| Message ID | 1502847935-20483-1-git-send-email-douly.fnst@cn.fujitsu.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
Hi Igor, I tested this patch with following guests: 1. RHEL 6.5 with Linux 2.6.32 2. RHEL 7.0 with Linux 3.10.0 3. Fedora 23 with Linux 4.13.0-rc5 4. window 2003 service 5. window 7 6. window 10 Thanks, dou. At 08/16/2017 09:45 AM, Dou Liyang wrote: > Currently, Using the fisrt node without memory on the machine makes > QEMU unhappy. With this example command line: > ... \ > -m 1024M,slots=4,maxmem=32G \ > -numa node,nodeid=0 \ > -numa node,mem=1024M,nodeid=1 \ > -numa node,nodeid=2 \ > -numa node,nodeid=3 \ > Guest reports "No NUMA configuration found" and the NUMA topology is > wrong. > > This is because when QEMU builds ACPI SRAT, it regards node0 as the > default node to deal with the memory hole(640K-1M). this means the > node0 must have some memory(>1M), but, actually it can have no > memory. > > Fix this problem by replace the node0 with the first node which has > memory on it. Add a new function for each node. Also do some cleanup. > > Signed-off-by: Dou Liyang <douly.fnst@cn.fujitsu.com> > --- > V3 --> V2 > -Modify the title > V2 --> V1: > -Fix a coding style problem > Replace > for (node = 0; > node < pcms->numa_nodes && pcms->node_mem[node] == 0; > node++); > > with > for (node = 0; node < pcms->numa_nodes; node++) { > if (pcms->node_mem[node] != 0) { > break; > } > > hw/i386/acpi-build.c | 78 +++++++++++++++++++++++++++++++++------------------- > 1 file changed, 50 insertions(+), 28 deletions(-) > > diff --git a/hw/i386/acpi-build.c b/hw/i386/acpi-build.c > index 98dd424..f93d712 100644 > --- a/hw/i386/acpi-build.c > +++ b/hw/i386/acpi-build.c > @@ -2318,15 +2318,43 @@ build_tpm2(GArray *table_data, BIOSLinker *linker) > (void *)tpm2_ptr, "TPM2", sizeof(*tpm2_ptr), 4, NULL, NULL); > } > > +static uint64_t > +build_srat_node_entry(GArray *table_data, PCMachineState *pcms, > + int i, uint64_t mem_base, uint64_t mem_len) > +{ > + AcpiSratMemoryAffinity *numamem; > + uint64_t next_base; > + > + next_base = mem_base + mem_len; > + > + /* Cut out the ACPI_PCI hole */ > + if (mem_base <= pcms->below_4g_mem_size && > + next_base > pcms->below_4g_mem_size) { > + mem_len -= next_base - pcms->below_4g_mem_size; > + if (mem_len > 0) { > + numamem = acpi_data_push(table_data, sizeof *numamem); > + build_srat_memory(numamem, mem_base, mem_len, i, > + MEM_AFFINITY_ENABLED); > + } > + mem_base = 1ULL << 32; > + mem_len = next_base - pcms->below_4g_mem_size; > + next_base += (1ULL << 32) - pcms->below_4g_mem_size; > + } > + numamem = acpi_data_push(table_data, sizeof *numamem); > + build_srat_memory(numamem, mem_base, mem_len, i, > + MEM_AFFINITY_ENABLED); > + return next_base; > +} > + > static void > build_srat(GArray *table_data, BIOSLinker *linker, MachineState *machine) > { > AcpiSystemResourceAffinityTable *srat; > AcpiSratMemoryAffinity *numamem; > > - int i; > + int i, node; > int srat_start, numa_start, slots; > - uint64_t mem_len, mem_base, next_base; > + uint64_t mem_len, mem_base; > MachineClass *mc = MACHINE_GET_CLASS(machine); > const CPUArchIdList *apic_ids = mc->possible_cpu_arch_ids(machine); > PCMachineState *pcms = PC_MACHINE(machine); > @@ -2370,36 +2398,30 @@ build_srat(GArray *table_data, BIOSLinker *linker, MachineState *machine) > /* the memory map is a bit tricky, it contains at least one hole > * from 640k-1M and possibly another one from 3.5G-4G. > */ > - next_base = 0; > + > numa_start = table_data->len; > > - numamem = acpi_data_push(table_data, sizeof *numamem); > - build_srat_memory(numamem, 0, 640 * 1024, 0, MEM_AFFINITY_ENABLED); > - next_base = 1024 * 1024; > - for (i = 1; i < pcms->numa_nodes + 1; ++i) { > - mem_base = next_base; > - mem_len = pcms->node_mem[i - 1]; > - if (i == 1) { > - mem_len -= 1024 * 1024; > + /* get the first node which has memory and map the hole from 640K-1M */ > + for (node = 0; node < pcms->numa_nodes; node++) { > + if (pcms->node_mem[node] != 0) { > + break; > } > - next_base = mem_base + mem_len; > - > - /* Cut out the ACPI_PCI hole */ > - if (mem_base <= pcms->below_4g_mem_size && > - next_base > pcms->below_4g_mem_size) { > - mem_len -= next_base - pcms->below_4g_mem_size; > - if (mem_len > 0) { > - numamem = acpi_data_push(table_data, sizeof *numamem); > - build_srat_memory(numamem, mem_base, mem_len, i - 1, > - MEM_AFFINITY_ENABLED); > - } > - mem_base = 1ULL << 32; > - mem_len = next_base - pcms->below_4g_mem_size; > - next_base += (1ULL << 32) - pcms->below_4g_mem_size; > + } > + numamem = acpi_data_push(table_data, sizeof *numamem); > + build_srat_memory(numamem, 0, 640 * 1024, node, MEM_AFFINITY_ENABLED); > + > + /* map the rest of memory from 1M */ > + mem_base = 1024 * 1024; > + mem_len = pcms->node_mem[node] - mem_base; > + mem_base = build_srat_node_entry(table_data, pcms, node, > + mem_base, mem_len); > + > + for (i = 0; i < pcms->numa_nodes; i++) { > + if (i == node) { > + continue; > } > - numamem = acpi_data_push(table_data, sizeof *numamem); > - build_srat_memory(numamem, mem_base, mem_len, i - 1, > - MEM_AFFINITY_ENABLED); > + mem_base = build_srat_node_entry(table_data, pcms, i, > + mem_base, pcms->node_mem[i]); > } > slots = (table_data->len - numa_start) / sizeof *numamem; > for (; slots < pcms->numa_nodes + 2; slots++) { >
