Message-ID: <09c4c066-6467-57d1-364b-be8964f6b0bb@suse.de>
Date:   Mon, 19 Jun 2023 15:18:11 +0200
From:   Thomas Zimmermann <tzimmermann@...e.de>
To:     Keith Zhao <keith.zhao@...rfivetech.com>,
        dri-devel@...ts.freedesktop.org, devicetree@...r.kernel.org,
        linux-kernel@...r.kernel.org, linux-riscv@...ts.infradead.org,
        linux-media@...r.kernel.org, linaro-mm-sig@...ts.linaro.org
Cc:     David Airlie <airlied@...il.com>, Daniel Vetter <daniel@...ll.ch>,
        Rob Herring <robh+dt@...nel.org>,
        Krzysztof Kozlowski <krzysztof.kozlowski+dt@...aro.org>,
        Conor Dooley <conor+dt@...nel.org>,
        Emil Renner Berthing <kernel@...il.dk>,
        Paul Walmsley <paul.walmsley@...ive.com>,
        Palmer Dabbelt <palmer@...belt.com>,
        Albert Ou <aou@...s.berkeley.edu>,
        Maarten Lankhorst <maarten.lankhorst@...ux.intel.com>,
        Maxime Ripard <mripard@...nel.org>,
        Philipp Zabel <p.zabel@...gutronix.de>,
        Sumit Semwal <sumit.semwal@...aro.org>,
        christian.koenig@....com, Bjorn Andersson <andersson@...nel.org>,
        Heiko Stuebner <heiko@...ech.de>,
        Shawn Guo <shawnguo@...nel.org>, Jagan Teki <jagan@...eble.ai>,
        Chris Morgan <macromorgan@...mail.com>,
        Jack Zhu <jack.zhu@...rfivetech.com>,
        Shengyang Chen <shengyang.chen@...rfivetech.com>,
        Changhuang Liang <changhuang.liang@...rfivetech.com>
Subject: Re: [PATCH 4/9] drm/verisilicon: Add gem driver for JH7110 SoC



