-----Original Message----- From: Wolfgang Denk [mailto:wd@denx.de] Sent: Wednesday, February 18, 2009 10:00 AM To: mjmarx Cc: u-boot@lists.denx.de Subject: Re: [U-Boot] [PATCH] ADS5121 Add mem config for current rev4 boards
Dear Martha,
in message 22077983.post@talk.nabble.com you wrote:
- im->mddrc.ddr_command = CONFIG_SYS_MICRON_PCHG_ALL;
- im->mddrc.ddr_command = CONFIG_SYS_MICRON_NOP;
- im->mddrc.ddr_command = CONFIG_SYS_MICRON_RFSH;
- im->mddrc.ddr_command = CONFIG_SYS_MICRON_NOP;
- im->mddrc.ddr_command = CONFIG_SYS_MICRON_RFSH;
- im->mddrc.ddr_command = CONFIG_SYS_MICRON_NOP;
im->mddrc.ddr_command = CONFIG_SYS_DDR_NOP;- im->mddrc.ddr_command = CONFIG_SYS_DDR_PCHG_ALL;
- im->mddrc.ddr_command = CONFIG_SYS_DDR_NOP;
- im->mddrc.ddr_command = CONFIG_SYS_DDR_RFSH;
- im->mddrc.ddr_command = CONFIG_SYS_DDR_NOP;
- im->mddrc.ddr_command = CONFIG_SYS_DDR_RFSH;
- im->mddrc.ddr_command = CONFIG_SYS_DDR_NOP;
While performing such a global rename, please let's get rid of the CONFIG_SYS_ names. These are NOT options that can be changed by the user in the board config file, so they should not look like such.
CONFIG_SYS_ was in there already for all the memory #defs (as of the global rename in October)
Yes, you are right. But as I mentioned - while performing such a global rename, we should use the opportunity and do it right this time.
As far as my name change -- I DID NOT DO A WHOLESALE RENAME OF THE MEMORY #defs
The init settings that Elpida and Micron share went from _MICRON_ to _DDR_ The setting they don't share stayed as _MICRON_ and now has a counterpart _ELPIDA_ This name change should not be combined with any other name change and belongs in the memory change patch. In fact it will be especially required if I keep one binary.
Code section with elipses
+#if defined(CONFIG_ADS5121_REV2) || \ + defined(CONFIG_ADS5121_REV3) || \ + defined(CONFIG_ADS5121_REV4M) + /* Micron init sequence */ im->mddrc.ddr_command = CONFIG_SYS_MICRON_INIT_DEV_OP; ... + im->mddrc.ddr_command = CONFIG_SYS_DDR_NOP; + im->mddrc.ddr_command = CONFIG_SYS_DDR_EM2; + im->mddrc.ddr_command = CONFIG_SYS_DDR_NOP; ... +#else + /* Elpida init sequence */ + im->mddrc.ddr_command = CONFIG_SYS_DDR_EM2; ... + im->mddrc.ddr_command = CONFIG_SYS_ELPIDA_INIT_DEV_OP; + udelay(200); +#endif
IF you look -- the only rename I did was to change _MICRON_ to _DDR_ and it is very appropriate for it to be in this patch.
But if we change the names anyway, we could correct the other mistake as well, right?
+#elif defined(CONFIG_ADS5121_REV3) || \
defined(CONFIG_ADS5121_REV4M)#define CONFIG_SYS_MDDRC_SYS_CFG 0xFA804A00 #define CONFIG_SYS_MDDRC_SYS_CFG_RUN 0xEA804A00 #define CONFIG_SYS_MDDRC_TIME_CFG1 0x68EC1168 #define CONFIG_SYS_MDDRC_TIME_CFG2 0x34310864 +#else +#define CONFIG_SYS_MDDRC_SYS_CFG 0xFA802b40 +#define CONFIG_SYS_MDDRC_SYS_CFG_RUN 0xEA802b40 +#define CONFIG_SYS_MDDRC_TIME_CFG1 0x690e1189 +#define CONFIG_SYS_MDDRC_TIME_CFG2 0x35410864 #endif
If I understand correctly, these are the only real changes to the memory init sequence right? This is virtually invisible in the huge number of variable renames. You should split this into two separate patches, one doing the renames only, and one adding the real new configuration.
See comment above.
In any case: please split this into two patches: one doing the rename only, and another one adding the new functionality.
The memory change was made due to a supply problem .. all new boards will have the new memory but unfortunately there is no new rev (or CPLD reg setting) for these boards (it is not an unwillingness on my part)
I understand this. On the other hand, there might be a new supply problem in N days/weeks/months from now, and we definitely don't want to have an ever growing number of incompatible configurations and binary images if it can be avoided (and I'm sure it can be avoided).
I will attempt to create one binary using Micron settings for rev 4.0 and below, Elpida settings for rev 4.1 and beyond and for the small number of 4.1 Micron boards out there I will attempt to have them run with Elpida settings (the memory will run slightly more slowly) I would like to keep an #ifdef in there to force Micron settings, however.
Let's try and find a solution that works with only one binary image, without additional target names and #ifdef's.
I will very much try my darn'dest!!!!!
-Martha
Best regards,
Wolfgang Denk
-- DENX Software Engineering GmbH, MD: Wolfgang Denk & Detlev Zundel HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd@denx.de "Engineering without management is art." - Jeff Johnson