| Message ID | 20190402165743.28106-2-d-gerlach@ti.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | ARM: OMAP2+: AM43x: Support DDR HW leveling suspend/resume | expand |
* Dave Gerlach <d-gerlach@ti.com> [190402 16:57]: > In certain situations, such as when returning from low power modes, the > EMIF must re-run hardware leveling to properly restore DDR3 access. > > This is accomplished by introducing a new ti-emif-sram-pm call, > ti_emif_run_hw_leveling, to check if DDR3 is in use and if so, trigger > the full write and read leveling processes. OK nice that you also consider devices with LPDDR2 then in case we start seeing those. Regards, Tony
Tony, On 4/2/19 12:24 PM, Tony Lindgren wrote: > * Dave Gerlach <d-gerlach@ti.com> [190402 16:57]: >> In certain situations, such as when returning from low power modes, the >> EMIF must re-run hardware leveling to properly restore DDR3 access. >> >> This is accomplished by introducing a new ti-emif-sram-pm call, >> ti_emif_run_hw_leveling, to check if DDR3 is in use and if so, trigger >> the full write and read leveling processes. > > OK nice that you also consider devices with LPDDR2 then in case we > start seeing those. Yes, LPDDR2 will work fine with these. Any thoughts on if this series should come through you or Santosh? Regards, Dave > > Regards, > > Tony >
* Dave Gerlach <d-gerlach@ti.com> [190408 18:31]: > Tony, > On 4/2/19 12:24 PM, Tony Lindgren wrote: > > * Dave Gerlach <d-gerlach@ti.com> [190402 16:57]: > > > In certain situations, such as when returning from low power modes, the > > > EMIF must re-run hardware leveling to properly restore DDR3 access. > > > > > > This is accomplished by introducing a new ti-emif-sram-pm call, > > > ti_emif_run_hw_leveling, to check if DDR3 is in use and if so, trigger > > > the full write and read leveling processes. > > > > OK nice that you also consider devices with LPDDR2 then in case we > > start seeing those. > > Yes, LPDDR2 will work fine with these. Any thoughts on if this series should > come through you or Santosh? Probably best that I queue these since there's also arch/arm/mach-omap2 code related changes. Regards, Tony
* Tony Lindgren <tony@atomide.com> [190408 19:34]: > * Dave Gerlach <d-gerlach@ti.com> [190408 18:31]: > > Tony, > > On 4/2/19 12:24 PM, Tony Lindgren wrote: > > > * Dave Gerlach <d-gerlach@ti.com> [190402 16:57]: > > > > In certain situations, such as when returning from low power modes, the > > > > EMIF must re-run hardware leveling to properly restore DDR3 access. > > > > > > > > This is accomplished by introducing a new ti-emif-sram-pm call, > > > > ti_emif_run_hw_leveling, to check if DDR3 is in use and if so, trigger > > > > the full write and read leveling processes. > > > > > > OK nice that you also consider devices with LPDDR2 then in case we > > > start seeing those. > > > > Yes, LPDDR2 will work fine with these. Any thoughts on if this series should > > come through you or Santosh? > > Probably best that I queue these since there's also > arch/arm/mach-omap2 code related changes. Santosh, is this OK with you? Care to ack the patch? Regards, Tony
