* [PATCH v3] drm: modify drm_global_item_ref to avoid two times of writing ref->object
@ 2016-09-08 2:07 Huang Rui
2016-09-08 6:36 ` Chris Wilson
2016-09-08 7:26 ` Christian König
0 siblings, 2 replies; 7+ messages in thread
From: Huang Rui @ 2016-09-08 2:07 UTC (permalink / raw)
To: dri-devel-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW, Alex Deucher,
Dave Airlie, Christian König, Daniel Vetter, Sean Paul
Cc: Ken Wang, Huang Rui, amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW
In previous drm_global_item_ref, there are two times of writing
ref->object if item->refcount is 0. So this patch does a minor update
to put alloc and init ref firstly, and then to modify the item of glob
array. Use "else" to avoid two times of writing ref->object. It can
make the code logic more clearly.
Signed-off-by: Huang Rui <ray.huang@amd.com>
---
Changes from V2 -> V3:
- Use duplicate mutex release to avoid "goto" in non-error patch.
- Rename error label
---
drivers/gpu/drm/drm_global.c | 24 ++++++++++++++----------
1 file changed, 14 insertions(+), 10 deletions(-)
diff --git a/drivers/gpu/drm/drm_global.c b/drivers/gpu/drm/drm_global.c
index 3d2e91c..b404287 100644
--- a/drivers/gpu/drm/drm_global.c
+++ b/drivers/gpu/drm/drm_global.c
@@ -65,30 +65,34 @@ void drm_global_release(void)
int drm_global_item_ref(struct drm_global_reference *ref)
{
- int ret;
+ int ret = 0;
struct drm_global_item *item = &glob[ref->global_type];
mutex_lock(&item->mutex);
if (item->refcount == 0) {
- item->object = kzalloc(ref->size, GFP_KERNEL);
- if (unlikely(item->object == NULL)) {
+ ref->object = kzalloc(ref->size, GFP_KERNEL);
+ if (unlikely(ref->object == NULL)) {
ret = -ENOMEM;
- goto out_err;
+ goto error_unlock;
}
-
- ref->object = item->object;
ret = ref->init(ref);
if (unlikely(ret != 0))
- goto out_err;
+ goto error_free;
+ item->object = ref->object;
+ } else {
+ ref->object = item->object;
}
+
++item->refcount;
- ref->object = item->object;
mutex_unlock(&item->mutex);
return 0;
-out_err:
+
+error_free:
+ kfree(ref->object);
+ ref->object = NULL;
+error_unlock:
mutex_unlock(&item->mutex);
- item->object = NULL;
return ret;
}
EXPORT_SYMBOL(drm_global_item_ref);
--
1.9.1
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
^ permalink raw reply related [flat|nested] 7+ messages in thread
* Re: [PATCH v3] drm: modify drm_global_item_ref to avoid two times of writing ref->object
2016-09-08 2:07 [PATCH v3] drm: modify drm_global_item_ref to avoid two times of writing ref->object Huang Rui
@ 2016-09-08 6:36 ` Chris Wilson
2016-09-08 7:22 ` Huang Rui
2016-09-08 7:26 ` Christian König
1 sibling, 1 reply; 7+ messages in thread
From: Chris Wilson @ 2016-09-08 6:36 UTC (permalink / raw)
To: Huang Rui
Cc: Daniel Vetter, dri-devel, amd-gfx, Alex Deucher, Ken Wang,
Christian König
On Wed, Sep 07, 2016 at 10:07:57PM -0400, Huang Rui wrote:
> In previous drm_global_item_ref, there are two times of writing
> ref->object if item->refcount is 0. So this patch does a minor update
> to put alloc and init ref firstly, and then to modify the item of glob
> array. Use "else" to avoid two times of writing ref->object. It can
> make the code logic more clearly.
>
> Signed-off-by: Huang Rui <ray.huang@amd.com>
> ---
>
> Changes from V2 -> V3:
> - Use duplicate mutex release to avoid "goto" in non-error patch.
