The memory device generation is guided by qemu paravirt info. Seabios uses the info to setup SRAT entries for the hotplug-able memory slots, and to generate appropriate memory device objects. One memory device (and corresponding SRAT entry) is generated for each hotplug-able qemu memslot. Currently no SSDT memory device is created for initial system memory.
We only support up to 255 DIMMs for now (PackageOp used for the MEON array can only describe an array of at most 255 elements. VarPackageOp would be needed to support more than 255 devices)
Signed-off-by: Vasilis Liaskovitis vasilis.liaskovitis@profitbricks.com Signed-off-by: Hu Tao hutao@cn.fujitsu.com --- src/acpi.c | 151 ++++++++++++++++++++++++++++++++++++++++++++++++++++++--- src/paravirt.c | 8 +++ 2 files changed, 152 insertions(+), 7 deletions(-)
diff --git a/src/acpi.c b/src/acpi.c index ce988e0..e9a0326 100644 --- a/src/acpi.c +++ b/src/acpi.c @@ -15,6 +15,8 @@ #include "config.h" // CONFIG_* #include "paravirt.h" // RamSize #include "dev-q35.h" +#include "memmap.h" +#include "paravirt.h"
#include "acpi-dsdt.hex"
@@ -250,11 +252,23 @@ encodeLen(u8 *ssdt_ptr, int length, int bytes) #define PCIHP_AML (ssdp_pcihp_aml + *ssdt_pcihp_start) #define PCI_SLOTS 32
+/* 0x5B 0x82 DeviceOp PkgLength NameString DimmID */ +#define MEM_BASE 0xaf80 +#define MEM_AML (ssdm_mem_aml + *ssdt_mem_start) +#define MEM_SIZEOF (*ssdt_mem_end - *ssdt_mem_start) +#define MEM_OFFSET_HEX (*ssdt_mem_name - *ssdt_mem_start + 2) +#define MEM_OFFSET_ID (*ssdt_mem_id - *ssdt_mem_start) +#define MEM_OFFSET_PXM 31 +#define MEM_OFFSET_START 55 +#define MEM_OFFSET_END 63 +#define MEM_OFFSET_SIZE 79 + #define SSDT_SIGNATURE 0x54445353 // SSDT #define SSDT_HEADER_LENGTH 36
#include "ssdt-misc.hex" #include "ssdt-pcihp.hex" +#include "ssdt-mem.hex"
#define PCI_RMV_BASE 0xae0c
@@ -306,9 +320,100 @@ static void patch_pcihp(int slot, u8 *ssdt_ptr, u32 eject) } }
+static void build_memdev(u8 *ssdt_ptr, int i, u64 mem_base, u64 mem_len, u8 node) +{ + memcpy(ssdt_ptr, MEM_AML, MEM_SIZEOF); + ssdt_ptr[MEM_OFFSET_HEX] = getHex(i >> 4); + ssdt_ptr[MEM_OFFSET_HEX+1] = getHex(i); + ssdt_ptr[MEM_OFFSET_ID] = i; + ssdt_ptr[MEM_OFFSET_PXM] = node; + *(u64*)(ssdt_ptr + MEM_OFFSET_START) = cpu_to_le64(mem_base); + *(u64*)(ssdt_ptr + MEM_OFFSET_END) = cpu_to_le64(mem_base + mem_len); + *(u64*)(ssdt_ptr + MEM_OFFSET_SIZE) = cpu_to_le64(mem_len); +} + +static u8 *build_memssdt(u8 *ssdt_ptr, int memssdt_len, + u64 *numadimmsmap, int nb_memdevs) +{ + u64 mem_base, mem_len; + u64 *dimm = numadimmsmap; + int node; + int i; + + // build Scope(_SB_) header + *(ssdt_ptr++) = 0x10; // ScopeOp + ssdt_ptr = encodeLen(ssdt_ptr, memssdt_len, 3); + *(ssdt_ptr++) = '_'; + *(ssdt_ptr++) = 'S'; + *(ssdt_ptr++) = 'B'; + *(ssdt_ptr++) = '_'; + + for (i = 0; i < nb_memdevs; i++) { + mem_base = *dimm++; + mem_len = *dimm++; + node = *dimm++; + build_memdev(ssdt_ptr, i, mem_base, mem_len, node); + ssdt_ptr += MEM_SIZEOF; + } + + // build "Method(MTFY, 2) {If (LEqual(Arg0, 0x00)) {Notify(CM00, Arg1)} ...