Am 02.06.23 um 09:40 schrieb Keith Zhao:
> This patch implements gem related APIs for JH7100 SoC.
> 
> Signed-off-by: Keith Zhao <keith.zhao@...rfivetech.com>
> ---
>   drivers/gpu/drm/verisilicon/Makefile |   3 +-
>   drivers/gpu/drm/verisilicon/vs_drv.c |   6 +
>   drivers/gpu/drm/verisilicon/vs_gem.c | 372 +++++++++++++++++++++++++++
>   drivers/gpu/drm/verisilicon/vs_gem.h |  72 ++++++
>   4 files changed, 452 insertions(+), 1 deletion(-)
>   create mode 100644 drivers/gpu/drm/verisilicon/vs_gem.c
>   create mode 100644 drivers/gpu/drm/verisilicon/vs_gem.h
> 
> diff --git a/drivers/gpu/drm/verisilicon/Makefile b/drivers/gpu/drm/verisilicon/Makefile
> index 64ce1b26546c..30360e370e47 100644
> --- a/drivers/gpu/drm/verisilicon/Makefile
> +++ b/drivers/gpu/drm/verisilicon/Makefile
> @@ -1,6 +1,7 @@
>   # SPDX-License-Identifier: GPL-2.0
>   
> -vs_drm-objs := vs_drv.o
> +vs_drm-objs := vs_drv.o \
> +		vs_gem.o
>   
>   obj-$(CONFIG_DRM_VERISILICON) += vs_drm.o
>   
> diff --git a/drivers/gpu/drm/verisilicon/vs_drv.c b/drivers/gpu/drm/verisilicon/vs_drv.c
> index 24d333598477..e0a2fc43b55f 100644
> --- a/drivers/gpu/drm/verisilicon/vs_drv.c
> +++ b/drivers/gpu/drm/verisilicon/vs_drv.c
> @@ -30,6 +30,7 @@
>   #include <drm/drm_vblank.h>
>   
>   #include "vs_drv.h"
> +#include "vs_gem.h"
>   
>   #define DRV_NAME	"starfive"
>   #define DRV_DESC	"Starfive DRM driver"
> @@ -47,6 +48,7 @@ static const struct file_operations fops = {
>   	.compat_ioctl	= drm_compat_ioctl,
>   	.poll			= drm_poll,
>   	.read			= drm_read,
> +	.mmap			= vs_gem_mmap,
>   };
>   
>   static struct drm_driver vs_drm_driver = {
> @@ -54,6 +56,10 @@ static struct drm_driver vs_drm_driver = {
>   	.lastclose		= drm_fb_helper_lastclose,
>   	.prime_handle_to_fd = drm_gem_prime_handle_to_fd,
>   	.prime_fd_to_handle = drm_gem_prime_fd_to_handle,
> +	.gem_prime_import	= vs_gem_prime_import,
> +	.gem_prime_import_sg_table = vs_gem_prime_import_sg_table,
> +	.gem_prime_mmap		= vs_gem_prime_mmap,
> +	.dumb_create		= vs_gem_dumb_create,
>   	.fops			= &fops,
>   	.name			= DRV_NAME,
>   	.desc			= DRV_DESC,
> diff --git a/drivers/gpu/drm/verisilicon/vs_gem.c b/drivers/gpu/drm/verisilicon/vs_gem.c
> new file mode 100644
> index 000000000000..3f963471c1ab
> --- /dev/null
> +++ b/drivers/gpu/drm/verisilicon/vs_gem.c
> @@ -0,0 +1,372 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/*
> + * Copyright (C) 2023 VeriSilicon Holdings Co., Ltd.
> + */
> +
> +#include <linux/dma-buf.h>
> +#include <linux/of_reserved_mem.h>
> +#include <drm/drm_gem_dma_helper.h>
> +
> +#include "vs_drv.h"
> +#include "vs_gem.h"
> +
> +static const struct drm_gem_object_funcs vs_gem_default_funcs;
> +
> +static int vs_gem_alloc_buf(struct vs_gem_object *vs_obj)
> +{
> +	struct drm_device *dev = vs_obj->base.dev;
> +	unsigned int nr_pages;
> +	struct sg_table sgt;
> +	int ret = -ENOMEM;
> +
> +	if (vs_obj->dma_addr) {
> +		DRM_DEV_DEBUG_KMS(dev->dev, "already allocated.\n");
> +		return 0;
> +	}
> +
> +	vs_obj->dma_attrs = DMA_ATTR_WRITE_COMBINE | DMA_ATTR_FORCE_CONTIGUOUS
> +			   | DMA_ATTR_NO_KERNEL_MAPPING;
> +
> +	nr_pages = vs_obj->size >> PAGE_SHIFT;
> +
> +	vs_obj->pages = kvmalloc_array(nr_pages, sizeof(struct page *),
> +				       GFP_KERNEL | __GFP_ZERO);
> +	if (!vs_obj->pages) {
> +		DRM_DEV_ERROR(dev->dev, "failed to allocate pages.\n");
> +		return -ENOMEM;
> +	}
> +
> +	vs_obj->cookie = dma_alloc_attrs(to_dma_dev(dev), vs_obj->size,
> +					 &vs_obj->dma_addr, GFP_KERNEL,
> +					 vs_obj->dma_attrs);
> +
> +	if (!vs_obj->cookie) {
> +		DRM_DEV_ERROR(dev->dev, "failed to allocate buffer.\n");
> +		goto err_free;
> +	}
> +
> +	vs_obj->iova = vs_obj->dma_addr;
> +
> +	ret = dma_get_sgtable_attrs(to_dma_dev(dev), &sgt,
> +				    vs_obj->cookie, vs_obj->dma_addr,
> +				    vs_obj->size, vs_obj->dma_attrs);
> +	if (ret < 0) {
> +		DRM_DEV_ERROR(dev->dev, "failed to get sgtable.\n");
> +		goto err_mem_free;
> +	}
> +
> +	if (drm_prime_sg_to_page_array(&sgt, vs_obj->pages, nr_pages)) {
> +		DRM_DEV_ERROR(dev->dev, "invalid sgtable.\n");
> +		ret = -EINVAL;
> +		goto err_sgt_free;
> +	}
> +
> +	sg_free_table(&sgt);
> +
> +	return 0;
> +
> +err_sgt_free:
> +	sg_free_table(&sgt);
> +err_mem_free:
> +		dma_free_attrs(to_dma_dev(dev), vs_obj->size, vs_obj->cookie,
> +			       vs_obj->dma_addr, vs_obj->dma_attrs);
> +err_free:
> +	kvfree(vs_obj->pages);
> +
> +	return ret;
> +}
> +
> +static void vs_gem_free_buf(struct vs_gem_object *vs_obj)
> +{
> +	struct drm_device *dev = vs_obj->base.dev;
> +
> +	if (!vs_obj->dma_addr) {
> +		DRM_DEV_DEBUG_KMS(dev->dev, "dma_addr is invalid.\n");
> +		return;
> +	}
> +
> +	dma_free_attrs(to_dma_dev(dev), vs_obj->size, vs_obj->cookie,
> +		       (dma_addr_t)vs_obj->dma_addr,
> +		       vs_obj->dma_attrs);
> +
> +	kvfree(vs_obj->pages);
> +}
> +
> +static void vs_gem_free_object(struct drm_gem_object *obj)
> +{
> +	struct vs_gem_object *vs_obj = to_vs_gem_object(obj);
> +
> +	if (obj->import_attach)
> +		drm_prime_gem_destroy(obj, vs_obj->sgt);
> +	else
> +		vs_gem_free_buf(vs_obj);
> +
> +	drm_gem_object_release(obj);
> +
> +	kfree(vs_obj);
> +}
> +
> +static struct vs_gem_object *vs_gem_alloc_object(struct drm_device *dev,
> +						 size_t size)
> +{
> +	struct vs_gem_object *vs_obj;
> +	struct drm_gem_object *obj;
> +	int ret;
> +
> +	vs_obj = kzalloc(sizeof(*vs_obj), GFP_KERNEL);
> +	if (!vs_obj)
> +		return ERR_PTR(-ENOMEM);
> +
> +	vs_obj->size = size;
> +	obj = &vs_obj->base;
> +
> +	ret = drm_gem_object_init(dev, obj, size);
> +	if (ret)
> +		goto err_free;
> +
> +	vs_obj->base.funcs = &vs_gem_default_funcs;
> +
> +	ret = drm_gem_create_mmap_offset(obj);
> +	if (ret) {
> +		drm_gem_object_release(obj);
> +		goto err_free;
> +	}
> +
> +	return vs_obj;
