2 * Copyright 2008 Advanced Micro Devices, Inc.
3 * Copyright 2008 Red Hat Inc.
4 * Copyright 2009 Jerome Glisse.
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the "Software"),
8 * to deal in the Software without restriction, including without limitation
9 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10 * and/or sell copies of the Software, and to permit persons to whom the
11 * Software is furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
20 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
21 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
22 * OTHER DEALINGS IN THE SOFTWARE.
26 #include <linux/kthread.h>
27 #include <linux/pci.h>
28 #include <linux/uaccess.h>
29 #include <linux/pm_runtime.h>
32 #include "amdgpu_pm.h"
33 #include "amdgpu_dm_debugfs.h"
34 #include "amdgpu_ras.h"
35 #include "amdgpu_rap.h"
36 #include "amdgpu_securedisplay.h"
37 #include "amdgpu_fw_attestation.h"
38 #include "amdgpu_umr.h"
40 #include "amdgpu_reset.h"
41 #include "amdgpu_psp_ta.h"
43 #if defined(CONFIG_DEBUG_FS)
46 * amdgpu_debugfs_process_reg_op - Handle MMIO register reads/writes
48 * @read: True if reading
49 * @f: open file handle
50 * @buf: User buffer to write/read to
51 * @size: Number of bytes to write/read
52 * @pos: Offset to seek to
54 * This debugfs entry has special meaning on the offset being sought.
55 * Various bits have different meanings:
57 * Bit 62: Indicates a GRBM bank switch is needed
58 * Bit 61: Indicates a SRBM bank switch is needed (implies bit 62 is
60 * Bits 24..33: The SE or ME selector if needed
61 * Bits 34..43: The SH (or SA) or PIPE selector if needed
62 * Bits 44..53: The INSTANCE (or CU/WGP) or QUEUE selector if needed
64 * Bit 23: Indicates that the PM power gating lock should be held
65 * This is necessary to read registers that might be
66 * unreliable during a power gating transistion.
68 * The lower bits are the BYTE offset of the register to read. This
69 * allows reading multiple registers in a single call and having
70 * the returned size reflect that.
72 static int amdgpu_debugfs_process_reg_op(bool read, struct file *f,
73 char __user *buf, size_t size, loff_t *pos)
75 struct amdgpu_device *adev = file_inode(f)->i_private;
78 bool pm_pg_lock, use_bank, use_ring;
79 unsigned instance_bank, sh_bank, se_bank, me, pipe, queue, vmid;
81 pm_pg_lock = use_bank = use_ring = false;
82 instance_bank = sh_bank = se_bank = me = pipe = queue = vmid = 0;
84 if (size & 0x3 || *pos & 0x3 ||
85 ((*pos & (1ULL << 62)) && (*pos & (1ULL << 61))))
88 /* are we reading registers for which a PG lock is necessary? */
89 pm_pg_lock = (*pos >> 23) & 1;
91 if (*pos & (1ULL << 62)) {
92 se_bank = (*pos & GENMASK_ULL(33, 24)) >> 24;
93 sh_bank = (*pos & GENMASK_ULL(43, 34)) >> 34;
94 instance_bank = (*pos & GENMASK_ULL(53, 44)) >> 44;
100 if (instance_bank == 0x3FF)
101 instance_bank = 0xFFFFFFFF;
103 } else if (*pos & (1ULL << 61)) {
105 me = (*pos & GENMASK_ULL(33, 24)) >> 24;
106 pipe = (*pos & GENMASK_ULL(43, 34)) >> 34;
107 queue = (*pos & GENMASK_ULL(53, 44)) >> 44;
108 vmid = (*pos & GENMASK_ULL(58, 54)) >> 54;
112 use_bank = use_ring = false;
115 *pos &= (1UL << 22) - 1;
117 r = pm_runtime_get_sync(adev_to_drm(adev)->dev);
119 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
123 r = amdgpu_virt_enable_access_debugfs(adev);
125 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
130 if ((sh_bank != 0xFFFFFFFF && sh_bank >= adev->gfx.config.max_sh_per_se) ||
131 (se_bank != 0xFFFFFFFF && se_bank >= adev->gfx.config.max_shader_engines)) {
132 pm_runtime_mark_last_busy(adev_to_drm(adev)->dev);
133 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
134 amdgpu_virt_disable_access_debugfs(adev);
137 mutex_lock(&adev->grbm_idx_mutex);
138 amdgpu_gfx_select_se_sh(adev, se_bank,
139 sh_bank, instance_bank);
140 } else if (use_ring) {
141 mutex_lock(&adev->srbm_mutex);
142 amdgpu_gfx_select_me_pipe_q(adev, me, pipe, queue, vmid);
146 mutex_lock(&adev->pm.mutex);
152 value = RREG32(*pos >> 2);
153 r = put_user(value, (uint32_t *)buf);
155 r = get_user(value, (uint32_t *)buf);
157 amdgpu_mm_wreg_mmio_rlc(adev, *pos >> 2, value);
172 amdgpu_gfx_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
173 mutex_unlock(&adev->grbm_idx_mutex);
174 } else if (use_ring) {
175 amdgpu_gfx_select_me_pipe_q(adev, 0, 0, 0, 0);
176 mutex_unlock(&adev->srbm_mutex);
180 mutex_unlock(&adev->pm.mutex);
182 pm_runtime_mark_last_busy(adev_to_drm(adev)->dev);
183 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
185 amdgpu_virt_disable_access_debugfs(adev);
190 * amdgpu_debugfs_regs_read - Callback for reading MMIO registers
192 static ssize_t amdgpu_debugfs_regs_read(struct file *f, char __user *buf,
193 size_t size, loff_t *pos)
195 return amdgpu_debugfs_process_reg_op(true, f, buf, size, pos);
199 * amdgpu_debugfs_regs_write - Callback for writing MMIO registers
201 static ssize_t amdgpu_debugfs_regs_write(struct file *f, const char __user *buf,
202 size_t size, loff_t *pos)
204 return amdgpu_debugfs_process_reg_op(false, f, (char __user *)buf, size, pos);
207 static int amdgpu_debugfs_regs2_open(struct inode *inode, struct file *file)
209 struct amdgpu_debugfs_regs2_data *rd;
211 rd = kzalloc(sizeof *rd, GFP_KERNEL);
214 rd->adev = file_inode(file)->i_private;
215 file->private_data = rd;
216 mutex_init(&rd->lock);
221 static int amdgpu_debugfs_regs2_release(struct inode *inode, struct file *file)
223 struct amdgpu_debugfs_regs2_data *rd = file->private_data;
224 mutex_destroy(&rd->lock);
225 kfree(file->private_data);
229 static ssize_t amdgpu_debugfs_regs2_op(struct file *f, char __user *buf, u32 offset, size_t size, int write_en)
231 struct amdgpu_debugfs_regs2_data *rd = f->private_data;
232 struct amdgpu_device *adev = rd->adev;
237 if (size & 0x3 || offset & 0x3)
