--- zzzz-none-000/linux-2.6.39.4/arch/x86/include/asm/mach_timer.h 2011-08-03 19:43:28.000000000 +0000 +++ puma6-arm-6490-729/linux-2.6.39.4/arch/x86/include/asm/mach_timer.h 2021-11-10 13:23:10.000000000 +0000 @@ -10,6 +10,16 @@ * directly because of the awkward 8-bit access mechanism of the 82C54 * device. */ + +/****************************************************************** + + Includes Intel Corporation's changes/modifications dated: 07/2011. + Changed/modified portions - Copyright(c) 2011, Intel Corporation. + +******************************************************************/ + + + #ifndef _ASM_X86_MACH_DEFAULT_MACH_TIMER_H #define _ASM_X86_MACH_DEFAULT_MACH_TIMER_H @@ -39,9 +49,31 @@ static inline void mach_countup(unsigned long *count_p) { unsigned long count = 0; +/* + * The following code is for Intel Media SOC Gen3 B0 and B1 workaround. +*/ +#ifdef CONFIG_ARCH_GEN3 +/* + * The 8254 timer flop has an issue between the reset and set pin of a flop. + * This causes a race condition to happen and the value can change based on + * clock sku. The workaround for this silicon issue uses a polling method which + * is not affected by race condition. If this workaround is not applied, then the + * cpu MHz entry in /proc/cpuinfo may show the wrong information. This workaround + * is for errata number 22 in Errata - A Step. +*/ + u8 count1; + u8 count2; + do { + count++; + outb(0x80, 0x43); + count1 = inb(0x42); + count2 = inb(0x42); + } while (count1 >= 0x20 || count2 != 0); +#else do { count++; } while ((inb_p(0x61) & 0x20) == 0); +#endif *count_p = count; }