Linux Audio

Check our new training course

Loading...
v4.6
 
  1/*
  2 * Tegra host1x driver
  3 *
  4 * Copyright (c) 2010-2013, NVIDIA Corporation.
  5 *
  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,
  8 * version 2, as published by the Free Software Foundation.
  9 *
 10 * This program is distributed in the hope it will be useful, but WITHOUT
 11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 12 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
 13 * more details.
 14 *
 15 * You should have received a copy of the GNU General Public License
 16 * along with this program.  If not, see <http://www.gnu.org/licenses/>.
 17 */
 18
 19#include <linux/module.h>
 20#include <linux/list.h>
 21#include <linux/slab.h>
 22#include <linux/of.h>
 23#include <linux/of_device.h>
 24#include <linux/clk.h>
 25#include <linux/io.h>
 26#include <linux/dma-mapping.h>
 
 
 
 
 
 
 
 
 
 27
 28#define CREATE_TRACE_POINTS
 29#include <trace/events/host1x.h>
 
 
 
 
 
 30
 31#include "bus.h"
 32#include "dev.h"
 33#include "intr.h"
 34#include "channel.h"
 
 35#include "debug.h"
 
 
 
 36#include "hw/host1x01.h"
 37#include "hw/host1x02.h"
 38#include "hw/host1x04.h"
 39#include "hw/host1x05.h"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 40
 41void host1x_sync_writel(struct host1x *host1x, u32 v, u32 r)
 42{
 43	void __iomem *sync_regs = host1x->regs + host1x->info->sync_offset;
 44
 45	writel(v, sync_regs + r);
 46}
 47
 48u32 host1x_sync_readl(struct host1x *host1x, u32 r)
 49{
 50	void __iomem *sync_regs = host1x->regs + host1x->info->sync_offset;
 51
 52	return readl(sync_regs + r);
 53}
 54
 55void host1x_ch_writel(struct host1x_channel *ch, u32 v, u32 r)
 56{
 57	writel(v, ch->regs + r);
 58}
 59
 60u32 host1x_ch_readl(struct host1x_channel *ch, u32 r)
 61{
 62	return readl(ch->regs + r);
 63}
 64
 65static const struct host1x_info host1x01_info = {
 66	.nb_channels	= 8,
 67	.nb_pts		= 32,
 68	.nb_mlocks	= 16,
 69	.nb_bases	= 8,
 70	.init		= host1x01_init,
 71	.sync_offset	= 0x3000,
 72	.dma_mask	= DMA_BIT_MASK(32),
 
 
 
 
 
 73};
 74
 75static const struct host1x_info host1x02_info = {
 76	.nb_channels = 9,
 77	.nb_pts = 32,
 78	.nb_mlocks = 16,
 79	.nb_bases = 12,
 80	.init = host1x02_init,
 81	.sync_offset = 0x3000,
 82	.dma_mask = DMA_BIT_MASK(32),
 
 
 
 
 
 83};
 84
 85static const struct host1x_info host1x04_info = {
 86	.nb_channels = 12,
 87	.nb_pts = 192,
 88	.nb_mlocks = 16,
 89	.nb_bases = 64,
 90	.init = host1x04_init,
 91	.sync_offset = 0x2100,
 92	.dma_mask = DMA_BIT_MASK(34),
 
 
 
 
 
 93};
 94
 95static const struct host1x_info host1x05_info = {
 96	.nb_channels = 14,
 97	.nb_pts = 192,
 98	.nb_mlocks = 16,
 99	.nb_bases = 64,
100	.init = host1x05_init,
101	.sync_offset = 0x2100,
102	.dma_mask = DMA_BIT_MASK(34),
 
 
 
 
 
103};
104
105static struct of_device_id host1x_of_match[] = {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
106	{ .compatible = "nvidia,tegra210-host1x", .data = &host1x05_info, },
107	{ .compatible = "nvidia,tegra124-host1x", .data = &host1x04_info, },
108	{ .compatible = "nvidia,tegra114-host1x", .data = &host1x02_info, },
109	{ .compatible = "nvidia,tegra30-host1x", .data = &host1x01_info, },
110	{ .compatible = "nvidia,tegra20-host1x", .data = &host1x01_info, },
111	{ },
112};
113MODULE_DEVICE_TABLE(of, host1x_of_match);
114
115static int host1x_probe(struct platform_device *pdev)
116{
117	const struct of_device_id *id;
118	struct host1x *host;
119	struct resource *regs;
120	int syncpt_irq;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
121	int err;
122
123	id = of_match_device(host1x_of_match, &pdev->dev);
124	if (!id)
125		return -EINVAL;
126
127	regs = platform_get_resource(pdev, IORESOURCE_MEM, 0);
128	if (!regs) {
129		dev_err(&pdev->dev, "failed to get registers\n");
130		return -ENXIO;
131	}
 