240 r = pm_runtime_get_sync(adev_to_drm(adev)->dev);
242 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
246 r = amdgpu_virt_enable_access_debugfs(adev);
248 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
252 mutex_lock(&rd->lock);
254 if (rd->id.use_grbm) {
255 if ((rd->id.grbm.sh != 0xFFFFFFFF && rd->id.grbm.sh >= adev->gfx.config.max_sh_per_se) ||
256 (rd->id.grbm.se != 0xFFFFFFFF && rd->id.grbm.se >= adev->gfx.config.max_shader_engines)) {
257 pm_runtime_mark_last_busy(adev_to_drm(adev)->dev);
258 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
259 amdgpu_virt_disable_access_debugfs(adev);
260 mutex_unlock(&rd->lock);
263 mutex_lock(&adev->grbm_idx_mutex);
264 amdgpu_gfx_select_se_sh(adev, rd->id.grbm.se,
266 rd->id.grbm.instance);
269 if (rd->id.use_srbm) {
270 mutex_lock(&adev->srbm_mutex);
271 amdgpu_gfx_select_me_pipe_q(adev, rd->id.srbm.me, rd->id.srbm.pipe,
272 rd->id.srbm.queue, rd->id.srbm.vmid);
276 mutex_lock(&adev->pm.mutex);
280 value = RREG32(offset >> 2);
281 r = put_user(value, (uint32_t *)buf);
283 r = get_user(value, (uint32_t *)buf);
285 amdgpu_mm_wreg_mmio_rlc(adev, offset >> 2, value);
297 if (rd->id.use_grbm) {
298 amdgpu_gfx_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
299 mutex_unlock(&adev->grbm_idx_mutex);
302 if (rd->id.use_srbm) {
303 amdgpu_gfx_select_me_pipe_q(adev, 0, 0, 0, 0);
304 mutex_unlock(&adev->srbm_mutex);
308 mutex_unlock(&adev->pm.mutex);
310 mutex_unlock(&rd->lock);
312 pm_runtime_mark_last_busy(adev_to_drm(adev)->dev);
313 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
315 amdgpu_virt_disable_access_debugfs(adev);
319 static long amdgpu_debugfs_regs2_ioctl(struct file *f, unsigned int cmd, unsigned long data)
321 struct amdgpu_debugfs_regs2_data *rd = f->private_data;
325 case AMDGPU_DEBUGFS_REGS2_IOC_SET_STATE:
326 mutex_lock(&rd->lock);
327 r = copy_from_user(&rd->id, (struct amdgpu_debugfs_regs2_iocdata *)data, sizeof rd->id);
328 mutex_unlock(&rd->lock);
329 return r ? -EINVAL : 0;
336 static ssize_t amdgpu_debugfs_regs2_read(struct file *f, char __user *buf, size_t size, loff_t *pos)
338 return amdgpu_debugfs_regs2_op(f, buf, *pos, size, 0);
341 static ssize_t amdgpu_debugfs_regs2_write(struct file *f, const char __user *buf, size_t size, loff_t *pos)
343 return amdgpu_debugfs_regs2_op(f, (char __user *)buf, *pos, size, 1);
348 * amdgpu_debugfs_regs_pcie_read - Read from a PCIE register
350 * @f: open file handle
351 * @buf: User buffer to store read data in
352 * @size: Number of bytes to read
353 * @pos: Offset to seek to
355 * The lower bits are the BYTE offset of the register to read. This
356 * allows reading multiple registers in a single call and having
357 * the returned size reflect that.
359 static ssize_t amdgpu_debugfs_regs_pcie_read(struct file *f, char __user *buf,
360 size_t size, loff_t *pos)
362 struct amdgpu_device *adev = file_inode(f)->i_private;
366 if (size & 0x3 || *pos & 0x3)
369 r = pm_runtime_get_sync(adev_to_drm(adev)->dev);
371 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
375 r = amdgpu_virt_enable_access_debugfs(adev);
377 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
384 value = RREG32_PCIE(*pos);
385 r = put_user(value, (uint32_t *)buf);
397 pm_runtime_mark_last_busy(adev_to_drm(adev)->dev);
398 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
399 amdgpu_virt_disable_access_debugfs(adev);
404 * amdgpu_debugfs_regs_pcie_write - Write to a PCIE register
406 * @f: open file handle
407 * @buf: User buffer to write data from
408 * @size: Number of bytes to write
409 * @pos: Offset to seek to
411 * The lower bits are the BYTE offset of the register to write. This
412 * allows writing multiple registers in a single call and having
413 * the returned size reflect that.
415 static ssize_t amdgpu_debugfs_regs_pcie_write(struct file *f, const char __user *buf,
416 size_t size, loff_t *pos)
418 struct amdgpu_device *adev = file_inode(f)->i_private;
422 if (size & 0x3 || *pos & 0x3)
425 r = pm_runtime_get_sync(adev_to_drm(adev)->dev);
427 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
431 r = amdgpu_virt_enable_access_debugfs(adev);
433 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
440 r = get_user(value, (uint32_t *)buf);
444 WREG32_PCIE(*pos, value);
454 pm_runtime_mark_last_busy(adev_to_drm(adev)->dev);
455 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
456 amdgpu_virt_disable_access_debugfs(adev);
461 * amdgpu_debugfs_regs_didt_read - Read from a DIDT register
463 * @f: open file handle
464 * @buf: User buffer to store read data in
465 * @size: Number of bytes to read
466 * @pos: Offset to seek to
468 * The lower bits are the BYTE offset of the register to read. This
469 * allows reading multiple registers in a single call and having
470 * the returned size reflect that.
472 static ssize_t amdgpu_debugfs_regs_didt_read(struct file *f, char __user *buf,
473 size_t size, loff_t *pos)
475 struct amdgpu_device *adev = file_inode(f)->i_private;
479 if (size & 0x3 || *pos & 0x3)
482 r = pm_runtime_get_sync(adev_to_drm(adev)->dev);
484 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
488 r = amdgpu_virt_enable_access_debugfs(adev);
490 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
497 value = RREG32_DIDT(*pos >> 2);
498 r = put_user(value, (uint32_t *)buf);
510 pm_runtime_mark_last_busy(adev_to_drm(adev)->dev);
511 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
512 amdgpu_virt_disable_access_debugfs(adev);
517 * amdgpu_debugfs_regs_didt_write - Write to a DIDT register
519 * @f: open file handle
520 * @buf: User buffer to write data from
521 * @size: Number of bytes to write
522 * @pos: Offset to seek to
524 * The lower bits are the BYTE offset of the register to write. This
525 * allows writing multiple registers in a single call and having
526 * the returned size reflect that.
