1/*
2 * Copyright 2014 Advanced Micro Devices, Inc.
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
13 *
14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
17 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20 * OTHER DEALINGS IN THE SOFTWARE.
21 *
22 */
23#include "drmP.h"
24#include "amdgpu.h"
25#include "amdgpu_ih.h"
26#include "vid.h"
27
28#include "oss/oss_3_0_1_d.h"
29#include "oss/oss_3_0_1_sh_mask.h"
30
31#include "bif/bif_5_1_d.h"
32#include "bif/bif_5_1_sh_mask.h"
33
34/*
35 * Interrupts
36 * Starting with r6xx, interrupts are handled via a ring buffer.
37 * Ring buffers are areas of GPU accessible memory that the GPU
38 * writes interrupt vectors into and the host reads vectors out of.
39 * There is a rptr (read pointer) that determines where the
40 * host is currently reading, and a wptr (write pointer)
41 * which determines where the GPU has written.  When the
42 * pointers are equal, the ring is idle.  When the GPU
43 * writes vectors to the ring buffer, it increments the
44 * wptr.  When there is an interrupt, the host then starts
45 * fetching commands and processing them until the pointers are
46 * equal again at which point it updates the rptr.
47 */
48
49static void cz_ih_set_interrupt_funcs(struct amdgpu_device *adev);
50
51/**
52 * cz_ih_enable_interrupts - Enable the interrupt ring buffer
53 *
54 * @adev: amdgpu_device pointer
55 *
56 * Enable the interrupt ring buffer (VI).
57 */
58static void cz_ih_enable_interrupts(struct amdgpu_device *adev)
59{
60	u32 ih_cntl = RREG32(mmIH_CNTL);
61	u32 ih_rb_cntl = RREG32(mmIH_RB_CNTL);
62
63	ih_cntl = REG_SET_FIELD(ih_cntl, IH_CNTL, ENABLE_INTR, 1);
64	ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, RB_ENABLE, 1);
65	WREG32(mmIH_CNTL, ih_cntl);
66	WREG32(mmIH_RB_CNTL, ih_rb_cntl);
67	adev->irq.ih.enabled = true;
68}
69
70/**
71 * cz_ih_disable_interrupts - Disable the interrupt ring buffer
72 *
73 * @adev: amdgpu_device pointer
74 *
75 * Disable the interrupt ring buffer (VI).
76 */
77static void cz_ih_disable_interrupts(struct amdgpu_device *adev)
78{
79	u32 ih_rb_cntl = RREG32(mmIH_RB_CNTL);
80	u32 ih_cntl = RREG32(mmIH_CNTL);
81
82	ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, RB_ENABLE, 0);
83	ih_cntl = REG_SET_FIELD(ih_cntl, IH_CNTL, ENABLE_INTR, 0);
84	WREG32(mmIH_RB_CNTL, ih_rb_cntl);
85	WREG32(mmIH_CNTL, ih_cntl);
86	/* set rptr, wptr to 0 */
87	WREG32(mmIH_RB_RPTR, 0);
88	WREG32(mmIH_RB_WPTR, 0);
89	adev->irq.ih.enabled = false;
90	adev->irq.ih.rptr = 0;
91}
92
93/**
94 * cz_ih_irq_init - init and enable the interrupt ring
95 *
96 * @adev: amdgpu_device pointer
97 *
98 * Allocate a ring buffer for the interrupt controller,
99 * enable the RLC, disable interrupts, enable the IH
100 * ring buffer and enable it (VI).
101 * Called at device load and reume.
102 * Returns 0 for success, errors for failure.
