summaryrefslogtreecommitdiffstats
path: root/legacy-driver-guest/vf
diff options
context:
space:
mode:
authorArthur Hanson Rasmusson <arthur@vgpu.io>2026-09-21 13:06:51 -0400
committerArthur Hanson Rasmusson <arthur@vgpu.io>2026-09-21 13:06:51 -0400
commitc3a295efd1b1132bbaa5d422cd05b3fa8e89a8bf (patch)
treef74bd7724153cd9d870688df3ff75274cd5000da /legacy-driver-guest/vf
parent5d4d26f8e773f2871c1b5c9266b82495d390970e (diff)
downloadopen-iov-c3a295efd1b1132bbaa5d422cd05b3fa8e89a8bf.tar.gz
open-iov-c3a295efd1b1132bbaa5d422cd05b3fa8e89a8bf.zip
kernel: Fix GSP emulation shadow binding, eventfd wiring, RPC livelocks + docs suite + repo reorg
CRITICAL FIXES (guest driver bring-up): - gsp_emulation.c: emulation engine bound to the WRONG BAR0 - it polled the host physical BAR0 (bars[0].vaddr ioremap) for the doorbell and read RPCs from a local dma_alloc_wc buffer, while the guest writes land in the vzalloc'd bar0_shadow. The RPC path could never work; PBA writes even corrupted the host GPU's physical BAR0+0x3000. gsp_emu_init now binds exclusively to bar0_shadow (doorbell/mailbox/PBA aliases; shared_mem_dma==0 marks the alias so cleanup never dma_free's it). - main.c SET_IRQS: DATA_EVENTFD eventfds were DISCARDED ('TODO' + kfree) - the guest never received MSI-X completions. Full wiring now: eventfd_ctx_fdget per vector, fd -1 clears, dev_close releases refs. gsp_emu_inject_completion_interrupt looks up vgpu->irqs[0].trigger LIVE (init-time copy was permanently NULL since QEMU wires after gsp_emu_init). - gsp_rpc_client.c: BAR_TRAP/MIGRATE in emulation mode called gsp_rpc_submit_raw which rings the SAME doorbell the thread polls - self-triggered RPC livelock + multi-second stalls. Now fail fast with -ENOTSUPP + pr_warn_ratelimited (guest BAR access is direct MMIO). MORE FIXES: - main.c: nv_vgpu_config_write returned SIZE_MAX (while(count--) wrap) and was dead code while the VFIO write handler did unfiltered byte copies (guest could rewrite vendor ID/BARs). Return fixed; VFIO write handler now routes through the per-register filter. - kernel/Makefile: M=$(PWD) used the env var which is stale when invoked via the top-level make (pointed at repo root -> kbuild silently built NOTHING; explains every 'stale .ko' incident). Now M=$(CURDIR). - Warnings sweep: header prototypes for mdev/type functions, format casts in gsp_comm.c DRY-RUN logs, unused variable. ROBUSTNESS (harness/tooling): - vgpu_vm.py: run had no timeout (start could block forever); guest_shell recv could block forever on a wedged guest. Both now bounded; guest_shell pre-flights guest-ping and reports wedge diagnostics. - stack_manager.sh: qemu/qemu-debug now launch via the vgpu_vm.py harness (workflow rule: never raw QEMU). REPO REORGANIZATION: - driver/ (legacy guest-side gsp-vgpu.ko) moved to legacy-driver-guest/ (reference only, not built; README explains). 