528 static ssize_t amdgpu_debugfs_regs_didt_write(struct file *f, const char __user *buf,
529 size_t size, loff_t *pos)
531 struct amdgpu_device *adev = file_inode(f)->i_private;
535 if (size & 0x3 || *pos & 0x3)
538 r = pm_runtime_get_sync(adev_to_drm(adev)->dev);
540 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
544 r = amdgpu_virt_enable_access_debugfs(adev);
546 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
553 r = get_user(value, (uint32_t *)buf);
557 WREG32_DIDT(*pos >> 2, value);
567 pm_runtime_mark_last_busy(adev_to_drm(adev)->dev);
568 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
569 amdgpu_virt_disable_access_debugfs(adev);
574 * amdgpu_debugfs_regs_smc_read - Read from a SMC register
576 * @f: open file handle
577 * @buf: User buffer to store read data in
578 * @size: Number of bytes to read
579 * @pos: Offset to seek to
581 * The lower bits are the BYTE offset of the register to read. This
582 * allows reading multiple registers in a single call and having
583 * the returned size reflect that.
585 static ssize_t amdgpu_debugfs_regs_smc_read(struct file *f, char __user *buf,
586 size_t size, loff_t *pos)
588 struct amdgpu_device *adev = file_inode(f)->i_private;
592 if (size & 0x3 || *pos & 0x3)
595 r = pm_runtime_get_sync(adev_to_drm(adev)->dev);
597 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
601 r = amdgpu_virt_enable_access_debugfs(adev);
603 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
610 value = RREG32_SMC(*pos);
611 r = put_user(value, (uint32_t *)buf);
623 pm_runtime_mark_last_busy(adev_to_drm(adev)->dev);
624 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
625 amdgpu_virt_disable_access_debugfs(adev);
630 * amdgpu_debugfs_regs_smc_write - Write to a SMC register
632 * @f: open file handle
633 * @buf: User buffer to write data from
634 * @size: Number of bytes to write
635 * @pos: Offset to seek to
637 * The lower bits are the BYTE offset of the register to write. This
638 * allows writing multiple registers in a single call and having
639 * the returned size reflect that.
641 static ssize_t amdgpu_debugfs_regs_smc_write(struct file *f, const char __user *buf,
642 size_t size, loff_t *pos)
644 struct amdgpu_device *adev = file_inode(f)->i_private;
648 if (size & 0x3 || *pos & 0x3)
651 r = pm_runtime_get_sync(adev_to_drm(adev)->dev);
653 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
657 r = amdgpu_virt_enable_access_debugfs(adev);
659 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
666 r = get_user(value, (uint32_t *)buf);
670 WREG32_SMC(*pos, value);
680 pm_runtime_mark_last_busy(adev_to_drm(adev)->dev);
681 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
682 amdgpu_virt_disable_access_debugfs(adev);
687 * amdgpu_debugfs_gca_config_read - Read from gfx config data
689 * @f: open file handle
690 * @buf: User buffer to store read data in
691 * @size: Number of bytes to read
692 * @pos: Offset to seek to
694 * This file is used to access configuration data in a somewhat
695 * stable fashion. The format is a series of DWORDs with the first
696 * indicating which revision it is. New content is appended to the
697 * end so that older software can still read the data.
700 static ssize_t amdgpu_debugfs_gca_config_read(struct file *f, char __user *buf,
701 size_t size, loff_t *pos)
703 struct amdgpu_device *adev = file_inode(f)->i_private;
706 uint32_t *config, no_regs = 0;
708 if (size & 0x3 || *pos & 0x3)
711 config = kmalloc_array(256, sizeof(*config), GFP_KERNEL);
715 /* version, increment each time something is added */
716 config[no_regs++] = 5;
717 config[no_regs++] = adev->gfx.config.max_shader_engines;
718 config[no_regs++] = adev->gfx.config.max_tile_pipes;
719 config[no_regs++] = adev->gfx.config.max_cu_per_sh;
720 config[no_regs++] = adev->gfx.config.max_sh_per_se;
721 config[no_regs++] = adev->gfx.config.max_backends_per_se;
722 config[no_regs++] = adev->gfx.config.max_texture_channel_caches;
723 config[no_regs++] = adev->gfx.config.max_gprs;
724 config[no_regs++] = adev->gfx.config.max_gs_threads;
725 config[no_regs++] = adev->gfx.config.max_hw_contexts;
726 config[no_regs++] = adev->gfx.config.sc_prim_fifo_size_frontend;
727 config[no_regs++] = adev->gfx.config.sc_prim_fifo_size_backend;
728 config[no_regs++] = adev->gfx.config.sc_hiz_tile_fifo_size;
729 config[no_regs++] = adev->gfx.config.sc_earlyz_tile_fifo_size;
730 config[no_regs++] = adev->gfx.config.num_tile_pipes;
731 config[no_regs++] = adev->gfx.config.backend_enable_mask;
732 config[no_regs++] = adev->gfx.config.mem_max_burst_length_bytes;
733 config[no_regs++] = adev->gfx.config.mem_row_size_in_kb;
734 config[no_regs++] = adev->gfx.config.shader_engine_tile_size;
735 config[no_regs++] = adev->gfx.config.num_gpus;
736 config[no_regs++] = adev->gfx.config.multi_gpu_tile_size;
737 config[no_regs++] = adev->gfx.config.mc_arb_ramcfg;
738 config[no_regs++] = adev->gfx.config.gb_addr_config;
739 config[no_regs++] = adev->gfx.config.num_rbs;
742 config[no_regs++] = adev->rev_id;
743 config[no_regs++] = lower_32_bits(adev->pg_flags);
744 config[no_regs++] = lower_32_bits(adev->cg_flags);
747 config[no_regs++] = adev->family;
748 config[no_regs++] = adev->external_rev_id;
751 config[no_regs++] = adev->pdev->device;
752 config[no_regs++] = adev->pdev->revision;
753 config[no_regs++] = adev->pdev->subsystem_device;
754 config[no_regs++] = adev->pdev->subsystem_vendor;
756 /* rev==4 APU flag */
757 config[no_regs++] = adev->flags & AMD_IS_APU ? 1 : 0;
759 /* rev==5 PG/CG flag upper 32bit */
760 config[no_regs++] = upper_32_bits(adev->pg_flags);
761 config[no_regs++] = upper_32_bits(adev->cg_flags);
763 while (size && (*pos < no_regs * 4)) {
766 value = config[*pos >> 2];
767 r = put_user(value, (uint32_t *)buf);
784 * amdgpu_debugfs_sensor_read - Read from the powerplay sensors
786 * @f: open file handle
787 * @buf: User buffer to store read data in
788 * @size: Number of bytes to read
789 * @pos: Offset to seek to
791 * The offset is treated as the BYTE address of one of the sensors
792 * enumerated in amd/include/kgd_pp_interface.h under the
793 * 'amd_pp_sensors' enumeration. For instance to read the UVD VCLK
794 * you would use the offset 3 * 4 = 12.
796 static ssize_t amdgpu_debugfs_sensor_read(struct file *f, char __user *buf,
797 size_t size, loff_t *pos)
799 struct amdgpu_device *adev = file_inode(f)->i_private;
800 int idx, x, outsize, r, valuesize;
803 if (size & 3 || *pos & 0x3)
806 if (!adev->pm.dpm_enabled)
809 /* convert offset to sensor number */
812 valuesize = sizeof(values);
814 r = pm_runtime_get_sync(adev_to_drm(adev)->dev);
816 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
820 r = amdgpu_virt_enable_access_debugfs(adev);
822 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
826 r = amdgpu_dpm_read_sensor(adev, idx, &values[0], &valuesize);
828 pm_runtime_mark_last_busy(adev_to_drm(adev)->dev);
829 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
832 amdgpu_virt_disable_access_debugfs(adev);
836 if (size > valuesize) {
837 amdgpu_virt_disable_access_debugfs(adev);
845 r = put_user(values[x++], (int32_t *)buf);
852 amdgpu_virt_disable_access_debugfs(adev);
853 return !r ? outsize : r;
856 /** amdgpu_debugfs_wave_read - Read WAVE STATUS data
858 * @f: open file handle
859 * @buf: User buffer to store read data in
860 * @size: Number of bytes to read
861 * @pos: Offset to seek to
863 * The offset being sought changes which wave that the status data
864 * will be returned for. The bits are used as follows:
866 * Bits 0..6: Byte offset into data
867 * Bits 7..14: SE selector
868 * Bits 15..22: SH/SA selector
869 * Bits 23..30: CU/{WGP+SIMD} selector
870 * Bits 31..36: WAVE ID selector
871 * Bits 37..44: SIMD ID selector
873 * The returned data begins with one DWORD of version information
874 * Followed by WAVE STATUS registers relevant to the GFX IP version
875 * being used. See gfx_v8_0_read_wave_data() for an example output.
877 static ssize_t amdgpu_debugfs_wave_read(struct file *f, char __user *buf,
878 size_t size, loff_t *pos)
880 struct amdgpu_device *adev = f->f_inode->i_private;
883 uint32_t offset, se, sh, cu, wave, simd, data[32];
885 if (size & 3 || *pos & 3)
889 offset = (*pos & GENMASK_ULL(6, 0));
890 se = (*pos & GENMASK_ULL(14, 7)) >> 7;
891 sh = (*pos & GENMASK_ULL(22, 15)) >> 15;
892 cu = (*pos & GENMASK_ULL(30, 23)) >> 23;
893 wave = (*pos & GENMASK_ULL(36, 31)) >> 31;
894 simd = (*pos & GENMASK_ULL(44, 37)) >> 37;
896 r = pm_runtime_get_sync(adev_to_drm(adev)->dev);
898 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
902 r = amdgpu_virt_enable_access_debugfs(adev);
904 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
908 /* switch to the specific se/sh/cu */
909 mutex_lock(&adev->grbm_idx_mutex);
910 amdgpu_gfx_select_se_sh(adev, se, sh, cu);
913 if (adev->gfx.funcs->read_wave_data)
914 adev->gfx.funcs->read_wave_data(adev, simd, wave, data, &x);
916 amdgpu_gfx_select_se_sh(adev, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF);
917 mutex_unlock(&adev->grbm_idx_mutex);
919 pm_runtime_mark_last_busy(adev_to_drm(adev)->dev);
920 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
923 amdgpu_virt_disable_access_debugfs(adev);
927 while (size && (offset < x * 4)) {
930 value = data[offset >> 2];
931 r = put_user(value, (uint32_t *)buf);
933 amdgpu_virt_disable_access_debugfs(adev);
943 amdgpu_virt_disable_access_debugfs(adev);
947 /** amdgpu_debugfs_gpr_read - Read wave gprs
949 * @f: open file handle
950 * @buf: User buffer to store read data in
951 * @size: Number of bytes to read
952 * @pos: Offset to seek to
954 * The offset being sought changes which wave that the status data
955 * will be returned for. The bits are used as follows:
957 * Bits 0..11: Byte offset into data
958 * Bits 12..19: SE selector
959 * Bits 20..27: SH/SA selector
960 * Bits 28..35: CU/{WGP+SIMD} selector
961 * Bits 36..43: WAVE ID selector
962 * Bits 37..44: SIMD ID selector
963 * Bits 52..59: Thread selector
964 * Bits 60..61: Bank selector (VGPR=0,SGPR=1)
966 * The return data comes from the SGPR or VGPR register bank for
967 * the selected operational unit.
969 static ssize_t amdgpu_debugfs_gpr_read(struct file *f, char __user *buf,
970 size_t size, loff_t *pos)
972 struct amdgpu_device *adev = f->f_inode->i_private;
975 uint32_t offset, se, sh, cu, wave, simd, thread, bank, *data;
977 if (size > 4096 || size & 3 || *pos & 3)
981 offset = (*pos & GENMASK_ULL(11, 0)) >> 2;
982 se = (*pos & GENMASK_ULL(19, 12)) >> 12;
983 sh = (*pos & GENMASK_ULL(27, 20)) >> 20;
984 cu = (*pos & GENMASK_ULL(35, 28)) >> 28;
985 wave = (*pos & GENMASK_ULL(43, 36)) >> 36;
986 simd = (*pos & GENMASK_ULL(51, 44)) >> 44;
987 thread = (*pos & GENMASK_ULL(59, 52)) >> 52;
988 bank = (*pos & GENMASK_ULL(61, 60)) >> 60;
990 data = kcalloc(1024, sizeof(*data), GFP_KERNEL);
994 r = pm_runtime_get_sync(adev_to_drm(adev)->dev);
998 r = amdgpu_virt_enable_access_debugfs(adev);
1002 /* switch to the specific se/sh/cu */
1003 mutex_lock(&adev->grbm_idx_mutex);
1004 amdgpu_gfx_select_se_sh(adev, se, sh, cu);
1007 if (adev->gfx.funcs->read_wave_vgprs)
1008 adev->gfx.funcs->read_wave_vgprs(adev, simd, wave, thread, offset, size>>2, data);
1010 if (adev->gfx.funcs->read_wave_sgprs)
1011 adev->gfx.funcs->read_wave_sgprs(adev, simd, wave, offset, size>>2, data);
1014 amdgpu_gfx_select_se_sh(adev, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF);
1015 mutex_unlock(&adev->grbm_idx_mutex);
1017 pm_runtime_mark_last_busy(adev_to_drm(adev)->dev);
1018 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
1023 value = data[result >> 2];
1024 r = put_user(value, (uint32_t *)buf);
1026 amdgpu_virt_disable_access_debugfs(adev);
1036 amdgpu_virt_disable_access_debugfs(adev);
1040 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
1046 * amdgpu_debugfs_gfxoff_residency_read - Read GFXOFF residency
1048 * @f: open file handle
1049 * @buf: User buffer to store read data in
1050 * @size: Number of bytes to read
1051 * @pos: Offset to seek to
1053 * Read the last residency value logged. It doesn't auto update, one needs to
1054 * stop logging before getting the current value.
