[PATCH v6 2/9] drm/hisilicon/hibmc: Add video memory management

Rongrong Zou zourongrong at huawei.com
Fri Nov 11 05:57:56 PST 2016


在 2016/11/11 21:25, Sean Paul 写道:
> On Fri, Nov 11, 2016 at 6:16 AM, Rongrong Zou <zourongrong at huawei.com> wrote:
>> 在 2016/11/11 1:35, Sean Paul 写道:
>>>
>>> On Fri, Oct 28, 2016 at 3:27 AM, Rongrong Zou <zourongrong at gmail.com>
>>> wrote:
>>>>
>>>> Hibmc have 32m video memory which can be accessed through PCIe by host,
>>>> we use ttm to manage these memory.
>>>>
>>>> Signed-off-by: Rongrong Zou <zourongrong at gmail.com>
>>>> ---
>>>>    drivers/gpu/drm/hisilicon/hibmc/Kconfig         |   1 +
>>>>    drivers/gpu/drm/hisilicon/hibmc/Makefile        |   2 +-
>>>>    drivers/gpu/drm/hisilicon/hibmc/hibmc_drm_drv.c |  12 +
>>>>    drivers/gpu/drm/hisilicon/hibmc/hibmc_drm_drv.h |  46 +++
>>>>    drivers/gpu/drm/hisilicon/hibmc/hibmc_ttm.c     | 490
>>>> ++++++++++++++++++++++++
>>>>    5 files changed, 550 insertions(+), 1 deletion(-)
>>>>    create mode 100644 drivers/gpu/drm/hisilicon/hibmc/hibmc_ttm.c
>>>>
>>>> diff --git a/drivers/gpu/drm/hisilicon/hibmc/Kconfig
>>>> b/drivers/gpu/drm/hisilicon/hibmc/Kconfig
>>>> index a9af90d..bcb8c18 100644
>>>> --- a/drivers/gpu/drm/hisilicon/hibmc/Kconfig
>>>> +++ b/drivers/gpu/drm/hisilicon/hibmc/Kconfig
>>>> @@ -1,6 +1,7 @@
>>>>    config DRM_HISI_HIBMC
>>>>           tristate "DRM Support for Hisilicon Hibmc"
>>>>           depends on DRM && PCI
>>>> +       select DRM_TTM
>>>>
>>>>           help
>>>>             Choose this option if you have a Hisilicon Hibmc soc chipset.
>>>> diff --git a/drivers/gpu/drm/hisilicon/hibmc/Makefile
>>>> b/drivers/gpu/drm/hisilicon/hibmc/Makefile
>>>> index 97cf4a0..d5c40b8 100644
>>>> --- a/drivers/gpu/drm/hisilicon/hibmc/Makefile
>>>> +++ b/drivers/gpu/drm/hisilicon/hibmc/Makefile
>>>> @@ -1,5 +1,5 @@
>>>>    ccflags-y := -Iinclude/drm
>>>> -hibmc-drm-y := hibmc_drm_drv.o hibmc_drm_power.o
>>>> +hibmc-drm-y := hibmc_drm_drv.o hibmc_drm_power.o hibmc_ttm.o
>>>>
>>>>    obj-$(CONFIG_DRM_HISI_HIBMC)   +=hibmc-drm.o
>>>>    #obj-y += hibmc-drm.o
>>>> diff --git a/drivers/gpu/drm/hisilicon/hibmc/hibmc_drm_drv.c
>>>> b/drivers/gpu/drm/hisilicon/hibmc/hibmc_drm_drv.c
>>>> index 4669d42..81f4301 100644
>>>> --- a/drivers/gpu/drm/hisilicon/hibmc/hibmc_drm_drv.c
>>>> +++ b/drivers/gpu/drm/hisilicon/hibmc/hibmc_drm_drv.c
>>>> @@ -31,6 +31,7 @@
>>>>    #ifdef CONFIG_COMPAT
>>>>           .compat_ioctl   = drm_compat_ioctl,
>>>>    #endif
>>>> +       .mmap           = hibmc_mmap,
>>>>           .poll           = drm_poll,
>>>>           .read           = drm_read,
>>>>           .llseek         = no_llseek,
>>>> @@ -46,6 +47,8 @@ static void hibmc_disable_vblank(struct drm_device
>>>> *dev, unsigned int pipe)
>>>>    }
>>>>
>>>>    static struct drm_driver hibmc_driver = {
>>>> +       .driver_features        = DRIVER_GEM,
>>>> +
>>>
>>>
>>> nit: extra space
>>>
>>>>           .fops                   = &hibmc_fops,
>>>>           .name                   = "hibmc",
>>>>           .date                   = "20160828",
>>>> @@ -55,6 +58,10 @@ static void hibmc_disable_vblank(struct drm_device
>>>> *dev, unsigned int pipe)
>>>>           .get_vblank_counter     = drm_vblank_no_hw_counter,
>>>>           .enable_vblank          = hibmc_enable_vblank,
>>>>           .disable_vblank         = hibmc_disable_vblank,
>>>> +       .gem_free_object_unlocked = hibmc_gem_free_object,
>>>> +       .dumb_create            = hibmc_dumb_create,
>>>> +       .dumb_map_offset        = hibmc_dumb_mmap_offset,
>>>> +       .dumb_destroy           = drm_gem_dumb_destroy,
>>>>    };
>>>>
>>>>    static int hibmc_pm_suspend(struct device *dev)
>>>> @@ -163,6 +170,7 @@ static int hibmc_unload(struct drm_device *dev)
>>>>    {
>>>>           struct hibmc_drm_device *hidev = dev->dev_private;
>>>>
>>>> +       hibmc_mm_fini(hidev);
>>>>           hibmc_hw_fini(hidev);
>>>>           dev->dev_private = NULL;
>>>>           return 0;
>>>> @@ -183,6 +191,10 @@ static int hibmc_load(struct drm_device *dev,
>>>> unsigned long flags)
>>>>           if (ret)
>>>>                   goto err;
>>>>
>>>> +       ret = hibmc_mm_init(hidev);
>>>> +       if (ret)
>>>> +               goto err;
>>>> +
>>>>           return 0;
>>>>
>>>>    err:
>>>> diff --git a/drivers/gpu/drm/hisilicon/hibmc/hibmc_drm_drv.h
>>>> b/drivers/gpu/drm/hisilicon/hibmc/hibmc_drm_drv.h
>>>> index 0037341..db8d80e 100644
>>>> --- a/drivers/gpu/drm/hisilicon/hibmc/hibmc_drm_drv.h
>>>> +++ b/drivers/gpu/drm/hisilicon/hibmc/hibmc_drm_drv.h
>>>> @@ -20,6 +20,8 @@
>>>>    #define HIBMC_DRM_DRV_H
>>>>
>>>>    #include <drm/drmP.h>
>>>> +#include <drm/ttm/ttm_bo_driver.h>
>>>> +#include <drm/drm_gem.h>
>>>
>>>
>>> nit: alphabetize
>>
>>
>> will fix it, thanks.
>>
>>>
>>>>
>>>>    struct hibmc_drm_device {
>>>>           /* hw */
>>>> @@ -30,6 +32,50 @@ struct hibmc_drm_device {
>>>>
>>>>           /* drm */
>>>>           struct drm_device  *dev;
>>>> +
>>>> +       /* ttm */
>>>> +       struct {
>>>> +               struct drm_global_reference mem_global_ref;
>>>> +               struct ttm_bo_global_ref bo_global_ref;
>>>> +               struct ttm_bo_device bdev;
>>>> +               bool initialized;
>>>> +       } ttm;
>>>
>>>
>>> I don't think you gain anything other than keystrokes from the substruct
>>
>>
>> I'm sorry i didn't catch you, i looked at the all drivers used ttm such
>> as ast/bochs/cirrus/mgag200/qxl/virtio_gpu, they all embedded the ttm
>> substruct
>> into the driver-private struct.
>>
>> so do you mean
>> struct hibmc_drm_device {
>>          /* hw */
>>          void __iomem   *mmio;
>>          void __iomem   *fb_map;
>>          unsigned long  fb_base;
>>          unsigned long  fb_size;
>>
>>          /* drm */
>>          struct drm_device  *dev;
>>          struct drm_plane plane;
>>          struct drm_crtc crtc;
>>          struct drm_encoder encoder;
>>          struct drm_connector connector;
>>          bool mode_config_initialized;
>>
>>          /* ttm */
>>          struct drm_global_reference mem_global_ref;
>>          struct ttm_bo_global_ref bo_global_ref;
>>          struct ttm_bo_device bdev;
>>          bool initialized;
>>          ...
>>          };
>> ?
>
> Yeah, that's what I was thinking
>
>>
>>>
>>>> +
>>>> +       bool mm_inited;
>>>>    };
>>>>
>>>> +struct hibmc_bo {
>>>> +       struct ttm_buffer_object bo;
>>>> +       struct ttm_placement placement;
>>>> +       struct ttm_bo_kmap_obj kmap;
>>>> +       struct drm_gem_object gem;
>>>> +       struct ttm_place placements[3];
>>>> +       int pin_count;
>>>> +};
>>>> +
>>>> +static inline struct hibmc_bo *hibmc_bo(struct ttm_buffer_object *bo)
>>>> +{
>>>> +       return container_of(bo, struct hibmc_bo, bo);
>>>> +}
>>>> +
>>>> +static inline struct hibmc_bo *gem_to_hibmc_bo(struct drm_gem_object
>>>> *gem)
>>>> +{
>>>> +       return container_of(gem, struct hibmc_bo, gem);
>>>> +}
>>>> +
>>>> +#define DRM_FILE_PAGE_OFFSET (0x100000000ULL >> PAGE_SHIFT)
>>>
>>>
>>> Hide this in ttm.c
>>
>>
>> ok, will do that.
>> thanks for pointing it out.