132
133	syncpt_irq = platform_get_irq(pdev, 0);
134	if (syncpt_irq < 0) {
135		dev_err(&pdev->dev, "failed to get IRQ\n");
136		return -ENXIO;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
137	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
138
139	host = devm_kzalloc(&pdev->dev, sizeof(*host), GFP_KERNEL);
140	if (!host)
141		return -ENOMEM;
142
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
143	mutex_init(&host->devices_lock);
144	INIT_LIST_HEAD(&host->devices);
145	INIT_LIST_HEAD(&host->list);
146	host->dev = &pdev->dev;
147	host->info = id->data;
148
149	/* set common host1x device data */
150	platform_set_drvdata(pdev, host);
151
152	host->regs = devm_ioremap_resource(&pdev->dev, regs);
153	if (IS_ERR(host->regs))
154		return PTR_ERR(host->regs);
155
156	dma_set_mask_and_coherent(host->dev, host->info->dma_mask);
157
158	if (host->info->init) {
159		err = host->info->init(host);
160		if (err)
161			return err;
162	}
163
164	host->clk = devm_clk_get(&pdev->dev, NULL);
165	if (IS_ERR(host->clk)) {
166		dev_err(&pdev->dev, "failed to get clock\n");
167		err = PTR_ERR(host->clk);
 
 
 
 
168		return err;
169	}
170
171	err = host1x_channel_list_init(host);
 
 
 
 
 
 
 
 
 
 
 
 
 
172	if (err) {
173		dev_err(&pdev->dev, "failed to initialize channel list\n");
174		return err;
175	}
176
177	err = clk_prepare_enable(host->clk);
178	if (err < 0) {
179		dev_err(&pdev->dev, "failed to enable clock\n");
180		return err;
181	}
182
183	err = host1x_syncpt_init(host);
184	if (err) {
185		dev_err(&pdev->dev, "failed to initialize syncpts\n");
186		goto fail_unprepare_disable;
187	}
188
189	err = host1x_intr_init(host, syncpt_irq);
190	if (err) {
191		dev_err(&pdev->dev, "failed to initialize interrupts\n");
192		goto fail_deinit_syncpt;
193	}
194
 
 
 
 
 
 
 
 
 
 
 
195	host1x_debug_init(host);
196
197	err = host1x_register(host);
198	if (err < 0)
199		goto fail_deinit_intr;
 
 
 
 
200
201	return 0;
202
203fail_deinit_intr:
 
 
 
 
 
 
 
 
204	host1x_intr_deinit(host);
205fail_deinit_syncpt:
206	host1x_syncpt_deinit(host);
207fail_unprepare_disable:
208	clk_disable_unprepare(host->clk);
 
 
 
 
 
 
 
209	return err;
210}
211
212static int host1x_remove(struct platform_device *pdev)
213{
214	struct host1x *host = platform_get_drvdata(pdev);
215
216	host1x_unregister(host);
 
 
 
 
217	host1x_intr_deinit(host);
218	host1x_syncpt_deinit(host);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
219	clk_disable_unprepare(host->clk);
 
220
221	return 0;
 
 
 
 
 
 
 
222}
223
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
224static struct platform_driver tegra_host1x_driver = {
225	.driver = {
226		.name = "tegra-host1x",
227		.of_match_table = host1x_of_match,
 
228	},
229	.probe = host1x_probe,
230	.remove = host1x_remove,
231};
232
233static struct platform_driver * const drivers[] = {
234	&tegra_host1x_driver,
235	&tegra_mipi_driver,
236};
237
238static int __init tegra_host1x_init(void)
239{
240	int err;
241
242	err = bus_register(&host1x_bus_type);
243	if (err < 0)
244		return err;
245
246	err = platform_register_drivers(drivers, ARRAY_SIZE(drivers));
247	if (err < 0)
248		bus_unregister(&host1x_bus_type);
249
250	return err;
251}
252module_init(tegra_host1x_init);
253
254static void __exit tegra_host1x_exit(void)
255{
256	platform_unregister_drivers(drivers, ARRAY_SIZE(drivers));
257	bus_unregister(&host1x_bus_type);
258}
259module_exit(tegra_host1x_exit);
 
