drm/msm: Fix refcnt underflow in error path

If we hit an error path in GEM obj creation before msm_gem_new_handle()
updates obj->resv to point to the gpuvm resv object, then obj->resv
still points to &obj->_resv.  In this case we don't want to decrement
the refcount of the object being freed (since the refcnt is already
zero).  This fixes the following splat:

   ------------[ cut here ]------------
   refcount_t: underflow; use-after-free.
   WARNING: CPU: 9 PID: 7013 at lib/refcount.c:28 refcount_warn_saturate+0xf4/0x148
   Modules linked in: uinput snd_seq_dummy snd_hrtimer aes_ce_ccm snd_soc_wsa884x regmap_sdw q6prm_clocks q6apm_lpass_da>
    qcom_pil_info i2c_hid drm_kms_helper qcom_common qcom_q6v5 phy_snps_eusb2 qcom_geni_serial drm qcom_sysmon pinctrl_s>
   CPU: 9 UID: 1000 PID: 7013 Comm: deqp-vk Not tainted 6.16.0-rc4-debug+ torvalds#25 PREEMPT(voluntary)
   Hardware name: LENOVO 83ED/LNVNB161216, BIOS NHCN53WW 08/02/2024
   pstate: 6140000 (nZCv daif +PAN -UAO -TCO +DIT -SSBS BTYPE=--)
   pc : refcount_warn_saturate+0xf4/0x148
   lr : refcount_warn_saturate+0xf4/0x148
   sp : ffff8000a2073920
   x29: ffff8000a2073920 x28: 0000000000000010 x27: 0000000000000010
   x26: 0000000000000042 x25: ffff000810e09800 x24: 0000000000000010
   x23: ffff8000a2073b94 x22: ffff000ddb22de00 x21: ffff000ddb22dc00
   x20: ffff000ddb22ddf8 x19: ffff0008024934e0 x18: 000000000000000a
   x17: 0000000000000000 x16: ffff9f8c67d77340 x15: 0000000000000000
   x14: 00000000ffffffff x13: 2e656572662d7265 x12: 7466612d65737520
   x11: 3b776f6c66726564 x10: 00000000ffff7fff x9 : ffff9f8c67506c70
   x8 : ffff9f8c69fa26f0 x7 : 00000000000bffe8 x6 : c0000000ffff7fff
   x5 : ffff000f53e14548 x4 : ffff6082ea2b2000 x3 : ffff0008b86ab080
   x2 : 0000000000000000 x1 : 0000000000000000 x0 : ffff0008b86ab080
   Call trace:
    refcount_warn_saturate+0xf4/0x148 (P)
    msm_gem_free_object+0x248/0x260 [msm]
    drm_gem_object_free+0x24/0x40 [drm]
    msm_gem_new+0x1c4/0x1e0 [msm]
    msm_gem_new_handle+0x3c/0x1a0 [msm]
    msm_ioctl_gem_new+0x38/0x70 [msm]
    drm_ioctl_kernel+0xc8/0x138 [drm]
    drm_ioctl+0x2c8/0x618 [drm]
    __arm64_sys_ioctl+0xac/0x108
    invoke_syscall.constprop.0+0x64/0xe8
    el0_svc_common.constprop.0+0x40/0xe8
    do_el0_svc+0x24/0x38
    el0_svc+0x54/0x1d8
    el0t_64_sync_handler+0x10c/0x138
    el0t_64_sync+0x19c/0x1a0
   irq event stamp: 3698694
   hardirqs last  enabled at (3698693): [<ffff9f8c675021dc>] __up_console_sem+0x74/0x90
   hardirqs last disabled at (3698694): [<ffff9f8c68ce8164>] el1_dbg+0x24/0x90
   softirqs last  enabled at (3697578): [<ffff9f8c6744ec5c>] handle_softirqs+0x454/0x4b0
   softirqs last disabled at (3697567): [<ffff9f8c67360244>] __do_softirq+0x1c/0x28
   ---[ end trace 0000000000000000 ]---

Fixes: b58e12a ("drm/msm: Add _NO_SHARE flag")
Signed-off-by: Rob Clark <[email protected]>
Patchwork: https://patchwork.freedesktop.org/patch/665355/

Author: Rob Clark

drivers/gpu/drm/msm/msm_gem.c

@@ -1120,10 +1120,12 @@ static void msm_gem_free_object(struct drm_gem_object *obj)
 		put_pages(obj);
 	}
 
-	if (msm_obj->flags & MSM_BO_NO_SHARE) {
+	if (obj->resv != &obj->_resv) {
 		struct drm_gem_object *r_obj =
 			container_of(obj->resv, struct drm_gem_object, _resv);
 
+		WARN_ON(!(msm_obj->flags & MSM_BO_NO_SHARE));
+
 		/* Drop reference we hold to shared resv obj: */
 		drm_gem_object_put(r_obj);
 	}