1056 static ssize_t amdgpu_debugfs_gfxoff_residency_read(struct file *f, char __user *buf,
1057 size_t size, loff_t *pos)
1059 struct amdgpu_device *adev = file_inode(f)->i_private;
1063 if (size & 0x3 || *pos & 0x3)
1066 r = pm_runtime_get_sync(adev_to_drm(adev)->dev);
1068 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
1075 r = amdgpu_get_gfx_off_residency(adev, &value);
1079 r = put_user(value, (uint32_t *)buf);
1091 pm_runtime_mark_last_busy(adev_to_drm(adev)->dev);
1092 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
1098 * amdgpu_debugfs_gfxoff_residency_write - Log GFXOFF Residency
1100 * @f: open file handle
1101 * @buf: User buffer to write data from
1102 * @size: Number of bytes to write
1103 * @pos: Offset to seek to
1105 * Write a 32-bit non-zero to start logging; write a 32-bit zero to stop
1107 static ssize_t amdgpu_debugfs_gfxoff_residency_write(struct file *f, const char __user *buf,
1108 size_t size, loff_t *pos)
1110 struct amdgpu_device *adev = file_inode(f)->i_private;
1114 if (size & 0x3 || *pos & 0x3)
1117 r = pm_runtime_get_sync(adev_to_drm(adev)->dev);
1119 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
1126 r = get_user(value, (uint32_t *)buf);
1130 amdgpu_set_gfx_off_residency(adev, value ? true : false);
1140 pm_runtime_mark_last_busy(adev_to_drm(adev)->dev);
1141 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
1148 * amdgpu_debugfs_gfxoff_count_read - Read GFXOFF entry count
1150 * @f: open file handle
1151 * @buf: User buffer to store read data in
1152 * @size: Number of bytes to read
1153 * @pos: Offset to seek to
1155 static ssize_t amdgpu_debugfs_gfxoff_count_read(struct file *f, char __user *buf,
1156 size_t size, loff_t *pos)
1158 struct amdgpu_device *adev = file_inode(f)->i_private;
1162 if (size & 0x3 || *pos & 0x3)
1165 r = pm_runtime_get_sync(adev_to_drm(adev)->dev);
1167 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
1174 r = amdgpu_get_gfx_off_entrycount(adev, &value);
1178 r = put_user(value, (u64 *)buf);
1190 pm_runtime_mark_last_busy(adev_to_drm(adev)->dev);
1191 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
1197 * amdgpu_debugfs_gfxoff_write - Enable/disable GFXOFF
1199 * @f: open file handle
1200 * @buf: User buffer to write data from
1201 * @size: Number of bytes to write
1202 * @pos: Offset to seek to
1204 * Write a 32-bit zero to disable or a 32-bit non-zero to enable
1206 static ssize_t amdgpu_debugfs_gfxoff_write(struct file *f, const char __user *buf,
1207 size_t size, loff_t *pos)
1209 struct amdgpu_device *adev = file_inode(f)->i_private;
1213 if (size & 0x3 || *pos & 0x3)
1216 r = pm_runtime_get_sync(adev_to_drm(adev)->dev);
1218 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
1225 r = get_user(value, (uint32_t *)buf);
1229 amdgpu_gfx_off_ctrl(adev, value ? true : false);
1239 pm_runtime_mark_last_busy(adev_to_drm(adev)->dev);
1240 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
1247 * amdgpu_debugfs_gfxoff_read - read gfxoff status
1249 * @f: open file handle
1250 * @buf: User buffer to store read data in
1251 * @size: Number of bytes to read
1252 * @pos: Offset to seek to
1254 static ssize_t amdgpu_debugfs_gfxoff_read(struct file *f, char __user *buf,
1255 size_t size, loff_t *pos)
1257 struct amdgpu_device *adev = file_inode(f)->i_private;
1261 if (size & 0x3 || *pos & 0x3)
1264 r = pm_runtime_get_sync(adev_to_drm(adev)->dev);
1266 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
1271 u32 value = adev->gfx.gfx_off_state;
1273 r = put_user(value, (u32 *)buf);
1285 pm_runtime_mark_last_busy(adev_to_drm(adev)->dev);
1286 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
1291 static ssize_t amdgpu_debugfs_gfxoff_status_read(struct file *f, char __user *buf,
1292 size_t size, loff_t *pos)
1294 struct amdgpu_device *adev = file_inode(f)->i_private;
1298 if (size & 0x3 || *pos & 0x3)
1301 r = pm_runtime_get_sync(adev_to_drm(adev)->dev);
1303 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
1310 r = amdgpu_get_gfx_off_status(adev, &value);
1314 r = put_user(value, (u32 *)buf);
1326 pm_runtime_mark_last_busy(adev_to_drm(adev)->dev);
1327 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
1332 static const struct file_operations amdgpu_debugfs_regs2_fops = {
1333 .owner = THIS_MODULE,
1334 .unlocked_ioctl = amdgpu_debugfs_regs2_ioctl,
1335 .read = amdgpu_debugfs_regs2_read,
1336 .write = amdgpu_debugfs_regs2_write,
1337 .open = amdgpu_debugfs_regs2_open,
1338 .release = amdgpu_debugfs_regs2_release,
1339 .llseek = default_llseek
1342 static const struct file_operations amdgpu_debugfs_regs_fops = {
1343 .owner = THIS_MODULE,
1344 .read = amdgpu_debugfs_regs_read,
1345 .write = amdgpu_debugfs_regs_write,
1346 .llseek = default_llseek
1348 static const struct file_operations amdgpu_debugfs_regs_didt_fops = {
1349 .owner = THIS_MODULE,
1350 .read = amdgpu_debugfs_regs_didt_read,
1351 .write = amdgpu_debugfs_regs_didt_write,
1352 .llseek = default_llseek
1354 static const struct file_operations amdgpu_debugfs_regs_pcie_fops = {
1355 .owner = THIS_MODULE,
1356 .read = amdgpu_debugfs_regs_pcie_read,
1357 .write = amdgpu_debugfs_regs_pcie_write,
1358 .llseek = default_llseek
