On Thu, Dec 12, 2019 at 12:34:49PM -0800, Lucas De Marchi wrote:
On Thu, Dec 12, 2019 at 09:37:17AM -0800, Matt Roper wrote:
On Wed, Dec 11, 2019 at 04:22:50PM -0800, Lucas De Marchi wrote:
On Wed, Dec 11, 2019 at 12:10:41PM +0530, Bharadiya,Pankaj wrote:
On Tue, Dec 10, 2019 at 09:57:39PM -0800, Lucas De Marchi wrote:
> On Mon, Dec 09, 2019 at 08:09:02PM +0530, Pankaj Bharadiya wrote:
> >intel_bw_state allocated memory is not getting freed even after
> >module removal.
> >
> >kmemleak reported backtrace:
> >
> > [<0000000079019739>] kmemdup+0x17/0x40
> > [<00000000d58c1b9d>] intel_bw_duplicate_state+0x1b/0x40 [i915]
> > [<000000007423ed0c>] drm_atomic_get_private_obj_state+0xca/0x140
> > [<00000000100e3533>] intel_bw_atomic_check+0x133/0x350 [i915]
> > [<00000000126d0e0c>] intel_atomic_check+0x1ab7/0x20d0 [i915]
> > [<00000000d5dfc004>] drm_atomic_check_only+0x563/0x810
> > [<00000000c9379611>] drm_atomic_commit+0xe/0x50
> > [<00000000ec82b765>] drm_atomic_helper_disable_all+0x133/0x160
> > [<000000003c44760c>] drm_atomic_helper_shutdown+0x65/0xc0
> > [<00000000414e3e5c>] i915_driver_remove+0xcb/0x130 [i915]
> > [<00000000f8544c2a>] i915_pci_remove+0x19/0x40 [i915]
> > [<000000002dcbd148>] pci_device_remove+0x36/0xb0
> > [<000000003c8c6b0a>] device_release_driver_internal+0xe0/0x1c0
> > [<00000000580e9566>] unbind_store+0xc3/0x120
> > [<00000000869d0df5>] kernfs_fop_write+0x104/0x190
> > [<000000004dc1a355>] vfs_write+0xb9/0x1d0
>
> what I find strange in this is that the last state was allocated by the
> "driver remove" code path.
>
> >
> >Call the drm_atomic_private_obj_fini(), which inturn calls the
> >intel_bw_destroy_state() to make sure the intel_bw_state memory is
> >freed properly.
> >
> >Signed-off-by: Pankaj Bharadiya <pankaj.laxminarayan.bharadiya@xxxxxxxxx>
> >---
> >drivers/gpu/drm/i915/display/intel_bw.c | 5 +++++
> >drivers/gpu/drm/i915/display/intel_bw.h | 1 +
> >drivers/gpu/drm/i915/display/intel_display.c | 2 ++
> >3 files changed, 8 insertions(+)
> >
> >diff --git a/drivers/gpu/drm/i915/display/intel_bw.c b/drivers/gpu/drm/i915/display/intel_bw.c
> >index dcb66a33be9b..b228671d5a5d 100644
> >--- a/drivers/gpu/drm/i915/display/intel_bw.c
> >+++ b/drivers/gpu/drm/i915/display/intel_bw.c
> >@@ -486,3 +486,8 @@ int intel_bw_init(struct drm_i915_private *dev_priv)
> >
> > return 0;
> >}
> >+
> >+void intel_bw_cleanup(struct drm_i915_private *dev_priv)
> >+{
> >+ drm_atomic_private_obj_fini(&dev_priv->bw_obj);
> >+}
> >diff --git a/drivers/gpu/drm/i915/display/intel_bw.h b/drivers/gpu/drm/i915/display/intel_bw.h
> >index 9db10af012f4..20b9ad241802 100644
> >--- a/drivers/gpu/drm/i915/display/intel_bw.h
> >+++ b/drivers/gpu/drm/i915/display/intel_bw.h
> >@@ -25,6 +25,7 @@ struct intel_bw_state {
> >
> >void intel_bw_init_hw(struct drm_i915_private *dev_priv);
> >int intel_bw_init(struct drm_i915_private *dev_priv);
> >+void intel_bw_cleanup(struct drm_i915_private *dev_priv);
> >int intel_bw_atomic_check(struct intel_atomic_state *state);
> >void intel_bw_crtc_update(struct intel_bw_state *bw_state,
> > const struct intel_crtc_state *crtc_state);
> >diff --git a/drivers/gpu/drm/i915/display/intel_display.c b/drivers/gpu/drm/i915/display/intel_display.c
> >index 3190aa27ffdc..756eb90b1bb1 100644
> >--- a/drivers/gpu/drm/i915/display/intel_display.c
> >+++ b/drivers/gpu/drm/i915/display/intel_display.c
> >@@ -17912,6 +17912,8 @@ void intel_modeset_driver_remove(struct drm_i915_private *i915)
> >
> > intel_gmbus_teardown(i915);
> >
> >+ intel_bw_cleanup(i915);
>
> This doesn't seem to match the (reverse) order of
> intel_modeset_init()... but it's actually the gmbus_teardown() that is
> out of place. Did you check if it's not a wrong shutdown ordering?
>
In intel_modeset_init(), intel_gmbus_setup() happens after
intel_bw_init().
I think the patch follows the reverse ordering properly.
Am I missing anything?
I said it seems that it's the gmbus_teardown() that is out of place.
Have you seen my comment above? Why are we duplicating the bw_state on
the module-remove code path?
I think that part is legitimate. Part of the module remove sequence
does an atomic commit to turn everything off. During atomic
transactions, we create duplicates of all modesetting state objects can
be modified; if/when the transaction succeeds, those duplicates are
swapped into the actual driver state and the old objects are destroyed.
Thus in cases like this where we forget to destroy a private object
state, that leaked state structure will be the one allocated during the
very last atomic transaction that happened (i.e., on the driver teardown
codepath).
humn, that makes sense. The new duplicate state will replace the
previous one and hence why we see it in the backtrace, rather than one
allocated previously.
thanks
Lucas De Marchi
and...
Reviewed-by: Lucas De Marchi <lucas.demarchi@xxxxxxxxx>
Lucas De Marchi
Matt
Lucas De Marchi
Thanks,
Pankaj
> thanks
> Lucas De Marchi
>
> >+
> > destroy_workqueue(i915->flip_wq);
> > destroy_workqueue(i915->modeset_wq);
> >
> >--
> >2.23.0
> >
> >_______________________________________________
> >Intel-gfx mailing list
> >Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
> >https://lists.freedesktop.org/mailman/listinfo/intel-gfx
--
Matt Roper
Graphics Software Engineer
VTT-OSGC Platform Enablement
Intel Corporation
(916) 356-2795
_______________________________________________
dri-devel mailing list
dri-devel@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/dri-devel
