[mirror_ubuntu-jammy-kernel.git] / drivers / gpu / host1x / job.c

// SPDX-License-Identifier: GPL-2.0-only
/*
 * Tegra host1x Job
 *
 * Copyright (c) 2010-2015, NVIDIA Corporation.
 */

#include <linux/dma-mapping.h>
#include <linux/err.h>
#include <linux/host1x.h>
#include <linux/iommu.h>
#include <linux/kref.h>
#include <linux/module.h>
#include <linux/scatterlist.h>
#include <linux/slab.h>
#include <linux/vmalloc.h>
#include <trace/events/host1x.h>

#include "channel.h"
#include "dev.h"
#include "job.h"
#include "syncpt.h"

#define HOST1X_WAIT_SYNCPT_OFFSET 0x8

struct host1x_job *host1x_job_alloc(struct host1x_channel *ch,
				    u32 num_cmdbufs, u32 num_relocs)
{
	struct host1x_job *job = NULL;
	unsigned int num_unpins = num_cmdbufs + num_relocs;
	u64 total;
	void *mem;

	/* Check that we're not going to overflow */
	total = sizeof(struct host1x_job) +
		(u64)num_relocs * sizeof(struct host1x_reloc) +
		(u64)num_unpins * sizeof(struct host1x_job_unpin_data) +
		(u64)num_cmdbufs * sizeof(struct host1x_job_gather) +
		(u64)num_unpins * sizeof(dma_addr_t) +
		(u64)num_unpins * sizeof(u32 *);
	if (total > ULONG_MAX)
		return NULL;

	mem = job = kzalloc(total, GFP_KERNEL);
	if (!job)
		return NULL;

	kref_init(&job->ref);
	job->channel = ch;

	/* Redistribute memory to the structs  */
	mem += sizeof(struct host1x_job);
	job->relocs = num_relocs ? mem : NULL;
	mem += num_relocs * sizeof(struct host1x_reloc);
	job->unpins = num_unpins ? mem : NULL;
	mem += num_unpins * sizeof(struct host1x_job_unpin_data);
	job->gathers = num_cmdbufs ? mem : NULL;
	mem += num_cmdbufs * sizeof(struct host1x_job_gather);
	job->addr_phys = num_unpins ? mem : NULL;

	job->reloc_addr_phys = job->addr_phys;
	job->gather_addr_phys = &job->addr_phys[num_relocs];

	return job;
}
EXPORT_SYMBOL(host1x_job_alloc);

struct host1x_job *host1x_job_get(struct host1x_job *job)
{
	kref_get(&job->ref);
	return job;
}
EXPORT_SYMBOL(host1x_job_get);

static void job_free(struct kref *ref)
{
	struct host1x_job *job = container_of(ref, struct host1x_job, ref);

	kfree(job);
}

void host1x_job_put(struct host1x_job *job)
{
	kref_put(&job->ref, job_free);
}
EXPORT_SYMBOL(host1x_job_put);

void host1x_job_add_gather(struct host1x_job *job, struct host1x_bo *bo,
			   unsigned int words, unsigned int offset)
{
	struct host1x_job_gather *gather = &job->gathers[job->num_gathers];

	gather->words = words;
	gather->bo = bo;
	gather->offset = offset;

	job->num_gathers++;
}
EXPORT_SYMBOL(host1x_job_add_gather);

static unsigned int pin_job(struct host1x *host, struct host1x_job *job)
{
	struct host1x_client *client = job->client;
	struct device *dev = client->dev;
	struct iommu_domain *domain;
	unsigned int i;
	int err;

	domain = iommu_get_domain_for_dev(dev);
	job->num_unpins = 0;

	for (i = 0; i < job->num_relocs; i++) {
		struct host1x_reloc *reloc = &job->relocs[i];
		dma_addr_t phys_addr, *phys;
		struct sg_table *sgt;

		reloc->target.bo = host1x_bo_get(reloc->target.bo);
		if (!reloc->target.bo) {
			err = -EINVAL;
			goto unpin;
		}

		/*
		 * If the client device is not attached to an IOMMU, the
		 * physical address of the buffer object can be used.
		 *
		 * Similarly, when an IOMMU domain is shared between all
		 * host1x clients, the IOVA is already available, so no
		 * need to map the buffer object again.
		 *
		 * XXX Note that this isn't always safe to do because it
		 * relies on an assumption that no cache maintenance is
		 * needed on the buffer objects.
		 */
		if (!domain || client->group)
			phys = &phys_addr;
		else
			phys = NULL;

		sgt = host1x_bo_pin(dev, reloc->target.bo, phys);
		if (IS_ERR(sgt)) {
			err = PTR_ERR(sgt);
			goto unpin;
		}

		if (sgt) {
			unsigned long mask = HOST1X_RELOC_READ |
					     HOST1X_RELOC_WRITE;
			enum dma_data_direction dir;

			switch (reloc->flags & mask) {
			case HOST1X_RELOC_READ:
				dir = DMA_TO_DEVICE;
				break;

			case HOST1X_RELOC_WRITE:
				dir = DMA_FROM_DEVICE;
				break;

			case HOST1X_RELOC_READ | HOST1X_RELOC_WRITE:
				dir = DMA_BIDIRECTIONAL;
				break;

			default:
				err = -EINVAL;
				goto unpin;
			}

			err = dma_map_sg(dev, sgt->sgl, sgt->nents, dir);
			if (!err) {
				err = -ENOMEM;
				goto unpin;
			}

			job->unpins[job->num_unpins].dev = dev;
			job->unpins[job->num_unpins].dir = dir;
			phys_addr = sg_dma_address(sgt->sgl);
		}

		job->addr_phys[job->num_unpins] = phys_addr;
		job->unpins[job->num_unpins].bo = reloc->target.bo;
		job->unpins[job->num_unpins].sgt = sgt;
		job->num_unpins++;
	}

	for (i = 0; i < job->num_gathers; i++) {
		struct host1x_job_gather *g = &job->gathers[i];
		size_t gather_size = 0;
		struct scatterlist *sg;
		struct sg_table *sgt;
		dma_addr_t phys_addr;
		unsigned long shift;
		struct iova *alloc;
		dma_addr_t *phys;
		unsigned int j;

		g->bo = host1x_bo_get(g->bo);
		if (!g->bo) {
			err = -EINVAL;
			goto unpin;
		}

