vgsp: report FB info sizes in KB and use 0 for unsupported context buffers

The FB info ABI documents RAM_SIZE/TOTAL_RAM_SIZE/USABLE_RAM_SIZE/HEAP_SIZE/
MAPPABLE_HEAP_SIZE/HEAP_FREE/BAR1_SIZE/BAR1_AVAIL_SIZE in kilobytes; the
emulation returned bytes, over-reporting the emulated framebuffer by 1024x
to the X DDX.  COMPRESSION_SIZE and L2CACHE_SIZE remain in bytes.

KGR_GET_CONTEXT_BUFFERS_INFO reported NV_U32_MAX for buffer kinds Turing
does not support.  kgraphics treats size==0 as unsupported (skip); an
NV_U32_MAX byte length rounds up to exactly 4 GiB in the page-count
computation, which is the shape of the transient guest sysmem allocations.
Report 0/0 instead.

Evidence: handoff-meta-plan.md CONT.435 (c15-c21 allocation traces, the
guest nvidia.ko disassembly of osAllocPagesInternal), kernel 570.144
ctrl2080fb.h units and kernel_graphics.c:1775-1800.

1 files changed, 23 insertions, 9 deletions

diff --git a/kernel/gsp_rm_emulation.c b/kernel/gsp_rm_emulation.c
index 9d9f351..581ed19 100644
--- a/kernel/gsp_rm_emulation.c
+++ b/kernel/gsp_rm_emulation.c
@@ -2892,11 +2892,16 @@ static void gsp_rm_handle_kgr_get_context_buffers_info(
 
 	for (e = 0; e < TU104_CTX_GPU_MAX_ENGINES; e++) {
 		for (i = 0; i < TU104_CTX_ENGINE_ID_COUNT; i++) {
-			/* engine IDs the chip does not support report
-			 * NV_U32_MAX (the client skips those entries) */
+			/* r1781/CONT.435: use 0 (not NV_U32_MAX) for buffer
+			 * kinds this chip does not support.  kgraphics treats
+			 * size==0 as "not supported" and skips the buffer
+			 * (kernel_graphics.c:1785); an NV_U32_MAX byte length
+			 * would instead round up to exactly 4 GiB in the
+			 * page-count computation.  The reserve-size sum also
+			 * guards on NV_U32_MAX, so 0 is both safe and honest. */
 			if (sizes[i].size == 0) {
-				info->engineContextBuffersInfo[e].engine[i].size = 0xFFFFFFFF;
-				info->engineContextBuffersInfo[e].engine[i].alignment = 0xFFFFFFFF;
+				info->engineContextBuffersInfo[e].engine[i].size = 0;
+				info->engineContextBuffersInfo[e].engine[i].alignment = 0;
 				continue;
 			}
 			info->engineContextBuffersInfo[e].engine[i].size = sizes[i].size;
@@ -2948,15 +2953,24 @@ static u32 tu104_fb_info_data(struct gsp_emu_context *ctx, u32 index)
 		return gsp_rm_fb_compr_mb << 20;
 	case TU104_FB_INFO_IDX_PARTITION_COUNT:
 		return 1;
+	/*
+	 * r1781/CONT.435: every size index below is documented in KB
+	 * (ctrl2080fb.h: RAM_SIZE/TOTAL_RAM_SIZE/USABLE_RAM_SIZE/HEAP_SIZE/
+	 * MAPPABLE_HEAP_SIZE/HEAP_FREE/BAR1_SIZE/BAR1_AVAIL_SIZE are all
+	 * "in kilobytes").  Returning the byte count over-reported the
+	 * emulated FB size by 1024x to the X DDX (FbTotalSizeKb =
+	 * 0x20000000 = 512 GB).  COMPRESSION_SIZE and L2CACHE_SIZE stay
+	 * in bytes per the same header.
+	 */
 	case TU104_FB_INFO_IDX_BAR1_SIZE:
-		return (u32)fb;
+		return (u32)(fb >> 10);
 	case TU104_FB_INFO_IDX_RAM_SIZE:
 	case TU104_FB_INFO_IDX_TOTAL_RAM_SIZE:
 	case TU104_FB_INFO_IDX_USABLE_RAM_SIZE:
-		return (u32)fb;
+		return (u32)(fb >> 10);
 	case TU104_FB_INFO_IDX_HEAP_SIZE:
 	case TU104_FB_INFO_IDX_MAPPABLE_HEAP_SIZE:
-		return (u32)fb;
+		return (u32)(fb >> 10);
 	case TU104_FB_INFO_IDX_BUS_WIDTH:
 		return 256;
 	case TU104_FB_INFO_IDX_RAM_TYPE:
@@ -2969,7 +2983,7 @@ static u32 tu104_fb_info_data(struct gsp_emu_context *ctx, u32 index)
 	case TU104_FB_INFO_IDX_PARTITION_MASK:
 		return 1;
 	case TU104_FB_INFO_IDX_HEAP_FREE:
-		return (u32)fb;
+		return (u32)(fb >> 10);
 	case TU104_FB_INFO_IDX_RAM_LOCATION:
 		return TU104_RAM_LOCATION_GPU_DEDICATED;
 	case TU104_FB_INFO_IDX_FBP_COUNT:
@@ -2982,7 +2996,7 @@ static u32 tu104_fb_info_data(struct gsp_emu_context *ctx, u32 index)
 		return TU104_MEMORYINFO_VENDOR_ID_SAMSUNG;
 	case TU104_FB_INFO_IDX_BAR1_AVAIL_SIZE:
 	case TU104_FB_INFO_IDX_BAR1_MAX_CONTIG_AVAIL_SIZE:
-		return (u32)fb;
+		return (u32)(fb >> 10);
 	case TU104_FB_INFO_IDX_LTC_COUNT:
 		return TU104_LTC_COUNT;
 	case TU104_FB_INFO_IDX_LTS_COUNT: