Re: rkisp in mainline (destaging) vs. rk3326/px30 uapi differences

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Hi,

Am 07.01.21 um 21:23 schrieb Heiko Stuebner:
Hi,

the rkisp driver in the mainline Linux kernel moved out of staging with
5.11-rc1, so the uapi will be fixed after 5.11 proper is released.

The rkisp driver currently only supports the rk3399 and while working
on porting the support for rk3326/px30 I noticed discrepancies.

Hence it would be somewhat urgent to clarify this, as later it will get
really cumbersome.

I see that we are now on 5.11-rc2 so that gives us about 4-5 weeks,


----

The rkisp on the px30 (v12) has some changes compared to the rk3399 (v10).

How do you know that the isp of rk3399 is v10 ? I looked at the RK3399 TRM
and the datasheet for the isp and could not find this information.

Some sub-blocks moved around or seem to have been replaced with newer
variants and the gist of changes can be seen in [0] with the important
part being the uapi changes [1] and those values also exist in mainline.


See functions in that patch:
- isp_goc_config_v12()
- rkisp1_stats_get_aec_meas_v12()
- rkisp1_stats_get_hst_meas_v12()

Looking at the code, the register locations are different, for gammas and
the histogram the actual amount of raw registers is the same, while the
"aec" seems to use 25 registers on V10 while 21 registers on V12. Though
their content gets split into multiple values in that v12 variant.


As somehow expected the whole thing is pretty undocumented and I
have no clue what these "bins" or "gammas" mean and why the amount of
entries now differs and how this relates to userspace at all.

Also looking through libcamera as the one open user of the driver,
the whole rkisp1_cif_isp_isp_other_cfg (containing the gamma config)
as well as the rkisp1_cif_isp_stat struct (for ae and histogram)
don't seem to be used so far.

yes, that's a shame. There is a simple implementation using the ae in
stuct rkisp1_cif_isp_stat in src/ipa/rkisp1.c


Hence I also added some Rockchip people in the hope of getting
a bit of clarification ;-) .


Ideas on how to proceed?

Thanks
Heiko


[0] https://github.com/rockchip-linux/kernel/commit/2ff670508e8fdfefd67318e885effb8cee4a0f4c
[1]
diff --git a/include/uapi/linux/rkisp1-config.h b/include/uapi/linux/rkisp1-config.h
index b471f01a8459..fbeb6b5dba03 100644
--- a/include/uapi/linux/rkisp1-config.h
+++ b/include/uapi/linux/rkisp1-config.h
@@ -32,8 +32,8 @@
  #define CIFISP_CTK_COEFF_MAX            0x100
  #define CIFISP_CTK_OFFSET_MAX           0x800
-#define CIFISP_AE_MEAN_MAX 25
-#define CIFISP_HIST_BIN_N_MAX           16
+#define CIFISP_AE_MEAN_MAX              81
+#define CIFISP_HIST_BIN_N_MAX           32
  #define CIFISP_AFM_MAX_WINDOWS          3
  #define CIFISP_DEGAMMA_CURVE_SIZE       17
@@ -69,7 +69,7 @@
   * Gamma out
   */
  /* Maximum number of color samples supported */
-#define CIFISP_GAMMA_OUT_MAX_SAMPLES       17
+#define CIFISP_GAMMA_OUT_MAX_SAMPLES       34

I see that in that code you use the old names of the registers.
The names are different in the current version of the driver,
in the media tree: git://linuxtv.org/media_tree.git
Also, I guess that instead of changing the values you should
add a separated define, something like:

-#define CIFISP_GAMMA_OUT_MAX_SAMPLES       17
+#define CIFISP_GAMMA_OUT_MAX_SAMPLES_V10       17
+#define CIFISP_GAMMA_OUT_MAX_SAMPLES_v12       34


Thanks for working on that, hope we could still fix this in 5.11,

I don't have a rk3326/px30 hardware so I can't test your patches.
Do you have a hardware to test it?
I suggest that you send a patchset to the mailing list then I can
review it and test it on rk3399. Unfortunately there is indeed no way
to thoroughly test the params/stats since there is no userspace for that.

Thanks,
Dafna







[Index of Archives]     [Linux Input]     [Video for Linux]     [Gstreamer Embedded]     [Mplayer Users]     [Linux USB Devel]     [Linux Audio Users]     [Linux Kernel]     [Linux SCSI]     [Yosemite Backpacking]

  Powered by Linux