thoughts about recent Samsung related patches
Hi,
while working on the esdhc-controller for the imx-platform, I noticed the recently comitted changes for Samsung controllers and have some remarks (sorry for being late):
1) Checking conditions for of host->ops->get_min_clock
The original commit e9510176ff728135383f0cdfc9c90cfe57f9e162 (sdhci: be more strict with get_min_clock() usage) states why the additional checks were added. Under this light, it could be argued that commit cfd1f82f20e0c557a061189f7d8c30d623fbe313 (sdhci: remove useless set_clock() check) could be reverted and this commit ce5f036bbbfc6c21d7b55b8fdaa2e2bd56392d94 (sdhci-s3c: add support for the non standard minimal clock value) could also be reverted if the samsung platform driver just uses SDHCI_QUIRK_NONSTANDARD_CLOCK without setting ops->set_clock?
2) 8-Bit data transfer support
The comitted version ae6d6c92212e94b12ab9365c23fb73acc2c3c2e7 (sdhci:
8-bit data transfer width support) looks different from another RFC posted in February:
http://www.mail-archive.com/[email protected]/msg01250.html
As those two already differ, I think it might be wiser to move 8-bit-mode-handling to the platform-specific code? Even the documented features of a SDHC differ across implementations, I fear side-effects when using this kind of undocumented feature (official spec says "reserved" when describing this bit).
3) NO_HI_SPD
Commit 5193250168ccdf87364e35a11965336dc088578c (sdhci: add no hi-speed bit quirk support) adds a quirk which can be avoided by using io-accessors like in sdhci-of-esdhc.c. Maybe we can even get rid of more, older quirks this way to save precious quirk flags. Have to check that later.
I hope my comments are applicable; because there is no freely available datasheet, I can't verify all of my assumptions. Looking forward to comments.
Kind regards,
Wolfram
Signature · 2 lines
Pengutronix e.K. | Wolfram Sang | Industrial Linux Solutions | http://www.pengutronix.de/ |
Hi
Quoted 8 lines · Wolfram Sang
Hi,
while working on the esdhc-controller for the imx-platform, I noticed the recently comitted changes for Samsung controllers and have some remarks (sorry for being late):
1) Checking conditions for of
host->ops->get_min_clock
Quoted 9 lines
The original commit
e9510176ff728135383f0cdfc9c90cfe57f9e162(sdhci: be more strict withget_min_clock()usage) states why the additional checks were added. Under this light, it could be argued that commitcfd1f82f20e0c557a061189f7d8c30d623fbe313(sdhci: remove uselessset_clock()check) could be reverted and this commitce5f036bbbfc6c21d7b55b8fdaa2e2bd56392d94(sdhci-s3c: add support for the non standard minimal clock value) could also be reverted if the samsung platform driver just usesSDHCI_QUIRK_NONSTANDARD_CLOCKwithout settingops->set_clock?
In our case we only want to use the min_clock at probe time without NONSTANDARD_CLOCK. and no need to NONSTANDARD_CLOCK at set_clock since we need other parts also at sdhci.c
Quoted 8 lines
2) 8-Bit data transfer support
The comitted version
ae6d6c92212e94b12ab9365c23fb73acc2c3c2e7(sdhci:
8-bit data transfer width support) looks different from another RFC posted in February:
http://www.mail-archive.com/[email protected]/msg01250.html
Quoted 5 lines
As those two already differ, I think it might be wiser to move 8-bit-mode-handling to the platform-specific code? Even the documented features of a SDHC differ across implementations, I fear side-effects when using this kind of undocumented feature (official spec says "reserved" when describing this bit).
Okay it looks different from Samsung Spec.
WIDE8[5]: Extended Data Transfer Width (It is for MMC 8-bit card). 1: 8-bit operatoin
0: Bit width is designated by the bit 1(Data Transfer Width) So no need to set the BIT 1 and BIT5 simultaneously.
I also wonder other Specs how described it.
Quoted 9 lines
3)
NO_HI_SPDCommit
5193250168ccdf87364e35a11965336dc088578c(sdhci: add no hi-speed bit quirk support) adds a quirk which can be avoided by using io-accessors like insdhci-of-esdhc.c. Maybe we can even get rid of more, older quirks this way to save precious quirk flags. Have to check that later.
Quoted 3 lines
I hope my comments are applicable; because there is no freely available datasheet, I can't verify all of my assumptions. Looking forward to comments.
Good, I think it's possible. I'll try and send a patch.
Quoted 18 lines
Kind regards,
Wolfram
--
Pengutronix e.K. | Wolfram Sang | Industrial Linux Solutions | http://www.pengutronix.de/ |-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.10 (GNU/Linux)iEYEARECAAYFAkyEyesACgkQD27XaX1/VRsm/QCePflpNSE0iQQWSUjf/PY723qK ZbEAoLxAIS7jIJY/lUQpQvbkJMN6ve19
=98xD
-----END PGP SIGNATURE-----
Quoted 14 lines
Quoted 8 lines
2) 8-Bit data transfer support
The comitted version
ae6d6c92212e94b12ab9365c23fb73acc2c3c2e7(sdhci:
8-bit data transfer width support) looks different from another RFC posted in February:
http://www.mail-archive.com/[email protected]/msg01250.htmlQuoted 5 lines
As those two already differ, I think it might be wiser to move 8-bit-mode-handling to the platform-specific code? Even the documented features of a SDHC differ across implementations, I fear side-effects when using this kind of undocumented feature (official spec says "reserved" when describing this bit).
