Re: [PATCH] drm/xlnx: fix kmemleak by sending vblank_event in atomic_disable

From: Daniel Vetter
Date: Wed Feb 03 2021 - 05:43:00 EST


On Tue, Feb 02, 2021 at 02:41:21PM +0800, quanyang.wang@xxxxxxxxxxxxx wrote:
> From: Quanyang Wang <quanyang.wang@xxxxxxxxxxxxx>
>
> When running xrandr to change resolution of DP, the kmemleak as below
> can be observed:
>
> unreferenced object 0xffff00080a351000 (size 256):
> comm "Xorg", pid 248, jiffies 4294899614 (age 19.960s)
> hex dump (first 32 bytes):
> 98 a0 bc 01 08 00 ff ff 01 00 00 00 00 00 00 00 ................
> ff ff ff ff 00 00 00 00 00 00 00 00 00 00 00 00 ................
> backtrace:
> [<00000000e0bd0f69>] kmemleak_alloc+0x30/0x40
> [<00000000cde2f318>] kmem_cache_alloc+0x3d4/0x588
> [<0000000088ea9bd7>] drm_atomic_helper_setup_commit+0x84/0x5f8
> [<000000002290a264>] drm_atomic_helper_commit+0x58/0x388
> [<00000000f6ea78c3>] drm_atomic_commit+0x4c/0x60
> [<00000000c8e0725e>] drm_atomic_connector_commit_dpms+0xe8/0x110
> [<0000000020ade187>] drm_mode_obj_set_property_ioctl+0x1b0/0x450
> [<00000000918206d6>] drm_connector_property_set_ioctl+0x3c/0x68
> [<000000008d51e7a5>] drm_ioctl_kernel+0xc4/0x118
> [<000000002a819b75>] drm_ioctl+0x214/0x448
> [<000000008ca4e588>] __arm64_sys_ioctl+0xa8/0xf0
> [<0000000034e15a35>] el0_svc_common.constprop.0+0x74/0x190
> [<000000001b93d916>] do_el0_svc+0x24/0x90
> [<00000000ce9230e0>] el0_svc+0x14/0x20
> [<00000000e3607d82>] el0_sync_handler+0xb0/0xb8
> [<000000003e79c15f>] el0_sync+0x174/0x180
>
> This is because there is a scenario that a drm_crtc_commit commit is
> allocated but not freed. The drm subsystem require/release references
> to a CRTC commit by calling drm_crtc_commit_get/put, and when
> drm_crtc_commit_put find that commit.ref.refcount is zero, it will
> call __drm_crtc_commit_free to free this CRTC commit. Among these
> drm_crtc_commit_get/put pairs, there is a drm_crtc_commit_get in
> drm_atomic_helper_setup_commit as below:
>
> ...
> new_crtc_state->event->base.completion = &commit->flip_done;
> new_crtc_state->event->base.completion_release = release_crtc_commit;
> drm_crtc_commit_get(commit);
> ...
>
> This reference to the CRTC commit should be released at the function
> release_crtc_commit by calling e->completion_release(e->completion) in
> drm_send_event_locked. So we need to call drm_send_event_locked at
> two places: handling vblank event in the irq handler and the crtc disable
> helper. But in zynqmp_disp_crtc_atomic_disable, it only marks the flip
> is done and not call drm_crtc_commit_put. This result that the refcount
> of this commit is always non-zero and this commit will never be freed.
>
> Since the function drm_crtc_send_vblank_event has operations both sending
> a flip_done signal and releasing reference to the CRTC commit, let's use
> it instead.
>
> Signed-off-by: Quanyang Wang <quanyang.wang@xxxxxxxxxxxxx>

Yeah that's the way to do it, not the hack. Thanks for your patch, I'll
push it to drm-misc-fixes with Cc: stable@xxxxxxxxxxxxxxx

Cheers, Daniel


> ---
> drivers/gpu/drm/xlnx/zynqmp_disp.c | 15 +++++++--------
> 1 file changed, 7 insertions(+), 8 deletions(-)
>
> diff --git a/drivers/gpu/drm/xlnx/zynqmp_disp.c b/drivers/gpu/drm/xlnx/zynqmp_disp.c
> index 68cc4ffff969..ee7657a48e6a 100644
> --- a/drivers/gpu/drm/xlnx/zynqmp_disp.c
> +++ b/drivers/gpu/drm/xlnx/zynqmp_disp.c
> @@ -1396,19 +1396,11 @@ static void zynqmp_disp_enable(struct zynqmp_disp *disp)
> */
> static void zynqmp_disp_disable(struct zynqmp_disp *disp)
> {
> - struct drm_crtc *crtc = &disp->crtc;
> -
> zynqmp_disp_audio_disable(&disp->audio);
>
> zynqmp_disp_avbuf_disable_audio(&disp->avbuf);
> zynqmp_disp_avbuf_disable_channels(&disp->avbuf);
> zynqmp_disp_avbuf_disable(&disp->avbuf);
> -
> - /* Mark the flip is done as crtc is disabled anyway */
> - if (crtc->state->event) {
> - complete_all(crtc->state->event->base.completion);
> - crtc->state->event = NULL;
> - }
> }
>
> static inline struct zynqmp_disp *crtc_to_disp(struct drm_crtc *crtc)
> @@ -1499,6 +1491,13 @@ zynqmp_disp_crtc_atomic_disable(struct drm_crtc *crtc,
>
> drm_crtc_vblank_off(&disp->crtc);
>
> + spin_lock_irq(&crtc->dev->event_lock);
> + if (crtc->state->event) {
> + drm_crtc_send_vblank_event(crtc, crtc->state->event);
> + crtc->state->event = NULL;
> + }
> + spin_unlock_irq(&crtc->dev->event_lock);
> +
> clk_disable_unprepare(disp->pclk);
> pm_runtime_put_sync(disp->dev);
> }
> --
> 2.25.1
>

--
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch