RE: [PATCH v2 05/16] remoteproc: modify rproc_handle_carveout to support preallocated region

[Loic PALLARDY]

> -----Original Message-----
> From: Bjorn Andersson [mailto:bjorn.andersson@xxxxxxxxxx]
> Sent: Thursday, December 14, 2017 1:59 AM
> To: Loic PALLARDY <loic.pallardy@xxxxxx>
> Cc: ohad@xxxxxxxxxx; linux-remoteproc@xxxxxxxxxxxxxxx; linux-
> kernel@xxxxxxxxxxxxxxx; Arnaud POULIQUEN <arnaud.pouliquen@xxxxxx>;
> benjamin.gaignard@xxxxxxxxxx
> Subject: Re: [PATCH v2 05/16] remoteproc: modify rproc_handle_carveout to
> support preallocated region
> 
> On Thu 30 Nov 08:46 PST 2017, Loic Pallardy wrote:
> 
> > In current version rproc_handle_carveout function support only dynamic
> > region allocation.
> > This patch extends rproc_handle_carveout function to support different
> carveout
> > configurations:
> > - fixed DA and fixed PA: check if already part of pre-registered carveouts
> > (platform driver). If no, return error.
> > - fixed DA and any PA: check if already part of pre-allocated carveouts
> > (platform driver). If not found and rproc supports iommu, continue with
> > dynamic allocation (DA will be used for iommu programming), else return
> > error as no way to force DA.
> > - any DA and any PA: use original dynamic allocation
> >
> > Signed-off-by: Loic Pallardy <loic.pallardy@xxxxxx>
> > ---
> >  drivers/remoteproc/remoteproc_core.c | 40
> ++++++++++++++++++++++++++++++++++++
> >  1 file changed, 40 insertions(+)
> >
> > diff --git a/drivers/remoteproc/remoteproc_core.c
> b/drivers/remoteproc/remoteproc_core.c
> > index 78525d1..515a17a 100644
> > --- a/drivers/remoteproc/remoteproc_core.c
> > +++ b/drivers/remoteproc/remoteproc_core.c
> > @@ -184,6 +184,10 @@ void *rproc_da_to_va(struct rproc *rproc, u64 da,
> int len)
> >  	struct rproc_mem_entry *carveout;
> >  	void *ptr = NULL;
> >
> > +	/*
> > +	 * da_to_va platform driver is deprecated. Driver should register
> > +	 * carveout thanks to rproc_add_carveout function
> > +	 */
> 
> I think this comment is unrelated to the rest of this patch. I also
> think that at the end of the carveout-rework we should have a patch
> removing this ops.

I'll remove this comment and add a da_to_va clean-up patch at the end of the series

> 
> >  	if (rproc->ops->da_to_va) {
> >  		ptr = rproc->ops->da_to_va(rproc, da, len);
> >  		if (ptr)
> > @@ -677,6 +681,7 @@ static int rproc_handle_carveout(struct rproc
> *rproc,
> >  	struct rproc_mem_entry *carveout, *mapping;
> >  	struct device *dev = &rproc->dev;
> >  	dma_addr_t dma;
> > +	phys_addr_t pa;
> >  	void *va;
> >  	int ret;
> >
> > @@ -698,6 +703,41 @@ static int rproc_handle_carveout(struct rproc
> *rproc,
> >  	if (!carveout)
> >  		return -ENOMEM;
> >
> > +	/* Check carveout rsc already part of a registered carveout */
> > +	if (rsc->da != FW_RSC_ADDR_ANY) {
> 
> As mentioned before, I consider it perfectly viable for rsc->da to be
> ANY and the driver providing a fixed carveout.

Yes I'll change sequence to lookup by name first and then verify exact parameters matching , not only da definition.

> 
> > +		va = rproc_find_carveout_by_da(rproc, rsc->da, rsc->len);
> > +
> > +		if (va) {
> 
> In a system with an iommu it's possible that rsc->len is larger than
> some carveout->len and va is NULL here so we fall through, allocate some
> memory and remap a segment of the carveout. (Or hopefully fails
> attempting).
> 
> > +			/* Registered region found */
> > +			pa = rproc_va_to_pa(va);
> > +			if (rsc->pa != FW_RSC_ADDR_ANY && rsc->pa !=
> (u32)pa) {
> > +				/* Carveout doesn't match request */
> > +				dev_err(dev->parent,
> > +					"Failed to find carveout fitting da and
> pa\n");
> > +				return -ENOMEM;
> > +			}
> > +
> > +			/* Update rsc table with physical address */
> > +			rsc->pa = (u32)pa;
> > +
> > +			/* Update carveouts list */
> > +			carveout->va = va;
> > +			carveout->len = rsc->len;
> > +			carveout->da = rsc->da;
> > +			carveout->priv = (void *)CARVEOUT_RSC;
> > +
> > +			list_add_tail(&carveout->node, &rproc->carveouts);
> 
> rproc_find_carveout_by_da() will return a reference into a carveout, now
> we add another overlapping carveout into the same list.
> 
> 
> I think it would be saner to not allow the resource table to describe
> subsets of carveouts registered by the driver.
> 
> In which case this would better find a carveout by name or exact da,
> then check that the pa, da, len and rsc->flags are adequate.

Agree
/Loic
> 
> > +
> > +			return 0;
> > +		}
> > +
> > +		if (!rproc->domain) {
> 
> Currently this function ignore invalid values of da when !domain, so I
> think it would be good you can submit this sanity check in it's own
> patch so that anyone bisecting this would know why their broken firmware
> suddenly isn't loadable.
> 
> > +			dev_err(dev->parent,
> > +				"Bad carveout rsc configuration\n");
> > +			return -ENOMEM;
> > +		}
> > +	}
> > +
> 
> Regards,
> Bjorn