* [bug report] drm/vmwgfx: Implement an infrastructure for read-coherent resources
@ 2021-07-29 14:39 Dan Carpenter
2021-07-29 14:51 ` Dan Carpenter
0 siblings, 1 reply; 2+ messages in thread
From: Dan Carpenter @ 2021-07-29 14:39 UTC (permalink / raw)
To: thellstrom; +Cc: Deepak Rawat, dri-devel
Hello Thomas Hellstrom,
The patch fb80edb0d766: "drm/vmwgfx: Implement an infrastructure for
read-coherent resources" from Mar 28, 2019, leads to the following
static checker warning:
drivers/gpu/drm/vmwgfx/vmwgfx_page_dirty.c:461 vmw_bo_vm_fault()
warn: missing conversion: 'page_offset + ((1) << 12)' 'page + byte'
drivers/gpu/drm/vmwgfx/vmwgfx_page_dirty.c:534 vmw_bo_vm_huge_fault()
warn: missing conversion: 'page_offset + ((1) << 12)' 'page + byte'
drivers/gpu/drm/vmwgfx/vmwgfx_page_dirty.c
435 vm_fault_t vmw_bo_vm_fault(struct vm_fault *vmf)
436 {
437 struct vm_area_struct *vma = vmf->vma;
438 struct ttm_buffer_object *bo = (struct ttm_buffer_object *)
439 vma->vm_private_data;
440 struct vmw_buffer_object *vbo =
441 container_of(bo, struct vmw_buffer_object, base);
442 pgoff_t num_prefault;
443 pgprot_t prot;
444 vm_fault_t ret;
445
446 ret = ttm_bo_vm_reserve(bo, vmf);
447 if (ret)
448 return ret;
449
450 num_prefault = (vma->vm_flags & VM_RAND_READ) ? 1 :
451 TTM_BO_VM_NUM_PREFAULT;
452
453 if (vbo->dirty) {
454 pgoff_t allowed_prefault;
455 unsigned long page_offset;
456
457 page_offset = vmf->pgoff -
458 drm_vma_node_start(&bo->base.vma_node);
459 if (page_offset >= bo->resource->num_pages ||
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
page_offset is in terms of pages
460 vmw_resources_clean(vbo, page_offset,
--> 461 page_offset + PAGE_SIZE,
^^^^^^^^^^^^^^^^^^^^^^^
It doesn't make sense to add PAGE_SIZE (which is bytes) to pages. The
code in vmw_bo_vm_huge_fault() has a similar bug.
462 &allowed_prefault)) {
463 ret = VM_FAULT_SIGBUS;
464 goto out_unlock;
465 }
466
467 num_prefault = min(num_prefault, allowed_prefault);
468 }
469
470 /*
471 * If we don't track dirty using the MKWRITE method, make sure
472 * sure the page protection is write-enabled so we don't get
473 * a lot of unnecessary write faults.
474 */
475 if (vbo->dirty && vbo->dirty->method == VMW_BO_DIRTY_MKWRITE)
476 prot = vm_get_page_prot(vma->vm_flags & ~VM_SHARED);
477 else
478 prot = vm_get_page_prot(vma->vm_flags);
479
480 ret = ttm_bo_vm_fault_reserved(vmf, prot, num_prefault, 1);
481 if (ret == VM_FAULT_RETRY && !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT))
482 return ret;
483
484 out_unlock:
485 dma_resv_unlock(bo->base.resv);
486
487 return ret;
488 }
regards,
dan carpenter
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: [bug report] drm/vmwgfx: Implement an infrastructure for read-coherent resources
2021-07-29 14:39 [bug report] drm/vmwgfx: Implement an infrastructure for read-coherent resources Dan Carpenter
@ 2021-07-29 14:51 ` Dan Carpenter
0 siblings, 0 replies; 2+ messages in thread
From: Dan Carpenter @ 2021-07-29 14:51 UTC (permalink / raw)
To: dri-devel
Huh... Vmware is blocking email to Thomas?