Quoted 6 lines
Okay it looks different from Samsung Spec.
WIDE8[5]: Extended Data Transfer Width (It is for MMC 8-bit card). 1: 8-bit operatoin
0: Bit width is designated by the bit 1(Data Transfer Width) So no need to set the BIT 1 and BIT5 simultaneously.
I also wonder other Specs how described it.
esdhc occupied BIT 2 (originally HI_SPD) for this :( %10 means 8 bit transfer. %11 is reserved.
Quoted 1 line
Good, I think it's possible. I'll try and send a patch.
Thanks!
Signature · 2 lines
Pengutronix e.K. | Wolfram Sang | Industrial Linux Solutions | http://www.pengutronix.de/ |
Quoted 4 lines
Quoted 3 lines
I hope my comments are applicable; because there is no freely available datasheet, I can't verify all of my assumptions. Looking forward to comments.
Good, I think it's possible. I'll try and send a patch.
Any news on this ?
Regards,
Wolfram
Signature · 2 lines
Pengutronix e.K. | Wolfram Sang | Industrial Linux Solutions | http://www.pengutronix.de/ |
Quoted 6 lines · Wolfram Sang
Quoted 4 lines
Quoted 3 lines
I hope my comments are applicable; because there is no freely available datasheet, I can't verify all of my assumptions. Looking forward to comments.
Good, I think it's possible. I'll try and send a patch.
Any news on this ?
I tried and no problem to work, but I'm thinking does it really required for reducing one quirk?
With this change sdhci-s3c should use the CONFIG_MMC_SDHCI_IO_ACCESSORS and it introduces the io functions for every io access.
So I'm not sure these change is required for s3c sdhci.
Give you opinions.
Thank you,
Kyungmin Park
Quoted 18 lines
Regards,
Wolfram
--
Pengutronix e.K. | Wolfram Sang | Industrial Linux Solutions | http://www.pengutronix.de/ |-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.10 (GNU/Linux)iEYEARECAAYFAkyOGYwACgkQD27XaX1/VRt27wCgkDJcNU61qyoor1Wzkrce5R/A 44YAnRvNMzJ1WbB5IbCgmV+kLuP1+h2C
=mOsI
-----END PGP SIGNATURE-----
cleaning up sdhci? (was Re: [RFC] thoughts about recent Samsung related patches)
Quoted 14 lines · Kyungmin Park
Quoted 6 lines · Wolfram Sang
Quoted 4 lines
Quoted 3 lines
I hope my comments are applicable; because there is no freely available datasheet, I can't verify all of my assumptions. Looking forward to comments.
Good, I think it's possible. I'll try and send a patch.
Any news on this ?
I tried and no problem to work, but I'm thinking does it really required for reducing one quirk?
With this change sdhci-s3c should use theCONFIG_MMC_SDHCI_IO_ACCESSORSand it introduces the io functions for every io access.
Quoted 3 lines
So I'm not sure these change is required for s3c sdhci.
Give you opinions.
I know what you mean. Still, I'd prefer to see that sdhci.c implements stuff as generic as possible and trying to follow the standard closely. Exceptions to that should be handled elsewhere if possible. That is, if some implementations of a controller do not adhere to that standard and change some bits in the register layout, well, then they have to live with the quirks and their performance issues IMHO.
BTW it is not only the HI_SPD issue, which is a candidate for the io- accessors. I am also interested to get the 8BITBUS patch reverted. It breaks some SD cards on some boards using the esdhc controller :( Support for a 8bit-bus needs a bit more thought, it seems...
Seeing Alan's patchset, it seems to be a good time to do some sdhci-cleanup. My impression is that we could get generalize a few quirks by using io-accessors and overriding capabilties-flags. I am open for other opinions, though. All in all, it would be great if we all could combine our efforts and get some kind of master-plan :) Will take a look at Alan's patches now...
Regards,
Wolfram
Signature · 2 lines
Pengutronix e.K. | Wolfram Sang | Industrial Linux Solutions | http://www.pengutronix.de/ |
Re: cleaning up sdhci? (was Re: [RFC] thoughts about recent Samsung related patches)
Quoted 3 lines
That is, if some implementations of a controller do not adhere to that standard and change some bits in the register layout, well, then they have to live with the quirks and their performance issues IMHO.
The ATA layer uses I/O accessors although it abstracts them a bit (so its got 'read status' type abstractions. For performance reasons one or two places the abstraction is higher level (eg block data transfers) but with those in place it doesn't seem to be measurable.
Quoted 3 lines
Seeing Alan's patchset, it seems to be a good time to do some sdhci-cleanup. My impression is that we could get generalize a few quirks by using io-accessors and overriding capabilties-flags. I am
I avoided doing any of it by making I/O access functions fake results because that seemed to be a recipe for hard to debug future problems. When dealing with different register layouts or specific bugs it seems sensible
(and 8250 serial does this succesfully with some devices faking certain register bits their hardware gets wrong)
Quoted 3 lines
open for other opinions, though. All in all, it would be great if we all could combine our efforts and get some kind of master-plan :) Will take a look at Alan's patches now...
If there is going to be a grand clean up my preference long term would be that there is no
if (host->ops->foo)
host->ops->foo(...)
else {
do foo
}but that host->ops gets all the NULL entries filled in and the core code simply calls host->ops->xxx in each case.
Alan