| Message ID | 20200605101222.10783-1-wsa+renesas@sang-engineering.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | mmc: renesas_sdhi: support manual calibration | expand |
On Fri, 5 Jun 2020 at 12:12, Wolfram Sang <wsa+renesas@sang-engineering.com> wrote: > > Some R-Car Gen3 SoCs need corrections after they tuned to HS400. This > series implements that. Please have a look at the patch descriptions for > details. The origins were patches in the BSP from Takeshi Saito. > > Changes since RFC v3: > > * moved detection in probe() a bit to the front before calling into the > TMIO driver ensuring all fields are properly initialized > * tested on Gen2 (R-Car H2), too; no regressions > > There is a branch for testing here (with the debug patch on top): > > git://git.kernel.org/pub/scm/linux/kernel/git/wsa/linux.git renesas/topic/sdhi-manual-calib > > Note: handling 'hs400_calib_table' in patch 2 uses some multidimensional > array vs pointers judo. I think it is correct C this way but I am > especially grateful for further review there. Wolfram, I tried applying this, but got some conflict. I guess you have based this on some of the other series that's in the pipe. Maybe you can give me some advice about the order or how to apply things? Kind regards Uffe > > > Current test results: > --------------------- > > An unaffected SDHI instance (no eMMC (R-Car H2) or no table (H3 ES2.0)): > > No calib table, so flag gets disabled during the reset in the beginning > of the initialization. The same reset routine will be called when tuning > fails. > > kworker/0:0-3 [000] .... 2.799451: renesas_sdhi_reset_hs400_mode: ee100000.sd: calib table 00000000 > kworker/0:0-3 [000] .... 2.799459: renesas_sdhi_hw_reset: ee100000.sd: reset! flag disabled > > An SDHI instance with a HS400 capable eMMC connected (R-Car M3-N): > > When preparing HS400 tuning, a calib table is found. So, (potential) old > HS400 tuning gets reset and the manual adjustment gets disabled, the > runtime flag (signalling new manual adjustment will be needed) is set: > > kworker/1:2-139 [001] .... 4.028232: renesas_sdhi_reset_hs400_mode: ee140000.sd: calib table ffff800010c76ee8 > kworker/1:2-139 [001] .... 4.028237: renesas_sdhi_reset_hs400_mode: ee140000.sd: adjustment disabled > kworker/1:2-139 [001] .... 4.028239: renesas_sdhi_prepare_hs400_tuning: ee140000.sd: flag enabled > > Lots of tuning CMDs: > > kworker/0:1-21 [000] .... 4.028370: renesas_sdhi_fixup_request: ee140000.sd: opcode 21, flag 1 > kworker/0:1-21 [000] .... 4.028576: renesas_sdhi_fixup_request: ee140000.sd: opcode 21, flag 1 > kworker/0:1-21 [000] .... 4.028668: renesas_sdhi_fixup_request: ee140000.sd: opcode 21, flag 1 > kworker/0:1-21 [000] .... 4.028754: renesas_sdhi_fixup_request: ee140000.sd: opcode 21, flag 1 > kworker/0:1-21 [000] .... 4.028832: renesas_sdhi_fixup_request: ee140000.sd: opcode 21, flag 1 > kworker/0:1-21 [000] .... 4.028909: renesas_sdhi_fixup_request: ee140000.sd: opcode 21, flag 1 > kworker/0:1-21 [000] .... 4.028989: renesas_sdhi_fixup_request: ee140000.sd: opcode 21, flag 1 > kworker/0:1-21 [000] .... 4.029065: renesas_sdhi_fixup_request: ee140000.sd: opcode 21, flag 1 > kworker/0:1-21 [000] .... 4.029149: renesas_sdhi_fixup_request: ee140000.sd: opcode 21, flag 1 > kworker/0:1-21 [000] .... 4.029228: renesas_sdhi_fixup_request: ee140000.sd: opcode 21, flag 1 > kworker/0:1-21 [000] .... 