On 4/9/19 8:19 AM, Tony Lindgren wrote: > * Tony Lindgren <tony@atomide.com> [190408 19:34]: >> * Dave Gerlach <d-gerlach@ti.com> [190408 18:31]: >>> Tony, >>> On 4/2/19 12:24 PM, Tony Lindgren wrote: >>>> * Dave Gerlach <d-gerlach@ti.com> [190402 16:57]: >>>>> In certain situations, such as when returning from low power modes, the >>>>> EMIF must re-run hardware leveling to properly restore DDR3 access. >>>>> >>>>> This is accomplished by introducing a new ti-emif-sram-pm call, >>>>> ti_emif_run_hw_leveling, to check if DDR3 is in use and if so, trigger >>>>> the full write and read leveling processes. >>>> >>>> OK nice that you also consider devices with LPDDR2 then in case we >>>> start seeing those. >>> >>> Yes, LPDDR2 will work fine with these. Any thoughts on if this series should >>> come through you or Santosh? >> >> Probably best that I queue these since there's also >> arch/arm/mach-omap2 code related changes. > > Santosh, is this OK with you? Care to ack the patch? > Sure Tony !! Acked-by: Santosh Shilimkar <ssantosh@kernel.org>
* santosh.shilimkar@oracle.com <santosh.shilimkar@oracle.com> [190409 15:26]: > On 4/9/19 8:19 AM, Tony Lindgren wrote: > > * Tony Lindgren <tony@atomide.com> [190408 19:34]: > > > * Dave Gerlach <d-gerlach@ti.com> [190408 18:31]: > > > > Tony, > > > > On 4/2/19 12:24 PM, Tony Lindgren wrote: > > > > > * Dave Gerlach <d-gerlach@ti.com> [190402 16:57]: > > > > > > In certain situations, such as when returning from low power modes, the > > > > > > EMIF must re-run hardware leveling to properly restore DDR3 access. > > > > > > > > > > > > This is accomplished by introducing a new ti-emif-sram-pm call, > > > > > > ti_emif_run_hw_leveling, to check if DDR3 is in use and if so, trigger > > > > > > the full write and read leveling processes. > > > > > > > > > > OK nice that you also consider devices with LPDDR2 then in case we > > > > > start seeing those. > > > > > > > > Yes, LPDDR2 will work fine with these. Any thoughts on if this series should > > > > come through you or Santosh? > > > > > > Probably best that I queue these since there's also > > > arch/arm/mach-omap2 code related changes. > > > > Santosh, is this OK with you? Care to ack the patch? > > > Sure Tony !! > > Acked-by: Santosh Shilimkar <ssantosh@kernel.org> Thanks, applying these two patches into omap-for-v5.2/am4-ddr3 branch. Reagrds, Tony
diff --git a/drivers/memory/emif.h b/drivers/memory/emif.h index 9e9f8037955d..6b71fadb3cfa 100644 --- a/drivers/memory/emif.h +++ b/drivers/memory/emif.h @@ -537,6 +537,9 @@ #define MCONNID_SHIFT 0 #define MCONNID_MASK (0xff << 0) +/* READ_WRITE_LEVELING_CONTROL */ +#define RDWRLVLFULL_START 0x80000000 + /* DDR_PHY_CTRL_1 - EMIF4D */ #define DLL_SLAVE_DLY_CTRL_SHIFT_4D 4 #define DLL_SLAVE_DLY_CTRL_MASK_4D (0xFF << 4) @@ -598,6 +601,7 @@ extern struct emif_regs_amx3 ti_emif_regs_amx3; void ti_emif_save_context(void); void ti_emif_restore_context(void); +void ti_emif_run_hw_leveling(void); void ti_emif_enter_sr(void); void ti_emif_exit_sr(void); void ti_emif_abort_sr(void); diff --git a/drivers/memory/ti-emif-pm.c b/drivers/memory/ti-emif-pm.c index 2250d03ea17f..ab07aa163138 100644 --- a/drivers/memory/ti-emif-pm.c +++ b/drivers/memory/ti-emif-pm.c @@ -138,6 +138,9 @@ static int ti_emif_alloc_sram(struct device *dev, emif_data->pm_functions.exit_sr = sram_resume_address(emif_data, (unsigned long)ti_emif_exit_sr); + emif_data->pm_functions.run_hw_leveling = + sram_resume_address(emif_data, + (unsigned long)ti_emif_run_hw_leveling); emif_data->pm_data.regs_virt = (struct emif_regs_amx3 *)emif_data->ti_emif_sram_data_virt; diff --git a/drivers/memory/ti-emif-sram-pm.S b/drivers/memory/ti-emif-sram-pm.S index a5369181e5c2..d75ae18efa7d 100644 --- a/drivers/memory/ti-emif-sram-pm.S +++ b/drivers/memory/ti-emif-sram-pm.S @@ -27,6 +27,7 @@ #define EMIF_POWER_MGMT_SELF_REFRESH_MODE_MASK 0x0700 #define EMIF_SDCFG_TYPE_DDR2 0x2 << SDRAM_TYPE_SHIFT +#define EMIF_SDCFG_TYPE_DDR3 0x3 << SDRAM_TYPE_SHIFT #define EMIF_STATUS_READY 0x4 #define AM43XX_EMIF_PHY_CTRL_REG_COUNT 0x120 @@ -244,6 +245,46 @@ emif_skip_restore_extra_regs: mov pc, lr ENDPROC(ti_emif_restore_context) +/* + * void ti_emif_run_hw_leveling(void) + * + * Used during resume to run hardware leveling again and restore the + * configuration of the EMIF PHY, only for DDR3. + */ +ENTRY(ti_emif_run_hw_leveling) + adr r4, ti_emif_pm_sram_data + ldr r0, [r4, #EMIF_PM_BASE_ADDR_PHYS_OFFSET] + + ldr r3, [r0, #EMIF_READ_WRITE_LEVELING_CONTROL] + orr r3, r3, #RDWRLVLFULL_START + ldr r2, [r0, #EMIF_SDRAM_CONFIG] + and r2, r2, #SDRAM_TYPE_MASK + cmp r2, #EMIF_SDCFG_TYPE_DDR3 + bne skip_hwlvl + + str r3, [r0, #EMIF_READ_WRITE_LEVELING_CONTROL] + + /* + * If EMIF registers are touched during initial stage of HW + * leveling sequence there will be an L3 NOC timeout error issued + * as the EMIF will not respond, which is not fatal, but it is + * avoidable. This small wait loop is enough time for this condition + * to clear, even at worst case of CPU running at max speed of 1Ghz. + */ + mov r2, #0x2000 +1: + subs r2, r2, #0x1 + bne 1b + + /* Bit clears when operation is complete */ +2: ldr r1, [r0, #EMIF_READ_WRITE_LEVELING_CONTROL] + tst r1, #RDWRLVLFULL_START + bne 2b + +skip_hwlvl: + mov pc, lr +ENDPROC(ti_emif_run_hw_leveling) + /* * void ti_emif_enter_sr(void) * diff --git a/include/linux/ti-emif-sram.h b/include/linux/ti-emif-sram.h index 53604b087f2c..2fc854155c27 100644 --- a/include/linux/ti-emif-sram.h +++ b/include/linux/ti-emif-sram.h @@ -55,6 +55,7 @@ struct ti_emif_pm_data { struct ti_emif_pm_functions { u32 save_context; u32 restore_context; + u32 run_hw_leveling; u32 enter_sr; u32 exit_sr; u32 abort_sr; @@ -126,6 +127,8 @@ static inline void ti_emif_asm_offsets(void) offsetof(struct ti_emif_pm_functions, save_context)); DEFINE(EMIF_PM_RESTORE_CONTEXT_OFFSET, offsetof(struct ti_emif_pm_functions, restore_context)); + DEFINE(EMIF_PM_RUN_HW_LEVELING, + offsetof(struct ti_emif_pm_functions, run_hw_leveling)); DEFINE(EMIF_PM_ENTER_SR_OFFSET, offsetof(struct ti_emif_pm_functions, enter_sr)); DEFINE(EMIF_PM_EXIT_SR_OFFSET,
In certain situations, such as when returning from low power modes, the EMIF must re-run hardware leveling to properly restore DDR3 access. This is accomplished by introducing a new ti-emif-sram-pm call, ti_emif_run_hw_leveling, to check if DDR3 is in use and if so, trigger the full write and read leveling processes. Suggested-by: Brad Griffis <bgriffis@ti.com> Signed-off-by: Dave Gerlach <d-gerlach@ti.com> --- drivers/memory/emif.h | 4 ++++ drivers/memory/ti-emif-pm.c | 3 +++ drivers/memory/ti-emif-sram-pm.S | 41 ++++++++++++++++++++++++++++++++ include/linux/ti-emif-sram.h | 3 +++ 4 files changed, 51 insertions(+)