diff options
Diffstat (limited to 'include/os/linux')
-rw-r--r-- | include/os/linux/debug.c | 4 | ||||
-rw-r--r-- | include/os/linux/driver_common.c | 69 | ||||
-rw-r--r-- | include/os/linux/ecc_linux.h | 49 | ||||
-rw-r--r-- | include/os/linux/ioctl_clk_arb.c | 17 | ||||
-rw-r--r-- | include/os/linux/ioctl_ctrl.c | 122 | ||||
-rw-r--r-- | include/os/linux/ioctl_tsg.c | 26 | ||||
-rw-r--r-- | include/os/linux/module.c | 22 | ||||
-rw-r--r-- | include/os/linux/os_linux.h | 7 | ||||
-rw-r--r-- | include/os/linux/platform_gv11b_tegra.c | 4 | ||||
-rw-r--r-- | include/os/linux/scale.c | 29 | ||||
-rw-r--r-- | include/os/linux/sdl.c | 341 | ||||
-rw-r--r-- | include/os/linux/vm.c | 8 |
12 files changed, 601 insertions, 97 deletions
diff --git a/include/os/linux/debug.c b/include/os/linux/debug.c index 5f0703c..b8c4596 100644 --- a/include/os/linux/debug.c +++ b/include/os/linux/debug.c | |||
@@ -1,5 +1,5 @@ | |||
1 | /* | 1 | /* |
2 | * Copyright (C) 2017-2018 NVIDIA Corporation. All rights reserved. | 2 | * Copyright (C) 2017-2021 NVIDIA Corporation. All rights reserved. |
3 | * | 3 | * |
4 | * This software is licensed under the terms of the GNU General Public | 4 | * This software is licensed under the terms of the GNU General Public |
5 | * License version 2, as published by the Free Software Foundation, and | 5 | * License version 2, as published by the Free Software Foundation, and |
@@ -224,7 +224,7 @@ static int railgate_residency_show(struct seq_file *s, void *data) | |||
224 | unsigned long total_rail_gate_time_ms; | 224 | unsigned long total_rail_gate_time_ms; |
225 | unsigned long total_rail_ungate_time_ms; | 225 | unsigned long total_rail_ungate_time_ms; |
226 | 226 | ||
227 | if (platform->is_railgated(dev_from_gk20a(g))) { | 227 | if (platform && platform->is_railgated && platform->is_railgated(dev_from_gk20a(g))) { |
228 | time_since_last_state_transition_ms = | 228 | time_since_last_state_transition_ms = |
229 | jiffies_to_msecs(jiffies - | 229 | jiffies_to_msecs(jiffies - |
230 | g->pstats.last_rail_gate_complete); | 230 | g->pstats.last_rail_gate_complete); |
diff --git a/include/os/linux/driver_common.c b/include/os/linux/driver_common.c index c76dabe..8f5872d 100644 --- a/include/os/linux/driver_common.c +++ b/include/os/linux/driver_common.c | |||
@@ -1,5 +1,5 @@ | |||
1 | /* | 1 | /* |
2 | * Copyright (c) 2016-2018, NVIDIA CORPORATION. All rights reserved. | 2 | * Copyright (c) 2016-2022, NVIDIA CORPORATION. All rights reserved. |
3 | * | 3 | * |
4 | * This program is free software; you can redistribute it and/or modify it | 4 | * This program is free software; you can redistribute it and/or modify it |
5 | * under the terms and conditions of the GNU General Public License, | 5 | * under the terms and conditions of the GNU General Public License, |
@@ -18,6 +18,7 @@ | |||
18 | #include <linux/dma-mapping.h> | 18 | #include <linux/dma-mapping.h> |
19 | #include <linux/mm.h> | 19 | #include <linux/mm.h> |
20 | #include <linux/slab.h> | 20 | #include <linux/slab.h> |
21 | #include <linux/of_platform.h> | ||
21 | #include <uapi/linux/nvgpu.h> | 22 | #include <uapi/linux/nvgpu.h> |
22 | 23 | ||
23 | #include <nvgpu/defaults.h> | 24 | #include <nvgpu/defaults.h> |
@@ -241,6 +242,8 @@ int nvgpu_probe(struct gk20a *g, | |||
241 | struct device *dev = dev_from_gk20a(g); | 242 | struct device *dev = dev_from_gk20a(g); |
242 | struct gk20a_platform *platform = dev_get_drvdata(dev); | 243 | struct gk20a_platform *platform = dev_get_drvdata(dev); |
243 | int err = 0; | 244 | int err = 0; |
245 | struct device_node *np = dev->of_node; | ||
246 | bool disable_l3_alloc = false; | ||
244 | 247 | ||
245 | nvgpu_init_vars(g); | 248 | nvgpu_init_vars(g); |
246 | nvgpu_init_gr_vars(g); | 249 | nvgpu_init_gr_vars(g); |
@@ -265,6 +268,12 @@ int nvgpu_probe(struct gk20a *g, | |||
265 | return err; | 268 | return err; |
266 | } | 269 | } |
267 | 270 | ||
271 | disable_l3_alloc = of_property_read_bool(np, "disable_l3_alloc"); | ||
272 | if (disable_l3_alloc) { | ||
273 | nvgpu_log_info(g, "L3 alloc is disabled\n"); | ||
274 | __nvgpu_set_enabled(g, NVGPU_DISABLE_L3_SUPPORT, true); | ||
275 | } | ||
276 | |||
268 | nvgpu_init_mm_vars(g); | 277 | nvgpu_init_mm_vars(g); |
269 | 278 | ||
270 | /* platform probe can defer do user init only if probe succeeds */ | 279 | /* platform probe can defer do user init only if probe succeeds */ |
@@ -312,30 +321,70 @@ static int cyclic_delta(int a, int b) | |||
312 | } | 321 | } |
313 | 322 | ||
314 | /** | 323 | /** |
315 | * nvgpu_wait_for_deferred_interrupts - Wait for interrupts to complete | 324 | * nvgpu_wait_for_stall_interrupts - Wait for the stalling interrupts to |
325 | * complete. | ||
316 | * | 326 | * |
317 | * @g - The GPU to wait on. | 327 | * @g - The GPU to wait on. |
328 | * @timeout - maximum time period to wait for. | ||
318 | * | 329 | * |
319 | * Waits until all interrupt handlers that have been scheduled to run have | 330 | * Waits until all stalling interrupt handlers that have been scheduled to run |
320 | * completed. | 331 | * have completed. |
321 | */ | 332 | */ |
322 | void nvgpu_wait_for_deferred_interrupts(struct gk20a *g) | 333 | int nvgpu_wait_for_stall_interrupts(struct gk20a *g, u32 timeout) |
323 | { | 334 | { |
324 | struct nvgpu_os_linux *l = nvgpu_os_linux_from_gk20a(g); | 335 | struct nvgpu_os_linux *l = nvgpu_os_linux_from_gk20a(g); |
325 | int stall_irq_threshold = atomic_read(&l->hw_irq_stall_count); | 336 | int stall_irq_threshold = atomic_read(&l->hw_irq_stall_count); |
326 | int nonstall_irq_threshold = atomic_read(&l->hw_irq_nonstall_count); | ||
327 | 337 | ||
328 | /* wait until all stalling irqs are handled */ | 338 | /* wait until all stalling irqs are handled */ |
329 | NVGPU_COND_WAIT(&l->sw_irq_stall_last_handled_wq, | 339 | return NVGPU_COND_WAIT(&l->sw_irq_stall_last_handled_wq, |
330 | cyclic_delta(stall_irq_threshold, | 340 | cyclic_delta(stall_irq_threshold, |
331 | atomic_read(&l->sw_irq_stall_last_handled)) | 341 | atomic_read(&l->sw_irq_stall_last_handled)) |
332 | <= 0, 0); | 342 | <= 0, timeout); |
343 | } | ||
344 | |||
345 | /** | ||
346 | * nvgpu_wait_for_nonstall_interrupts - Wait for the nonstalling interrupts to | ||
347 | * complete. | ||
348 | * | ||
349 | * @g - The GPU to wait on. | ||
350 | * @timeout - maximum time period to wait for. | ||
351 | * | ||
352 | * Waits until all non-stalling interrupt handlers that have been scheduled to | ||
353 | * run have completed. | ||
354 | */ | ||
355 | int nvgpu_wait_for_nonstall_interrupts(struct gk20a *g, u32 timeout) | ||
356 | { | ||
357 | struct nvgpu_os_linux *l = nvgpu_os_linux_from_gk20a(g); | ||
358 | int nonstall_irq_threshold = atomic_read(&l->hw_irq_nonstall_count); | ||
333 | 359 | ||
334 | /* wait until all non-stalling irqs are handled */ | 360 | /* wait until all non-stalling irqs are handled */ |
335 | NVGPU_COND_WAIT(&l->sw_irq_nonstall_last_handled_wq, | 361 | return NVGPU_COND_WAIT(&l->sw_irq_nonstall_last_handled_wq, |
336 | cyclic_delta(nonstall_irq_threshold, | 362 | cyclic_delta(nonstall_irq_threshold, |
337 | atomic_read(&l->sw_irq_nonstall_last_handled)) | 363 | atomic_read(&l->sw_irq_nonstall_last_handled)) |
338 | <= 0, 0); | 364 | <= 0, timeout); |
365 | } | ||
366 | |||
367 | /** | ||
368 | * nvgpu_wait_for_deferred_interrupts - Wait for interrupts to complete | ||
369 | * | ||
370 | * @g - The GPU to wait on. | ||
371 | * | ||
372 | * Waits until all interrupt handlers that have been scheduled to run have | ||
373 | * completed. | ||
374 | */ | ||
375 | void nvgpu_wait_for_deferred_interrupts(struct gk20a *g) | ||
376 | { | ||
377 | int ret; | ||
378 | |||
379 | ret = nvgpu_wait_for_stall_interrupts(g, 0U); | ||
380 | if (ret != 0) { | ||
381 | nvgpu_err(g, "wait for stall interrupts failed %d", ret); | ||
382 | } | ||
383 | |||
384 | ret = nvgpu_wait_for_nonstall_interrupts(g, 0U); | ||
385 | if (ret != 0) { | ||
386 | nvgpu_err(g, "wait for nonstall interrupts failed %d", ret); | ||
387 | } | ||
339 | } | 388 | } |
340 | 389 | ||
341 | static void nvgpu_free_gk20a(struct gk20a *g) | 390 | static void nvgpu_free_gk20a(struct gk20a *g) |
diff --git a/include/os/linux/ecc_linux.h b/include/os/linux/ecc_linux.h new file mode 100644 index 0000000..7e0f650 --- /dev/null +++ b/include/os/linux/ecc_linux.h | |||
@@ -0,0 +1,49 @@ | |||
1 | /* | ||
2 | * | ||
3 | * Copyright (c) 2021, NVIDIA CORPORATION. All rights reserved. | ||
4 | * | ||
5 | * Permission is hereby granted, free of charge, to any person obtaining a | ||
6 | * copy of this software and associated documentation files (the "Software"), | ||
7 | * to deal in the Software without restriction, including without limitation | ||
8 | * the rights to use, copy, modify, merge, publish, distribute, sublicense, | ||
9 | * and/or sell copies of the Software, and to permit persons to whom the | ||
10 | * Software is furnished to do so, subject to the following conditions: | ||
11 | * | ||
12 | * The above copyright notice and this permission notice shall be included in | ||
13 | * all copies or substantial portions of the Software. | ||
14 | * | ||
15 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | ||
16 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | ||
17 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL | ||
18 | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | ||
19 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING | ||
20 | * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER | ||
21 | * DEALINGS IN THE SOFTWARE. | ||
22 | */ | ||
23 | |||
24 | #ifndef NVGPU_OS_ECC_LINUX_H | ||
25 | #define NVGPU_OS_ECC_LINUX_H | ||
26 | |||
27 | #ifdef CONFIG_NVGPU_SUPPORT_LINUX_ECC_ERROR_REPORTING | ||
28 | |||
29 | #include <linux/tegra_l1ss_kernel_interface.h> | ||
30 | #include <linux/tegra_l1ss_ioctl.h> | ||
31 | #include <linux/tegra_nv_guard_service_id.h> | ||
32 | #include <linux/tegra_nv_guard_group_id.h> | ||
33 | |||
34 | #include <nvgpu/nvgpu_err.h> | ||
35 | |||
36 | struct nvgpu_ecc_reporting_linux { | ||
37 | struct nvgpu_ecc_reporting common; | ||
38 | client_param_t priv; | ||
39 | }; | ||
40 | |||
41 | static inline struct nvgpu_ecc_reporting_linux *get_ecc_reporting_linux( | ||
42 | struct nvgpu_ecc_reporting *ecc_report) | ||
43 | { | ||
44 | return container_of(ecc_report, struct nvgpu_ecc_reporting_linux, common); | ||
45 | } | ||
46 | |||
47 | #endif /* CONFIG_NVGPU_SUPPORT_LINUX_ECC_ERROR_REPORTING */ | ||
48 | |||
49 | #endif \ No newline at end of file | ||
diff --git a/include/os/linux/ioctl_clk_arb.c b/include/os/linux/ioctl_clk_arb.c index 477222d..9f32102 100644 --- a/include/os/linux/ioctl_clk_arb.c +++ b/include/os/linux/ioctl_clk_arb.c | |||
@@ -1,5 +1,5 @@ | |||
1 | /* | 1 | /* |
2 | * Copyright (c) 2016-2018, NVIDIA CORPORATION. All rights reserved. | 2 | * Copyright (c) 2016-2021, NVIDIA CORPORATION. All rights reserved. |
3 | * | 3 | * |
4 | * This software is licensed under the terms of the GNU General Public | 4 | * This software is licensed under the terms of the GNU General Public |
5 | * License version 2, as published by the Free Software Foundation, and | 5 | * License version 2, as published by the Free Software Foundation, and |
@@ -51,19 +51,28 @@ static int nvgpu_clk_arb_release_completion_dev(struct inode *inode, | |||
51 | { | 51 | { |
52 | struct nvgpu_clk_dev *dev = filp->private_data; | 52 | struct nvgpu_clk_dev *dev = filp->private_data; |
53 | struct nvgpu_clk_session *session = dev->session; | 53 | struct nvgpu_clk_session *session = dev->session; |
54 | struct gk20a *g = session->g; | ||
55 | struct nvgpu_clk_arb *arb = g->clk_arb; | ||
54 | 56 | ||
57 | clk_arb_dbg(g, " "); | ||
55 | 58 | ||
56 | clk_arb_dbg(session->g, " "); | 59 | nvgpu_spinlock_acquire(&session->session_lock); |
60 | nvgpu_spinlock_acquire(&arb->requests_lock); | ||
61 | |||
62 | nvgpu_list_del(&dev->node); | ||
63 | |||
64 | nvgpu_spinlock_release(&arb->requests_lock); | ||
65 | nvgpu_spinlock_release(&session->session_lock); | ||
57 | 66 | ||
58 | /* This is done to account for the extra refcount taken in | 67 | /* This is done to account for the extra refcount taken in |
59 | * nvgpu_clk_arb_commit_request_fd without events support in iGPU | 68 | * nvgpu_clk_arb_commit_request_fd without events support in iGPU |
60 | */ | 69 | */ |
61 | if (!session->g->clk_arb->clk_arb_events_supported) { | 70 | if (!arb->clk_arb_events_supported) { |
62 | nvgpu_ref_put(&dev->refcount, nvgpu_clk_arb_free_fd); | 71 | nvgpu_ref_put(&dev->refcount, nvgpu_clk_arb_free_fd); |
63 | } | 72 | } |
64 | 73 | ||
65 | nvgpu_ref_put(&session->refcount, nvgpu_clk_arb_free_session); | ||
66 | nvgpu_ref_put(&dev->refcount, nvgpu_clk_arb_free_fd); | 74 | nvgpu_ref_put(&dev->refcount, nvgpu_clk_arb_free_fd); |
75 | nvgpu_ref_put(&session->refcount, nvgpu_clk_arb_free_session); | ||
67 | return 0; | 76 | return 0; |
68 | } | 77 | } |
69 | 78 | ||
diff --git a/include/os/linux/ioctl_ctrl.c b/include/os/linux/ioctl_ctrl.c index ee141ff..841d345 100644 --- a/include/os/linux/ioctl_ctrl.c +++ b/include/os/linux/ioctl_ctrl.c | |||
@@ -1,5 +1,5 @@ | |||
1 | /* | 1 | /* |
2 | * Copyright (c) 2011-2020, NVIDIA Corporation. All rights reserved. | 2 | * Copyright (c) 2011-2021, NVIDIA Corporation. All rights reserved. |
3 | * | 3 | * |
4 | * This program is free software; you can redistribute it and/or modify it | 4 | * This program is free software; you can redistribute it and/or modify it |
5 | * under the terms and conditions of the GNU General Public License, | 5 | * under the terms and conditions of the GNU General Public License, |
@@ -60,7 +60,6 @@ struct gk20a_ctrl_priv { | |||
60 | struct nvgpu_list_node list; | 60 | struct nvgpu_list_node list; |
61 | struct { | 61 | struct { |
62 | struct vm_area_struct *vma; | 62 | struct vm_area_struct *vma; |
63 | unsigned long flags; | ||
64 | bool vma_mapped; | 63 | bool vma_mapped; |
65 | } usermode_vma; | 64 | } usermode_vma; |
66 | }; | 65 | }; |
@@ -488,27 +487,26 @@ static int gk20a_ctrl_alloc_as( | |||
488 | 487 | ||
489 | snprintf(name, sizeof(name), "nvhost-%s-fd%d", g->name, fd); | 488 | snprintf(name, sizeof(name), "nvhost-%s-fd%d", g->name, fd); |
490 | 489 | ||
491 | file = anon_inode_getfile(name, l->as_dev.cdev.ops, NULL, O_RDWR); | ||
492 | if (IS_ERR(file)) { | ||
493 | err = PTR_ERR(file); | ||
494 | goto clean_up; | ||
495 | } | ||
496 | |||
497 | err = gk20a_as_alloc_share(g, args->big_page_size, | 490 | err = gk20a_as_alloc_share(g, args->big_page_size, |
498 | gk20a_as_translate_as_alloc_flags(g, | 491 | gk20a_as_translate_as_alloc_flags(g, |
499 | args->flags), | 492 | args->flags), |
500 | &as_share); | 493 | &as_share); |
501 | if (err) | 494 | if (err) |
502 | goto clean_up_file; | 495 | goto clean_up; |
496 | |||
497 | file = anon_inode_getfile(name, l->as_dev.cdev.ops, as_share, O_RDWR); | ||
498 | if (IS_ERR(file)) { | ||
499 | err = PTR_ERR(file); | ||
500 | goto clean_up_as; | ||
501 | } | ||
503 | 502 | ||
504 | fd_install(fd, file); | 503 | fd_install(fd, file); |
505 | file->private_data = as_share; | ||
506 | 504 | ||
507 | args->as_fd = fd; | 505 | args->as_fd = fd; |
508 | return 0; | 506 | return 0; |
509 | 507 | ||
510 | clean_up_file: | 508 | clean_up_as: |
511 | fput(file); | 509 | gk20a_as_release_share(as_share); |
512 | clean_up: | 510 | clean_up: |
513 | put_unused_fd(fd); | 511 | put_unused_fd(fd); |
514 | return err; | 512 | return err; |
@@ -692,12 +690,15 @@ static int nvgpu_gpu_ioctl_trigger_suspend(struct gk20a *g) | |||
692 | 690 | ||
693 | err = gk20a_busy(g); | 691 | err = gk20a_busy(g); |
694 | if (err) | 692 | if (err) |
695 | return err; | 693 | return err; |
696 | 694 | ||
697 | nvgpu_mutex_acquire(&g->dbg_sessions_lock); | 695 | if (g->ops.gr.trigger_suspend) { |
698 | err = gr_gk20a_elpg_protected_call(g, | 696 | nvgpu_mutex_acquire(&g->dbg_sessions_lock); |
697 | err = gr_gk20a_elpg_protected_call(g, | ||
699 | g->ops.gr.trigger_suspend(g)); | 698 | g->ops.gr.trigger_suspend(g)); |
700 | nvgpu_mutex_release(&g->dbg_sessions_lock); | 699 | nvgpu_mutex_release(&g->dbg_sessions_lock); |
700 | } else | ||
701 | err = -EINVAL; | ||
701 | 702 | ||
702 | gk20a_idle(g); | 703 | gk20a_idle(g); |
703 | 704 | ||
@@ -731,8 +732,13 @@ static int nvgpu_gpu_ioctl_wait_for_pause(struct gk20a *g, | |||
731 | goto out_free; | 732 | goto out_free; |
732 | 733 | ||
733 | nvgpu_mutex_acquire(&g->dbg_sessions_lock); | 734 | nvgpu_mutex_acquire(&g->dbg_sessions_lock); |
734 | (void)gr_gk20a_elpg_protected_call(g, | 735 | if (g->ops.gr.wait_for_pause) { |
736 | (void)gr_gk20a_elpg_protected_call(g, | ||
735 | g->ops.gr.wait_for_pause(g, w_state)); | 737 | g->ops.gr.wait_for_pause(g, w_state)); |
738 | } else { | ||
739 | err = -EINVAL; | ||
740 | goto out_idle; | ||
741 | } | ||
736 | 742 | ||
737 | for (sm_id = 0; sm_id < g->gr.no_of_sm; sm_id++) { | 743 | for (sm_id = 0; sm_id < g->gr.no_of_sm; sm_id++) { |
738 | ioctl_w_state[sm_id].valid_warps[0] = | 744 | ioctl_w_state[sm_id].valid_warps[0] = |
@@ -755,6 +761,7 @@ static int nvgpu_gpu_ioctl_wait_for_pause(struct gk20a *g, | |||
755 | err = -EFAULT; | 761 | err = -EFAULT; |
756 | } | 762 | } |
757 | 763 | ||
764 | out_idle: | ||
758 | nvgpu_mutex_release(&g->dbg_sessions_lock); | 765 | nvgpu_mutex_release(&g->dbg_sessions_lock); |
759 | 766 | ||
760 | gk20a_idle(g); | 767 | gk20a_idle(g); |
@@ -772,12 +779,15 @@ static int nvgpu_gpu_ioctl_resume_from_pause(struct gk20a *g) | |||
772 | 779 | ||
773 | err = gk20a_busy(g); | 780 | err = gk20a_busy(g); |
774 | if (err) | 781 | if (err) |
775 | return err; | 782 | return err; |
776 | 783 | ||
777 | nvgpu_mutex_acquire(&g->dbg_sessions_lock); | 784 | if (g->ops.gr.resume_from_pause) { |
778 | err = gr_gk20a_elpg_protected_call(g, | 785 | nvgpu_mutex_acquire(&g->dbg_sessions_lock); |
786 | err = gr_gk20a_elpg_protected_call(g, | ||
779 | g->ops.gr.resume_from_pause(g)); | 787 | g->ops.gr.resume_from_pause(g)); |
780 | nvgpu_mutex_release(&g->dbg_sessions_lock); | 788 | nvgpu_mutex_release(&g->dbg_sessions_lock); |
789 | } else | ||
790 | err = -EINVAL; | ||
781 | 791 | ||
782 | gk20a_idle(g); | 792 | gk20a_idle(g); |
783 | 793 | ||
@@ -792,8 +802,11 @@ static int nvgpu_gpu_ioctl_clear_sm_errors(struct gk20a *g) | |||
792 | if (err) | 802 | if (err) |
793 | return err; | 803 | return err; |
794 | 804 | ||
795 | err = gr_gk20a_elpg_protected_call(g, | 805 | if (g->ops.gr.clear_sm_errors) { |
806 | err = gr_gk20a_elpg_protected_call(g, | ||
796 | g->ops.gr.clear_sm_errors(g)); | 807 | g->ops.gr.clear_sm_errors(g)); |
808 | } else | ||
809 | err = -EINVAL; | ||
797 | 810 | ||
798 | gk20a_idle(g); | 811 | gk20a_idle(g); |
799 | 812 | ||
@@ -806,9 +819,12 @@ static int nvgpu_gpu_ioctl_has_any_exception( | |||
806 | { | 819 | { |
807 | u32 tpc_exception_en; | 820 | u32 tpc_exception_en; |
808 | 821 | ||
809 | nvgpu_mutex_acquire(&g->dbg_sessions_lock); | 822 | if (g->ops.gr.tpc_enabled_exceptions) { |
810 | tpc_exception_en = g->ops.gr.tpc_enabled_exceptions(g); | 823 | nvgpu_mutex_acquire(&g->dbg_sessions_lock); |
811 | nvgpu_mutex_release(&g->dbg_sessions_lock); | 824 | tpc_exception_en = g->ops.gr.tpc_enabled_exceptions(g); |
825 | nvgpu_mutex_release(&g->dbg_sessions_lock); | ||
826 | } else | ||
827 | return -EINVAL; | ||
812 | 828 | ||
813 | args->tpc_exception_en_sm_mask = tpc_exception_en; | 829 | args->tpc_exception_en_sm_mask = tpc_exception_en; |
814 | 830 | ||
@@ -2023,7 +2039,6 @@ int gk20a_ctrl_dev_mmap(struct file *filp, struct vm_area_struct *vma) | |||
2023 | vma->vm_end - vma->vm_start, vma->vm_page_prot); | 2039 | vma->vm_end - vma->vm_start, vma->vm_page_prot); |
2024 | if (!err) { | 2040 | if (!err) { |
2025 | priv->usermode_vma.vma = vma; | 2041 | priv->usermode_vma.vma = vma; |
2026 | priv->usermode_vma.flags = vma->vm_flags; | ||
2027 | vma->vm_private_data = priv; | 2042 | vma->vm_private_data = priv; |
2028 | priv->usermode_vma.vma_mapped = true; | 2043 | priv->usermode_vma.vma_mapped = true; |
2029 | } | 2044 | } |
@@ -2034,7 +2049,7 @@ int gk20a_ctrl_dev_mmap(struct file *filp, struct vm_area_struct *vma) | |||
2034 | return err; | 2049 | return err; |
2035 | } | 2050 | } |
2036 | 2051 | ||
2037 | static void alter_usermode_mapping(struct gk20a *g, | 2052 | static int alter_usermode_mapping(struct gk20a *g, |
2038 | struct gk20a_ctrl_priv *priv, | 2053 | struct gk20a_ctrl_priv *priv, |
2039 | bool poweroff) | 2054 | bool poweroff) |
2040 | { | 2055 | { |
@@ -2042,57 +2057,80 @@ static void alter_usermode_mapping(struct gk20a *g, | |||
2042 | struct vm_area_struct *vma = priv->usermode_vma.vma; | 2057 | struct vm_area_struct *vma = priv->usermode_vma.vma; |
2043 | bool vma_mapped = priv->usermode_vma.vma_mapped; | 2058 | bool vma_mapped = priv->usermode_vma.vma_mapped; |
2044 | u64 addr; | 2059 | u64 addr; |
2045 | int err; | 2060 | int err = 0; |
2046 | 2061 | ||
2047 | if (!vma) { | 2062 | if (!vma) { |
2048 | /* Nothing to do - no mmap called */ | 2063 | /* Nothing to do - no mmap called */ |
2049 | return; | 2064 | return 0; |
2050 | } | 2065 | } |
2051 | 2066 | ||
2052 | addr = l->regs_bus_addr + g->ops.fifo.usermode_base(g); | 2067 | addr = l->regs_bus_addr + g->ops.fifo.usermode_base(g); |
2053 | 2068 | ||
2054 | down_write(&vma->vm_mm->mmap_sem); | ||
2055 | |||
2056 | /* | 2069 | /* |
2057 | * This is a no-op for the below cases | 2070 | * This is a no-op for the below cases |
2058 | * a) poweroff and !vma_mapped - > do nothing as no map exists | 2071 | * a) poweroff and !vma_mapped - > do nothing as no map exists |
2059 | * b) !poweroff and vmap_mapped -> do nothing as already mapped | 2072 | * b) !poweroff and vmap_mapped -> do nothing as already mapped |
2060 | */ | 2073 | */ |
2061 | if (poweroff && vma_mapped) { | 2074 | if (poweroff != vma_mapped) { |
2075 | return 0; | ||
2076 | } | ||
2077 | |||
2078 | /* | ||
2079 | * We use trylock due to lock inversion: we need to acquire | ||
2080 | * mmap_lock while holding ctrl_privs_lock. usermode_vma_close | ||
2081 | * does it in reverse order. Trylock is a way to avoid deadlock. | ||
2082 | */ | ||
2083 | if (!down_write_trylock(&vma->vm_mm->mmap_sem)) { | ||
2084 | return -EBUSY; | ||
2085 | } | ||
2086 | |||
2087 | if (poweroff) { | ||
2062 | err = zap_vma_ptes(vma, vma->vm_start, SZ_4K); | 2088 | err = zap_vma_ptes(vma, vma->vm_start, SZ_4K); |
2063 | if (err == 0) { | 2089 | if (err == 0) { |
2064 | vma->vm_flags = VM_NONE; | ||
2065 | priv->usermode_vma.vma_mapped = false; | 2090 | priv->usermode_vma.vma_mapped = false; |
2066 | } else { | 2091 | } else { |
2067 | nvgpu_err(g, "can't remove usermode mapping"); | 2092 | nvgpu_err(g, "can't remove usermode mapping"); |
2068 | } | 2093 | } |
2069 | } else if (!poweroff && !vma_mapped) { | 2094 | } else { |
2070 | vma->vm_flags = priv->usermode_vma.flags; | ||
2071 | err = io_remap_pfn_range(vma, vma->vm_start, | 2095 | err = io_remap_pfn_range(vma, vma->vm_start, |
2072 | addr >> PAGE_SHIFT, | 2096 | addr >> PAGE_SHIFT, |
2073 | SZ_4K, vma->vm_page_prot); | 2097 | SZ_4K, vma->vm_page_prot); |
2074 | if (err != 0) { | 2098 | if (err != 0) { |
2075 | nvgpu_err(g, "can't restore usermode mapping"); | 2099 | nvgpu_err(g, "can't restore usermode mapping"); |
2076 | vma->vm_flags = VM_NONE; | ||
2077 | } else { | 2100 | } else { |
2078 | priv->usermode_vma.vma_mapped = true; | 2101 | priv->usermode_vma.vma_mapped = true; |
2079 | } | 2102 | } |
2080 | } | 2103 | } |
2081 | 2104 | ||
2082 | up_write(&vma->vm_mm->mmap_sem); | 2105 | up_write(&vma->vm_mm->mmap_sem); |
2106 | |||
2107 | return err; | ||
2083 | } | 2108 | } |
2084 | 2109 | ||
2085 | static void alter_usermode_mappings(struct gk20a *g, bool poweroff) | 2110 | static void alter_usermode_mappings(struct gk20a *g, bool poweroff) |
2086 | { | 2111 | { |
2087 | struct gk20a_ctrl_priv *priv; | 2112 | struct gk20a_ctrl_priv *priv; |
2088 | struct nvgpu_os_linux *l = nvgpu_os_linux_from_gk20a(g); | 2113 | struct nvgpu_os_linux *l = nvgpu_os_linux_from_gk20a(g); |
2114 | int err = 0; | ||
2089 | 2115 | ||
2090 | nvgpu_mutex_acquire(&l->ctrl.privs_lock); | 2116 | do { |
2091 | nvgpu_list_for_each_entry(priv, &l->ctrl.privs, | 2117 | nvgpu_mutex_acquire(&l->ctrl.privs_lock); |
2092 | gk20a_ctrl_priv, list) { | 2118 | nvgpu_list_for_each_entry(priv, &l->ctrl.privs, |
2093 | alter_usermode_mapping(g, priv, poweroff); | 2119 | gk20a_ctrl_priv, list) { |
2094 | } | 2120 | err = alter_usermode_mapping(g, priv, poweroff); |
2095 | nvgpu_mutex_release(&l->ctrl.privs_lock); | 2121 | if (err != 0) { |
2122 | break; | ||
2123 | } | ||
2124 | } | ||
2125 | nvgpu_mutex_release(&l->ctrl.privs_lock); | ||
2126 | |||
2127 | if (err == -EBUSY) { | ||
2128 | nvgpu_log_info(g, "ctrl_privs_lock lock contended. retry altering usermode mappings"); | ||
2129 | nvgpu_udelay(10); | ||
2130 | } else if (err != 0) { | ||
2131 | nvgpu_err(g, "can't alter usermode mapping. err = %d", err); | ||
2132 | } | ||
2133 | } while (err == -EBUSY); | ||
2096 | } | 2134 | } |
2097 | 2135 | ||
2098 | void nvgpu_hide_usermode_for_poweroff(struct gk20a *g) | 2136 | void nvgpu_hide_usermode_for_poweroff(struct gk20a *g) |
diff --git a/include/os/linux/ioctl_tsg.c b/include/os/linux/ioctl_tsg.c index 2f8cb3a..296b02b 100644 --- a/include/os/linux/ioctl_tsg.c +++ b/include/os/linux/ioctl_tsg.c | |||
@@ -1,5 +1,5 @@ | |||
1 | /* | 1 | /* |
2 | * Copyright (c) 2014-2020, NVIDIA CORPORATION. All rights reserved. | 2 | * Copyright (c) 2014-2021, NVIDIA CORPORATION. All rights reserved. |
3 | * | 3 | * |
4 | * This program is free software; you can redistribute it and/or modify it | 4 | * This program is free software; you can redistribute it and/or modify it |
5 | * under the terms and conditions of the GNU General Public License, | 5 | * under the terms and conditions of the GNU General Public License, |
@@ -134,7 +134,10 @@ static int gk20a_tsg_unbind_channel_fd(struct tsg_gk20a *tsg, int ch_fd) | |||
134 | goto out; | 134 | goto out; |
135 | } | 135 | } |
136 | 136 | ||
137 | err = gk20a_tsg_unbind_channel(ch); | 137 | err = gk20a_tsg_unbind_channel(ch, false); |
138 | if (err == -EAGAIN) { | ||
139 | goto out; | ||
140 | } | ||
138 | 141 | ||
139 | /* | 142 | /* |
140 | * Mark the channel timedout since channel unbound from TSG | 143 | * Mark the channel timedout since channel unbound from TSG |
@@ -307,17 +310,10 @@ static int gk20a_tsg_event_id_enable(struct tsg_gk20a *tsg, | |||
307 | snprintf(name, sizeof(name), "nvgpu-event%d-fd%d", | 310 | snprintf(name, sizeof(name), "nvgpu-event%d-fd%d", |
308 | event_id, local_fd); | 311 | event_id, local_fd); |
309 | 312 | ||
310 | file = anon_inode_getfile(name, &gk20a_event_id_ops, | ||
311 | NULL, O_RDWR); | ||
312 | if (IS_ERR(file)) { | ||
313 | err = PTR_ERR(file); | ||
314 | goto clean_up; | ||
315 | } | ||
316 | |||
317 | event_id_data = nvgpu_kzalloc(tsg->g, sizeof(*event_id_data)); | 313 | event_id_data = nvgpu_kzalloc(tsg->g, sizeof(*event_id_data)); |
318 | if (!event_id_data) { | 314 | if (!event_id_data) { |
319 | err = -ENOMEM; | 315 | err = -ENOMEM; |
320 | goto clean_up_file; | 316 | goto clean_up; |
321 | } | 317 | } |
322 | event_id_data->g = g; | 318 | event_id_data->g = g; |
323 | event_id_data->id = tsg->tsgid; | 319 | event_id_data->id = tsg->tsgid; |
@@ -330,12 +326,18 @@ static int gk20a_tsg_event_id_enable(struct tsg_gk20a *tsg, | |||
330 | 326 | ||
331 | nvgpu_init_list_node(&event_id_data->event_id_node); | 327 | nvgpu_init_list_node(&event_id_data->event_id_node); |
332 | 328 | ||
329 | file = anon_inode_getfile(name, &gk20a_event_id_ops, | ||
330 | event_id_data, O_RDWR); | ||