On Thu, Aug 17, 2017 at 01:58:40PM +0800, Dou Liyang wrote: > Hi Igor, > > I tested this patch with following guests: > > 1. RHEL 6.5 with Linux 2.6.32 > 2. RHEL 7.0 with Linux 3.10.0 > 3. Fedora 23 with Linux 4.13.0-rc5 > 4. window 2003 service > 5. window 7 > 6. window 10 What's the command-line(s) you have tested with each OS? Have you tested both the node0-with-RAM and node0-without-RAM cases? I would be interested in a demonstration that no bytes in the ACPI table are changed by this patch when there is some RAM configured in node 0. (Is that already covered by our existing ACPI test cases?) A new test case in bios-tables-test.c for the bug you are fixing would be nice to have. > > Thanks, > dou. > > At 08/16/2017 09:45 AM, Dou Liyang wrote: > > Currently, Using the fisrt node without memory on the machine makes > > QEMU unhappy. With this example command line: > > ... \ > > -m 1024M,slots=4,maxmem=32G \ > > -numa node,nodeid=0 \ > > -numa node,mem=1024M,nodeid=1 \ > > -numa node,nodeid=2 \ > > -numa node,nodeid=3 \ > > Guest reports "No NUMA configuration found" and the NUMA topology is > > wrong. > > > > This is because when QEMU builds ACPI SRAT, it regards node0 as the > > default node to deal with the memory hole(640K-1M). this means the > > node0 must have some memory(>1M), but, actually it can have no > > memory. > > > > Fix this problem by replace the node0 with the first node which has > > memory on it. Add a new function for each node. Also do some cleanup. > > > > Signed-off-by: Dou Liyang <douly.fnst@cn.fujitsu.com> > > --- > > V3 --> V2 > > -Modify the title > > V2 --> V1: > > -Fix a coding style problem > > Replace > > for (node = 0; > > node < pcms->numa_nodes && pcms->node_mem[node] == 0; > > node++); > > > > with > > for (node = 0; node < pcms->numa_nodes; node++) { > > if (pcms->node_mem[node] != 0) { > > break; > > } > > > > hw/i386/acpi-build.c | 78 +++++++++++++++++++++++++++++++++------------------- > > 1 file changed, 50 insertions(+), 28 deletions(-) > > > > diff --git a/hw/i386/acpi-build.c b/hw/i386/acpi-build.c > > index 98dd424..f93d712 100644 > > --- a/hw/i386/acpi-build.c > > +++ b/hw/i386/acpi-build.c > > @@ -2318,15 +2318,43 @@ build_tpm2(GArray *table_data, BIOSLinker *linker) > > (void *)tpm2_ptr, "TPM2", sizeof(*tpm2_ptr), 4, NULL, NULL); > > } > > > > +static uint64_t > > +build_srat_node_entry(GArray *table_data, PCMachineState *pcms, > > + int i, uint64_t mem_base, uint64_t mem_len) > > +{ > > + AcpiSratMemoryAffinity *numamem; > > + uint64_t next_base; > > + > > + next_base = mem_base + mem_len; > > + > > + /* Cut out the ACPI_PCI hole */ > > + if (mem_base <= pcms->below_4g_mem_size && > > + next_base > pcms->below_4g_mem_size) { > > + mem_len -= next_base - pcms->below_4g_mem_size; > > + if (mem_len > 0) { > > + numamem = acpi_data_push(table_data, sizeof *numamem); > > + build_srat_memory(numamem, mem_base, mem_len, i, > > + MEM_AFFINITY_ENABLED); > > + } > > + mem_base = 1ULL << 32; > > + mem_len = next_base - pcms->below_4g_mem_size; > > + next_base += (1ULL << 32) - pcms->below_4g_mem_size; > > + } > > + numamem = acpi_data_push(table_data, sizeof *numamem); > > + build_srat_memory(numamem, mem_base, mem_len, i, > > + MEM_AFFINITY_ENABLED); > > + return next_base; > > +} > > + > > static void > > build_srat(GArray *table_data, BIOSLinker *linker, MachineState *machine) > > { > > AcpiSystemResourceAffinityTable *srat; > > AcpiSratMemoryAffinity *numamem; > > > > - int i; > > + int i, node; > > int srat_start, numa_start, slots; > > - uint64_t mem_len, mem_base, next_base; > > + uint64_t mem_len, mem_base; > > MachineClass *mc = MACHINE_GET_CLASS(machine); > > const CPUArchIdList *apic_ids = mc->possible_cpu_arch_ids(machine); > > PCMachineState *pcms = PC_MACHINE(machine); > > @@ -2370,36 +2398,30 @@ build_srat(GArray *table_data, BIOSLinker *linker, MachineState *machine) > > /* the memory map is a bit tricky, it contains at least one hole > > * from 640k-1M and possibly another one from 3.5G-4G. > > */ > > - next_base = 0; > > + > > numa_start = table_data->len; > > > > - numamem = acpi_data_push(table_data, sizeof *numamem); > > - build_srat_memory(numamem, 0, 640 * 1024, 0, MEM_AFFINITY_ENABLED); > > - next_base = 1024 * 1024; > > - for (i = 1; i < pcms->numa_nodes + 1; ++i) { > > - mem_base = next_base; > > - mem_len = pcms->node_mem[i - 1]; > > - if (i == 1) { > > - mem_len -= 1024 * 1024; > > + /* get the first node which has memory and map the hole from 640K-1M */ > > + for (node = 0; node < pcms->numa_nodes; node++) { > > + if (pcms->node_mem[node] != 0) { > > + break; > > } > > - next_base = mem_base + mem_len; > > - > > - /* Cut out the ACPI_PCI hole */ > > - if (mem_base <= pcms->below_4g_mem_size && > > - next_base > pcms->below_4g_mem_size) { > > - mem_len -= next_base - pcms->below_4g_mem_size; > > - if (mem_len > 0) { > > - numamem = acpi_data_push(table_data, sizeof *numamem); > > - build_srat_memory(numamem, mem_base, mem_len, i - 1, > > - MEM_AFFINITY_ENABLED); > > - } > > - mem_base = 1ULL << 32; > > - mem_len = next_base - pcms->below_4g_mem_size; > > - next_base += (1ULL << 32) - pcms->below_4g_mem_size; > > + } > > + numamem = acpi_data_push(table_data, sizeof *numamem); > > + build_srat_memory(numamem, 0, 640 * 1024, node, MEM_AFFINITY_ENABLED); > > + > > + /* map the rest of memory from 1M */ > > + mem_base = 1024 * 1024; > > + mem_len = pcms->node_mem[node] - mem_base; > > + mem_base = build_srat_node_entry(table_data, pcms, node, > > + mem_base, mem_len); > > + > > + for (i = 0; i < pcms->numa_nodes; i++) { > > + if (i == node) { > > + continue; > > } > > - numamem = acpi_data_push(table_data, sizeof *numamem); > > - build_srat_memory(numamem, mem_base, mem_len, i - 1, > > - MEM_AFFINITY_ENABLED); > > + mem_base = build_srat_node_entry(table_data, pcms, i, > > + mem_base, pcms->node_mem[i]); > > } > > slots = (table_data->len - numa_start) / sizeof *numamem; > > for (; slots < pcms->numa_nodes + 2; slots++) { > > > >
On Wed, Aug 16, 2017 at 09:45:35AM +0800, Dou Liyang wrote: > Currently, Using the fisrt node without memory on the machine makes > QEMU unhappy. With this example command line: > ... \ > -m 1024M,slots=4,maxmem=32G \ > -numa node,nodeid=0 \ > -numa node,mem=1024M,nodeid=1 \ > -numa node,nodeid=2 \ > -numa node,nodeid=3 \ > Guest reports "No NUMA configuration found" and the NUMA topology is > wrong. > > This is because when QEMU builds ACPI SRAT, it regards node0 as the > default node to deal with the memory hole(640K-1M). this means the > node0 must have some memory(>1M), but, actually it can have no > memory. > > Fix this problem by replace the node0 with the first node which has > memory on it. Add a new function for each node. Also do some cleanup. > > Signed-off-by: Dou Liyang <douly.fnst@cn.fujitsu.com> This isn't a regression, is it? If it isn't, it's not a 2.10 candidate IMHO. > --- > V3 --> V2 > -Modify the title > V2 --> V1: > -Fix a coding style problem > Replace > for (node = 0; > node < pcms->numa_nodes && pcms->node_mem[node] == 0; > node++); > > with > for (node = 0; node < pcms->numa_nodes; node++) { > if (pcms->node_mem[node] != 0) { > break; > } > > hw/i386/acpi-build.c | 78 +++++++++++++++++++++++++++++++++------------------- > 1 file changed, 50 insertions(+), 28 deletions(-) > > diff --git a/hw/i386/acpi-build.c b/hw/i386/acpi-build.c > index 98dd424..f93d712 100644 > --- a/hw/i386/acpi-build.c > +++ b/hw/i386/acpi-build.c > @@ -2318,15 +2318,43 @@ build_tpm2(GArray *table_data, BIOSLinker *linker) > (void *)tpm2_ptr, "TPM2", sizeof(*tpm2_ptr), 4, NULL, NULL); > } > > +static uint64_t > +build_srat_node_entry(GArray *table_data, PCMachineState *pcms, > + int i, uint64_t mem_base, uint64_t mem_len) > +{ > + AcpiSratMemoryAffinity *numamem; > + uint64_t next_base; > + > + next_base = mem_base + mem_len; > + > + /* Cut out the ACPI_PCI hole */ > + if (mem_base <= pcms->below_4g_mem_size && > + next_base > pcms->below_4g_mem_size) { > + mem_len -= next_base - pcms->below_4g_mem_size; > + if (mem_len > 0) { > + numamem = acpi_data_push(table_data, sizeof *numamem); > + build_srat_memory(numamem, mem_base, mem_len, i, > + MEM_AFFINITY_ENABLED); > + } > + mem_base = 1ULL << 32; > + mem_len = next_base - pcms->below_4g_mem_size; > + next_base += (1ULL << 32) - pcms->below_4g_mem_size; > + } > + numamem = acpi_data_push(table_data, sizeof *numamem); > + build_srat_memory(numamem, mem_base, mem_len, i, > + MEM_AFFINITY_ENABLED); > + return next_base; > +} > + > static void > build_srat(GArray *table_data, BIOSLinker *linker, MachineState *machine) > { > AcpiSystemResourceAffinityTable *srat; > AcpiSratMemoryAffinity *numamem; > > - int i; > + int i, node; > int srat_start, numa_start, slots; > - uint64_t mem_len, mem_base, next_base; > + uint64_t mem_len, mem_base; > MachineClass *mc = MACHINE_GET_CLASS(machine); > const CPUArchIdList *apic_ids = mc->possible_cpu_arch_ids(machine); > PCMachineState *pcms = PC_MACHINE(machine); > @@ -2370,36 +2398,30 @@ build_srat(GArray *table_data, BIOSLinker *linker, MachineState *machine) > /* the memory map is a bit tricky, it contains at least one hole > * from 640k-1M and possibly another one from 3.5G-4G. > */ > - next_base = 0; > + > numa_start = table_data->len; > > - numamem = acpi_data_push(table_data, sizeof *numamem); > - build_srat_memory(numamem, 0, 640 * 1024, 0, MEM_AFFINITY_ENABLED); > - next_base = 1024 * 1024; > - for (i = 1; i < pcms->numa_nodes + 1; ++i) { > - mem_base = next_base; > - mem_len = pcms->node_mem[i - 1]; > - if (i == 1) { > - mem_len -= 1024 * 1024; > + /* get the first node which has memory and map the hole from 640K-1M */ > + for (node = 0; node < pcms->numa_nodes; node++) { > + if (pcms->node_mem[node] != 0) { > + break; > } > - next_base = mem_base + mem_len; > - > - /* Cut out the ACPI_PCI hole */ > - if (mem_base <= pcms->below_4g_mem_size && > - next_base > pcms->below_4g_mem_size) { > - mem_len -= next_base - pcms->below_4g_mem_size; > - if (mem_len > 0) { > - numamem = acpi_data_push(table_data, sizeof *numamem); > - build_srat_memory(numamem, mem_base, mem_len, i - 1, > - MEM_AFFINITY_ENABLED); > - } > - mem_base = 1ULL << 32; > - mem_len = next_base - pcms->below_4g_mem_size; > - next_base += (1ULL << 32) - pcms->below_4g_mem_size; > + } > + numamem = acpi_data_push(table_data, sizeof *numamem); > + build_srat_memory(numamem, 0, 640 * 1024, node, MEM_AFFINITY_ENABLED); > + > + /* map the rest of memory from 1M */ > + mem_base = 1024 * 1024; > + mem_len = pcms->node_mem[node] - mem_base; > + mem_base = build_srat_node_entry(table_data, pcms, node, > + mem_base, mem_len); > + > + for (i = 0; i < pcms->numa_nodes; i++) { > + if (i == node) { > + continue; > } > - numamem = acpi_data_push(table_data, sizeof *numamem); > - build_srat_memory(numamem, mem_base, mem_len, i - 1, > - MEM_AFFINITY_ENABLED); > + mem_base = build_srat_node_entry(table_data, pcms, i, > + mem_base, pcms->node_mem[i]); > } > slots = (table_data->len - numa_start) / sizeof *numamem; > for (; slots < pcms->numa_nodes + 2; slots++) { > -- > 2.5.5 > >
On Fri, Aug 18, 2017 at 09:35:00PM +0300, Michael S. Tsirkin wrote: > On Wed, Aug 16, 2017 at 09:45:35AM +0800, Dou Liyang wrote: > > Currently, Using the fisrt node without memory on the machine makes > > QEMU unhappy. With this example command line: > > ... \ > > -m 1024M,slots=4,maxmem=32G \ > > -numa node,nodeid=0 \ > > -numa node,mem=1024M,nodeid=1 \ > > -numa node,nodeid=2 \ > > -numa node,nodeid=3 \ > > Guest reports "No NUMA configuration found" and the NUMA topology is > > wrong. > > > > This is because when QEMU builds ACPI SRAT, it regards node0 as the > > default node to deal with the memory hole(640K-1M). this means the > > node0 must have some memory(>1M), but, actually it can have no > > memory. > > > > Fix this problem by replace the node0 with the first node which has > > memory on it. Add a new function for each node. Also do some cleanup. > > > > Signed-off-by: Dou Liyang <douly.fnst@cn.fujitsu.com> > > This isn't a regression, is it? If it isn't, it's not a 2.10 candidate > IMHO. As noted in another reply to v2, I agree and I'm treating it as a 2.11 candidate.
Hi Eduardo, At 08/19/2017 12:48 AM, Eduardo Habkost wrote: > On Thu, Aug 17, 2017 at 01:58:40PM +0800, Dou Liyang wrote: >> Hi Igor, >> >> I tested this patch with following guests: >> >> 1. RHEL 6.5 with Linux 2.6.32 >> 2. RHEL 7.0 with Linux 3.10.0 >> 3. Fedora 23 with Linux 4.13.0-rc5 >> 4. window 2003 service >> 5. window 7 >> 6. window 10 > > What's the command-line(s) you have tested with each OS? Have > I am sorry, one of the command-lines for the node0-without-RAM: ./x86_64-softmmu/qemu-system-x86_64 \ -hda /home/douly/image/rhel6.5.qcow2 \ -m 2G,slots=4,maxmem=32G \ -enable-kvm \ -smp 2,maxcpus=8,sockets=2,cores=2,threads=2\ -object memory-backend-ram,id=mem0,size=1024M \ -object memory-backend-ram,id=mem1,size=1024M \ -object memory-backend-ram,id=mem2,size=1024M \ -object memory-backend-ram,id=mem3,size=1024M \ -device pc-dimm,id=dimm0,memdev=mem0,slot=0,node=0 \ -device pc-dimm,id=dimm1,memdev=mem1,slot=1,node=1 \ -device pc-dimm,id=dimm2,memdev=mem2,slot=2,node=2 \ -device pc-dimm,id=dimm3,memdev=mem3,slot=3,node=3 \ -device qemu64-x86_64-cpu,id=cpu2,socket-id=0,core-id=1,thread-id=0 \ -device qemu64-x86_64-cpu,id=cpu3,socket-id=0,core-id=1,thread-id=1 \ -device qemu64-x86_64-cpu,id=cpu4,socket-id=1,core-id=0,thread-id=0 \ -device qemu64-x86_64-cpu,id=cpu5,socket-id=1,core-id=0,thread-id=1 \ -device qemu64-x86_64-cpu,id=cpu6,socket-id=1,core-id=1,thread-id=0 \ -device qemu64-x86_64-cpu,id=cpu7,socket-id=1,core-id=1,thread-id=1 \ -numa node,nodeid=0,cpus=0-1 \ -numa node,mem=2G,nodeid=1,cpus=2-3 \ -numa node,nodeid=2,cpus=4-5 \ -numa node,nodeid=3,cpus=6-7 \ -serial stdio \ -monitor telnet:127.0.0.1:4444,server,nowait \ # -kernel /home/douly/openSource/linux/arch/x86_64/boot/bzImage \ # -append "root=/dev/mapper/fedora_s3101490-root console=tty0 console=ttyS0 earlyprintk=ttyS0,115200n8 movable_node" \ # -initrd /home/douly/openSource/initramfs_image/4.13.0.img \ you tested both the node0-with-RAM and node0-without-RAM cases? > Yes, I tested three cases for each guests: Case 1) all node with RAM. -numa node,nodeid=0,cpus=0-1 \ -numa node,nodeid=1,cpus=2-3 \ -numa node,nodeid=2,cpus=4-5 \ -numa node,nodeid=3,cpus=6-7 \ Case 2) the node0-with-RAM: -numa node,mem=2G,nodeid=0,cpus=0-1 \ -numa node,nodeid=1,cpus=2-3 \ -numa node,nodeid=2,cpus=4-5 \ -numa node,nodeid=3,cpus=6-7 \ Case 3) the node0-without-RAM: -numa node,nodeid=0,cpus=0-1 \ -numa node,mem=2G,nodeid=1,cpus=2-3 \ -numa node,nodeid=2,cpus=4-5 \ -numa node,nodeid=3,cpus=6-7 \ > I would be