This patch fixes the DDR3 initialization procedure in
order to comply with DDR3 standard. A 500 us delay is specified
between the DDR3 reset and clock enable signal. Until now,
this delay was not respected. Some DDR3 chips don't bother
but the bigger the RAM becomes the more likely it seems that
this delay is needed. We observed that DRAM > 256 MB from
the manufacturer Samsung have an issue when the specification
is not respected.
Changes:
1) Add a 1 ms wait for L3 timeout error trigger
2) Don't delay DDR3 initialization
Bit 31 of emif_sdram_ref_ctrl shouldn't be set because his
suppresses the initialization of DDR3
Signed-off-by: Samuel Egli <samuel.egli@siemens.com>
Reviewed-by: James Doublesin <doublesin@ti.com>
Cc: Tom Rini <trini@konsulko.com>
Cc: Felipe Balbi <balbi@ti.com>
Cc: Roger Meier <r.meier@siemens.com>
Cc: Heiko Schocher <hs@denx.de>
The recent changes for hw leveling on am33xx were not intended for
DDR2 boards, only DDR3. Update emif_sdram_type to take a sdram_config
value to check against. This lets us pass in the value we would use to
configure, when we have not yet configured the board yet. In other cases
update the call to be as functional as before and check an already
programmed value in.
Tested-by: Yan Liu <yan-liu@ti.com>
Signed-off-by: Tom Rini <trini@ti.com>
Signed-off-by: Lokesh Vutla <lokeshvutla@ti.com>
Back in fc46bae a "clean up" was introduced that intended to reconcile
some of the AM335x codepaths based on how AM43xx operates.
Unfortunately this introduced a regression on the DDR2 platforms. This
was un-noticed on DDR3 (everything except for Beaglebone White) as we
had already populated sdram_config correctly in sequence. This change
brings us back to the older behavior and is fine on all platforms.
Tested on Beaglebone White, Beaglebone Black and AM335x GP EVM
Reported-by: Matt Ranostay <mranostay@gmail.com>
Signed-off-by: Tom Rini <trini@konsulko.com>
Switch to using hardware leveling for certain parameters on the EMIF
rather than using precalculated values. Doing this also means we have a
common place now between am437x and am335x for setting
emif_sdram_ref_ctrl with a value for the correct delay length.
Tested-by: Felipe Balbi <balbi@ti.com>
Tested-by: Tom Rini <trini@ti.com>
Signed-off-by: James Doublesin <doublesin@ti.com>
Signed-off-by: Felipe Balbi <balbi@ti.com>
The register secure_emif_sdram_config in control module is copied to
the EMIF sdram_config register when it is coming out of DeepSleep0 in
order to ensure that the EMIF comes up for the correct type of DDR.
Without this, resume can hang from within the kernel.
Signed-off-by: Dave Gerlach <d-gerlach@ti.com>
Per a suggestion from the hardware team, program the emif_pwr_mgmt_ctrl
and emif_pwr_mgmt_ctrl_shdw registers within the EMIF to hold the
desired delay in cycles that the EMIF waits without an access to enter
self-refresh, in this case 8192 cycles. With this, code desiring to
enter self refresh only has to toggle one bit to enable it.
Signed-off-by: Dave Gerlach <d-gerlach@ti.com>
GP EVM has 1GB DDR3 attached(Part no: MT41K512M8RH).
Adding details for the same.
Below is the brief description of DDR3 init sequence(SW leveling):
-> Enable VTT regulator
-> Configure VTP
-> Configure DDR IO settings
-> Disable initialization and refreshes until EMIF registers are programmed.
-> Program Timing registers
-> Program leveling registers
-> Program PHY control and Temp alert and ZQ config registers.
-> Enable initialization and refreshes and configure SDRAM CONFIG register
Signed-off-by: Lokesh Vutla <lokeshvutla@ti.com>
AM4372 EPOS EVM has 1GB LPDDR2(Part no: MT42L256M32D2LG-25 WT:A)
Adding LPDDR2 init sequence and register details for the same.
Below is the brief description of LPDDR2 init sequence:
-> Configure VTP
-> Configure DDR IO settings
-> Disable initialization and refreshes until EMIF registers are programmed.
-> Program Timing registers
-> Program PHY control and Temp alert and ZQ config registers.
-> Enable initialization and refreshes and configure SDRAM CONFIG register
-> Wait till initialization is complete and the configure MR registers.
Signed-off-by: Lokesh Vutla <lokeshvutla@ti.com>
Currently same value is programmed for all ioregs. This is not
the case for all SoC's like AM4372. So adding a structure for ioregs
and updating in all board files. And also return from config_cmd_ctrl()
and config_ddr_data() functions if data is not passed.
