Re: [PATCH 1/3] drm/i915: Switch obj->mm.lock lockdep annotations on its head

From: Chris Wilson
Date: Tue Aug 20 2019 - 06:05:28 EST

Quoting Daniel Vetter (2019-08-20 09:19:49)
> +#include <linux/sched/mm.h>
> +
> #include "display/intel_frontbuffer.h"
> #include "gt/intel_gt.h"
> #include "i915_drv.h"
> @@ -51,6 +53,15 @@ void i915_gem_object_init(struct drm_i915_gem_object *obj,
> {
> mutex_init(&obj->mm.lock);
>
> + if (IS_ENABLED(CONFIG_LOCKDEP)) {
> + mutex_lock_nested(&obj->mm.lock, I915_MM_GET_PAGES);
> + fs_reclaim_acquire(GFP_KERNEL);
> + might_lock(&obj->mm.lock);
> + fs_reclaim_release(GFP_KERNEL);
> + mutex_unlock(&obj->mm.lock);
> + }
> +
> +

It's good, but nothing is worth angering checkpatch over an extra '\n'.

> spin_lock_init(&obj->vma.lock);
> INIT_LIST_HEAD(&obj->vma.list);
>
> @@ -176,7 +187,7 @@ static void __i915_gem_free_objects(struct drm_i915_private *i915,
> GEM_BUG_ON(!list_empty(&obj->lut_list));
>
> atomic_set(&obj->mm.pages_pin_count, 0);
> - __i915_gem_object_put_pages(obj, I915_MM_NORMAL);
> + __i915_gem_object_put_pages(obj);
> GEM_BUG_ON(i915_gem_object_has_pages(obj));
> bitmap_free(obj->bit_17);
>
> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_object.h b/drivers/gpu/drm/i915/gem/i915_gem_object.h
> index 5efb9936e05b..a0b1fa8a3224 100644
> --- a/drivers/gpu/drm/i915/gem/i915_gem_object.h
> +++ b/drivers/gpu/drm/i915/gem/i915_gem_object.h
> @@ -281,11 +281,21 @@ i915_gem_object_unpin_pages(struct drm_i915_gem_object *obj)
>
> enum i915_mm_subclass { /* lockdep subclass for obj->mm.lock/struct_mutex */
> I915_MM_NORMAL = 0,
> - I915_MM_SHRINKER /* called "recursively" from direct-reclaim-esque */
> + /*
> + * Only used by struct_mutex, when called "recursively" from
> + * direct-reclaim-esque. Safe because there is only every one
> + * struct_mutex in the entire system. */
> + I915_MM_SHRINKER = 1,

MM_SHRINKER is no longer part of this subclass, and I intend to remove
shortly.

> + /*
> + * Used for obj->mm.lock when allocating pages. Safe because the object
> + * isn't yet on any LRU, and therefore the shrinker can't deadlock on
> + * it. As soon as the object has pages, obj->mm.lock nests within
> + * fs_reclaim.
> + */
> + I915_MM_GET_PAGES = 1,

So I was thinking that this can just become SINGLE_DEPTH_NESTING, but
then remembered that I want a proxy object with its own recursion inside
get-pages.

> };
>
> -int __i915_gem_object_put_pages(struct drm_i915_gem_object *obj,
> - enum i915_mm_subclass subclass);
> +int __i915_gem_object_put_pages(struct drm_i915_gem_object *obj);
> void i915_gem_object_truncate(struct drm_i915_gem_object *obj);
> void i915_gem_object_writeback(struct drm_i915_gem_object *obj);
>
> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_object_types.h b/drivers/gpu/drm/i915/gem/i915_gem_object_types.h
> index ede0eb4218a8..7b7cf711a21a 100644
> --- a/drivers/gpu/drm/i915/gem/i915_gem_object_types.h
> +++ b/drivers/gpu/drm/i915/gem/i915_gem_object_types.h
> @@ -156,7 +156,11 @@ struct drm_i915_gem_object {
> unsigned int pin_global;
>
> struct {
> - struct mutex lock; /* protects the pages and their use */
> + /*
> + * Protects the pages and their use. Do not use directly, but
> + * instead go through the pin/unpin interfaces.
> + */
> + struct mutex lock;

I am really tempted to rename this pin_mutex.

> atomic_t pages_pin_count;

>
> struct sg_table *pages;
> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_pages.c b/drivers/gpu/drm/i915/gem/i915_gem_pages.c
> index 18f0ce0135c1..202526e8910f 100644
> --- a/drivers/gpu/drm/i915/gem/i915_gem_pages.c
> +++ b/drivers/gpu/drm/i915/gem/i915_gem_pages.c
> @@ -101,7 +101,7 @@ int __i915_gem_object_get_pages(struct drm_i915_gem_object *obj)
> {
> int err;
>
> - err = mutex_lock_interruptible(&obj->mm.lock);
> + err = mutex_lock_interruptible_nested(&obj->mm.lock, I915_MM_GET_PAGES);
> if (err)
> return err;

> @@ -285,7 +284,7 @@ void *i915_gem_object_pin_map(struct drm_i915_gem_object *obj,
> if (unlikely(!i915_gem_object_has_struct_page(obj)))
> return ERR_PTR(-ENXIO);
>
> - err = mutex_lock_interruptible(&obj->mm.lock);
> + err = mutex_lock_interruptible_nested(&obj->mm.lock, I915_MM_GET_PAGES);
> if (err)
> return ERR_PTR(err);
>
> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_phys.c b/drivers/gpu/drm/i915/gem/i915_gem_phys.c
> index 768356908160..2aea8960f0f1 100644
> --- a/drivers/gpu/drm/i915/gem/i915_gem_phys.c
> +++ b/drivers/gpu/drm/i915/gem/i915_gem_phys.c
> @@ -163,7 +163,7 @@ int i915_gem_object_attach_phys(struct drm_i915_gem_object *obj, int align)
> if (err)
> return err;
>
> - mutex_lock(&obj->mm.lock);
> + mutex_lock_nested(&obj->mm.lock, I915_MM_GET_PAGES);

> @@ -514,7 +514,7 @@ __i915_gem_userptr_get_pages_worker(struct work_struct *_work)
> }
> }
>
> - mutex_lock(&obj->mm.lock);
> + mutex_lock_nested(&obj->mm.lock, I915_MM_GET_PAGES);
> if (obj->userptr.work == &work->work) {
> struct sg_table *pages = ERR_PTR(ret);
>

Ok. That should be all of the allocators.

And the put are selfchecking.

The only caveat I have is that GET_PAGES is simply about the recursive
behaviour of allocating pages (to allocate one set of pages we may have
to free another), and giving it a distinct subclass tricks me, at least,
into considering it to be a distinct lockclass, and so start assuming it
needs to be used in e.g. i915_gem_madvise_ioctl.

That's just a poorly educated user problem (subclass should always be
about recursion levels, if you have distinct lockclasses, make them
distinct!) Hammer, meet screw.

Reviewed-by: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>
-Chris