interested in a demonstration that no bytes in the > ACPI table are changed by this patch when there is some RAM Yes, you are right, there is no change if node 0 has some RAM. This patch just works for the node0-without-RAM situation. > configured in node 0. (Is that already covered by our existing > ACPI test cases?) I am not sure it, could you tell me where are the SRAT test cases? Seems there is no test cases about SRAT in bios-tables-test.c. > > A new test case in bios-tables-test.c for the bug you are fixing > would be nice to have. > Ok, I will do it right now. Thanks, dou. >> >> Thanks, >> dou. >> >> At 08/16/2017 09:45 AM, Dou Liyang wrote: >>> Currently, Using the fisrt node without memory on the machine makes >>> QEMU unhappy. With this example command line: >>> ... \ >>> -m 1024M,slots=4,maxmem=32G \ >>> -numa node,nodeid=0 \ >>> -numa node,mem=1024M,nodeid=1 \ >>> -numa node,nodeid=2 \ >>> -numa node,nodeid=3 \ >>> Guest reports "No NUMA configuration found" and the NUMA topology is >>> wrong. >>> >>> This is because when QEMU builds ACPI SRAT, it regards node0 as the >>> default node to deal with the memory hole(640K-1M). this means the >>> node0 must have some memory(>1M), but, actually it can have no >>> memory. >>> >>> Fix this problem by replace the node0 with the first node which has >>> memory on it. Add a new function for each node. Also do some cleanup. >>> >>> Signed-off-by: Dou Liyang <douly.fnst@cn.fujitsu.com> >>> --- >>> V3 --> V2 >>> -Modify the title >>> V2 --> V1: >>> -Fix a coding style problem >>> Replace >>> for (node = 0; >>> node < pcms->numa_nodes && pcms->node_mem[node] == 0; >>> node++); >>> >>> with >>> for (node = 0; node < pcms->numa_nodes; node++) { >>> if (pcms->node_mem[node] != 0) { >>> break; >>> } >>> >>> hw/i386/acpi-build.c | 78 +++++++++++++++++++++++++++++++++------------------- >>> 1 file changed, 50 insertions(+), 28 deletions(-) >>> >>> diff --git a/hw/i386/acpi-build.c b/hw/i386/acpi-build.c >>> index 98dd424..f93d712 100644 >>> --- a/hw/i386/acpi-build.c >>> +++ b/hw/i386/acpi-build.c >>> @@ -2318,15 +2318,43 @@ build_tpm2(GArray *table_data, BIOSLinker *linker) >>> (void *)tpm2_ptr, "TPM2", sizeof(*tpm2_ptr), 4, NULL, NULL); >>> } >>> >>> +static uint64_t >>> +build_srat_node_entry(GArray *table_data, PCMachineState *pcms, >>> + int i, uint64_t mem_base, uint64_t mem_len) >>> +{ >>> + AcpiSratMemoryAffinity *numamem; >>> + uint64_t next_base; >>> + >>> + next_base = mem_base + mem_len; >>> + >>> + /* Cut out the ACPI_PCI hole */ >>> + if (mem_base <= pcms->below_4g_mem_size && >>> + next_base > pcms->below_4g_mem_size) { >>> + mem_len -= next_base - pcms->below_4g_mem_size; >>> + if (mem_len > 0) { >>> + numamem = acpi_data_push(table_data, sizeof *numamem); >>> + build_srat_memory(numamem, mem_base, mem_len, i, >>> + MEM_AFFINITY_ENABLED); >>> + } >>> + mem_base = 1ULL << 32; >>> + mem_len = next_base - pcms->below_4g_mem_size; >>> + next_base += (1ULL << 32) - pcms->below_4g_mem_size; >>> + } >>> + numamem = acpi_data_push(table_data, sizeof *numamem); >>> + build_srat_memory(numamem, mem_base, mem_len, i, >>> + MEM_AFFINITY_ENABLED); >>> + return next_base; >>> +} >>> + >>> static void >>> build_srat(GArray *table_data, BIOSLinker *linker, MachineState *machine) >>> { >>> AcpiSystemResourceAffinityTable *srat; >>> AcpiSratMemoryAffinity *numamem; >>> >>> - int i; >>> + int i, node; >>> int srat_start, numa_start, slots; >>> - uint64_t mem_len, mem_base, next_base; >>> + uint64_t mem_len, mem_base; >>> MachineClass *mc = MACHINE_GET_CLASS(machine); >>> const CPUArchIdList *apic_ids = mc->possible_cpu_arch_ids(machine); >>> PCMachineState *pcms = PC_MACHINE(machine); >>> @@ -2370,36 +2398,30 @@ build_srat(GArray *table_data, BIOSLinker *linker, MachineState *machine) >>> /* the memory map is a bit tricky, it contains at least one hole >>> * from 640k-1M and possibly another one from 3.5G-4G. >>> */ >>> - next_base = 0; >>> + >>> numa_start = table_data->len; >>> >>> - numamem = acpi_data_push(table_data, sizeof *numamem); >>> - build_srat_memory(numamem, 0, 640 * 1024, 0, MEM_AFFINITY_ENABLED); >>> - next_base = 1024 * 1024; >>> - for (i = 1; i < pcms->numa_nodes + 1; ++i) { >>> - mem_base = next_base; >>> - mem_len = pcms->node_mem[i - 1]; >>> - if (i == 1) { >>> - mem_len -= 1024 * 1024; >>> + /* get the first node which has memory and map the hole from 640K-1M */ >>> + for (node = 0; node < pcms->numa_nodes; node++) { >>> + if (pcms->node_mem[node] != 0) { >>> + break; >>> } >>> - next_base = mem_base + mem_len; >>> - >>> - /* Cut out the ACPI_PCI hole */ >>> - if (mem_base <= pcms->below_4g_mem_size && >>> - next_base > pcms->below_4g_mem_size) { >>> - mem_len -= next_base - pcms->below_4g_mem_size; >>> - if (mem_len > 0) { >>> - numamem = acpi_data_push(table_data, sizeof *numamem); >>> - build_srat_memory(numamem, mem_base, mem_len, i - 1, >>> - MEM_AFFINITY_ENABLED); >>> - } >>> - mem_base = 1ULL << 32; >>> - mem_len = next_base - pcms->below_4g_mem_size; >>> - next_base += (1ULL << 32) - pcms->below_4g_mem_size; >>> + } >>> + numamem = acpi_data_push(table_data, sizeof *numamem); >>> + build_srat_memory(numamem, 0, 640 * 1024, node, MEM_AFFINITY_ENABLED); >>> + >>> + /* map the rest of memory from 1M */ >>> + mem_base = 1024 * 1024; >>> + mem_len = pcms->node_mem[node] - mem_base; >>> + mem_base = build_srat_node_entry(table_data, pcms, node, >>> + mem_base, mem_len); >>> + >>> + for (i = 0; i < pcms->numa_nodes; i++) { >>> + if (i == node) { >>> + continue; >>> } >>> - numamem = acpi_data_push(table_data, sizeof *numamem); >>> - build_srat_memory(numamem, mem_base, mem_len, i - 1, >>> - MEM_AFFINITY_ENABLED); >>> + mem_base = build_srat_node_entry(table_data, pcms, i, >>> + mem_base, pcms->node_mem[i]); >>> } >>> slots = (table_data->len - numa_start) / sizeof *numamem; >>> for (; slots < pcms->numa_nodes + 2; slots++) { >>> >> >> >
On Sun, 20 Aug 2017 19:52:11 +0800 Dou Liyang <douly.fnst@cn.fujitsu.com> wrote: > Hi Eduardo, > > At 08/19/2017 12:48 AM, Eduardo Habkost wrote: > > On Thu, Aug 17, 2017 at 01:58:40PM +0800, Dou Liyang wrote: > >> Hi Igor, > >> > >> I tested this patch with following guests: > >> > >> 1. RHEL 6.5 with Linux 2.6.32 > >> 2. RHEL 7.0 with Linux 3.10.0 > >> 3. Fedora 23 with Linux 4.13.0-rc5 > >> 4. window 2003 service > >> 5. window 7 > >> 6. window 10 > > > > What's the command-line(s) you have tested with each OS? Have > > > > I am sorry, one of the command-lines for the node0-without-RAM: > > ./x86_64-softmmu/qemu-system-x86_64 \ > -hda /home/douly/image/rhel6.5.qcow2 \ > -m 2G,slots=4,maxmem=32G \ > -enable-kvm \ > -smp 2,maxcpus=8,sockets=2,cores=2,threads=2\ > -object memory-backend-ram,id=mem0,size=1024M \ > -object memory-backend-ram,id=mem1,size=1024M \ > -object memory-backend-ram,id=mem2,size=1024M \ > -object memory-backend-ram,id=mem3,size=1024M \ > -device pc-dimm,id=dimm0,memdev=mem0,slot=0,node=0 \ > -device pc-dimm,id=dimm1,memdev=mem1,slot=1,node=1 \ > -device pc-dimm,id=dimm2,memdev=mem2,slot=2,node=2 \ > -device pc-dimm,id=dimm3,memdev=mem3,slot=3,node=3 \ > -device qemu64-x86_64-cpu,id=cpu2,socket-id=0,core-id=1,thread-id=0 \ > -device qemu64-x86_64-cpu,id=cpu3,socket-id=0,core-id=1,thread-id=1 \ > -device qemu64-x86_64-cpu,id=cpu4,socket-id=1,core-id=0,thread-id=0 \ > -device qemu64-x86_64-cpu,id=cpu5,socket-id=1,core-id=0,thread-id=1 \ > -device qemu64-x86_64-cpu,id=cpu6,socket-id=1,core-id=1,thread-id=0 \ > -device qemu64-x86_64-cpu,id=cpu7,socket-id=1,core-id=1,thread-id=1 \ > -numa node,nodeid=0,cpus=0-1 \ > -numa node,mem=2G,nodeid=1,cpus=2-3 \ > -numa node,nodeid=2,cpus=4-5 \ > -numa node,nodeid=3,cpus=6-7 \ > -serial stdio \ > -monitor telnet:127.0.0.1:4444,server,nowait \ > # -kernel /home/douly/openSource/linux/arch/x86_64/boot/bzImage \ > # -append "root=/dev/mapper/fedora_s3101490-root console=tty0 > console=ttyS0 earlyprintk=ttyS0,115200n8 movable_node" \ > # -initrd /home/douly/openSource/initramfs_image/4.13.0.img \ > you tested both the node0-with-RAM and node0-without-RAM cases? > > > > Yes, I tested three cases for each guests: > > Case 1) all node with RAM. > > -numa node,nodeid=0,cpus=0-1 \ > -numa node,nodeid=1,cpus=2-3 \ > -numa node,nodeid=2,cpus=4-5 \ > -numa node,nodeid=3,cpus=6-7 \ > > Case 2) the node0-with-RAM: > > -numa node,mem=2G,nodeid=0,cpus=0-1 \ > -numa node,nodeid=1,cpus=2-3 \ > -numa node,nodeid=2,cpus=4-5 \ > -numa node,nodeid=3,cpus=6-7 \ > > Case 3) the node0-without-RAM: > > -numa node,nodeid=0,cpus=0-1 \ > -numa node,mem=2G,nodeid=1,cpus=2-3 \ > -numa node,nodeid=2,cpus=4-5 \ > -numa node,nodeid=3,cpus=6-7 \ > > > I would be interested in a demonstration that no bytes in the > > ACPI table are changed by this patch when there is some RAM > > Yes, you are right, there is no change if node 0 has some RAM. > This patch just works for the node0-without-RAM situation. > > > configured in node 0. (Is that already covered by our existing > > ACPI test cases?) > > I am not sure it, could you tell me where are the SRAT test cases? > Seems there is no test cases about SRAT in bios-tables-test.c. > > > > > A new test case in bios-tables-test.c for the bug you are fixing > > would be nice to have. > > > > Ok, I will do it right now. see commit 6b9c1dd2c for example of adding test case, currently test has base SRAT tables that is reused by all tests. When you create testcase that produces different SRAT table, 'make check' will print warnings that ables do not match you can add "V=1" env. var. 'make V=1 check' to get detailed diff of what's changed. also see how 'test_data.variant' is handled to get idea how to use it to differentiate specific test case vs shared ACPI tables baseline blobs. > > Thanks, > dou. > >> > > > >> Thanks, > >> dou. > >> > >> At 08/16/2017 09:45 AM, Dou Liyang wrote: > >>> Currently, Using the fisrt node without memory on the machine makes > >>> QEMU unhappy. With this example command line: > >>> ... \ > >>> -m 1024M,slots=4,maxmem=32G \ > >>> -numa node,nodeid=0 \ > >>> -numa node,mem=1024M,nodeid=1 \ > >>> -numa node,nodeid=2 \ > >>> -numa node,nodeid=3 \ > >>> Guest reports "No NUMA configuration found" and the NUMA topology is > >>> wrong. > >>> > >>> This is because when QEMU builds ACPI SRAT, it regards node0 as the > >>> default node to deal with the memory hole(640K-1M). this means the > >>> node0 must have some memory(>1M), but, actually it can have no > >>> memory. > >>> > >>> Fix this problem by replace the node0 with the first node which has > >>> memory on it. Add a new function for each node. Also do some cleanup. > >>> > >>> Signed-off-by: Dou Liyang <douly.fnst@cn.fujitsu.com> > >>> --- > >>> V3 --> V2 > >>> -Modify the title > >>> V2 --> V1: > >>> -Fix a coding style problem > >>> Replace > >>> for (node = 0; > >>> node < pcms->numa_nodes && pcms->node_mem[node] == 0; > >>> node++); > >>> > >>> with > >>> for (node = 0; node < pcms->numa_nodes; node++) { > >>> if (pcms->node_mem[node] != 0) { > >>> break; > >>> } > >>> > >>> hw/i386/acpi-build.c | 78 +++++++++++++++++++++++++++++++++------------------- > >>> 1 file changed, 50 insertions(+), 28 deletions(-) > >>> > >>> diff --git a/hw/i386/acpi-build.c b/hw/i386/acpi-build.c > >>> index 98dd424..f93d712 100644 > >>> --- a/hw/i386/acpi-build.c > >>> +++ b/hw/i386/acpi-build.c > >>> @@ -2318,15 +2318,43 @@ build_tpm2(GArray *table_data, BIOSLinker *linker) > >>> (void *)tpm2_ptr, "TPM2", sizeof(*tpm2_ptr), 4, NULL, NULL); > >>> } > >>> > >>> +static uint64_t > >>> +build_srat_node_entry(GArray *table_data, PCMachineState *pcms, > >>> + int i, uint64_t mem_base, uint64_t mem_len) > >>> +{ > >>> + AcpiSratMemoryAffinity *numamem; > >>> + uint64_t next_base; > >>> + > >>> + next_base = mem_base + mem_len; > >>> + > >>> + /* Cut out the ACPI_PCI hole */ > >>> + if (mem_base <= pcms->below_4g_mem_size && > >>> + next_base > pcms->below_4g_mem_size) { > >>> + mem_len -= next_base - pcms->below_4g_mem_size; > >>> + if (mem_len > 0) { > >>> + numamem = acpi_data_push(table_data, sizeof *numamem); > >>> + build_srat_memory(numamem, mem_base, mem_len, i, > >>> + MEM_AFFINITY_ENABLED); > >>> + } > >>> + mem_base = 1ULL << 32; > >>> + mem_len = next_base - pcms->below_4g_mem_size; > >>> + next_base += (1ULL << 32) - pcms->below_4g_mem_size; > >>> + } > >>> + numamem = acpi_data_push(table_data, sizeof *numamem); > >>> + build_srat_memory(numamem, mem_base, mem_len, i, > >>> + MEM_AFFINITY_ENABLED); > >>> + return next_base; > >>> +} > >>> + > >>> static void > >>> build_srat(GArray *table_data, BIOSLinker *linker, MachineState *machine) > >>> { > >>> AcpiSystemResourceAffinityTable *srat; > >>> AcpiSratMemoryAffinity *numamem; > >>> > >>> - int i; > >>> + int i, node; > >>> int srat_start, numa_start, slots; > >>> - uint64_t mem_len, mem_base, next_base; > >>> + uint64_t mem_len, mem_base; > >>> MachineClass *mc = MACHINE_GET_CLASS(machine); > >>> const CPUArchIdList *apic_ids = mc->possible_cpu_arch_ids(machine); > >>> PCMachineState *pcms = PC_MACHINE(machine); > >>> @@ -2370,36 +2398,30 @@ build_srat(GArray *table_data, BIOSLinker *linker, MachineState *machine) > >>> /* the memory map is a bit tricky, it contains at least one hole > >>> * from 640k-1M and possibly another one from 3.5G-4G. > >>> */ > >>> - next_base = 0; > >>> + > >>> numa_start = table_data->len; > >>> > >>> - numamem = acpi_data_push(table_data, sizeof *numamem); > >>> - build_srat_memory(numamem, 0, 640 * 1024, 0, MEM_AFFINITY_ENABLED); > >>> - next_base = 1024 * 1024; > >>> - for (i = 1; i < pcms->numa_nodes + 1; ++i) { > >>> - mem_base = next_base; > >>> - mem_len = pcms->node_mem[i - 1]; > >>> - if (i == 1) { > >>> - mem_len -= 1024 * 1024; > >>> + /* get the first node which has memory and map the hole from 640K-1M */ > >>> + for (node = 0; node < pcms->numa_nodes; node++) { > >>> + if (pcms->node_mem[node] != 0) { > >>> + break; > >>> } > >>> - next_base = mem_base + mem_len; > >>> - > >>> - /* Cut out the ACPI_PCI hole */ > >>> - if (mem_base <= pcms->below_4g_mem_size && > >>> - next_base > pcms->below_4g_mem_size) { > >>> - mem_len -= next_base - pcms->below_4g_mem_size; > >>> - if (mem_len > 0) { > >>> - numamem = acpi_data_push(table_data, sizeof *numamem); > >>> - build_srat_memory(numamem, mem_base, mem_len, i - 1, > >>> - MEM_AFFINITY_ENABLED); > >>> - } > >>> - mem_base = 1ULL << 32; > >>> - mem_len = next_base - pcms->below_4g_mem_size; > >>> - next_base += (1ULL << 32) - pcms->below_4g_mem_size; > >>> + } > >>> + numamem = acpi_data_push(table_data, sizeof *numamem); > >>> + build_srat_memory(numamem, 0, 640 * 1024, node, MEM_AFFINITY_ENABLED); > >>> + > >>> + /* map the rest of memory from 1M */ > >>> + mem_base = 1024 * 1024; > >>> + mem_len = pcms->node_mem[node] - mem_base; > >>> + mem_base = build_srat_node_entry(table_data, pcms, node, > >>> + mem_base, mem_len); > >>> + > >>> + for (i = 0; i < pcms->numa_nodes; i++) { > >>> + if (i == node) { > >>> + continue; > >>> } > >>> - numamem = acpi_data_push(table_data, sizeof *numamem); > >>> - build_srat_memory(numamem, mem_base, mem_len, i - 1, > >>> - MEM_AFFINITY_ENABLED); > >>> + mem_base = build_srat_node_entry(table_data, pcms, i, > >>> + mem_base, pcms->node_mem[i]); > >>> } > >>> slots = (table_data->len - numa_start) / sizeof *numamem; > >>> for (; slots < pcms->numa_nodes + 2; slots++) { > >>> > >> > >> > > > >
