[PATCHv3 5/9] ARM: OMAP2+: AM33XX: Add assembly code for PM operations
Santosh Shilimkar
santosh.shilimkar at ti.com
Thu Aug 8 10:50:35 EDT 2013
On Tuesday 06 August 2013 01:49 PM, Dave Gerlach wrote:
> From: Vaibhav Bedia <vaibhav.bedia at ti.com>
>
> In preparation for suspend-resume support for AM33XX, add
> the assembly file with the code which is copied to internal
> memory (OCMC RAM) during bootup and runs from there.
>
> As part of the low power entry (DeepSleep0 mode in AM33XX TRM),
> the code running from OCMC RAM does the following
> 1. Stores the EMIF configuration
> 2. Puts external memory in self-refresh
> 3. Disables EMIF clock
> 4. Executes WFI after writing to MPU_CLKCTRL register.
>
> If no interrupts have come, WFI execution on MPU gets registered
> as an interrupt with the WKUP-M3. WKUP-M3 takes care of disabling
> some clocks which MPU should not (L3, L4, OCMC RAM etc) and takes
> care of clockdomain and powerdomain transitions as part of the
> DeepSleep0 mode entry.
>
> In case a late interrupt comes in, WFI ends up as a NOP and MPU
> continues execution from internal memory. The 'abort path' code
> undoes whatever was done as part of the low power entry and indicates
> a suspend failure by passing a non-zero value to the cpu_resume routine.
>
> The 'resume path' code is similar to the 'abort path' with the key
> difference of MMU being enabled in the 'abort path' but being
> disabled in the 'resume path' due to MPU getting powered off.
>
> Signed-off-by: Vaibhav Bedia <vaibhav.bedia at ti.com>
> Signed-off-by: Dave Gerlach <d-gerlach at ti.com>
> Cc: Santosh Shilimkar <santosh.shilimkar at ti.com>
> Cc: Kevin Hilman <khilman at linaro.org>
> ---
> arch/arm/mach-omap2/sleep33xx.S | 350 +++++++++++++++++++++++++++++++++++++++
> 1 file changed, 350 insertions(+)
> create mode 100644 arch/arm/mach-omap2/sleep33xx.S
>
> diff --git a/arch/arm/mach-omap2/sleep33xx.S b/arch/arm/mach-omap2/sleep33xx.S
> new file mode 100644
> index 0000000..834c7d4
> --- /dev/null
> +++ b/arch/arm/mach-omap2/sleep33xx.S
> @@ -0,0 +1,350 @@
> +/*
> + * Low level suspend code for AM33XX SoCs
> + *
> + * Copyright (C) 2012 Texas Instruments Incorporated - http://www.ti.com/
> + * Vaibhav Bedia <vaibhav.bedia at ti.com>
> + *
> + * This program is free software; you can redistribute it and/or
> + * modify it under the terms of the GNU General Public License as
> + * published by the Free Software Foundation version 2.
> + *
> + * This program is distributed "as is" WITHOUT ANY WARRANTY of any
> + * kind, whether express or implied; without even the implied warranty
> + * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> + * GNU General Public License for more details.
> + */
> +
> +#include <linux/linkage.h>
> +#include <linux/ti_emif.h>
> +#include <asm/memory.h>
> +#include <asm/assembler.h>
> +
> +#include "cm33xx.h"
> +#include "pm33xx.h"
> +#include "prm33xx.h"
> +
> + .text
> + .align 3
> +
> +/*
> + * This routine is executed from internal RAM and expects some
> + * parameters to be passed in r0 _strictly_ in following order:
> + * 1) emif_addr_virt - ioremapped EMIF address
> + * 2) mem_type - 2 -> DDR2, 3-> DDR3
> + * 3) dram_sync_word - uncached word in SDRAM
> + *
> + * The code loads these values taking r0 value as reference to
> + * the array in registers starting from r0, i.e emif_addr_virt
> + * goes to r1, mem_type goes to r2 and and so on. These are
> + * then saved into memory locations before proceeding with the
> + * sleep sequence and hence registers r0, r1 etc can still be
> + * used in the rest of the sleep code.
