Merge branch 'refs/heads/550.127.08-p2p' into 550.142-p2p

# Conflicts: # README.md
ORDIS-Co-Ltd · Dec 19, 2024 · 4638499 · 4638499
2 parents 346011d + 1d243c2
commit 4638499
Show file tree

Hide file tree

Showing 11 changed files with 238 additions and 975 deletions.
diff --git a/.github/workflows/build.yml b/.github/workflows/build.yml
@@ -0,0 +1,51 @@
+name: Build
+
+on:
+  push:
+
+permissions:
+  contents: write
+
+jobs:
+  build-deb:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - name: Install dependencies
+        run: sudo apt-get install -y build-essential dpkg-dev
+      - name: Download source deb
+        run: |
+          branch=$(echo $GITHUB_REF | cut -d'/' -f3)
+          branch=${branch/-p2p/}
+          # download source deb
+          curl -o source.deb https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/x86_64/nvidia-kernel-source-550-open_${branch}-0ubuntu1_amd64.deb
+      - name: Extract source deb
+        run: dpkg-deb -R source.deb source-deb
+      - name: Build kernel module
+        run: |
+          make modules -j$(nproc)
+      - name: Replace kernel module source
+        run: |
+          branch=$(echo $GITHUB_REF | cut -d'/' -f3)
+          branch=${branch/-p2p/}
+          cp kernel-open/nvidia/nv-pci.c source-deb/usr/src/nvidia-${branch}/nvidia/nv-pci.c
+      - name: Replace kernel module
+        run: |
+          branch=$(echo $GITHUB_REF | cut -d'/' -f3)
+          branch=${branch/-p2p/}
+          cp src/nvidia/_out/Linux_x86_64/nv-kernel.o source-deb/usr/src/nvidia-${branch}/nvidia/nv-kernel.o_binary
+          cp src/nvidia-modeset/_out/Linux_x86_64/nv-modeset-kernel.o source-deb/usr/src/nvidia-${branch}/nvidia-modeset/nv-modeset-kernel.o_binary
+      - name: Repack deb
+        run: dpkg-deb -b source-deb nvidia-kernel-source-550-open-0ubuntu1_amd64.deb
+      - name: Upload deb
+        uses: actions/upload-artifact@v4
+        with:
+          name: nvidia-kernel-source-550-open-0ubuntu1_amd64.deb
+          path: nvidia-kernel-source-550-open-0ubuntu1_amd64.deb
+      - name: Create release
+        uses: softprops/action-gh-release@v2
+        if: startsWith(github.ref, 'refs/tags/')
+        with:
+          files: nvidia-kernel-source-550-open-0ubuntu1_amd64.deb
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
diff --git a/README.md b/README.md
diff --git a/install.sh b/install.sh
@@ -0,0 +1,7 @@
+#!/bin/bash
+sudo rmmod nvidia_drm nvidia_modeset nvidia_uvm nvidia
+set -e
+make modules -j$(nproc)
+sudo make modules_install -j$(nproc)
+sudo depmod
+nvidia-smi
diff --git a/kernel-open/nvidia/nv-pci.c b/kernel-open/nvidia/nv-pci.c
@@ -184,11 +184,11 @@ static int nv_resize_pcie_bars(struct pci_dev *pci_dev) {
     struct pci_host_bridge *host;
 #endif
 
-    if (NVreg_EnableResizableBar == 0)
+    /*if (NVreg_EnableResizableBar == 0)
     {
         nv_printf(NV_DBG_INFO, "NVRM: resizable BAR disabled by regkey, skipping\n");
         return 0;
-    }
+    }*/
 
     // Check if BAR1 has PCIe rebar capabilities
     sizes = pci_rebar_get_possible_sizes(pci_dev, NV_GPU_BAR1);
@@ -404,7 +404,7 @@ nv_init_coherent_link_info
             /* Fail for the baremetal case */
             goto failed;
         }
-        
+
         /*
          * For the virtualization usecase on SHH, the coherent GPU memory
          * PA is exposed as BAR2 to the VM and the "nvidia,gpu-mem-base-pa"
@@ -1019,7 +1019,7 @@ nv_pci_remove(struct pci_dev *pci_dev)
                 nv_printf(NV_DBG_ERRORS,
                           "NVRM: Failed removal of device %04x:%02x:%02x.%x!\n",
                           NV_PCI_DOMAIN_NUMBER(pci_dev), NV_PCI_BUS_NUMBER(pci_dev),
-                          NV_PCI_SLOT_NUMBER(pci_dev), PCI_FUNC(pci_dev->devfn));  
+                          NV_PCI_SLOT_NUMBER(pci_dev), PCI_FUNC(pci_dev->devfn));
                 WARN_ON(1);
                 goto done;
             }

diff --git a/src/nvidia/src/kernel/gpu/bif/kernel_bif.c b/src/nvidia/src/kernel/gpu/bif/kernel_bif.c
@@ -709,7 +709,7 @@ _kbifInitRegistryOverrides
     NvU32 data32;
 
     // P2P Override
-    pKernelBif->p2pOverride = BIF_P2P_NOT_OVERRIDEN;
+    pKernelBif->p2pOverride = 0x11;
     if (osReadRegistryDword(pGpu, NV_REG_STR_CL_FORCE_P2P, &data32) == NV_OK)
     {
         pKernelBif->p2pOverride = data32;
@@ -718,7 +718,7 @@ _kbifInitRegistryOverrides
     }
 
     // P2P force type override
-    pKernelBif->forceP2PType = NV_REG_STR_RM_FORCE_P2P_TYPE_DEFAULT;
+    pKernelBif->forceP2PType = NV_REG_STR_RM_FORCE_P2P_TYPE_BAR1P2P;
     if (osReadRegistryDword(pGpu, NV_REG_STR_RM_FORCE_P2P_TYPE, &data32) == NV_OK &&
         (data32 <= NV_REG_STR_RM_FORCE_P2P_TYPE_MAX))
     {

diff --git a/src/nvidia/src/kernel/gpu/bus/arch/hopper/kern_bus_gh100.c b/src/nvidia/src/kernel/gpu/bus/arch/hopper/kern_bus_gh100.c
@@ -333,7 +333,7 @@ kbusVerifyBar2_GH100
         {
             memdescCreateExisting(&memDesc, pGpu, size, ADDR_FBMEM, NV_MEMORY_UNCACHED, MEMDESC_FLAGS_NONE);
         }
-        memdescTagAlloc(status, NV_FB_ALLOC_RM_INTERNAL_OWNER_UNNAMED_TAG_95, 
+        memdescTagAlloc(status, NV_FB_ALLOC_RM_INTERNAL_OWNER_UNNAMED_TAG_95,
                         (&memDesc));
         if (status != NV_OK)
         {
@@ -765,7 +765,7 @@ kbusCreateP2PMapping_GH100
 
     if (FLD_TEST_DRF(_P2PAPI, _ATTRIBUTES, _CONNECTION_TYPE, _PCIE_BAR1, attributes))
     {
-        return kbusCreateP2PMappingForBar1P2P_HAL(pGpu0, pKernelBus0, pGpu1, pKernelBus1, attributes);
+        return kbusCreateP2PMappingForBar1P2P_GH100(pGpu0, pKernelBus0, pGpu1, pKernelBus1, attributes);
     }
 
     if (FLD_TEST_DRF(_P2PAPI, _ATTRIBUTES, _CONNECTION_TYPE, _PCIE, attributes))
@@ -816,7 +816,7 @@ kbusRemoveP2PMapping_GH100
 
     if (FLD_TEST_DRF(_P2PAPI, _ATTRIBUTES, _CONNECTION_TYPE, _PCIE_BAR1, attributes))
     {
-        return kbusRemoveP2PMappingForBar1P2P_HAL(pGpu0, pKernelBus0, pGpu1, pKernelBus1, attributes);
+        return kbusRemoveP2PMappingForBar1P2P_GH100(pGpu0, pKernelBus0, pGpu1, pKernelBus1, attributes);
     }
 