>>
>>
>>>
>>>> +
>>>> +int hibmc_gem_create(struct drm_device *dev, u32 size, bool iskernel,
>>>> +                    struct drm_gem_object **obj);
>>>> +
>>>> +int hibmc_mm_init(struct hibmc_drm_device *hibmc);
>>>> +void hibmc_mm_fini(struct hibmc_drm_device *hibmc);
>>>> +int hibmc_bo_pin(struct hibmc_bo *bo, u32 pl_flag, u64 *gpu_addr);
>>>> +void hibmc_gem_free_object(struct drm_gem_object *obj);
>>>> +int hibmc_dumb_create(struct drm_file *file, struct drm_device *dev,
>>>> +                     struct drm_mode_create_dumb *args);
>>>> +int hibmc_dumb_mmap_offset(struct drm_file *file, struct drm_device
>>>> *dev,
>>>> +                          u32 handle, u64 *offset);
>>>> +int hibmc_mmap(struct file *filp, struct vm_area_struct *vma);
>>>> +
>>>>    #endif
>>>> diff --git a/drivers/gpu/drm/hisilicon/hibmc/hibmc_ttm.c
>>>> b/drivers/gpu/drm/hisilicon/hibmc/hibmc_ttm.c
>>>> new file mode 100644
>>>> index 0000000..0802ebd
>>>> --- /dev/null
>>>> +++ b/drivers/gpu/drm/hisilicon/hibmc/hibmc_ttm.c
>>>> @@ -0,0 +1,490 @@
>>>> +/* Hisilicon Hibmc SoC drm driver
>>>> + *
>>>> + * Based on the bochs drm driver.
>>>> + *
>>>> + * Copyright (c) 2016 Huawei Limited.
>>>> + *
>>>> + * Author:
>>>> + *     Rongrong Zou <zourongrong at huawei.com>
>>>> + *     Rongrong Zou <zourongrong at gmail.com>
>>>> + *     Jianhua Li <lijianhua at huawei.com>
>>>> + *
>>>> + * This program is free software; you can redistribute it and/or modify
>>>> + * it under the terms of the GNU General Public License as published by
>>>> + * the Free Software Foundation; either version 2 of the License, or
>>>> + * (at your option) any later version.
>>>> + *
>>>> + */
>>>> +
>>>> +#include "hibmc_drm_drv.h"
>>>> +#include <ttm/ttm_page_alloc.h>
>>>> +#include <drm/drm_crtc_helper.h>
>>>> +#include <drm/drm_atomic_helper.h>
>>>> +
>>>> +static inline struct hibmc_drm_device *
>>>> +hibmc_bdev(struct ttm_bo_device *bd)
>>>> +{
>>>> +       return container_of(bd, struct hibmc_drm_device, ttm.bdev);
>>>> +}
>>>> +
>>>> +static int
>>>> +hibmc_ttm_mem_global_init(struct drm_global_reference *ref)
>>>> +{
>>>> +       return ttm_mem_global_init(ref->object);
>>>> +}
>>>> +
>>>> +static void
>>>> +hibmc_ttm_mem_global_release(struct drm_global_reference *ref)
>>>> +{
>>>> +       ttm_mem_global_release(ref->object);
>>>> +}
>>>> +
>>>> +static int hibmc_ttm_global_init(struct hibmc_drm_device *hibmc)
>>>> +{
>>>> +       struct drm_global_reference *global_ref;
>>>> +       int r;
>>>
>>>
>>> nit: try not to use one character variable names unless it's for the
>>> purpose of a loop (ie: i,j). You also use ret elsewhere in the driver,
>>> so it'd be nice to remain consistent
>>
>>
>> the whole file is delivered from bochs ttm, i didn't modify anything except
>> some checkpatch warnings and the 'hibmc_' prefix. Unfortunately, some
>> problems were delivered too.
>
> Yeah, seems like it. Perhaps you can post patches to fix these issues
> in the other drivers too :)

i will do after the this one get merged :)

>
>>
>>>
>>>> +
>>>> +       global_ref = &hibmc->ttm.mem_global_ref;
>>>
>>>
>>> I think using the global_ref local obfuscates what you're doing here.
>>> It saves you 6 characters while typing, but adds a layer of
>>> indirection for all future readers.
>>>
>>>> +       global_ref->global_type = DRM_GLOBAL_TTM_MEM;
>>>> +       global_ref->size = sizeof(struct ttm_mem_global);
>>>> +       global_ref->init = &hibmc_ttm_mem_global_init;
>>>> +       global_ref->release = &hibmc_ttm_mem_global_release;
>>>> +       r = drm_global_item_ref(global_ref);
>>>> +       if (r != 0) {
>>>
>>>
>>> nit: if (r)
>>
>>
>> will fix it,
>> thanks.
>> BTW, i wonder why checkpatch.pl didn't report it.
>>
>>
>>>
>>>> +               DRM_ERROR("Failed setting up TTM memory accounting
>>>> subsystem.\n"
>>>> +                        );
>>>
>>>
>>> Breaking up the line for one character is probably not worthwhile, and
>>> you should really print the error. How about:
>>>
>>> DRM_ERROR("Could not get ref on ttm global ret=%d.\n", ret);
>>
>>
>> i like your solution, thanks.