3 legacy un-buildable test files moved to legacy-driver-guest/kernel-tests/. Top-level Makefile cleaned of the dead DRIVER_DIR reference. DOCUMENTATION SUITE (new): - docs/GSP_EMULATION_ARCHITECTURE.md - emulator deep-dive: shadow BAR0 model, doorbell protocol, dispatch table, MSI-X path, lifecycle, memory- model caveats - docs/DEBUGGING.md - QEMU GDB stub workflow, serial console, QMP NMI, failure taxonomy, full bring-up validation checklist, EPT-livelock case study, version-matrix discipline - docs/ONBOARDING.md - repo tour, coding standards (incl. virt_to_phys / dma_addr_t / get-put rules), first build & test - docs/VERSION_UPGRADE.md - end-to-end new-driver-version procedure (firmware fetch, Ghidra analysis, configs, signatures, validation) - docs/PERFORMANCE.md - bare-metal A/B methodology + targets - README.md doc index; STATE.md refreshed; DEVELOPMENT_WORKFLOW.md corrected (load order: open-iov-vfio BEFORE nvidia; stock-driver blacklist note)
Diffstat (limited to 'legacy-driver-guest/vf')
-rw-r--r--legacy-driver-guest/vf/gsp_vf.c312
1 files changed, 312 insertions, 0 deletions
diff --git a/legacy-driver-guest/vf/gsp_vf.c b/legacy-driver-guest/vf/gsp_vf.c
new file mode 100644
index 0000000..951c84d
--- /dev/null
+++ b/legacy-driver-guest/vf/gsp_vf.c
@@ -0,0 +1,312 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * GSP vGPU-IO Driver - VF Management
+ */
+
+#include "gsp_core.h"
+#include <linux/slab.h>
+#include <linux/idr.h>
+#include <linux/list.h>
+
+/* VF Reference Counting */
+static void gsp_vf_release(struct kref *ref)
+{
+ struct gsp_vf *vf = container_of(ref, struct gsp_vf, ref);
+ kfree(vf);
+}
+
+struct gsp_vf *gsp_vf_get(struct gsp_device *gsp, uint32_t id)
+{
+ struct gsp_vf *vf;
+
+ mutex_lock(&gsp->vf_lock);
+ vf = idr_find(&gsp->vf_idr, id);
+ if (vf && vf->active)
+ kref_get(&vf->ref);
+ else
+ vf = NULL;
+ mutex_unlock(&gsp->vf_lock);
+
+ return vf;
+}
+
+void gsp_vf_put(struct gsp_vf *vf)
+{
+ if (vf)
+ kref_put(&vf->ref, gsp_vf_release);
+}
+
+/* VF Creation */
+int gsp_vf_create(struct gsp_device *gsp, const struct gsp_vf_config *config)
+{
+ struct gsp_vf *vf;
+ int ret, id;
+
+ if (!config || config->id == 0 || config->id > gsp->max_vfs)
+ return -EINVAL;
+
+ vf = kzalloc(sizeof(*vf), GFP_KERNEL);
+ if (!vf)
+ return -ENOMEM;
+
+ vf->gsp = gsp;
+ vf->id = config->id;
+ kref_init(&vf->ref);
+ mutex_init(&vf->lock);
+ mutex_init(&vf->wpr_lock);
+ INIT_LIST_HEAD(&vf->wpr_list);
+ INIT_LIST_HEAD(&vf->list);
+ atomic_set(&vf->refcount, 1);
+
+ /* Copy configuration */
+ vf->bar1_offset = config->bar1_offset;
+ vf->bar1_size = config->bar1_size;
+ vf->pt_base = config->pt_base;
+ vf->pt_size = config->pt_size;
+ vf->cmd_base = config->cmd_base;
+ vf->cmd_size = config->cmd_size;
+
+ /* Allocate VF ID */
+ mutex_lock(&gsp->vf_lock);
+ id = idr_alloc(&gsp->vf_idr, vf, config->id, config->id + 1, GFP_KERNEL);
+ if (id < 0) {
+ ret = id;
+ mutex_unlock(&gsp->vf_lock);