> - Rename error label
>
> ---
> drivers/gpu/drm/drm_global.c | 24 ++++++++++++++----------
> 1 file changed, 14 insertions(+), 10 deletions(-)
>
> diff --git a/drivers/gpu/drm/drm_global.c b/drivers/gpu/drm/drm_global.c
> index 3d2e91c..b404287 100644
> --- a/drivers/gpu/drm/drm_global.c
> +++ b/drivers/gpu/drm/drm_global.c
> @@ -65,30 +65,34 @@ void drm_global_release(void)
>
> int drm_global_item_ref(struct drm_global_reference *ref)
> {
> - int ret;
> + int ret = 0;
> struct drm_global_item *item = &glob[ref->global_type];
>
> mutex_lock(&item->mutex);
> if (item->refcount == 0) {
> - item->object = kzalloc(ref->size, GFP_KERNEL);
> - if (unlikely(item->object == NULL)) {
> + ref->object = kzalloc(ref->size, GFP_KERNEL);
So the item refcount is 0, we operate on ref, whereas previous we
inspected item and operated on item. Not an improvement.
-Chris
--
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [PATCH v3] drm: modify drm_global_item_ref to avoid two times of writing ref->object
2016-09-08 6:36 ` Chris Wilson
@ 2016-09-08 7:22 ` Huang Rui
2016-09-08 7:35 ` Chris Wilson
0 siblings, 1 reply; 7+ messages in thread
From: Huang Rui @ 2016-09-08 7:22 UTC (permalink / raw)
To: Chris Wilson
Cc: Daniel Vetter, amd-gfx, dri-devel, Deucher, Alexander, Wang, Ken,
Koenig, Christian
On Thu, Sep 08, 2016 at 02:36:06PM +0800, Chris Wilson wrote:
> On Wed, Sep 07, 2016 at 10:07:57PM -0400, Huang Rui wrote:
> > In previous drm_global_item_ref, there are two times of writing
> > ref->object if item->refcount is 0. So this patch does a minor update
> > to put alloc and init ref firstly, and then to modify the item of glob
> > array. Use "else" to avoid two times of writing ref->object. It can
> > make the code logic more clearly.
> >
> > Signed-off-by: Huang Rui <ray.huang@amd.com>
> > ---
> >
> > Changes from V2 -> V3:
> > - Use duplicate mutex release to avoid "goto" in non-error patch.
> > - Rename error label
> >
> > ---
> > drivers/gpu/drm/drm_global.c | 24 ++++++++++++++----------
> > 1 file changed, 14 insertions(+), 10 deletions(-)
> >
> > diff --git a/drivers/gpu/drm/drm_global.c b/drivers/gpu/drm/drm_global.c
> > index 3d2e91c..b404287 100644
> > --- a/drivers/gpu/drm/drm_global.c
> > +++ b/drivers/gpu/drm/drm_global.c
> > @@ -65,30 +65,34 @@ void drm_global_release(void)
> >
> > int drm_global_item_ref(struct drm_global_reference *ref)
> > {
> > - int ret;
> > + int ret = 0;
> > struct drm_global_item *item = &glob[ref->global_type];
> >
> > mutex_lock(&item->mutex);
> > if (item->refcount == 0) {
> > - item->object = kzalloc(ref->size, GFP_KERNEL);
> > - if (unlikely(item->object == NULL)) {
> > + ref->object = kzalloc(ref->size, GFP_KERNEL);
>
> So the item refcount is 0, we operate on ref, whereas previous we
> inspected item and operated on item. Not an improvement.
Hmm, when item refcount is 0, we operate on ref to create object and init
it for item, so although item->object and ref->object here should point the
same thing, but we should alloc and init ref firstly and pass the
ref->object address to item (item actually points a static area). This make
the code logic more clearly and readable. So I updated it to "ref" here.
:-)
Thanks,
Rui
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [PATCH v3] drm: modify drm_global_item_ref to avoid two times of writing ref->object
2016-09-08 2:07 [PATCH v3] drm: modify drm_global_item_ref to avoid two times of writing ref->object Huang Rui
2016-09-08 6:36 ` Chris Wilson
@ 2016-09-08 7:26 ` Christian König
1 sibling, 0 replies; 7+ messages in thread
From: Christian König @ 2016-09-08 7:26 UTC (permalink / raw)
To: Huang Rui, dri-devel, Alex Deucher, Dave Airlie, Daniel Vetter,
Sean Paul
Cc: Ken Wang, amd-gfx
Am 08.09.2016 um 04:07 schrieb Huang Rui:
> In previous drm_global_item_ref, there are two times of writing
> ref->object if item->refcount is 0. So this patch does a minor update
> to put alloc and init ref firstly, and then to modify the item of glob
> array. Use "else" to avoid two times of writing ref->object. It can
> make the code logic more clearly.
