Linux Audio

Check our new training course

Loading...
v4.6
  1/*
  2 * Copyright 1993-2003 NVIDIA, Corporation
  3 * Copyright 2007-2009 Stuart Bennett
  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 BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
 19 * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF
 20 * OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 21 * SOFTWARE.
 22 */
 23
 24#include <drm/drmP.h>
 25
 26#include "nouveau_drm.h"
 27#include "nouveau_reg.h"
 28#include "hw.h"
 29
 30/****************************************************************************\
 31*                                                                            *
 32* The video arbitration routines calculate some "magic" numbers.  Fixes      *
 33* the snow seen when accessing the framebuffer without it.                   *
 34* It just works (I hope).                                                    *
 35*                                                                            *
 36\****************************************************************************/
 37
 38struct nv_fifo_info {
 39	int lwm;
 40	int burst;
 41};
 42
 43struct nv_sim_state {
 44	int pclk_khz;
 45	int mclk_khz;
 46	int nvclk_khz;
 47	int bpp;
 48	int mem_page_miss;
 49	int mem_latency;
 50	int memory_type;
 51	int memory_width;
 52	int two_heads;
 53};
 54
 55static void
 56nv04_calc_arb(struct nv_fifo_info *fifo, struct nv_sim_state *arb)
 57{
 58	int pagemiss, cas, width, bpp;
 59	int nvclks, mclks, pclks, crtpagemiss;
 60	int found, mclk_extra, mclk_loop, cbs, m1, p1;
 61	int mclk_freq, pclk_freq, nvclk_freq;
 62	int us_m, us_n, us_p, crtc_drain_rate;
 63	int cpm_us, us_crt, clwm;
 64
 65	pclk_freq = arb->pclk_khz;
 66	mclk_freq = arb->mclk_khz;
 67	nvclk_freq = arb->nvclk_khz;
 68	pagemiss = arb->mem_page_miss;
 69	cas = arb->mem_latency;
 70	width = arb->memory_width >> 6;
 71	bpp = arb->bpp;
 72	cbs = 128;
 73
 74	pclks = 2;
 75	nvclks = 10;
 76	mclks = 13 + cas;
 77	mclk_extra = 3;
 78	found = 0;
 79
 80	while (!found) {
 81		found = 1;
 82
 83		mclk_loop = mclks + mclk_extra;
 84		us_m = mclk_loop * 1000 * 1000 / mclk_freq;
 85		us_n = nvclks * 1000 * 1000 / nvclk_freq;
 86		us_p = nvclks * 1000 * 1000 / pclk_freq;
 87
 88		crtc_drain_rate = pclk_freq * bpp / 8;
 89		crtpagemiss = 2;
 90		crtpagemiss += 1;
 91		cpm_us = crtpagemiss * pagemiss * 1000 * 1000 / mclk_freq;
 92		us_crt = cpm_us + us_m + us_n + us_p;
 93		clwm = us_crt * crtc_drain_rate / (1000 * 1000);
 94		clwm++;
 95
 96		m1 = clwm + cbs - 512;
 97		p1 = m1 * pclk_freq / mclk_freq;
 98		p1 = p1 * bpp / 8;
 99		if ((p1 < m1 && m1 > 0) || clwm > 519) {
100			found = !mclk_extra;
101			mclk_extra--;
102		}
103		if (clwm < 384)
104			clwm = 384;
105
106		fifo->lwm = clwm;
107		fifo->burst = cbs;
108	}
109}
110
111static void
112nv10_calc_arb(struct nv_fifo_info *fifo, struct nv_sim_state *arb)
113{
114	int fill_rate, drain_rate;
115	int pclks, nvclks, mclks, xclks;
116	int pclk_freq, nvclk_freq, mclk_freq;
117	int fill_lat, extra_lat;
118	int max_burst_o, max_burst_l;
119	int fifo_len, min_lwm, max_lwm;
120	const int burst_lat = 80; /* Maximum allowable latency due
121				   * to the CRTC FIFO burst. (ns) */
122
123	pclk_freq = arb->pclk_khz;
124	nvclk_freq = arb->nvclk_khz;
125	mclk_freq = arb->mclk_khz;
126
127	fill_rate = mclk_freq * arb->memory_width / 8; /* kB/s */
128	drain_rate = pclk_freq * arb->bpp / 8; /* kB/s */
129
130	fifo_len = arb->two_heads ? 1536 : 1024; /* B */
131
132	/* Fixed FIFO refill latency. */
133
134	pclks = 4;	/* lwm detect. */
135
136	nvclks = 3	/* lwm -> sync. */
137		+ 2	/* fbi bus cycles (1 req + 1 busy) */
138		+ 1	/* 2 edge sync.  may be very close to edge so
139			 * just put one. */
140		+ 1	/* fbi_d_rdv_n */
141		+ 1	/* Fbi_d_rdata */
142		+ 1;	/* crtfifo load */
143
144	mclks = 1	/* 2 edge sync.  may be very close to edge so
145			 * just put one. */
146		+ 1	/* arb_hp_req */
147		+ 5	/* tiling pipeline */
148		+ 2	/* latency fifo */
149		+ 2	/* memory request to fbio block */
150		+ 7;	/* data returned from fbio block */
151
152	/* Need to accumulate 256 bits for read */
153	mclks += (arb->memory_type == 0 ? 2 : 1)
154		* arb->memory_width / 32;
155
156	fill_lat = mclks * 1000 * 1000 / mclk_freq   /* minimum mclk latency */
157		+ nvclks * 1000 * 1000 / nvclk_freq  /* nvclk latency */
158		+ pclks * 1000 * 1000 / pclk_freq;   /* pclk latency */
159
160	/* Conditional FIFO refill latency. */
161
162	xclks = 2 * arb->mem_page_miss + mclks /* Extra latency due to
163						* the overlay. */
164		+ 2 * arb->mem_page_miss       /* Extra pagemiss latency. */
165		+ (arb->bpp == 32 ? 8 : 4);    /* Margin of error. */
166
167	extra_lat = xclks * 1000 * 1000 / mclk_freq;
168
169	if (arb->two_heads)
170		/* Account for another CRTC. */
171		extra_lat += fill_lat + extra_lat + burst_lat;
172
173	/* FIFO burst */
174
175	/* Max burst not leading to overflows. */
176	max_burst_o = (1 + fifo_len - extra_lat * drain_rate / (1000 * 1000))
177		* (fill_rate / 1000) / ((fill_rate - drain_rate) / 1000);
178	fifo->burst = min(max_burst_o, 1024);
179
180	/* Max burst value with an acceptable latency. */
181	max_burst_l = burst_lat * fill_rate / (1000 * 1000);
182	fifo->burst = min(max_burst_l, fifo->burst);
183
184	fifo->burst = rounddown_pow_of_two(fifo->burst);
185
186	/* FIFO low watermark */
187
188	min_lwm = (fill_lat + extra_lat) * drain_rate / (1000 * 1000) + 1;
189	max_lwm = fifo_len - fifo->burst
190		+ fill_lat * drain_rate / (1000 * 1000)
191		+ fifo->burst * drain_rate / fill_rate;
192
193	fifo->lwm = min_lwm + 10 * (max_lwm - min_lwm) / 100; /* Empirical. */
194}
195
196static void
197nv04_update_arb(struct drm_device *dev, int VClk, int bpp,
198		int *burst, int *lwm)
199{
200	struct nouveau_drm *drm = nouveau_drm(dev);
201	struct nvif_object *device = &nouveau_drm(dev)->device.object;
202	struct nv_fifo_info fifo_data;
203	struct nv_sim_state sim_data;
204	int MClk = nouveau_hw_get_clock(dev, PLL_MEMORY);
205	int NVClk = nouveau_hw_get_clock(dev, PLL_CORE);
206	uint32_t cfg1 = nvif_rd32(device, NV04_PFB_CFG1);
207
208	sim_data.pclk_khz = VClk;
209	sim_data.mclk_khz = MClk;
210	sim_data.nvclk_khz = NVClk;
211	sim_data.bpp = bpp;
212	sim_data.two_heads = nv_two_heads(dev);
213	if ((dev->pdev->device & 0xffff) == 0x01a0 /*CHIPSET_NFORCE*/ ||
214	    (dev->pdev->device & 0xffff) == 0x01f0 /*CHIPSET_NFORCE2*/) {
215		uint32_t type;
 
216
217		pci_read_config_dword(pci_get_bus_and_slot(0, 1), 0x7c, &type);
 
218
219		sim_data.memory_type = (type >> 12) & 1;
220		sim_data.memory_width = 64;
221		sim_data.mem_latency = 3;
222		sim_data.mem_page_miss = 10;
223	} else {
224		sim_data.memory_type = nvif_rd32(device, NV04_PFB_CFG0) & 0x1;
225		sim_data.memory_width = (nvif_rd32(device, NV_PEXTDEV_BOOT_0) & 0x10) ? 128 : 64;
226		sim_data.mem_latency = cfg1 & 0xf;
227		sim_data.mem_page_miss = ((cfg1 >> 4) & 0xf) + ((cfg1 >> 31) & 0x1);
228	}
229
230	if (drm->device.info.family == NV_DEVICE_INFO_V0_TNT)
231		nv04_calc_arb(&fifo_data, &sim_data);
232	else
233		nv10_calc_arb(&fifo_data, &sim_data);
234
235	*burst = ilog2(fifo_data.burst >> 4);
236	*lwm = fifo_data.lwm >> 3;
237}
238
239static void
240nv20_update_arb(int *burst, int *lwm)
241{
242	unsigned int fifo_size, burst_size, graphics_lwm;
243
244	fifo_size = 2048;
245	burst_size = 512;
246	graphics_lwm = fifo_size - burst_size;
247
248	*burst = ilog2(burst_size >> 5);
249	*lwm = graphics_lwm >> 3;
250}
251
252void
253nouveau_calc_arb(struct drm_device *dev, int vclk, int bpp, int *burst, int *lwm)
254{
255	struct nouveau_drm *drm = nouveau_drm(dev);
256
257	if (drm->device.info.family < NV_DEVICE_INFO_V0_KELVIN)
258		nv04_update_arb(dev, vclk, bpp, burst, lwm);
259	else if ((dev->pdev->device & 0xfff0) == 0x0240 /*CHIPSET_C51*/ ||
260		 (dev->pdev->device & 0xfff0) == 0x03d0 /*CHIPSET_C512*/) {
261		*burst = 128;
262		*lwm = 0x0480;
263	} else
264		nv20_update_arb(burst, lwm);
265}
v5.4
  1/*
  2 * Copyright 1993-2003 NVIDIA, Corporation
  3 * Copyright 2007-2009 Stuart Bennett
  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 BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
 19 * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF
 20 * OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 21 * SOFTWARE.
 22 */
 23
 24#include "nouveau_drv.h"
 
 
 25#include "nouveau_reg.h"
 26#include "hw.h"
 27
 28/****************************************************************************\
 29*                                                                            *
 30* The video arbitration routines calculate some "magic" numbers.  Fixes      *
 31* the snow seen when accessing the framebuffer without it.                   *
 32* It just works (I hope).                                                    *
 33*                                                                            *
 34\****************************************************************************/
 35
 36struct nv_fifo_info {
 37	int lwm;
 38	int burst;
 39};
 40
 41struct nv_sim_state {
 42	int pclk_khz;
 43	int mclk_khz;
 44	int nvclk_khz;
 45	int bpp;
 46	int mem_page_miss;
 47	int mem_latency;
 48	int memory_type;
 49	int memory_width;
 50	int two_heads;
 51};
 52
 53static void
 54nv04_calc_arb(struct nv_fifo_info *fifo, struct nv_sim_state *arb)
 55{
 56	int pagemiss, cas, width, bpp;
 57	int nvclks, mclks, pclks, crtpagemiss;
 58	int found, mclk_extra, mclk_loop, cbs, m1, p1;
 59	int mclk_freq, pclk_freq, nvclk_freq;
 60	int us_m, us_n, us_p, crtc_drain_rate;