>>
>>
>>>
>>>
>>>> +               return r;
>>>> +       }
>>>> +
>>>> +       hibmc->ttm.bo_global_ref.mem_glob =
>>>> +               hibmc->ttm.mem_global_ref.object;
>>>> +       global_ref = &hibmc->ttm.bo_global_ref.ref;
>>>> +       global_ref->global_type = DRM_GLOBAL_TTM_BO;
>>>> +       global_ref->size = sizeof(struct ttm_bo_global);
>>>> +       global_ref->init = &ttm_bo_global_init;
>>>> +       global_ref->release = &ttm_bo_global_release;
>>>> +       r = drm_global_item_ref(global_ref);
>>>> +       if (r != 0) {
>>>> +               DRM_ERROR("Failed setting up TTM BO subsystem.\n");
>>>> +               drm_global_item_unref(&hibmc->ttm.mem_global_ref);
>>>> +               return r;
>>>> +       }
>>>> +       return 0;
>>>> +}
>>>> +
>>>> +static void
>>>> +hibmc_ttm_global_release(struct hibmc_drm_device *hibmc)
>>>> +{
>>>> +       if (!hibmc->ttm.mem_global_ref.release)
>>>
>>>
>>> Are you actually hitting this condition? This seems like it's papering
>>> over something else.
>>
>>
>> it was also delivered from others, i looked at the xxx_ttm_global_init
>> function, 'mem_global_ref.release' is assigned unconditionally, so i
>> think this condition never be hit, it may be hit when release twice,
>> but this won't take place in my driver.
>>
>
> Yeah, that's what I was hoping for. So perhaps we can remove this?

yes, we can.

Regards,
Rongrong.

>
>>>
>>>> +               return;
>>>> +
>>>> +       drm_global_item_unref(&hibmc->ttm.bo_global_ref.ref);
>>>> +       drm_global_item_unref(&hibmc->ttm.mem_global_ref);
>>>> +       hibmc->ttm.mem_global_ref.release = NULL;
>>>> +}
>>>> +
>>>> +static void hibmc_bo_ttm_destroy(struct ttm_buffer_object *tbo)
>>>> +{
>>>> +       struct hibmc_bo *bo;
>>>> +
>>>> +       bo = container_of(tbo, struct hibmc_bo, bo);
>>>
>>>
>>> nit: No need to split this into a separate line.
>>
>>
>> agreed, thanks.
>>
>>>
>>>> +
>>>> +       drm_gem_object_release(&bo->gem);
>>>> +       kfree(bo);
>>>> +}
>>>> +
>>>> +static bool hibmc_ttm_bo_is_hibmc_bo(struct ttm_buffer_object *bo)
>>>> +{
>>>> +       if (bo->destroy == &hibmc_bo_ttm_destroy)
>>>> +               return true;
>>>> +       return false;
>>>
>>>
>>> return bo->destroy == &hibmc_bo_ttm_destroy;
>>
>>
>> looks better to me.
>>
>>
>>>
>>>> +}
>>>> +
>>>> +static int
>>>> +hibmc_bo_init_mem_type(struct ttm_bo_device *bdev, u32 type,
>>>> +                      struct ttm_mem_type_manager *man)
>>>> +{
>>>> +       switch (type) {
>>>> +       case TTM_PL_SYSTEM:
>>>> +               man->flags = TTM_MEMTYPE_FLAG_MAPPABLE;
>>>> +               man->available_caching = TTM_PL_MASK_CACHING;
>>>> +               man->default_caching = TTM_PL_FLAG_CACHED;
>>>> +               break;
>>>> +       case TTM_PL_VRAM:
>>>> +               man->func = &ttm_bo_manager_func;
>>>> +               man->flags = TTM_MEMTYPE_FLAG_FIXED |
>>>> +                       TTM_MEMTYPE_FLAG_MAPPABLE;
>>>> +               man->available_caching = TTM_PL_FLAG_UNCACHED |
>>>> +                       TTM_PL_FLAG_WC;
>>>> +               man->default_caching = TTM_PL_FLAG_WC;
>>>> +               break;
>>>> +       default:
>>>> +               DRM_ERROR("Unsupported memory type %u\n", type);
>>>> +               return -EINVAL;
>>>> +       }
>>>> +       return 0;
>>>> +}
>>>> +
>>>> +void hibmc_ttm_placement(struct hibmc_bo *bo, int domain)
>>>> +{
>>>> +       u32 c = 0;
>>>
>>>
>>> Can you please use a more descriptive name than 'c'?
>>
>>
>> ok, will do that.
>>
>>>
>>>> +       u32 i;
>>>> +
>>>> +       bo->placement.placement = bo->placements;
>>>> +       bo->placement.busy_placement = bo->placements;
>>>> +       if (domain & TTM_PL_FLAG_VRAM)
>>>> +               bo->placements[c++].flags = TTM_PL_FLAG_WC |
>>>> +               TTM_PL_FLAG_UNCACHED | TTM_PL_FLAG_VRAM;
>>>
>>>
>>> nit: you're alignment is off here and below
>>
>>
>> is it correct?
