fix: Steam Controller (Triton) output/feature reports are no longer overwritten unread (#1032)

This commit is contained in:
Resul Övür authored and GitHub committed 2026-10-21 23:55:00 -04:00
1 parent 6b2ca398e2
commit a863ca7b31
2 files changed
+30 -9

No files matched your search

+18 -3
View File
@@ -1184,8 +1184,9 @@ struct TritonSharedData {
uint32_t input_length;
uint8_t input[64];
uint64_t output_seq;
uint32_t output_feature;
uint32_t output_length;
uint16_t output_feature;
uint16_t output_length;
uint32_t output_ack;
uint8_t output[64];
uint64_t get_request;
uint64_t get_reply;
@@ -1202,6 +1203,8 @@ static_assert(offsetof(TritonSharedData, report_id) == 16, "Triton report ID off
static_assert(offsetof(TritonSharedData, input) == 24, "Triton input offset");
static_assert(offsetof(TritonSharedData, output_seq) == 88, "Triton output sequence offset");
static_assert(offsetof(TritonSharedData, output_feature) == 96, "Triton output feature offset");
static_assert(offsetof(TritonSharedData, output_length) == 98, "Triton output length offset");
static_assert(offsetof(TritonSharedData, output_ack) == 100, "Triton output ack offset");
static_assert(offsetof(TritonSharedData, output) == 104, "Triton output offset");
static_assert(offsetof(TritonSharedData, get_request) == 168, "Triton feature request offset");
static_assert(offsetof(TritonSharedData, get_reply) == 176, "Triton feature reply offset");
@@ -1219,6 +1222,8 @@ static constexpr int TRITON_REPORT_ID_BYTES = 1;
static constexpr int TRITON_STATE_BYTES = 45;
static constexpr int TRITON_REPORT_INTERVAL_US = 4000;
static constexpr int TRITON_FEATURE_TIMEOUT_MS = 500;
// The app clears output_length once it has taken a report; a writer waits this long for that before overwriting.
static constexpr int TRITON_OUTPUT_TIMEOUT_MS = 50;
static constexpr const char *TRITON_NAME = "Steam Controller";
static constexpr const char *TRITON_SERIAL = "DROIDDECKSC01";
static const uint8_t kTritonReportDescriptor[] = {
@@ -2574,11 +2579,21 @@ __attribute__((visibility("hidden"))) static void triton_queue_output(DeckHidraw
if (!data || !size || size > 64 || triton.mapping_size < TRITON_SHARED_OFFSET + TRITON_SHARED_BLOCK_SIZE) return;
auto *shared = reinterpret_cast<TritonSharedData *>(reinterpret_cast<uint8_t *>(triton.ring) + TRITON_SHARED_OFFSET);
if (__atomic_load_n(&shared->magic, __ATOMIC_ACQUIRE) != TRITON_SHARED_MAGIC) return;
static std::mutex one_at_a_time;
std::lock_guard<std::mutex> guard(one_at_a_time);
for (int waited = 0;
static_cast<uint32_t>(__atomic_load_n(&shared->output_seq, __ATOMIC_ACQUIRE)) !=
__atomic_load_n(&shared->output_ack, __ATOMIC_ACQUIRE) &&
__atomic_load_n(&shared->connected, __ATOMIC_ACQUIRE) != 0 && waited < TRITON_OUTPUT_TIMEOUT_MS * 4;
waited++) {
struct timespec interval = {0, 250 * 1000L};
nanosleep(&interval, nullptr);
}
uint64_t seq = __atomic_load_n(&shared->output_seq, __ATOMIC_RELAXED);
if (seq & 1) seq++;
__atomic_store_n(&shared->output_seq, seq + 1, __ATOMIC_RELEASE);
shared->output_feature = feature ? 1 : 0;
shared->output_length = static_cast<uint32_t>(size);
shared->output_length = static_cast<uint16_t>(size);
memcpy(shared->output, data, size);
if (size < sizeof(shared->output)) memset(shared->output + size, 0, sizeof(shared->output) - size);
__atomic_thread_fence(__ATOMIC_RELEASE);
@@ -99,7 +99,8 @@ public class FakeInputWriter {
private static final int TRITON_REPORT_OFFSET = TRITON_OFFSET + 24;
private static final int TRITON_OUTPUT_SEQ_OFFSET = TRITON_OFFSET + 88;
private static final int TRITON_OUTPUT_FEATURE_OFFSET = TRITON_OFFSET + 96;
private static final int TRITON_OUTPUT_LENGTH_OFFSET = TRITON_OFFSET + 100;
private static final int TRITON_OUTPUT_LENGTH_OFFSET = TRITON_OFFSET + 98;
private static final int TRITON_OUTPUT_ACK_OFFSET = TRITON_OFFSET + 100;
private static final int TRITON_OUTPUT_OFFSET = TRITON_OFFSET + 104;
private static final int TRITON_GET_REQUEST_OFFSET = TRITON_OFFSET + 168;
private static final int TRITON_GET_REPLY_OFFSET = TRITON_OFFSET + 176;
@@ -229,8 +230,9 @@ public class FakeInputWriter {
data.putInt(TRITON_REPORT_ID_OFFSET, 0);
data.putInt(TRITON_REPORT_LENGTH_OFFSET, 0);
data.putLong(TRITON_OUTPUT_SEQ_OFFSET, 0L);
data.putInt(TRITON_OUTPUT_FEATURE_OFFSET, 0);
data.putInt(TRITON_OUTPUT_LENGTH_OFFSET, 0);
data.putShort(TRITON_OUTPUT_FEATURE_OFFSET, (short) 0);
data.putShort(TRITON_OUTPUT_LENGTH_OFFSET, (short) 0);
data.putInt(TRITON_OUTPUT_ACK_OFFSET, 0);
data.putLong(TRITON_GET_REQUEST_OFFSET, 0L);
data.putLong(TRITON_GET_REPLY_OFFSET, 0L);
}
@@ -587,14 +589,18 @@ public class FakeInputWriter {
nativeStoreFence();
long seq = ring.getLong(TRITON_OUTPUT_SEQ_OFFSET);
if ((seq & 1) != 0 || seq <= lastSequence) return null;
int length = ring.getInt(TRITON_OUTPUT_LENGTH_OFFSET);
int length = ring.getShort(TRITON_OUTPUT_LENGTH_OFFSET) & 0xffff;
if (length <= 0 || length > TRITON_REPORT_MAX) return null;
byte[] output = new byte[9 + length];
ByteBuffer.wrap(output).order(ByteOrder.LITTLE_ENDIAN).putLong(seq);
output[8] = (byte) (ring.getInt(TRITON_OUTPUT_FEATURE_OFFSET) != 0 ? 1 : 0);
output[8] = (byte) (ring.getShort(TRITON_OUTPUT_FEATURE_OFFSET) != 0 ? 1 : 0);
for (int i = 0; i < length; i++) output[9 + i] = ring.get(TRITON_OUTPUT_OFFSET + i);
nativeStoreFence();
return seq == ring.getLong(TRITON_OUTPUT_SEQ_OFFSET) ? output : null;
if (seq != ring.getLong(TRITON_OUTPUT_SEQ_OFFSET)) return null;
// Acknowledge the exact report sequence taken so a concurrent timeout overwrite cannot lose its length.
ring.putInt(TRITON_OUTPUT_ACK_OFFSET, (int) seq);
nativeStoreFence();
return output;
}
}