Based-on: [PATCH v5 00/13] acpi: NUMA nodes for CXL HB as GP + complex NUMA test Based-on: Message-ID: 20240712110837.1439736-1-jonathan.came...@huawei.com
The Generic Ports support added the ability to describe the bandwidth and Latency within a host to a CXL host bridge. To be able to test the of the discovery path used by Linux [1] we also need to be able to create bottlenecks at difference places in the topology. There are two parts to this * CXL link characteristics as described by PCI Express Capability Link status etc. * Bandwidth and latency across CXL Switches (via CDAT data from the switch USP) * Bandwidth and latency from the CXL type 3 device port to the actual memory (Via CDAT data from the EP). Currently we have fixed values for the CXL CDAT tables, and to test this I recommend changing those as per the patch at the end of this cover letter (so they aren't always the bottleneck). Making those configurable will be handled in a future patch set. RFC question: - I could enable this for all PCIe device (including ports). Does that makes sense, or is it better to limit this to my cases. It is quite easy to build broken setups (downstream device reports faster link than the port etc) because QEMU 'link' training' is simplistic. I'm not sure it is worth making it more clever. Test case snippets I promised Dave Jiang. These rely on the tweaks to CDAT at end of this cover letter. Case 1: Direct connected pair of type 3 nodes on same HB - type 3 limited ========================================================================== -object memory-backend-file,id=cxl-mem1,share=on,mem-path=/tmp/t3_cxl1.raw,size=1G,align=256M \ -object memory-backend-file,id=cxl-mem2,share=on,mem-path=/tmp/t3_cxl2.raw,size=1G,align=256M \ -object memory-backend-file,id=cxl-lsa1,share=on,mem-path=/tmp/t3_lsa1.raw,size=1M,align=1M \ -object memory-backend-file,id=cxl-mem3,share=on,mem-path=/tmp/t3_cxl3.raw,size=1G,align=256M \ -object memory-backend-file,id=cxl-mem4,share=on,mem-path=/tmp/t3_cxl4.raw,size=1G,align=256M \ -object memory-backend-file,id=cxl-lsa2,share=on,mem-path=/tmp/t3_lsa2.raw,size=1M,align=1M \ -device pxb-cxl,bus_nr=12,bus=pcie.0,id=cxl.1,hdm_for_passthrough=true,numa_node=0 \ -device cxl-rp,port=0,bus=cxl.1,id=cxl_rp_port0,chassis=0,slot=2 \ -device cxl-rp,port=1,bus=cxl.1,id=cxl_rp_port1,chassis=0,slot=3 \ -device cxl-type3,bus=cxl_rp_port0,volatile-memdev=cxl-mem1,id=cxl-pmem1,lsa=cxl-lsa1,sn=3,x-speed=32,x-width=16 \ -device cxl-type3,bus=cxl_rp_port1,volatile-memdev=cxl-mem3,id=cxl-pmem2,lsa=cxl-lsa2,sn=5,x-speed=32,x-width=16 \ -machine cxl-fmw.0.targets.0=cxl.1,cxl-fmw.0.size=8G,cxl-fmw.0.interleave-granularity=1k \ -numa node,nodeid=0,cpus=0-3,memdev=mem0 \ -numa node,nodeid=1 \ -object acpi-generic-initiator,id=bob2,pci-dev=bob,node=1 \ -numa node,nodeid=2 \ -object acpi-generic-port,id=bob11,pci-bus=cxl.1,node=2 \ -numa dist,src=0,dst=0,val=10 -numa dist,src=0,dst=1,val=21 -numa dist,src=0,dst=2,val=21 \ -numa