diff mbox series

[v3,05/28] drm/i915/gvt: Explicitly check that vGPU is attached before shadowing

Message ID 20230513003600.818142-6-seanjc@google.com
State New
Headers
Series drm/i915/gvt: KVM: KVMGT fixes and page-track cleanups |

Commit Message

Sean Christopherson May 13, 2023, 12:35 a.m. UTC 
  Move the check that a vGPU is attacked from is_2MB_gtt_possible() to its
sole caller, ppgtt_populate_shadow_entry().  All of the paths in
ppgtt_populate_shadow_entry() eventually check for attachment by way of
intel_gvt_dma_map_guest_page(), but explicitly checking can avoid
unnecessary work and will make it more obvious that a future cleanup of
is_2MB_gtt_possible() isn't introducing a bug.

Signed-off-by: Sean Christopherson <seanjc@google.com>
---
 drivers/gpu/drm/i915/gvt/gtt.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

Comments

Wang, Zhi A May 15, 2023, 11:24 a.m. UTC | #1 
On 5/13/2023 8:35 AM, Sean Christopherson wrote:
> Move the check that a vGPU is attacked from is_2MB_gtt_possible() to its
> sole caller, ppgtt_populate_shadow_entry().  All of the paths in
> ppgtt_populate_shadow_entry() eventually check for attachment by way of
> intel_gvt_dma_map_guest_page(), but explicitly checking can avoid
> unnecessary work and will make it more obvious that a future cleanup of
> is_2MB_gtt_possible() isn't introducing a bug.
> 

It might be better move this check to shadow_ppgtt_mm() which is used
in both shadow page table creation and pinning path so that the path
can bail out even earlier when creating a shadow page table but a vGPU
has not been attached to KVM yet.

> Signed-off-by: Sean Christopherson <seanjc@google.com>
> ---
>   drivers/gpu/drm/i915/gvt/gtt.c | 5 +++--
>   1 file changed, 3 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/gpu/drm/i915/gvt/gtt.c b/drivers/gpu/drm/i915/gvt/gtt.c
> index 5426a27c1b71..2aed31b497c9 100644
> --- a/drivers/gpu/drm/i915/gvt/gtt.c
> +++ b/drivers/gpu/drm/i915/gvt/gtt.c
> @@ -1163,8 +1163,6 @@ static int is_2MB_gtt_possible(struct intel_vgpu *vgpu,
>   	if (!HAS_PAGE_SIZES(vgpu->gvt->gt->i915, I915_GTT_PAGE_SIZE_2M))
>   		return 0;
>   
> -	if (!test_bit(INTEL_VGPU_STATUS_ATTACHED, vgpu->status))
> -		return -EINVAL;
>   	pfn = gfn_to_pfn(vgpu->vfio_device.kvm, ops->get_pfn(entry));
>   	if (is_error_noslot_pfn(pfn))
>   		return -EINVAL;
> @@ -1277,6 +1275,9 @@ static int ppgtt_populate_shadow_entry(struct intel_vgpu *vgpu,
>   	if (!pte_ops->test_present(ge))
>   		return 0;
>   
> +	if (!test_bit(INTEL_VGPU_STATUS_ATTACHED, vgpu->status))
> +		return -EINVAL;
> +
>   	gfn = pte_ops->get_pfn(ge);
>   
>   	switch (ge->type) {
Sean Christopherson May 15, 2023, 5:57 p.m. UTC | #2 
On Mon, May 15, 2023, Zhi A Wang wrote:
> On 5/13/2023 8:35 AM, Sean Christopherson wrote:
> > Move the check that a vGPU is attacked from is_2MB_gtt_possible() to its
> > sole caller, ppgtt_populate_shadow_entry().  All of the paths in
> > ppgtt_populate_shadow_entry() eventually check for attachment by way of
> > intel_gvt_dma_map_guest_page(), but explicitly checking can avoid
> > unnecessary work and will make it more obvious that a future cleanup of
> > is_2MB_gtt_possible() isn't introducing a bug.
> > 
> 
> It might be better move this check to shadow_ppgtt_mm() which is used
> in both shadow page table creation and pinning path so that the path
> can bail out even earlier when creating a shadow page table but a vGPU
> has not been attached to KVM yet.

Ah, yes, that'll work.  I traced through all of the paths that lead to
ppgtt_populate_shadow_entry(), and shadow_ppgtt_mm() is the only caller that isn't
already gated by INTEL_VGPU_STATUS_ATTACHED or INTEL_VGPU_STATUS_ACTIVE (ACTIVE
is set iff ATTACHED is set).

I'll move the check up to shadow_ppgtt_mm() in the next version.

Thanks!


workload_thread() <= pick_next_workload() => INTEL_VGPU_STATUS_ACTIVE
|
-> dispatch_workload()
   |
   |-> prepare_workload()
       |
       -> intel_vgpu_sync_oos_pages()
       |  |
       |  |-> ppgtt_set_guest_page_sync()
       |      |
       |      |-> sync_oos_page()
       |          |
       |          |-> ppgtt_populate_shadow_entry()
       |
       |-> intel_vgpu_flush_post_shadow()
           |
1:         |-> ppgtt_handle_guest_write_page_table()
               |
               |-> ppgtt_handle_guest_entry_add()
                   |
2:                 | -> ppgtt_populate_spt_by_guest_entry()
                   |    |
                   |    |-> ppgtt_populate_spt()
                   |        |
                   |        |-> ppgtt_populate_shadow_entry()
                   |            |
                   |            |-> ppgtt_populate_spt_by_guest_entry() [see 2]
                   |
                   |-> ppgtt_populate_shadow_entry()


kvmgt_page_track_write()  <= KVM callback => INTEL_VGPU_STATUS_ATTACHED
|
|-> intel_vgpu_page_track_handler()
    |
    |-> ppgtt_write_protection_handler()
        |
        |-> ppgtt_handle_guest_write_page_table_bytes()
            |
            |-> ppgtt_handle_guest_write_page_table() [see 1]
diff mbox series

Patch

diff --git a/drivers/gpu/drm/i915/gvt/gtt.c b/drivers/gpu/drm/i915/gvt/gtt.c
index 5426a27c1b71..2aed31b497c9 100644
--- a/drivers/gpu/drm/i915/gvt/gtt.c
+++ b/drivers/gpu/drm/i915/gvt/gtt.c
@@ -1163,8 +1163,6 @@  static int is_2MB_gtt_possible(struct intel_vgpu *vgpu,
 	if (!HAS_PAGE_SIZES(vgpu->gvt->gt->i915, I915_GTT_PAGE_SIZE_2M))
 		return 0;
 
-	if (!test_bit(INTEL_VGPU_STATUS_ATTACHED, vgpu->status))
-		return -EINVAL;
 	pfn = gfn_to_pfn(vgpu->vfio_device.kvm, ops->get_pfn(entry));
 	if (is_error_noslot_pfn(pfn))
 		return -EINVAL;
@@ -1277,6 +1275,9 @@  static int ppgtt_populate_shadow_entry(struct intel_vgpu *vgpu,
 	if (!pte_ops->test_present(ge))
 		return 0;
 
+	if (!test_bit(INTEL_VGPU_STATUS_ATTACHED, vgpu->status))
+		return -EINVAL;
+
 	gfn = pte_ops->get_pfn(ge);
 
 	switch (ge->type) {