1360 static const struct file_operations amdgpu_debugfs_regs_smc_fops = {
1361 .owner = THIS_MODULE,
1362 .read = amdgpu_debugfs_regs_smc_read,
1363 .write = amdgpu_debugfs_regs_smc_write,
1364 .llseek = default_llseek
1367 static const struct file_operations amdgpu_debugfs_gca_config_fops = {
1368 .owner = THIS_MODULE,
1369 .read = amdgpu_debugfs_gca_config_read,
1370 .llseek = default_llseek
1373 static const struct file_operations amdgpu_debugfs_sensors_fops = {
1374 .owner = THIS_MODULE,
1375 .read = amdgpu_debugfs_sensor_read,
1376 .llseek = default_llseek
1379 static const struct file_operations amdgpu_debugfs_wave_fops = {
1380 .owner = THIS_MODULE,
1381 .read = amdgpu_debugfs_wave_read,
1382 .llseek = default_llseek
1384 static const struct file_operations amdgpu_debugfs_gpr_fops = {
1385 .owner = THIS_MODULE,
1386 .read = amdgpu_debugfs_gpr_read,
1387 .llseek = default_llseek
1390 static const struct file_operations amdgpu_debugfs_gfxoff_fops = {
1391 .owner = THIS_MODULE,
1392 .read = amdgpu_debugfs_gfxoff_read,
1393 .write = amdgpu_debugfs_gfxoff_write,
1394 .llseek = default_llseek
1397 static const struct file_operations amdgpu_debugfs_gfxoff_status_fops = {
1398 .owner = THIS_MODULE,
1399 .read = amdgpu_debugfs_gfxoff_status_read,
1400 .llseek = default_llseek
1403 static const struct file_operations amdgpu_debugfs_gfxoff_count_fops = {
1404 .owner = THIS_MODULE,
1405 .read = amdgpu_debugfs_gfxoff_count_read,
1406 .llseek = default_llseek
1409 static const struct file_operations amdgpu_debugfs_gfxoff_residency_fops = {
1410 .owner = THIS_MODULE,
1411 .read = amdgpu_debugfs_gfxoff_residency_read,
1412 .write = amdgpu_debugfs_gfxoff_residency_write,
1413 .llseek = default_llseek
1416 static const struct file_operations *debugfs_regs[] = {
1417 &amdgpu_debugfs_regs_fops,
1418 &amdgpu_debugfs_regs2_fops,
1419 &amdgpu_debugfs_regs_didt_fops,
1420 &amdgpu_debugfs_regs_pcie_fops,
1421 &amdgpu_debugfs_regs_smc_fops,
1422 &amdgpu_debugfs_gca_config_fops,
1423 &amdgpu_debugfs_sensors_fops,
1424 &amdgpu_debugfs_wave_fops,
1425 &amdgpu_debugfs_gpr_fops,
1426 &amdgpu_debugfs_gfxoff_fops,
1427 &amdgpu_debugfs_gfxoff_status_fops,
1428 &amdgpu_debugfs_gfxoff_count_fops,
1429 &amdgpu_debugfs_gfxoff_residency_fops,
1432 static const char *debugfs_regs_names[] = {
1438 "amdgpu_gca_config",
1443 "amdgpu_gfxoff_status",
1444 "amdgpu_gfxoff_count",
1445 "amdgpu_gfxoff_residency",
1449 * amdgpu_debugfs_regs_init - Initialize debugfs entries that provide
1452 * @adev: The device to attach the debugfs entries to
1454 int amdgpu_debugfs_regs_init(struct amdgpu_device *adev)
1456 struct drm_minor *minor = adev_to_drm(adev)->primary;
1457 struct dentry *ent, *root = minor->debugfs_root;
1460 for (i = 0; i < ARRAY_SIZE(debugfs_regs); i++) {
1461 ent = debugfs_create_file(debugfs_regs_names[i],
1462 S_IFREG | S_IRUGO, root,
1463 adev, debugfs_regs[i]);
1464 if (!i && !IS_ERR_OR_NULL(ent))
1465 i_size_write(ent->d_inode, adev->rmmio_size);
1471 static int amdgpu_debugfs_test_ib_show(struct seq_file *m, void *unused)
1473 struct amdgpu_device *adev = (struct amdgpu_device *)m->private;
1474 struct drm_device *dev = adev_to_drm(adev);
1477 r = pm_runtime_get_sync(dev->dev);
1479 pm_runtime_put_autosuspend(dev->dev);
1483 /* Avoid accidently unparking the sched thread during GPU reset */
1484 r = down_write_killable(&adev->reset_domain->sem);
1488 /* hold on the scheduler */
1489 for (i = 0; i < AMDGPU_MAX_RINGS; i++) {
1490 struct amdgpu_ring *ring = adev->rings[i];
1492 if (!ring || !ring->sched.thread)
1494 kthread_park(ring->sched.thread);
1497 seq_printf(m, "run ib test:\n");
1498 r = amdgpu_ib_ring_tests(adev);
1500 seq_printf(m, "ib ring tests failed (%d).\n", r);
1502 seq_printf(m, "ib ring tests passed.\n");
1504 /* go on the scheduler */
1505 for (i = 0; i < AMDGPU_MAX_RINGS; i++) {
1506 struct amdgpu_ring *ring = adev->rings[i];
1508 if (!ring || !ring->sched.thread)
1510 kthread_unpark(ring->sched.thread);
1513 up_write(&adev->reset_domain->sem);
1515 pm_runtime_mark_last_busy(dev->dev);
1516 pm_runtime_put_autosuspend(dev->dev);
1521 static int amdgpu_debugfs_evict_vram(void *data, u64 *val)
1523 struct amdgpu_device *adev = (struct amdgpu_device *)data;
1524 struct drm_device *dev = adev_to_drm(adev);
1527 r = pm_runtime_get_sync(dev->dev);
1529 pm_runtime_put_autosuspend(dev->dev);
1533 *val = amdgpu_ttm_evict_resources(adev, TTM_PL_VRAM);
1535 pm_runtime_mark_last_busy(dev->dev);
1536 pm_runtime_put_autosuspend(dev->dev);
1542 static int amdgpu_debugfs_evict_gtt(void *data, u64 *val)
1544 struct amdgpu_device *adev = (struct amdgpu_device *)data;
1545 struct drm_device *dev = adev_to_drm(adev);
1548 r = pm_runtime_get_sync(dev->dev);
1550 pm_runtime_put_autosuspend(dev->dev);
1554 *val = amdgpu_ttm_evict_resources(adev, TTM_PL_TT);
1556 pm_runtime_mark_last_busy(dev->dev);
1557 pm_runtime_put_autosuspend(dev->dev);
1562 static int amdgpu_debugfs_benchmark(void *data, u64 val)
1564 struct amdgpu_device *adev = (struct amdgpu_device *)data;
1565 struct drm_device *dev = adev_to_drm(adev);
1568 r = pm_runtime_get_sync(dev->dev);
1570 pm_runtime_put_autosuspend(dev->dev);
1574 r = amdgpu_benchmark(adev, val);
