On Tue, Oct 11, 2016 at 11:01:43AM -0600, Jason Gunthorpe wrote: > On Tue, Oct 11, 2016 at 12:23:04PM +0300, Jarkko Sakkinen wrote: > > In order to provide access to locality registers, this commits adds > > mapping of the head of the CRB registers, which are located right > > before the control area. > > I think you should squash this into the prior patch, no sense in > changing all these lines twice > > But looks better to me.
So... do you suggest to squash with request/reliquish locality stuff? Just checking because this is 1/3 :) > > - priv->cca = crb_map_res(dev, priv, &io_res, buf->control_address, > > - sizeof(struct crb_control_area)); > > - if (IS_ERR(priv->cca)) > > - return PTR_ERR(priv->cca); > > + if (priv->flags & CRB_FL_CRB_START) { > > + priv->regs_h = crb_map_res(dev, priv, &io_res, io_res.start, > > + sizeof(struct crb_regs_head)); > > + if (IS_ERR(priv->regs_h)) > > + return PTR_ERR(priv->regs_h); > > + } > > + > > + priv->regs_t = crb_map_res(dev, priv, &io_res, buf->control_address, > > + sizeof(struct crb_regs_tail)); > > + if (IS_ERR(priv->regs_t)) > > + return PTR_ERR(priv->regs_t); > > So.. The ACPI IO region starts at the head area and continues to > include the control area, as one nice sane region - except for some > older stuff that puts the control area outside the ACPI IO region? Yes. The old hardware triggered SMM to do a DMA transfer (those that so called ACPI start). As a workaround for some of the hardware the driver always sets the CRB start flag also in the control area. That's why I also propose that we replace them with a single flag. > That makes a lot more sense.. > > Maybe chuck in a > > if (priv->flags & CRB_FL_CRB_START) { > if (buf->control_address != io_res.start + sizeof(struct > crb_regs_head)) > dev_warn(dev, FIRMWARE_BUG "Bad ACPI memory layout") > > Jason As a sanity check this would probably make sense. /Jarkko