dist,src=1,dst=0,val=21 -numa dist,src=1,dst=1,val=10 -numa dist,src=1,dst=2,val=21 \ -numa dist,src=2,dst=0,val=21 -numa dist,src=2,dst=1,val=21 -numa dist,src=2,dst=2,val=10 \ -numa hmat-lb,initiator=0,target=0,hierarchy=memory,data-type=access-latency,latency=10 \ -numa hmat-lb,initiator=0,target=0,hierarchy=memory,data-type=access-bandwidth,bandwidth=3G \ -numa hmat-lb,initiator=0,target=2,hierarchy=memory,data-type=access-latency,latency=100 \ -numa hmat-lb,initiator=0,target=2,hierarchy=memory,data-type=access-bandwidth,bandwidth=100G \ -numa hmat-lb,initiator=1,target=0,hierarchy=memory,data-type=access-latency,latency=500 \ -numa hmat-lb,initiator=1,target=0,hierarchy=memory,data-type=access-bandwidth,bandwidth=100M \ -numa hmat-lb,initiator=1,target=2,hierarchy=memory,data-type=access-latency,latency=50 \ -numa hmat-lb,initiator=1,target=2,hierarchy=memory,data-type=access-bandwidth,bandwidth=100G Set x-width to 2 and it will be link limited. -device cxl-type3,bus=cxl_rp_port0,volatile-memdev=cxl-mem1,id=cxl-pmem1,lsa=cxl-lsa1,sn=3,x-speed=32,x-width=2 \ -device cxl-type3,bus=cxl_rp_port1,volatile-memdev=cxl-mem3,id=cxl-pmem2,lsa=cxl-lsa2,sn=5,x-speed=32,x-width=2 \ Case 2: Switch connected type 3. ================================ -object memory-backend-file,id=cxl-mem1,share=on,mem-path=/tmp/t3_cxl1.raw,size=1G,align=256M \ -object memory-backend-file,id=cxl-mem3,share=on,mem-path=/tmp/t3_cxl3.raw,size=1G,align=256M \ -device pxb-cxl,bus_nr=12,bus=pcie.0,id=cxl.1,hdm_for_passthrough=true \ -device cxl-rp,port=0,bus=cxl.1,id=cxl_rp_port0,chassis=0,slot=2,mem-reserve=128M,hotplug=true \ -device cxl-upstream,port=33,bus=cxl_rp_port0,id=us0,multifunction=on,addr=0.0,sn=33,x-speed=64,x-width=16 \ -device cxl-downstream,port=0,bus=us0,id=swport0,chassis=0,slot=4,hotplug=true \ -device cxl-downstream,port=1,bus=us0,id=swport1,chassis=0,slot=5,hotplug=true \ -device cxl-type3,bus=swport0,volatile-memdev=cxl-mem1,id=cxl-pmem1,sn=3,x-speed=32,x-width=16 \ -device cxl-type3,bus=swport1,volatile-memdev=cxl-mem3,id=cxl-pmem2,sn=5,x-speed=32,x-width=16 \ -machine cxl-fmw.0.targets.0=cxl.1,cxl-fmw.0.size=4G,cxl-fmw.0.interleave-granularity=1k \ -numa node,nodeid=0,cpus=0-3,memdev=mem0 \ -numa node,nodeid=1 \ -object acpi-generic-initiator,id=bob2,pci-dev=bob,node=1 \ -numa node,nodeid=2 \ -object acpi-generic-port,id=bob11,pci-bus=cxl.1,node=2 \ -numa dist,src=0,dst=0,val=10 -numa dist,src=0,dst=1,val=21 -numa dist,src=0,dst=2,val=21 \ -numa dist,src=1,dst=0,val=21 -numa dist,src=1,dst=1,val=10 -numa dist,src=1,dst=2,val=21 \ -numa dist,src=2,dst=0,val=21 -numa dist,src=2,dst=1,val=21 -numa dist,src=2,dst=2,val=10 \ -numa hmat-lb,initiator=0,target=0,hierarchy=memory,data-type=access-latency,latency=10 \ -numa hmat-lb,initiator=0,target=0,hierarchy=memory,data-type=access-bandwidth,bandwidth=800M \ -numa hmat-lb,initiator=0,target=2,hierarchy=memory,data-type=access-latency,latency=100 \ -numa hmat-lb,initiator=0,target=2,hierarchy=memory,data-type=access-bandwidth,bandwidth=200G \ -numa hmat-lb,initiator=1,target=0,hierarchy=memory,data-type=access-latency,latency=500 \ -numa hmat-lb,initiator=1,target=0,hierarchy=memory,data-type=access-bandwidth,bandwidth=100M \ -numa hmat-lb,initiator=1,target=2,hierarchy=memory,data-type=access-latency,latency=50 \ -numa hmat-lb,initiator=1,target=2,hierarchy=memory,data-type=access-bandwidth,bandwidth=200G Starting from this and... * Changing links for the two type 3 devices to x-speed=16,x-width=4 will make it leaf link limited. * Changing link for the USP to x-speed=32,x-width=8 will make it head link limited. Case 3: 2 GP, 2 HB, direct connected type 3. ============================================ GP limited configuration (found bug in v6 of kernel series) -object memory-backend-file,id=cxl-mem1,share=on,mem-path=/tmp/t3_cxl1.raw,size=1G,align=256M \ -object memory-backend-file,id=cxl-mem2,share=on,mem-path=/tmp/t3_cxl2.raw,size=1G,align=256M \ -object memory-backend-file,id=cxl-lsa1,share=on,mem-path=/tmp/t3_lsa1.raw,size=1M,align=1M \ -object memory-backend-file,id=cxl-mem3,share=on,mem-path=/tmp/t3_cxl3.raw,size=1G,align=256M \ -object memory-backend-file,id=cxl-mem4,share=on,mem-path=/tmp/t3_cxl4.raw,size=1G,align=256M \ -object memory-backend-file,id=cxl-lsa2,share=on,mem-path=/tmp/t3_lsa2.raw,size=1M,align=1M \ -device pxb-cxl,bus_nr=12,bus=pcie.0,id=cxl.1,hdm_for_passthrough=true,numa_node=2 \ -device pxb-cxl,bus_nr=22,bus=pcie.0,id=cxl.2,hdm_for_passthrough=true,numa_node=3 \ -device cxl-rp,port=0,bus=cxl.1,id=cxl_rp_port0,chassis=0,slot=2 \ -device cxl-rp,port=1,bus=cxl.2,id=cxl_rp_port1,chassis=0,slot=3 \ -device cxl-type3,bus=cxl_rp_port0,volatile-memdev=cxl-mem1,id=cxl-pmem1,lsa=cxl-lsa1,sn=3,x-speed=32,x-width=16 \ -device cxl-type3,bus=cxl_rp_port1,volatile-memdev=cxl-mem3,id=cxl-pmem2,lsa=cxl-lsa2,sn=5,x-speed=32,x-width=16 \ -machine cxl-fmw.0.targets.0=cxl.1,cxl-fmw.0.targets.1=cxl.2,cxl-fmw.0.size=8G,cxl-fmw.0.interleave-granularity=1k \ -numa node,nodeid=0,cpus=0-3,memdev=mem0 \ -numa node,nodeid=1 \ -object acpi-generic-initiator,id=bob2,pci-dev=bob,node=1 \ -numa node,nodeid=2 \ -object acpi-generic-port,id=bob11,pci-bus=cxl.1,node=2 \ -numa node,nodeid=3 \ -object acpi-generic-port,id=bob12,pci-bus=cxl.2,node=3 \ -numa dist,src=0,dst=0,val=10 -numa dist,src=0,dst=1,val=21 -numa dist,src=0,dst=2,val=21 -numa dist,src=0,dst=3,val=21 \ -numa dist,src=1,dst=0,val=21 -numa dist,src=1,dst=1,val=10 -numa dist,src=1,dst=2,val=21 -numa dist,src=1,dst=3,val=21 \ -numa dist,src=2,dst=0,val=21 -numa dist,src=2,dst=1,val=21 -numa dist,src=2,dst=2,val=10 -numa dist,src=2,dst=3,val=21 \ -numa dist,src=3,dst=0,val=21 -numa dist,src=3,dst=1,val=21 -numa dist,src=3,dst=2,val=21 -numa dist,src=3,dst=3,val=10 \ -numa hmat-lb,initiator=0,target=0,hierarchy=memory,data-type=access-latency,latency=10 \ -numa hmat-lb,initiator=0,target=0,hierarchy=memory,data-type=access-bandwidth,bandwidth=3G \ -numa