[Intel-gfx] [PATCH v4 1/8] drm/i915/skl: Add support to load SKL CSR firmware.

[Imre Deak]

On Thu, 2015-04-16 at 19:53 +0530, Animesh Manna wrote:
> 
> On 4/16/2015 4:55 PM, Imre Deak wrote:
> > On to, 2015-04-16 at 17:29 +0530, Animesh Manna wrote:
> >> + Jesse, Rodrigo
> >>
> >> On 04/16/2015 02:51 PM, Imre Deak wrote:
> >>> On to, 2015-04-16 at 14:22 +0530, Animesh Manna wrote:
> >>>> [...]
> >>>> +#include <linux/firmware.h>
> >>>> +#include "i915_drv.h"
> >>>> +#include "i915_reg.h"
> >>>> +
> >>>> +#define I915_CSR_SKL "i915/skl_dmc_ver1.bin"
> >>> The latest version on the FW download page is skl_dmc_ver4.bin as Damien
> >>> pointed out and skl_dmc_ver1.bin is not available there. So why did you
> >>> choose ver1? If ver4 is not compatible with the current driver code,
> >>> then ver1 must be made available too on the download page, otherwise we
> >>> should use here ver4.
> >>>
> >>> The patch otherwise looks ok to me.
> >>>
> >>> --Imre
> >> Yes, 01.org will be updated with skl_dmc_ver1.bin. Thinking of  future
> >> scaling it is decided that we will follow the naming convention
> >> <platform>_dmc_<api-version>.bin and this will be a symbolic link of
> >> the actual firmware.
> >>
> >> Will be requesting firmware team to release firmware in
> >> <platform>_dmc_<api-version>_<minor-version>.bin Here in
> >> skl_dmc_ver4.bin api version is 1.0 and minor version is 4, so changed
> >> the file name to skl_dmc_ver1.bin.
> > I just repeat here Damien, but I don't think there is a need to
> > complicate things with symlinks. IMO I915_CSR_SKL should point to the
> > exact version of the firmware, not just to a symlink with the API
> > version. The driver will be tested with that particular firmware version
> > and so the driver writers should be in full control of what version will
> > be loaded with a given driver version.
>
> Agree, if we want to support only latest firmware for a particular
> platform we do not need symlink but during discussion found that if
> user want to rollback the newest firmware for same API version which
> can be achievable by using symlink and introducing api-version and
> minor-version. Anyways I am ok with both the approaches.

The only reason I can see for rolling back would be a regression. Fixing
that would need to go through the usual bug reporting flow and eventual
firmware and driver update. A symlink could anyway be created - as a
kind of hack - to force loading a different firmware version, but making
this the regular way would only allow for confusion about which firmware
version is actually loaded.

--Imre