diff options
Diffstat (limited to 'drivers/gpu')
-rw-r--r-- | drivers/gpu/nvgpu/vgpu/gk20a/vgpu_gr_gk20a.c | 30 | ||||
-rw-r--r-- | drivers/gpu/nvgpu/vgpu/gm20b/vgpu_gr_gm20b.c | 25 | ||||
-rw-r--r-- | drivers/gpu/nvgpu/vgpu/gr_vgpu.c | 48 |
3 files changed, 97 insertions, 6 deletions
diff --git a/drivers/gpu/nvgpu/vgpu/gk20a/vgpu_gr_gk20a.c b/drivers/gpu/nvgpu/vgpu/gk20a/vgpu_gr_gk20a.c index 0a2ca743..2cfe16da 100644 --- a/drivers/gpu/nvgpu/vgpu/gk20a/vgpu_gr_gk20a.c +++ b/drivers/gpu/nvgpu/vgpu/gk20a/vgpu_gr_gk20a.c | |||
@@ -1,5 +1,5 @@ | |||
1 | /* | 1 | /* |
2 | * Copyright (c) 2015, NVIDIA CORPORATION. All rights reserved. | 2 | * Copyright (c) 2015-2016, 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, |
@@ -45,7 +45,35 @@ static void vgpu_gk20a_detect_sm_arch(struct gk20a *g) | |||
45 | gr_gpc0_tpc0_sm_arch_warp_count_v(v); | 45 | gr_gpc0_tpc0_sm_arch_warp_count_v(v); |
46 | } | 46 | } |
47 | 47 | ||
48 | static int vgpu_gk20a_init_fs_state(struct gk20a *g) | ||
49 | { | ||
50 | struct gr_gk20a *gr = &g->gr; | ||
51 | u32 tpc_index, gpc_index; | ||
52 | u32 sm_id = 0; | ||
53 | |||
54 | gk20a_dbg_fn(""); | ||
55 | |||
56 | for (tpc_index = 0; tpc_index < gr->max_tpc_per_gpc_count; | ||
57 | tpc_index++) { | ||
58 | for (gpc_index = 0; gpc_index < gr->gpc_count; gpc_index++) { | ||
59 | if (tpc_index < gr->gpc_tpc_count[gpc_index]) { | ||
60 | g->gr.sm_to_cluster[sm_id].tpc_index = | ||
61 | tpc_index; | ||
62 | g->gr.sm_to_cluster[sm_id].gpc_index = | ||
63 | gpc_index; | ||
64 | |||
65 | sm_id++; | ||
66 | } | ||
67 | } | ||
68 | } | ||
69 | |||
70 | gr->no_of_sm = sm_id; | ||
71 | |||
72 | return 0; | ||
73 | } | ||
74 | |||
48 | void vgpu_gk20a_init_gr_ops(struct gpu_ops *gops) | 75 | void vgpu_gk20a_init_gr_ops(struct gpu_ops *gops) |
49 | { | 76 | { |
50 | gops->gr.detect_sm_arch = vgpu_gk20a_detect_sm_arch; | 77 | gops->gr.detect_sm_arch = vgpu_gk20a_detect_sm_arch; |
78 | gops->gr.init_fs_state = vgpu_gk20a_init_fs_state; | ||
51 | } | 79 | } |
diff --git a/drivers/gpu/nvgpu/vgpu/gm20b/vgpu_gr_gm20b.c b/drivers/gpu/nvgpu/vgpu/gm20b/vgpu_gr_gm20b.c index 07d41b2e..fb1f31d8 100644 --- a/drivers/gpu/nvgpu/vgpu/gm20b/vgpu_gr_gm20b.c +++ b/drivers/gpu/nvgpu/vgpu/gm20b/vgpu_gr_gm20b.c | |||
@@ -1,5 +1,5 @@ | |||
1 | /* | 1 | /* |
2 | * Copyright (c) 2015, NVIDIA CORPORATION. All rights reserved. | 2 | * Copyright (c) 2015-2016, 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, |
@@ -37,7 +37,30 @@ static void vgpu_gm20b_detect_sm_arch(struct gk20a *g) | |||
