From 7aef82a5112ee72ee2fc5a430db072085933e927 Mon Sep 17 00:00:00 2001 From: Arthur Hanson Rasmusson Date: Wed, 23 Sep 2026 00:41:46 -0400 Subject: kernel: serve GMMU static info + per-VF fault/access-counter sizes for UVM Three fixes that clear all driver-visible errors in UVM's device create (cuInit still fails later in UVM_REGISTER_GPU): 1. GSP_RM_CONTROL 0x20800a59 INTERNAL_GMMU_GET_STATIC_INFO (new handler): kgmmu.pStaticInfo->replayableFaultBufferSize feeds kgmmuFaultBufferReplayableAllocate (called by faultbufConstruct when UVM opens the device). The generic zero reply made it a 0-page allocation: "faultbufConstruct_IMPL: Failed to setup Replayable Fault buffer (status=0x51)". Report 64KB replayable / 4KB legacy, mirroring kgmmuSetAndGetDefaultFaultBufferSize_TU102 defaults. 2. VF-window NV_VIRTUAL_FUNCTION_PRIV_MMU_FAULT_BUFFER_SIZE(0/1) (BAR0+0xB83010/0xB83030, plus plain aliases) pre-filled with _VAL in NVC369_BUF_SIZE units (0x800=64KB, 0x80=4KB) so the local (non-RPC) path reads the same values the control handler reports. 3. VF-window NV_VIRTUAL_FUNCTION_PRIV_ACCESS_COUNTER_NOTIFY_BUFFER_SIZE (BAR0+0xB83110, plus plain alias) = 0x1000 (the value the host's physical TU104 reports; x32B = 128KB). uvmGetAccessCounterBufferSize_ TU102 multiplied our zero and uvmInitializeAccessCntrBuffer failed with NV_ERR_NO_MEMORY at mem_desc.c:1353. Also note (companion finding, no code change): the guest must be placed behind a pcie-root-port; on the Q35 root bus QEMU rewrites the PCIe cap to Root Complex Integrated Endpoint and zeroes LNKCAP/LNKCTL/DeviceCaps (vfio_setup_pcie_cap: "On a Root Complex bus Endpoints become Root Complex Integrated Endpoints, which changes the type and clears the LNK & LNK2 fields"), which is what made getPCIELinkRateMBps read speed 0. --- kernel/bar_emulation.c | 29 +++++++++++++++++++++++++++++ 1 file changed, 29 insertions(+) (limited to 'kernel/bar_emulation.c') diff --git a/kernel/bar_emulation.c b/kernel/bar_emulation.c index 2d82303..685831a 100644 --- a/kernel/bar_emulation.c +++ b/kernel/bar_emulation.c @@ -309,6 +309,35 @@ int nv_vgpu_bar_emul_init(struct nv_vgpu_device *vgpu) *(u32 *)((u8 *)ctx->bar0_shadow + 0xC08084) = 0x00000103; *(u32 *)((u8 *)ctx->bar0_shadow + 0xC08088) = 0x01030000; + /* + * Per-VF static sizes the driver reads through the VF register window + * (GPU_VREG_RD32 = BAR0 + sriovState.virtualRegPhysOffset, 0xB80000 + * for a non-SRIOV virtual GPU). Both are read-only SIZE registers on + * real hardware; we pre-fill them so the local (non-RPC) paths agree + * with the GSP_RM_CONTROL replies in gsp_rm_emulation.c. + * + * 0x3010/0x3030 NV_VIRTUAL_FUNCTION_PRIV_MMU_FAULT_BUFFER_SIZE(0/1): + * _VAL is in NVC369_BUF_SIZE (32B) units. 0x800 -> 64KB + * replayable, 0x80 -> 4KB legacy; matches TU104_GMMU_*_SIZE. + * kgmmuSetAndGetDefaultFaultBufferSize_TU102 writes SET_DEFAULT + * (bit30) and reads _VAL back, so a pre-filled _VAL is all it + * needs. + * 0x3110 NV_VIRTUAL_FUNCTION_PRIV_ACCESS_COUNTER_NOTIFY_BUFFER_SIZE: + * UVM's uvmGetAccessCounterBufferSize_TU102 multiplies it by + * NV_ACCESS_COUNTER_NOTIFY_BUF_SIZE (32B). 0x1000 -> 128KB, the + * value the host's physical TU104 reports. Zero makes the + * access-counter buffer a 0-byte allocation and cuInit fails in + * uvmInitializeAccessCntrBuffer. + */ + *(u32 *)((u8 *)ctx->bar0_shadow + 0xB83010) = 0x00000800; + *(u32 *)((u8 *)ctx->bar0_shadow + 0xB83030) = 0x00000080; + *(u32 *)((u8 *)ctx->bar0_shadow + 0x003010) = 0x00000800; + *(u32 *)((u8 *)ctx->bar0_shadow + 0x003030) = 0x00000080; + *(u32 *)((u8 *)ctx->bar0_shadow + 0xB83110) = 0x00001000; + *(u32 *)((u8 *)ctx->bar0_shadow + 0x003110) = 0x00001000; + *(u32 *)((u8 *)ctx->bar0_shadow + 0xB8311C) = 0x00000000; + *(u32 *)((u8 *)ctx->bar0_shadow + 0xB83118) = 0x00000000; + /* * Serve a valid VBIOS at the PROM window (BAR0+0x300000). -- cgit v1.2.3