On 11/11/2015 03:08 PM, Alexandre Courbot wrote: > On Wed, Nov 11, 2015 at 9:19 AM, Ben Skeggs wrote: >> On 11/09/2015 05:37 PM, Alexandre Courbot wrote: >>> The LRU list used for recycling CPU mappings was handling concurrency >>> very poorly. For instance, if an instobj was acquired twice before being >>> released once, it would end up into the LRU list even though there is >>> still a client accessing it. >>> >>> This patch fixes this by properly counting how many clients are >>> currently using a given instobj. >> Out of curiosity, which instobjs are being accessed concurrently? > > PGTs it seems - at least this is what dumping a stack when detecting a > concurrent usage on an instobj indicates: Ah, of course. That makes sense, as the mmu code has its own methods of dealing with concurrent access to particular areas of a page table. I'll merge your fix shortly, after another quick eyeball. Thanks, Ben. > > [ 270.547052] [] (gk20a_instobj_acquire [nouveau]) from > [] (gf100_vm_map_sg+0x58/0x124 [nouveau]) > [ 270.557568] [] (gf100_vm_map_sg [nouveau]) from > [] (nvkm_vm_map+0x300/0x3bc [nouveau]) > [ 270.567333] [] (nvkm_vm_map [nouveau]) from [] > (nouveau_bo_vma_add+0x74/0xa0 [nouveau]) > [ 270.577189] [] (nouveau_bo_vma_add [nouveau]) from > [] (nouveau_gem_object_open+0x124/0x158 [nouveau]) > [ 270.588196] [] (nouveau_gem_object_open [nouveau]) from > [] (drm_gem_handle_create_tail+0x104/0x19c) > [ 270.599025] [] (drm_gem_handle_create_tail) from > [] (nouveau_gem_ioctl_new+0x90/0x18c [nouveau]) > [ 270.609594] [] (nouveau_gem_ioctl_new [nouveau]) from > [] (drm_ioctl+0x284/0x440) > [ 270.618777] [] (drm_ioctl) from [] > (nouveau_drm_ioctl+0x54/0x98 [nouveau]) > [ 270.627441] [] (nouveau_drm_ioctl [nouveau]) from > [] (do_vfs_ioctl+0x458/0x6dc) > [ 270.636471] [] (do_vfs_ioctl) from [] > (SyS_ioctl+0x34/0x5c) > [ 270.643767] [] (SyS_ioctl) from [] > (ret_fast_syscall+0x0/0x3c) >