37 | gr_gpc0_tpc0_sm_arch_warp_count_v(v); | 37 | gr_gpc0_tpc0_sm_arch_warp_count_v(v); |
38 | } | 38 | } |
39 | 39 | ||
40 | static int vgpu_gm20b_init_fs_state(struct gk20a *g) | ||
41 | { | ||
42 | struct gr_gk20a *gr = &g->gr; | ||
43 | u32 tpc_index, gpc_index; | ||
44 | u32 sm_id = 0; | ||
45 | |||
46 | gk20a_dbg_fn(""); | ||
47 | |||
48 | for (gpc_index = 0; gpc_index < gr->gpc_count; gpc_index++) { | ||
49 | for (tpc_index = 0; tpc_index < gr->gpc_tpc_count[gpc_index]; | ||
50 | tpc_index++) { | ||
51 | g->gr.sm_to_cluster[sm_id].tpc_index = tpc_index; | ||
52 | g->gr.sm_to_cluster[sm_id].gpc_index = gpc_index; | ||
53 | |||
54 | sm_id++; | ||
55 | } | ||
56 | } | ||
57 | |||
58 | gr->no_of_sm = sm_id; | ||
59 | return 0; | ||
60 | } | ||
61 | |||
40 | void vgpu_gm20b_init_gr_ops(struct gpu_ops *gops) | 62 | void vgpu_gm20b_init_gr_ops(struct gpu_ops *gops) |
41 | { | 63 | { |
42 | gops->gr.detect_sm_arch = vgpu_gm20b_detect_sm_arch; | 64 | gops->gr.detect_sm_arch = vgpu_gm20b_detect_sm_arch; |
65 | gops->gr.init_fs_state = vgpu_gm20b_init_fs_state; | ||
43 | } | 66 | } |
diff --git a/drivers/gpu/nvgpu/vgpu/gr_vgpu.c b/drivers/gpu/nvgpu/vgpu/gr_vgpu.c index 16d51ad3..420c714e 100644 --- a/drivers/gpu/nvgpu/vgpu/gr_vgpu.c +++ b/drivers/gpu/nvgpu/vgpu/gr_vgpu.c | |||
@@ -567,6 +567,19 @@ static int vgpu_gr_free_obj_ctx(struct channel_gk20a *c, | |||
567 | return 0; | 567 | return 0; |
568 | } | 568 | } |
569 | 569 | ||
570 | static u32 vgpu_gr_get_gpc_tpc_count(struct gk20a *g, u32 gpc_index) | ||
571 | { | ||
572 | struct gk20a_platform *platform = gk20a_get_platform(g->dev); | ||
573 | u32 data; | ||
574 | |||
575 | WARN_ON(gpc_index > 0); | ||
576 | |||
577 | if (vgpu_get_attribute(platform->virt_handle, | ||
578 | TEGRA_VGPU_ATTRIB_GPC0_TPC_COUNT, &data)) | ||
579 | gk20a_err(dev_from_gk20a(g), "failed to retrieve gpc0_tpc_count"); | ||
580 | return data; | ||
581 | } | ||
582 | |||
570 | static int vgpu_gr_init_gr_config(struct gk20a *g, struct gr_gk20a *gr) | 583 | static int vgpu_gr_init_gr_config(struct gk20a *g, struct gr_gk20a *gr) |
571 | { | 584 | { |
572 | struct gk20a_platform *platform = gk20a_get_platform(g->dev); | 585 | struct gk20a_platform *platform = gk20a_get_platform(g->dev); |
@@ -593,13 +606,23 @@ static int vgpu_gr_init_gr_config(struct gk20a *g, struct gr_gk20a *gr) | |||
593 | &gr->tpc_count)) | 606 | &gr->tpc_count)) |
594 | return -ENOMEM; | 607 | return -ENOMEM; |
595 | 608 | ||
609 | gr->gpc_tpc_count = kzalloc(gr->gpc_count * sizeof(u32), GFP_KERNEL); | ||
610 | if (!gr->gpc_tpc_count) | ||
611 | goto cleanup; | ||