		/**
		 * If the host1x is not attached to an IOMMU, there is no need
		 * to map the buffer object for the host1x, since the physical
		 * address can simply be used.
		 */
		if (!iommu_get_domain_for_dev(host->dev))
			phys = &phys_addr;
		else
			phys = NULL;

		sgt = host1x_bo_pin(host->dev, g->bo, phys);
		if (IS_ERR(sgt)) {
			err = PTR_ERR(sgt);
			goto unpin;
		}

		if (!IS_ENABLED(CONFIG_TEGRA_HOST1X_FIREWALL) && host->domain) {
			for_each_sg(sgt->sgl, sg, sgt->nents, j)
				gather_size += sg->length;
			gather_size = iova_align(&host->iova, gather_size);

			shift = iova_shift(&host->iova);
			alloc = alloc_iova(&host->iova, gather_size >> shift,
					   host->iova_end >> shift, true);
			if (!alloc) {
				err = -ENOMEM;
				goto unpin;
			}

			err = iommu_map_sg(host->domain,
					iova_dma_addr(&host->iova, alloc),
					sgt->sgl, sgt->nents, IOMMU_READ);
			if (err == 0) {
				__free_iova(&host->iova, alloc);
				err = -EINVAL;
				goto unpin;
			}

			job->unpins[job->num_unpins].size = gather_size;
			phys_addr = iova_dma_addr(&host->iova, alloc);
		} else if (sgt) {
			err = dma_map_sg(host->dev, sgt->sgl, sgt->nents,
					 DMA_TO_DEVICE);
			if (!err) {
				err = -ENOMEM;
				goto unpin;
			}

			job->unpins[job->num_unpins].dir = DMA_TO_DEVICE;
			job->unpins[job->num_unpins].dev = host->dev;
			phys_addr = sg_dma_address(sgt->sgl);
		}

		job->addr_phys[job->num_unpins] = phys_addr;
		job->gather_addr_phys[i] = phys_addr;

		job->unpins[job->num_unpins].bo = g->bo;
		job->unpins[job->num_unpins].sgt = sgt;
		job->num_unpins++;
	}

	return 0;

unpin:
	host1x_job_unpin(job);
	return err;
}

static int do_relocs(struct host1x_job *job, struct host1x_job_gather *g)
{
	void *cmdbuf_addr = NULL;
	struct host1x_bo *cmdbuf = g->bo;
	unsigned int i;

	/* pin & patch the relocs for one gather */
	for (i = 0; i < job->num_relocs; i++) {
		struct host1x_reloc *reloc = &job->relocs[i];
		u32 reloc_addr = (job->reloc_addr_phys[i] +
				  reloc->target.offset) >> reloc->shift;
		u32 *target;

		/* skip all other gathers */
		if (cmdbuf != reloc->cmdbuf.bo)
			continue;

		if (IS_ENABLED(CONFIG_TEGRA_HOST1X_FIREWALL)) {
			target = (u32 *)job->gather_copy_mapped +
					reloc->cmdbuf.offset / sizeof(u32) +
						g->offset / sizeof(u32);
			goto patch_reloc;
		}

		if (!cmdbuf_addr) {
			cmdbuf_addr = host1x_bo_mmap(cmdbuf);

			if (unlikely(!cmdbuf_addr)) {
				pr_err("Could not map cmdbuf for relocation\n");
				return -ENOMEM;
			}
		}

		target = cmdbuf_addr + reloc->cmdbuf.offset;
patch_reloc:
		*target = reloc_addr;
	}

	if (cmdbuf_addr)
		host1x_bo_munmap(cmdbuf, cmdbuf_addr);

	return 0;
}

static bool check_reloc(struct host1x_reloc *reloc, struct host1x_bo *cmdbuf,
			unsigned int offset)
{
	offset *= sizeof(u32);

	if (reloc->cmdbuf.bo != cmdbuf || reloc->cmdbuf.offset != offset)
		return false;

	/* relocation shift value validation isn't implemented yet */
	if (reloc->shift)
		return false;

	return true;
}

struct host1x_firewall {
	struct host1x_job *job;
	struct device *dev;

	unsigned int num_relocs;
	struct host1x_reloc *reloc;

	struct host1x_bo *cmdbuf;
	unsigned int offset;

	u32 words;
	u32 class;
	u32 reg;
	u32 mask;
	u32 count;
};

static int check_register(struct host1x_firewall *fw, unsigned long offset)
{
	if (!fw->job->is_addr_reg)
		return 0;

	if (fw->job->is_addr_reg(fw->dev, fw->class, offset)) {
		if (!fw->num_relocs)
			return -EINVAL;

		if (!check_reloc(fw->reloc, fw->cmdbuf, fw->offset))
			return -EINVAL;

		fw->num_relocs--;
		fw->reloc++;
	}

	return 0;
}

static int check_class(struct host1x_firewall *fw, u32 class)
{
	if (!fw->job->is_valid_class) {
		if (fw->class != class)
			return -EINVAL;
	} else {
		if (!fw->job->is_valid_class(fw->class))
			return -EINVAL;
	}

	return 0;
}

static int check_mask(struct host1x_firewall *fw)
{
	u32 mask = fw->mask;
	u32 reg = fw->reg;
	int ret;

	while (mask) {
		if (fw->words == 0)
			return -EINVAL;

