo_wait_engine_idle(g); - if (err) { - nvgpu_err(g, "failed to idle engines, err=%d", - err); - goto finish; - } - - if (gk20a_gpu_is_virtual(&pdev->dev)) - err = vgpu_pm_prepare_poweroff(&pdev->dev); - else - err = gk20a_pm_prepare_poweroff(&pdev->dev); - if (err) { - nvgpu_err(g, "failed to prepare for poweroff, err=%d", - err); - goto finish; - } - - err = gk20a_pm_railgate(&pdev->dev); - if (err) - nvgpu_err(g, "failed to railgate, err=%d", err); - -finish: - nvgpu_info(g, "shut down complete\n"); -} - -#ifdef CONFIG_PM -static int gk20a_pm_runtime_resume(struct device *dev) -{ - int err = 0; - - err = gk20a_pm_unrailgate(dev); - if (err) - goto fail; - - err = gk20a_pm_finalize_poweron(dev); - if (err) - goto fail_poweron; - - return 0; - -fail_poweron: - gk20a_pm_railgate(dev); -fail: - return err; -} - -static int gk20a_pm_runtime_suspend(struct device *dev) -{ - int err = 0; - - err = gk20a_pm_prepare_poweroff(dev); - if (err) - goto fail; - - err = gk20a_pm_railgate(dev); - if (err) - goto fail_railgate; - - return 0; - -fail_railgate: - gk20a_pm_finalize_poweron(dev); -fail: - pm_runtime_mark_last_busy(dev); - return err; -} - -static int gk20a_pm_suspend(struct device *dev) -{ - struct gk20a_platform *platform = dev_get_drvdata(dev); - struct gk20a *g = get_gk20a(dev); - int ret = 0; - - if (platform->user_railgate_disabled) - gk20a_idle_nosuspend(dev); - - if (atomic_read(&dev->power.usage_count) > 1) { - ret = -EBUSY; - goto fail; - } - - if (!g->power_on) + if (g->driver_is_dying) return 0; - - ret = gk20a_pm_runtime_suspend(dev); - if (ret) - goto fail; - - if (platform->suspend) - platform->suspend(dev); - - g->suspended = true; - - return 0; - -fail: - if (platform->user_railgate_disabled) - gk20a_busy_noresume(dev); - - return ret; -} - -static int gk20a_pm_resume(struct device *dev) -{ - struct gk20a *g = get_gk20a(dev); - struct gk20a_platform *platform = dev_get_drvdata(dev); - int ret = 0; - - if (platform->user_railgate_disabled) - gk20a_busy_noresume(dev); - - if (!g->suspended) - return 0; - - ret = gk20a_pm_runtime_resume(dev); - - g->suspended = false; - - return ret; -} - -static const struct dev_pm_ops gk20a_pm_ops = { - .runtime_resume = gk20a_pm_runtime_resume, - .runtime_suspend = gk20a_pm_runtime_suspend, - .resume = gk20a_pm_resume, - .suspend = gk20a_pm_suspend, -}; -#endif - -int gk20a_pm_init(struct device *dev) -{ - struct gk20a_platform *platform = dev_get_drvdata(dev); - int err = 0; - - gk20a_dbg_fn(""); - - /* Initialise pm runtime */ - if (platform->railgate_delay) { - pm_runtime_set_autosuspend_delay(dev, - platform->railgate_delay); - pm_runtime_use_autosuspend(dev); - } - - if (platform->can_railgate) { - pm_runtime_enable(dev); - if (!pm_runtime_enabled(dev)) - gk20a_pm_unrailgate(dev); - else - gk20a_pm_railgate(dev); - } else { - __pm_runtime_disable(dev, false); - gk20a_pm_unrailgate(dev); - } - - return err; -} - -int gk20a_secure_page_alloc(struct device *dev) -{ - struct gk20a_platform *platform = dev_get_drvdata(dev); - int err = 0; - - if (platform->secure_page_alloc) { - err = platform->secure_page_alloc(dev); - if (!err) - platform->secure_alloc_ready = true; - } - - return err; -} - -static int gk20a_probe(struct platform_device *dev) -{ - struct gk20a *gk20a; - int err; - struct gk20a_platform *platform = NULL; - - if (dev->dev.of_node) { - const struct of_device_id *match; - - match = of_match_device(tegra_gk20a_of_match, &dev->dev); - if (match) - platform = (struct gk20a_platform *)match->data; - } else - platform = (struct gk20a_platform *)dev->dev.platform_data; - - if (!platform) { - dev_err(&dev->dev, "no platform data\n"); - return -ENODATA; - } - - gk20a_dbg_fn(""); - - platform_set_drvdata(dev, platform); - - if (gk20a_gpu_is_virtual(&dev->dev)) - return vgpu_probe(dev); - - gk20a = kzalloc(sizeof(struct gk20a), GFP_KERNEL); - if (!gk20a) { - dev_err(&dev->dev, "couldn't allocate gk20a support"); - return -ENOMEM; - } - - set_gk20a(dev, gk20a); - gk20a->dev = &dev->dev; - - if (nvgpu_platform_is_simulation(gk20a)) - platform->is_fmodel = true; - - nvgpu_kmem_init(gk20a); - - gk20a->irq_stall = platform_get_irq(dev, 0); - gk20a->irq_nonstall = platform_get_irq(dev, 1); - if (gk20a->irq_stall < 0 || gk20a->irq_nonstall < 0) - return -ENXIO; - - err = devm_request_threaded_irq(&dev->dev, - gk20a->irq_stall, - gk20a_intr_isr_stall, - gk20a_intr_thread_stall, - 0, "gk20a_stall", gk20a); - if (err) { - dev_err(&dev->dev, - "failed to request stall intr irq @ %d\n", - gk20a->irq_stall); - return err; - } - err = devm_request_irq(&dev->dev, - gk20a->irq_nonstall, - gk20a_intr_isr_nonstall, - 0, "gk20a_nonstall", gk20a); - if (err) { - dev_err(&dev->dev, - "failed to request non-stall intr irq @ %d\n", - gk20a->irq_nonstall); - return err; - } - disable_irq(gk20a->irq_stall); - if (gk20a->irq_stall != gk20a->irq_nonstall) - disable_irq(gk20a->irq_nonstall); - - /* - * is_fmodel needs to be in gk20a struct for deferred teardown - */ - gk20a->is_fmodel = platform->is_fmodel; - - err = gk20a_init_support(dev); - if (err) - return err; - -#ifdef CONFIG_RESET_CONTROLLER - platform->reset_control = devm_reset_control_get(&dev->dev, NULL); - if (IS_ERR(platform->reset_control)) - platform->reset_control = NULL; -#endif - - err = nvgpu_probe(gk20a, "gpu.0", INTERFACE_NAME, &nvgpu_class); - if (err) - return err; - - err = gk20a_pm_init(&dev->dev); - if (err) { - dev_err(&dev->dev, "pm init failed"); - return err; - } - - gk20a->mm.has_physical_mode = !nvgpu_is_hypervisor_mode(gk20a); - - return 0; -} - -static int __exit gk20a_remove(struct platform_device *pdev) -{ - struct device *dev = &pdev->dev; - struct gk20a *g = get_gk20a(dev); - struct gk20a_platform *platform = gk20a_get_platform(dev); - - gk20a_dbg_fn(""); - - if (gk20a_gpu_is_virtual(dev)) - return vgpu_remove(pdev); - - if (platform->has_cde) - gk20a_cde_destroy(g); - - gk20a_ctxsw_trace_cleanup(g); - - gk20a_sched_ctrl_cleanup(g); - - if (IS_ENABLED(CONFIG_GK20A_DEVFREQ)) - gk20a_scale_exit(dev); - -#ifdef CONFIG_ARCH_TEGRA_18x_SOC - nvgpu_clk_arb_cleanup_arbiter(g); -#endif - - gk20a_user_deinit(dev, &nvgpu_class); - - debugfs_remove_recursive(platform->debugfs); - debugfs_remove_recursive(platform->debugfs_alias); - - gk20a_remove_sysfs(dev); - - if (platform->secure_buffer.destroy) - platform->secure_buffer.destroy(dev, - &platform->secure_buffer); - - if (pm_runtime_enabled(dev)) - pm_runtime_disable(dev); - - if (platform->remove) - platform->remove(dev); - - set_gk20a(pdev, NULL); - gk20a_put(g); - - gk20a_dbg_fn("removed"); - - return 0; -} - -static struct platform_driver gk20a_driver = { - .probe = gk20a_probe, - .remove = __exit_p(gk20a_remove), - .shutdown = gk20a_pm_shutdown, - .driver = { - .owner = THIS_MODULE, - .name = "gk20a", -#if LINUX_VERSION_CODE >= KERNEL_VERSION(3,18,0) - .probe_type = PROBE_PREFER_ASYNCHRONOUS, -#endif -#ifdef CONFIG_OF - .of_match_table = tegra_gk20a_of_match, -#endif -#ifdef CONFIG_PM - .pm = &gk20a_pm_ops, -#endif - .suppress_bind_attrs = true, - } -}; - -struct class nvgpu_class = { - .owner = THIS_MODULE, - .name = CLASS_NAME, -}; - -static int __init gk20a_init(void) -{ - - int ret; - - ret = class_register(&nvgpu_class); - if (ret) - return ret; - - ret = nvgpu_pci_init(); - if (ret) - return ret; - - return platform_driver_register(&gk20a_driver); -} - -static void __exit gk20a_exit(void) -{ - nvgpu_pci_exit(); - platform_driver_unregister(&gk20a_driver); - class_unregister(&nvgpu_class); -} - -void gk20a_busy_noresume(struct device *dev) -{ - pm_runtime_get_noresume(dev); + return 1; } /* @@ -1193,292 +418,6 @@ int gk20a_wait_for_idle(struct device *dev) return 0; } -/* - * Check if the device can go busy. Basically if the driver is currently - * in the process of dying then do not let new places make the driver busy. - */ -static int gk20a_can_busy(struct gk20a *g) -{ - if (g->driver_is_dying) - return 0; - return 1; -} - -int gk20a_busy(struct gk20a *g) -{ - int ret = 0; - struct device *dev; - - if (!g) - return -ENODEV; - - atomic_inc(&g->usage_count); - - down_read(&g->busy_lock); - - if (!gk20a_can_busy(g)) { - ret = -ENODEV; - atomic_dec(&g->usage_count); - goto fail; - } - - dev = g->dev; - - if (pm_runtime_enabled(dev)) { - ret = pm_runtime_get_sync(dev); - if (ret < 0) { - pm_runtime_put_noidle(dev); - atomic_dec(&g->usage_count); - goto fail; - } - } else { - if (!g->power_on) { - ret = gk20a_gpu_is_virtual(dev) ? - vgpu_pm_finalize_poweron(dev) - : gk20a_pm_finalize_poweron(dev); - if (ret) { - atomic_dec(&g->usage_count); - goto fail; - } - } - } - - gk20a_scale_notify_busy(dev); - -fail: - up_read(&g->busy_lock); - - return ret < 0 ? ret : 0; -} - -void gk20a_idle_nosuspend(struct device *dev) -{ - pm_runtime_put_noidle(dev); -} - -void gk20a_idle(struct gk20a *g) -{ - struct device *dev; - - atomic_dec(&g->usage_count); - down_read(&g->busy_lock); - - dev = g->dev; - - if (!(dev && gk20a_can_busy(g))) - goto fail; - - if (pm_runtime_enabled(dev)) { -#ifdef CONFIG_PM - if (atomic_read(&g->dev->power.usage_count) == 1) - gk20a_scale_notify_idle(dev); -#endif - - pm_runtime_mark_last_busy(dev); - pm_runtime_put_sync_autosuspend(dev); - - } else { - gk20a_scale_notify_idle(dev); - } -fail: - up_read(&g->busy_lock); -} - -#ifdef CONFIG_PM -/** - * __gk20a_do_idle() - force the GPU to idle and railgate - * - * In success, this call MUST be balanced by caller with __gk20a_do_unidle() - * - * Acquires two locks : &g->busy_lock and &platform->railgate_lock - * In success, we hold these locks and return - * In failure, we release these locks and return - */ -int __gk20a_do_idle(struct device *dev, bool force_reset) -{ - struct gk20a *g = get_gk20a(dev); - struct gk20a_platform *platform = dev_get_drvdata(dev); - struct nvgpu_timeout timeout; - int ref_cnt; - int target_ref_cnt = 0; - bool is_railgated; - int err = 0; - - /* acquire busy lock to block other busy() calls */ - down_write(&g->busy_lock); - - /* acquire railgate lock to prevent unrailgate in midst of do_idle() */ - nvgpu_mutex_acquire(&platform->railgate_lock); - - /* check if it is already railgated ? */ - if (platform->is_railgated(dev)) - return 0; - - /* - * release railgate_lock, prevent suspend by incrementing usage counter, - * re-acquire railgate_lock - */ - nvgpu_mutex_release(&platform->railgate_lock); - pm_runtime_get_sync(dev); - - /* - * One refcount taken in this API - * If User disables rail gating, we take one more - * extra refcount - */ - if (platform->user_railgate_disabled) - target_ref_cnt = 2; - else - target_ref_cnt = 1; - nvgpu_mutex_acquire(&platform->railgate_lock); - - nvgpu_timeout_init(g, &timeout, GK20A_WAIT_FOR_IDLE_MS, - NVGPU_TIMER_CPU_TIMER); - - /* check and wait until GPU is idle (with a timeout) */ - do { - nvgpu_msleep(1); - ref_cnt = atomic_read(&dev->power.usage_count); - } while (ref_cnt != target_ref_cnt && !nvgpu_timeout_expired(&timeout)); - - if (ref_cnt != target_ref_cnt) { - nvgpu_err(g, "failed to idle - refcount %d != 1\n", - ref_cnt); - goto fail_drop_usage_count; - } - - /* check if global force_reset flag is set */ - force_reset |= platform->force_reset_in_do_idle; - - nvgpu_timeout_init(g, &timeout, GK20A_WAIT_FOR_IDLE_MS, - NVGPU_TIMER_CPU_TIMER); - - if (platform->can_railgate && !force_reset) { - /* - * Case 1 : GPU railgate is supported - * - * if GPU is now idle, we will have only one ref count, - * drop this ref which will rail gate the GPU - */ - pm_runtime_put_sync(dev); - - /* add sufficient delay to allow GPU to rail gate */ - nvgpu_msleep(platform->railgate_delay); - - /* check in loop if GPU is railgated or not */ - do { - nvgpu_msleep(1); - is_railgated = platform->is_railgated(dev); - } while (!is_railgated && !nvgpu_timeout_expired(&timeout)); - - if (is_railgated) { - return 0; - } else { - nvgpu_err(g, "failed to idle in timeout\n"); - goto fail_timeout; - } - } else { - /* - * Case 2 : GPU railgate is not supported or we explicitly - * do not want to depend on runtime PM - * - * if GPU is now idle, call prepare_poweroff() to save the - * state and then do explicit railgate - * - * __gk20a_do_unidle() needs to unrailgate, call - * finalize_poweron(), and then call pm_runtime_put_sync() - * to balance the GPU usage counter - */ - - /* Save the GPU state */ - err = gk20a_pm_prepare_poweroff(dev); - if (err) - goto fail_drop_usage_count; - - /* railgate GPU */ - platform->railgate(dev); - - nvgpu_udelay(10); - - g->forced_reset = true; - return 0; - } - -fail_drop_usage_count: - pm_runtime_put_noidle(dev); -fail_timeout: - nvgpu_mutex_release(&platform->railgate_lock); - up_write(&g->busy_lock); - return -EBUSY; -} - -/** - * gk20a_do_idle() - wrap up for __gk20a_do_idle() to be called - * from outside of GPU driver - * - * In success, this call MUST be balanced by caller with gk20a_do_unidle() - */ -int gk20a_do_idle(void) -{ - struct device_node *node = - of_find_matching_node(NULL, tegra_gk20a_of_match); - struct platform_device *pdev = of_find_device_by_node(node); - - int ret = __gk20a_do_idle(&pdev->dev, true); - - of_node_put(node); - - return ret; -} - -/** - * __gk20a_do_unidle() - unblock all the tasks blocked by __gk20a_do_idle() - */ -int __gk20a_do_unidle(struct device *dev) -{ - struct gk20a *g = get_gk20a(dev); - struct gk20a_platform *platform = dev_get_drvdata(dev); - - if (g->forced_reset) { - /* - * If we did a forced-reset/railgate - * then unrailgate the GPU here first - */ - platform->unrailgate(dev); - - /* restore the GPU state */ - gk20a_pm_finalize_poweron(dev); - - /* balance GPU usage counter */ - pm_runtime_put_sync(dev); - - g->forced_reset = false; - } - - /* release the lock and open up all other busy() calls */ - nvgpu_mutex_release(&platform->railgate_lock); - up_write(&g->busy_lock); - - return 0; -} - -/** - * gk20a_do_unidle() - wrap up for __gk20a_do_unidle() - */ -int gk20a_do_unidle(void) -{ - struct device_node *node = - of_find_matching_node(NULL, tegra_gk20a_of_match); - struct platform_device *pdev = of_find_device_by_node(node); - - int ret = __gk20a_do_unidle(&pdev->dev); - - of_node_put(node); - - return ret; -} -#endif - int gk20a_init_gpu_characteristics(struct gk20a *g) { struct nvgpu_gpu_characteristics *gpu = &g->gpu_characteristics; @@ -1651,7 +590,3 @@ void gk20a_put(struct gk20a *g) kref_put(&g->refcount, gk20a_free_cb); } - -MODULE_LICENSE("GPL v2"); -module_init(gk20a_init); -module_exit(gk20a_exit); diff --git a/drivers/gpu/nvgpu/gk20a/gk20a.h b/drivers/gpu/nvgpu/gk20a/gk20a.h index 60d04b644..9860910cf 100644 --- a/drivers/gpu/nvgpu/gk20a/gk20a.h +++ b/drivers/gpu/nvgpu/gk20a/gk20a.h @@ -1389,6 +1389,7 @@ int gk20a_do_unidle(void); int __gk20a_do_idle(struct device *dev, bool force_reset); int __gk20a_do_unidle(struct device *dev); +int gk20a_can_busy(struct gk20a *g); void gk20a_driver_start_unload(struct gk20a *g); int gk20a_wait_for_idle(struct device *dev); @@ -1431,9 +1432,8 @@ extern struct class nvgpu_class; #define INTERFACE_NAME "nvhost%s-gpu" -int gk20a_pm_init(struct device *dev); -int gk20a_pm_finalize_poweron(struct device *dev); -void gk20a_remove_support(struct gk20a *g); +int gk20a_prepare_poweroff(struct gk20a *g); +int gk20a_finalize_poweron(struct gk20a *g); static inline struct tsg_gk20a *tsg_gk20a_from_ch(struct channel_gk20a *ch) { @@ -1476,8 +1476,4 @@ void gk20a_put(struct gk20a *g); int gk20a_railgating_debugfs_init(struct device *dev); #endif -int gk20a_secure_page_alloc(struct device *dev); -void __iomem *gk20a_ioremap_resource(struct platform_device *dev, int i, - struct resource **out); - #endif /* GK20A_H */ diff --git a/drivers/gpu/nvgpu/gk20a/sim_gk20a.c b/drivers/gpu/nvgpu/gk20a/sim_gk20a.c index 8951d5a47..5c11c1181 100644 --- a/drivers/gpu/nvgpu/gk20a/sim_gk20a.c +++ b/drivers/gpu/nvgpu/gk20a/sim_gk20a.c @@ -103,15 +103,6 @@ int gk20a_init_sim_support(struct platform_device *pdev) struct gk20a *g = get_gk20a(dev); u64 phys; - g->sim.g = g; - g->sim.regs = gk20a_ioremap_resource(pdev, GK20A_SIM_IORESOURCE_MEM, - &g->sim.reg_mem); - if (IS_ERR(g->sim.regs)) { - nvgpu_err(g, "failed to remap gk20a sim regs\n"); - err = PTR_ERR(g->sim.regs); - goto fail; - } - /* allocate sim event/msg buffers */ err = alloc_and_kmap_iopage(g, &g->sim.send_bfr.kvaddr, &g->sim.send_bfr.phys,