The code that reprograms the SDRC memory controller during CORE DVFS, mach-omap2/sram34xx.S:omap3_sram_configure_core_dpll(), does not ensure that all L3 initiators are prevented from accessing the SDRAM before modifying the SDRC AC timing and MR registers. This can cause memory to be corrupted or cause the SDRC to enter an unpredictable state. This patch comments out that code for now and adds a note explaining what is going on. Ideally it can be added back in once supporting code is present to ensure that other initiators aren't touching the SDRAM. At the very least, these registers should be reprogrammable during kernel init to deal with buggy bootloaders. This is a modification of a patch originally written by Rajendra Nayak <rnayak@xxxxxx> (the original is at http://patchwork.kernel.org/patch/51927/). Rather than removing the code completely, this patch just comments it out. Thanks to Benoît Cousson <b-cousson@xxxxxx> and Christophe Sucur <c-sucur@xxxxxx> for explaining the technical basis for this and for explaining what can be done to make this path work in future code. Thanks to Richard Woodruff <r-woodruff2@xxxxxx> for his comments. Signed-off-by: Paul Walmsley <paul@xxxxxxxxx> Cc: Rajendra Nayak <rnayak@xxxxxx> Cc: Christophe Sucur <c-sucur@xxxxxx> Cc: Benoît Cousson <b-cousson@xxxxxx> Cc: Richard Woodruff <r-woodruff2@xxxxxx> --- arch/arm/mach-omap2/sram34xx.S | 18 ++++++++++++++++-- 1 files changed, 16 insertions(+), 2 deletions(-) diff --git a/arch/arm/mach-omap2/sram34xx.S b/arch/arm/mach-omap2/sram34xx.S index 82aa4a3..8fa8955 100644 --- a/arch/arm/mach-omap2/sram34xx.S +++ b/arch/arm/mach-omap2/sram34xx.S @@ -91,8 +91,18 @@ * new SDRC_ACTIM_CTRL_B_1 register contents * new SDRC_MR_1 register value * - * If the param SDRC_RFR_CTRL_1 is 0, the parameters - * are not programmed into the SDRC CS1 registers + * If the param SDRC_RFR_CTRL_1 is 0, the parameters are not programmed into + * the SDRC CS1 registers + * + * NOTE: This code no longer attempts to program the SDRC AC timing and MR + * registers. This is because the code currently cannot ensure that all + * L3 initiators (e.g., sDMA, IVA, DSS DISPC, etc.) are not accessing the + * SDRAM when the registers are written. If the registers are changed while + * an initiator is accessing SDRAM, memory can be corrupted and/or the SDRC + * may enter an unpredictable state. The code to reprogram the registers, + * however, has been left in -- commented out in "#if 0" .. "#endif" blocks -- + * since in the future, the intent is to re-enable this code in cases where we + * can ensure that no initiators are touching the SDRAM. */ ENTRY(omap3_sram_configure_core_dpll) stmfd sp!, {r1-r12, lr} @ store regs to stack @@ -219,6 +229,7 @@ configure_sdrc: ldr r12, omap_sdrc_rfr_ctrl_0_val @ fetch value from SRAM ldr r11, omap3_sdrc_rfr_ctrl_0 @ fetch addr from SRAM str r12, [r11] @ store +#if 0 ldr r12, omap_sdrc_actim_ctrl_a_0_val ldr r11, omap3_sdrc_actim_ctrl_a_0 str r12, [r11] @@ -228,11 +239,13 @@ configure_sdrc: ldr r12, omap_sdrc_mr_0_val ldr r11, omap3_sdrc_mr_0 str r12, [r11] +#endif ldr r12, omap_sdrc_rfr_ctrl_1_val cmp r12, #0 @ if SDRC_RFR_CTRL_1 is 0, beq skip_cs1_prog @ do not program cs1 params ldr r11, omap3_sdrc_rfr_ctrl_1 str r12, [r11] +#if 0 ldr r12, omap_sdrc_actim_ctrl_a_1_val ldr r11, omap3_sdrc_actim_ctrl_a_1 str r12, [r11] @@ -242,6 +255,7 @@ configure_sdrc: ldr r12, omap_sdrc_mr_1_val ldr r11, omap3_sdrc_mr_1 str r12, [r11] +#endif skip_cs1_prog: ldr r12, [r11] @ posted-write barrier for SDRC bx lr -- 1.6.5.GIT