 
 
 
 
 
 
 
 
 
 
 
 
260
261MODULE_AUTHOR("Thierry Reding <thierry.reding@avionic-design.de>");
262MODULE_AUTHOR("Terje Bergstrom <tbergstrom@nvidia.com>");
263MODULE_DESCRIPTION("Host1x driver for Tegra products");
264MODULE_LICENSE("GPL");
v6.2
  1// SPDX-License-Identifier: GPL-2.0-only
  2/*
  3 * Tegra host1x driver
  4 *
  5 * Copyright (c) 2010-2013, NVIDIA Corporation.
 
 
 
 
 
 
 
 
 
 
 
 
  6 */
  7
 
 
 
 
 
  8#include <linux/clk.h>
  9#include <linux/delay.h>
 10#include <linux/dma-mapping.h>
 11#include <linux/io.h>
 12#include <linux/list.h>
 13#include <linux/module.h>
 14#include <linux/of_device.h>
 15#include <linux/of.h>
 16#include <linux/pm_runtime.h>
 17#include <linux/slab.h>
 18
 19#include <soc/tegra/common.h>
 20
 21#define CREATE_TRACE_POINTS
 22#include <trace/events/host1x.h>
 23#undef CREATE_TRACE_POINTS
 24
 25#if IS_ENABLED(CONFIG_ARM_DMA_USE_IOMMU)
 26#include <asm/dma-iommu.h>
 27#endif
 28
 29#include "bus.h"
 