331 | if (IS_ERR(file)) { | ||
332 | err = PTR_ERR(file); | ||
333 | goto clean_up_free; | ||
334 | } | ||
335 | |||
333 | nvgpu_mutex_acquire(&tsg->event_id_list_lock); | 336 | nvgpu_mutex_acquire(&tsg->event_id_list_lock); |
334 | nvgpu_list_add_tail(&event_id_data->event_id_node, &tsg->event_id_list); | 337 | nvgpu_list_add_tail(&event_id_data->event_id_node, &tsg->event_id_list); |
335 | nvgpu_mutex_release(&tsg->event_id_list_lock); | 338 | nvgpu_mutex_release(&tsg->event_id_list_lock); |
336 | 339 | ||
337 | fd_install(local_fd, file); | 340 | fd_install(local_fd, file); |
338 | file->private_data = event_id_data; | ||
339 | 341 | ||
340 | *fd = local_fd; | 342 | *fd = local_fd; |
341 | 343 | ||
@@ -343,8 +345,6 @@ static int gk20a_tsg_event_id_enable(struct tsg_gk20a *tsg, | |||
343 | 345 | ||
344 | clean_up_free: | 346 | clean_up_free: |
345 | nvgpu_kfree(g, event_id_data); | 347 | nvgpu_kfree(g, event_id_data); |
346 | clean_up_file: | ||
347 | fput(file); | ||
348 | clean_up: | 348 | clean_up: |
349 | put_unused_fd(local_fd); | 349 | put_unused_fd(local_fd); |
350 | free_ref: | 350 | free_ref: |
diff --git a/include/os/linux/module.c b/include/os/linux/module.c index 807df2c..fdbab46 100644 --- a/include/os/linux/module.c +++ b/include/os/linux/module.c | |||
@@ -1,7 +1,7 @@ | |||
1 | /* | 1 | /* |
2 | * GK20A Graphics | 2 | * GK20A Graphics |
3 | * | 3 | * |
4 | * Copyright (c) 2011-2020, NVIDIA CORPORATION. All rights reserved. | 4 | * Copyright (c) 2011-2021, NVIDIA CORPORATION. All rights reserved. |
5 | * | 5 | * |
6 | * This program is free software; you can redistribute it and/or modify it | 6 | * This program is free software; you can redistribute it and/or modify it |
7 | * under the terms and conditions of the GNU General Public License, | 7 | * under the terms and conditions of the GNU General Public License, |
@@ -49,6 +49,7 @@ | |||
49 | #include <nvgpu/clk_arb.h> | 49 | #include <nvgpu/clk_arb.h> |
50 | #include <nvgpu/timers.h> | 50 | #include <nvgpu/timers.h> |
51 | #include <nvgpu/channel.h> | 51 | #include <nvgpu/channel.h> |
52 | #include <nvgpu/nvgpu_err.h> | ||
52 | 53 | ||
53 | #include "platform_gk20a.h" | 54 | #include "platform_gk20a.h" |
54 | #include "sysfs.h" | 55 | #include "sysfs.h" |
@@ -355,6 +356,10 @@ int gk20a_pm_finalize_poweron(struct device *dev) | |||
355 | gk20a_init_cde_support(l); | 356 | gk20a_init_cde_support(l); |
356 | #endif | 357 | #endif |
357 | 358 | ||
359 | #ifdef CONFIG_NVGPU_SUPPORT_LINUX_ECC_ERROR_REPORTING | ||
360 | nvgpu_enable_ecc_reporting(g); | ||
361 | #endif | ||
362 | |||
358 | err = gk20a_sched_ctrl_init(g); | 363 | err = gk20a_sched_ctrl_init(g); |
359 | if (err) { | 364 | if (err) { |
360 | nvgpu_err(g, "failed to init sched control"); | 365 | nvgpu_err(g, "failed to init sched control"); |
@@ -364,9 +369,14 @@ int gk20a_pm_finalize_poweron(struct device *dev) | |||
364 | g->sw_ready = true; | 369 | g->sw_ready = true; |
365 | 370 | ||
366 | done: | 371 | done: |
367 | if (err) | 372 | if (err) { |
368 | g->power_on = false; | 373 | g->power_on = false; |
369 | 374 | ||
375 | #ifdef CONFIG_NVGPU_SUPPORT_LINUX_ECC_ERROR_REPORTING | ||
376 | nvgpu_disable_ecc_reporting(g); | ||
377 | #endif | ||
378 | } | ||
379 | |||
370 | nvgpu_mutex_release(&g->power_lock); | 380 | nvgpu_mutex_release(&g->power_lock); |
371 | return err; | 381 | return err; |
372 | } | 382 | } |
@@ -433,6 +443,10 @@ static int gk20a_pm_prepare_poweroff(struct device *dev) | |||
433 | /* Stop CPU from accessing the GPU registers. */ | 443 | /* Stop CPU from accessing the GPU registers. */ |
434 | gk20a_lockout_registers(g); | 444 | gk20a_lockout_registers(g); |
435 | 445 | ||
446 | #ifdef CONFIG_NVGPU_SUPPORT_LINUX_ECC_ERROR_REPORTING | ||
447 | nvgpu_disable_ecc_reporting(g); | ||
448 | #endif | ||
449 | |||
436 | nvgpu_hide_usermode_for_poweroff(g); | 450 | nvgpu_hide_usermode_for_poweroff(g); |
437 | nvgpu_mutex_release(&g->power_lock); | 451 | nvgpu_mutex_release(&g->power_lock); |
438 | return 0; | 452 | return 0; |
@@ -1382,6 +1396,10 @@ static int gk20a_probe(struct platform_device *dev) | |||
1382 | goto return_err; | 1396 | goto return_err; |
1383 | } | 1397 | } |
1384 | 1398 | ||
1399 | #ifdef CONFIG_NVGPU_SUPPORT_LINUX_ECC_ERROR_REPORTING | ||
1400 | nvgpu_init_ecc_reporting(gk20a); | ||
1401 | #endif | ||
1402 | |||
1385 | gk20a->nvgpu_reboot_nb.notifier_call = | 1403 | gk20a->nvgpu_reboot_nb.notifier_call = |
1386 | nvgpu_kernel_shutdown_notification; | 1404 | nvgpu_kernel_shutdown_notification; |
1387 | err = register_reboot_notifier(&gk20a->nvgpu_reboot_nb); | 1405 | err = register_reboot_notifier(&gk20a->nvgpu_reboot_nb); |
diff --git a/include/os/linux/os_linux.h b/include/os/linux/os_linux.h index 25c6c03..adcfdb2 100644 --- a/include/os/linux/os_linux.h +++ b/include/os/linux/os_linux.h | |||
@@ -1,5 +1,5 @@ | |||
1 | /* | 1 | /* |
2 | * Copyright (c) 2017-2020, NVIDIA CORPORATION. All rights reserved. | 2 | * Copyright (c) 2017-2021, NVIDIA CORPORATION. All rights reserved. |
3 | * | 3 | * |
4 | * This program is free software; you can redistribute it and/or modify it | 4 | * This program is free software; you can redistribute it and/or modify it |
5 | * under the terms and conditions of the GNU General Public License, | 5 | * under the terms and conditions of the GNU General Public License, |
@@ -25,6 +25,7 @@ | |||
25 | 25 | ||
26 | #include "cde.h" | 26 | #include "cde.h" |
27 | #include "sched.h" | 27 | #include "sched.h" |
28 | #include "ecc_linux.h" | ||
28 | 29 | ||
29 | struct nvgpu_os_linux_ops { | 30 | struct nvgpu_os_linux_ops { |
30 | struct { | 31 | struct { |
@@ -134,6 +135,10 @@ struct nvgpu_os_linux { | |||
134 | 135 | ||
135 | u64 regs_bus_addr; | 136 | u64 regs_bus_addr; |
136 | 137 | ||
138 | #ifdef CONFIG_NVGPU_SUPPORT_LINUX_ECC_ERROR_REPORTING | ||
139 | struct nvgpu_ecc_reporting_linux ecc_reporting_linux; | ||
140 | #endif | ||
141 | |||
137 | struct nvgpu_os_linux_ops ops; | 142 | struct nvgpu_os_linux_ops ops; |
138 | 143 | ||
139 | #ifdef CONFIG_DEBUG_FS | 144 | #ifdef CONFIG_DEBUG_FS |
diff --git a/include/os/linux/platform_gv11b_tegra.c b/include/os/linux/platform_gv11b_tegra.c index 6c9d0f5..7900eaa 100644 --- a/include/os/linux/platform_gv11b_tegra.c +++ b/include/os/linux/platform_gv11b_tegra.c | |||
@@ -1,7 +1,7 @@ | |||
1 | /* | 1 | /* |
2 | * GV11B Tegra Platform Interface | 2 | * GV11B Tegra Platform Interface |
3 | * | 3 | * |
4 | * Copyright (c) 2016-2018, NVIDIA CORPORATION. All rights reserved. | 4 | * Copyright (c) 2016-2022, NVIDIA CORPORATION. All rights reserved. |
5 | * | 5 | * |
6 | * This program is free software; you can redistribute it and/or modify it | 6 | * This program is free software; you can redistribute it and/or modify it |
7 | * under the terms and conditions of the GNU General Public License, | 7 | * under the terms and conditions of the GNU General Public License, |