> +
> +err_free:
> +	kfree(vs_obj);
> +	return ERR_PTR(ret);
> +}
> +
> +struct vs_gem_object *vs_gem_create_object(struct drm_device *dev,
> +					   size_t size)
> +{
> +	struct vs_gem_object *vs_obj;
> +	int ret;
> +
> +	size = PAGE_ALIGN(size);
> +
> +	vs_obj = vs_gem_alloc_object(dev, size);
> +	if (IS_ERR(vs_obj))
> +		return vs_obj;
> +
> +	ret = vs_gem_alloc_buf(vs_obj);
> +	if (ret) {
> +		drm_gem_object_release(&vs_obj->base);
> +		kfree(vs_obj);
> +		return ERR_PTR(ret);
> +	}
> +
> +	return vs_obj;
> +}
> +
> +static struct vs_gem_object *vs_gem_create_with_handle(struct drm_device *dev,
> +						       struct drm_file *file,
> +						       size_t size,
> +						       unsigned int *handle)
> +{
> +	struct vs_gem_object *vs_obj;
> +	struct drm_gem_object *obj;
> +	int ret;
> +
> +	vs_obj = vs_gem_create_object(dev, size);
> +	if (IS_ERR(vs_obj))
> +		return vs_obj;
> +
> +	obj = &vs_obj->base;
> +
> +	ret = drm_gem_handle_create(file, obj, handle);
> +
> +	drm_gem_object_put(obj);
> +
> +	if (ret)
> +		return ERR_PTR(ret);
> +
> +	return vs_obj;
> +}
> +
> +static int vs_gem_mmap_obj(struct drm_gem_object *obj,
> +			   struct vm_area_struct *vma)
> +{
> +	struct vs_gem_object *vs_obj = to_vs_gem_object(obj);
> +	struct drm_device *drm_dev = vs_obj->base.dev;
> +	unsigned long vm_size;
> +	int ret = 0;
> +
> +	vm_size = vma->vm_end - vma->vm_start;
> +	if (vm_size > vs_obj->size)
> +		return -EINVAL;
> +
> +	vma->vm_pgoff = 0;
> +
> +	/*
> +	 * We allocated a struct page table for starfive_obj, so clear
> +	 * VM_PFNMAP flag that was set by drm_gem_mmap_obj()/drm_gem_mmap().
> +	 */
> +	vm_flags_mod(vma, VM_IO | VM_DONTEXPAND | VM_DONTDUMP, VM_PFNMAP);
> +
> +	vma->vm_page_prot = pgprot_writecombine(vm_get_page_prot(vma->vm_flags));
> +	vma->vm_page_prot = pgprot_decrypted(vma->vm_page_prot);
> +
> +	ret = dma_mmap_attrs(to_dma_dev(drm_dev), vma, vs_obj->cookie,
> +			     vs_obj->dma_addr, vs_obj->size,
> +			     vs_obj->dma_attrs);
> +
> +	if (ret)
> +		drm_gem_vm_close(vma);
> +
> +	return ret;
> +}
> +
> +struct sg_table *vs_gem_prime_get_sg_table(struct drm_gem_object *obj)
> +{
> +	struct vs_gem_object *vs_obj = to_vs_gem_object(obj);
> +
> +	return drm_prime_pages_to_sg(obj->dev, vs_obj->pages,
> +					 vs_obj->size >> PAGE_SHIFT);
> +}
> +
> +int vs_gem_prime_vmap(struct drm_gem_object *obj, struct iosys_map *map)
> +{
> +	struct vs_gem_object *vs_obj = to_vs_gem_object(obj);
> +
> +	void *vaddr = vs_obj->dma_attrs & DMA_ATTR_NO_KERNEL_MAPPING ?
> +		       page_address(vs_obj->cookie) : vs_obj->cookie;
> +
> +	iosys_map_set_vaddr(map, vaddr);
> +
> +	return 0;
> +}
> +
> +void vs_gem_prime_vunmap(struct drm_gem_object *obj, struct iosys_map *map)
> +{
> +	/* Nothing to do */
> +}
> +
> +static const struct vm_operations_struct vs_vm_ops = {
> +	.open  = drm_gem_vm_open,
> +	.close = drm_gem_vm_close,