"Recipient is not authorized to accept external mail"
This seems like potentially a serious bug and I don't know how to report
it.
regards,
dan carpenter
On Thu, Jul 29, 2021 at 05:39:45PM +0300, Dan Carpenter wrote:
> Hello Thomas Hellstrom,
>
> The patch fb80edb0d766: "drm/vmwgfx: Implement an infrastructure for
> read-coherent resources" from Mar 28, 2019, leads to the following
> static checker warning:
>
> drivers/gpu/drm/vmwgfx/vmwgfx_page_dirty.c:461 vmw_bo_vm_fault()
> warn: missing conversion: 'page_offset + ((1) << 12)' 'page + byte'
>
> drivers/gpu/drm/vmwgfx/vmwgfx_page_dirty.c:534 vmw_bo_vm_huge_fault()
> warn: missing conversion: 'page_offset + ((1) << 12)' 'page + byte'
>
> drivers/gpu/drm/vmwgfx/vmwgfx_page_dirty.c
> 435 vm_fault_t vmw_bo_vm_fault(struct vm_fault *vmf)
> 436 {
> 437 struct vm_area_struct *vma = vmf->vma;
> 438 struct ttm_buffer_object *bo = (struct ttm_buffer_object *)
> 439 vma->vm_private_data;
> 440 struct vmw_buffer_object *vbo =
> 441 container_of(bo, struct vmw_buffer_object, base);
> 442 pgoff_t num_prefault;
> 443 pgprot_t prot;
> 444 vm_fault_t ret;
> 445
> 446 ret = ttm_bo_vm_reserve(bo, vmf);
> 447 if (ret)
> 448 return ret;
> 449
> 450 num_prefault = (vma->vm_flags & VM_RAND_READ) ? 1 :
> 451 TTM_BO_VM_NUM_PREFAULT;
> 452
> 453 if (vbo->dirty) {
> 454 pgoff_t allowed_prefault;
> 455 unsigned long page_offset;
> 456
> 457 page_offset = vmf->pgoff -
> 458 drm_vma_node_start(&bo->base.vma_node);
> 459 if (page_offset >= bo->resource->num_pages ||
> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
> page_offset is in terms of pages
>
> 460 vmw_resources_clean(vbo, page_offset,
> --> 461 page_offset + PAGE_SIZE,
> ^^^^^^^^^^^^^^^^^^^^^^^
> It doesn't make sense to add PAGE_SIZE (which is bytes) to pages. The
> code in vmw_bo_vm_huge_fault() has a similar bug.
>
> 462 &allowed_prefault)) {
> 463 ret = VM_FAULT_SIGBUS;
> 464 goto out_unlock;
> 465 }
> 466
> 467 num_prefault = min(num_prefault, allowed_prefault);
> 468 }
> 469
> 470 /*
> 471 * If we don't track dirty using the MKWRITE method, make sure
> 472 * sure the page protection is write-enabled so we don't get
> 473 * a lot of unnecessary write faults.
> 474 */
> 475 if (vbo->dirty && vbo->dirty->method == VMW_BO_DIRTY_MKWRITE)
> 476 prot = vm_get_page_prot(vma->vm_flags & ~VM_SHARED);
> 477 else
> 478 prot = vm_get_page_prot(vma->vm_flags);
> 479
> 480 ret = ttm_bo_vm_fault_reserved(vmf, prot, num_prefault, 1);
> 481 if (ret == VM_FAULT_RETRY && !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT))
> 482 return ret;
> 483
> 484 out_unlock:
> 485 dma_resv_unlock(bo->base.resv);
> 486
> 487 return ret;
> 488 }
>
> regards,
> dan carpenter
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2021-07-29 14:52 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-07-29 14:39 [bug report] drm/vmwgfx: Implement an infrastructure for read-coherent resources Dan Carpenter
2021-07-29 14:51 ` Dan Carpenter
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.