     if (FLD_TEST_DRF(_P2PAPI, _ATTRIBUTES, _CONNECTION_TYPE, _PCIE, attributes))
@@ -1139,7 +1139,7 @@ kbusIsBar1P2PCapable_GH100
     //
     // GPU BAR1 supports the SYSMEM mapping for the bar1 doorbell, RM needs
     // to make sure that BAR1 VA has 128KB space left for such cases after all
-    // FB statically mapped in BAR1. Bug 3869651 #14. 
+    // FB statically mapped in BAR1. Bug 3869651 #14.
     //
     if ((bar1VASize < (32 * RM_PAGE_SIZE)) ||
         ((bar1VASize - (32 * RM_PAGE_SIZE)) < RM_ALIGN_UP(fbSize, RM_PAGE_SIZE_2M)))
@@ -1448,7 +1448,7 @@ kbusCreateP2PMappingForBar1P2P_GH100
         return NV_ERR_NOT_SUPPORTED;
     }
 
-    if (!kbusIsPcieBar1P2PMappingSupported_HAL(pGpu0, pKernelBus0, pGpu1, pKernelBus1))
+    if (!kbusIsPcieBar1P2PMappingSupported_GH100(pGpu0, pKernelBus0, pGpu1, pKernelBus1))
     {
         return NV_ERR_NOT_SUPPORTED;
     }
@@ -1871,8 +1871,8 @@ kbusEnableStaticBar1Mapping_GH100
     pKernelBus->bar1[gfid].staticBar1.pDmaMemDesc = pDmaMemDesc;
     pKernelBus->bar1[gfid].staticBar1.size = bar1MapSize;
 
-    NV_PRINTF(LEVEL_INFO, "Static bar1 mapped offset 0x%llx size 0x%llx\n",
-                           bar1Offset, bar1MapSize);
+    NV_PRINTF(LEVEL_ERROR, "Static bar1 mapped offset 0x%llx size 0x%llx bar1BusAddr 0x%llx\n",
+                           bar1Offset, bar1MapSize, bar1BusAddr);
 
     return NV_OK;
 

diff --git a/src/nvidia/src/kernel/gpu/bus/arch/maxwell/kern_bus_gm107.c b/src/nvidia/src/kernel/gpu/bus/arch/maxwell/kern_bus_gm107.c
@@ -1008,7 +1008,7 @@ kbusInitBar1_GM107(OBJGPU *pGpu, KernelBus *pKernelBus, NvU32 gfid)
     //
     NV_ASSERT(pKernelBus->bar1[gfid].apertureLength <= kbusGetPciBarSize(pKernelBus, 1));
 
-    bBar1P2PCapable = kbusIsBar1P2PCapable(pGpu, pKernelBus, gfid);
+    bBar1P2PCapable = kbusIsBar1P2PCapable_GH100(pGpu, pKernelBus, gfid);
 
     //
     // If we need to preserve a console mapping at the start of BAR1, we
@@ -1094,7 +1094,7 @@ kbusInitBar1_GM107(OBJGPU *pGpu, KernelBus *pKernelBus, NvU32 gfid)
     {
         // Enable the static BAR1 mapping for the BAR1 P2P
         NV_ASSERT_OK_OR_GOTO(rmStatus,
-                             kbusEnableStaticBar1Mapping_HAL(pGpu, pKernelBus, gfid),
+                             kbusEnableStaticBar1Mapping_GH100(pGpu, pKernelBus, gfid),
                              kbusInitBar1_failed);
     }
     else
@@ -1605,7 +1605,7 @@ kbusSetupBar2GpuVaSpace_GM107
             NV_ASSERT_OR_RETURN(status == NV_OK, status);
         }
 
-        memdescTagAlloc(status, NV_FB_ALLOC_RM_INTERNAL_OWNER_UNNAMED_TAG_63, 
+        memdescTagAlloc(status, NV_FB_ALLOC_RM_INTERNAL_OWNER_UNNAMED_TAG_63,
                         pKernelBus->bar2[gfid].pInstBlkMemDesc);
         NV_ASSERT_OR_RETURN(status == NV_OK, status);
 
@@ -1687,7 +1687,7 @@ kbusSetupBar2GpuVaSpace_GM107
 
         if (IS_GFID_VF(gfid))
         {
-            memdescTagAlloc(status, NV_FB_ALLOC_RM_INTERNAL_OWNER_UNNAMED_TAG_64, 
+            memdescTagAlloc(status, NV_FB_ALLOC_RM_INTERNAL_OWNER_UNNAMED_TAG_64,
                             pPageLevelsMemDesc);
             NV_ASSERT_OR_GOTO(status == NV_OK, cleanup);
 