}" + *(ssdt_ptr++) = 0x14; // MethodOp + ssdt_ptr = encodeLen(ssdt_ptr, 2+5+(12*nb_memdevs), 2); + *(ssdt_ptr++) = 'M'; + *(ssdt_ptr++) = 'T'; + *(ssdt_ptr++) = 'F'; + *(ssdt_ptr++) = 'Y'; + *(ssdt_ptr++) = 0x02; + for (i=0; i<nb_memdevs; i++) { + *(ssdt_ptr++) = 0xA0; // IfOp + ssdt_ptr = encodeLen(ssdt_ptr, 11, 1); + *(ssdt_ptr++) = 0x93; // LEqualOp + *(ssdt_ptr++) = 0x68; // Arg0Op + *(ssdt_ptr++) = 0x0A; // BytePrefix + *(ssdt_ptr++) = i; + *(ssdt_ptr++) = 0x86; // NotifyOp + *(ssdt_ptr++) = 'M'; + *(ssdt_ptr++) = 'P'; + *(ssdt_ptr++) = getHex(i >> 4); + *(ssdt_ptr++) = getHex(i); + *(ssdt_ptr++) = 0x69; // Arg1Op + } + + // build "Name(MEON, Package() { One, One, ..., Zero, Zero, ... })" + *(ssdt_ptr++) = 0x08; // NameOp + *(ssdt_ptr++) = 'M'; + *(ssdt_ptr++) = 'E'; + *(ssdt_ptr++) = 'O'; + *(ssdt_ptr++) = 'N'; + *(ssdt_ptr++) = 0x12; // PackageOp + ssdt_ptr = encodeLen(ssdt_ptr, 2+1+(1*nb_memdevs), 2); + *(ssdt_ptr++) = nb_memdevs; + + dimm = numadimmsmap; + u8 memslot_status = 0, enabled; + + for (i = 0; i < nb_memdevs; i++) { + enabled = 0; + if (i % 8 == 0) + memslot_status = inb(MEM_BASE + i/8); + enabled = memslot_status & 1; + mem_base = *dimm++; + mem_len = *dimm++; + dimm++; // node + *(ssdt_ptr++) = enabled ? 0x01 : 0x00; + if (enabled) + add_e820(mem_base, mem_len, E820_RAM); + memslot_status = memslot_status >> 1; + } + + return ssdt_ptr; +} + static void* build_ssdt(void) { + int nb_memdevs; + u64 *numadimmsmap; int acpi_cpus = MaxCountCPUs > 0xff ? 0xff : MaxCountCPUs; int length = (sizeof(ssdp_misc_aml) // _S3_ / _S4_ / _S5_ + (1+3+4) // Scope(_SB_) @@ -318,9 +423,20 @@ build_ssdt(void) + (1+3+4) // Scope(PCI0) + ((PCI_SLOTS - 1) * PCIHP_SIZEOF) // slots + (1+2+5+(12*(PCI_SLOTS - 1)))); // PCNT - u8 *ssdt = malloc_high(length); + + numadimmsmap = romfile_loadfile("etc/numa-dimm-map", &nb_memdevs); + nb_memdevs /= 3 * sizeof(u64); + + // for build_memssdt + int memssdt_length = (1+3+4) + + (nb_memdevs * MEM_SIZEOF) + + (1+2+5+(12*nb_memdevs)) + + (6+2+1+(1*nb_memdevs)); + + u8 *ssdt = malloc_high(length + memssdt_length); if (! ssdt) { warn_noalloc(); + free(numadimmsmap); return NULL; } u8 *ssdt_ptr = ssdt; @@ -411,10 +527,13 @@ build_ssdt(void)
ssdt_ptr = build_notify(ssdt_ptr, "PCNT", 1, PCI_SLOTS, "S00_", 1);
+ ssdt_ptr = build_memssdt(ssdt_ptr, memssdt_length, numadimmsmap, nb_memdevs); + build_header((void*)ssdt, SSDT_SIGNATURE, ssdt_ptr - ssdt, 1);
//hexdump(ssdt, ssdt_ptr - ssdt);
+ free(numadimmsmap); return ssdt; }