1576 pm_runtime_mark_last_busy(dev->dev);
1577 pm_runtime_put_autosuspend(dev->dev);
1582 static int amdgpu_debugfs_vm_info_show(struct seq_file *m, void *unused)
1584 struct amdgpu_device *adev = (struct amdgpu_device *)m->private;
1585 struct drm_device *dev = adev_to_drm(adev);
1586 struct drm_file *file;
1589 r = mutex_lock_interruptible(&dev->filelist_mutex);
1593 list_for_each_entry(file, &dev->filelist, lhead) {
1594 struct amdgpu_fpriv *fpriv = file->driver_priv;
1595 struct amdgpu_vm *vm = &fpriv->vm;
1597 seq_printf(m, "pid:%d\tProcess:%s ----------\n",
1598 vm->task_info.pid, vm->task_info.process_name);
1599 r = amdgpu_bo_reserve(vm->root.bo, true);
1602 amdgpu_debugfs_vm_bo_info(vm, m);
1603 amdgpu_bo_unreserve(vm->root.bo);
1606 mutex_unlock(&dev->filelist_mutex);
1611 DEFINE_SHOW_ATTRIBUTE(amdgpu_debugfs_test_ib);
1612 DEFINE_SHOW_ATTRIBUTE(amdgpu_debugfs_vm_info);
1613 DEFINE_DEBUGFS_ATTRIBUTE(amdgpu_evict_vram_fops, amdgpu_debugfs_evict_vram,
1615 DEFINE_DEBUGFS_ATTRIBUTE(amdgpu_evict_gtt_fops, amdgpu_debugfs_evict_gtt,
1617 DEFINE_DEBUGFS_ATTRIBUTE(amdgpu_benchmark_fops, NULL, amdgpu_debugfs_benchmark,
1620 static void amdgpu_ib_preempt_fences_swap(struct amdgpu_ring *ring,
1621 struct dma_fence **fences)
1623 struct amdgpu_fence_driver *drv = &ring->fence_drv;
1624 uint32_t sync_seq, last_seq;
1626 last_seq = atomic_read(&ring->fence_drv.last_seq);
1627 sync_seq = ring->fence_drv.sync_seq;
1629 last_seq &= drv->num_fences_mask;
1630 sync_seq &= drv->num_fences_mask;
1633 struct dma_fence *fence, **ptr;
1636 last_seq &= drv->num_fences_mask;
1637 ptr = &drv->fences[last_seq];
1639 fence = rcu_dereference_protected(*ptr, 1);
1640 RCU_INIT_POINTER(*ptr, NULL);
1645 fences[last_seq] = fence;
1647 } while (last_seq != sync_seq);
1650 static void amdgpu_ib_preempt_signal_fences(struct dma_fence **fences,
1654 struct dma_fence *fence;
1656 for (i = 0; i < length; i++) {
1660 dma_fence_signal(fence);
1661 dma_fence_put(fence);
1665 static void amdgpu_ib_preempt_job_recovery(struct drm_gpu_scheduler *sched)
1667 struct drm_sched_job *s_job;
1668 struct dma_fence *fence;
1670 spin_lock(&sched->job_list_lock);
1671 list_for_each_entry(s_job, &sched->pending_list, list) {
1672 fence = sched->ops->run_job(s_job);
1673 dma_fence_put(fence);
1675 spin_unlock(&sched->job_list_lock);
1678 static void amdgpu_ib_preempt_mark_partial_job(struct amdgpu_ring *ring)
1680 struct amdgpu_job *job;
1681 struct drm_sched_job *s_job, *tmp;
1682 uint32_t preempt_seq;
1683 struct dma_fence *fence, **ptr;
1684 struct amdgpu_fence_driver *drv = &ring->fence_drv;
1685 struct drm_gpu_scheduler *sched = &ring->sched;
1686 bool preempted = true;
1688 if (ring->funcs->type != AMDGPU_RING_TYPE_GFX)
1691 preempt_seq = le32_to_cpu(*(drv->cpu_addr + 2));
1692 if (preempt_seq <= atomic_read(&drv->last_seq)) {
1697 preempt_seq &= drv->num_fences_mask;
1698 ptr = &drv->fences[preempt_seq];
1699 fence = rcu_dereference_protected(*ptr, 1);
1702 spin_lock(&sched->job_list_lock);
1703 list_for_each_entry_safe(s_job, tmp, &sched->pending_list, list) {
1704 if (dma_fence_is_signaled(&s_job->s_fence->finished)) {
1705 /* remove job from ring_mirror_list */
1706 list_del_init(&s_job->list);
1707 sched->ops->free_job(s_job);
1710 job = to_amdgpu_job(s_job);
1711 if (preempted && (&job->hw_fence) == fence)
1712 /* mark the job as preempted */
1713 job->preemption_status |= AMDGPU_IB_PREEMPTED;
1715 spin_unlock(&sched->job_list_lock);
1718 static int amdgpu_debugfs_ib_preempt(void *data, u64 val)
1721 struct amdgpu_ring *ring;
1722 struct dma_fence **fences = NULL;
1723 struct amdgpu_device *adev = (struct amdgpu_device *)data;
1725 if (val >= AMDGPU_MAX_RINGS)
1728 ring = adev->rings[val];
1730 if (!ring || !ring->funcs->preempt_ib || !ring->sched.thread)
1733 /* the last preemption failed */
1734 if (ring->trail_seq != le32_to_cpu(*ring->trail_fence_cpu_addr))
1737 length = ring->fence_drv.num_fences_mask + 1;
1738 fences = kcalloc(length, sizeof(void *), GFP_KERNEL);
1742 /* Avoid accidently unparking the sched thread during GPU reset */
1743 r = down_read_killable(&adev->reset_domain->sem);
1747 /* stop the scheduler */
1748 kthread_park(ring->sched.thread);
1750 /* preempt the IB */
1751 r = amdgpu_ring_preempt_ib(ring);
1753 DRM_WARN("failed to preempt ring %d\n", ring->idx);
1757 amdgpu_fence_process(ring);
1759 if (atomic_read(&ring->fence_drv.last_seq) !=
1760 ring->fence_drv.sync_seq) {
1761 DRM_INFO("ring %d was preempted\n", ring->idx);
1763 amdgpu_ib_preempt_mark_partial_job(ring);
1765 /* swap out the old fences */
1766 amdgpu_ib_preempt_fences_swap(ring, fences);
1768 amdgpu_fence_driver_force_completion(ring);
1770 /* resubmit unfinished jobs */
1771 amdgpu_ib_preempt_job_recovery(&ring->sched);
1773 /* wait for jobs finished */
1774 amdgpu_fence_wait_empty(ring);
1776 /* signal the old fences */
1777 amdgpu_ib_preempt_signal_fences(fences, length);
1781 /* restart the scheduler */
1782 kthread_unpark(ring->sched.thread);
1784 up_read(&adev->reset_domain->sem);
1792 static int amdgpu_debugfs_sclk_set(void *data, u64 val)
1795 uint32_t max_freq, min_freq;
1796 struct amdgpu_device *adev = (struct amdgpu_device *)data;