hmat-lb,initiator=0,target=2,hierarchy=memory,data-type=access-latency,latency=100 \ -numa hmat-lb,initiator=0,target=2,hierarchy=memory,data-type=access-bandwidth,bandwidth=13G \ -numa hmat-lb,initiator=0,target=3,hierarchy=memory,data-type=access-latency,latency=100 \ -numa hmat-lb,initiator=0,target=3,hierarchy=memory,data-type=access-bandwidth,bandwidth=13G \ -numa hmat-lb,initiator=1,target=0,hierarchy=memory,data-type=access-latency,latency=500 \ -numa hmat-lb,initiator=1,target=0,hierarchy=memory,data-type=access-bandwidth,bandwidth=100M \ -numa hmat-lb,initiator=1,target=2,hierarchy=memory,data-type=access-latency,latency=50 \ -numa hmat-lb,initiator=1,target=2,hierarchy=memory,data-type=access-bandwidth,bandwidth=20G \ -numa hmat-lb,initiator=1,target=3,hierarchy=memory,data-type=access-latency,latency=50 \ -numa hmat-lb,initiator=1,target=3,hierarchy=memory,data-type=access-bandwidth,bandwidth=20G \ [1] https://lore.kernel.org/linux-cxl/20240710222716.797267-1-dave.ji...@intel.com Suggested tweaks to the CDAT values. diff --git a/hw/mem/cxl_type3.c b/hw/mem/cxl_type3.c index 88f117576d..fe5fc8cb6c 100644 --- a/hw/mem/cxl_type3.c +++ b/hw/mem/cxl_type3.c @@ -109,7 +109,7 @@ static void ct3_build_cdat_entries_for_mr(CDATSubHeader **cdat_table, .flags = HMAT_LB_MEM_MEMORY, .data_type = HMAT_LB_DATA_READ_BANDWIDTH, .entry_base_unit = 1000, /* GB/s */ - .entry[0] = 16, + .entry[0] = 48, }; dslbis3 = g_malloc(sizeof(*dslbis3)); @@ -122,7 +122,7 @@ static void ct3_build_cdat_entries_for_mr(CDATSubHeader **cdat_table, .flags = HMAT_LB_MEM_MEMORY, .data_type = HMAT_LB_DATA_WRITE_BANDWIDTH, .entry_base_unit = 1000, /* GB/s */ - .entry[0] = 16, + .entry[0] = 48, }; dsemts = g_malloc(sizeof(*dsemts)); diff --git a/hw/pci-bridge/cxl_upstream.c b/hw/pci-bridge/cxl_upstream.c index 70cb06436d..85b114d3ce 100644 --- a/hw/pci-bridge/cxl_upstream.c +++ b/hw/pci-bridge/cxl_upstream.c @@ -319,5 +319,5 @@ static int build_cdat_table(CDATSubHeader ***cdat_table, void *priv) sslbis_bandwidth->sslbe[i] = (CDATSslbe) { .port_x_id = CDAT_PORT_ID_USP, .port_y_id = port_ids[i], - .latency_bandwidth = 16, /* 16 GB/s */ + .latency_bandwidth = 128, /* 128 GB/s */ }; Jonathan Cameron (6): hw/pci-bridge/cxl_root_port: Provide x-speed and x-width properties. hw/pci-bridge/cxl_upstream: Provide x-speed and x-width properties. hw/pcie: Factor out PCI Express link register filing common to EP. hw/pcie: Provide a utility function for control of EP / SW USP link hw/mem/cxl-type3: Add properties to control link speed and width hw/pci-bridge/cxl-upstream: Add properties to control link speed and width include/hw/cxl/cxl_device.h | 4 + include/hw/pci-bridge/cxl_upstream_port.h | 4 + include/hw/pci/pcie.h | 2 + hw/mem/cxl_type3.c | 6 ++ hw/pci-bridge/cxl_downstream.c | 23 +++-- hw/pci-bridge/cxl_root_port.c | 5 ++ hw/pci-bridge/cxl_upstream.c | 6 ++ hw/pci/pcie.c | 105 ++++++++++++++-------- 8 files changed, 103 insertions(+), 52 deletions(-) -- 2.43.0