 
 30#include "channel.h"
 31#include "context.h"
 32#include "debug.h"
 33#include "dev.h"
 34#include "intr.h"
 35
 36#include "hw/host1x01.h"
 37#include "hw/host1x02.h"
 38#include "hw/host1x04.h"
 39#include "hw/host1x05.h"
 40#include "hw/host1x06.h"
 41#include "hw/host1x07.h"
 42#include "hw/host1x08.h"
 43
 44void host1x_common_writel(struct host1x *host1x, u32 v, u32 r)
 45{
 46	writel(v, host1x->common_regs + r);
 47}
 48
 49void host1x_hypervisor_writel(struct host1x *host1x, u32 v, u32 r)
 50{
 51	writel(v, host1x->hv_regs + r);
 52}
 53
 54u32 host1x_hypervisor_readl(struct host1x *host1x, u32 r)
 55{
 56	return readl(host1x->hv_regs + r);
 57}
 58
 59void host1x_sync_writel(struct host1x *host1x, u32 v, u32 r)
 60{
 61	void __iomem *sync_regs = host1x->regs + host1x->info->sync_offset;
 62
 63	writel(v, sync_regs + r);
 64}
 65
 66u32 host1x_sync_readl(struct host1x *host1x, u32 r)
 67{
 68	void __iomem *sync_regs = host1x->regs + host1x->info->sync_offset;
 69
 70	return readl(sync_regs + r);
 71}
 72
 73void host1x_ch_writel(struct host1x_channel *ch, u32 v, u32 r)
 74{
 75	writel(v, ch->regs + r);
 76}
 77
 78u32 host1x_ch_readl(struct host1x_channel *ch, u32 r)
 79{
 80	return readl(ch->regs + r);
 81}
 82
 83static const struct host1x_info host1x01_info = {
 84	.nb_channels = 8,
 85	.nb_pts = 32,
 86	.nb_mlocks = 16,
 87	.nb_bases = 8,
 88	.init = host1x01_init,
 89	.sync_offset = 0x3000,
 90	.dma_mask = DMA_BIT_MASK(32),
 91	.has_wide_gather = false,
 92	.has_hypervisor = false,
 93	.num_sid_entries = 0,
 94	.sid_table = NULL,
 95	.reserve_vblank_syncpts = true,
 96};
 97
 98static const struct host1x_info host1x02_info = {
 99	.nb_channels = 9,
100	.nb_pts = 32,
101	.nb_mlocks = 16,
102	.nb_bases = 12,
103	.init = host1x02_init,
104	.sync_offset = 0x3000,
105	.dma_mask = DMA_BIT_MASK(32),
106	.has_wide_gather = false,
107	.has_hypervisor = false,
108	.num_sid_entries = 0,
109	.sid_table = NULL,
110	.reserve_vblank_syncpts = true,
111};
112
113static const struct host1x_info host1x04_info = {
114	.nb_channels = 12,
115	.nb_pts = 192,
116	.nb_mlocks = 16,
117	.nb_bases = 64,
118	.init = host1x04_init,
119	.sync_offset = 0x2100,
120	.dma_mask = DMA_BIT_MASK(34),
121	.has_wide_gather = false,
122	.has_hypervisor = false,
123	.num_sid_entries = 0,
124	.sid_table = NULL,
125	.reserve_vblank_syncpts = false,
126};
127
128static const struct host1x_info host1x05_info = {
129	.nb_channels = 14,
130	.nb_pts = 192,
131	.nb_mlocks = 16,
132	.nb_bases = 64,
133	.init = host1x05_init,
134	.sync_offset = 0x2100,
135	.dma_mask = DMA_BIT_MASK(34),
136	.has_wide_gather = false,
137	.has_hypervisor = false,
138	.num_sid_entries = 0,
139	.sid_table = NULL,
140	.reserve_vblank_syncpts = false,
141};
142
143static const struct host1x_sid_entry tegra186_sid_table[] = {
144	{
145		/* VIC */
146		.base = 0x1af0,
147		.offset = 0x30,
148		.limit = 0x34
149	},
150	{
151		/* NVDEC */
152		.base = 0x1b00,
153		.offset = 0x30,
154		.limit = 0x34
155	},
156};
157
158static const struct host1x_info host1x06_info = {
159	.nb_channels = 63,
160	.nb_pts = 576,
161	.nb_mlocks = 24,
162	.nb_bases = 16,
163	.init = host1x06_init,
164	.sync_offset = 0x0,
165	.dma_mask = DMA_BIT_MASK(40),
166	.has_wide_gather = true,
167	.has_hypervisor = true,
168	.num_sid_entries = ARRAY_SIZE(tegra186_sid_table),
169	.sid_table = tegra186_sid_table,
170	.reserve_vblank_syncpts = false,
171};
172
173static const struct host1x_sid_entry tegra194_sid_table[] = {
174	{