4.029302: renesas_sdhi_fixup_request: ee140000.sd: opcode 21, flag 1 > kworker/0:1-21 [000] .... 4.029378: renesas_sdhi_fixup_request: ee140000.sd: opcode 21, flag 1 > kworker/0:1-21 [000] .... 4.029453: renesas_sdhi_fixup_request: ee140000.sd: opcode 21, flag 1 > kworker/0:1-21 [000] .... 4.029532: renesas_sdhi_fixup_request: ee140000.sd: opcode 21, flag 1 > kworker/0:1-21 [000] .... 4.029612: renesas_sdhi_fixup_request: ee140000.sd: opcode 21, flag 1 > kworker/0:1-21 [000] .... 4.029694: renesas_sdhi_fixup_request: ee140000.sd: opcode 21, flag 1 > kworker/0:1-21 [000] .... 4.029775: renesas_sdhi_fixup_request: ee140000.sd: opcode 6, flag 1 > > First CMD13 after tuning discovered, enable adjustment, clear runtime > flag, show replacement value: > > kworker/0:1-21 [000] .... 4.030294: renesas_sdhi_fixup_request: ee140000.sd: opcode 13, flag 1 > kworker/0:1-21 [000] .... 4.030299: renesas_sdhi_fixup_request: ee140000.sd: adjustment enabled, flag disabled > kworker/0:1-21 [000] .... 4.030300: renesas_sdhi_fixup_request: code 0 replacement 1 > > And then checksumming a large file works without performance > regressions. Retunes happen once occasionally and follow the above > pattern exactly (with different replacement values, though). > > So, again a lot of text but I want interested parties to verify results. > > Looking forward to comments, although I won't complain if this is the > final version ;) > > Thanks and kind regards, > > Wolfram > > > Wolfram Sang (2): > mmc: tmio: add generic hook to fixup after a completed request > mmc: renesas_sdhi: support manual calibration > > drivers/mmc/host/renesas_sdhi.h | 5 + > drivers/mmc/host/renesas_sdhi_core.c | 152 ++++++++++++++++++++++++++- > drivers/mmc/host/tmio_mmc.h | 1 + > drivers/mmc/host/tmio_mmc_core.c | 3 + > 4 files changed, 159 insertions(+), 2 deletions(-) > > -- > 2.20.1 >
> I tried applying this, but got some conflict. I guess you have based > this on some of the other series that's in the pipe. Maybe you can > give me some advice about the order or how to apply things? Sorry, this depends on the stalled SCC patchset and I need to send a second version of that (hopefully next week). Should have mentioned it here, mea culpa! So, this needs to wait a little.
Some R-Car Gen3 SoCs need corrections after they tuned to HS400. This series implements that. Please have a look at the patch descriptions for details. The origins were patches in the BSP from Takeshi Saito. Changes since RFC v3: * moved detection in probe() a bit to the front before calling into the TMIO driver ensuring all fields are properly initialized * tested on Gen2 (R-Car H2), too; no regressions There is a branch for testing here (with the debug patch on top): git://git.kernel.org/pub/scm/linux/kernel/git/wsa/linux.git renesas/topic/sdhi-manual-calib Note: handling 'hs400_calib_table' in patch 2 uses some multidimensional array vs pointers judo. I think it is correct C this way but I am especially grateful for further review there. Current test results: --------------------- An unaffected SDHI instance (no eMMC (R-Car H2) or no table (H3 ES2.0)): No calib table, so flag gets disabled during the reset in the beginning of the initialization. The same reset routine will be called when tuning fails. kworker/0:0-3 [000] .... 2.799451: renesas_sdhi_reset_hs400_mode: ee100000.sd: calib table 00000000 kworker/0:0-3 [000] .... 