gsp-phys: capture GSP-RM NOCAT journal records from the status queue

1 files changed, 261 insertions, 8 deletions

diff --git a/kernel/gsp_phys_booter.c b/kernel/gsp_phys_booter.c
index aaf8f6d..19abbce 100644
--- a/kernel/gsp_phys_booter.c
+++ b/kernel/gsp_phys_booter.c
@@ -60,6 +60,11 @@ module_param_named(gsp_phys_boot_dryrun, gsp_phys_boot_dryrun_param, bool, 0644)
 MODULE_PARM_DESC(gsp_phys_boot_dryrun,
 		 "Prepare the stage-3d host-memory staging (radix3, queues, args) on the claimed GPU without any BAR0 write or falcon reset (default 0; read-only)");
 
+static bool gsp_phys_nocat_param;
+module_param_named(gsp_phys_nocat, gsp_phys_nocat_param, bool, 0644);
+MODULE_PARM_DESC(gsp_phys_nocat,
+		 "Log every GSP-RM NOCAT journal record (status-queue function 0x1020) individually during the stage-3e RM-init poll (default 0; the unique-record summary is always accumulated and dumped when the poll ends)");
+
 bool gsp_phys_boot_enabled(void)
 {
 	return gsp_phys_boot_param;
@@ -125,6 +130,7 @@ bool gsp_phys_boot_requested(void)
 #define GSP_PHYS_MSG_ELEMENT_RPC_RESULT_OFF	64
 #define GSP_PHYS_MSG_FUNC_GSP_INIT_DONE	0x1001
 #define GSP_PHYS_MSG_FUNC_RUN_CPU_SEQUENCER	0x1002
+#define GSP_PHYS_MSG_FUNC_GSP_POST_NOCAT_RECORD	0x1020
 
 /*
  * Status-queue element layout. The 48-byte element header
@@ -1862,6 +1868,204 @@ static void gsp_phys_rx_record_free(struct gsp_phys_rx_record *rec)
 	memset(rec, 0, sizeof(*rec));
 }
 