103 */
104static int cz_ih_irq_init(struct amdgpu_device *adev)
105{
106	int ret = 0;
107	int rb_bufsz;
108	u32 interrupt_cntl, ih_cntl, ih_rb_cntl;
109	u64 wptr_off;
110
111	/* disable irqs */
112	cz_ih_disable_interrupts(adev);
113
114	/* setup interrupt control */
115	WREG32(mmINTERRUPT_CNTL2, adev->dummy_page.addr >> 8);
116	interrupt_cntl = RREG32(mmINTERRUPT_CNTL);
117	/* INTERRUPT_CNTL__IH_DUMMY_RD_OVERRIDE_MASK=0 - dummy read disabled with msi, enabled without msi
118	 * INTERRUPT_CNTL__IH_DUMMY_RD_OVERRIDE_MASK=1 - dummy read controlled by IH_DUMMY_RD_EN
119	 */
120	interrupt_cntl = REG_SET_FIELD(interrupt_cntl, INTERRUPT_CNTL, IH_DUMMY_RD_OVERRIDE, 0);
121	/* INTERRUPT_CNTL__IH_REQ_NONSNOOP_EN_MASK=1 if ring is in non-cacheable memory, e.g., vram */
122	interrupt_cntl = REG_SET_FIELD(interrupt_cntl, INTERRUPT_CNTL, IH_REQ_NONSNOOP_EN, 0);
123	WREG32(mmINTERRUPT_CNTL, interrupt_cntl);
124
125	/* Ring Buffer base. [39:8] of 40-bit address of the beginning of the ring buffer*/
126	WREG32(mmIH_RB_BASE, adev->irq.ih.gpu_addr >> 8);
127
128	rb_bufsz = order_base_2(adev->irq.ih.ring_size / 4);
129	ih_rb_cntl = REG_SET_FIELD(0, IH_RB_CNTL, WPTR_OVERFLOW_ENABLE, 1);
130	ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, WPTR_OVERFLOW_CLEAR, 1);
131	ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, RB_SIZE, rb_bufsz);
132
133	/* Ring Buffer write pointer writeback. If enabled, IH_RB_WPTR register value is written to memory */
134	ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, WPTR_WRITEBACK_ENABLE, 1);
135
136	/* set the writeback address whether it's enabled or not */
137	wptr_off = adev->wb.gpu_addr + (adev->irq.ih.wptr_offs * 4);
138	WREG32(mmIH_RB_WPTR_ADDR_LO, lower_32_bits(wptr_off));
139	WREG32(mmIH_RB_WPTR_ADDR_HI, upper_32_bits(wptr_off) & 0xFF);
140
141	WREG32(mmIH_RB_CNTL, ih_rb_cntl);
142
143	/* set rptr, wptr to 0 */
144	WREG32(mmIH_RB_RPTR, 0);
145	WREG32(mmIH_RB_WPTR, 0);
146
147	/* Default settings for IH_CNTL (disabled at first) */
148	ih_cntl = RREG32(mmIH_CNTL);
149	ih_cntl = REG_SET_FIELD(ih_cntl, IH_CNTL, MC_VMID, 0);
150
151	if (adev->irq.msi_enabled)
152		ih_cntl = REG_SET_FIELD(ih_cntl, IH_CNTL, RPTR_REARM, 1);
153	WREG32(mmIH_CNTL, ih_cntl);
154
155	pci_set_master(adev->pdev);
156
157	/* enable interrupts */
158	cz_ih_enable_interrupts(adev);
159
160	return ret;
161}
162
163/**
164 * cz_ih_irq_disable - disable interrupts
165 *
166 * @adev: amdgpu_device pointer
167 *
168 * Disable interrupts on the hw (VI).
169 */
170static void cz_ih_irq_disable(struct amdgpu_device *adev)
171{
172	cz_ih_disable_interrupts(adev);
173
174	/* Wait and acknowledge irq */
175	mdelay(1);
176}
177
178/**
179 * cz_ih_get_wptr - get the IH ring buffer wptr
180 *
181 * @adev: amdgpu_device pointer
182 *
183 * Get the IH ring buffer wptr from either the register
184 * or the writeback memory buffer (VI).  Also check for
185 * ring buffer overflow and deal with it.
186 * Used by cz_irq_process(VI).
187 * Returns the value of the wptr.
188 */
189static u32 cz_ih_get_wptr(struct amdgpu_device *adev)
190{
191	u32 wptr, tmp;
192
193	wptr = le32_to_cpu(adev->wb.wb[adev->irq.ih.wptr_offs]);
194
195	if (REG_GET_FIELD(wptr, IH_RB_WPTR, RB_OVERFLOW)) {
196		wptr = REG_SET_FIELD(wptr, IH_RB_WPTR, RB_OVERFLOW, 0);
197		/* When a ring buffer overflow happen start parsing interrupt
198		 * from the last not overwritten vector (wptr + 16). Hopefully
199		 * this should allow us to catchup.
