aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authoruser <user@clank>2026-09-25 18:07:33 +0200
committeruser <user@clank>2026-09-25 18:07:33 +0200
commitcbfea8d37b012cd52942e7f580342b2dc6129b29 (patch)
tree91a024b309f4ad7525aa9d38805622a8039b5cc3
parentIdentify RPC 0x1a5 as the display-presence command and drive cmd 0x0F over th... (diff)
downloadamd-psp-re-cbfea8d37b012cd52942e7f580342b2dc6129b29.tar.gz
amd-psp-re-cbfea8d37b012cd52942e7f580342b2dc6129b29.tar.xz
amd-psp-re-cbfea8d37b012cd52942e7f580342b2dc6129b29.zip
Locate the exact HDCP message id (0x10) that drives the display-presence RPC
sub_4020E0, which cmd 0x0F calls twice in every ring-driven run (return sites 0x40231e and 0x402352), gates on its second argument as a protocol message id: subs r6,r1,#1 / cmp r6,#0x12 / bhi -> must be 0x81 else exit / tbb [pc,r6]. Decoded that table (base 0x4020FE, ids 1..0x13) against the call site 0x40218c that reaches 0x403990 -> 0x404604 -> RPC 0x1a5. Message id 0x10's handler starts at 0x402186 and the call is six bytes later, so HDCP protocol message 0x10 is what drives RPC 0x1a5 and therefore what tells ASD a display is present. In every run so far r1 is neither 1..0x13 nor 0x81, so sub_4020E0 exits at 0x402126 immediately. r1 is loaded from the stack at 0x40233E, derived from the request rather than from the TA subcommand field this harness has been varying, which explains why sweeping subcommands 0..19 never moved it. The full chain is now identified end to end -- ring cmd 0x0F -> sub_4020E0 -> message 0x10 -> 0x403990 -> RPC 0x1a5 -> ASD sub_D0FEAC -> presence flag -> RPC 0x1a0 stops failing -> cmd 0x0E reaches RPC 0x103 -> ASD reads DDC via svc 0x7b, where the sink model attaches -- with exactly one unknown left: which request field supplies r1. That is a request-layout question with no hardware dependency.
-rw-r--r--psp_out/ANGLE-DISPLAY-SINK-INPUT-SURFACE.md44
1 files changed, 44 insertions, 0 deletions
diff --git a/psp_out/ANGLE-DISPLAY-SINK-INPUT-SURFACE.md b/psp_out/ANGLE-DISPLAY-SINK-INPUT-SURFACE.md
index 0447ad8..89265d7 100644
--- a/psp_out/ANGLE-DISPLAY-SINK-INPUT-SURFACE.md
+++ b/psp_out/ANGLE-DISPLAY-SINK-INPUT-SURFACE.md
@@ -899,3 +899,47 @@ dispatcher over the ring, and some subcommand path from there leads through
Stage 2 cmd 0x0E -> RPC 0x1a0 -> ASD DONE
Stage 2b cmd 0x0F -> subcommand dispatch DONE (slot 1/2, sub 0)
Stage 3 presence set, sink model driven NOT YET
+
+## The exact message id located: HDCP protocol message 0x10
+
+`sub_4020E0` — the function cmd 0x0F calls twice (return sites `0x40231e` and
+`0x402352` appear in every ring-driven run) — gates on its second argument, an
+HDCP protocol message id:
+
+ 004020f4 subs r6, r1, #1
+ 004020f6 cmp r6, #0x12
+ 004020f8 bhi 0x402116 ; not 1..0x13 -> must be 0x81, else exit
+ 004020fa tbb [pc, r6]
+
+Decoding that table (base `0x4020FE`, ids 1..0x13) against the call site
+`0x40218c` that reaches `0x403990 -> 0x404604 (RPC 0x1a5)`:
+
+ id 0x09 -> 0x402156 id 0x0b -> 0x402162 id 0x0e -> 0x40216e
+ id 0x0f -> 0x40217a id 0x10 -> 0x402186 id 0x13 -> 0x402192
+
+`id 0x10`'s handler starts at `0x402186` and the call at `0x40218c` is six bytes
+later, so **HDCP protocol message `0x10` is what drives RPC `0x1a5`**, i.e. what
+tells ASD a display is present.
+
+In every ring-driven run so far `r1` is neither in 1..0x13 nor `0x81`, so
+`sub_4020E0` exits at `0x402126` immediately. `r1` is loaded from the stack at
+`0x40233E` (`ldr r1,[sp,#0x268]`), i.e. it is derived from the request, not from
+the TA subcommand field this harness has been varying — which is why sweeping
+subcommands 0..19 never moved it.
+
+### Complete chain, with only one unknown left
+
+ host ring INVOKE cmd 0x0F [DRIVABLE]
+ -> sub_4021A0 -> sub_4020E0 [REACHED, 2x]
+ -> message id 0x10 selects handler 0x402186 [UNKNOWN: which
+ request field becomes r1]
+ -> 0x403990 -> wrapper 0x404604 -> RPC 0x1a5 [identified]
+ -> ASD handler 0xd01742 -> sub_D0FEAC [identified]
+ -> display-presence flag [0xD42890+(i*3)*0x20+4] = 1
+ -> RPC 0x1a0 stops returning 5
+ -> cmd 0x0E proceeds to RPC 0x103 (gate B)
+ -> ASD reads DDC via svc 0x7b -> SINK MODEL ATTACHES HERE
+
+Everything in that chain is identified except which request field supplies `r1`.
+That is the single remaining unknown between this emulator and an end-to-end
+sink test, and it is a pure request-layout question with no hardware dependency.