		if (mask & 1) {
			ret = check_register(fw, reg);
			if (ret < 0)
				return ret;

			fw->words--;
			fw->offset++;
		}
		mask >>= 1;
		reg++;
	}

	return 0;
}

static int check_incr(struct host1x_firewall *fw)
{
	u32 count = fw->count;
	u32 reg = fw->reg;
	int ret;

	while (count) {
		if (fw->words == 0)
			return -EINVAL;

		ret = check_register(fw, reg);
		if (ret < 0)
			return ret;

		reg++;
		fw->words--;
		fw->offset++;
		count--;
	}

	return 0;
}

static int check_nonincr(struct host1x_firewall *fw)
{
	u32 count = fw->count;
	int ret;

	while (count) {
		if (fw->words == 0)
			return -EINVAL;

		ret = check_register(fw, fw->reg);
		if (ret < 0)
			return ret;

		fw->words--;
		fw->offset++;
		count--;
	}

	return 0;
}

static int validate(struct host1x_firewall *fw, struct host1x_job_gather *g)
{
	u32 *cmdbuf_base = (u32 *)fw->job->gather_copy_mapped +
		(g->offset / sizeof(u32));
	u32 job_class = fw->class;
	int err = 0;

	fw->words = g->words;
	fw->cmdbuf = g->bo;
	fw->offset = 0;

	while (fw->words && !err) {
		u32 word = cmdbuf_base[fw->offset];
		u32 opcode = (word & 0xf0000000) >> 28;

		fw->mask = 0;
		fw->reg = 0;
		fw->count = 0;
		fw->words--;
		fw->offset++;

		switch (opcode) {
		case 0:
			fw->class = word >> 6 & 0x3ff;
			fw->mask = word & 0x3f;
			fw->reg = word >> 16 & 0xfff;
			err = check_class(fw, job_class);
			if (!err)
				err = check_mask(fw);
			if (err)
				goto out;
			break;
		case 1:
			fw->reg = word >> 16 & 0xfff;
			fw->count = word & 0xffff;
			err = check_incr(fw);
			if (err)
				goto out;
			break;

		case 2:
			fw->reg = word >> 16 & 0xfff;
			fw->count = word & 0xffff;
			err = check_nonincr(fw);
			if (err)
				goto out;
			break;

		case 3:
			fw->mask = word & 0xffff;
			fw->reg = word >> 16 & 0xfff;
			err = check_mask(fw);
			if (err)
				goto out;
			break;
		case 4:
		case 14:
			break;
		default:
			err = -EINVAL;
			break;
		}
	}

out:
	return err;
}

static inline int copy_gathers(struct device *host, struct host1x_job *job,
			       struct device *dev)
{
	struct host1x_firewall fw;
	size_t size = 0;
	size_t offset = 0;
	unsigned int i;

	fw.job = job;
	fw.dev = dev;
	fw.reloc = job->relocs;
	fw.num_relocs = job->num_relocs;
	fw.class = job->class;

	for (i = 0; i < job->num_gathers; i++) {
		struct host1x_job_gather *g = &job->gathers[i];

		size += g->words * sizeof(u32);
	}

	/*
	 * Try a non-blocking allocation from a higher priority pools first,
	 * as awaiting for the allocation here is a major performance hit.
	 */
	job->gather_copy_mapped = dma_alloc_wc(host, size, &job->gather_copy,
					       GFP_NOWAIT);

	/* the higher priority allocation failed, try the generic-blocking */
	if (!job->gather_copy_mapped)
		job->gather_copy_mapped = dma_alloc_wc(host, size,
						       &job->gather_copy,
						       GFP_KERNEL);
	if (!job->gather_copy_mapped)
		return -ENOMEM;

	job->gather_copy_size = size;

	for (i = 0; i < job->num_gathers; i++) {
		struct host1x_job_gather *g = &job->gathers[i];
		void *gather;

		/* Copy the gather */
		gather = host1x_bo_mmap(g->bo);
		memcpy(job->gather_copy_mapped + offset, gather + g->offset,
		       g->words * sizeof(u32));
		host1x_bo_munmap(g->bo, gather);

		/* Store the location in the buffer */
		g->base = job->gather_copy;
		g->offset = offset;

		/* Validate the job */
		if (validate(&fw, g))
			return -EINVAL;

		offset += g->words * sizeof(u32);
	}

	/* No relocs should remain at this point */
	if (fw.num_relocs)
		return -EINVAL;

	return 0;
}

int host1x_job_pin(struct host1x_job *job, struct device *dev)
{
	int err;
	unsigned int i, j;
	struct host1x *host = dev_get_drvdata(dev->parent);

	/* pin memory */
	err = pin_job(host, job);
	if (err)
		goto out;

	if (IS_ENABLED(CONFIG_TEGRA_HOST1X_FIREWALL)) {
		err = copy_gathers(host->dev, job, dev);
		if (err)
			goto out;
	}

	/* patch gathers */
	for (i = 0; i < job->num_gathers; i++) {
		struct host1x_job_gather *g = &job->gathers[i];

		/* process each gather mem only once */
		if (g->handled)
			continue;