@@ -3013,15 +3013,15 @@ _kbusMapAperture_GM107
     NvU32               swizzId = KMIGMGR_SWIZZID_INVALID;
     NvU32               gfid;
 
-    if (kbusIsStaticBar1Enabled(pGpu, pKernelBus) &&
+    /*if (kbusIsStaticBar1Enabled(pGpu, pKernelBus) &&
         (memdescGetAddressSpace(pMemDesc) == ADDR_FBMEM))
     {
         NV_ASSERT_OK_OR_RETURN(vgpuGetCallingContextGfid(pGpu, &gfid));
 
         return kbusGetStaticFbAperture_HAL(pGpu, pKernelBus, pMemDesc,
                                            offset, pAperOffset,
                                            pLength, gfid);
-    }
+    }*/
 
     // Ensure that the BAR1 VA space is the same across all subdevices
     if (IsSLIEnabled(pGpu) && ((mapFlags & BUS_MAP_FB_FLAGS_MAP_UNICAST) == 0))
@@ -3136,12 +3136,12 @@ _kbusUnmapAperture_GM107
     NV_STATUS           rmStatus = NV_OK;
     VirtMemAllocator   *pDma = GPU_GET_DMA(pGpu);
 
-    if (kbusIsStaticBar1Enabled(pGpu, pKernelBus) &&
+    /*if (kbusIsStaticBar1Enabled(pGpu, pKernelBus) &&
         (memdescGetAddressSpace(pMemDesc) == ADDR_FBMEM))
     {
         // No op for the static bar1 mode
         return NV_OK;
-    }
+    }*/
 
     rmStatus = dmaFreeMapping_HAL(pGpu, pDma, pVAS, aperOffset, pMemDesc, 0, NULL);
 
@@ -3749,7 +3749,7 @@ kbusVerifyBar2_GM107
         {
             memdescCreateExisting(&memDesc, pGpu, size, ADDR_FBMEM, NV_MEMORY_UNCACHED, MEMDESC_FLAGS_NONE);
         }
-        memdescTagAlloc(status, NV_FB_ALLOC_RM_INTERNAL_OWNER_UNNAMED_TAG_65, 
+        memdescTagAlloc(status, NV_FB_ALLOC_RM_INTERNAL_OWNER_UNNAMED_TAG_65,
                         (&memDesc));
         if (status != NV_OK)
         {
@@ -4805,7 +4805,7 @@ kbusBar1InstBlkVasUpdate_GM107
             NV_ASSERT_OR_RETURN(status == NV_OK, status);
         }
 
-        memdescTagAlloc(status, NV_FB_ALLOC_RM_INTERNAL_OWNER_UNNAMED_TAG_66, 
+        memdescTagAlloc(status, NV_FB_ALLOC_RM_INTERNAL_OWNER_UNNAMED_TAG_66,
                         pKernelBus->bar1[gfid].pInstBlkMemDesc);
         NV_ASSERT_OR_RETURN(status == NV_OK, status);
 

diff --git a/src/nvidia/src/kernel/gpu/bus/arch/pascal/kern_bus_gp100.c b/src/nvidia/src/kernel/gpu/bus/arch/pascal/kern_bus_gp100.c
@@ -70,6 +70,11 @@ kbusCreateP2PMapping_GP100
         return kbusCreateP2PMappingForMailbox_HAL(pGpu0, pKernelBus0, pGpu1, pKernelBus1, peer0, peer1, attributes);
     }
 
+    if (FLD_TEST_DRF(_P2PAPI, _ATTRIBUTES, _CONNECTION_TYPE, _PCIE_BAR1, attributes))
+    {
+        return kbusCreateP2PMappingForBar1P2P_GH100(pGpu0, pKernelBus0, pGpu1, pKernelBus1, attributes);
+    }
+
     NV_PRINTF(LEVEL_ERROR, "P2P type %d is not supported\n", DRF_VAL(_P2PAPI, _ATTRIBUTES, _CONNECTION_TYPE, attributes));
 
     return NV_ERR_NOT_SUPPORTED;

diff --git a/src/nvidia/src/kernel/gpu/bus/p2p_api.c b/src/nvidia/src/kernel/gpu/bus/p2p_api.c
@@ -575,20 +575,20 @@ p2papiConstruct_IMPL
         //
         // TODO: This function need to have a cleanup path when this function
         //       fails after kbusCreateP2PMaping(), busBindLocalGfidForP2P()
-        //       and busBindRemoteGfidForP2P(). The current state, the 
+        //       and busBindRemoteGfidForP2P(). The current state, the
         //       function just returns an error. Bug 4016670 filed to track
         //       the effort.
         //
 