> + */
> +
> +ENTRY(am33xx_do_wfi)
> + stmfd sp!, {r4 - r11, lr} @ save registers on stack
> +
> + ldm r0, {r1-r3} @ gather values passed
> +
> + /* Save the values passed */
> + str r1, emif_addr_virt
> + str r2, mem_type
> + str r3, dram_sync_word
None of this parameter are going to change for every suspend entry and
exit so saving them once and accessing them should be fine. Just
create a structure with above, save them in init from C code and
then access that structure where you need to.
> +
> + /*
> + * Flush all data from the L1 data cache before disabling
> + * SCTLR.C bit.
> + */
> + ldr r1, kernel_flush
> + blx r1
> +
> + /*
> + * Clear the SCTLR.C bit to prevent further data cache
> + * allocation. Clearing SCTLR.C would make all the data accesses
> + * strongly ordered and would not hit the cache.
> + */
> + mrc p15, 0, r0, c1, c0, 0
> + bic r0, r0, #(1 << 2) @ Disable the C bit
> + mcr p15, 0, r0, c1, c0, 0
> + isb
> +
> + /*
> + * Invalidate L1 data cache. Even though only invalidate is
> + * necessary exported flush API is used here. Doing clean
> + * on already clean cache would be almost NOP.
> + */
Comment is stale for AM33XX since below flush will clean l1 and l2
together. We need to first flush and then invalidate. Please update it.
> + ldr r1, kernel_flush
> + blx r1
> +
> + ldr r0, emif_addr_virt
> + /* Save EMIF configuration */
> + ldr r1, [r0, #EMIF_SDRAM_CONFIG]
> + str r1, emif_sdcfg_val
> + ldr r1, [r0, #EMIF_SDRAM_REFRESH_CONTROL]
> + str r1, emif_ref_ctrl_val
> + ldr r1, [r0, #EMIF_SDRAM_TIMING_1]
> + str r1, emif_timing1_val
> + ldr r1, [r0, #EMIF_SDRAM_TIMING_2]
> + str r1, emif_timing2_val
> + ldr r1, [r0, #EMIF_SDRAM_TIMING_3]
> + str r1, emif_timing3_val
> + ldr r1, [r0, #EMIF_POWER_MANAGEMENT_CONTROL]
> + str r1, emif_pmcr_val
> + ldr r1, [r0, #EMIF_POWER_MANAGEMENT_CTRL_SHDW]
> + str r1, emif_pmcr_shdw_val
> + ldr r1, [r0, #EMIF_SDRAM_OUTPUT_IMPEDANCE_CALIBRATION_CONFIG]
> + str r1, emif_zqcfg_val
> + ldr r1, [r0, #EMIF_DDR_PHY_CTRL_1]
> + str r1, emif_rd_lat_val
> +
> + /* Put SDRAM in self-refresh */
> + ldr r1, [r0, #EMIF_POWER_MANAGEMENT_CONTROL]
> + orr r1, r1, #0xa0
No magic numbers please. here and rest of the patch. Have proper
defines.
> + str r1, [r0, #EMIF_POWER_MANAGEMENT_CTRL_SHDW]
> + str r1, [r0, #4]
> +
> + ldr r1, dram_sync_word @ a dummy access to DDR as per spec
> + ldr r2, [r1, #0]
> + ldr r1, [r0, #EMIF_POWER_MANAGEMENT_CONTROL]
> + orr r1, r1, #0x200
> + str r1, [r0, #EMIF_POWER_MANAGEMENT_CONTROL]
> +
> + mov r1, #0x1000 @ Wait for system to enter SR
> +wait_sr:
What is sr ? I know you mean self-refresh, but sr can be smart reflex
as well. Fix that please.
> + subs r1, r1, #1
> + bne wait_sr
> +
So how did you derive this 0x1000 number. Whats the real
delay need ? For e.g a CPU running at 600 MHz vs running
at 1.2 GHz, the above loop becomes 2 times faster.
> + /* Disable EMIF */
> + ldr r1, virt_emif_clkctrl
> + ldr r2, [r1]
> + bic r2, r2, #0x03
> + str r2, [r1]
> +
> + ldr r1, virt_emif_clkctrl
> +wait_emif_disable:
> + ldr r2, [r1]
> + ldr r3, module_disabled_val
> + cmp r2, r3
> + bne wait_emif_disable
> +
> + /*
> + * For the MPU WFI to be registered as an interrupt
> + * to WKUP_M3, MPU_CLKCTRL.MODULEMODE needs to be set
> + * to DISABLED
> + */
> + ldr r1, virt_mpu_clkctrl
> + ldr r2, [r1]
> + bic r2, r2, #0x03
Magic value
> + str r2, [r1]
> +
> + /*
> + * Execute an ISB instruction to ensure that all of the
> + * CP15 register changes have been committed.