+/*
+ * GSP-RM NOCAT journal records (status-queue function 0x1020,
+ * NV_VGPU_MSG_EVENT_GSP_POST_NOCAT_RECORD). The RPC payload is the reference
+ * NV2080CtrlNocatJournalInsertRecord (ctrl2080nvd.h): a u64 timestamp, then
+ * two 65-byte strings around a packed u32 subsystem and an 8-byte-aligned u64
+ * errorCode, then a diagnostic buffer of up to 1 KB. Offsets below are
+ * relative to the RPC payload (element offset 80) and were verified by
+ * compiling the reference struct with NV_DECLARE_ALIGNED semantics and
+ * offsetof; the static asserts at the end of this file re-check them against
+ * the local re-declaration. In particular the 65-byte source ends at 89 and
+ * subsystem is padded to 92; errorCode is 8-byte aligned at 96.
+ *
+ * A producer may truncate a record (diagBufferLen, or a short element), so
+ * every field read is bounded by rec->payload_len: no byte is ever read past
+ * the received record, and the strings are copied into terminated buffers
+ * before they are printed.
+ */
+#define GSP_PHYS_NOCAT_OFF_RECTYPE	16
+#define GSP_PHYS_NOCAT_OFF_BUGCHECK	20
+#define GSP_PHYS_NOCAT_OFF_SOURCE	24
+#define GSP_PHYS_NOCAT_LEN_SOURCE	65
+#define GSP_PHYS_NOCAT_OFF_SUBSYSTEM	92
+#define GSP_PHYS_NOCAT_OFF_ERRORCODE	96
+#define GSP_PHYS_NOCAT_OFF_FAULTENGINE	104
+#define GSP_PHYS_NOCAT_OFF_TDRREASON	172
+#define GSP_PHYS_NOCAT_OFF_DIAGBUFLEN	176
+#define GSP_PHYS_NOCAT_OFF_DIAGBUF	180
+#define GSP_PHYS_NOCAT_MIN_PAYLOAD	GSP_PHYS_NOCAT_OFF_DIAGBUF
+#define GSP_PHYS_NOCAT_FULL_PAYLOAD	1208
+
+/* Local re-declaration of the reference record, only for the offset asserts. */
+struct gsp_phys_nocat_journal_record {
+	u32 flags;
+	u64 timestamp __attribute__((aligned(8)));
+	u8 rec_type;
+	u32 bugcheck;
+	char source[GSP_PHYS_NOCAT_LEN_SOURCE];
+	u32 subsystem;
+	u64 error_code __attribute__((aligned(8)));
+	char faulting_engine[GSP_PHYS_NOCAT_LEN_SOURCE];
+	u32 tdr_reason;
+	u32 diag_buffer_len;
+	u8 diag_buffer[1024];
+};
+
+/*
+ * Unique-record summary, keyed on recType/errorCode/subsystem/bugcheck plus
+ * the truncated source/faultingEngine strings. Static storage so the capture
+ * never adds a large frame; 48 slots then count overflow separately.
+ */
+#define GSP_PHYS_NOCAT_SUMMARY_SLOTS	48
+#define GSP_PHYS_NOCAT_SUMMARY_STR	48
+
+struct gsp_phys_nocat_slot {
+	u8 rec_type;
+	u64 error_code;
+	u32 subsystem;
+	u32 bugcheck;
+	u32 count;
+	u32 first_seq;
+	u32 last_seq;
+	char source[GSP_PHYS_NOCAT_SUMMARY_STR];
+	char faulting_engine[GSP_PHYS_NOCAT_SUMMARY_STR];
+};
+
+static struct gsp_phys_nocat_slot
+	gsp_phys_nocat_slots[GSP_PHYS_NOCAT_SUMMARY_SLOTS];
+static u32 gsp_phys_nocat_slot_count;
+static u32 gsp_phys_nocat_record_count;
+static u32 gsp_phys_nocat_overflow;
+
+static void gsp_phys_nocat_reset(void)
+{
+	memset(gsp_phys_nocat_slots, 0, sizeof(gsp_phys_nocat_slots));
+	gsp_phys_nocat_slot_count = 0;
+	gsp_phys_nocat_record_count = 0;
+	gsp_phys_nocat_overflow = 0;
+}
+
+/*
+ * Copy at most max_chars bytes of payload[off .. payload_len) into dst and
+ * NUL-terminate. A field outside the received payload yields an empty
+ * string; no read ever crosses payload_len.
+ */
+static void gsp_phys_nocat_copy_str(char *dst, u32 dst_size,
+				    const u8 *payload, u32 payload_len,
+				    u32 off, u32 max_chars)
+{
+	u32 n = 0;
+
+	if (off < payload_len) {
+		n = payload_len - off;
+		if (n > max_chars)
+			n = max_chars;
+	}
+	if (n >= dst_size)
+		n = dst_size - 1;
+	if (n)
+		memcpy(dst, payload + off, n);
+	dst[n] = '\0';
+}
+
+static void gsp_phys_nocat_accumulate(u8 rec_type, u64 error_code,