		/* copy_gathers() sets gathers base if firewall is enabled */
		if (!IS_ENABLED(CONFIG_TEGRA_HOST1X_FIREWALL))
			g->base = job->gather_addr_phys[i];

		for (j = i + 1; j < job->num_gathers; j++) {
			if (job->gathers[j].bo == g->bo) {
				job->gathers[j].handled = true;
				job->gathers[j].base = g->base;
			}
		}

		err = do_relocs(job, g);
		if (err)
			break;
	}

out:
	if (err)
		host1x_job_unpin(job);
	wmb();

	return err;
}
EXPORT_SYMBOL(host1x_job_pin);

void host1x_job_unpin(struct host1x_job *job)
{
	struct host1x *host = dev_get_drvdata(job->channel->dev->parent);
	unsigned int i;

	for (i = 0; i < job->num_unpins; i++) {
		struct host1x_job_unpin_data *unpin = &job->unpins[i];
		struct device *dev = unpin->dev ?: host->dev;
		struct sg_table *sgt = unpin->sgt;

		if (!IS_ENABLED(CONFIG_TEGRA_HOST1X_FIREWALL) &&
		    unpin->size && host->domain) {
			iommu_unmap(host->domain, job->addr_phys[i],
				    unpin->size);
			free_iova(&host->iova,
				iova_pfn(&host->iova, job->addr_phys[i]));
		}

		if (unpin->dev && sgt)
			dma_unmap_sg(unpin->dev, sgt->sgl, sgt->nents,
				     unpin->dir);

		host1x_bo_unpin(dev, unpin->bo, sgt);
		host1x_bo_put(unpin->bo);
	}

	job->num_unpins = 0;

	if (job->gather_copy_size)
		dma_free_wc(host->dev, job->gather_copy_size,
			    job->gather_copy_mapped, job->gather_copy);
}
EXPORT_SYMBOL(host1x_job_unpin);