         // setup the p2p resources
         NV_CHECK_OK_OR_RETURN(LEVEL_ERROR,
-                              kbusCreateP2PMapping_HAL(pLocalGpu, pLocalKernelBus, pRemoteGpu,
+                              kbusCreateP2PMapping_GH100(pLocalGpu, pLocalKernelBus, pRemoteGpu,
                                                        pRemoteKernelBus, &peer1, &peer2,
                                                        pP2PApi->attributes));
         if (bEgmPeer)
         {
             NV_CHECK_OK_OR_RETURN(LEVEL_ERROR,
-                                  kbusCreateP2PMapping_HAL(pLocalGpu, pLocalKernelBus, pRemoteGpu,
+                                  kbusCreateP2PMapping_GH100(pLocalGpu, pLocalKernelBus, pRemoteGpu,
                                                            pRemoteKernelBus, &egmPeer1, &egmPeer2,
                                                            pP2PApi->attributes |
                                                            DRF_DEF(_P2PAPI, _ATTRIBUTES, _REMOTE_EGM, _YES)));
@@ -598,14 +598,14 @@ p2papiConstruct_IMPL
             (pCallContext->secInfo.privLevel >= RS_PRIV_LEVEL_KERNEL))
         {
             NV_CHECK_OK_OR_RETURN(LEVEL_ERROR,
-                                  kbusGetBar1P2PDmaInfo_HAL(pLocalGpu, pRemoteGpu,
+                                  kbusGetBar1P2PDmaInfo_GH100(pLocalGpu, pRemoteGpu,
                                       pRemoteKernelBus,
                                       &pNv503bAllocParams->l2pBar1P2PDmaInfo.dma_address,
                                       &pNv503bAllocParams->l2pBar1P2PDmaInfo.dma_size));
 
             NV_CHECK_OK_OR_RETURN(LEVEL_ERROR,
-                                  kbusGetBar1P2PDmaInfo_HAL(pRemoteGpu, pLocalGpu,
-                                      pLocalKernelBus, 
+                                  kbusGetBar1P2PDmaInfo_GH100(pRemoteGpu, pLocalGpu,
+                                      pLocalKernelBus,
                                       &pNv503bAllocParams->p2lBar1P2PDmaInfo.dma_address,
                                       &pNv503bAllocParams->p2lBar1P2PDmaInfo.dma_size));
         }
@@ -740,7 +740,7 @@ p2papiDestruct_IMPL
     {
         // remove any resources associated with this mapping
         NV_CHECK_OK_OR_GOTO(status, LEVEL_ERROR,
-                            kbusRemoveP2PMapping_HAL(pLocalGpu, pLocalKernelBus,
+                            kbusRemoveP2PMapping_GH100(pLocalGpu, pLocalKernelBus,
                                                      pRemoteGpu, pRemoteKernelBus,
                                                      pP2PApi->peerId1, pP2PApi->peerId2,
                                                      pP2PApi->attributes), end);
@@ -749,7 +749,7 @@ p2papiDestruct_IMPL
             memmgrIsLocalEgmEnabled(GPU_GET_MEMORY_MANAGER(pRemoteGpu)) &&
             !GPU_IS_NVSWITCH_DETECTED(pLocalGpu))
         {
-            status = kbusRemoveP2PMapping_HAL(pLocalGpu, pLocalKernelBus,
+            status = kbusRemoveP2PMapping_GH100(pLocalGpu, pLocalKernelBus,
                                               pRemoteGpu, pRemoteKernelBus,
                                               pP2PApi->egmPeerId1, pP2PApi->egmPeerId2,
                                               pP2PApi->attributes |