 61	int cpm_us, us_crt, clwm;
 62
 63	pclk_freq = arb->pclk_khz;
 64	mclk_freq = arb->mclk_khz;
 65	nvclk_freq = arb->nvclk_khz;
 66	pagemiss = arb->mem_page_miss;
 67	cas = arb->mem_latency;
 68	width = arb->memory_width >> 6;
 69	bpp = arb->bpp;
 70	cbs = 128;
 71
 72	pclks = 2;
 73	nvclks = 10;
 74	mclks = 13 + cas;
 75	mclk_extra = 3;
 76	found = 0;
 77
 78	while (!found) {
 79		found = 1;
 80
 81		mclk_loop = mclks + mclk_extra;
 82		us_m = mclk_loop * 1000 * 1000 / mclk_freq;
 83		us_n = nvclks * 1000 * 1000 / nvclk_freq;
 84		us_p = nvclks * 1000 * 1000 / pclk_freq;
 85
 86		crtc_drain_rate = pclk_freq * bpp / 8;
 87		crtpagemiss = 2;
 88		crtpagemiss += 1;
 89		cpm_us = crtpagemiss * pagemiss * 1000 * 1000 / mclk_freq;
 90		us_crt = cpm_us + us_m + us_n + us_p;
 91		clwm = us_crt * crtc_drain_rate / (1000 * 1000);
 92		clwm++;
 93
 94		m1 = clwm + cbs - 512;
 95		p1 = m1 * pclk_freq / mclk_freq;
 96		p1 = p1 * bpp / 8;
 97		if ((p1 < m1 && m1 > 0) || clwm > 519) {
 98			found = !mclk_extra;
 99			mclk_extra--;
100		}
101		if (clwm < 384)
102			clwm = 384;
103
104		fifo->lwm = clwm;
105		fifo->burst = cbs;
106	}
107}
108
109static void
110nv10_calc_arb(struct nv_fifo_info *fifo, struct nv_sim_state *arb)
111{
112	int fill_rate, drain_rate;
113	int pclks, nvclks, mclks, xclks;
114	int pclk_freq, nvclk_freq, mclk_freq;
115	int fill_lat, extra_lat;
116	int max_burst_o, max_burst_l;
117	int fifo_len, min_lwm, max_lwm;
118	const int burst_lat = 80; /* Maximum allowable latency due
119				   * to the CRTC FIFO burst. (ns) */
120
121	pclk_freq = arb->pclk_khz;
122	nvclk_freq = arb->nvclk_khz;
123	mclk_freq = arb->mclk_khz;
124
125	fill_rate = mclk_freq * arb->memory_width / 8; /* kB/s */
126	drain_rate = pclk_freq * arb->bpp / 8; /* kB/s */
127
128	fifo_len = arb->two_heads ? 1536 : 1024; /* B */
129
130	/* Fixed FIFO refill latency. */
131
132	pclks = 4;	/* lwm detect. */
133
134	nvclks = 3	/* lwm -> sync. */
135		+ 2	/* fbi bus cycles (1 req + 1 busy) */
136		+ 1	/* 2 edge sync.  may be very close to edge so
137			 * just put one. */
138		+ 1	/* fbi_d_rdv_n */
139		+ 1	/* Fbi_d_rdata */
140		+ 1;	/* crtfifo load */
141
142	mclks = 1	/* 2 edge sync.  may be very close to edge so
143			 * just put one. */
144		+ 1	/* arb_hp_req */
145		+ 5	/* tiling pipeline */
146		+ 2	/* latency fifo */
147		+ 2	/* memory request to fbio block */
148		+ 7;	/* data returned from fbio block */
149
150	/* Need to accumulate 256 bits for read */
151	mclks += (arb->memory_type == 0 ? 2 : 1)
152		* arb->memory_width / 32;
153
154	fill_lat = mclks * 1000 * 1000 / mclk_freq   /* minimum mclk latency */
155		+ nvclks * 1000 * 1000 / nvclk_freq  /* nvclk latency */
156		+ pclks * 1000 * 1000 / pclk_freq;   /* pclk latency */
157
158	/* Conditional FIFO refill latency. */
159
160	xclks = 2 * arb->mem_page_miss + mclks /* Extra latency due to
161						* the overlay. */
162		+ 2 * arb->mem_page_miss       /* Extra pagemiss latency. */
163		+ (arb->bpp == 32 ? 8 : 4);    /* Margin of error. */
164
165	extra_lat = xclks * 1000 * 1000 / mclk_freq;
166
167	if (arb->two_heads)
168		/* Account for another CRTC. */
169		extra_lat += fill_lat + extra_lat + burst_lat;
170