@@ -322,7 +322,7 @@ struct gk20a_platform gv11b_tegra_platform = { | |||
322 | 322 | ||
323 | .honors_aperture = true, | 323 | .honors_aperture = true, |
324 | .unified_memory = true, | 324 | .unified_memory = true, |
325 | .dma_mask = DMA_BIT_MASK(36), | 325 | .dma_mask = DMA_BIT_MASK(38), |
326 | 326 | ||
327 | .reset_assert = gp10b_tegra_reset_assert, | 327 | .reset_assert = gp10b_tegra_reset_assert, |
328 | .reset_deassert = gp10b_tegra_reset_deassert, | 328 | .reset_deassert = gp10b_tegra_reset_deassert, |
diff --git a/include/os/linux/scale.c b/include/os/linux/scale.c index 388e168..f8f0ef9 100644 --- a/include/os/linux/scale.c +++ b/include/os/linux/scale.c | |||
@@ -1,7 +1,7 @@ | |||
1 | /* | 1 | /* |
2 | * gk20a clock scaling profile | 2 | * gk20a clock scaling profile |
3 | * | 3 | * |
4 | * Copyright (c) 2013-2020, NVIDIA Corporation. All rights reserved. | 4 | * Copyright (c) 2013-2023, NVIDIA Corporation. All rights reserved. |
5 | * | 5 | * |
6 | * This program is free software; you can redistribute it and/or modify it | 6 | * This program is free software; you can redistribute it and/or modify it |
7 | * under the terms and conditions of the GNU General Public License, | 7 | * under the terms and conditions of the GNU General Public License, |
@@ -148,32 +148,24 @@ static int gk20a_scale_target(struct device *dev, unsigned long *freq, | |||
148 | { | 148 | { |
149 | struct gk20a_platform *platform = dev_get_drvdata(dev); | 149 | struct gk20a_platform *platform = dev_get_drvdata(dev); |
150 | struct gk20a *g = platform->g; | 150 | struct gk20a *g = platform->g; |
151 | struct nvgpu_os_linux *l = nvgpu_os_linux_from_gk20a(g); | ||
152 | struct gk20a_scale_profile *profile = g->scale_profile; | 151 | struct gk20a_scale_profile *profile = g->scale_profile; |
153 | struct devfreq *devfreq = l->devfreq; | ||
154 | unsigned long local_freq = *freq; | 152 | unsigned long local_freq = *freq; |
155 | unsigned long rounded_rate; | 153 | unsigned long rounded_rate; |
154 | #ifdef CONFIG_GK20A_PM_QOS | ||
156 | unsigned long min_freq = 0, max_freq = 0; | 155 | unsigned long min_freq = 0, max_freq = 0; |
156 | #endif | ||
157 | 157 | ||
158 | if (nvgpu_clk_arb_has_active_req(g)) | 158 | if (nvgpu_clk_arb_has_active_req(g)) |
159 | return 0; | 159 | return 0; |
160 | |||
161 | #ifdef CONFIG_GK20A_PM_QOS | ||
160 | /* | 162 | /* |
161 | * Calculate floor and cap frequency values | 163 | * devfreq takes care of min/max freq clipping in update_devfreq() then |
162 | * | 164 | * invoked devfreq->profile->target(), thus we only need to do freq |
163 | * Policy : | 165 | * clipping based on pm_qos constraint |
164 | * We have two APIs to clip the frequency | ||
165 | * 1. devfreq | ||
166 | * 2. pm_qos | ||
167 | * | ||
168 | * To calculate floor (min) freq, we select MAX of floor frequencies | ||
169 | * requested from both APIs | ||
170 | * To get cap (max) freq, we select MIN of max frequencies | ||
171 | * | ||
172 | * In case we have conflict (min_freq > max_freq) after above | ||
173 | * steps, we ensure that max_freq wins over min_freq | ||
174 | */ | 166 | */ |
175 | min_freq = max_t(u32, devfreq->min_freq, profile->qos_min_freq); | 167 | min_freq = profile->qos_min_freq; |
176 | max_freq = min_t(u32, devfreq->max_freq, profile->qos_max_freq); | 168 | max_freq = profile->qos_max_freq; |
177 | 169 | ||
178 | if (min_freq > max_freq) | 170 | if (min_freq > max_freq) |
179 | min_freq = max_freq; | 171 | min_freq = max_freq; |
@@ -184,6 +176,7 @@ static int gk20a_scale_target(struct device *dev, unsigned long *freq, | |||
184 | 176 | ||
185 | if (local_freq > max_freq) | 177 | if (local_freq > max_freq) |
186 | local_freq = max_freq; | 178 | local_freq = max_freq; |
179 | #endif | ||
187 | 180 | ||
188 | /* set the final frequency */ | 181 | /* set the final frequency */ |
189 | rounded_rate = platform->clk_round_rate(dev, local_freq); | 182 | rounded_rate = platform->clk_round_rate(dev, local_freq); |
diff --git a/include/os/linux/sdl.c b/include/os/linux/sdl.c new file mode 100644 index 0000000..c4dccdc --- /dev/null +++ b/include/os/linux/sdl.c | |||
@@ -0,0 +1,341 @@ | |||
1 | /* | ||
2 | * Copyright (c) 2021, NVIDIA Corporation. All rights reserved. | ||
3 | * | ||
4 | * This program is free software; you can redistribute it and/or modify it | ||
5 | * under the terms and conditions of the GNU General Public License, | ||
6 | * version 2, as published by the Free Software Foundation. | ||
7 | * | ||
8 | * This program is distributed in the hope it will be useful, but WITHOUT | ||
9 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | ||
10 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for | ||
11 | * more details. | ||
12 | * | ||
13 | * You should have received a copy of the GNU General Public License | ||
14 | * along with this program. If not, see <http://www.gnu.org/licenses/>. | ||
15 | */ | ||
16 | |||
17 | #include <nvgpu/gk20a.h> | ||
18 | #include <nvgpu/types.h> | ||
19 | #include <nvgpu/nvgpu_err.h> | ||
20 | #include <nvgpu/timers.h> | ||
21 | #include <nvgpu/bug.h> | ||
22 | |||
23 | #include "ecc_linux.h" | ||
24 | #include "os_linux.h" | ||
25 | #include "module.h" | ||
26 | |||
27 | /* This look-up table initializes the list of hw units and their errors. | ||
28 | * It also specifies the error injection mechanism supported, for each error. | ||
29 | * In case of hw error injection support, this initialization will be overriden | ||
30 | * by the values provided from the hal layes of corresponding hw units. | ||
31 | */ | ||
32 | static struct nvgpu_err_hw_module gv11b_err_lut[] = { | ||
33 | { | ||
34 | .name = "sm", | ||
35 | .hw_unit = (u32)NVGPU_ERR_MODULE_SM, | ||
36 | .num_errs = 21U, | ||
37 | .base_ecc_service_id = | ||
38 | NVGUARD_SERVICE_IGPU_SM_SWERR_L1_TAG_ECC_CORRECTED, | ||
39 | .errs = (struct nvgpu_err_desc[]) { | ||
40 | GPU_NONCRITERR("l1_tag_ecc_corrected", | ||
41 | GPU_SM_L1_TAG_ECC_CORRECTED, 0, 0), | ||
42 | GPU_CRITERR("l1_tag_ecc_uncorrected", | ||
43 | GPU_SM_L1_TAG_ECC_UNCORRECTED, 0, 0), | ||
44 | GPU_NONCRITERR("cbu_ecc_corrected", 0, 0, 0), | ||
45 | GPU_CRITERR("cbu_ecc_uncorrected", | ||
46 | GPU_SM_CBU_ECC_UNCORRECTED, 0, 0), | ||
47 | GPU_NONCRITERR("lrf_ecc_corrected", 0, 0, 0), | ||
48 | GPU_CRITERR("lrf_ecc_uncorrected", | ||
49 | GPU_SM_LRF_ECC_UNCORRECTED, 0, 0), | ||
50 | GPU_NONCRITERR("l1_data_ecc_corrected", 0, 0, 0), | ||
51 | GPU_CRITERR("l1_data_ecc_uncorrected", | ||
52 | GPU_SM_L1_DATA_ECC_UNCORRECTED, 0, 0), | ||
53 | GPU_NONCRITERR("icache_l0_data_ecc_corrected", 0, 0, 0), | ||
54 | GPU_CRITERR("icache_l0_data_ecc_uncorrected", | ||
55 | GPU_SM_ICACHE_L0_DATA_ECC_UNCORRECTED, 0, 0), | ||
56 | GPU_NONCRITERR("icache_l1_data_ecc_corrected", 0, 0, 0), | ||
57 | GPU_CRITERR("icache_l1_data_ecc_uncorrected", | ||
58 | GPU_SM_ICACHE_L1_DATA_ECC_UNCORRECTED, 0, 0), | ||
