Donate to e Foundation | Murena handsets with /e/OS | Own a part of Murena! Learn more

Commit 6b16d08c authored by Kamal Agrawal's avatar Kamal Agrawal Committed by Hareesh Gundu
Browse files

msm: kgsl: Remove sysfs entries after releasing memory 



Consider a scenario when device is in OOM situation. Assume a process is
doing sysfs operation, grabs kernfs_mutex and wants to allocate memory.
LMKD tries to kill processes but kgsl processes will be blocked waiting
for kernfs_mutex in sysfs_remove_file.

kgsl_process_private_close > kgsl_process_uninit_sysfs > sysfs_remove_file

KGSL won't free up memory as it is done after sysfs removal leading to a
livelock. Fix it by releasing memory before removing sysfs entries.

Change-Id: I99640d7a653faffa671d5b035abb78e9473da12e
Signed-off-by: default avatarKamal Agrawal <kamaagra@codeaurora.org>
parent f9416d64

drivers/gpu/msm/kgsl.c

+4 −8
Original line number Diff line number Diff line
@@ -1057,11 +1057,14 @@ static void kgsl_process_private_close(struct kgsl_device_private *dev_priv, 
	 * directories and garbage collect any outstanding resources

	 */



	kgsl_process_uninit_sysfs(private);

	process_release_memory(private);



	/* Release all syncsource objects from process private */

	kgsl_syncsource_process_release_syncsources(private);



	debugfs_remove_recursive(private->debug_root);

	kgsl_process_uninit_sysfs(private);



	/* When using global pagetables, do not detach global pagetable */

	if (private->pagetable->name != KGSL_MMU_GLOBAL_PT)

		kgsl_mmu_detach_pagetable(private->pagetable);
@@ -1071,15 +1074,8 @@ static void kgsl_process_private_close(struct kgsl_device_private *dev_priv, 
	list_del(&private->list);

	spin_unlock(&kgsl_driver.proclist_lock);



	debugfs_remove_recursive(private->debug_root);



	/*

	 * Unlock the mutex before releasing the memory - this prevents a

	 * deadlock with the IOMMU mutex if a page fault occurs.

	 */

	mutex_unlock(&kgsl_driver.process_mutex);



	process_release_memory(private);

	kgsl_process_private_put(private);

}