171	/* FIFO burst */
172
173	/* Max burst not leading to overflows. */
174	max_burst_o = (1 + fifo_len - extra_lat * drain_rate / (1000 * 1000))
175		* (fill_rate / 1000) / ((fill_rate - drain_rate) / 1000);
176	fifo->burst = min(max_burst_o, 1024);
177
178	/* Max burst value with an acceptable latency. */
179	max_burst_l = burst_lat * fill_rate / (1000 * 1000);
180	fifo->burst = min(max_burst_l, fifo->burst);
181
182	fifo->burst = rounddown_pow_of_two(fifo->burst);
183
184	/* FIFO low watermark */
185
186	min_lwm = (fill_lat + extra_lat) * drain_rate / (1000 * 1000) + 1;
187	max_lwm = fifo_len - fifo->burst
188		+ fill_lat * drain_rate / (1000 * 1000)
189		+ fifo->burst * drain_rate / fill_rate;
190
191	fifo->lwm = min_lwm + 10 * (max_lwm - min_lwm) / 100; /* Empirical. */
192}
193
194static void
195nv04_update_arb(struct drm_device *dev, int VClk, int bpp,
196		int *burst, int *lwm)
197{
198	struct nouveau_drm *drm = nouveau_drm(dev);
199	struct nvif_object *device = &nouveau_drm(dev)->client.device.object;
200	struct nv_fifo_info fifo_data;
201	struct nv_sim_state sim_data;
202	int MClk = nouveau_hw_get_clock(dev, PLL_MEMORY);
203	int NVClk = nouveau_hw_get_clock(dev, PLL_CORE);
204	uint32_t cfg1 = nvif_rd32(device, NV04_PFB_CFG1);
205
206	sim_data.pclk_khz = VClk;
207	sim_data.mclk_khz = MClk;
208	sim_data.nvclk_khz = NVClk;
209	sim_data.bpp = bpp;
210	sim_data.two_heads = nv_two_heads(dev);
211	if ((dev->pdev->device & 0xffff) == 0x01a0 /*CHIPSET_NFORCE*/ ||
212	    (dev->pdev->device & 0xffff) == 0x01f0 /*CHIPSET_NFORCE2*/) {
213		uint32_t type;
214		int domain = pci_domain_nr(dev->pdev->bus);
215
216		pci_read_config_dword(pci_get_domain_bus_and_slot(domain, 0, 1),
217				      0x7c, &type);
218
219		sim_data.memory_type = (type >> 12) & 1;
220		sim_data.memory_width = 64;
221		sim_data.mem_latency = 3;
222		sim_data.mem_page_miss = 10;
223	} else {
224		sim_data.memory_type = nvif_rd32(device, NV04_PFB_CFG0) & 0x1;
225		sim_data.memory_width = (nvif_rd32(device, NV_PEXTDEV_BOOT_0) & 0x10) ? 128 : 64;
226		sim_data.mem_latency = cfg1 & 0xf;
227		sim_data.mem_page_miss = ((cfg1 >> 4) & 0xf) + ((cfg1 >> 31) & 0x1);
228	}
229
230	if (drm->client.device.info.family == NV_DEVICE_INFO_V0_TNT)
231		nv04_calc_arb(&fifo_data, &sim_data);
232	else
233		nv10_calc_arb(&fifo_data, &sim_data);
234
235	*burst = ilog2(fifo_data.burst >> 4);
236	*lwm = fifo_data.lwm >> 3;
237}
238
239static void
240nv20_update_arb(int *burst, int *lwm)
241{
242	unsigned int fifo_size, burst_size, graphics_lwm;
243
244	fifo_size = 2048;
245	burst_size = 512;
246	graphics_lwm = fifo_size - burst_size;
247
248	*burst = ilog2(burst_size >> 5);
249	*lwm = graphics_lwm >> 3;
250}
251
252void
253nouveau_calc_arb(struct drm_device *dev, int vclk, int bpp, int *burst, int *lwm)
254{
255	struct nouveau_drm *drm = nouveau_drm(dev);
256
257	if (drm->client.device.info.family < NV_DEVICE_INFO_V0_KELVIN)
258		nv04_update_arb(dev, vclk, bpp, burst, lwm);
259	else if ((dev->pdev->device & 0xfff0) == 0x0240 /*CHIPSET_C51*/ ||
260		 (dev->pdev->device & 0xfff0) == 0x03d0 /*CHIPSET_C512*/) {
261		*burst = 128;
262		*lwm = 0x0480;
263	} else
264		nv20_update_arb(burst, lwm);
265}