2.799459: renesas_sdhi_hw_reset: ee100000.sd: reset! flag disabled An SDHI instance with a HS400 capable eMMC connected (R-Car M3-N): When preparing HS400 tuning, a calib table is found. So, (potential) old HS400 tuning gets reset and the manual adjustment gets disabled, the runtime flag (signalling new manual adjustment will be needed) is set: kworker/1:2-139 [001] .... 4.028232: renesas_sdhi_reset_hs400_mode: ee140000.sd: calib table ffff800010c76ee8 kworker/1:2-139 [001] .... 4.028237: renesas_sdhi_reset_hs400_mode: ee140000.sd: adjustment disabled kworker/1:2-139 [001] .... 4.028239: renesas_sdhi_prepare_hs400_tuning: ee140000.sd: flag enabled Lots of tuning CMDs: kworker/0:1-21 [000] .... 4.028370: renesas_sdhi_fixup_request: ee140000.sd: opcode 21, flag 1 kworker/0:1-21 [000] .... 4.028576: renesas_sdhi_fixup_request: ee140000.sd: opcode 21, flag 1 kworker/0:1-21 [000] .... 4.028668: renesas_sdhi_fixup_request: ee140000.sd: opcode 21, flag 1 kworker/0:1-21 [000] .... 4.028754: renesas_sdhi_fixup_request: ee140000.sd: opcode 21, flag 1 kworker/0:1-21 [000] .... 4.028832: renesas_sdhi_fixup_request: ee140000.sd: opcode 21, flag 1 kworker/0:1-21 [000] .... 4.028909: renesas_sdhi_fixup_request: ee140000.sd: opcode 21, flag 1 kworker/0:1-21 [000] .... 4.028989: renesas_sdhi_fixup_request: ee140000.sd: opcode 21, flag 1 kworker/0:1-21 [000] .... 4.029065: renesas_sdhi_fixup_request: ee140000.sd: opcode 21, flag 1 kworker/0:1-21 [000] .... 4.029149: renesas_sdhi_fixup_request: ee140000.sd: opcode 21, flag 1 kworker/0:1-21 [000] .... 4.029228: renesas_sdhi_fixup_request: ee140000.sd: opcode 21, flag 1 kworker/0:1-21 [000] .... 4.029302: renesas_sdhi_fixup_request: ee140000.sd: opcode 21, flag 1 kworker/0:1-21 [000] .... 4.029378: renesas_sdhi_fixup_request: ee140000.sd: opcode 21, flag 1 kworker/0:1-21 [000] .... 4.029453: renesas_sdhi_fixup_request: ee140000.sd: opcode 21, flag 1 kworker/0:1-21 [000] .... 4.029532: renesas_sdhi_fixup_request: ee140000.sd: opcode 21, flag 1 kworker/0:1-21 [000] .... 4.029612: renesas_sdhi_fixup_request: ee140000.sd: opcode 21, flag 1 kworker/0:1-21 [000] .... 4.029694: renesas_sdhi_fixup_request: ee140000.sd: opcode 21, flag 1 kworker/0:1-21 [000] .... 4.029775: renesas_sdhi_fixup_request: ee140000.sd: opcode 6, flag 1 First CMD13 after tuning discovered, enable adjustment, clear runtime flag, show replacement value: kworker/0:1-21 [000] .... 4.030294: renesas_sdhi_fixup_request: ee140000.sd: opcode 13, flag 1 kworker/0:1-21 [000] .... 4.030299: renesas_sdhi_fixup_request: ee140000.sd: adjustment enabled, flag disabled kworker/0:1-21 [000] .... 4.030300: renesas_sdhi_fixup_request: code 0 replacement 1 And then checksumming a large file works without performance regressions. Retunes happen once occasionally and follow the above pattern exactly (with different replacement values, though). So, again a lot of text but I want interested parties to verify results. Looking forward to comments, although I won't complain if this is the final version ;) Thanks and kind regards, Wolfram Wolfram Sang (2): mmc: tmio: add generic hook to fixup after a completed request mmc: renesas_sdhi: support manual calibration drivers/mmc/host/renesas_sdhi.h | 5 + drivers/mmc/host/renesas_sdhi_core.c | 152 ++++++++++++++++++++++++++- drivers/mmc/host/tmio_mmc.h | 1 + drivers/mmc/host/tmio_mmc_core.c | 3 + 4 files changed, 159 insertions(+), 2 deletions(-)