Signed-off-by: Lokesh Vutla <lokeshvutla@ti.com>
[trini: Fixup dxr2, cm_t335, adapt pcm051 rev3]
Signed-off-by: Tom Rini <trini@ti.com>
Based on the definitive guide to EMIF configuration[1] certain registers
that we have been modifying (and are documented registers) should be
left in their reset values rather than modified. This has been tested
on AM335x GP EVM and Beaglebone White.
[1]: http://processors.wiki.ti.com/index.php/AM335x_EMIF_Configuration_tips
Cc: Enric Balletbo i Serra <eballetbo@iseebcn.com>
Cc: Javier Martinez Canillas <javier@dowhile0.org>
Cc: Heiko Schocher <hs@denx.de>
Cc: Lars Poeschel <poeschel@lemonage.de>
Signed-off-by: Tom Rini <trini@ti.com>
Tested-by: Matt Porter <matt.porter@linaro.org>
The AM33xx emif4/ddr support closely matches what is need to support
TI814x except that TI814x has two EMIF instances. Refactor all the
emif4 helper calls and the config_ddr() init function to use an
additional instance number argument.
Signed-off-by: Matt Porter <mporter@ti.com>
Reviewed-by: Tom Rini <trini@ti.com>
The original write to sdram_config is correct for DDR3 but incorrect
for DDR2 so SPL was hanging. For DDR2, the write to sdram_config
should be after the writes to ref_ctrl. This was working for DDR3
because there was a write of 0x2800 to ref_ctrl before a write
to sdram_config.
Tested on: GP EVM 1.1A (DDR2), GP EVM 1.5A (DDR3),
Beaglebone A6 (DDR2), Beagleone Blacd A4A (DDR3)
Signed-off-by: Steve Kipisz <s-kipisz2@ti.com>
Based on
http://processors.wiki.ti.com/index.php/AM335x_EMIF_Configuration_tips
we need to re-work our sequence in config_sdram slightly to match what
the TRM describes as the correct sequence. In our current (incorrect)
sequence some edge cases may fail to initalize correctly.
Signed-off-by: Tom Rini <trini@ti.com>
This patch updates secure_emif_sdram_config with the
same value written to sdram_config during ddr3 initialization.
During suspend/resume, this value is copied into sdram_config.
With this, a write to sdram_config at the end of resume sequence
which triggers an init sequence can be avoided.
Without this register write in place, the DDR_RESET line goes
low for a few cycles during resume which is a violation of the
JEDEC spec.
Signed-off-by: Satyanarayana, Sandhya <sandhya.satyanarayana@ti.com>
The ddr_regs struct was incorrectly offset after the dt0wiratio0 entry.
Correct this by documenting a missing register that will be used at some
point in the future (when write leveling is supported). Further, the
cmdNcs{force,delay} fields are undocumented and we have been setting
them to zero, remove. Next, setting of the
'DATAn_REG_PHY_USE_RANK0_DELAYS field belongs with the rest of the
ddr_data entries, so program it there. Finally, comment on how we are
configuring the DATA1 registers that correspond to the DATA0 (dt0)
registers defined in the struct.
Signed-off-by: Tom Rini <trini@ti.com>
The various ratio1 fields are not documented in any of the documentation
I can find. Removing these and testing has yielded success, so remove
the code that sets them and move their locations into the reserved
fields.
Signed-off-by: Tom Rini <trini@ti.com>
This function sets a number of related registers to the same value (the
registers in question all have the same field descriptions and are
related in operation). Rather than defining a struct and setting the
value repeatedly, just pass in the value.
Signed-off-by: Tom Rini <trini@ti.com>
Rather than defining our own structs to note what to use when
programming the EMIF and related re-use the emif_regs struct.
Signed-off-by: Tom Rini <trini@ti.com>
A number of memory initalization functions were int and always returned
0. Further it's not feasible to be doing error checking here, so simply
turn them into void functions.
Signed-off-by: Tom Rini <trini@ti.com>
Rework the EMIF4/DDR code slightly to setup the structs that
config_cmd_ctrl and config_ddr_data take to be setup at compile time and
mark them as const. This lets us simplify the calling path slightly as
well as making it easier to deal with DDR3.
Signed-off-by: Tom Rini <trini@ti.com>
When we change SDRAM_CONFIG this triggers a refresh based on all of the
parameters that we have programmed so we must do this last.
Signed-off-by: Tom Rini <trini@ti.com>
This patch adds AM33xx emif/ddr support along with board specific
defines.
Signed-off-by: Chandan Nath <chandan.nath@ti.com>
Signed-off-by: Sandeep Paulraj <s-paulraj@ti.com>