175		/* VIC */
176		.base = 0x1af0,
177		.offset = 0x30,
178		.limit = 0x34
179	},
180	{
181		/* NVDEC */
182		.base = 0x1b00,
183		.offset = 0x30,
184		.limit = 0x34
185	},
186	{
187		/* NVDEC1 */
188		.base = 0x1bc0,
189		.offset = 0x30,
190		.limit = 0x34
191	},
192};
193
194static const struct host1x_info host1x07_info = {
195	.nb_channels = 63,
196	.nb_pts = 704,
197	.nb_mlocks = 32,
198	.nb_bases = 0,
199	.init = host1x07_init,
200	.sync_offset = 0x0,
201	.dma_mask = DMA_BIT_MASK(40),
202	.has_wide_gather = true,
203	.has_hypervisor = true,
204	.num_sid_entries = ARRAY_SIZE(tegra194_sid_table),
205	.sid_table = tegra194_sid_table,
206	.reserve_vblank_syncpts = false,
207};
208
209/*
210 * Tegra234 has two stream ID protection tables, one for setting stream IDs
211 * through the channel path via SETSTREAMID, and one for setting them via
212 * MMIO. We program each engine's data stream ID in the channel path table
213 * and firmware stream ID in the MMIO path table.
214 */
215static const struct host1x_sid_entry tegra234_sid_table[] = {
216	{
217		/* VIC channel */
218		.base = 0x17b8,
219		.offset = 0x30,
220		.limit = 0x30
221	},
222	{
223		/* VIC MMIO */
224		.base = 0x1688,
225		.offset = 0x34,
226		.limit = 0x34
227	},
228	{
229		/* NVDEC channel */
230		.base = 0x17c8,
231		.offset = 0x30,
232		.limit = 0x30,
233	},
234	{
235		/* NVDEC MMIO */
236		.base = 0x1698,
237		.offset = 0x34,
238		.limit = 0x34,
239	},
240};
241
242static const struct host1x_info host1x08_info = {
243	.nb_channels = 63,
244	.nb_pts = 1024,
245	.nb_mlocks = 24,
246	.nb_bases = 0,
247	.init = host1x08_init,
248	.sync_offset = 0x0,
249	.dma_mask = DMA_BIT_MASK(40),
250	.has_wide_gather = true,
251	.has_hypervisor = true,
252	.has_common = true,
253	.num_sid_entries = ARRAY_SIZE(tegra234_sid_table),
254	.sid_table = tegra234_sid_table,
255	.streamid_vm_table = { 0x1004, 128 },
256	.classid_vm_table = { 0x1404, 25 },
257	.mmio_vm_table = { 0x1504, 25 },
258	.reserve_vblank_syncpts = false,
259};
260
261static const struct of_device_id host1x_of_match[] = {
262	{ .compatible = "nvidia,tegra234-host1x", .data = &host1x08_info, },
263	{ .compatible = "nvidia,tegra194-host1x", .data = &host1x07_info, },
264	{ .compatible = "nvidia,tegra186-host1x", .data = &host1x06_info, },
265	{ .compatible = "nvidia,tegra210-host1x", .data = &host1x05_info, },
266	{ .compatible = "nvidia,tegra124-host1x", .data = &host1x04_info, },
267	{ .compatible = "nvidia,tegra114-host1x", .data = &host1x02_info, },
268	{ .compatible = "nvidia,tegra30-host1x", .data = &host1x01_info, },
269	{ .compatible = "nvidia,tegra20-host1x", .data = &host1x01_info, },
270	{ },
271};
272MODULE_DEVICE_TABLE(of, host1x_of_match);
273
274static void host1x_setup_virtualization_tables(struct host1x *host)
275{
276	const struct host1x_info *info = host->info;
277	unsigned int i;
278
279	if (!info->has_hypervisor)
280		return;
281
282	for (i = 0; i < info->num_sid_entries; i++) {
283		const struct host1x_sid_entry *entry = &info->sid_table[i];
284
285		host1x_hypervisor_writel(host, entry->offset, entry->base);
286		host1x_hypervisor_writel(host, entry->limit, entry->base + 4);
287	}
288
289	for (i = 0; i < info->streamid_vm_table.count; i++) {
290		/* Allow access to all stream IDs to all VMs. */
291		host1x_hypervisor_writel(host, 0xff, info->streamid_vm_table.base + 4 * i);
292	}
293
294	for (i = 0; i < info->classid_vm_table.count; i++) {
295		/* Allow access to all classes to all VMs. */
