Re: [PATCH v2 09/11] drm/mediatek: Add secure flow support to mediatek-drm

["Jason-JH Lin (林睿祥)" <Jason-JH.Lin@mediatek.com>]

[-- Attachment #1: Type: text/html, Size: 12355 bytes --]

[-- Attachment #2: Type: text/plain, Size: 5818 bytes --]

On Mon, 2023-11-06 at 01:27 +0000, CK Hu (胡俊光) wrote:
> On Sun, 2023-11-05 at 13:04 +0000, Jason-JH Lin (林睿祥) wrote:
> > On Tue, 2023-10-31 at 06:01 +0000, CK Hu (胡俊光) wrote:
> > > Hi, Jason:
> > > 
> > > On Mon, 2023-10-23 at 12:45 +0800, Jason-JH.Lin wrote:
> > > > To add secure flow support for mediatek-drm, each crtc have to
> > > > create a secure cmdq mailbox channel. Then cmdq packets with
> > > > display HW configuration will be sent to secure cmdq mailbox
> > > > channel
> > > > and configured in the secure world.
> > > > 
> > > > Each crtc have to use secure cmdq interface to configure some
> > > > secure
> > > > settings for display HW before sending cmdq packets to secure
> > > > cmdq
> > > > mailbox channel.
> > > > 
> > > > If any of fb get from current drm_atomic_state is secure, then
> > > > crtc
> > > > will switch to the secure flow to configure display HW.
> > > > If all fbs are not secure in current drm_atomic_state, then
> > > > crtc
> > > > will
> > > > switch to the normal flow.
> > > > 
> > > > Signed-off-by: Jason-JH.Lin <jason-jh.lin@mediatek.com>
> > > > ---
> > > >  drivers/gpu/drm/mediatek/mtk_drm_crtc.c  | 272
> > > > ++++++++++++++++++++++-
> > > >  drivers/gpu/drm/mediatek/mtk_drm_crtc.h  |   1 +
> > > >  drivers/gpu/drm/mediatek/mtk_drm_plane.c |   7 +
> > > >  3 files changed, 269 insertions(+), 11 deletions(-)
> > > > 
> 
> [snip]
> 
> > > > +
> > > > +#if IS_REACHABLE(CONFIG_MTK_CMDQ)
> > > > +static void mtk_crtc_enable_secure_state(struct drm_crtc
> > > > *crtc)
> > > > +{
> > > > +	enum cmdq_sec_scenario sec_scn = CMDQ_MAX_SEC_COUNT;
> > > > +	int i;
> > > > +	struct mtk_ddp_comp *ddp_first_comp;
> > > > +	struct mtk_drm_crtc *mtk_crtc = to_mtk_crtc(crtc);
> > > > +	u64 sec_engine = 0; /* for hw engine write output
> > > > secure fb */
> > > > +	u64 sec_port = 0; /* for larb port read input secure fb
> > > > */
> > > > +
> > > > +	cmdq_pkt_wfe(&mtk_crtc->sec_cmdq_handle, mtk_crtc-
> > > > >cmdq_event,
> > > > false);
> > > > +
> > > > +	ddp_first_comp = mtk_crtc->ddp_comp[0];
> > > > +	for (i = 0; i < mtk_crtc->layer_nr; i++)
> > > > +		if (mtk_crtc->planes[i].type ==
> > > > DRM_PLANE_TYPE_CURSOR)
> > > > +			sec_port |=
> > > > mtk_ddp_comp_layer_get_sec_port(ddp_first_comp, i);
> > > > +
> > > > +	if (drm_crtc_index(crtc) == 0)
> > > > +		sec_scn = CMDQ_SEC_PRIMARY_DISP;
> > > > +	else if (drm_crtc_index(crtc) == 1)
> > > > +		sec_scn = CMDQ_SEC_SUB_DISP;
> > > > +
> > > > +	cmdq_sec_pkt_set_data(&mtk_crtc->sec_cmdq_handle,
> > > > sec_engine,
> > > > sec_port, sec_scn);
> > > 
> > > In cmdq driver, sec_engine means engine which need dapc. You set
> > > 0
> > > to
> > > sec_engine, does it mean that no engine is protected by dapc? If
> > > OVL
> > > is
> > > not protected by dapc, I think we could use cmdq normal thread to
> > > write
> > > OVL register instead of cmdq secure thread.
> > > 
> > 
> > We enable DPAC protection for the engine that is able to write data
> > to
> > the DRAM address set on their register, such as WDMA and WROT, to
> > avoid
> > their register being set to the normal DRAM address.
> > 
> > We enable larb port protection for the engine that is able to read
> > data
> > from the DRAM address, such as OVL, RDMA and MDP_RDMA, to avoid
> > secure
> > DRAM being read by the non-secure larb port. So we don't need
> > toenable
> > DAPC for these engines.
> > 
> > No mater DAPC protection or larb port protection, they both need to
> > use
> > sec_engine to tell TEE which engines need to be protected.
> > 
> > But OVL is a special HW engine, it can only set its
> > DISP_REG_OVL_SECURE
> > register in [PATCH v2 07/11] to enable its larb port protection, so
> > OVL
> > no need to set the sec_engine. But we'll move that part to the TEE
> > secure world, so that means OVL sec_engine will be set here in the
> > next
> > version.
> > 
> > > It's weird that normal world could decide which engine is
> > > protected
> > > by
> > > dapc. If hacker set 0 for output engine, then outout engine would
> > > not
> > > be protected?
> > > 
> > 
> > If hacker set 0 for output engine, TEE world will check that output
> > engine didn't set sec_engine from normal world by verifying
> > instruction
> > where the output engine instruction set the secure handle.
> > 
> > We still need to set sec_engine to check that all the sec_engine
> > fags
> > are matched to the scenario and instruction verification in the
> > secure
> > world.
> 
> So after secure scenario is set, TEE already have a sec_engine list.
> Let's call it TEE sec_engine list. And normal world has another
> sec_engine list, let's call it normal sec_engine list. Normal world
> pass normal sec_engine list to TEE by cmdq_sec_pkt_set_data() and TEE
> would check normal sec_engine list is identical to TEE sec_engine
> list
> or not. If TEE already have a TEE sec_engine list, I think it's not
> necessary that normal world have another normal sec_engine list. So
> drop this normal sec_engine list parameter.
> 

The TEE sec_engine list is align to the sec_engine list in normal
world. I think the sec_engine for DAPC can be dropped and it can be
enabled/disabled by the TEE sec_engine, but the sec_engine for larb
port won't do that verification in the secure world. If sec_engine for
larb port is not set in normal world, it'll cause iova translation
fault. So we still need this sec_engine for larb port.

Regards,
Jason-JH.Lin

> Regards,
> CK
> 
> > 
> > Regards,
> > Jason-JH.Lin
> > 
> > > Regards,
> > > CK
> > > 
> > > > +
> > > > +	pr_debug("crtc-%d enable secure plane!\n",
> > > > drm_crtc_index(crtc));
> > > > +}
> > > > +#endif
> > > >