+				      u32 subsystem, u32 bugcheck,
+				      const char *source,
+				      const char *faulting_engine, u32 seq)
+{
+	struct gsp_phys_nocat_slot *slot;
+	u32 i;
+
+	gsp_phys_nocat_record_count++;
+	for (i = 0; i < gsp_phys_nocat_slot_count; i++) {
+		slot = &gsp_phys_nocat_slots[i];
+		if (slot->rec_type == rec_type &&
+		    slot->error_code == error_code &&
+		    slot->subsystem == subsystem &&
+		    slot->bugcheck == bugcheck &&
+		    strcmp(slot->source, source) == 0 &&
+		    strcmp(slot->faulting_engine, faulting_engine) == 0) {
+			slot->count++;
+			slot->last_seq = seq;
+			return;
+		}
+	}
+	if (gsp_phys_nocat_slot_count >= GSP_PHYS_NOCAT_SUMMARY_SLOTS) {
+		gsp_phys_nocat_overflow++;
+		return;
+	}
+	slot = &gsp_phys_nocat_slots[gsp_phys_nocat_slot_count++];
+	slot->rec_type = rec_type;
+	slot->error_code = error_code;
+	slot->subsystem = subsystem;
+	slot->bugcheck = bugcheck;
+	slot->count = 1;
+	slot->first_seq = seq;
+	slot->last_seq = seq;
+	strscpy(slot->source, source, sizeof(slot->source));
+	strscpy(slot->faulting_engine, faulting_engine,
+		sizeof(slot->faulting_engine));
+}
+
+static void gsp_phys_nocat_record(const struct gsp_phys_rx_record *rec)
+{
+	const u8 *p = rec->buf + GSP_PHYS_MSG_ELEMENT_PAYLOAD_OFF;
+	u32 len = rec->payload_len;
+	char source[GSP_PHYS_NOCAT_LEN_SOURCE];
+	char engine[GSP_PHYS_NOCAT_LEN_SOURCE];
+	u8 rec_type = 0;
+	u32 subsystem = 0;
+	u32 bugcheck = 0;
+	u64 error_code = 0;
+
+	if (len > rec->buf_len - GSP_PHYS_MSG_ELEMENT_PAYLOAD_OFF)
+		len = rec->buf_len - GSP_PHYS_MSG_ELEMENT_PAYLOAD_OFF;
+
+	if (len > GSP_PHYS_NOCAT_OFF_RECTYPE)
+		rec_type = p[GSP_PHYS_NOCAT_OFF_RECTYPE];
+	if (len >= GSP_PHYS_NOCAT_OFF_BUGCHECK + 4)
+		bugcheck = get_unaligned_le32(p + GSP_PHYS_NOCAT_OFF_BUGCHECK);
+	if (len >= GSP_PHYS_NOCAT_OFF_SUBSYSTEM + 4)
+		subsystem = get_unaligned_le32(p + GSP_PHYS_NOCAT_OFF_SUBSYSTEM);
+	if (len >= GSP_PHYS_NOCAT_OFF_ERRORCODE + 8)
+		error_code = get_unaligned_le64(p +
+						GSP_PHYS_NOCAT_OFF_ERRORCODE);
+	gsp_phys_nocat_copy_str(source, sizeof(source), p, len,
+				GSP_PHYS_NOCAT_OFF_SOURCE,
+				GSP_PHYS_NOCAT_LEN_SOURCE);
+	gsp_phys_nocat_copy_str(engine, sizeof(engine), p, len,
+				GSP_PHYS_NOCAT_OFF_FAULTENGINE,
+				GSP_PHYS_NOCAT_LEN_SOURCE);
+
+	gsp_phys_nocat_accumulate(rec_type, error_code, subsystem, bugcheck,
+				  source, engine, rec->seq_num);
+
+	if (gsp_phys_nocat_param)
+		pr_info("gsp-nocat: rec seq=%u recType=%u errorCode=0x%llx subsystem=0x%x bugcheck=0x%x payload=%u source=\"%s\" faultingEngine=\"%s\"\n",
+			rec->seq_num, rec_type, error_code, subsystem,
+			bugcheck, len, source, engine);
+}
+
+static void gsp_phys_nocat_summary_dump(void)
+{
+	u32 i;
+
+	pr_info("gsp-nocat: summary: %u unique, %u total, %u overflow\n",
+		gsp_phys_nocat_slot_count, gsp_phys_nocat_record_count,
+		gsp_phys_nocat_overflow);
+	for (i = 0; i < gsp_phys_nocat_slot_count; i++) {
+		const struct gsp_phys_nocat_slot *slot =
+			&gsp_phys_nocat_slots[i];
+
+		pr_info("gsp-nocat: entry[%u]: count=%u recType=%u errorCode=0x%llx subsystem=0x%x bugcheck=0x%x firstSeq=%u lastSeq=%u source=\"%s\" faultingEngine=\"%s\"\n",
+			i, slot->count, slot->rec_type, slot->error_code,
+			slot->subsystem, slot->bugcheck, slot->first_seq,
+			slot->last_seq, slot->source, slot->faulting_engine);
+	}
+}
+
 static void gsp_phys_status_queue_init(struct gsp_phys_status_queue *sq)
 {
 	memset(sq, 0, sizeof(*sq));
@@ -2158,11 +2362,18 @@ void gsp_phys_boot_rmwait(struct gsp_phys_boot *boot, void __iomem *bar0,
 	struct gsp_phys_status_queue sq;
 	unsigned long start_jiffies, deadline;
 