+ goto err_free;
+ }
+
+ list_add(&vf->list, &gsp->vf_list);
+ vf->active = true;
+ vf->state = GSP_VF_ACTIVE;
+ mutex_unlock(&gsp->vf_lock);
+
+ /* Initialize subsystems */
+ ret = gsp_vf_wpr_init(vf);
+ if (ret) {
+ dev_err(gsp->dev, "VF %u: WPR init failed: %d\n", vf->id, ret);
+ goto err_wpr;
+ }
+
+ ret = gsp_vf_bar_init(vf);
+ if (ret) {
+ dev_err(gsp->dev, "VF %u: BAR init failed: %d\n", vf->id, ret);
+ goto err_bar;
+ }
+
+ dev_info(gsp->dev, "VF %u created (bar1=0x%llx-0x%llx, pt=0x%llx-0x%llx)\n",
+ vf->id, vf->bar1_offset, vf->bar1_offset + vf->bar1_size,
+ vf->pt_base, vf->pt_base + vf->pt_size);
+
+ return 0;
+
+err_bar:
+ gsp_vf_wpr_fini(vf);
+err_wpr:
+ mutex_lock(&gsp->vf_lock);
+ list_del(&vf->list);
+ idr_remove(&gsp->vf_idr, vf->id);
+ vf->active = false;
+ mutex_unlock(&gsp->vf_lock);
+err_free:
+ kfree(vf);
+ return ret;
+}
+
+/* VF Destruction */
+void gsp_vf_destroy(struct gsp_vf *vf)
+{
+ struct gsp_device *gsp = vf->gsp;
+ uint32_t id = vf->id;
+
+ if (!vf->active)
+ return;
+
+ /* Cancel any ongoing migration */
+ mutex_lock(&vf->lock);
+ if (vf->migration) {
+ gsp_migration_cancel(vf->migration);
+ vf->migration = NULL;
+ }
+ vf->active = false;
+ vf->state = GSP_VF_DESTROYING;
+ mutex_unlock(&vf->lock);
+
+ /* Remove from device lists */
+ mutex_lock(&gsp->vf_lock);
+ list_del(&vf->list);
+ idr_remove(&gsp->vf_idr, id);
+ mutex_unlock(&gsp->vf_lock);
+
+ /* Cleanup subsystems */
+ gsp_vf_bar_fini(vf);
+ gsp_vf_wpr_fini(vf);
+
+ dev_info(gsp->dev, "VF %u destroyed\n", id);
+
+ gsp_vf_put(vf);
+}
+
+/* Memory Allocation */
+int gsp_vf_mem_alloc(struct gsp_vf *vf, uint64_t size, uint32_t flags,
+ uint64_t *handle, uint64_t *va)
+{
+ struct gsp_alloc_payload payload = {
+ .size = size,
+ .type = flags & 0x3, /* Low bits: mem type */
+ .alignment = 0x1000,
+ .va_hint = 0,
+ .flags = flags >> 2,
+ .pid = current->pid,
+ .vf_id = vf->id,
+ };
+ struct gsp_alloc_response resp;
+ size_t resp_size = sizeof(resp);
+ int ret;
+
+ if (!handle || !va)
+ return -EINVAL;
+
+ /* In emulation mode, handle locally */
+ if (emulation_mode) {
+ static uint64_t next_handle = 0x1000;
+ static uint64_t next_va = 0x100000000ULL; /* 4GB base */
+
+ *handle = next_handle++;
+ *va = next_va;
+ next_va += (size + 0xFFF) & ~0xFFF; /* Align to 4KB */
+
+ dev_dbg(vf->gsp->dev, "VF %u: alloc handle=0x%llx va=0x%llx size=0x%llx (emulation)\n",
+ vf->id, *handle, *va, size);
+
+ return 0;
+ }
+
+ ret = gsp_rpc_send_sync(vf->gsp->rpc, GSP_RPC_ALLOC,
+ &payload, sizeof(payload),
+ &resp, &resp_size);
+ if (ret)
+ return ret;
+
+ *handle = resp.handle;
+ *va = resp.va;
+
+ dev_dbg(vf->gsp->dev, "VF %u: alloc handle=0x%llx va=0x%llx size=0x%llx\n",
+ vf->id, *handle, *va, size);
+
+ return 0;
+}
+
+int gsp_vf_mem_free(struct gsp_vf *vf, uint64_t handle)
+{
+ /* FREE uses primary dispatcher sub-command */
+ struct gsp_subcmd_desc desc;
+ size_t desc_size = sizeof(desc);