200		 */
201		dev_warn(adev->dev, "IH ring buffer overflow (0x%08X, 0x%08X, 0x%08X)\n",
202			wptr, adev->irq.ih.rptr, (wptr + 16) & adev->irq.ih.ptr_mask);
203		adev->irq.ih.rptr = (wptr + 16) & adev->irq.ih.ptr_mask;
204		tmp = RREG32(mmIH_RB_CNTL);
205		tmp = REG_SET_FIELD(tmp, IH_RB_CNTL, WPTR_OVERFLOW_CLEAR, 1);
206		WREG32(mmIH_RB_CNTL, tmp);
207	}
208	return (wptr & adev->irq.ih.ptr_mask);
209}
210
211/**
212 * cz_ih_decode_iv - decode an interrupt vector
213 *
214 * @adev: amdgpu_device pointer
215 *
216 * Decodes the interrupt vector at the current rptr
217 * position and also advance the position.
218 */
219static void cz_ih_decode_iv(struct amdgpu_device *adev,
220				 struct amdgpu_iv_entry *entry)
221{
222	/* wptr/rptr are in bytes! */
223	u32 ring_index = adev->irq.ih.rptr >> 2;
224	uint32_t dw[4];
225
226	dw[0] = le32_to_cpu(adev->irq.ih.ring[ring_index + 0]);
227	dw[1] = le32_to_cpu(adev->irq.ih.ring[ring_index + 1]);
228	dw[2] = le32_to_cpu(adev->irq.ih.ring[ring_index + 2]);
229	dw[3] = le32_to_cpu(adev->irq.ih.ring[ring_index + 3]);
230
231	entry->src_id = dw[0] & 0xff;
232	entry->src_data = dw[1] & 0xfffffff;
233	entry->ring_id = dw[2] & 0xff;
234	entry->vm_id = (dw[2] >> 8) & 0xff;
235	entry->pas_id = (dw[2] >> 16) & 0xffff;
236
237	/* wptr/rptr are in bytes! */
238	adev->irq.ih.rptr += 16;
239}
240
241/**
242 * cz_ih_set_rptr - set the IH ring buffer rptr
243 *
244 * @adev: amdgpu_device pointer
245 *
246 * Set the IH ring buffer rptr.
247 */
248static void cz_ih_set_rptr(struct amdgpu_device *adev)
249{
250	WREG32(mmIH_RB_RPTR, adev->irq.ih.rptr);
251}
252
253static int cz_ih_early_init(void *handle)
254{
255	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
256
257	cz_ih_set_interrupt_funcs(adev);
258	return 0;
259}
260
261static int cz_ih_sw_init(void *handle)
262{
263	int r;
264	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
265
266	r = amdgpu_ih_ring_init(adev, 64 * 1024, false);
267	if (r)
268		return r;
269
270	r = amdgpu_irq_init(adev);
271
272	return r;
273}
274
275static int cz_ih_sw_fini(void *handle)
276{
277	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
278
279	amdgpu_irq_fini(adev);
280	amdgpu_ih_ring_fini(adev);
281
282	return 0;
283}
284
285static int cz_ih_hw_init(void *handle)
286{
287	int r;
288	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
289
290	r = cz_ih_irq_init(adev);
291	if (r)
292		return r;
293
294	return 0;
295}
296
297static int cz_ih_hw_fini(void *handle)
298{
299	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
300
301	cz_ih_irq_disable(adev);
302
303	return 0;
304}
305
306static int cz_ih_suspend(void *handle)
307{
308	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
309
310	return cz_ih_hw_fini(adev);
311}
312
313static int cz_ih_resume(void *handle)
314{
315	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
316
317	return cz_ih_hw_init(adev);
318}
319
320static bool cz_ih_is_idle(void *handle)
321{
322	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
323	u32 tmp = RREG32(mmSRBM_STATUS);
324
325	if (REG_GET_FIELD(tmp, SRBM_STATUS, IH_BUSY))
326		return false;
327
328	return true;
329}
330
331static int cz_ih_wait_for_idle(void *handle)
332{
333	unsigned i;
334	u32 tmp;
335	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
336
