Re: kern/52347: ww mutex class mismatch

[coypu%sdf.org@localhost]

The following reply was made to PR kern/52347; it has been noted by GNATS.

From: coypu%sdf.org@localhost
To: gnats-bugs%NetBSD.org@localhost
Cc: 
Subject: Re: kern/52347: ww mutex class mismatch
Date: Wed, 28 Jun 2017 17:26:50 +0000

 One thing I see in linux is that it has (and we don't)
 
 in radeon_cs_ioctl:
         if (rdev->in_reset) {
                 up_read(&rdev->exclusive_lock);
                 r = radeon_gpu_reset(rdev);
                 if (!r)
                         r = -EAGAIN;
                 return r;
         }
 
 And in radeon_gpu_rest (which we do have)
 
 	rdev->in_reset = true;
 
 	... precarious things, I assume ...
 	drm_helper_resume_force_mode(rdev->ddev);
 
 	rdev->in_reset = false;
 
 We do drm_helper_resume_force_mode etc without in_reset without
 something blocking entry to radeon_cs_ioctl.
 
 It's possible I misunderstand rwlocks (hence the question to tech-kern)
 and the lock prevents this too.