>>
>>          if (domain & TTM_PL_FLAG_VRAM)
>>                  bo->placements[c++].flags = TTM_PL_FLAG_WC |
>>                          TTM_PL_FLAG_UNCACHED | TTM_PL_FLAG_VRAM;
>>          if (domain & TTM_PL_FLAG_SYSTEM)
>>                  bo->placements[c++].flags = TTM_PL_MASK_CACHING |
>>                          TTM_PL_FLAG_SYSTEM;
>>          if (!c)
>>                  bo->placements[c++].flags = TTM_PL_MASK_CACHING |
>>                          TTM_PL_FLAG_SYSTEM;
>>
>
> Pretty much anything other than lining them up one under the other is better
>
>>>
>>>> +       if (domain & TTM_PL_FLAG_SYSTEM)
>>>> +               bo->placements[c++].flags = TTM_PL_MASK_CACHING |
>>>> +               TTM_PL_FLAG_SYSTEM;
>>>> +       if (!c)
>>>> +               bo->placements[c++].flags = TTM_PL_MASK_CACHING |
>>>> +               TTM_PL_FLAG_SYSTEM;
>>>> +
>>>> +       bo->placement.num_placement = c;
>>>> +       bo->placement.num_busy_placement = c;
>>>> +       for (i = 0; i < c; ++i) {
>>>
>>>
>>> nit: we tend towards post-increment in kernel
>>
>>
>> agreed, thanks.
>>
>>
>>>
>>>> +               bo->placements[i].fpfn = 0;
>>>> +               bo->placements[i].lpfn = 0;
>>>> +       }
>>>> +}
>>>> +
>>>> +static void
>>>> +hibmc_bo_evict_flags(struct ttm_buffer_object *bo, struct ttm_placement
>>>> *pl)
>>>> +{
>>>> +       struct hibmc_bo *hibmcbo = hibmc_bo(bo);
>>>> +
>>>> +       if (!hibmc_ttm_bo_is_hibmc_bo(bo))
>>>> +               return;
>>>> +
>>>> +       hibmc_ttm_placement(hibmcbo, TTM_PL_FLAG_SYSTEM);
>>>> +       *pl = hibmcbo->placement;
>>>> +}
>>>> +
>>>> +static int hibmc_bo_verify_access(struct ttm_buffer_object *bo,
>>>> +                                 struct file *filp)
>>>> +{
>>>> +       struct hibmc_bo *hibmcbo = hibmc_bo(bo);
>>>> +
>>>> +       return drm_vma_node_verify_access(&hibmcbo->gem.vma_node,
>>>> +                                         filp->private_data);
>>>> +}
>>>> +
>>>> +static int hibmc_ttm_io_mem_reserve(struct ttm_bo_device *bdev,
>>>> +                                   struct ttm_mem_reg *mem)
>>>> +{
>>>> +       struct ttm_mem_type_manager *man = &bdev->man[mem->mem_type];
>>>> +       struct hibmc_drm_device *hibmc = hibmc_bdev(bdev);
>>>> +
>>>> +       mem->bus.addr = NULL;
>>>> +       mem->bus.offset = 0;
>>>> +       mem->bus.size = mem->num_pages << PAGE_SHIFT;
>>>> +       mem->bus.base = 0;
>>>> +       mem->bus.is_iomem = false;
>>>> +       if (!(man->flags & TTM_MEMTYPE_FLAG_MAPPABLE))
>>>> +               return -EINVAL;
>>>> +       switch (mem->mem_type) {
>>>> +       case TTM_PL_SYSTEM:
>>>> +               /* system memory */
>>>> +               return 0;
>>>> +       case TTM_PL_VRAM:
>>>> +               mem->bus.offset = mem->start << PAGE_SHIFT;
>>>> +               mem->bus.base = pci_resource_start(hibmc->dev->pdev, 0);
>>>> +               mem->bus.is_iomem = true;
>>>> +               break;
>>>> +       default:
>>>> +               return -EINVAL;
>>>> +       }
>>>> +       return 0;
>>>> +}
>>>> +
>>>> +static void hibmc_ttm_io_mem_free(struct ttm_bo_device *bdev,
>>>> +                                 struct ttm_mem_reg *mem)
>>>> +{
>>>> +}
>>>
>>>
>>> No need to stub this, the caller does a NULL-check before invoking
>>
>>
>> will delete it, thanks.
>>
>>>
>>>> +
>>>> +static void hibmc_ttm_backend_destroy(struct ttm_tt *tt)
>>>> +{
>>>> +       ttm_tt_fini(tt);
>>>> +       kfree(tt);
>>>> +}
>>>> +
>>>> +static struct ttm_backend_func hibmc_tt_backend_func = {
>>>> +       .destroy = &hibmc_ttm_backend_destroy,
>>>> +};
>>>> +
>>>> +static struct ttm_tt *hibmc_ttm_tt_create(struct ttm_bo_device *bdev,
>>>> +                                         unsigned long size,
>>>> +                                         u32 page_flags,
>>>> +                                         struct page *dummy_read_page)
>>>> +{
>>>> +       struct ttm_tt *tt;
>>>> +
>>>> +       tt = kzalloc(sizeof(*tt), GFP_KERNEL);
>>>> +       if (!tt)
>>>
>>>
>>> Print error
>>
>>
>> ok, will do that, thanks.
