| 123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101 |
- /*
- * MTRR write-combining.
- *
- * The VESA LFB is mapped uncacheable (UC) by paging.c, so every dword the
- * 30 Hz shadow flush writes to VRAM is a separate uncached PCI transaction.
- * Marking the LFB range write-combining (WC) via an MTRR lets the CPU batch
- * those writes into burst transfers, which is the single biggest rendering
- * win on real hardware. (The "P5 USWC" write-combining comment in vga_copy.S
- * does not hold on the Pentium 4 / Athlon class machines that ship with
- * Radeon RV280 cards.)
- */
- #include "mtrr.h"
- #include "serial.h"
- #define MSR_MTRRCAP 0x0FE
- #define MSR_MTRR_DEF_TYPE 0x2FF
- #define MSR_MTRR_PHYSBASE0 0x200
- #define MSR_MTRR_PHYSMASK0 0x201
- #define MTRR_TYPE_WC 0x01
- static void rdmsr(unsigned int msr, unsigned int *lo, unsigned int *hi)
- {
- __asm__ volatile ("rdmsr" : "=a"(*lo), "=d"(*hi) : "c"(msr));
- }
- static void wrmsr(unsigned int msr, unsigned int lo, unsigned int hi)
- {
- __asm__ volatile ("wrmsr" : : "a"(lo), "d"(hi), "c"(msr));
- }
- static int cpu_has_mtrr(void)
- {
- unsigned int eax, ebx, ecx, edx;
- __asm__ volatile ("cpuid"
- : "=a"(eax), "=b"(ebx), "=c"(ecx), "=d"(edx)
- : "a"(1), "c"(0));
- return (edx & (1u << 12)) != 0;
- }
- void mtrr_set_wc(unsigned int base, unsigned int size)
- {
- unsigned int lo, hi;
- unsigned int def_type;
- unsigned int vcnt;
- unsigned int region_size, align;
- unsigned int physmask;
- if (!cpu_has_mtrr())
- return;
- if (size == 0)
- return;
- rdmsr(MSR_MTRRCAP, &lo, &hi);
- vcnt = lo & 0xFF;
- if (vcnt == 0)
- return;
- /* Round size up to the next power of two (min 4 KB). */
- region_size = 0x1000;
- while (region_size < size) {
- region_size <<= 1;
- if (region_size == 0)
- return; /* overflow (size > 2 GB) -- give up */
- }
- /* A variable-range MTRR must be aligned to its own size. The largest
- * region we may use is therefore the base address's own power-of-two
- * alignment (base & -base). */
- align = base & (~base + 1u);
- if (align == 0 || align > 0x80000000u)
- align = 0x80000000u;
- if (region_size > align)
- region_size = align;
- if (region_size < 0x1000)
- return;
- base &= ~(region_size - 1);
- /* PhysMask: address bits [11:0] are implicitly don't-care; bits
- * [35:12] are the mask, with a 1 = don't-care. For a <4 GB region the
- * high 32-bit mask is 0 (bits 32..35 must match physbase == 0). */
- physmask = ((region_size - 1) & 0xFFFFF000u) | 0x800;
- /* Update MTRR with the global enable temporarily cleared so an
- * in-progress region never exposes a bogus memory type. */
- rdmsr(MSR_MTRR_DEF_TYPE, &def_type, &hi);
- wrmsr(MSR_MTRR_DEF_TYPE, def_type & ~0x800u, 0);
- wrmsr(MSR_MTRR_PHYSBASE0, base | MTRR_TYPE_WC, 0);
- wrmsr(MSR_MTRR_PHYSMASK0, physmask, 0);
- wrmsr(MSR_MTRR_DEF_TYPE, def_type, 0);
- serial_puts(SERIAL_COM1, "mtrr: LFB WC [");
- serial_puthex(SERIAL_COM1, base);
- serial_puts(SERIAL_COM1, " +");
- serial_puthex(SERIAL_COM1, region_size);
- serial_puts(SERIAL_COM1, "]\n");
- }
|