296		host1x_hypervisor_writel(host, 0xff, info->classid_vm_table.base + 4 * i);
297	}
298
299	for (i = 0; i < info->mmio_vm_table.count; i++) {
300		/* Use VM1 (that's us) as originator VMID for engine MMIO accesses. */
301		host1x_hypervisor_writel(host, 0x1, info->mmio_vm_table.base + 4 * i);
302	}
303}
304
305static bool host1x_wants_iommu(struct host1x *host1x)
306{
307	/* Our IOMMU usage policy doesn't currently play well with GART */
308	if (of_machine_is_compatible("nvidia,tegra20"))
309		return false;
310
311	/*
312	 * If we support addressing a maximum of 32 bits of physical memory
313	 * and if the host1x firewall is enabled, there's no need to enable
314	 * IOMMU support. This can happen for example on Tegra20, Tegra30
315	 * and Tegra114.
316	 *
317	 * Tegra124 and later can address up to 34 bits of physical memory and
318	 * many platforms come equipped with more than 2 GiB of system memory,
319	 * which requires crossing the 4 GiB boundary. But there's a catch: on
320	 * SoCs before Tegra186 (i.e. Tegra124 and Tegra210), the host1x can
321	 * only address up to 32 bits of memory in GATHER opcodes, which means
322	 * that command buffers need to either be in the first 2 GiB of system
323	 * memory (which could quickly lead to memory exhaustion), or command
324	 * buffers need to be treated differently from other buffers (which is
325	 * not possible with the current ABI).
326	 *
327	 * A third option is to use the IOMMU in these cases to make sure all
328	 * buffers will be mapped into a 32-bit IOVA space that host1x can
329	 * address. This allows all of the system memory to be used and works
330	 * within the limitations of the host1x on these SoCs.
331	 *
332	 * In summary, default to enable IOMMU on Tegra124 and later. For any
333	 * of the earlier SoCs, only use the IOMMU for additional safety when
334	 * the host1x firewall is disabled.
335	 */
336	if (host1x->info->dma_mask <= DMA_BIT_MASK(32)) {
337		if (IS_ENABLED(CONFIG_TEGRA_HOST1X_FIREWALL))
338			return false;
339	}
340
341	return true;
342}
343
344static struct iommu_domain *host1x_iommu_attach(struct host1x *host)
345{
346	struct iommu_domain *domain = iommu_get_domain_for_dev(host->dev);
347	int err;
348
349#if IS_ENABLED(CONFIG_ARM_DMA_USE_IOMMU)
350	if (host->dev->archdata.mapping) {
351		struct dma_iommu_mapping *mapping =
352				to_dma_iommu_mapping(host->dev);
353		arm_iommu_detach_device(host->dev);
354		arm_iommu_release_mapping(mapping);
355
356		domain = iommu_get_domain_for_dev(host->dev);
357	}
358#endif
359
360	/*
361	 * We may not always want to enable IOMMU support (for example if the
362	 * host1x firewall is already enabled and we don't support addressing
363	 * more than 32 bits of physical memory), so check for that first.
364	 *
365	 * Similarly, if host1x is already attached to an IOMMU (via the DMA
366	 * API), don't try to attach again.
367	 */
368	if (!host1x_wants_iommu(host) || domain)
369		return domain;
370
371	host->group = iommu_group_get(host->dev);
372	if (host->group) {
373		struct iommu_domain_geometry *geometry;
374		dma_addr_t start, end;
375		unsigned long order;
376
377		err = iova_cache_get();
378		if (err < 0)
379			goto put_group;
380
381		host->domain = iommu_domain_alloc(&platform_bus_type);
382		if (!host->domain) {
383			err = -ENOMEM;
384			goto put_cache;
385		}
386
387		err = iommu_attach_group(host->domain, host->group);
388		if (err) {
389			if (err == -ENODEV)
390				err = 0;
391
392			goto free_domain;