>>
>>>
>>>> +               return NULL;
>>>> +       tt->func = &hibmc_tt_backend_func;
>>>> +       if (ttm_tt_init(tt, bdev, size, page_flags, dummy_read_page)) {
>>>
>>>
>>> Here too?
>>
>>
>> ditto
>>
>>
>>>
>>>> +               kfree(tt);
>>>> +               return NULL;
>>>> +       }
>>>> +       return tt;
>>>> +}
>>>> +
>>>> +static int hibmc_ttm_tt_populate(struct ttm_tt *ttm)
>>>> +{
>>>> +       return ttm_pool_populate(ttm);
>>>> +}
>>>> +
>>>> +static void hibmc_ttm_tt_unpopulate(struct ttm_tt *ttm)
>>>> +{
>>>> +       ttm_pool_unpopulate(ttm);
>>>> +}
>>>> +
>>>> +struct ttm_bo_driver hibmc_bo_driver = {
>>>> +       .ttm_tt_create          = hibmc_ttm_tt_create,
>>>> +       .ttm_tt_populate        = hibmc_ttm_tt_populate,
>>>> +       .ttm_tt_unpopulate      = hibmc_ttm_tt_unpopulate,
>>>> +       .init_mem_type          = hibmc_bo_init_mem_type,
>>>> +       .evict_flags            = hibmc_bo_evict_flags,
>>>> +       .move                   = NULL,
>>>> +       .verify_access          = hibmc_bo_verify_access,
>>>> +       .io_mem_reserve         = &hibmc_ttm_io_mem_reserve,
>>>> +       .io_mem_free            = &hibmc_ttm_io_mem_free,
>>>> +       .lru_tail               = &ttm_bo_default_lru_tail,
>>>> +       .swap_lru_tail          = &ttm_bo_default_swap_lru_tail,
>>>> +};
>>>> +
>>>> +int hibmc_mm_init(struct hibmc_drm_device *hibmc)
>>>> +{
>>>> +       int ret;
>>>> +       struct drm_device *dev = hibmc->dev;
>>>> +       struct ttm_bo_device *bdev = &hibmc->ttm.bdev;
>>>> +
>>>> +       ret = hibmc_ttm_global_init(hibmc);
>>>> +       if (ret)
>>>> +               return ret;
>>>> +
>>>> +       ret = ttm_bo_device_init(&hibmc->ttm.bdev,
>>>> +                                hibmc->ttm.bo_global_ref.ref.object,
>>>> +                                &hibmc_bo_driver,
>>>> +                                dev->anon_inode->i_mapping,
>>>> +                                DRM_FILE_PAGE_OFFSET,
>>>> +                                true);
>>>> +       if (ret) {
>>>
>>>
>>> Call hibmc_ttm_global_release here?
>>
>>
>> agreed, thanks for pointing it out.
>>
>>>
>>>> +               DRM_ERROR("Error initialising bo driver; %d\n", ret);
>>>> +               return ret;
>>>> +       }
>>>> +
>>>> +       ret = ttm_bo_init_mm(bdev, TTM_PL_VRAM,
>>>> +                            hibmc->fb_size >> PAGE_SHIFT);
>>>> +       if (ret) {
>>>
>>>
>>> Clean up here as well?
>>
>>
>> ditto
>>
>>
>>>
>>>> +               DRM_ERROR("Failed ttm VRAM init: %d\n", ret);
>>>> +               return ret;
>>>> +       }
>>>> +
>>>> +       hibmc->mm_inited = true;
>>>> +       return 0;
>>>> +}
>>>> +
>>>> +void hibmc_mm_fini(struct hibmc_drm_device *hibmc)
>>>> +{
>>>> +       if (!hibmc->mm_inited)
>>>> +               return;
>>>> +
>>>> +       ttm_bo_device_release(&hibmc->ttm.bdev);
>>>> +       hibmc_ttm_global_release(hibmc);
>>>> +       hibmc->mm_inited = false;
>>>> +}
>>>> +
>>>> +int hibmc_bo_create(struct drm_device *dev, int size, int align,
>>>> +                   u32 flags, struct hibmc_bo **phibmcbo)
>>>> +{
>>>> +       struct hibmc_drm_device *hibmc = dev->dev_private;
>>>> +       struct hibmc_bo *hibmcbo;
>>>> +       size_t acc_size;
>>>> +       int ret;
>>>> +
>>>> +       hibmcbo = kzalloc(sizeof(*hibmcbo), GFP_KERNEL);
>>>> +       if (!hibmcbo)
>>>> +               return -ENOMEM;
>>>> +
>>>> +       ret = drm_gem_object_init(dev, &hibmcbo->gem, size);
>>>> +       if (ret) {
>>>> +               kfree(hibmcbo);
>>>> +               return ret;
>>>> +       }
>>>> +
>>>> +       hibmcbo->bo.bdev = &hibmc->ttm.bdev;
>>>> +
>>>> +       hibmc_ttm_placement(hibmcbo, TTM_PL_FLAG_VRAM |
>>>> TTM_PL_FLAG_SYSTEM);
>>>> +
>>>> +       acc_size = ttm_bo_dma_acc_size(&hibmc->ttm.bdev, size,
>>>> +                                      sizeof(struct hibmc_bo));
>>>> +
>>>> +       ret = ttm_bo_init(&hibmc->ttm.bdev, &hibmcbo->bo, size,
>>>> +                         ttm_bo_type_device, &hibmcbo->placement,
>>>> +                         align >> PAGE_SHIFT, false, NULL, acc_size,
>>>> +                         NULL, NULL, hibmc_bo_ttm_destroy);
>>>> +       if (ret)
>>>
>>>
>>> Missing hibmcbo clean up here
>>
>>
>> i looked at all other ttm drivers and all of them return directly when
>> ttm_bo_init
>> failed, however, i think it is better to clean up here, should i call
>> hibmc_bo_unref(&hibmc_bo) here ?