> + */
> + isb
> +
> + /*
> + * Execute a barrier instruction to ensure that all cache,
> + * TLB and branch predictor maintenance operations issued
> + * have completed.
> + */
> + dsb
> + dmb
> +
> + /*
> + * Execute a WFI instruction and wait until the
> + * STANDBYWFI output is asserted to indicate that the
> + * CPU is in idle and low power state. CPU can specualatively
> + * prefetch the instructions so add NOPs after WFI. Thirteen
> + * NOPs as per Cortex-A8 pipeline.
> + */
> + wfi
> +
> + nop
> + nop
> + nop
> + nop
> + nop
> + nop
> + nop
> + nop
> + nop
> + nop
> + nop
> + nop
> + nop
> +
> + /* We come here in case of an abort due to a late interrupt */
> +
> + /* Set MPU_CLKCTRL.MODULEMODE back to ENABLE */
> + ldr r1, virt_mpu_clkctrl
> + mov r2, #0x02
> + str r2, [r1]
> +
> + /* Re-enable EMIF */
> + ldr r1, virt_emif_clkctrl
> + mov r2, #0x02
> + str r2, [r1]
> +wait_emif_enable:
> + ldr r3, [r1]
> + cmp r2, r3
> + bne wait_emif_enable
> +
> + /* Disable EMIF self-refresh */
> + ldr r0, emif_addr_virt
> + ldr r1, [r0, #EMIF_POWER_MANAGEMENT_CONTROL]
> + bic r1, r1, #LP_MODE_MASK
> + str r1, [r0, #EMIF_POWER_MANAGEMENT_CONTROL]
> + str r1, [r0, #EMIF_POWER_MANAGEMENT_CTRL_SHDW]
> +
> + /*
> + * A write to SDRAM CONFIG register triggers
> + * an init sequence and hence it must be done
> + * at the end for DDR2
> + */
> + ldr r0, emif_addr_virt
> + add r0, r0, #EMIF_SDRAM_CONFIG
> + ldr r4, emif_sdcfg_val
> + str r4, [r0]
> +
> + /*
> + * Set SCTLR.C bit to allow data cache allocation
> + */
> + mrc p15, 0, r0, c1, c0, 0
> + orr r0, r0, #(1 << 2) @ Enable the C bit
> + mcr p15, 0, r0, c1, c0, 0
> + isb
> +
> + /* Kill some time for sanity to settle in */
Really ?
> + mov r0, #0x1000
> +wait_abt:
> + subs r0, r0, #1
> + bne wait_abt
Why do you want to kill time ? How
does this 0x1000 delay sanities it.