393		}
394
395		geometry = &host->domain->geometry;
396		start = geometry->aperture_start & host->info->dma_mask;
397		end = geometry->aperture_end & host->info->dma_mask;
398
399		order = __ffs(host->domain->pgsize_bitmap);
400		init_iova_domain(&host->iova, 1UL << order, start >> order);
401		host->iova_end = end;
402
403		domain = host->domain;
404	}
405
406	return domain;
407
408free_domain:
409	iommu_domain_free(host->domain);
410	host->domain = NULL;
411put_cache:
412	iova_cache_put();
413put_group:
414	iommu_group_put(host->group);
415	host->group = NULL;
416
417	return ERR_PTR(err);
418}
419
420static int host1x_iommu_init(struct host1x *host)
421{
422	u64 mask = host->info->dma_mask;
423	struct iommu_domain *domain;
424	int err;
425
426	domain = host1x_iommu_attach(host);
427	if (IS_ERR(domain)) {
428		err = PTR_ERR(domain);
429		dev_err(host->dev, "failed to attach to IOMMU: %d\n", err);
430		return err;
431	}
432
433	/*
434	 * If we're not behind an IOMMU make sure we don't get push buffers
435	 * that are allocated outside of the range addressable by the GATHER
436	 * opcode.
437	 *
438	 * Newer generations of Tegra (Tegra186 and later) support a wide
439	 * variant of the GATHER opcode that allows addressing more bits.
440	 */
441	if (!domain && !host->info->has_wide_gather)
442		mask = DMA_BIT_MASK(32);
443
444	err = dma_coerce_mask_and_coherent(host->dev, mask);
445	if (err < 0) {
446		dev_err(host->dev, "failed to set DMA mask: %d\n", err);
447		return err;
448	}
449
450	return 0;
451}
452
453static void host1x_iommu_exit(struct host1x *host)
454{
455	if (host->domain) {
456		put_iova_domain(&host->iova);
457		iommu_detach_group(host->domain, host->group);
458
459		iommu_domain_free(host->domain);
460		host->domain = NULL;
461
462		iova_cache_put();
463
464		iommu_group_put(host->group);
465		host->group = NULL;
466	}
467}
468
469static int host1x_get_resets(struct host1x *host)
470{
471	int err;
472
473	host->resets[0].id = "mc";
474	host->resets[1].id = "host1x";
475	host->nresets = ARRAY_SIZE(host->resets);
476
477	err = devm_reset_control_bulk_get_optional_exclusive_released(
478				host->dev, host->nresets, host->resets);
479	if (err) {
480		dev_err(host->dev, "failed to get reset: %d\n", err);
481		return err;
482	}
483
484	return 0;
485}
486
487static int host1x_probe(struct platform_device *pdev)
488{
489	struct host1x *host;
490	int syncpt_irq;
491	int err;
492
493	host = devm_kzalloc(&pdev->dev, sizeof(*host), GFP_KERNEL);
494	if (!host)
495		return -ENOMEM;
496
497	host->info = of_device_get_match_data(&pdev->dev);
498
499	if (host->info->has_hypervisor) {
500		host->regs = devm_platform_ioremap_resource_byname(pdev, "vm");
501		if (IS_ERR(host->regs))
502			return PTR_ERR(host->regs);
503
504		host->hv_regs = devm_platform_ioremap_resource_byname(pdev, "hypervisor");
505		if (IS_ERR(host->hv_regs))
506			return PTR_ERR(host->hv_regs);
507
508		if (host->info->has_common) {
509			host->common_regs = devm_platform_ioremap_resource_byname(pdev, "common");
510			if (IS_ERR(host->common_regs))
511				return PTR_ERR(host->common_regs);
512		}
513	} else {
514		host->regs = devm_platform_ioremap_resource(pdev, 0);
515		if (IS_ERR(host->regs))
516			return PTR_ERR(host->regs);
517	}
518
519	syncpt_irq = platform_get_irq(pdev, 0);
520	if (syncpt_irq < 0)
521		return syncpt_irq;
522
523	mutex_init(&host->devices_lock);
524	INIT_LIST_HEAD(&host->devices);
525	INIT_LIST_HEAD(&host->list);
526	host->dev = &pdev->dev;
 