+	/*
+	 * The summary is per stage-3e poll: reset it on every entry and dump it
+	 * on every exit (success, transport error, GSP-RM failure or timeout)
+	 * so a failed boot always leaves the NOCAT record summary in dmesg.
+	 */
 	if (!boot || !boot_stage.ran) {
 		pr_warn("gsp-phys-boot: stage 3e skipped: stage 3d did not complete\n");
-		return;
+		gsp_phys_nocat_reset();
+		goto out;
 	}
 
+	gsp_phys_nocat_reset();
 	gsp_phys_status_queue_init(&sq);
 	start_jiffies = jiffies;
 	deadline = start_jiffies + msecs_to_jiffies(GSP_PHYS_RM_INIT_TIMEOUT_MS);
@@ -2184,7 +2395,7 @@ void gsp_phys_boot_rmwait(struct gsp_phys_boot *boot, void __iomem *bar0,
 				if (rret) {
 					pr_err("gsp-phys-boot: stage 3e incomplete: status-queue record walk failed (%d); stage 3d done, WPR2 teardown remains valid\n",
 					       rret);
-					return;
+					goto out;
 				}
 
 				pr_info("gsp-phys-boot: status queue element slot %u (seqNum=%u elemCount=%u): function=0x%x rpc_result=%u\n",
@@ -2192,6 +2403,9 @@ void gsp_phys_boot_rmwait(struct gsp_phys_boot *boot, void __iomem *bar0,
 					rec.elem_count, rec.function,
 					rec.rpc_result);
 				if (rec.function ==
+				    GSP_PHYS_MSG_FUNC_GSP_POST_NOCAT_RECORD)
+					gsp_phys_nocat_record(&rec);
+				if (rec.function ==
 				    GSP_PHYS_MSG_FUNC_RUN_CPU_SEQUENCER) {
 					u32 seq_buffer = 0, seq_index = 0;
 					u32 seq_cmds = 0, seq_off = 0;
@@ -2203,7 +2417,7 @@ void gsp_phys_boot_rmwait(struct gsp_phys_boot *boot, void __iomem *bar0,
 						pr_err("gsp-phys-boot: stage 3e incomplete: sequencer 0x%x arrived without a BAR0 window; stage 3d done, WPR2 teardown remains valid\n",
 						       rec.function);
 						gsp_phys_rx_record_free(&rec);
-						return;
+						goto out;
 					}
 					if (rec.rpc_length >
 					    GSP_PHYS_MSG_SIZE -
@@ -2214,7 +2428,7 @@ void gsp_phys_boot_rmwait(struct gsp_phys_boot *boot, void __iomem *bar0,
 						       GSP_PHYS_MSG_SIZE,
 						       rec.first_slot);
 						gsp_phys_rx_record_free(&rec);
-						return;
+						goto out;
 					}
 					sret = gsp_phys_seq_run_buffer(boot, bar0,
 							bar0_len,
@@ -2229,7 +2443,7 @@ void gsp_phys_boot_rmwait(struct gsp_phys_boot *boot, void __iomem *bar0,
 						       rec.first_slot,
 						       rec.seq_num);
 						gsp_phys_rx_record_free(&rec);
-						return;
+						goto out;
 					}
 					seq_payload = rec.buf + seq_off;
 					gsp_phys_fwsec_scratch14_read(bar0, &s14);
@@ -2258,13 +2472,13 @@ void gsp_phys_boot_rmwait(struct gsp_phys_boot *boot, void __iomem *bar0,
 								 start_jiffies),
 							sq.last_wp);
 						gsp_phys_rx_record_free(&rec);
-						return;
+						goto out;
 					}
 					pr_info("gsp-phys-boot: stage 3e incomplete: GSP-RM reported init failure (function=0x%x rpc_result=0x%x, seqNum=%u); stage 3d done, WPR2 teardown remains valid\n",
 						rec.function, rec.rpc_result,
 						rec.seq_num);
 					gsp_phys_rx_record_free(&rec);
-					return;
+					goto out;
 				}
 				gsp_phys_rx_record_free(&rec);
 			}
@@ -2275,10 +2489,13 @@ void gsp_phys_boot_rmwait(struct gsp_phys_boot *boot, void __iomem *bar0,
 				GSP_PHYS_MSG_FUNC_GSP_INIT_DONE,
 				GSP_PHYS_RM_INIT_TIMEOUT_MS,
 				sq.linked ? sq.last_wp : 0);
-			return;
+			goto out;
 		}
 		msleep(GSP_PHYS_RM_INIT_POLL_MS);
 	}
+
+out:
+	gsp_phys_nocat_summary_dump();
 }
 