>
> Signed-off-by: Huang Rui <ray.huang@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>.
> ---
>
> Changes from V2 -> V3:
> - Use duplicate mutex release to avoid "goto" in non-error patch.
> - Rename error label
>
> ---
> drivers/gpu/drm/drm_global.c | 24 ++++++++++++++----------
> 1 file changed, 14 insertions(+), 10 deletions(-)
>
> diff --git a/drivers/gpu/drm/drm_global.c b/drivers/gpu/drm/drm_global.c
> index 3d2e91c..b404287 100644
> --- a/drivers/gpu/drm/drm_global.c
> +++ b/drivers/gpu/drm/drm_global.c
> @@ -65,30 +65,34 @@ void drm_global_release(void)
>
> int drm_global_item_ref(struct drm_global_reference *ref)
> {
> - int ret;
> + int ret = 0;
> struct drm_global_item *item = &glob[ref->global_type];
>
> mutex_lock(&item->mutex);
> if (item->refcount == 0) {
> - item->object = kzalloc(ref->size, GFP_KERNEL);
> - if (unlikely(item->object == NULL)) {
> + ref->object = kzalloc(ref->size, GFP_KERNEL);
> + if (unlikely(ref->object == NULL)) {
> ret = -ENOMEM;
> - goto out_err;
> + goto error_unlock;
> }
> -
> - ref->object = item->object;
> ret = ref->init(ref);
> if (unlikely(ret != 0))
> - goto out_err;
> + goto error_free;
>
> + item->object = ref->object;
> + } else {
> + ref->object = item->object;
> }
> +
> ++item->refcount;
> - ref->object = item->object;
> mutex_unlock(&item->mutex);
> return 0;
> -out_err:
> +
> +error_free:
> + kfree(ref->object);
> + ref->object = NULL;
> +error_unlock:
> mutex_unlock(&item->mutex);
> - item->object = NULL;
> return ret;
> }
> EXPORT_SYMBOL(drm_global_item_ref);
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [PATCH v3] drm: modify drm_global_item_ref to avoid two times of writing ref->object
2016-09-08 7:22 ` Huang Rui
@ 2016-09-08 7:35 ` Chris Wilson
2016-09-08 7:43 ` Christian König
0 siblings, 1 reply; 7+ messages in thread
From: Chris Wilson @ 2016-09-08 7:35 UTC (permalink / raw)
To: Huang Rui
Cc: Daniel Vetter, amd-gfx, dri-devel, Deucher, Alexander, Wang, Ken,
Koenig, Christian
On Thu, Sep 08, 2016 at 03:22:48PM +0800, Huang Rui wrote:
> On Thu, Sep 08, 2016 at 02:36:06PM +0800, Chris Wilson wrote:
> > On Wed, Sep 07, 2016 at 10:07:57PM -0400, Huang Rui wrote:
> > > In previous drm_global_item_ref, there are two times of writing
> > > ref->object if item->refcount is 0. So this patch does a minor update
> > > to put alloc and init ref firstly, and then to modify the item of glob
> > > array. Use "else" to avoid two times of writing ref->object. It can
> > > make the code logic more clearly.
> > >
> > > Signed-off-by: Huang Rui <ray.huang@amd.com>
> > > ---
> > >
> > > Changes from V2 -> V3:
> > > - Use duplicate mutex release to avoid "goto" in non-error patch.
> > > - Rename error label
> > >
> > > ---
> > > drivers/gpu/drm/drm_global.c | 24 ++++++++++++++----------
> > > 1 file changed, 14 insertions(+), 10 deletions(-)
> > >
> > > diff --git a/drivers/gpu/drm/drm_global.c b/drivers/gpu/drm/drm_global.c
> > > index 3d2e91c..b404287 100644
> > > --- a/drivers/gpu/drm/drm_global.c
> > > +++ b/drivers/gpu/drm/drm_global.c
> > > @@ -65,30 +65,34 @@ void drm_global_release(void)
> > >
> > > int drm_global_item_ref(struct drm_global_reference *ref)
> > > {
> > > - int ret;
> > > + int ret = 0;
> > > struct drm_global_item *item = &glob[ref->global_type];
> > >
> > > mutex_lock(&item->mutex);
> > > if (item->refcount == 0) {
> > > - item->object = kzalloc(ref->size, GFP_KERNEL);
> > > - if (unlikely(item->object == NULL)) {
> > > + ref->object = kzalloc(ref->size, GFP_KERNEL);
> >
> > So the item refcount is 0, we operate on ref, whereas previous we
> > inspected item and operated on item. Not an improvement.