> +};
> +
> +static const struct drm_gem_object_funcs vs_gem_default_funcs = {
> +	.free = vs_gem_free_object,
> +	.get_sg_table = vs_gem_prime_get_sg_table,
> +	.vmap = vs_gem_prime_vmap,
> +	.vunmap = vs_gem_prime_vunmap,
> +	.vm_ops = &vs_vm_ops,
> +};
> +
> +int vs_gem_dumb_create(struct drm_file *file,
> +		       struct drm_device *dev,
> +		       struct drm_mode_create_dumb *args)
> +{
> +	struct vs_drm_private *priv = dev->dev_private;
> +	struct vs_gem_object *vs_obj;
> +	unsigned int pitch = DIV_ROUND_UP(args->width * args->bpp, 8);
> +
> +	if (args->bpp % 10)
> +		args->pitch = ALIGN(pitch, priv->pitch_alignment);
> +	else
> +		/* for costum 10bit format with no bit gaps */
> +		args->pitch = pitch;
> +	args->size = PAGE_ALIGN(args->pitch * args->height);
> +	vs_obj = vs_gem_create_with_handle(dev, file, args->size,
> +					   &args->handle);
> +	return PTR_ERR_OR_ZERO(vs_obj);
> +}
> +
> +struct drm_gem_object *vs_gem_prime_import(struct drm_device *dev,
> +					   struct dma_buf *dma_buf)
> +{
> +	return drm_gem_prime_import_dev(dev, dma_buf, to_dma_dev(dev));
> +}
> +
> +struct drm_gem_object *
> +vs_gem_prime_import_sg_table(struct drm_device *dev,
> +			     struct dma_buf_attachment *attach,
> +			     struct sg_table *sgt)
> +{
> +	struct vs_gem_object *vs_obj;
> +	int npages;
> +	int ret;
> +	struct scatterlist *s;
> +	u32 i;
> +	dma_addr_t expected;
> +	size_t size = attach->dmabuf->size;
> +
> +	size = PAGE_ALIGN(size);
> +
> +	vs_obj = vs_gem_alloc_object(dev, size);
> +	if (IS_ERR(vs_obj))
> +		return ERR_CAST(vs_obj);
> +
> +	expected = sg_dma_address(sgt->sgl);
> +	for_each_sg(sgt->sgl, s, sgt->nents, i) {
> +		if (sg_dma_address(s) != expected) {
> +			DRM_ERROR("sg_table is not contiguous");
> +			ret = -EINVAL;
> +			goto err;
> +		}
> +		if (sg_dma_len(s) & (PAGE_SIZE - 1)) {
> +			ret = -EINVAL;
> +			goto err;
> +		}
> +		if (i == 0)
> +			vs_obj->iova = sg_dma_address(s);
> +		expected = sg_dma_address(s) + sg_dma_len(s);
> +	}
> +
> +	vs_obj->dma_addr = sg_dma_address(sgt->sgl);
> +
> +	npages = vs_obj->size >> PAGE_SHIFT;
> +	vs_obj->pages = kvmalloc_array(npages, sizeof(struct page *),
> +				       GFP_KERNEL);
> +	if (!vs_obj->pages) {
> +		ret = -ENOMEM;
> +		goto err;
> +	}
> +
> +	ret = drm_prime_sg_to_page_array(sgt, vs_obj->pages, npages);
> +	if (ret)
> +		goto err_free_page;
> +
> +	vs_obj->sgt = sgt;
> +
> +	return &vs_obj->base;
> +
> +err_free_page:
> +	kvfree(vs_obj->pages);
> +err:
> +	vs_gem_free_object(&vs_obj->base);
> +
> +	return ERR_PTR(ret);
> +}
> +
> +int vs_gem_prime_mmap(struct drm_gem_object *obj, struct vm_area_struct *vma)
> +{
> +	int ret = 0;
> +
> +	ret = drm_gem_mmap_obj(obj, obj->size, vma);
> +	if (ret < 0)
> +		return ret;
> +
> +	return vs_gem_mmap_obj(obj, vma);
> +}
> +
> +int vs_gem_mmap(struct file *filp, struct vm_area_struct *vma)

This function needs to go away.

> +{
> +	struct drm_gem_object *obj;
> +	int ret;
> +
> +	ret = drm_gem_mmap(filp, vma);

Set drm_gem_mmap() as your fops.mmap callback.

> +	if (ret)
> +		return ret;
> +
> +	obj = vma->vm_private_data;
> +
> +	if (obj->import_attach)
> +		return dma_buf_mmap(obj->dma_buf, vma, 0);
> +
> +	return vs_gem_mmap_obj(obj, vma);

Both, dma_buf and regular objects, should be handled in struct 
drm_gem_object_funcs.mmap. drm_gem_mmap() will call it for you. 
vs_gem_mmap_obj() can then be removed.

you can find example code in drm_gem_shmem_helper.c and 
drm_gem_dma_helper.c on how to write GEM object's mmap function.


> +}
> diff --git a/drivers/gpu/drm/verisilicon/vs_gem.h b/drivers/gpu/drm/verisilicon/vs_gem.h
> new file mode 100644
> index 000000000000..3a6d7452cb06
> --- /dev/null
> +++ b/drivers/gpu/drm/verisilicon/vs_gem.h
> @@ -0,0 +1,72 @@
> +/* SPDX-License-Identifier: GPL-2.0 */
> +/*
> + * Copyright (C) 2023 VeriSilicon Holdings Co., Ltd.
> + */
> +
> +#ifndef __VS_GEM_H__
> +#define __VS_GEM_H__
> +
> +#include <linux/dma-buf.h>
> +
> +#include <drm/drm_gem.h>
> +#include <drm/drm_prime.h>
> +
> +#include "vs_drv.h"
> +/*
> + *
> + * @base: drm gem object.
> + * @size: size requested from user
> + * @cookie: cookie returned by dma_alloc_attrs
> + *	- not kernel virtual address with DMA_ATTR_NO_KERNEL_MAPPING
> + * @dma_addr: bus address(accessed by dma) to allocated memory region.
> + *	- this address could be physical address without IOMMU and
> + *	device address with IOMMU.
> + * @dma_attrs: attribute for DMA API
> + * @get_pages: flag for manually applying for non-contiguous memory.
> + * @pages: Array of backing pages.
> + * @sgt: Imported sg_table.
> + *
> + */
> +struct vs_gem_object {
> +	struct drm_gem_object	base;
> +	size_t			size;
> +	void			*cookie;
> +	dma_addr_t		dma_addr;
> +	u32				iova;
> +	unsigned long	dma_attrs;
> +	bool			get_pages;
> +	struct page		**pages;
> +	struct sg_table *sgt;
> +};
> +
> +static inline
> +struct vs_gem_object *to_vs_gem_object(struct drm_gem_object *obj)
> +{
> +	return container_of(obj, struct vs_gem_object, base);
> +}
> +
> +struct vs_gem_object *vs_gem_create_object(struct drm_device *dev,
> +					   size_t size);
> +
> +int vs_gem_prime_vmap(struct drm_gem_object *obj, struct iosys_map *map);
> +void vs_gem_prime_vunmap(struct drm_gem_object *obj, struct iosys_map *map);



> +
> +int vs_gem_prime_mmap(struct drm_gem_object *obj,
> +		      struct vm_area_struct *vma);
> +
> +int vs_gem_dumb_create(struct drm_file *file_priv,
> +		       struct drm_device *drm,
> +		       struct drm_mode_create_dumb *args);
> +
> +int vs_gem_mmap(struct file *filp, struct vm_area_struct *vma);
> +
> +struct sg_table *vs_gem_prime_get_sg_table(struct drm_gem_object *obj);
> +
> +struct drm_gem_object *vs_gem_prime_import(struct drm_device *dev,
> +					   struct dma_buf *dma_buf);
> +struct drm_gem_object *
> +vs_gem_prime_import_sg_table(struct drm_device *dev,
> +			     struct dma_buf_attachment *attach,
> +			     struct sg_table *sgt);
> +
> +#endif /* __VS_GEM_H__ */

-- 
Thomas Zimmermann
Graphics Driver Developer
SUSE Software Solutions Germany GmbH
Frankenstrasse 146, 90461 Nuernberg, Germany
GF: Ivo Totev, Andrew Myers, Andrew McDonald, Boudien Moerman
HRB 36809 (AG Nuernberg)

Download attachment "OpenPGP_signature" of type "application/pgp-signature" (841 bytes)

Powered by blists - more mailing lists