/*
 * Debug routine used to dump job entries
 */
void host1x_job_dump(struct device *dev, struct host1x_job *job)
{
	dev_dbg(dev, "    SYNCPT_ID   %d\n", job->syncpt_id);
	dev_dbg(dev, "    SYNCPT_VAL  %d\n", job->syncpt_end);
	dev_dbg(dev, "    FIRST_GET   0x%x\n", job->first_get);
	dev_dbg(dev, "    TIMEOUT     %d\n", job->timeout);
	dev_dbg(dev, "    NUM_SLOTS   %d\n", job->num_slots);
	dev_dbg(dev, "    NUM_HANDLES %d\n", job->num_unpins);
}
Commit	Line	Data
9952f691	1	// SPDX-License-Identifier: GPL-2.0-only
6579324a TB	2	/*
	3	* Tegra host1x Job
	4	*
f08ef2d1	5	* Copyright (c) 2010-2015, NVIDIA Corporation.
6579324a TB	6	*/
	7
	8	#include <linux/dma-mapping.h>
	9	#include <linux/err.h>
35d747a8	10	#include <linux/host1x.h>
273da5a0	11	#include <linux/iommu.h>
6579324a TB	12	#include <linux/kref.h>
	13	#include <linux/module.h>
	14	#include <linux/scatterlist.h>
	15	#include <linux/slab.h>
	16	#include <linux/vmalloc.h>
	17	#include <trace/events/host1x.h>
	18
	19	#include "channel.h"
	20	#include "dev.h"
6579324a TB	21	#include "job.h"
	22	#include "syncpt.h"
	23
a47ac10e DO	24	#define HOST1X_WAIT_SYNCPT_OFFSET 0x8
a47ac10e DO	25
6579324a	26	struct host1x_job host1x_job_alloc(struct host1x_channel ch,
24c94e16	27	u32 num_cmdbufs, u32 num_relocs)
6579324a TB	28	{
	29	struct host1x_job *job = NULL;
	30	unsigned int num_unpins = num_cmdbufs + num_relocs;
	31	u64 total;
	32	void *mem;
	33
	34	/* Check that we're not going to overflow */
	35	total = sizeof(struct host1x_job) +
f5fda676 DC	36	(u64)num_relocs * sizeof(struct host1x_reloc) +
f5fda676 DC	37	(u64)num_unpins * sizeof(struct host1x_job_unpin_data) +
f5fda676 DC	38	(u64)num_cmdbufs * sizeof(struct host1x_job_gather) +
	39	(u64)num_unpins * sizeof(dma_addr_t) +
	40	(u64)num_unpins * sizeof(u32 *);
6579324a TB	41	if (total > ULONG_MAX)
	42	return NULL;
	43
	44	mem = job = kzalloc(total, GFP_KERNEL);
	45	if (!job)
	46	return NULL;
	47
	48	kref_init(&job->ref);
	49	job->channel = ch;
	50
	51	/* Redistribute memory to the structs */
	52	mem += sizeof(struct host1x_job);
06490bb9	53	job->relocs = num_relocs ? mem : NULL;
6579324a TB	54	mem += num_relocs * sizeof(struct host1x_reloc);
	55	job->unpins = num_unpins ? mem : NULL;
	56	mem += num_unpins * sizeof(struct host1x_job_unpin_data);
6579324a TB	57	job->gathers = num_cmdbufs ? mem : NULL;
	58	mem += num_cmdbufs * sizeof(struct host1x_job_gather);
	59	job->addr_phys = num_unpins ? mem : NULL;
	60
	61	job->reloc_addr_phys = job->addr_phys;
	62	job->gather_addr_phys = &job->addr_phys[num_relocs];
	63
	64	return job;
	65	}
fae798a1	66	EXPORT_SYMBOL(host1x_job_alloc);
6579324a TB	67
	68	struct host1x_job host1x_job_get(struct host1x_job job)
	69	{
	70	kref_get(&job->ref);
	71	return job;
	72	}
fae798a1	73	EXPORT_SYMBOL(host1x_job_get);
6579324a TB	74
	75	static void job_free(struct kref *ref)
	76	{
	77	struct host1x_job *job = container_of(ref, struct host1x_job, ref);
	78
	79	kfree(job);
	80	}
	81
	82	void host1x_job_put(struct host1x_job *job)
	83	{
	84	kref_put(&job->ref, job_free);
	85	}
fae798a1	86	EXPORT_SYMBOL(host1x_job_put);
6579324a TB	87
6579324a TB	88	void host1x_job_add_gather(struct host1x_job job, struct host1x_bo bo,
326bbd79	89	unsigned int words, unsigned int offset)
6579324a	90	{
326bbd79 TR	91	struct host1x_job_gather *gather = &job->gathers[job->num_gathers];
	92
	93	gather->words = words;
	94	gather->bo = bo;
	95	gather->offset = offset;
6579324a	96
6579324a TB	97	job->num_gathers++;
6579324a TB	98	}
fae798a1	99	EXPORT_SYMBOL(host1x_job_add_gather);
6579324a	100
404bfb78	101	static unsigned int pin_job(struct host1x host, struct host1x_job job)
6579324a	102	{
af1cbfb9 TR	103	struct host1x_client *client = job->client;
af1cbfb9 TR	104	struct device *dev = client->dev;
273da5a0	105	struct iommu_domain *domain;
6579324a	106	unsigned int i;
404bfb78	107	int err;
6579324a	108
273da5a0	109	domain = iommu_get_domain_for_dev(dev);
6579324a TB	110	job->num_unpins = 0;
	111
	112	for (i = 0; i < job->num_relocs; i++) {
06490bb9	113	struct host1x_reloc *reloc = &job->relocs[i];
af1cbfb9	114	dma_addr_t phys_addr, *phys;
6579324a	115	struct sg_table *sgt;
6579324a	116
961e3bea	117	reloc->target.bo = host1x_bo_get(reloc->target.bo);
404bfb78 MP	118	if (!reloc->target.bo) {
404bfb78 MP	119	err = -EINVAL;
6579324a	120	goto unpin;
404bfb78	121	}
6579324a	122
273da5a0 TR	123	/*
	124	* If the client device is not attached to an IOMMU, the
	125	* physical address of the buffer object can be used.
	126	*
	127	* Similarly, when an IOMMU domain is shared between all
	128	* host1x clients, the IOVA is already available, so no
	129	* need to map the buffer object again.
	130	*
	131	* XXX Note that this isn't always safe to do because it
	132	* relies on an assumption that no cache maintenance is
	133	* needed on the buffer objects.
	134	*/
	135	if (!domain \|\| client->group)
af1cbfb9 TR	136	phys = &phys_addr;
	137	else
	138	phys = NULL;
	139
	140	sgt = host1x_bo_pin(dev, reloc->target.bo, phys);
80327ce3 TR	141	if (IS_ERR(sgt)) {
	142	err = PTR_ERR(sgt);
	143	goto unpin;
	144	}
6579324a	145
af1cbfb9 TR	146	if (sgt) {
	147	unsigned long mask = HOST1X_RELOC_READ \|
	148	HOST1X_RELOC_WRITE;
	149	enum dma_data_direction dir;
	150
	151	switch (reloc->flags & mask) {
	152	case HOST1X_RELOC_READ:
	153	dir = DMA_TO_DEVICE;
	154	break;
	155
	156	case HOST1X_RELOC_WRITE:
	157	dir = DMA_FROM_DEVICE;
	158	break;
	159
	160	case HOST1X_RELOC_READ \| HOST1X_RELOC_WRITE:
	161	dir = DMA_BIDIRECTIONAL;
	162	break;
	163
	164	default:
	165	err = -EINVAL;
	166	goto unpin;
	167	}
	168
	169	err = dma_map_sg(dev, sgt->sgl, sgt->nents, dir);
	170	if (!err) {
	171	err = -ENOMEM;
	172	goto unpin;
	173	}
	174
	175	job->unpins[job->num_unpins].dev = dev;
	176	job->unpins[job->num_unpins].dir = dir;
	177	phys_addr = sg_dma_address(sgt->sgl);
	178	}
	179
6579324a	180	job->addr_phys[job->num_unpins] = phys_addr;
961e3bea	181	job->unpins[job->num_unpins].bo = reloc->target.bo;
6579324a TB	182	job->unpins[job->num_unpins].sgt = sgt;
	183	job->num_unpins++;
	184	}
	185
	186	for (i = 0; i < job->num_gathers; i++) {
	187	struct host1x_job_gather *g = &job->gathers[i];
404bfb78 MP	188	size_t gather_size = 0;
404bfb78 MP	189	struct scatterlist *sg;
6579324a TB	190	struct sg_table *sgt;
6579324a TB	191	dma_addr_t phys_addr;
404bfb78 MP	192	unsigned long shift;
404bfb78 MP	193	struct iova *alloc;
273da5a0	194	dma_addr_t *phys;
404bfb78	195	unsigned int j;
6579324a TB	196
6579324a TB	197	g->bo = host1x_bo_get(g->bo);
404bfb78 MP	198	if (!g->bo) {
404bfb78 MP	199	err = -EINVAL;
6579324a	200	goto unpin;
404bfb78	201	}
6579324a	202
273da5a0 TR	203	/**
	204	* If the host1x is not attached to an IOMMU, there is no need
	205	* to map the buffer object for the host1x, since the physical
	206	* address can simply be used.
	207	*/
	208	if (!iommu_get_domain_for_dev(host->dev))
	209	phys = &phys_addr;
	210	else
	211	phys = NULL;
	212
	213	sgt = host1x_bo_pin(host->dev, g->bo, phys);
80327ce3 TR	214	if (IS_ERR(sgt)) {
	215	err = PTR_ERR(sgt);
	216	goto unpin;
	217	}
404bfb78 MP	218
	219	if (!IS_ENABLED(CONFIG_TEGRA_HOST1X_FIREWALL) && host->domain) {
	220	for_each_sg(sgt->sgl, sg, sgt->nents, j)
	221	gather_size += sg->length;
	222	gather_size = iova_align(&host->iova, gather_size);
	223
	224	shift = iova_shift(&host->iova);
	225	alloc = alloc_iova(&host->iova, gather_size >> shift,
	226	host->iova_end >> shift, true);
	227	if (!alloc) {
	228	err = -ENOMEM;
	229	goto unpin;
	230	}
	231
	232	err = iommu_map_sg(host->domain,
	233	iova_dma_addr(&host->iova, alloc),
	234	sgt->sgl, sgt->nents, IOMMU_READ);
	235	if (err == 0) {
	236	__free_iova(&host->iova, alloc);
	237	err = -EINVAL;
	238	goto unpin;
	239	}
	240
404bfb78	241	job->unpins[job->num_unpins].size = gather_size;
af1cbfb9	242	phys_addr = iova_dma_addr(&host->iova, alloc);
273da5a0	243	} else if (sgt) {
af1cbfb9 TR	244	err = dma_map_sg(host->dev, sgt->sgl, sgt->nents,
	245	DMA_TO_DEVICE);
	246	if (!err) {
	247	err = -ENOMEM;
	248	goto unpin;
	249	}
	250
98ae41ad	251	job->unpins[job->num_unpins].dir = DMA_TO_DEVICE;
af1cbfb9 TR	252	job->unpins[job->num_unpins].dev = host->dev;
af1cbfb9 TR	253	phys_addr = sg_dma_address(sgt->sgl);
404bfb78 MP	254	}
404bfb78 MP	255
af1cbfb9 TR	256	job->addr_phys[job->num_unpins] = phys_addr;
af1cbfb9 TR	257	job->gather_addr_phys[i] = phys_addr;
6579324a	258
6579324a TB	259	job->unpins[job->num_unpins].bo = g->bo;
	260	job->unpins[job->num_unpins].sgt = sgt;
	261	job->num_unpins++;
	262	}
	263
404bfb78	264	return 0;
6579324a TB	265
	266	unpin:
	267	host1x_job_unpin(job);
404bfb78	268	return err;
6579324a TB	269	}
6579324a TB	270
47f89c10	271	static int do_relocs(struct host1x_job job, struct host1x_job_gather g)
6579324a	272	{
7a8139c5	273	void *cmdbuf_addr = NULL;
47f89c10	274	struct host1x_bo *cmdbuf = g->bo;
d4ad3ad9	275	unsigned int i;
6579324a TB	276
6579324a TB	277	/* pin & patch the relocs for one gather */
3364cd28	278	for (i = 0; i < job->num_relocs; i++) {
06490bb9	279	struct host1x_reloc *reloc = &job->relocs[i];
6579324a	280	u32 reloc_addr = (job->reloc_addr_phys[i] +
961e3bea	281	reloc->target.offset) >> reloc->shift;
6579324a TB	282	u32 *target;
	283
	284	/* skip all other gathers */
961e3bea	285	if (cmdbuf != reloc->cmdbuf.bo)
6579324a	286	continue;
6579324a	287
47f89c10 DO	288	if (IS_ENABLED(CONFIG_TEGRA_HOST1X_FIREWALL)) {
	289	target = (u32 *)job->gather_copy_mapped +
	290	reloc->cmdbuf.offset / sizeof(u32) +
	291	g->offset / sizeof(u32);
	292	goto patch_reloc;
	293	}
	294
7a8139c5 DV	295	if (!cmdbuf_addr) {
7a8139c5 DV	296	cmdbuf_addr = host1x_bo_mmap(cmdbuf);
6579324a	297
7a8139c5	298	if (unlikely(!cmdbuf_addr)) {
6579324a TB	299	pr_err("Could not map cmdbuf for relocation\n");
	300	return -ENOMEM;
	301	}
	302	}
	303
7a8139c5	304	target = cmdbuf_addr + reloc->cmdbuf.offset;
47f89c10	305	patch_reloc:
6579324a	306	*target = reloc_addr;
6579324a TB	307	}
6579324a TB	308