@@ -458,7 +577,7 @@ acpi_build_srat_memory(struct srat_memory_affinity *numamem, numamem->length = sizeof(*numamem); memset(numamem->proximity, 0, 4); numamem->proximity[0] = node; - numamem->flags = cpu_to_le32(!!enabled); + numamem->flags = cpu_to_le32(!!enabled) | cpu_to_le32(0x2); numamem->base_addr = cpu_to_le64(base); numamem->range_length = cpu_to_le64(len); } @@ -466,18 +585,22 @@ acpi_build_srat_memory(struct srat_memory_affinity *numamem, static void * build_srat(void) { - int numadatasize, numacpusize; + int numadatasize, numacpusize, nb_numa_dimms; + u64 *numadimmsmap; u64 *numadata = romfile_loadfile("etc/numa-nodes", &numadatasize); u64 *numacpumap = romfile_loadfile("etc/numa-cpu-map", &numacpusize); - if (!numadata || !numacpumap) - goto fail; + int max_cpu = numacpusize / sizeof(u64); int nb_numa_nodes = numadatasize / sizeof(u64);
+ numadimmsmap = romfile_loadfile("etc/numa-dimm-map", &nb_numa_dimms); + + nb_numa_dimms /= 3 * sizeof(u64); + struct system_resource_affinity_table *srat; int srat_size = sizeof(*srat) + sizeof(struct srat_processor_affinity) * max_cpu + - sizeof(struct srat_memory_affinity) * (nb_numa_nodes + 2); + sizeof(struct srat_memory_affinity) * (nb_numa_nodes + nb_numa_dimms + 2);
srat = malloc_high(srat_size); if (!srat) { @@ -512,6 +635,7 @@ build_srat(void) */ struct srat_memory_affinity *numamem = (void*)core; int slots = 0; + int node; u64 mem_len, mem_base, next_base = 0;
acpi_build_srat_memory(numamem, 0, 640*1024, 0, 1); @@ -541,7 +665,18 @@ build_srat(void) numamem++; slots++; } - for (; slots < nb_numa_nodes + 2; slots++) { + if (nb_numa_dimms) { + for (i = 1; i < nb_numa_dimms + 1; ++i) { + mem_base = *numadimmsmap++; + mem_len = *numadimmsmap++; + node = *numadimmsmap++; + acpi_build_srat_memory(numamem, mem_base, mem_len, node, 1); + numamem++; + slots++; + } + } + + for (; slots < nb_numa_nodes + nb_numa_dimms + 2; slots++) { acpi_build_srat_memory(numamem, 0, 0, 0, 0); numamem++; } @@ -550,10 +685,12 @@ build_srat(void)
free(numadata); free(numacpumap); + free(numadimmsmap); return srat; fail: free(numadata); free(numacpumap); + free(numadimmsmap); return NULL; }
diff --git a/src/paravirt.c b/src/paravirt.c index d1a5d3e..5925c63 100644 --- a/src/paravirt.c +++ b/src/paravirt.c @@ -240,6 +240,14 @@ qemu_cfg_legacy(void) , sizeof(numacount) + max_cpu*sizeof(u64) , numacount*sizeof(u64));
+ u64 dimm_count; + qemu_cfg_select(QEMU_CFG_NUMA); + qemu_cfg_skip((1 + max_cpu + numacount) * sizeof(u64)); + qemu_cfg_read(&dimm_count, sizeof(dimm_count)); + qemu_romfile_add("etc/numa-dimm-map", QEMU_CFG_NUMA + , (2 + max_cpu + numacount) * sizeof(u64), + dimm_count * 3 * sizeof(u64)); + // e820 data u32 count32; qemu_cfg_read_entry(&count32, QEMU_CFG_E820_TABLE, sizeof(count32));