527
528	/* set common host1x device data */
529	platform_set_drvdata(pdev, host);
530
531	host->dev->dma_parms = &host->dma_parms;
532	dma_set_max_seg_size(host->dev, UINT_MAX);
 
 
 
533
534	if (host->info->init) {
535		err = host->info->init(host);
536		if (err)
537			return err;
538	}
539
540	host->clk = devm_clk_get(&pdev->dev, NULL);
541	if (IS_ERR(host->clk)) {
 
542		err = PTR_ERR(host->clk);
543
544		if (err != -EPROBE_DEFER)
545			dev_err(&pdev->dev, "failed to get clock: %d\n", err);
546
547		return err;
548	}
549
550	err = host1x_get_resets(host);
551	if (err)
552		return err;
553
554	host1x_bo_cache_init(&host->cache);
555
556	err = host1x_iommu_init(host);
557	if (err < 0) {
558		dev_err(&pdev->dev, "failed to setup IOMMU: %d\n", err);
559		goto destroy_cache;
560	}
561
562	err = host1x_channel_list_init(&host->channel_list,
563				       host->info->nb_channels);
564	if (err) {
565		dev_err(&pdev->dev, "failed to initialize channel list\n");
566		goto iommu_exit;
567	}
568
569	err = host1x_memory_context_list_init(host);
570	if (err) {
571		dev_err(&pdev->dev, "failed to initialize context list\n");
572		goto free_channels;
573	}
574
575	err = host1x_syncpt_init(host);
576	if (err) {
577		dev_err(&pdev->dev, "failed to initialize syncpts\n");
578		goto free_contexts;
579	}
580
581	err = host1x_intr_init(host, syncpt_irq);
582	if (err) {
583		dev_err(&pdev->dev, "failed to initialize interrupts\n");
584		goto deinit_syncpt;
585	}
586
587	pm_runtime_enable(&pdev->dev);
588
589	err = devm_tegra_core_dev_init_opp_table_common(&pdev->dev);
590	if (err)
591		goto pm_disable;
592
593	/* the driver's code isn't ready yet for the dynamic RPM */
594	err = pm_runtime_resume_and_get(&pdev->dev);
595	if (err)
596		goto pm_disable;
597
598	host1x_debug_init(host);
599
600	err = host1x_register(host);
601	if (err < 0)
602		goto deinit_debugfs;
603
604	err = devm_of_platform_populate(&pdev->dev);
605	if (err < 0)
606		goto unregister;
607
608	return 0;
609
610unregister:
611	host1x_unregister(host);
612deinit_debugfs:
613	host1x_debug_deinit(host);
614
615	pm_runtime_put_sync_suspend(&pdev->dev);
616pm_disable:
617	pm_runtime_disable(&pdev->dev);
618
619	host1x_intr_deinit(host);
620deinit_syncpt:
621	host1x_syncpt_deinit(host);
622free_contexts:
623	host1x_memory_context_list_free(&host->context_list);
624free_channels:
625	host1x_channel_list_free(&host->channel_list);
626iommu_exit:
627	host1x_iommu_exit(host);
628destroy_cache:
629	host1x_bo_cache_destroy(&host->cache);
630
631	return err;
632}
633
634static int host1x_remove(struct platform_device *pdev)
635{
636	struct host1x *host = platform_get_drvdata(pdev);
637
638	host1x_unregister(host);
639	host1x_debug_deinit(host);
640
641	pm_runtime_force_suspend(&pdev->dev);
642
643	host1x_intr_deinit(host);
644	host1x_syncpt_deinit(host);
645	host1x_memory_context_list_free(&host->context_list);
646	host1x_channel_list_free(&host->channel_list);
647	host1x_iommu_exit(host);
648	host1x_bo_cache_destroy(&host->cache);
649
650	return 0;
651}
652
653static int __maybe_unused host1x_runtime_suspend(struct device *dev)
654{
655	struct host1x *host = dev_get_drvdata(dev);
656	int err;
657
658	host1x_intr_stop(host);
659	host1x_syncpt_save(host);
660
661	err = reset_control_bulk_assert(host->nresets, host->resets);
662	if (err) {
663		dev_err(dev, "failed to assert reset: %d\n", err);
664		goto resume_host1x;
665	}
666
667	usleep_range(1000, 2000);
668
669	clk_disable_unprepare(host->clk);
670	reset_control_bulk_release(host->nresets, host->resets);
671
672	return 0;
673
674resume_host1x:
675	host1x_setup_virtualization_tables(host);
676	host1x_syncpt_restore(host);
677	host1x_intr_start(host);
678
679	return err;
680}
681
682static int __maybe_unused host1x_runtime_resume(struct device *dev)
683{
684	struct host1x *host = dev_get_drvdata(dev);
685	int err;
686
687	err = reset_control_bulk_acquire(host->nresets, host->resets);
688	if (err) {
689		dev_err(dev, "failed to acquire reset: %d\n", err);
690		return err;
691	}
692
693	err = clk_prepare_enable(host->clk);
694	if (err) {
695		dev_err(dev, "failed to enable clock: %d\n", err);
696		goto release_reset;
697	}
698
699	err = reset_control_bulk_deassert(host->nresets, host->resets);
700	if (err < 0) {
701		dev_err(dev, "failed to deassert reset: %d\n", err);
702		goto disable_clk;
703	}
704
705	host1x_setup_virtualization_tables(host);
706	host1x_syncpt_restore(host);
707	host1x_intr_start(host);
708
709	return 0;
710
711disable_clk:
712	clk_disable_unprepare(host->clk);
713release_reset:
714	reset_control_bulk_release(host->nresets, host->resets);
715
716	return err;
717}
718
719static const struct dev_pm_ops host1x_pm_ops = {
720	SET_RUNTIME_PM_OPS(host1x_runtime_suspend, host1x_runtime_resume,
721			   NULL)
722	/* TODO: add system suspend-resume once driver will be ready for that */
723};
724
725static struct platform_driver tegra_host1x_driver = {
726	.driver = {
727		.name = "tegra-host1x",
728		.of_match_table = host1x_of_match,
729		.pm = &host1x_pm_ops,
730	},
731	.probe = host1x_probe,
732	.remove = host1x_remove,
733};
734
735static struct platform_driver * const drivers[] = {
736	&tegra_host1x_driver,
737	&tegra_mipi_driver,
738};
739
740static int __init tegra_host1x_init(void)
741{
742	int err;
743
744	err = bus_register(&host1x_bus_type);
745	if (err < 0)
746		return err;
747
748	err = platform_register_drivers(drivers, ARRAY_SIZE(drivers));
749	if (err < 0)
750		bus_unregister(&host1x_bus_type);
751
752	return err;
753}
754module_init(tegra_host1x_init);
755
756static void __exit tegra_host1x_exit(void)
757{
758	platform_unregister_drivers(drivers, ARRAY_SIZE(drivers));
759	bus_unregister(&host1x_bus_type);
760}
761module_exit(tegra_host1x_exit);
762
763/**
764 * host1x_get_dma_mask() - query the supported DMA mask for host1x
765 * @host1x: host1x instance
766 *
767 * Note that this returns the supported DMA mask for host1x, which can be
768 * different from the applicable DMA mask under certain circumstances.
769 */
770u64 host1x_get_dma_mask(struct host1x *host1x)
771{
772	return host1x->info->dma_mask;
773}
774EXPORT_SYMBOL(host1x_get_dma_mask);
775
776MODULE_AUTHOR("Thierry Reding <thierry.reding@avionic-design.de>");
777MODULE_AUTHOR("Terje Bergstrom <tbergstrom@nvidia.com>");
778MODULE_DESCRIPTION("Host1x driver for Tegra products");
779MODULE_LICENSE("GPL");