Re: [PATCH 3/3] drm/i915: Track OACONTROL register enable/disable during parsing

["Volkin, Bradley D" <bradley.d.volkin@intel.com>]

On Thu, Mar 27, 2014 at 02:58:01PM -0700, Kenneth Graunke wrote:
> On 03/27/2014 11:43 AM, bradley.d.volkin@intel.com wrote:
> > From: Brad Volkin <bradley.d.volkin@intel.com>
> > 
> > There is some thought that the data from the performance counters enabled
> > via OACONTROL should only be available to the process that enabled counting.
> > To limit snooping, require that any batch buffer which sets OACONTROL to a
> > non-zero value also sets it back to 0 before the end of the batch.
> > 
> > This requires limiting OACONTROL writes to happen via MI_LOAD_REGISTER_IMM
> > so that we can access the value being written. This should be in line with
> > the expected use case for writing OACONTROL.
> > 
> > Cc: Kenneth Graunke <kenneth@whitecape.org>
> > Signed-off-by: Brad Volkin <bradley.d.volkin@intel.com>
> > ---
> >  drivers/gpu/drm/i915/i915_cmd_parser.c | 35 ++++++++++++++++++++++++++--------
> >  1 file changed, 27 insertions(+), 8 deletions(-)
> > 
> > diff --git a/drivers/gpu/drm/i915/i915_cmd_parser.c b/drivers/gpu/drm/i915/i915_cmd_parser.c
> > index 2eb2aca..779e14c 100644
> > --- a/drivers/gpu/drm/i915/i915_cmd_parser.c
> > +++ b/drivers/gpu/drm/i915/i915_cmd_parser.c
> > @@ -407,12 +407,7 @@ static const u32 gen7_render_regs[] = {
> >  	REG64(CL_PRIMITIVES_COUNT),
> >  	REG64(PS_INVOCATION_COUNT),
> >  	REG64(PS_DEPTH_COUNT),
> > -	/*
> > -	 * FIXME: This is just to keep mesa working for now, we need to check
> > -	 * that mesa resets this again and that it doesn't use any of the
> > -	 * special modes which write into the gtt.
> > -	 */
> > -	OACONTROL,
> > +	OACONTROL, /* Only allowed for LRI and SRM. See below. */
> >  	REG64(GEN7_SO_NUM_PRIMS_WRITTEN(0)),
> >  	REG64(GEN7_SO_NUM_PRIMS_WRITTEN(1)),
> >  	REG64(GEN7_SO_NUM_PRIMS_WRITTEN(2)),
> > @@ -761,7 +756,8 @@ bool i915_needs_cmd_parser(struct intel_ring_buffer *ring)
> >  static bool check_cmd(const struct intel_ring_buffer *ring,
> >  		      const struct drm_i915_cmd_descriptor *desc,
> >  		      const u32 *cmd,
> > -		      const bool is_master)
> > +		      const bool is_master,
> > +		      bool *oacontrol_set)
> >  {
> >  	if (desc->flags & CMD_DESC_REJECT) {
> >  		DRM_DEBUG_DRIVER("CMD: Rejected command: 0x%08X\n", *cmd);
> > @@ -777,6 +773,23 @@ static bool check_cmd(const struct intel_ring_buffer *ring,
> >  	if (desc->flags & CMD_DESC_REGISTER) {
> >  		u32 reg_addr = cmd[desc->reg.offset] & desc->reg.mask;
> >  
> > +		/*
> > +		 * OACONTROL requires some special handling for writes. We
> > +		 * want to make sure that any batch which enables OA also
> > +		 * disables it before the end of the batch. The goal is to
> > +		 * prevent one process from snooping on the perf data from
> > +		 * another process. To do that, we need to check the value
> > +		 * that will be written to the register. Hence, limit
> > +		 * OACONTROL writes to only MI_LOAD_REGISTER_IMM commands.
> > +		 */
> > +		if (reg_addr == OACONTROL) {
> > +			if (desc->cmd.value == MI_LOAD_REGISTER_MEM)
> > +				return false;
> 
> While I don't need the ability to use LRM on OACONTROL, I don't see any
> specific reason to prohibit it, as long as there's a later LRI of 0.
> 
> I think you could just do:
> 
> if (desc->cmd.value == MI_LOAD_REGISTER_MEM)
>    oacontrol_set = true;
> 
> which will later get cleared if it sees a LRI of 0, and if not, you
> reject the batch.

Fair enough. Rejecting LRM was as much a carryover from thinking we needed to
check bits within the value. I'll make this change if there are no objections.

> 
> > +
> > +			if (desc->cmd.value == MI_LOAD_REGISTER_IMM(1))
> > +				*oacontrol_set = (cmd[2] != 0) ? true : false;
> 
> You shouldn't need to do both != 0 and ? true : false...
> 
> *oacontrol_set = cmd[2] != 0;

Will fix.

> 
> Thanks for implementing this!  That was surprisingly less code than I
> thought.

This is the "you told me there's only one register that needs this" implementation :)
A more general solution would probably be uglier.

Thanks,
Brad

> 
> > +		}
> > +
> >  		if (!valid_reg(ring->reg_table,
> >  			       ring->reg_count, reg_addr)) {
> >  			if (!is_master ||
> > @@ -851,6 +864,7 @@ int i915_parse_cmds(struct intel_ring_buffer *ring,
> >  	u32 *cmd, *batch_base, *batch_end;
> >  	struct drm_i915_cmd_descriptor default_desc = { 0 };
> >  	int needs_clflush = 0;
> > +	bool oacontrol_set = false; /* OACONTROL tracking. See check_cmd() */
> >  
> >  	ret = i915_gem_obj_prepare_shmem_read(batch_obj, &needs_clflush);
> >  	if (ret) {
> > @@ -900,7 +914,7 @@ int i915_parse_cmds(struct intel_ring_buffer *ring,
> >  			break;
> >  		}
> >  
> > -		if (!check_cmd(ring, desc, cmd, is_master)) {
> > +		if (!check_cmd(ring, desc, cmd, is_master, &oacontrol_set)) {
> >  			ret = -EINVAL;
> >  			break;
> >  		}
> > @@ -908,6 +922,11 @@ int i915_parse_cmds(struct intel_ring_buffer *ring,
> >  		cmd += length;
> >  	}
> >  
> > +	if (oacontrol_set) {
> > +		DRM_DEBUG_DRIVER("CMD: batch set OACONTROL but did not clear it\n");
> > +		ret = -EINVAL;
> > +	}
> > +
> >  	if (cmd >= batch_end) {
> >  		DRM_DEBUG_DRIVER("CMD: Got to the end of the buffer w/o a BBE cmd!\n");
> >  		ret = -EINVAL;
> > 
> 
>