7a8139c5 DV	309	if (cmdbuf_addr)
7a8139c5 DV	310	host1x_bo_munmap(cmdbuf, cmdbuf_addr);
6579324a TB	311
	312	return 0;
	313	}
	314
5060d8ec	315	static bool check_reloc(struct host1x_reloc reloc, struct host1x_bo cmdbuf,
37857cd2	316	unsigned int offset)
6579324a TB	317	{
	318	offset *= sizeof(u32);
	319
961e3bea	320	if (reloc->cmdbuf.bo != cmdbuf \|\| reloc->cmdbuf.offset != offset)
5060d8ec	321	return false;
6579324a	322
571cbf70 DO	323	/* relocation shift value validation isn't implemented yet */
	324	if (reloc->shift)
	325	return false;
	326
5060d8ec	327	return true;
6579324a TB	328	}
	329
	330	struct host1x_firewall {
	331	struct host1x_job *job;
	332	struct device *dev;
	333
	334	unsigned int num_relocs;
	335	struct host1x_reloc *reloc;
	336
d7fbcf47	337	struct host1x_bo *cmdbuf;
6579324a TB	338	unsigned int offset;
	339
	340	u32 words;
	341	u32 class;
	342	u32 reg;
	343	u32 mask;
	344	u32 count;
	345	};
	346
d77563ff TR	347	static int check_register(struct host1x_firewall *fw, unsigned long offset)
d77563ff TR	348	{
0f563a4b DO	349	if (!fw->job->is_addr_reg)
	350	return 0;
	351
d77563ff TR	352	if (fw->job->is_addr_reg(fw->dev, fw->class, offset)) {
	353	if (!fw->num_relocs)
	354	return -EINVAL;
	355
	356	if (!check_reloc(fw->reloc, fw->cmdbuf, fw->offset))
	357	return -EINVAL;
	358
	359	fw->num_relocs--;
	360	fw->reloc++;
	361	}
	362
	363	return 0;
	364	}
	365
0f563a4b DO	366	static int check_class(struct host1x_firewall *fw, u32 class)
	367	{
	368	if (!fw->job->is_valid_class) {
	369	if (fw->class != class)
	370	return -EINVAL;
	371	} else {
	372	if (!fw->job->is_valid_class(fw->class))
	373	return -EINVAL;
	374	}
	375
	376	return 0;
	377	}
	378
6579324a TB	379	static int check_mask(struct host1x_firewall *fw)
	380	{
	381	u32 mask = fw->mask;
	382	u32 reg = fw->reg;
d77563ff	383	int ret;
6579324a TB	384
	385	while (mask) {
	386	if (fw->words == 0)
	387	return -EINVAL;
	388
	389	if (mask & 1) {
d77563ff TR	390	ret = check_register(fw, reg);
	391	if (ret < 0)
	392	return ret;
	393
6579324a TB	394	fw->words--;
	395	fw->offset++;
	396	}
	397	mask >>= 1;
	398	reg++;
	399	}
	400
	401	return 0;
	402	}
	403
	404	static int check_incr(struct host1x_firewall *fw)
	405	{
	406	u32 count = fw->count;
	407	u32 reg = fw->reg;
d77563ff	408	int ret;
6579324a	409
64c173d3	410	while (count) {
6579324a TB	411	if (fw->words == 0)
	412	return -EINVAL;
	413
d77563ff TR	414	ret = check_register(fw, reg);
	415	if (ret < 0)
	416	return ret;
	417
6579324a TB	418	reg++;
	419	fw->words--;
	420	fw->offset++;
	421	count--;
	422	}
	423
	424	return 0;
	425	}
	426
	427	static int check_nonincr(struct host1x_firewall *fw)
	428	{
6579324a	429	u32 count = fw->count;
d77563ff	430	int ret;
6579324a TB	431
	432	while (count) {
	433	if (fw->words == 0)
	434	return -EINVAL;
	435
d77563ff TR	436	ret = check_register(fw, fw->reg);
	437	if (ret < 0)
	438	return ret;
	439
6579324a TB	440	fw->words--;
	441	fw->offset++;
	442	count--;
	443	}
	444
	445	return 0;
	446	}
	447
afac0e43	448	static int validate(struct host1x_firewall fw, struct host1x_job_gather g)
6579324a	449	{
3364cd28 AM	450	u32 cmdbuf_base = (u32 )fw->job->gather_copy_mapped +
3364cd28 AM	451	(g->offset / sizeof(u32));
0f563a4b	452	u32 job_class = fw->class;
6579324a	453	int err = 0;
6579324a	454
afac0e43	455	fw->words = g->words;
d7fbcf47	456	fw->cmdbuf = g->bo;
afac0e43	457	fw->offset = 0;
6579324a	458
afac0e43 TB	459	while (fw->words && !err) {
afac0e43 TB	460	u32 word = cmdbuf_base[fw->offset];
6579324a TB	461	u32 opcode = (word & 0xf0000000) >> 28;
6579324a TB	462
afac0e43 TB	463	fw->mask = 0;
	464	fw->reg = 0;
	465	fw->count = 0;
	466	fw->words--;
	467	fw->offset++;
6579324a TB	468
	469	switch (opcode) {
	470	case 0:
afac0e43 TB	471	fw->class = word >> 6 & 0x3ff;
	472	fw->mask = word & 0x3f;
	473	fw->reg = word >> 16 & 0xfff;
0f563a4b DO	474	err = check_class(fw, job_class);
	475	if (!err)
	476	err = check_mask(fw);
6579324a TB	477	if (err)
	478	goto out;
	479	break;
	480	case 1:
afac0e43 TB	481	fw->reg = word >> 16 & 0xfff;
	482	fw->count = word & 0xffff;
	483	err = check_incr(fw);
6579324a TB	484	if (err)
	485	goto out;
	486	break;
	487
	488	case 2:
afac0e43 TB	489	fw->reg = word >> 16 & 0xfff;
	490	fw->count = word & 0xffff;
	491	err = check_nonincr(fw);
6579324a TB	492	if (err)
	493	goto out;
	494	break;
	495
	496	case 3:
afac0e43 TB	497	fw->mask = word & 0xffff;
	498	fw->reg = word >> 16 & 0xfff;
	499	err = check_mask(fw);
6579324a TB	500	if (err)
	501	goto out;
	502	break;
	503	case 4:
6579324a TB	504	case 14:
	505	break;
	506	default:
	507	err = -EINVAL;
	508	break;
	509	}
	510	}
	511
6579324a	512	out:
6579324a TB	513	return err;
	514	}
	515
b78e70c0 TR	516	static inline int copy_gathers(struct device host, struct host1x_job job,
b78e70c0 TR	517	struct device *dev)
6579324a	518	{
3364cd28	519	struct host1x_firewall fw;
6579324a TB	520	size_t size = 0;
6579324a TB	521	size_t offset = 0;
d4ad3ad9	522	unsigned int i;
