|
| From: | Gavin Shan |
| Subject: | Re: [PATCH 2/3] hw/acpi/aml-build: Fix {socket, cluster, core} IDs in PPTT |
| Date: | Thu, 26 May 2022 22:40:05 +0800 |
| User-agent: | Mozilla/5.0 (X11; Linux x86_64; rv:68.0) Gecko/20100101 Thunderbird/68.2.0 |
Hi Igor, On 5/26/22 8:25 PM, Igor Mammedov wrote:
On Wed, 18 May 2022 17:21:40 +0800 Gavin Shan <gshan@redhat.com> wrote:The {socket, cluster, core} IDs detected from Linux guest aren't matching with what have been provided in PPTT. The flag used for 'ACPI Processor ID valid' is missed for {socket, cluster, core} nodes.To permit this flag set on no leaf nodes we have to have a corresponding containers built for them in DSDT so that 'ACPI Processor ID' could be matched with containers '_UID's. If we don not build such containers then setting this flag is not correct. And I don't recall QEMU building CPU hierarchy in DSDT.
It's true that we don't have containers in DSDT. In Linux implementation,
the corresponding IDs are fetched if 'ACPI Processor ID valid' is set in
PPTT node (entry), without checking DSDT table.
I don't know how the PPTT entry is linked to DSDT for _UID, after rechecking
ACPI specification. I was thinking 'Private Resources' fields are used for
the linking, but I should be wrong after checking PPTT tables on my host.
I'm not sure if you have idea how PPTT entry (node) is linked with one
specific device in DSDT table?
On my host, one of the cluster node resides at offset 10B0h and it's ID
has been marked as valid. The 'Private Resources' fields point to the
type-1 cache structures, which resides in PPTT table either. The cluster
ID ('0x109') isn't appearing in DSDT table.
[C9Ch 3228 1] Subtable Type : 01 [Cache Type]
[C9Dh 3229 1] Length : 18
[C9Eh 3230 2] Reserved : 0000
[CA0h 3232 4] Flags (decoded below) : 0000005F
Size valid : 1
Number of Sets valid : 1
Associativity valid : 1
Allocation Type valid : 1
Cache Type valid : 1
Write Policy valid : 0
Line Size valid : 1
Cache ID valid : 0
:
:
[CB4h 3252 1] Subtable Type : 01 [Cache Type]
[CB5h 3253 1] Length : 18
[CB6h 3254 2] Reserved : 0000
[CB8h 3256 4] Flags (decoded below) : 0000007F
Size valid : 1
Number of Sets valid : 1
Associativity valid : 1
Allocation Type valid : 1
Cache Type valid : 1
Write Policy valid : 1
Line Size valid : 1
Cache ID valid : 0
[CBCh 3260 4] Next Level of Cache : 00000CCC
:
:
[10B0h 4272 1] Subtable Type : 00 [Processor Hierarchy Node]
[10B1h 4273 1] Length : 1C
[10B2h 4274 2] Reserved : 0000
[10B4h 4276 4] Flags (decoded below) : 00000002
Physical package : 0
ACPI Processor ID valid : 1
Processor is a thread : 0
Node is a leaf : 0
Identical Implementation : 0
[10B8h 4280 4] Parent : 00000C6C
[10BCh 4284 4] ACPI Processor ID : 00000109
[10C0h 4288 4] Private Resource Number : 00000002
[10C4h 4292 4] Private Resource : 00000C9C
[10C8h 4296 4] Private Resource : 00000CB4
[gwshan@gshan tmp]$ grep -i 109 dsdt.dsl
[gwshan@gshan tmp]$ grep -i 265 dsdt.dsl
In this case, Linux guest takes the offset between the node and PPTT header as the corresponding IDs, as the following logs show.perhaps it's kernel which should be fixed to handle not set 'ACPI Processor ID valid' correctly.
This case is already handled by kernel. If 'ACPI Processor ID valid' flag is missed, the offset between PPTT node and header is taken as the ID. It's correct behaviour because 'ACPI Processor ID valid' flag isn't mandatory for PPTT nodes. The problem is that the offset isn't the desired ID, which is not understandable and mismatch with the ID (value) we feed to PPTT table in QEMU.
/home/gavin/sandbox/qemu.main/build/qemu-system-aarch64 \ -accel kvm -machine virt,gic-version=host -cpu host \ -smp 8,sockets=2,clusters=2,cores=2,threads=1 : # cd /sys/devices/system/cpu # for i in `seq 0 15`; do cat cpu$i/topology/physical_package_id; done 36 36 36 36 36 36 36 36 336 336 336 336 336 336 336 336 # for i in `seq 0 15`; do cat cpu$i/topology/cluster_id; done 56 56 56 56 196 196 196 196 356 356 356 356 496 496 496 496 # for i in `seq 0 15`; do cat cpu$i/topology/core_id; done 76 76 136 136 216 216 276 276 376 376 436 436 516 516 576 576 This fixes the issue by setting 'ACPI Processor ID valid' flag for {socket, cluster, core} nodes. With this applied, the IDs are exactly what have been provided in PPTT. # for i in `seq 0 15`; do cat cpu$i/topology/physical_package_id; done 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 # for i in `seq 0 15`; do cat cpu$i/topology/cluster_id; done 0 0 0 0 1 1 1 1 0 0 0 0 1 1 1 1 # for i in `seq 0 15`; do cat cpu$i/topology/core_id; done 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 Signed-off-by: Gavin Shan <gshan@redhat.com> --- hw/acpi/aml-build.c | 9 ++++++--- 1 file changed, 6 insertions(+), 3 deletions(-) diff --git a/hw/acpi/aml-build.c b/hw/acpi/aml-build.c index e6bfac95c7..89f191fd3b 100644 --- a/hw/acpi/aml-build.c +++ b/hw/acpi/aml-build.c @@ -2026,7 +2026,8 @@ void build_pptt(GArray *table_data, BIOSLinker *linker, MachineState *ms, core_id = -1; socket_offset = table_data->len - pptt_start; build_processor_hierarchy_node(table_data, - (1 << 0), /* Physical package */ + (1 << 0) | /* Physical package */ + (1 << 1), /* ACPI Processor ID valid */ 0, socket_id, NULL, 0); }@@ -2037,7 +2038,8 @@ void build_pptt(GArray *table_data, BIOSLinker *linker, MachineState *ms,core_id = -1; cluster_offset = table_data->len - pptt_start; build_processor_hierarchy_node(table_data, - (0 << 0), /* Not a physical package */ + (0 << 0) | /* Not a physical package */ + (1 << 1), /* ACPI Processor ID valid */ socket_offset, cluster_id, NULL, 0); } } else { @@ -2055,7 +2057,8 @@ void build_pptt(GArray *table_data, BIOSLinker *linker, MachineState *ms, core_id = cpus->cpus[n].props.core_id; core_offset = table_data->len - pptt_start; build_processor_hierarchy_node(table_data, - (0 << 0), /* Not a physical package */ + (0 << 0) | /* Not a physical package */ + (1 << 1), /* ACPI Processor ID valid */ cluster_offset, core_id, NULL, 0); }
Thanks, Gavin
| [Prev in Thread] | Current Thread | [Next in Thread] |