diff --git a/hw/i386/acpi-build.c b/hw/i386/acpi-build.c index 98dd424..f93d712 100644 --- a/hw/i386/acpi-build.c +++ b/hw/i386/acpi-build.c @@ -2318,15 +2318,43 @@ build_tpm2(GArray *table_data, BIOSLinker *linker) (void *)tpm2_ptr, "TPM2", sizeof(*tpm2_ptr), 4, NULL, NULL); } +static uint64_t +build_srat_node_entry(GArray *table_data, PCMachineState *pcms, + int i, uint64_t mem_base, uint64_t mem_len) +{ + AcpiSratMemoryAffinity *numamem; + uint64_t next_base; + + next_base = mem_base + mem_len; + + /* Cut out the ACPI_PCI hole */ + if (mem_base <= pcms->below_4g_mem_size && + next_base > pcms->below_4g_mem_size) { + mem_len -= next_base - pcms->below_4g_mem_size; + if (mem_len > 0) { + numamem = acpi_data_push(table_data, sizeof *numamem); + build_srat_memory(numamem, mem_base, mem_len, i, + MEM_AFFINITY_ENABLED); + } + mem_base = 1ULL << 32; + mem_len = next_base - pcms->below_4g_mem_size; + next_base += (1ULL << 32) - pcms->below_4g_mem_size; + } + numamem = acpi_data_push(table_data, sizeof *numamem); + build_srat_memory(numamem, mem_base, mem_len, i, + MEM_AFFINITY_ENABLED); + return next_base; +} + static void build_srat(GArray *table_data, BIOSLinker *linker, MachineState *machine) { AcpiSystemResourceAffinityTable *srat; AcpiSratMemoryAffinity *numamem; - int i; + int i, node; int srat_start, numa_start, slots; - uint64_t mem_len, mem_base, next_base; + uint64_t mem_len, mem_base; MachineClass *mc = MACHINE_GET_CLASS(machine); const CPUArchIdList *apic_ids = mc->possible_cpu_arch_ids(machine); PCMachineState *pcms = PC_MACHINE(machine); @@ -2370,36 +2398,30 @@ build_srat(GArray *table_data, BIOSLinker *linker, MachineState *machine) /* the memory map is a bit tricky, it contains at least one hole * from 640k-1M and possibly another one from 3.5G-4G. */ - next_base = 0; + numa_start = table_data->len; - numamem = acpi_data_push(table_data, sizeof *numamem); - build_srat_memory(numamem, 0, 640 * 1024, 0, MEM_AFFINITY_ENABLED); - next_base = 1024 * 1024; - for (i = 1; i < pcms->numa_nodes + 1; ++i) { - mem_base = next_base; - mem_len = pcms->node_mem[i - 1]; - if (i == 1) { - mem_len -= 1024 * 1024; + /* get the first node which has memory and map the hole from 640K-1M */ + for (node = 0; node < pcms->numa_nodes; node++) { + if (pcms->node_mem[node] != 0) { + break; } - next_base = mem_base + mem_len; - - /* Cut out the ACPI_PCI hole */ - if (mem_base <= pcms->below_4g_mem_size && - next_base > pcms->below_4g_mem_size) { - mem_len -= next_base - pcms->below_4g_mem_size; - if (mem_len > 0) { - numamem = acpi_data_push(table_data, sizeof *numamem); - build_srat_memory(numamem, mem_base, mem_len, i - 1, - MEM_AFFINITY_ENABLED); - } - mem_base = 1ULL << 32; - mem_len = next_base - pcms->below_4g_mem_size; - next_base += (1ULL << 32) - pcms->below_4g_mem_size; + } + numamem = acpi_data_push(table_data, sizeof *numamem); + build_srat_memory(numamem, 0, 640 * 1024, node, MEM_AFFINITY_ENABLED); + + /* map the rest of memory from 1M */ + mem_base = 1024 * 1024; + mem_len = pcms->node_mem[node] - mem_base; + mem_base = build_srat_node_entry(table_data, pcms, node, + mem_base, mem_len); + + for (i = 0; i < pcms->numa_nodes; i++) { + if (i == node) { + continue; } - numamem = acpi_data_push(table_data, sizeof *numamem); - build_srat_memory(numamem, mem_base, mem_len, i - 1, - MEM_AFFINITY_ENABLED); + mem_base = build_srat_node_entry(table_data, pcms, i, + mem_base, pcms->node_mem[i]); } slots = (table_data->len - numa_start) / sizeof *numamem; for (; slots < pcms->numa_nodes + 2; slots++) {
Currently, Using the fisrt node without memory on the machine makes QEMU unhappy. With this example command line: ... \ -m 1024M,slots=4,maxmem=32G \ -numa node,nodeid=0 \ -numa node,mem=1024M,nodeid=1 \ -numa node,nodeid=2 \ -numa node,nodeid=3 \ Guest reports "No NUMA configuration found" and the NUMA topology is wrong. This is because when QEMU builds ACPI SRAT, it regards node0 as the default node to deal with the memory hole(640K-1M). this means the node0 must have some memory(>1M), but, actually it can have no memory. Fix this problem by replace the node0 with the first node which has memory on it. Add a new function for each node. Also do some cleanup. Signed-off-by: Dou Liyang <douly.fnst@cn.fujitsu.com> --- V3 --> V2 -Modify the title V2 --> V1: -Fix a coding style problem Replace for (node = 0; node < pcms->numa_nodes && pcms->node_mem[node] == 0; node++); with for (node = 0; node < pcms->numa_nodes; node++) { if (pcms->node_mem[node] != 0) { break; } hw/i386/acpi-build.c | 78 +++++++++++++++++++++++++++++++++------------------- 1 file changed, 50 insertions(+), 28 deletions(-)