From 81c11ca09adf85bd314f7406a8e0d7d637726a80 Mon Sep 17 00:00:00 2001 From: Arthur Hanson Rasmusson Date: Thu, 15 Oct 2026 17:53:52 -0400 Subject: vgsp-dual: stage B+C minimal per-mdev separation (G1-G3, G5, G6, G7, G8a) Goal 1.0: two mdevs + two QEMU guests running CUDA simultaneously. Single-mdev behaviour is kept identical: the legacy guest-RAM record still serves the first/only open mdev, the channel table degenerates to the historical key, and the session flush fires at the same last-channel-closed transition. - G6: guest-RAM backing {file, path, ranges} per mdev (legacy record + up to 4 vgpu-bound records); debugfs gsp-phys-guest-ram accepts an mdev UUID as an extra first token and keeps the 4-token form; records released on mdev cleanup and physical release. - G7: gsp_phys_gpa_to_hpa_vgpu(vgpu, gpa, *hpa); NULL keeps the historical first-open resolution. Forwarding operands pass the request's vgpu; the single-guest boot-ladder pin passes NULL. - G1: the sema-service thread services every open mdev each pass. - G2: channel table keyed by (vgpu, h_object); per-guest ranks; counter/scrub/opdef/pump snapshot+writeback matches per vgpu; scrubber records per guest. - G3: per-guest session records drive the flush; the shared learned tables (staged G4) are only flushed when no guest has live channels. - G5: hold/abort keyed by (vgpu, handle); dead-owner guards on the hold resolution paths; single-inflight serialization kept. - G8a: guest-GPA pin refused while more than one mdev is open. Build: make -j8 clean success (vgsp.ko). No module load. See DUAL-REPORT.md for the per-item status and live-validation plan. --- kernel/gsp_phys_gpu.h | 35 +++++++++++++++++++++++++++++++++-- 1 file changed, 33 insertions(+), 2 deletions(-) (limited to 'kernel/gsp_phys_gpu.h') diff --git a/kernel/gsp_phys_gpu.h b/kernel/gsp_phys_gpu.h index 73b0822..e2be7a8 100644 --- a/kernel/gsp_phys_gpu.h +++ b/kernel/gsp_phys_gpu.h @@ -3,6 +3,8 @@ #include +struct nv_vgpu_device; + /* * Physical GPU ownership (real-GSP execution backend, bring-up rung 1). * Claim is disabled unless the gsp_phys_gpu module parameter is set. @@ -428,21 +430,50 @@ bool gsp_phys_rmcall_vas_shared_value(void); */ int gsp_phys_gpa_to_hpa(u64 gpa, u64 *hpa_out); +/* + * Per-mdev variant: translate the GPA for the named owning vgpu (a captured + * forwarding request, a held release). The vgpu's own guest-RAM record is + * authoritative; the legacy record still serves the FIRST open mdev so + * single-guest callers are unchanged. NULL keeps the historical resolution + * (first open mdev). A named vgpu whose session is closed returns -ENODEV + * rather than being retargeted at another guest. + */ +int gsp_phys_gpa_to_hpa_vgpu(struct nv_vgpu_device *vgpu, u64 gpa, + u64 *hpa_out); + /* * Guest RAM backing-file configuration (debugfs gsp-phys-guest-ram). One - * shared file per module cycle, up to four GPA->file-offset ranges keyed by + * record per guest session, up to four GPA->file-offset ranges keyed by * gpa_base. The first configure opens the file; the same path adds/replaces * ranges, a different path is refused -EBUSY, and the ranges are dropped and - * the file closed by gsp_phys_guest_ram_clear() on physical-GPU release. The + * the file closed by gsp_phys_guest_ram_clear() on physical-GPU release (or + * gsp_phys_guest_ram_vgpu_clear() when one mdev is removed). The + * gsp_phys_guest_ram_configure() entry point configures the legacy record + * (the historical top-level debugfs write: serves the first/only open mdev); + * gsp_phys_guest_ram_configure_vgpu() binds the record to one mdev. The * state accessor is lock-safe and writes "active=0" or the active path, * range count and per-range gpa/off/size into the caller's buffer. */ int gsp_phys_guest_ram_configure(const char *path, u64 gpa_base, u64 file_offset, u64 size); +int gsp_phys_guest_ram_configure_vgpu(struct nv_vgpu_device *vgpu, + const char *path, u64 gpa_base, + u64 file_offset, u64 size); void gsp_phys_guest_ram_clear(void); +void gsp_phys_guest_ram_vgpu_clear(struct nv_vgpu_device *vgpu); bool gsp_phys_guest_ram_active(void); void gsp_phys_guest_ram_state(char *buf, size_t len); +/* + * Open-mdev helpers for the per-mdev debugfs selector and the single-guest + * guards: count the mdevs whose VFIO device is open, or resolve one open + * mdev by its device name (the mdev UUID). The returned device is not + * reference-counted; the caller must only use it while that mdev's fd is + * known to be open. + */ +int gsp_phys_open_mdev_count(void); +struct nv_vgpu_device *gsp_phys_open_mdev_by_name(const char *name); + /* * Domain shape of the physical (claimed) device, as the P6a engine path * needs it: the module programs raw host PAs into the CE page tables, so the -- cgit v1.2.3