612 | |||
596 | gr->gpc_tpc_mask = kzalloc(gr->gpc_count * sizeof(u32), GFP_KERNEL); | 613 | gr->gpc_tpc_mask = kzalloc(gr->gpc_count * sizeof(u32), GFP_KERNEL); |
597 | if (!gr->gpc_tpc_mask) { | 614 | if (!gr->gpc_tpc_mask) |
598 | gk20a_err(dev_from_gk20a(g), "%s: out of memory\n", __func__); | 615 | goto cleanup; |
599 | return -ENOMEM; | 616 | |
600 | } | 617 | gr->sm_to_cluster = kzalloc(gr->gpc_count * gr->max_tpc_per_gpc_count * |
618 | sizeof(struct sm_info), GFP_KERNEL); | ||
619 | if (!gr->sm_to_cluster) | ||
620 | goto cleanup; | ||
601 | 621 | ||
602 | for (gpc_index = 0; gpc_index < gr->gpc_count; gpc_index++) { | 622 | for (gpc_index = 0; gpc_index < gr->gpc_count; gpc_index++) { |
623 | gr->gpc_tpc_count[gpc_index] = | ||
624 | vgpu_gr_get_gpc_tpc_count(g, gpc_index); | ||
625 | |||
603 | if (g->ops.gr.get_gpc_tpc_mask) | 626 | if (g->ops.gr.get_gpc_tpc_mask) |
604 | gr->gpc_tpc_mask[gpc_index] = | 627 | gr->gpc_tpc_mask[gpc_index] = |
605 | g->ops.gr.get_gpc_tpc_mask(g, gpc_index); | 628 | g->ops.gr.get_gpc_tpc_mask(g, gpc_index); |
@@ -608,7 +631,18 @@ static int vgpu_gr_init_gr_config(struct gk20a *g, struct gr_gk20a *gr) | |||
608 | g->ops.gr.bundle_cb_defaults(g); | 631 | g->ops.gr.bundle_cb_defaults(g); |
609 | g->ops.gr.cb_size_default(g); | 632 | g->ops.gr.cb_size_default(g); |
610 | g->ops.gr.calc_global_ctx_buffer_size(g); | 633 | g->ops.gr.calc_global_ctx_buffer_size(g); |
634 | g->ops.gr.init_fs_state(g); | ||
611 | return 0; | 635 | return 0; |
636 | cleanup: | ||
637 | gk20a_err(dev_from_gk20a(g), "%s: out of memory\n", __func__); | ||
638 | |||
639 | kfree(gr->gpc_tpc_count); | ||
640 | gr->gpc_tpc_count = NULL; | ||
641 | |||
642 | kfree(gr->gpc_tpc_mask); | ||
643 | gr->gpc_tpc_mask = NULL; | ||
644 | |||
645 | return -ENOMEM; | ||
612 | } | 646 | } |
613 | 647 | ||
614 | static int vgpu_gr_bind_ctxsw_zcull(struct gk20a *g, struct gr_gk20a *gr, | 648 | static int vgpu_gr_bind_ctxsw_zcull(struct gk20a *g, struct gr_gk20a *gr, |
@@ -823,6 +857,12 @@ static void vgpu_remove_gr_support(struct gr_gk20a *gr) | |||
823 | 857 | ||
824 | kfree(gr->gpc_tpc_mask); | 858 | kfree(gr->gpc_tpc_mask); |
825 | gr->gpc_tpc_mask = NULL; | 859 | gr->gpc_tpc_mask = NULL; |
860 | |||
861 | kfree(gr->sm_to_cluster); | ||
862 | gr->sm_to_cluster = NULL; | ||
863 | |||
864 | kfree(gr->gpc_tpc_count); | ||
865 | gr->gpc_tpc_count = NULL; | ||
826 | } | 866 | } |
827 | 867 | ||
828 | static int vgpu_gr_init_gr_setup_sw(struct gk20a *g) | 868 | static int vgpu_gr_init_gr_setup_sw(struct gk20a *g) |