>
> Hmm, when item refcount is 0, we operate on ref to create object and init
> it for item, so although item->object and ref->object here should point the
> same thing, but we should alloc and init ref firstly and pass the
> ref->object address to item (item actually points a static area). This make
> the code logic more clearly and readable. So I updated it to "ref" here.
> :-)
The object is owned by item. ref is just a pointer to it.
-Chris
--
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [PATCH v3] drm: modify drm_global_item_ref to avoid two times of writing ref->object
2016-09-08 7:35 ` Chris Wilson
@ 2016-09-08 7:43 ` Christian König
2016-09-08 8:08 ` Chris Wilson
0 siblings, 1 reply; 7+ messages in thread
From: Christian König @ 2016-09-08 7:43 UTC (permalink / raw)
To: Chris Wilson, Huang Rui, Daniel Vetter, dri-devel, amd-gfx,
Deucher, Alexander, Wang, Ken
Am 08.09.2016 um 09:35 schrieb Chris Wilson:
> On Thu, Sep 08, 2016 at 03:22:48PM +0800, Huang Rui wrote:
>> On Thu, Sep 08, 2016 at 02:36:06PM +0800, Chris Wilson wrote:
>>> On Wed, Sep 07, 2016 at 10:07:57PM -0400, Huang Rui wrote:
>>>> In previous drm_global_item_ref, there are two times of writing
>>>> ref->object if item->refcount is 0. So this patch does a minor update
>>>> to put alloc and init ref firstly, and then to modify the item of glob
>>>> array. Use "else" to avoid two times of writing ref->object. It can
>>>> make the code logic more clearly.
>>>>
>>>> Signed-off-by: Huang Rui <ray.huang@amd.com>
>>>> ---
>>>>
>>>> Changes from V2 -> V3:
>>>> - Use duplicate mutex release to avoid "goto" in non-error patch.
>>>> - Rename error label
>>>>
>>>> ---
>>>> drivers/gpu/drm/drm_global.c | 24 ++++++++++++++----------
>>>> 1 file changed, 14 insertions(+), 10 deletions(-)
>>>>
>>>> diff --git a/drivers/gpu/drm/drm_global.c b/drivers/gpu/drm/drm_global.c
>>>> index 3d2e91c..b404287 100644
>>>> --- a/drivers/gpu/drm/drm_global.c
>>>> +++ b/drivers/gpu/drm/drm_global.c
>>>> @@ -65,30 +65,34 @@ void drm_global_release(void)
>>>>
>>>> int drm_global_item_ref(struct drm_global_reference *ref)
>>>> {
>>>> - int ret;
>>>> + int ret = 0;
>>>> struct drm_global_item *item = &glob[ref->global_type];
>>>>
>>>> mutex_lock(&item->mutex);
>>>> if (item->refcount == 0) {
>>>> - item->object = kzalloc(ref->size, GFP_KERNEL);
>>>> - if (unlikely(item->object == NULL)) {
>>>> + ref->object = kzalloc(ref->size, GFP_KERNEL);
>>> So the item refcount is 0, we operate on ref, whereas previous we
>>> inspected item and operated on item. Not an improvement.
>> Hmm, when item refcount is 0, we operate on ref to create object and init
>> it for item, so although item->object and ref->object here should point the
>> same thing, but we should alloc and init ref firstly and pass the
>> ref->object address to item (item actually points a static area). This make
>> the code logic more clearly and readable. So I updated it to "ref" here.
>> :-)
> The object is owned by item. ref is just a pointer to it.
Yeah, so what? We initialized ref->item when the refcount was 0 before
as well.
Why not create the reference first and initialize the item later on?
Regards,
Christian.
