Commit 991dfefd authored by Vasilis Liaskovitis's avatar Vasilis Liaskovitis Committed by Marcelo Tosatti
Browse files

Set numa topology for max_cpus 



qemu-kvm passes numa/SRAT topology information for smp_cpus to SeaBIOS. However
SeaBIOS always expects to setup max_cpus number of SRAT cpu entries
(MaxCountCPUs variable in build_srat function of Seabios). When qemu-kvm runs
with smp_cpus != max_cpus (e.g. -smp 2,maxcpus=4), Seabios will mistakenly use
memory SRAT info for setting up CPU SRAT entries for the offline CPUs. Wrong
SRAT memory entries are also created. This breaks NUMA in a guest.
Fix by setting up SRAT info for max_cpus in qemu-kvm.

Signed-off-by: default avatarVasilis Liaskovitis <vasilis.liaskovitis@profitbricks.com>
Signed-off-by: default avatarMarcelo Tosatti <mtosatti@redhat.com>
parent cce47516

hw/pc.c

+4 −4
Original line number Diff line number Diff line
@@ -624,9 +624,9 @@ static void *bochs_bios_init(void) 
     * of nodes, one word for each VCPU->node and one word for each node to

     * hold the amount of memory.

     */

    numa_fw_cfg = g_malloc0((1 + smp_cpus + nb_numa_nodes) * 8);

    numa_fw_cfg = g_malloc0((1 + max_cpus + nb_numa_nodes) * 8);

    numa_fw_cfg[0] = cpu_to_le64(nb_numa_nodes);

    for (i = 0; i < smp_cpus; i++) {

    for (i = 0; i < max_cpus; i++) {

        for (j = 0; j < nb_numa_nodes; j++) {

            if (node_cpumask[j] & (1 << i)) {

                numa_fw_cfg[i + 1] = cpu_to_le64(j);
@@ -635,10 +635,10 @@ static void *bochs_bios_init(void) 
        }

    }

    for (i = 0; i < nb_numa_nodes; i++) {

        numa_fw_cfg[smp_cpus + 1 + i] = cpu_to_le64(node_mem[i]);

        numa_fw_cfg[max_cpus + 1 + i] = cpu_to_le64(node_mem[i]);

    }

    fw_cfg_add_bytes(fw_cfg, FW_CFG_NUMA, (uint8_t *)numa_fw_cfg,

                     (1 + smp_cpus + nb_numa_nodes) * 8);

                     (1 + max_cpus + nb_numa_nodes) * 8);



    return fw_cfg;

}

vl.c

+1 −1
Original line number Diff line number Diff line
@@ -3305,7 +3305,7 @@ int main(int argc, char **argv, char **envp) 
         * real machines which also use this scheme.

         */

        if (i == nb_numa_nodes) {

            for (i = 0; i < smp_cpus; i++) {

            for (i = 0; i < max_cpus; i++) {

                node_cpumask[i % nb_numa_nodes] |= 1 << i;

            }

        }