6579324a	523
3364cd28 AM	524	fw.job = job;
3364cd28 AM	525	fw.dev = dev;
06490bb9	526	fw.reloc = job->relocs;
3364cd28	527	fw.num_relocs = job->num_relocs;
3833d16f	528	fw.class = job->class;
3364cd28	529
6579324a TB	530	for (i = 0; i < job->num_gathers; i++) {
6579324a TB	531	struct host1x_job_gather *g = &job->gathers[i];
6df633d0	532
6579324a TB	533	size += g->words * sizeof(u32);
	534	}
	535
43240bbd DO	536	/*
	537	* Try a non-blocking allocation from a higher priority pools first,
	538	* as awaiting for the allocation here is a major performance hit.
	539	*/
b78e70c0	540	job->gather_copy_mapped = dma_alloc_wc(host, size, &job->gather_copy,
43240bbd DO	541	GFP_NOWAIT);
	542
	543	/* the higher priority allocation failed, try the generic-blocking */
	544	if (!job->gather_copy_mapped)
b78e70c0	545	job->gather_copy_mapped = dma_alloc_wc(host, size,
43240bbd DO	546	&job->gather_copy,
	547	GFP_KERNEL);
	548	if (!job->gather_copy_mapped)
745cecc0	549	return -ENOMEM;
6579324a TB	550
	551	job->gather_copy_size = size;
	552
	553	for (i = 0; i < job->num_gathers; i++) {
	554	struct host1x_job_gather *g = &job->gathers[i];
	555	void *gather;
	556
3364cd28	557	/* Copy the gather */
6579324a TB	558	gather = host1x_bo_mmap(g->bo);
	559	memcpy(job->gather_copy_mapped + offset, gather + g->offset,
	560	g->words * sizeof(u32));
	561	host1x_bo_munmap(g->bo, gather);
	562
3364cd28	563	/* Store the location in the buffer */
6579324a TB	564	g->base = job->gather_copy;
6579324a TB	565	g->offset = offset;
3364cd28 AM	566
	567	/* Validate the job */
	568	if (validate(&fw, g))
	569	return -EINVAL;
6579324a TB	570
	571	offset += g->words * sizeof(u32);
	572	}
	573
24c94e16 TR	574	/* No relocs should remain at this point */
24c94e16 TR	575	if (fw.num_relocs)
a9ff9995 EFL	576	return -EINVAL;
a9ff9995 EFL	577
6579324a TB	578	return 0;
	579	}
	580
	581	int host1x_job_pin(struct host1x_job job, struct device dev)
	582	{
	583	int err;
	584	unsigned int i, j;
	585	struct host1x *host = dev_get_drvdata(dev->parent);
6579324a TB	586
6579324a TB	587	/* pin memory */
404bfb78 MP	588	err = pin_job(host, job);
404bfb78 MP	589	if (err)
6579324a TB	590	goto out;
6579324a TB	591
47f89c10	592	if (IS_ENABLED(CONFIG_TEGRA_HOST1X_FIREWALL)) {
b78e70c0	593	err = copy_gathers(host->dev, job, dev);
47f89c10 DO	594	if (err)
	595	goto out;
	596	}
	597
6579324a TB	598	/* patch gathers */
	599	for (i = 0; i < job->num_gathers; i++) {
	600	struct host1x_job_gather *g = &job->gathers[i];
	601
	602	/* process each gather mem only once */
	603	if (g->handled)
	604	continue;
	605
47f89c10 DO	606	/* copy_gathers() sets gathers base if firewall is enabled */
	607	if (!IS_ENABLED(CONFIG_TEGRA_HOST1X_FIREWALL))
	608	g->base = job->gather_addr_phys[i];
6579324a	609
f08ef2d1 AM	610	for (j = i + 1; j < job->num_gathers; j++) {
f08ef2d1 AM	611	if (job->gathers[j].bo == g->bo) {
6579324a	612	job->gathers[j].handled = true;
f08ef2d1 AM	613	job->gathers[j].base = g->base;
	614	}
	615	}
6579324a	616
47f89c10	617	err = do_relocs(job, g);
6579324a	618	if (err)
47f89c10	619	break;
6579324a TB	620	}
6579324a TB	621
6579324a	622	out:
e5855aa3 DO	623	if (err)
e5855aa3 DO	624	host1x_job_unpin(job);
6579324a TB	625	wmb();
	626
	627	return err;
	628	}
fae798a1	629	EXPORT_SYMBOL(host1x_job_pin);
6579324a TB	630
	631	void host1x_job_unpin(struct host1x_job *job)
	632	{
404bfb78	633	struct host1x *host = dev_get_drvdata(job->channel->dev->parent);
6579324a TB	634	unsigned int i;
	635
	636	for (i = 0; i < job->num_unpins; i++) {
	637	struct host1x_job_unpin_data *unpin = &job->unpins[i];
af1cbfb9 TR	638	struct device *dev = unpin->dev ?: host->dev;
af1cbfb9 TR	639	struct sg_table *sgt = unpin->sgt;
6df633d0	640
ec589232 DO	641	if (!IS_ENABLED(CONFIG_TEGRA_HOST1X_FIREWALL) &&
ec589232 DO	642	unpin->size && host->domain) {
404bfb78 MP	643	iommu_unmap(host->domain, job->addr_phys[i],
	644	unpin->size);
	645	free_iova(&host->iova,
	646	iova_pfn(&host->iova, job->addr_phys[i]));
	647	}
	648
af1cbfb9 TR	649	if (unpin->dev && sgt)
	650	dma_unmap_sg(unpin->dev, sgt->sgl, sgt->nents,
	651	unpin->dir);
	652
	653	host1x_bo_unpin(dev, unpin->bo, sgt);
6579324a TB	654	host1x_bo_put(unpin->bo);
6579324a TB	655	}
0b8070d1	656
6579324a TB	657	job->num_unpins = 0;
	658
	659	if (job->gather_copy_size)
b78e70c0	660	dma_free_wc(host->dev, job->gather_copy_size,
0b8070d1	661	job->gather_copy_mapped, job->gather_copy);
6579324a	662	}
fae798a1	663	EXPORT_SYMBOL(host1x_job_unpin);
6579324a TB	664
	665	/*
	666	* Debug routine used to dump job entries
	667	*/
	668	void host1x_job_dump(struct device dev, struct host1x_job job)
	669	{
	670	dev_dbg(dev, " SYNCPT_ID %d\n", job->syncpt_id);
	671	dev_dbg(dev, " SYNCPT_VAL %d\n", job->syncpt_end);
	672	dev_dbg(dev, " FIRST_GET 0x%x\n", job->first_get);
	673	dev_dbg(dev, " TIMEOUT %d\n", job->timeout);
	674	dev_dbg(dev, " NUM_SLOTS %d\n", job->num_slots);
	675	dev_dbg(dev, " NUM_HANDLES %d\n", job->num_unpins);
	676	}