337	for (i = 0; i < adev->usec_timeout; i++) {
338		/* read MC_STATUS */
339		tmp = RREG32(mmSRBM_STATUS);
340		if (!REG_GET_FIELD(tmp, SRBM_STATUS, IH_BUSY))
341			return 0;
342		udelay(1);
343	}
344	return -ETIMEDOUT;
345}
346
347static void cz_ih_print_status(void *handle)
348{
349	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
350
351	dev_info(adev->dev, "CZ IH registers\n");
352	dev_info(adev->dev, "  SRBM_STATUS=0x%08X\n",
353		RREG32(mmSRBM_STATUS));
354	dev_info(adev->dev, "  SRBM_STATUS2=0x%08X\n",
355		RREG32(mmSRBM_STATUS2));
356	dev_info(adev->dev, "  INTERRUPT_CNTL=0x%08X\n",
357		 RREG32(mmINTERRUPT_CNTL));
358	dev_info(adev->dev, "  INTERRUPT_CNTL2=0x%08X\n",
359		 RREG32(mmINTERRUPT_CNTL2));
360	dev_info(adev->dev, "  IH_CNTL=0x%08X\n",
361		 RREG32(mmIH_CNTL));
362	dev_info(adev->dev, "  IH_RB_CNTL=0x%08X\n",
363		 RREG32(mmIH_RB_CNTL));
364	dev_info(adev->dev, "  IH_RB_BASE=0x%08X\n",
365		 RREG32(mmIH_RB_BASE));
366	dev_info(adev->dev, "  IH_RB_WPTR_ADDR_LO=0x%08X\n",
367		 RREG32(mmIH_RB_WPTR_ADDR_LO));
368	dev_info(adev->dev, "  IH_RB_WPTR_ADDR_HI=0x%08X\n",
369		 RREG32(mmIH_RB_WPTR_ADDR_HI));
370	dev_info(adev->dev, "  IH_RB_RPTR=0x%08X\n",
371		 RREG32(mmIH_RB_RPTR));
372	dev_info(adev->dev, "  IH_RB_WPTR=0x%08X\n",
373		 RREG32(mmIH_RB_WPTR));
374}
375
376static int cz_ih_soft_reset(void *handle)
377{
378	u32 srbm_soft_reset = 0;
379	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
380	u32 tmp = RREG32(mmSRBM_STATUS);
381
382	if (tmp & SRBM_STATUS__IH_BUSY_MASK)
383		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET,
384						SOFT_RESET_IH, 1);
385
386	if (srbm_soft_reset) {
387		cz_ih_print_status((void *)adev);
388
389		tmp = RREG32(mmSRBM_SOFT_RESET);
390		tmp |= srbm_soft_reset;
391		dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
392		WREG32(mmSRBM_SOFT_RESET, tmp);
393		tmp = RREG32(mmSRBM_SOFT_RESET);
394
395		udelay(50);
396
397		tmp &= ~srbm_soft_reset;
398		WREG32(mmSRBM_SOFT_RESET, tmp);
399		tmp = RREG32(mmSRBM_SOFT_RESET);
400
401		/* Wait a little for things to settle down */
402		udelay(50);
403
404		cz_ih_print_status((void *)adev);
405	}
406
407	return 0;
408}
409
410static int cz_ih_set_clockgating_state(void *handle,
411					  enum amd_clockgating_state state)
412{
413	// TODO
414	return 0;
415}
416
417static int cz_ih_set_powergating_state(void *handle,
418					  enum amd_powergating_state state)
419{
420	// TODO
421	return 0;
422}
423
424const struct amd_ip_funcs cz_ih_ip_funcs = {
425	.early_init = cz_ih_early_init,
426	.late_init = NULL,
427	.sw_init = cz_ih_sw_init,
428	.sw_fini = cz_ih_sw_fini,
429	.hw_init = cz_ih_hw_init,
430	.hw_fini = cz_ih_hw_fini,
431	.suspend = cz_ih_suspend,
432	.resume = cz_ih_resume,
433	.is_idle = cz_ih_is_idle,
434	.wait_for_idle = cz_ih_wait_for_idle,
435	.soft_reset = cz_ih_soft_reset,
436	.print_status = cz_ih_print_status,
437	.set_clockgating_state = cz_ih_set_clockgating_state,
438	.set_powergating_state = cz_ih_set_powergating_state,
439};
440
441static const struct amdgpu_ih_funcs cz_ih_funcs = {
442	.get_wptr = cz_ih_get_wptr,
443	.decode_iv = cz_ih_decode_iv,
444	.set_rptr = cz_ih_set_rptr
445};
446
447static void cz_ih_set_interrupt_funcs(struct amdgpu_device *adev)
448{
449	if (adev->irq.ih_funcs == NULL)
450		adev->irq.ih_funcs = &cz_ih_funcs;
451}
452
453