From 70e4e560b55e7155e6103601f078945c1d13c615 Mon Sep 17 00:00:00 2001 From: Arthur Hanson Rasmusson Date: Mon, 16 Nov 2026 01:14:56 -0500 Subject: vgsp: eng77 - capture and replay the client's VERTEX_STREAM_TURING CALL(2) The beef 728-byte push carries CALL_MME_MACRO(2) (method 0x3810, golden slot 2 = VERTEX_STREAM_TURING) with payload [0,0,0x0450e100]: the Mesa mme_tu104_sim of golden slot 2 with those args emits SET_VERTEX_STREAM_A_LOCATION 0x1c04<-0, 0x1c08<-0, and the Turing size pair 0x600<-0, 0x604<- - the VTG stream state the per-frame DRAW window slices omit. The census (eng76, 0x100..0x1ff) finds it at 0x148 of the 728 push; the 247-word window carries the relocated call at 0x2e0 (d2=0x1e22100), the 151-word DRAW window does not. - service: sticky svc_gl_vs_call[3] from every 0x3810 run (VS-CALL receipt), getter gsp_sema_gl_vs_call(). - booter release_save: when the window has no 0x3810 run, prepend the captured ONE_INCR run (0x3810<-payload) ahead of the DRAW window; the normal patch relocates the VA payload (sema_gl_mme_reloc). - vs_pre buffer 12->24 words. Census range stays 0x100..0x1ff (eng76). --- kernel/gsp_phys_booter.c | 50 ++++++++++++++++++++++++++++++++++++++++++++++- kernel/gsp_sema_service.c | 39 ++++++++++++++++++++++++++++++++++++ kernel/gsp_sema_service.h | 1 + 3 files changed, 89 insertions(+), 1 deletion(-) diff --git a/kernel/gsp_phys_booter.c b/kernel/gsp_phys_booter.c index 10fe9c3..3a8bb55 100644 --- a/kernel/gsp_phys_booter.c +++ b/kernel/gsp_phys_booter.c @@ -19295,6 +19295,28 @@ static u32 gsp_phys_gr_vs_scan(const struct gsp_phys_gr_submit *req, return 0; } +/* Does the window carry the VERTEX_STREAM_TURING CALL(2) run (0x3810)? */ +static bool gsp_phys_gr_mme_window_has_call2(const u32 *win, u32 n) +{ + u32 r = 0; + + while (r < n) { + u32 h = win[r]; + u32 op = (h >> 29) & 7u; + u32 cnt = (h >> 16) & 0x1fffu; + u32 span = 1u; + u32 m0 = (h & 0xfffu) << 2; + + if ((op == 1u || op == 3u || op == 5u) && cnt && + r + 1u + cnt <= n) + span = 1u + cnt; + if (m0 == 0x3810u) + return true; + r += span; + } + return false; +} + /* Does the window already carry the Turing stream-size run (0x600)? */ static bool gsp_phys_gr_mme_window_has_vs_size(const u32 *win, u32 n) { @@ -19425,7 +19447,7 @@ static void gsp_phys_gr_mme_release_save(const struct gsp_phys_gr_submit *req, bool draw_ia, keep = false; const char *prefer = NULL; u32 old_h = 0, old_win = 0; - u32 vs_pre[12]; + u32 vs_pre[24]; u32 vs_n = 0; static u32 prefer_log; @@ -19567,6 +19589,32 @@ static void gsp_phys_gr_mme_release_save(const struct gsp_phys_gr_submit *req, pr_info("vgsp-gr: MME-VS-SIZE synth j=0 size=0x100000 (window has VS) h=0x%x push=0x%llx\n", req->h_object, req->push_va); } + /* + * eng77: the client's VERTEX_STREAM_TURING macro + * (CALL_MME_MACRO(2) = 0x3810) payload carries the + * VTG stream size/base (the 0x0450e100-family VA). + * The DRAW-window slices omit it; replay the last + * captured call before the