> +
> + /* Let the suspend code know about the abort */
> + mov r0, #1
> + ldmfd sp!, {r4 - r11, pc} @ restore regs and return
> +ENDPROC(am33xx_do_wfi)
> +
> + .align
> +ENTRY(am33xx_resume_offset)
> + .word . - am33xx_do_wfi
> +
> +ENTRY(am33xx_resume_from_deep_sleep)
> + /* Re-enable EMIF */
> + ldr r0, phys_emif_clkctrl
> + mov r1, #0x02
> + str r1, [r0]
> +wait_emif_enable1:
> + ldr r2, [r0]
> + cmp r1, r2
> + bne wait_emif_enable1
> +
> + /* Config EMIF Timings */
> + ldr r0, emif_phys_addr
> + ldr r1, emif_rd_lat_val
> + str r1, [r0, #EMIF_DDR_PHY_CTRL_1]
> + str r1, [r0, #EMIF_DDR_PHY_CTRL_1_SHDW]
> + ldr r1, emif_timing1_val
> + str r1, [r0, #EMIF_SDRAM_TIMING_1]
> + str r1, [r0, #EMIF_SDRAM_TIMING_1_SHDW]
> + ldr r1, emif_timing2_val
> + str r1, [r0, #EMIF_SDRAM_TIMING_2]
> + str r1, [r0, #EMIF_SDRAM_TIMING_2_SHDW]
> + ldr r1, emif_timing3_val
> + str r1, [r0, #EMIF_SDRAM_TIMING_3]
> + str r1, [r0, #EMIF_SDRAM_TIMING_3_SHDW]
> + ldr r1, emif_ref_ctrl_val
> + str r1, [r0, #EMIF_SDRAM_REFRESH_CONTROL]
> + str r1, [r0, #EMIF_SDRAM_REFRESH_CTRL_SHDW]
> + ldr r1, emif_pmcr_val
> + str r1, [r0, #EMIF_POWER_MANAGEMENT_CONTROL]
> + ldr r1, emif_pmcr_shdw_val
> + str r1, [r0, #EMIF_POWER_MANAGEMENT_CTRL_SHDW]
> +
> + /*
> + * Output impedence calib needed only for DDR3
> + * but since the initial state of this will be
> + * disabled for DDR2 no harm in restoring the
> + * old configuration
> + */
> + ldr r1, emif_zqcfg_val
> + str r1, [r0, #EMIF_SDRAM_OUTPUT_IMPEDANCE_CALIBRATION_CONFIG]
> +
> + /* Write to SDRAM_CONFIG only for DDR2 */
> + ldr r2, mem_type
> + cmp r2, #MEM_TYPE_DDR2
> + bne resume_to_ddr
> +
> + /*
> + * A write to SDRAM CONFIG register triggers
> + * an init sequence and hence it must be done
> + * at the end for DDR2
> + */
> + ldr r1, emif_sdcfg_val
> + str r1, [r0, #EMIF_SDRAM_CONFIG]
> +
> +resume_to_ddr:
> + /* Back from la-la-land. Kill some time for sanity to settle in */
> + mov r0, #0x1000
> +wait_resume:
> + subs r0, r0, #1
> + bne wait_resume
> +
You are killing too much time ;-)
without mentioning why ?
> + /* We are back. Branch to the common CPU resume routine */
> + mov r0, #0
> + ldr pc, resume_addr
Why can't you resume to "cpu_resume" directly.
> +ENDPROC(am33xx_resume_from_deep_sleep)
> +
> +
> +/*
> + * Local variables
> + */
> + .align
> +resume_addr:
> + .word cpu_resume - PAGE_OFFSET + 0x80000000
Do you really need above math ?
> +kernel_flush:
> + .word v7_flush_dcache_all
> +ddr_start:
> + .word PAGE_OFFSET
> +emif_phys_addr:
> + .word AM33XX_EMIF_BASE
> +virt_mpu_clkctrl:
> + .word AM33XX_CM_MPU_MPU_CLKCTRL
> +virt_emif_clkctrl:
> + .word AM33XX_CM_PER_EMIF_CLKCTRL
> +phys_emif_clkctrl:
> + .word (AM33XX_CM_BASE + AM33XX_CM_PER_MOD + \
> + AM33XX_CM_PER_EMIF_CLKCTRL_OFFSET)
> +module_disabled_val:
> + .word 0x30000
> +
> +/* DDR related defines */
> +dram_sync_word:
> + .word 0xDEADBEEF
> +mem_type:
> + .word 0xDEADBEEF
> +emif_addr_virt:
> + .word 0xDEADBEEF
> +emif_rd_lat_val:
> + .word 0xDEADBEEF
> +emif_timing1_val:
> + .word 0xDEADBEEF
> +emif_timing2_val:
> + .word 0xDEADBEEF
> +emif_timing3_val:
> + .word 0xDEADBEEF
> +emif_sdcfg_val:
> + .word 0xDEADBEEF
> +emif_ref_ctrl_val:
> + .word 0xDEADBEEF
> +emif_zqcfg_val:
> + .word 0xDEADBEEF
> +emif_pmcr_val:
> + .word 0xDEADBEEF
> +emif_pmcr_shdw_val:
> + .word 0xDEADBEEF
> +
You can create a structure above above regs.
refer 'cache-l2x0.h' struct l2x0_regs in case
you need an example.
Looks like you don't care about secure devices ?
Just confirm it.
Regards,
Santosh
More information about the linux-arm-kernel
mailing list