+ int ret;
+
+ /* Subcmd for FREE: need to determine which subcmd corresponds to FREE */
+ /* For now, use a generic FREE RPC if implemented */
+
+ /* Placeholder - actual implementation depends on FREE subcmd mapping */
+ dev_dbg(vf->gsp->dev, "VF %u: free handle=0x%llx\n", vf->id, handle);
+
+ return 0;
+}
+
+/* Memory Pinning */
+int gsp_vf_mem_pin(struct gsp_vf *vf, uint64_t handle,
+ uint64_t offset, uint64_t size,
+ uint64_t *dma_addr)
+{
+ struct gsp_pin_payload payload = {
+ .handle = handle,
+ .offset = offset,
+ .size = size,
+ .vf_id = vf->id,
+ .flags = GSP_PIN_READ | GSP_PIN_WRITE,
+ };
+ struct gsp_pin_response resp;
+ size_t resp_size = sizeof(resp);
+ int ret;
+
+ if (!dma_addr)
+ return -EINVAL;
+
+ /* In emulation mode, handle locally */
+ if (emulation_mode) {
+ static uint64_t next_dma = 0x200000000ULL; /* 8GB base */
+ *dma_addr = next_dma;
+ next_dma += (size + 0xFFF) & ~0xFFF;
+
+ dev_dbg(vf->gsp->dev, "VF %u: pin handle=0x%llx offset=0x%llx size=0x%llx -> dma=0x%llx (emulation)\n",
+ vf->id, handle, offset, size, *dma_addr);
+
+ return 0;
+ }
+
+ ret = gsp_rpc_send_sync(vf->gsp->rpc, GSP_RPC_PIN,
+ &payload, sizeof(payload),
+ &resp, &resp_size);
+ if (ret)
+ return ret;
+
+ *dma_addr = resp.dma_addr;
+
+ dev_dbg(vf->gsp->dev, "VF %u: pin handle=0x%llx offset=0x%llx size=0x%llx -> dma=0x%llx\n",
+ vf->id, handle, offset, size, *dma_addr);
+
+ return 0;
+}
+
+int gsp_vf_mem_unpin(struct gsp_vf *vf, uint64_t handle)
+{
+ dev_dbg(vf->gsp->dev, "VF %u: unpin handle=0x%llx\n", vf->id, handle);
+ return 0;
+}
+
+/* Memory Mapping */
+int gsp_vf_mem_map(struct gsp_vf *vf, uint64_t handle, uint64_t gpu_va,
+ uint64_t size, uint32_t prot, uint32_t flags,
+ uint64_t offset, uint64_t *mapped_va)
+{
+ /* In emulation mode, just return the requested VA */
+ if (emulation_mode) {
+ if (mapped_va)
+ *mapped_va = gpu_va;
+
+ dev_dbg(vf->gsp->dev, "VF %u: map handle=0x%llx gpu_va=0x%llx size=0x%llx -> mapped_va=0x%llx (emulation)\n",
+ vf->id, handle, gpu_va, size, gpu_va);
+
+ return 0;
+ }
+
+ /* TODO: Implement GSP RPC for MAP */
+ dev_dbg(vf->gsp->dev, "VF %u: map handle=0x%llx gpu_va=0x%llx size=0x%llx\n",
+ vf->id, handle, gpu_va, size);
+
+ if (mapped_va)
+ *mapped_va = gpu_va;
+
+ return 0;
+}
+
+int gsp_vf_mem_unmap(struct gsp_vf *vf, uint64_t handle, uint64_t gpu_va, uint64_t size)
+{
+ dev_dbg(vf->gsp->dev, "VF %u: unmap handle=0x%llx gpu_va=0x%llx size=0x%llx\n",
+ vf->id, handle, gpu_va, size);
+ return 0;
+}
+
+/* Scheduling */
+int gsp_vf_schedule(struct gsp_vf *vf, uint32_t priority, uint64_t time_slice_ns)
+{
+ /* Use primary dispatcher SCHEDULE subcmd or secondary dispatcher */
+ dev_dbg(vf->gsp->dev, "VF %u: schedule prio=%u slice=%llu ns\n",
+ vf->id, priority, time_slice_ns);
+ return 0;
+}
+
+int gsp_vf_yield(struct gsp_vf *vf)
+{
+ dev_dbg(vf->gsp->dev, "VF %u: yield\n", vf->id);
+ return 0;
+} \ No newline at end of file