> -Chris
>
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [PATCH v3] drm: modify drm_global_item_ref to avoid two times of writing ref->object
2016-09-08 7:43 ` Christian König
@ 2016-09-08 8:08 ` Chris Wilson
0 siblings, 0 replies; 7+ messages in thread
From: Chris Wilson @ 2016-09-08 8:08 UTC (permalink / raw)
To: Christian König
Cc: Daniel Vetter, amd-gfx, Huang Rui, dri-devel, Deucher, Alexander,
Wang, Ken
On Thu, Sep 08, 2016 at 09:43:52AM +0200, Christian König wrote:
> Am 08.09.2016 um 09:35 schrieb Chris Wilson:
> >On Thu, Sep 08, 2016 at 03:22:48PM +0800, Huang Rui wrote:
> >>On Thu, Sep 08, 2016 at 02:36:06PM +0800, Chris Wilson wrote:
> >>>On Wed, Sep 07, 2016 at 10:07:57PM -0400, Huang Rui wrote:
> >>>>In previous drm_global_item_ref, there are two times of writing
> >>>>ref->object if item->refcount is 0. So this patch does a minor update
> >>>>to put alloc and init ref firstly, and then to modify the item of glob
> >>>>array. Use "else" to avoid two times of writing ref->object. It can
> >>>>make the code logic more clearly.
> >>>>
> >>>>Signed-off-by: Huang Rui <ray.huang@amd.com>
> >>>>---
> >>>>
> >>>>Changes from V2 -> V3:
> >>>>- Use duplicate mutex release to avoid "goto" in non-error patch.
> >>>>- Rename error label
> >>>>
> >>>>---
> >>>> drivers/gpu/drm/drm_global.c | 24 ++++++++++++++----------
> >>>> 1 file changed, 14 insertions(+), 10 deletions(-)
> >>>>
> >>>>diff --git a/drivers/gpu/drm/drm_global.c b/drivers/gpu/drm/drm_global.c
> >>>>index 3d2e91c..b404287 100644
> >>>>--- a/drivers/gpu/drm/drm_global.c
> >>>>+++ b/drivers/gpu/drm/drm_global.c
> >>>>@@ -65,30 +65,34 @@ void drm_global_release(void)
> >>>> int drm_global_item_ref(struct drm_global_reference *ref)
> >>>> {
> >>>>- int ret;
> >>>>+ int ret = 0;
> >>>> struct drm_global_item *item = &glob[ref->global_type];
> >>>> mutex_lock(&item->mutex);
> >>>> if (item->refcount == 0) {
> >>>>- item->object = kzalloc(ref->size, GFP_KERNEL);
> >>>>- if (unlikely(item->object == NULL)) {
> >>>>+ ref->object = kzalloc(ref->size, GFP_KERNEL);
> >>>So the item refcount is 0, we operate on ref, whereas previous we
> >>>inspected item and operated on item. Not an improvement.
> >>Hmm, when item refcount is 0, we operate on ref to create object and init
> >>it for item, so although item->object and ref->object here should point the
> >>same thing, but we should alloc and init ref firstly and pass the
> >>ref->object address to item (item actually points a static area). This make
> >>the code logic more clearly and readable. So I updated it to "ref" here.
> >>:-)
> >The object is owned by item. ref is just a pointer to it.
>
> Yeah, so what? We initialized ref->item when the refcount was 0
> before as well.
>
> Why not create the reference first and initialize the item later on?
More to the point, there is a genuine bug in the code. It happens to be
fixed by this patch, but not purported to be. There is also a leak of
item->object, and the globals conflict between drivers. If radeon
instantiates the DRM_GLOBAL_TTM_MEM, nouveau bumps the reference and
radeon subsequently unloaded, when nouveau unloads it will crash due to
executing a random address. As a singleton class for ttm, it could do
with a revamp.
-Chris
--
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel
^ permalink raw reply [flat|nested] 7+ messages in thread
end of thread, other threads:[~2016-09-08 8:08 UTC | newest]
Thread overview: 7+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2016-09-08 2:07 [PATCH v3] drm: modify drm_global_item_ref to avoid two times of writing ref->object Huang Rui
2016-09-08 6:36 ` Chris Wilson
2016-09-08 7:22 ` Huang Rui
2016-09-08 7:35 ` Chris Wilson
2016-09-08 7:43 ` Christian König
2016-09-08 8:08 ` Chris Wilson
2016-09-08 7:26 ` Christian König
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.