Re: [PATCH] drm/msm: Set max segment size earlier

[Rob Clark]

On Mon, May 1, 2023 at 1:56 PM Dmitry Baryshkov
<dmitry.baryshkov@xxxxxxxxxx> wrote:
>
> On 01/05/2023 23:44, Rob Clark wrote:
> > From: Rob Clark <robdclark@xxxxxxxxxxxx>
> >
> > Fixes the following splat on a6xx gen2+ (a640, a650, a660 families),
> > a6xx gen1 has smaller GMU allocations so they fit under the default
> > 64K max segment size.
> >
> >     ------------[ cut here ]------------
> >     DMA-API: msm_dpu ae01000.display-controller: mapping sg segment longer than device claims to support [len=126976] [max=65536]
> >     WARNING: CPU: 5 PID: 9 at kernel/dma/debug.c:1160 debug_dma_map_sg+0x288/0x314
> >     Modules linked in:
> >     CPU: 5 PID: 9 Comm: kworker/u16:0 Not tainted 6.3.0-rc2-debug+ #629
> >     Hardware name: Google Villager (rev1+) with LTE (DT)
> >     Workqueue: events_unbound deferred_probe_work_func
> >     pstate: 60400009 (nZCv daif +PAN -UAO -TCO -DIT -SSBS BTYPE=--)
> >     pc : debug_dma_map_sg+0x288/0x314
> >     lr : debug_dma_map_sg+0x288/0x314
> >     sp : ffffffc00809b560
> >     x29: ffffffc00809b560 x28: 0000000000000060 x27: 0000000000000000
> >     x26: 0000000000010000 x25: 0000000000000004 x24: 0000000000000004
> >     x23: ffffffffffffffff x22: ffffffdb31693cc0 x21: ffffff8080935800
> >     x20: ffffff8087417400 x19: ffffff8087a45010 x18: 0000000000000000
> >     x17: 0000000000000000 x16: 0000000000000000 x15: 0000000000010000
> >     x14: 0000000000000001 x13: ffffffffffffffff x12: ffffffffffffffff
> >     x11: 0000000000000000 x10: 000000000000000a x9 : ffffffdb2ff05e14
> >     x8 : ffffffdb31275000 x7 : ffffffdb2ff08908 x6 : 0000000000000000
> >     x5 : 0000000000000001 x4 : ffffffdb2ff08a74 x3 : ffffffdb31275008
> >     x2 : 0000000000000000 x1 : 0000000000000000 x0 : ffffff80803a9a80
> >     Call trace:
> >      debug_dma_map_sg+0x288/0x314
> >      __dma_map_sg_attrs+0x80/0xe4
> >      dma_map_sgtable+0x30/0x4c
> >      get_pages+0x1d4/0x1e4
> >      msm_gem_pin_pages_locked+0xbc/0xf8
> >      msm_gem_pin_vma_locked+0x58/0xa0
> >      msm_gem_get_and_pin_iova_range+0x98/0xac
> >      a6xx_gmu_memory_alloc+0x7c/0x128
> >      a6xx_gmu_init+0x16c/0x9b0
> >      a6xx_gpu_init+0x38c/0x3e4
> >      adreno_bind+0x214/0x264
> >      component_bind_all+0x128/0x1f8
> >      msm_drm_bind+0x2b8/0x608
> >      try_to_bring_up_aggregate_device+0x88/0x1a4
> >      __component_add+0xec/0x13c
> >      component_add+0x1c/0x28
> >      dp_display_probe+0x3f8/0x43c
> >      platform_probe+0x70/0xc4
> >      really_probe+0x148/0x280
> >      __driver_probe_device+0xc8/0xe0
> >      driver_probe_device+0x44/0x100
> >      __device_attach_driver+0x64/0xdc
> >      bus_for_each_drv+0xb0/0xd8
> >      __device_attach+0xd8/0x168
> >      device_initial_probe+0x1c/0x28
> >      bus_probe_device+0x44/0xb0
> >      deferred_probe_work_func+0xc8/0xe0
> >      process_one_work+0x2e0/0x488
> >      process_scheduled_works+0x4c/0x50
> >      worker_thread+0x218/0x274
> >      kthread+0xf0/0x100
> >      ret_from_fork+0x10/0x20
> >     irq event stamp: 293712
> >     hardirqs last  enabled at (293711): [<ffffffdb2ff0893c>] vprintk_emit+0x160/0x25c
> >     hardirqs last disabled at (293712): [<ffffffdb30b48130>] el1_dbg+0x24/0x80
> >     softirqs last  enabled at (279520): [<ffffffdb2fe10420>] __do_softirq+0x21c/0x4bc
> >     softirqs last disabled at (279515): [<ffffffdb2fe16708>] ____do_softirq+0x18/0x24
> >     ---[ end trace 0000000000000000 ]---
> >
> > Signed-off-by: Rob Clark <robdclark@xxxxxxxxxxxx>
>
> I think this should be:
>
> Fixes: db735fc4036b ("drm/msm: Set dma maximum segment size for mdss")

yeah, or perhaps just that commit didn't fix the issue hard enough.. I
was thinking that it was introduced by memory allocations out of
hw_init() but actually this has been here the whole time (on newer
a6xx gens).  There was an internal bug about it, but somehow it didn't
get routed to me.

BR,
-R

> Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@xxxxxxxxxx>
>
> > ---
> >   drivers/gpu/drm/msm/msm_drv.c | 4 ++--
> >   1 file changed, 2 insertions(+), 2 deletions(-)
> >
> > diff --git a/drivers/gpu/drm/msm/msm_drv.c b/drivers/gpu/drm/msm/msm_drv.c
> > index 3a74b5653e96..6dec1a3534f2 100644
> > --- a/drivers/gpu/drm/msm/msm_drv.c
> > +++ b/drivers/gpu/drm/msm/msm_drv.c
> > @@ -440,27 +440,27 @@ static int msm_drm_init(struct device *dev, const struct drm_driver *drv)
> >       fs_reclaim_acquire(GFP_KERNEL);
> >       might_lock(&priv->lru.lock);
> >       fs_reclaim_release(GFP_KERNEL);
> >
> >       drm_mode_config_init(ddev);
> >
> >       ret = msm_init_vram(ddev);
> >       if (ret)
> >               goto err_drm_dev_put;
> >
> > +     dma_set_max_seg_size(dev, UINT_MAX);
> > +
> >       /* Bind all our sub-components: */
> >       ret = component_bind_all(dev, ddev);
> >       if (ret)
> >               goto err_drm_dev_put;
> >
> > -     dma_set_max_seg_size(dev, UINT_MAX);
> > -
> >       msm_gem_shrinker_init(ddev);
> >
> >       if (priv->kms_init) {
> >               ret = priv->kms_init(ddev);
> >               if (ret) {
> >                       DRM_DEV_ERROR(dev, "failed to load kms\n");
> >                       priv->kms = NULL;
> >                       goto err_msm_uninit;
> >               }
> >               kms = priv->kms;
>
> --
> With best wishes
> Dmitry
>