59 | GPU_NONCRITERR("icache_l0_predecode_ecc_corrected", 0, 0, 0), | ||
60 | GPU_CRITERR("icache_l0_predecode_ecc_uncorrected", | ||
61 | GPU_SM_ICACHE_L0_PREDECODE_ECC_UNCORRECTED, 0, 0), | ||
62 | GPU_NONCRITERR("l1_tag_miss_fifo_ecc_corrected", 0, 0, 0), | ||
63 | GPU_CRITERR("l1_tag_miss_fifo_ecc_uncorrected", | ||
64 | GPU_SM_L1_TAG_MISS_FIFO_ECC_UNCORRECTED, 0, 0), | ||
65 | GPU_NONCRITERR("l1_tag_s2r_pixprf_ecc_corrected", 0, 0, 0), | ||
66 | GPU_CRITERR("l1_tag_s2r_pixprf_ecc_uncorrected", | ||
67 | GPU_SM_L1_TAG_S2R_PIXPRF_ECC_UNCORRECTED, 0, 0), | ||
68 | GPU_CRITERR("machine_check_error", 0, 0, 0), | ||
69 | GPU_NONCRITERR("icache_l1_predecode_ecc_corrected", 0, 0, 0), | ||
70 | GPU_CRITERR("icache_l1_predecode_ecc_uncorrected", | ||
71 | GPU_SM_ICACHE_L1_PREDECODE_ECC_UNCORRECTED, 0, 0), | ||
72 | }, | ||
73 | }, | ||
74 | { | ||
75 | .name = "fecs", | ||
76 | .hw_unit = (u32)NVGPU_ERR_MODULE_FECS, | ||
77 | .num_errs = 4U, | ||
78 | .base_ecc_service_id = | ||
79 | NVGUARD_SERVICE_IGPU_FECS_SWERR_FALCON_IMEM_ECC_CORRECTED, | ||
80 | .errs = (struct nvgpu_err_desc[]) { | ||
81 | GPU_NONCRITERR("falcon_imem_ecc_corrected", | ||
82 | GPU_FECS_FALCON_IMEM_ECC_CORRECTED, 0, 0), | ||
83 | GPU_CRITERR("falcon_imem_ecc_uncorrected", | ||
84 | GPU_FECS_FALCON_IMEM_ECC_UNCORRECTED, 0, 0), | ||
85 | GPU_NONCRITERR("falcon_dmem_ecc_corrected", 0, 0, 0), | ||
86 | GPU_CRITERR("falcon_dmem_ecc_uncorrected", | ||
87 | GPU_FECS_FALCON_DMEM_ECC_UNCORRECTED, 0, 0), | ||
88 | }, | ||
89 | }, | ||
90 | { | ||
91 | .name = "pmu", | ||
92 | .hw_unit = NVGPU_ERR_MODULE_PMU, | ||
93 | .num_errs = 4U, | ||
94 | .base_ecc_service_id = | ||
95 | NVGUARD_SERVICE_IGPU_PMU_SWERR_FALCON_IMEM_ECC_CORRECTED, | ||
96 | .errs = (struct nvgpu_err_desc[]) { | ||
97 | GPU_NONCRITERR("falcon_imem_ecc_corrected", | ||
98 | GPU_PMU_FALCON_IMEM_ECC_CORRECTED, 0, 0), | ||
99 | GPU_CRITERR("falcon_imem_ecc_uncorrected", | ||
100 | GPU_PMU_FALCON_IMEM_ECC_UNCORRECTED, 0, 0), | ||
101 | GPU_NONCRITERR("falcon_dmem_ecc_corrected", 0, 0, 0), | ||
102 | GPU_CRITERR("falcon_dmem_ecc_uncorrected", | ||
103 | GPU_PMU_FALCON_DMEM_ECC_UNCORRECTED, 0, 0), | ||
104 | }, | ||
105 | }, | ||
106 | }; | ||
107 | |||
108 | static void nvgpu_init_err_msg_header(struct gpu_err_header *header) | ||
109 | { | ||
110 | header->version.major = (u16)1U; | ||
111 | header->version.minor = (u16)0U; | ||
112 | header->sub_err_type = 0U; | ||
113 | header->sub_unit_id = 0UL; | ||
114 | header->address = 0UL; | ||
115 | header->timestamp_ns = 0UL; | ||
116 | } | ||
117 | |||
118 | static void nvgpu_init_ecc_err_msg(struct gpu_ecc_error_info *err_info) | ||
119 | { | ||
120 | nvgpu_init_err_msg_header(&err_info->header); | ||
121 | err_info->err_cnt = 0UL; | ||
122 | } | ||
123 | |||
124 | static void nvgpu_report_ecc_error_linux(struct gk20a *g, u32 hw_unit, u32 inst, | ||
125 | u32 err_id, u64 err_addr, u64 err_count) | ||
126 | { | ||
127 | int err = 0; | ||
128 | u32 s_id = 0; | ||
129 | u8 err_status = 0; | ||
130 | u8 err_info_size = 0; | ||
131 | u64 timestamp = 0ULL; | ||
132 | int err_threshold_counter = 0; | ||
133 | struct gpu_ecc_error_info err_pkt; | ||
134 | struct nvgpu_err_desc *err_desc = NULL; | ||
135 | struct nvgpu_err_hw_module *hw_module = NULL; | ||
136 | nv_guard_request_t req; | ||
137 | |||
138 | memset(&req, 0, sizeof(req)); | ||
139 | nvgpu_init_ecc_err_msg(&err_pkt); | ||
140 | if (hw_unit >= sizeof(gv11b_err_lut)/sizeof(gv11b_err_lut[0])) { | ||
141 | err = -EINVAL; | ||
142 | goto done; | ||
143 | } | ||
144 | |||
145 | hw_module = &gv11b_err_lut[hw_unit]; | ||
146 | if (err_id >= hw_module->num_errs) { | ||
147 | nvgpu_err(g, "invalid err_id (%u) for hw module (%u)", | ||
148 | err_id, hw_module->hw_unit); | ||
149 | err = -EINVAL; | ||
150 | goto done; | ||
151 | } | ||
152 | err_desc = &hw_module->errs[err_id]; | ||
153 | timestamp = (u64)nvgpu_current_time_ns(); | ||
154 | |||
155 | err_pkt.header.timestamp_ns = timestamp; | ||
156 | err_pkt.header.sub_unit_id = inst; | ||
157 | err_pkt.header.address = err_addr; | ||
158 | err_pkt.err_cnt = err_count; | ||
159 | err_info_size = sizeof(err_pkt); | ||
160 | |||
161 | s_id = hw_module->base_ecc_service_id + err_id; | ||
162 | |||
163 | if (err_desc->is_critical) { | ||
164 | err_status = NVGUARD_ERROR_DETECTED; | ||
165 | } else { | ||
166 | err_status = NVGUARD_NO_ERROR; | ||
167 | } | ||
168 | |||
169 | nvgpu_atomic_inc(&err_desc->err_count); | ||
170 | err_threshold_counter = nvgpu_atomic_cmpxchg(&err_desc->err_count, | ||
171 | err_desc->err_threshold + 1, 0); | ||
172 | |||
173 | if (unlikely(err_threshold_counter != err_desc->err_threshold + 1)) { | ||
174 | goto done; | ||
175 | } | ||
176 | |||
177 | nvgpu_log(g, gpu_dbg_ecc, "ECC reporting hw: %s, desc:%s, count:%llu", | ||
178 | hw_module->name, err_desc->name, err_count); | ||
179 | |||
180 | req.srv_id_cmd = NVGUARD_SERVICESTATUS_NOTIFICATION; | ||
181 | req.srv_status.srv_id = (nv_guard_service_id_t)s_id; | ||
182 | req.srv_status.status = err_status; | ||
183 | req.srv_status.timestamp = timestamp; | ||
184 | req.srv_status.error_info_size = err_info_size; | ||
185 | memcpy(req.srv_status.error_info, (u8*)&err_pkt, err_info_size); | ||
186 | |||
187 | /* | ||
188 | * l1ss_submit_rq may fail due to kmalloc failures but may pass in | ||
189 | * subsequent calls | ||
190 | */ | ||
191 | err = l1ss_submit_rq(&req, true); | ||
192 | if (err != 0) { | ||
193 | nvgpu_err(g, "Error returned from L1SS submit %d", err); | ||
194 | } | ||
195 | |||
196 | if (err_desc->is_critical) { | ||
197 | nvgpu_quiesce(g); | ||
198 | } | ||
199 | |||
200 | done: | ||
201 | return; | ||
202 | } | ||
203 | |||
204 | static void nvgpu_report_ecc_error_empty(struct gk20a *g, u32 hw_unit, u32 inst, | ||
205 | u32 err_id, u64 err_addr, u64 err_count) { | ||
206 | nvgpu_log(g, gpu_dbg_ecc, "ECC reporting empty"); | ||
207 | } | ||
208 | |||
209 | const struct nvgpu_ecc_reporting_ops default_disabled_ecc_report_ops = { | ||
210 | .report_ecc_err = nvgpu_report_ecc_error_empty, | ||
211 | }; | ||
212 | |||
213 | const struct nvgpu_ecc_reporting_ops ecc_enable_report_ops = { | ||
214 | .report_ecc_err = nvgpu_report_ecc_error_linux, | ||
215 | }; | ||
216 | |||
217 | static int nvgpu_l1ss_callback(l1ss_cli_callback_param param, void *data) | ||
218 | { | ||
219 | struct gk20a *g = (struct gk20a *)data; | ||
220 | struct nvgpu_os_linux *l = NULL; | ||
221 | struct nvgpu_ecc_reporting_linux *ecc_reporting_linux = NULL; | ||
222 | int err = 0; | ||
223 | /* Ensure we have a valid gk20a struct before proceeding */ | ||
224 | if ((g == NULL) || (gk20a_get(g) == NULL)) { | ||
225 | return -ENODEV; | ||
226 | } | ||
227 | |||
228 | l = nvgpu_os_linux_from_gk20a(g); | ||
229 | ecc_reporting_linux = &l->ecc_reporting_linux; | ||
230 | |||
231 | nvgpu_spinlock_acquire(&ecc_reporting_linux->common.lock); | ||
232 | if (param == L1SS_READY) { | ||
233 | if (!ecc_reporting_linux->common.ecc_reporting_service_enabled) { | ||
234 | ecc_reporting_linux->common.ecc_reporting_service_enabled = true; | ||
235 | ecc_reporting_linux->common.ops = &ecc_enable_report_ops; | ||
236 | nvgpu_log(g, gpu_dbg_ecc, "ECC reporting is enabled"); | ||
237 | } | ||
238 | } else if (param == L1SS_NOT_READY) { | ||
239 | if (ecc_reporting_linux->common.ecc_reporting_service_enabled) { | ||
240 | ecc_reporting_linux->common.ecc_reporting_service_enabled = false; | ||
241 | ecc_reporting_linux->common.ops = &default_disabled_ecc_report_ops; | ||
242 | nvgpu_log(g, gpu_dbg_ecc, "ECC reporting is disabled"); | ||
243 | } | ||
244 | } else { | ||
245 | err = -EINVAL; | ||
246 | } | ||
247 | nvgpu_spinlock_release(&ecc_reporting_linux->common.lock); | ||
248 | |||
249 | gk20a_put(g); | ||
250 | |||
251 | return err; | ||
252 | } | ||
253 | |||
254 | void nvgpu_init_ecc_reporting(struct gk20a *g) | ||
255 | { | ||
256 | struct nvgpu_os_linux *l = nvgpu_os_linux_from_gk20a(g); | ||
257 | struct nvgpu_ecc_reporting_linux *ecc_report_linux = &l->ecc_reporting_linux; | ||
258 | int err = 0; | ||
259 | /* This will invoke the registration API */ | ||
260 | nvgpu_spinlock_init(&ecc_report_linux->common.lock); | ||
261 | ecc_report_linux->priv.id = (NVGUARD_GROUPID_IGPU & NVGUARD_GROUPINDEX_FIELDMASK); | ||
262 | ecc_report_linux->priv.cli_callback = nvgpu_l1ss_callback; | ||
263 | ecc_report_linux->priv.data = g; | ||
264 | ecc_report_linux->common.ops = &default_disabled_ecc_report_ops; | ||
265 | |||
266 | nvgpu_log(g, gpu_dbg_ecc, "ECC reporting Init"); | ||
267 | |||
268 | /* | ||
269 | * err == 0 indicates service is available but not active yet. | ||
270 | * err == 1 indicates service is available and active | ||
271 | * error for other cases. | ||
272 | */ | ||
273 | err = l1ss_register_client(&ecc_report_linux->priv); | ||
274 | if (err == 0) { | ||
275 | ecc_report_linux->common.ecc_reporting_service_enabled = false; | ||
276 | nvgpu_log(g, gpu_dbg_ecc, "ECC reporting init success"); | ||
277 | } else if (err == 1) { | ||
278 | ecc_report_linux->common.ecc_reporting_service_enabled = true; | ||
279 | /* Actual Ops will be replaced during nvgpu_enable_ecc_reporting | ||
280 | * called as part of gk20a_busy() | ||
281 | */ | ||
282 | } else { | ||
283 | nvgpu_log(g, gpu_dbg_ecc, "ECC reporting init failure %d", err); | ||
284 | } | ||
285 | } | ||
286 | |||
287 | void nvgpu_deinit_ecc_reporting(struct gk20a *g) | ||
288 | { | ||
289 | struct nvgpu_os_linux *l = nvgpu_os_linux_from_gk20a(g); | ||
290 | struct nvgpu_ecc_reporting_linux *ecc_report_linux = &l->ecc_reporting_linux; | ||
291 | |||
292 | if (ecc_report_linux->common.ecc_reporting_service_enabled) { | ||
293 | ecc_report_linux->common.ecc_reporting_service_enabled = false; | ||
294 | l1ss_deregister_client(ecc_report_linux->priv.id); | ||
295 | memset(ecc_report_linux, 0, sizeof(*ecc_report_linux)); | ||
296 | nvgpu_log(g, gpu_dbg_ecc, "ECC reporting de-init success"); | ||
297 | } | ||
298 | |||
299 | } | ||
300 | |||
301 | void nvgpu_enable_ecc_reporting(struct gk20a *g) | ||
302 | { | ||
303 | struct nvgpu_os_linux *l = nvgpu_os_linux_from_gk20a(g); | ||
304 | struct nvgpu_ecc_reporting_linux *ecc_report_linux = &l->ecc_reporting_linux; | ||
305 | struct nvgpu_ecc_reporting *error_reporting = &ecc_report_linux->common; | ||
306 | |||
307 | nvgpu_spinlock_acquire(&ecc_report_linux->common.lock); | ||
308 | if (error_reporting->ecc_reporting_service_enabled) { | ||
309 | error_reporting->ops = &ecc_enable_report_ops; | ||
310 | nvgpu_log(g, gpu_dbg_ecc, "ECC reporting is enabled"); | ||
311 | } | ||
312 | nvgpu_spinlock_release(&ecc_report_linux->common.lock); | ||
313 | } | ||
314 | |||
315 | void nvgpu_disable_ecc_reporting(struct gk20a *g) | ||
316 | { | ||
317 | struct nvgpu_os_linux *l = nvgpu_os_linux_from_gk20a(g); | ||
318 | struct nvgpu_ecc_reporting_linux *ecc_report_linux = &l->ecc_reporting_linux; | ||
319 | struct nvgpu_ecc_reporting *error_reporting = &ecc_report_linux->common; | ||
320 | |||
321 | nvgpu_spinlock_acquire(&ecc_report_linux->common.lock); | ||
322 | error_reporting->ops = &default_disabled_ecc_report_ops; | ||
323 | nvgpu_log(g, gpu_dbg_ecc, "ECC reporting is disabled"); | ||
324 | nvgpu_spinlock_release(&ecc_report_linux->common.lock); | ||
325 | } | ||
326 | |||
327 | void nvgpu_report_ecc_err(struct gk20a *g, u32 hw_unit, u32 inst, | ||
328 | u32 err_id, u64 err_addr, u64 err_count) | ||
329 | { | ||
330 | struct nvgpu_os_linux *l = nvgpu_os_linux_from_gk20a(g); | ||
331 | struct nvgpu_ecc_reporting_linux *ecc_report_linux = &l->ecc_reporting_linux; | ||
332 | struct nvgpu_ecc_reporting *error_reporting = &ecc_report_linux->common; | ||
333 | void (*report_ecc_err_func)(struct gk20a *g, u32 hw_unit, u32 inst, | ||
334 | u32 err_id, u64 err_addr, u64 err_count); | ||
335 | |||
336 | nvgpu_spinlock_acquire(&ecc_report_linux->common.lock); | ||
337 | report_ecc_err_func = error_reporting->ops->report_ecc_err; | ||
338 | nvgpu_spinlock_release(&ecc_report_linux->common.lock); | ||
339 | |||
340 | report_ecc_err_func(g, hw_unit, inst, err_id, err_addr, err_count); | ||
341 | } | ||
diff --git a/include/os/linux/vm.c b/include/os/linux/vm.c index dc807ab..8956cce 100644 --- a/include/os/linux/vm.c +++ b/include/os/linux/vm.c | |||
@@ -1,5 +1,5 @@ | |||
1 | /* | 1 | /* |
2 | * Copyright (c) 2017-2020, NVIDIA CORPORATION. All rights reserved. | 2 | * Copyright (c) 2017-2022, NVIDIA CORPORATION. All rights reserved. |
3 | * | 3 | * |
4 | * This program is free software; you can redistribute it and/or modify it | 4 | * This program is free software; you can redistribute it and/or modify it |
5 | * under the terms and conditions of the GNU General Public License, | 5 | * under the terms and conditions of the GNU General Public License, |
@@ -50,8 +50,10 @@ static u32 nvgpu_vm_translate_linux_flags(struct gk20a *g, u32 flags) | |||
50 | core_flags |= NVGPU_VM_MAP_IO_COHERENT; | 50 | core_flags |= NVGPU_VM_MAP_IO_COHERENT; |
51 | if (flags & NVGPU_AS_MAP_BUFFER_FLAGS_UNMAPPED_PTE) | 51 | if (flags & NVGPU_AS_MAP_BUFFER_FLAGS_UNMAPPED_PTE) |
52 | core_flags |= NVGPU_VM_MAP_UNMAPPED_PTE; | 52 | core_flags |= NVGPU_VM_MAP_UNMAPPED_PTE; |
53 | if (flags & NVGPU_AS_MAP_BUFFER_FLAGS_L3_ALLOC) | 53 | if (!nvgpu_is_enabled(g, NVGPU_DISABLE_L3_SUPPORT)) { |
54 | core_flags |= NVGPU_VM_MAP_L3_ALLOC; | 54 | if (flags & NVGPU_AS_MAP_BUFFER_FLAGS_L3_ALLOC) |
55 | core_flags |= NVGPU_VM_MAP_L3_ALLOC; | ||
56 | } | ||
55 | if (flags & NVGPU_AS_MAP_BUFFER_FLAGS_DIRECT_KIND_CTRL) | 57 | if (flags & NVGPU_AS_MAP_BUFFER_FLAGS_DIRECT_KIND_CTRL) |
56 | core_flags |= NVGPU_VM_MAP_DIRECT_KIND_CTRL; | 58 | core_flags |= NVGPU_VM_MAP_DIRECT_KIND_CTRL; |
57 | if (flags & NVGPU_AS_MAP_BUFFER_FLAGS_PLATFORM_ATOMIC) | 59 | if (flags & NVGPU_AS_MAP_BUFFER_FLAGS_PLATFORM_ATOMIC) |