1798 if (amdgpu_sriov_vf(adev) && !amdgpu_sriov_is_pp_one_vf(adev))
1801 ret = pm_runtime_get_sync(adev_to_drm(adev)->dev);
1803 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
1807 ret = amdgpu_dpm_get_dpm_freq_range(adev, PP_SCLK, &min_freq, &max_freq);
1808 if (ret == -EOPNOTSUPP) {
1812 if (ret || val > max_freq || val < min_freq) {
1817 ret = amdgpu_dpm_set_soft_freq_range(adev, PP_SCLK, (uint32_t)val, (uint32_t)val);
1822 pm_runtime_mark_last_busy(adev_to_drm(adev)->dev);
1823 pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
1828 DEFINE_DEBUGFS_ATTRIBUTE(fops_ib_preempt, NULL,
1829 amdgpu_debugfs_ib_preempt, "%llu\n");
1831 DEFINE_DEBUGFS_ATTRIBUTE(fops_sclk_set, NULL,
1832 amdgpu_debugfs_sclk_set, "%llu\n");
1834 static ssize_t amdgpu_reset_dump_register_list_read(struct file *f,
1835 char __user *buf, size_t size, loff_t *pos)
1837 struct amdgpu_device *adev = (struct amdgpu_device *)file_inode(f)->i_private;
1838 char reg_offset[12];
1839 int i, ret, len = 0;
1844 memset(reg_offset, 0, 12);
1845 ret = down_read_killable(&adev->reset_domain->sem);
1849 for (i = 0; i < adev->num_regs; i++) {
1850 sprintf(reg_offset, "0x%x\n", adev->reset_dump_reg_list[i]);
1851 up_read(&adev->reset_domain->sem);
1852 if (copy_to_user(buf + len, reg_offset, strlen(reg_offset)))
1855 len += strlen(reg_offset);
1856 ret = down_read_killable(&adev->reset_domain->sem);
1861 up_read(&adev->reset_domain->sem);
1867 static ssize_t amdgpu_reset_dump_register_list_write(struct file *f,
1868 const char __user *buf, size_t size, loff_t *pos)
1870 struct amdgpu_device *adev = (struct amdgpu_device *)file_inode(f)->i_private;
1871 char reg_offset[11];
1872 uint32_t *new = NULL, *tmp = NULL;
1873 int ret, i = 0, len = 0;
1876 memset(reg_offset, 0, 11);
1877 if (copy_from_user(reg_offset, buf + len,
1878 min(10, ((int)size-len)))) {
1883 new = krealloc_array(tmp, i + 1, sizeof(uint32_t), GFP_KERNEL);
1889 if (sscanf(reg_offset, "%X %n", &tmp[i], &ret) != 1) {
1896 } while (len < size);
1898 new = kmalloc_array(i, sizeof(uint32_t), GFP_KERNEL);
1903 ret = down_write_killable(&adev->reset_domain->sem);
1907 swap(adev->reset_dump_reg_list, tmp);
1908 swap(adev->reset_dump_reg_value, new);
1910 up_write(&adev->reset_domain->sem);
1920 static const struct file_operations amdgpu_reset_dump_register_list = {
1921 .owner = THIS_MODULE,
1922 .read = amdgpu_reset_dump_register_list_read,
1923 .write = amdgpu_reset_dump_register_list_write,
1924 .llseek = default_llseek
1927 int amdgpu_debugfs_init(struct amdgpu_device *adev)
1929 struct dentry *root = adev_to_drm(adev)->primary->debugfs_root;
1933 if (!debugfs_initialized())
1936 debugfs_create_x32("amdgpu_smu_debug", 0600, root,
1937 &adev->pm.smu_debug_mask);
1939 ent = debugfs_create_file("amdgpu_preempt_ib", 0600, root, adev,
1942 DRM_ERROR("unable to create amdgpu_preempt_ib debugsfs file\n");
1943 return PTR_ERR(ent);
1946 ent = debugfs_create_file("amdgpu_force_sclk", 0200, root, adev,
1949 DRM_ERROR("unable to create amdgpu_set_sclk debugsfs file\n");
1950 return PTR_ERR(ent);
1953 /* Register debugfs entries for amdgpu_ttm */
1954 amdgpu_ttm_debugfs_init(adev);
1955 amdgpu_debugfs_pm_init(adev);
1956 amdgpu_debugfs_sa_init(adev);
1957 amdgpu_debugfs_fence_init(adev);
1958 amdgpu_debugfs_gem_init(adev);
1960 r = amdgpu_debugfs_regs_init(adev);
1962 DRM_ERROR("registering register debugfs failed (%d).\n", r);
1964 amdgpu_debugfs_firmware_init(adev);
1965 amdgpu_ta_if_debugfs_init(adev);
1967 #if defined(CONFIG_DRM_AMD_DC)
1968 if (adev->dc_enabled)
1969 dtn_debugfs_init(adev);
1972 for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
1973 struct amdgpu_ring *ring = adev->rings[i];
1978 amdgpu_debugfs_ring_init(adev, ring);
1981 for ( i = 0; i < adev->vcn.num_vcn_inst; i++) {
1982 if (!amdgpu_vcnfw_log)
1985 if (adev->vcn.harvest_config & (1 << i))
1988 amdgpu_debugfs_vcn_fwlog_init(adev, i, &adev->vcn.inst[i]);
1991 amdgpu_ras_debugfs_create_all(adev);
1992 amdgpu_rap_debugfs_init(adev);
1993 amdgpu_securedisplay_debugfs_init(adev);
1994 amdgpu_fw_attestation_debugfs_init(adev);
1996 debugfs_create_file("amdgpu_evict_vram", 0444, root, adev,
1997 &amdgpu_evict_vram_fops);
1998 debugfs_create_file("amdgpu_evict_gtt", 0444, root, adev,
1999 &amdgpu_evict_gtt_fops);
2000 debugfs_create_file("amdgpu_test_ib", 0444, root, adev,
2001 &amdgpu_debugfs_test_ib_fops);
2002 debugfs_create_file("amdgpu_vm_info", 0444, root, adev,
2003 &amdgpu_debugfs_vm_info_fops);
2004 debugfs_create_file("amdgpu_benchmark", 0200, root, adev,
2005 &amdgpu_benchmark_fops);
2006 debugfs_create_file("amdgpu_reset_dump_register_list", 0644, root, adev,
2007 &amdgpu_reset_dump_register_list);
2009 adev->debugfs_vbios_blob.data = adev->bios;
2010 adev->debugfs_vbios_blob.size = adev->bios_size;
2011 debugfs_create_blob("amdgpu_vbios", 0444, root,
2012 &adev->debugfs_vbios_blob);
2014 adev->debugfs_discovery_blob.data = adev->mman.discovery_bin;
2015 adev->debugfs_discovery_blob.size = adev->mman.discovery_tmr_size;
2016 debugfs_create_blob("amdgpu_discovery", 0444, root,
2017 &adev->debugfs_discovery_blob);
2023 int amdgpu_debugfs_init(struct amdgpu_device *adev)
2027 int amdgpu_debugfs_regs_init(struct amdgpu_device *adev)