At 09/21/2012 07:17 PM, Vasilis Liaskovitis Wrote:
> pcimem_start and pcimem64_start are adjusted from srat entries. For this reason,
> paravirt info (NUMA SRAT entries and number of cpus) need to be read before pci_setup.
> Imho, this is an ugly code change since SRAT bios tables and number of
> cpus have to be read earlier. But the advantage is that no new paravirt interface
> is introduced. Suggestions to make the code change cleaner are welcome.
>
> The alternative patch (will be sent as a reply to this patch) implements a
> paravirt interface to read the starting values of pcimem_start and
> pcimem64_start from QEMU.
>
> Signed-off-by: Vasilis Liaskovitis <vasilis.liaskovitis@xxxxxxxxxxxxxxxx>
> ---
> src/acpi.c | 82 ++++++++++++++++++++++++++++++++++++++++----------------
> src/acpi.h | 3 ++
> src/pciinit.c | 6 +++-
> src/post.c | 3 ++
> src/smp.c | 4 +++
> 5 files changed, 72 insertions(+), 26 deletions(-)
>
> diff --git a/src/acpi.c b/src/acpi.c
> index 1223b52..9e99aa7 100644
> --- a/src/acpi.c
> +++ b/src/acpi.c
> @@ -428,7 +428,10 @@ encodeLen(u8 *ssdt_ptr, int length, int bytes)
> #define MEM_OFFSET_END 63
> #define MEM_OFFSET_SIZE 79
>
> -u64 nb_hp_memslots = 0;
> +u64 nb_hp_memslots = 0, nb_numanodes;
> +u64 *numa_data, *hp_memdata;
> +u64 below_4g_hp_mem_size = 0;
> +u64 above_4g_hp_mem_size = 0;
> struct srat_memory_affinity *mem;
>
> #define SSDT_SIGNATURE 0x54445353 // SSDT
> @@ -763,17 +766,7 @@ acpi_build_srat_memory(struct srat_memory_affinity *numamem,
> static void *
> build_srat(void)
> {
> - int nb_numa_nodes = qemu_cfg_get_numa_nodes();
> -
> - u64 *numadata = malloc_tmphigh(sizeof(u64) * (MaxCountCPUs + nb_numa_nodes));
> - if (!numadata) {
> - warn_noalloc();
> - return NULL;
> - }
> -
> - qemu_cfg_get_numa_data(numadata, MaxCountCPUs + nb_numa_nodes);
> -
> - qemu_cfg_get_numa_data(&nb_hp_memslots, 1);
> + int nb_numa_nodes = nb_numanodes;
> struct system_resource_affinity_table *srat;
> int srat_size = sizeof(*srat) +
> sizeof(struct srat_processor_affinity) * MaxCountCPUs +
> @@ -782,7 +775,7 @@ build_srat(void)
> srat = malloc_high(srat_size);
> if (!srat) {
> warn_noalloc();
> - free(numadata);
> + free(numa_data);
> return NULL;
> }
>
> @@ -791,6 +784,7 @@ build_srat(void)
> struct srat_processor_affinity *core = (void*)(srat + 1);
> int i;
> u64 curnode;
> + u64 *numadata = numa_data;
>
> for (i = 0; i < MaxCountCPUs; ++i) {
> core->type = SRAT_PROCESSOR;
> @@ -847,15 +841,7 @@ build_srat(void)
> mem = (void*)numamem;
>
> if (nb_hp_memslots) {
> - u64 *hpmemdata = malloc_tmphigh(sizeof(u64) * (3 * nb_hp_memslots));
> - if (!hpmemdata) {
> - warn_noalloc();
> - free(hpmemdata);
> - free(numadata);
> - return NULL;
> - }
> -
> - qemu_cfg_get_numa_data(hpmemdata, 3 * nb_hp_memslots);
> + u64 *hpmemdata = hp_memdata;
>
> for (i = 1; i < nb_hp_memslots + 1; ++i) {
> mem_base = *hpmemdata++;
> @@ -865,7 +851,7 @@ build_srat(void)
> numamem++;
> slots++;
> }
> - free(hpmemdata);
> + free(hp_memdata);
> }
>
> for (; slots < nb_numa_nodes + nb_hp_memslots + 2; slots++) {
> @@ -875,10 +861,58 @@ build_srat(void)
>
> build_header((void*)srat, SRAT_SIGNATURE, srat_size, 1);
>
> - free(numadata);
> + free(numa_data);
> return srat;
> }
>
> +/* QEMU paravirt SRAT entries need to be read in before pci initilization */
> +void read_srat_early(void)
> +{
> + int i;
> +
> + nb_numanodes = qemu_cfg_get_numa_nodes();
> + u64 *hpmemdata;
> + u64 mem_len, mem_base;
> +
> + numa_data = malloc_tmphigh(sizeof(u64) * (MaxCountCPUs + nb_numanodes));
> + if (!numa_data) {
> + warn_noalloc();
> + }
> +
> + qemu_cfg_get_numa_data(numa_data, MaxCountCPUs + nb_numanodes);
> + qemu_cfg_get_numa_data(&nb_hp_memslots, 1);
> +
> + if (nb_hp_memslots) {
> + hp_memdata = malloc_tmphigh(sizeof(u64) * (3 * nb_hp_memslots));
> + if (!hp_memdata) {
> + warn_noalloc();
> + free(hp_memdata);
> + free(numa_data);
> + }
> +
> + qemu_cfg_get_numa_data(hp_memdata, 3 * nb_hp_memslots);
> + hpmemdata = hp_memdata;
> +
> + for (i = 1; i < nb_hp_memslots + 1; ++i) {
> + mem_base = *hpmemdata++;
> + mem_len = *hpmemdata++;
> + hpmemdata++;
> + if (mem_base >= 0x100000000LL) {
> + above_4g_hp_mem_size += mem_len;
> + }
> + /* if dimm fits before pci hole, append it normally */
> + else if (mem_base + mem_len <= BUILD_PCIMEM_START) {
> + below_4g_hp_mem_size += mem_len;
> + }
> + /* otherwise place it above 4GB */
> + else {
> + above_4g_hp_mem_size += mem_len;
> + }
> + }
> +
> + }
> +}
> +
> static const struct pci_device_id acpi_find_tbl[] = {
> /* PIIX4 Power Management device. */
> PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_82371AB_3, NULL),
> diff --git a/src/acpi.h b/src/acpi.h
> index cb21561..d29837f 100644
> --- a/src/acpi.h
> +++ b/src/acpi.h
> @@ -5,6 +5,9 @@
>
> void acpi_bios_init(void);
> u32 find_resume_vector(void);
> +void read_srat_early(void);
> +extern u64 below_4g_hp_mem_size;
> +extern u64 above_4g_hp_mem_size;
>
> #define RSDP_SIGNATURE 0x2052545020445352LL // "RSD PTR "
>
> diff --git a/src/pciinit.c b/src/pciinit.c
> index 31115ee..c5a4b24 100644
> --- a/src/pciinit.c
> +++ b/src/pciinit.c
> @@ -12,6 +12,7 @@
> #include "ioport.h" // PORT_ATA1_CMD_BASE
> #include "config.h" // CONFIG_*
> #include "xen.h" // usingXen
> +#include "acpi.h"
>
> #define PCI_DEVICE_MEM_MIN 0x1000
> #define PCI_BRIDGE_IO_MIN 0x1000
> @@ -597,7 +598,7 @@ static void pci_region_map_entries(struct pci_bus *busses, struct pci_region *r)
>
> static void pci_bios_map_devices(struct pci_bus *busses)
> {
> - pcimem_start = RamSize;
> + pcimem_start = RamSize + below_4g_hp_mem_size;
>
> if (pci_bios_init_root_regions(busses)) {
> struct pci_region r64_mem, r64_pref;
> @@ -616,7 +617,8 @@ static void pci_bios_map_devices(struct pci_bus *busses)
> u64 align_mem = pci_region_align(&r64_mem);
> u64 align_pref = pci_region_align(&r64_pref);
>
> - r64_mem.base = ALIGN(0x100000000LL + RamSizeOver4G, align_mem);
> + r64_mem.base = ALIGN(0x100000000LL + RamSizeOver4G +
> + above_4g_hp_mem_size, align_mem);
> r64_pref.base = ALIGN(r64_mem.base + sum_mem, align_pref);
> pcimem64_start = r64_mem.base;
> pcimem64_end = r64_pref.base + sum_pref;
> diff --git a/src/post.c b/src/post.c
> index 924b311..c37730b 100644
> --- a/src/post.c
> +++ b/src/post.c
> @@ -234,6 +234,9 @@ maininit(void)
> // Initialize mtrr
> mtrr_setup();
>
> + smp_get_ncpus();
> + read_srat_early();
> +
> // Initialize pci
> pci_setup();
> smm_init();
> diff --git a/src/smp.c b/src/smp.c
> index 4975412..3922776 100644
> --- a/src/smp.c
> +++ b/src/smp.c
> @@ -138,7 +138,11 @@ smp_probe(void)
>
> // Restore memory.
> *(u64*)BUILD_AP_BOOT_ADDR = old;
> +}
>
> +void
> +smp_get_ncpus(void)
You don't declare this function, and use it in another file. It will break
building:
src/post.c: In function ‘maininit’:
src/post.c:237: warning: implicit declaration of function ‘smp_get_ncpus’
src/smp.c:144: note: previous definition of ‘smp_get_ncpus’ was here
src/post.c:237: error: incompatible implicit declaration of function ‘smp_get_ncpus’
src/smp.c:144: note: previous definition of ‘smp_get_ncpus’ was here
Thanks
Wen Congyang
> +{
> MaxCountCPUs = qemu_cfg_get_max_cpus();
> if (!MaxCountCPUs || MaxCountCPUs < CountCPUs)
> MaxCountCPUs = CountCPUs;
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