 /* ============================================================
@@ -3473,3 +3690,39 @@ static_assert(GSP_PHYS_LOG_BUFFER_SIZE % GSP_PHYS_PAGE_SIZE == 0,
 static_assert(GSP_PHYS_LOG_PA_TABLE_OFF +
 	      (GSP_PHYS_LOG_BUFFER_PAGES + 1) * 8 <= GSP_PHYS_PAGE_SIZE,
 	      "libos log PA table (put pointer + one entry per page) must fit page 0");
+static_assert(offsetof(struct gsp_phys_nocat_journal_record, rec_type) ==
+	      GSP_PHYS_NOCAT_OFF_RECTYPE,
+	      "NOCAT recType must live at payload offset 16 (reference ABI)");
+static_assert(offsetof(struct gsp_phys_nocat_journal_record, bugcheck) ==
+	      GSP_PHYS_NOCAT_OFF_BUGCHECK,
+	      "NOCAT bugcheck must live at payload offset 20 (reference ABI)");
+static_assert(offsetof(struct gsp_phys_nocat_journal_record, source) ==
+	      GSP_PHYS_NOCAT_OFF_SOURCE,
+	      "NOCAT source[65] must live at payload offset 24 (reference ABI)");
+static_assert(offsetof(struct gsp_phys_nocat_journal_record, subsystem) ==
+	      GSP_PHYS_NOCAT_OFF_SUBSYSTEM,
+	      "NOCAT subsystem must live at payload offset 92: source[65] ends at 89 and is padded to the next u32 (reference ABI)");
+static_assert(offsetof(struct gsp_phys_nocat_journal_record, error_code) ==
+	      GSP_PHYS_NOCAT_OFF_ERRORCODE,
+	      "NOCAT errorCode u64 must be 8-byte aligned at payload offset 96 (reference ABI)");
+static_assert(offsetof(struct gsp_phys_nocat_journal_record, faulting_engine) ==
+	      GSP_PHYS_NOCAT_OFF_FAULTENGINE,
+	      "NOCAT faultingEngine[65] must live at payload offset 104 (reference ABI)");
+static_assert(offsetof(struct gsp_phys_nocat_journal_record, tdr_reason) ==
+	      GSP_PHYS_NOCAT_OFF_TDRREASON,
+	      "NOCAT tdrReason must live at payload offset 172 (reference ABI)");
+static_assert(offsetof(struct gsp_phys_nocat_journal_record,
+		       diag_buffer_len) == GSP_PHYS_NOCAT_OFF_DIAGBUFLEN,
+	      "NOCAT diagBufferLen must live at payload offset 176 (reference ABI)");
+static_assert(sizeof(struct gsp_phys_nocat_journal_record) ==
+	      GSP_PHYS_NOCAT_FULL_PAYLOAD,
+	      "the NOCAT journal record must be 1208 bytes including the 1 KB diag buffer (reference ABI)");
+static_assert(GSP_PHYS_NOCAT_OFF_SOURCE + GSP_PHYS_NOCAT_LEN_SOURCE <=
+	      GSP_PHYS_NOCAT_OFF_SUBSYSTEM,
+	      "NOCAT source must end before subsystem");
+static_assert(GSP_PHYS_NOCAT_OFF_FAULTENGINE + GSP_PHYS_NOCAT_LEN_SOURCE <=
+	      GSP_PHYS_NOCAT_OFF_TDRREASON,
+	      "NOCAT faultingEngine must end before tdrReason");
+static_assert(offsetof(struct gsp_phys_nocat_journal_record, diag_buffer) ==
+	      GSP_PHYS_NOCAT_OFF_DIAGBUF,
+	      "NOCAT diagBuffer must live at payload offset 180 (reference ABI)");