>>
>
> Yeah, that should work (might want to test it, though ;)
>
>
>>>
>>>> +               return ret;
>>>> +
>>>> +       *phibmcbo = hibmcbo;
>>>> +       return 0;
>>>> +}
>>>> +
>>>> +static inline u64 hibmc_bo_gpu_offset(struct hibmc_bo *bo)
>>>> +{
>>>> +       return bo->bo.offset;
>>>> +}
>>>
>>>
>>> I don't think this function provides any value
>>
>>
>> do you nean i use bo->bo.offset instead of calling hibmc_bo_gpu_offset()?
>>
>
> yes
>
>>>
>>>> +
>>>> +int hibmc_bo_pin(struct hibmc_bo *bo, u32 pl_flag, u64 *gpu_addr)
>>>> +{
>>>> +       int i, ret;
>>>> +
>>>> +       if (bo->pin_count) {
>>>> +               bo->pin_count++;
>>>> +               if (gpu_addr)
>>>> +                       *gpu_addr = hibmc_bo_gpu_offset(bo);
>>>
>>>
>>> Are you missing a return here?
>>
>>
>> Thanks for pointing it out!
>>
>>
>>>
>>>> +       }
>>>> +
>>>> +       hibmc_ttm_placement(bo, pl_flag);
>>>> +       for (i = 0; i < bo->placement.num_placement; i++)
>>>> +               bo->placements[i].flags |= TTM_PL_FLAG_NO_EVICT;
>>>> +       ret = ttm_bo_validate(&bo->bo, &bo->placement, false, false);
>>>> +       if (ret)
>>>> +               return ret;
>>>> +
>>>> +       bo->pin_count = 1;
>>>> +       if (gpu_addr)
>>>> +               *gpu_addr = hibmc_bo_gpu_offset(bo);
>>>> +       return 0;
>>>> +}
>>>> +
>>>> +int hibmc_bo_push_sysram(struct hibmc_bo *bo)
>>>> +{
>>>> +       int i, ret;
>>>> +
>>>> +       if (!bo->pin_count) {
>>>> +               DRM_ERROR("unpin bad %p\n", bo);
>>>> +               return 0;
>>>> +       }
>>>> +       bo->pin_count--;
>>>> +       if (bo->pin_count)
>>>> +               return 0;
>>>> +
>>>> +       if (bo->kmap.virtual)
>>>
>>>
>>> ttm_bo_kunmap already does this check so you don't have to
>>
>>
>> agreed. will remove this condition.
>>
>>>
>>>> +               ttm_bo_kunmap(&bo->kmap);
>>>> +
>>>> +       hibmc_ttm_placement(bo, TTM_PL_FLAG_SYSTEM);
>>>> +       for (i = 0; i < bo->placement.num_placement ; i++)
>>>> +               bo->placements[i].flags |= TTM_PL_FLAG_NO_EVICT;
>>>> +
>>>> +       ret = ttm_bo_validate(&bo->bo, &bo->placement, false, false);
>>>> +       if (ret) {
>>>> +               DRM_ERROR("pushing to VRAM failed\n");
>>>
>>>
>>> Print ret
>>
>>
>> ok, thanks.
>>
>>
>>>
>>>> +               return ret;
>>>> +       }
>>>> +       return 0;
>>>> +}
>>>> +
>>>> +int hibmc_mmap(struct file *filp, struct vm_area_struct *vma)
>>>> +{
>>>> +       struct drm_file *file_priv;
>>>> +       struct hibmc_drm_device *hibmc;
>>>> +
>>>> +       if (unlikely(vma->vm_pgoff < DRM_FILE_PAGE_OFFSET))
>>>> +               return -EINVAL;
>>>> +
>>>> +       file_priv = filp->private_data;
>>>> +       hibmc = file_priv->minor->dev->dev_private;
>>>> +       return ttm_bo_mmap(filp, vma, &hibmc->ttm.bdev);
>>>> +}
>>>> +
>>>> +int hibmc_gem_create(struct drm_device *dev, u32 size, bool iskernel,
>>>> +                    struct drm_gem_object **obj)
>>>> +{
>>>> +       struct hibmc_bo *hibmcbo;
>>>> +       int ret;
>>>> +
>>>> +       *obj = NULL;
>>>> +
>>>> +       size = PAGE_ALIGN(size);
>>>> +       if (size == 0)
>>>
>>>
>>> Print error
>>
>>
>> ditto
>>
>>>
>>>> +               return -EINVAL;
>>>> +
>>>> +       ret = hibmc_bo_create(dev, size, 0, 0, &hibmcbo);
>>>> +       if (ret) {
>>>> +               if (ret != -ERESTARTSYS)
>>>> +                       DRM_ERROR("failed to allocate GEM object\n");
>>>
>>>
>>> Print ret
>>
>>
>> ditto
>>
>>>
>>>> +               return ret;
>>>> +       }
>>>> +       *obj = &hibmcbo->gem;
>>>> +       return 0;
>>>> +}
>>>> +
>>>> +int hibmc_dumb_create(struct drm_file *file, struct drm_device *dev,
>>>> +                     struct drm_mode_create_dumb *args)
>>>> +{
>>>> +       struct drm_gem_object *gobj;
>>>> +       u32 handle;
>>>> +       int ret;
>>>> +
>>>> +       args->pitch = ALIGN(args->width * ((args->bpp + 7) / 8), 16);
>>>
>>>
>>> What's up with the bpp + 7 here? Perhaps you're looking for DIV_ROUND_UP?
>>
>>
>> Yes, that sounds sane.
>>
>
> sane is what i usually aim for :)
>
> Sean
>
>
>>>
>>>
>>>> +       args->size = args->pitch * args->height;
>>>> +
>>>> +       ret = hibmc_gem_create(dev, args->size, false,
>>>> +                              &gobj);
>>>> +       if (ret)
>>>> +               return ret;
>>>> +
>>>> +       ret = drm_gem_handle_create(file, gobj, &handle);
>>>> +       drm_gem_object_unreference_unlocked(gobj);
>>>> +       if (ret)
>>>
>>>
>>> Print error here
>>
>>
>> agreed.
>>
>>
>>>
>>>> +               return ret;
>>>> +
>>>> +       args->handle = handle;
>>>> +       return 0;
>>>> +}
>>>> +
>>>> +static void hibmc_bo_unref(struct hibmc_bo **bo)
>>>> +{
>>>> +       struct ttm_buffer_object *tbo;
>>>> +
>>>> +       if ((*bo) == NULL)
>>>> +               return;
>>>> +
>>>> +       tbo = &((*bo)->bo);
>>>> +       ttm_bo_unref(&tbo);
>>>> +       *bo = NULL;
>>>> +}
>>>> +
>>>> +void hibmc_gem_free_object(struct drm_gem_object *obj)
>>>> +{
>>>> +       struct hibmc_bo *hibmcbo = gem_to_hibmc_bo(obj);
>>>> +
>>>> +       hibmc_bo_unref(&hibmcbo);
>>>> +}
>>>> +
>>>> +static u64 hibmc_bo_mmap_offset(struct hibmc_bo *bo)
>>>> +{
>>>> +       return drm_vma_node_offset_addr(&bo->bo.vma_node);
>>>> +}
>>>> +
>>>> +int hibmc_dumb_mmap_offset(struct drm_file *file, struct drm_device
>>>> *dev,
>>>> +                          u32 handle, u64 *offset)
>>>> +{
>>>> +       struct drm_gem_object *obj;
>>>> +       struct hibmc_bo *bo;
>>>> +
>>>> +       obj = drm_gem_object_lookup(file, handle);
>>>> +       if (!obj)
>>>> +               return -ENOENT;
>>>> +
>>>> +       bo = gem_to_hibmc_bo(obj);
>>>> +       *offset = hibmc_bo_mmap_offset(bo);
>>>> +
>>>> +       drm_gem_object_unreference_unlocked(obj);
>>>> +       return 0;
>>>> +}
>>
>>
>> Regards,
>> Rongrong.
>>
>>>> --
>>>> 1.9.1
>>>>
>>>>
>>>> _______________________________________________
>>>> linux-arm-kernel mailing list
>>>> linux-arm-kernel at lists.infradead.org
>>>> http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
>>>
>>> _______________________________________________
>>> linuxarm mailing list
>>> linuxarm at huawei.com
>>> http://rnd-openeuler.huawei.com/mailman/listinfo/linuxarm
>>>
>>> .
>>>
>>
>>
>> _______________________________________________
>> linux-arm-kernel mailing list
>> linux-arm-kernel at lists.infradead.org
>> http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
>
> .
>


-- 
Regards, Rongrong



More information about the linux-arm-kernel mailing list