[PATCH 1/4] drm/rockchip: Do not use DMA mapping API if attached to IOMMU domain
Mark yao
mark.yao at rock-chips.com
Tue Feb 7 08:03:02 UTC 2017
On 2017年02月07日 14:53, Tomasz Figa wrote:
> Hi Mark,
>
> Thanks for reviving this series and sorry for not taking care of it
> myself. Please see some comments inline.
Hi Tomasz
Thanks for review,
I will add the patches you mentioned into v2 version.
>
> On Tue, Feb 7, 2017 at 3:09 PM, Mark Yao <mark.yao at rock-chips.com> wrote:
>> From: Tomasz Figa <tfiga at chromium.org>
>>
>> The API is not suitable for subsystems consisting of multiple devices
>> and requires severe hacks to use it. To mitigate this, this patch
>> implements allocation and address space management locally by using
>> helpers provided by DRM framework, like other DRM drivers do, e.g.
>> Tegra.
>>
>> This patch should not introduce any functional changes until the driver
>> is made to attach subdevices into an IOMMU domain with the generic IOMMU
>> API, which will happen in following patch. Based heavily on GEM
>> implementation of Tegra DRM driver.
>>
>> Signed-off-by: Tomasz Figa <tfiga at chromium.org>
>> Signed-off-by: Shunqian Zheng <zhengsq at rock-chips.com>
>> Acked-by: Mark Yao <mark.yao at rock-chips.com>
> I believe you need your Signed-off-by here.
>
>> ---
>> drivers/gpu/drm/rockchip/rockchip_drm_drv.h | 4 +-
>> drivers/gpu/drm/rockchip/rockchip_drm_gem.c | 217 ++++++++++++++++++++++++++--
>> drivers/gpu/drm/rockchip/rockchip_drm_gem.h | 8 +
>> 3 files changed, 219 insertions(+), 10 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/rockchip/rockchip_drm_drv.h b/drivers/gpu/drm/rockchip/rockchip_drm_drv.h
>> index fb6226c..7c123d9 100644
>> --- a/drivers/gpu/drm/rockchip/rockchip_drm_drv.h
>> +++ b/drivers/gpu/drm/rockchip/rockchip_drm_drv.h
>> @@ -30,6 +30,7 @@
>>
>> struct drm_device;
>> struct drm_connector;
>> +struct iommu_domain;
>>
>> /*
>> * Rockchip drm private crtc funcs.
>> @@ -60,7 +61,8 @@ struct rockchip_drm_private {
>> struct drm_gem_object *fbdev_bo;
>> const struct rockchip_crtc_funcs *crtc_funcs[ROCKCHIP_MAX_CRTC];
>> struct drm_atomic_state *state;
>> -
>> + struct iommu_domain *domain;
>> + struct drm_mm mm;
>> struct list_head psr_list;
>> spinlock_t psr_list_lock;
>> };
>> diff --git a/drivers/gpu/drm/rockchip/rockchip_drm_gem.c b/drivers/gpu/drm/rockchip/rockchip_drm_gem.c
>> index b70f942..5209392 100644
>> --- a/drivers/gpu/drm/rockchip/rockchip_drm_gem.c
>> +++ b/drivers/gpu/drm/rockchip/rockchip_drm_gem.c
>> @@ -16,11 +16,135 @@
>> #include <drm/drmP.h>
>> #include <drm/drm_gem.h>
>> #include <drm/drm_vma_manager.h>
>> +#include <linux/iommu.h>
>>
>> #include "rockchip_drm_drv.h"
>> #include "rockchip_drm_gem.h"
>>
>> -static int rockchip_gem_alloc_buf(struct rockchip_gem_object *rk_obj,
>> +static int rockchip_gem_iommu_map(struct rockchip_gem_object *rk_obj)
>> +{
>> + struct drm_device *drm = rk_obj->base.dev;
>> + struct rockchip_drm_private *private = drm->dev_private;
>> + int prot = IOMMU_READ | IOMMU_WRITE;
>> + ssize_t ret;
>> +
>> + ret = drm_mm_insert_node_generic(&private->mm, &rk_obj->mm,
>> + rk_obj->base.size, PAGE_SIZE,
>> + 0, 0);
>> + if (ret < 0) {
>> + DRM_ERROR("out of I/O virtual memory: %zd\n", ret);
>> + return ret;
>> + }
>> +
>> + rk_obj->dma_addr = rk_obj->mm.start;
>> +
>> + ret = iommu_map_sg(private->domain, rk_obj->dma_addr, rk_obj->sgt->sgl,
>> + rk_obj->sgt->nents, prot);
>> + if (ret < 0) {
>> + DRM_ERROR("failed to map buffer: %zd\n", ret);
>> + goto err_remove_node;
>> + }
>> +
>> + rk_obj->size = ret;
>> +
>> + return 0;
>> +
>> +err_remove_node:
>> + drm_mm_remove_node(&rk_obj->mm);
>> +
>> + return ret;
>> +}
>> +
>> +static int rockchip_gem_iommu_unmap(struct rockchip_gem_object *rk_obj)
>> +{
>> + struct drm_device *drm = rk_obj->base.dev;
>> + struct rockchip_drm_private *private = drm->dev_private;
>> +
>> + iommu_unmap(private->domain, rk_obj->dma_addr, rk_obj->size);
>> + drm_mm_remove_node(&rk_obj->mm);
>> +
>> + return 0;
>> +}
>> +
>> +static int rockchip_gem_get_pages(struct rockchip_gem_object *rk_obj)
>> +{
>> + struct drm_device *drm = rk_obj->base.dev;
>> + int ret, i;
>> + struct scatterlist *s;
>> +
>> + rk_obj->pages = drm_gem_get_pages(&rk_obj->base);
>> + if (IS_ERR(rk_obj->pages))
>> + return PTR_ERR(rk_obj->pages);
>> +
>> + rk_obj->num_pages = rk_obj->base.size >> PAGE_SHIFT;
>> +
>> + rk_obj->sgt = drm_prime_pages_to_sg(rk_obj->pages, rk_obj->num_pages);
>> + if (IS_ERR(rk_obj->sgt)) {
>> + ret = PTR_ERR(rk_obj->sgt);
>> + goto err_put_pages;
>> + }
>> +
>> + /*
>> + * Fake up the SG table so that dma_sync_sg_for_device() can be used
>> + * to flush the pages associated with it.
>> + *
>> + * TODO: Replace this by drm_clflush_sg() once it can be implemented
>> + * without relying on symbols that are not exported.
>> + */
>> + for_each_sg(rk_obj->sgt->sgl, s, rk_obj->sgt->nents, i)
>> + sg_dma_address(s) = sg_phys(s);
>> +
>> + dma_sync_sg_for_device(drm->dev, rk_obj->sgt->sgl, rk_obj->sgt->nents,
>> + DMA_TO_DEVICE);
>> +
>> + return 0;
>> +
>> +err_put_pages:
>> + drm_gem_put_pages(&rk_obj->base, rk_obj->pages, false, false);
>> + return ret;
>> +}
>> +
>> +static void rockchip_gem_put_pages(struct rockchip_gem_object *rk_obj)
>> +{
>> + sg_free_table(rk_obj->sgt);
>> + kfree(rk_obj->sgt);
>> + drm_gem_put_pages(&rk_obj->base, rk_obj->pages, false, false);
> We need to call drm_gem_put_pages with two last arguments true, not
> false, because we don't have any way of tracking the accesses, so we
> need to assume the page was both accessed and dirty. See
> https://chromium-review.googlesource.com/382934.
>
>> +}
>> +
>> +static int rockchip_gem_alloc_iommu(struct rockchip_gem_object *rk_obj,
>> + bool alloc_kmap)
>> +{
>> + int ret;
>> +
>> + ret = rockchip_gem_get_pages(rk_obj);
>> + if (ret < 0)
>> + return ret;
>> +
>> + ret = rockchip_gem_iommu_map(rk_obj);
>> + if (ret < 0)
>> + goto err_free;
>> +
>> + if (alloc_kmap) {
>> + rk_obj->kvaddr = vmap(rk_obj->pages, rk_obj->num_pages, VM_MAP,
>> + pgprot_writecombine(PAGE_KERNEL));
>> + if (!rk_obj->kvaddr) {
>> + DRM_ERROR("failed to vmap() buffer\n");
>> + ret = -ENOMEM;
>> + goto err_unmap;
>> + }
>> + }
>> +
>> + return 0;
>> +
>> +err_unmap:
>> + rockchip_gem_iommu_unmap(rk_obj);
>> +err_free:
>> + rockchip_gem_put_pages(rk_obj);
>> +
>> + return ret;
>> +}
>> +
>> +static int rockchip_gem_alloc_dma(struct rockchip_gem_object *rk_obj,
>> bool alloc_kmap)
>> {
>> struct drm_gem_object *obj = &rk_obj->base;
>> @@ -42,7 +166,27 @@ static int rockchip_gem_alloc_buf(struct rockchip_gem_object *rk_obj,
>> return 0;
>> }
>>
>> -static void rockchip_gem_free_buf(struct rockchip_gem_object *rk_obj)
>> +static int rockchip_gem_alloc_buf(struct rockchip_gem_object *rk_obj,
>> + bool alloc_kmap)
>> +{
>> + struct drm_gem_object *obj = &rk_obj->base;
>> + struct drm_device *drm = obj->dev;
>> + struct rockchip_drm_private *private = drm->dev_private;
>> +
>> + if (private->domain)
>> + return rockchip_gem_alloc_iommu(rk_obj, alloc_kmap);
>> + else
>> + return rockchip_gem_alloc_dma(rk_obj, alloc_kmap);
>> +}
>> +
>> +static void rockchip_gem_free_iommu(struct rockchip_gem_object *rk_obj)
>> +{
>> + vunmap(rk_obj->kvaddr);
>> + rockchip_gem_iommu_unmap(rk_obj);
>> + rockchip_gem_put_pages(rk_obj);
>> +}
>> +
>> +static void rockchip_gem_free_dma(struct rockchip_gem_object *rk_obj)
>> {
>> struct drm_gem_object *obj = &rk_obj->base;
>> struct drm_device *drm = obj->dev;
>> @@ -51,23 +195,64 @@ static void rockchip_gem_free_buf(struct rockchip_gem_object *rk_obj)
>> rk_obj->dma_attrs);
>> }
>>
>> -static int rockchip_drm_gem_object_mmap(struct drm_gem_object *obj,
>> - struct vm_area_struct *vma)
>> +static void rockchip_gem_free_buf(struct rockchip_gem_object *rk_obj)
>> +{
>> + if (rk_obj->pages)
>> + rockchip_gem_free_iommu(rk_obj);
>> + else
>> + rockchip_gem_free_dma(rk_obj);
>> +}
>>
>> +static int rockchip_drm_gem_object_mmap_iommu(struct drm_gem_object *obj,
>> + struct vm_area_struct *vma)
>> {
>> + struct rockchip_gem_object *rk_obj = to_rockchip_obj(obj);
>> + unsigned int i, count = obj->size >> PAGE_SHIFT;
>> + unsigned long user_count = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
>> + unsigned long uaddr = vma->vm_start;
>> int ret;
>> +
>> + if (user_count == 0 || user_count > count)
>> + return -ENXIO;
>> +
>> + for (i = 0; i < user_count; i++) {
> Even though DRM mmap doesn't support page offset, we need to respect
> it here for PRIME/DMA-buf mmap. See
> https://chromium-review.googlesource.com/386477.
>
>> + ret = vm_insert_page(vma, uaddr, rk_obj->pages[i]);
>> + if (ret)
>> + return ret;
>> + uaddr += PAGE_SIZE;
>> + }
>> +
>> + return 0;
>> +}
>> +
>> +static int rockchip_drm_gem_object_mmap_dma(struct drm_gem_object *obj,
>> + struct vm_area_struct *vma)
>> +{
>> struct rockchip_gem_object *rk_obj = to_rockchip_obj(obj);
>> struct drm_device *drm = obj->dev;
>>
>> + return dma_mmap_attrs(drm->dev, vma, rk_obj->kvaddr, rk_obj->dma_addr,
>> + obj->size, rk_obj->dma_attrs);
>> +}
>> +
>> +static int rockchip_drm_gem_object_mmap(struct drm_gem_object *obj,
>> + struct vm_area_struct *vma)
>> +{
>> + int ret;
>> + struct rockchip_gem_object *rk_obj = to_rockchip_obj(obj);
>> +
>> /*
>> - * dma_alloc_attrs() allocated a struct page table for rk_obj, so clear
>> + * We allocated a struct page table for rk_obj, so clear
>> * VM_PFNMAP flag that was set by drm_gem_mmap_obj()/drm_gem_mmap().
>> */
>> vma->vm_flags &= ~VM_PFNMAP;
>> vma->vm_pgoff = 0;
>>
>> - ret = dma_mmap_attrs(drm->dev, vma, rk_obj->kvaddr, rk_obj->dma_addr,
>> - obj->size, rk_obj->dma_attrs);
>> + if (rk_obj->pages)
>> + ret = rockchip_drm_gem_object_mmap_iommu(obj, vma);
>> + else
>> + ret = rockchip_drm_gem_object_mmap_dma(obj, vma);
>> +
>> if (ret)
>> drm_gem_vm_close(vma);
>>
>> @@ -117,7 +302,7 @@ struct rockchip_gem_object *
>>
>> obj = &rk_obj->base;
>>
>> - drm_gem_private_object_init(drm, obj, size);
>> + drm_gem_object_init(drm, obj, size);
> In error path of this function and in all functions that destroy the
> GEM we need to call drm_gem_object_release(), as it's the counterpart
> to drm_gem_object_init(). See
> https://chromium-review.googlesource.com/385456.
>
> Best regards,
> Tomasz
>
>
>
--
Ｍark Yao
More information about the dri-devel
mailing list