DRAW_IAB. + */ + { + u32 vcall[3]; + + if (!gsp_phys_gr_mme_window_has_call2( + req->stream, req->stream_n) && + gsp_sema_gl_vs_call(vcall) && + vs_n + 4u <= ARRAY_SIZE(vs_pre)) { + vs_pre[vs_n++] = (5u << 29) | + (3u << 16) | + ((0x3810u >> 2) & + 0xfffu); + vs_pre[vs_n++] = vcall[0]; + vs_pre[vs_n++] = vcall[1]; + vs_pre[vs_n++] = vcall[2]; + pr_info("vgsp-gr: MME-VS-CALL synth 0x3810 payload=%08x,%08x,%08x h=0x%x push=0x%llx\n", + vcall[0], vcall[1], vcall[2], + req->h_object, req->push_va); + } + } } if (req->stream_n + vs_n > GSP_PHYS_GR_MME_RELEASE_WIN_MAX) vs_n = 0; diff --git a/kernel/gsp_sema_service.c b/kernel/gsp_sema_service.c index 2b49d24..499247d 100644 --- a/kernel/gsp_sema_service.c +++ b/kernel/gsp_sema_service.c @@ -9801,6 +9801,26 @@ static u8 svc_gl_shader_type[32] = { static u64 svc_gl_vs_base[8]; static u8 svc_gl_vs_hi[8]; /* LOCATION_A upper byte per stream */ static u32 svc_gl_vs_log; +/* + * eng77: the client's VERTEX_STREAM_TURING macro call (CALL_MME_MACRO(2) + * = 0x3810) whose payload carries the VTG vertex-stream state (the + * 0x0450e100-family VA). The per-frame DRAW window slices exclude this + * call; capture the last observed payload (3 words) so the booter can + * replay it before the DRAW_IAB. The census scans every beef push. + */ +static u32 svc_gl_vs_call[3]; +static u8 svc_gl_vs_call_ok; +static u32 svc_gl_vs_call_log; + +bool gsp_sema_gl_vs_call(u32 out[3]) +{ + if (!out || !READ_ONCE(svc_gl_vs_call_ok)) + return false; + out[0] = READ_ONCE(svc_gl_vs_call[0]); + out[1] = READ_ONCE(svc_gl_vs_call[1]); + out[2] = READ_ONCE(svc_gl_vs_call[2]); + return true; +} u64 gsp_sema_gl_vs_base(u32 j) { @@ -18329,6 +18349,25 @@ static void svc_mme_census(const u8 *buf, u32 len, u64 push_va, u32 h_object) memcpy(&v1, buf + off + 8, 4); if (off + 16 <= len) memcpy(&v2, buf + off + 12, 4); + /* + * eng77: remember the VERTEX_STREAM_TURING + * CALL_MME_MACRO(2) payload (the VTG stream state + * the DRAW window slices omit). + */ + if (addr == 0x3810u && off + 16 <= len) { + bool changed = !READ_ONCE(svc_gl_vs_call_ok) || + READ_ONCE(svc_gl_vs_call[2]) != v2; + + WRITE_ONCE(svc_gl_vs_call[0], v0); + WRITE_ONCE(svc_gl_vs_call[1], v1); + WRITE_ONCE(svc_gl_vs_call[2], v2); + WRITE_ONCE(svc_gl_vs_call_ok, 1); + if (changed && svc_gl_vs_call_log < 8u) { + svc_gl_vs_call_log++; + pr_info("sema-gl: VS-CALL 0x3810 payload=%08x,%08x,%08x\n", + v0, v1, v2); + } + } lc += scnprintf(line + lc, sizeof(line) - lc, "%s0x%x@0x%x(%08x,%08x,%08x)", nhit ? "," : "", addr, off, v0, v1, v2); diff --git a/kernel/gsp_sema_service.h b/kernel/gsp_sema_service.h index 926f81e..4a739da 100644 --- a/kernel/gsp_sema_service.h +++ b/kernel/gsp_sema_service.h @@ -629,6 +629,7 @@ bool gsp_sema_gr_surf_recs(u64 *va, u32 *n); */ u64 gsp_sema_mme_sa_get(u32 *entries, u32 n); u64 gsp_sema_gl_vs_base(u32 j); +bool gsp_sema_gl_vs_call(u32 out[3]); /* * Golden SA completion (sema_gl_mme_sa_golden_fill): set the host-side -- cgit v1.2.3