Add support to capture kernel performance counters for different
kernel level operations. These counters collects the information
for remote call and copies the information to a buffer shared
by user.

Collection of DSP performance counters is also added as part of
this change. DSP updates the performance information in the
metadata which is then copied to a buffer passed by the users.

Signed-off-by: Ekansh Gupta <quic_ekangupt@quicinc.com>
---
 drivers/misc/fastrpc.c      | 143 ++++++++++++++++++++++++++++++++++--
 include/uapi/misc/fastrpc.h |   4 +-
 2 files changed, 139 insertions(+), 8 deletions(-)

diff --git a/drivers/misc/fastrpc.c b/drivers/misc/fastrpc.c
index 8e817a763d1d..54a562fc94fb 100644
--- a/drivers/misc/fastrpc.c
+++ b/drivers/misc/fastrpc.c
@@ -19,6 +19,7 @@
 #include <linux/rpmsg.h>
 #include <linux/scatterlist.h>
 #include <linux/slab.h>
+#include <linux/delay.h>
 #include <linux/firmware/qcom/qcom_scm.h>
 #include <uapi/misc/fastrpc.h>
 #include <linux/of_reserved_mem.h>
@@ -38,6 +39,7 @@
 #define FASTRPC_CTX_MAX (256)
 #define FASTRPC_INIT_HANDLE	1
 #define FASTRPC_DSP_UTILITIES_HANDLE	2
+#define FASTRPC_MAX_STATIC_HANDLE (20)
 #define FASTRPC_CTXID_MASK (0xFF0)
 #define INIT_FILELEN_MAX (2 * 1024 * 1024)
 #define INIT_FILE_NAMELEN_MAX (128)
@@ -106,6 +108,9 @@
 
 #define miscdev_to_fdevice(d) container_of(d, struct fastrpc_device, miscdev)
 
+#define FASTRPC_KERNEL_PERF_LIST (PERF_KEY_MAX)
+#define FASTRPC_DSP_PERF_LIST 12
+
 static const char *domains[FASTRPC_DEV_MAX] = { "adsp", "mdsp",
 						"sdsp", "cdsp", "cdsp1" };
 struct fastrpc_phy_page {
@@ -229,6 +234,18 @@ struct fastrpc_map {
 	struct kref refcount;
 };
 
+struct fastrpc_perf {
+	u64 count;
+	u64 flush;
+	u64 map;
+	u64 copy;
+	u64 link;
+	u64 getargs;
+	u64 putargs;
+	u64 invargs;
+	u64 invoke;
+};
+
 struct fastrpc_invoke_ctx {
 	int nscalars;
 	int nbufs;
@@ -237,6 +254,8 @@ struct fastrpc_invoke_ctx {
 	int tgid;
 	u32 sc;
 	u32 *crc;
+	u64 *perf_kernel;
+	u64 *perf_dsp;
 	u64 ctxid;
 	u64 msg_sz;
 	struct kref refcount;
@@ -251,6 +270,7 @@ struct fastrpc_invoke_ctx {
 	struct fastrpc_invoke_args *args;
 	struct fastrpc_buf_overlap *olaps;
 	struct fastrpc_channel_ctx *cctx;
+	struct fastrpc_perf *perf;
 };
 
 struct fastrpc_session_ctx {
@@ -308,6 +328,51 @@ struct fastrpc_user {
 	struct mutex mutex;
 };
 
+enum fastrpc_perfkeys {
+	PERF_COUNT = 0,
+	PERF_FLUSH = 1,
+	PERF_MAP = 2,
+	PERF_COPY = 3,
+	PERF_LINK = 4,
+	PERF_GETARGS = 5,
+	PERF_PUTARGS = 6,
+	PERF_INVARGS = 7,
+	PERF_INVOKE = 8,
+	PERF_KEY_MAX = 9,
+};
+
+#define PERF_END ((void)0)
+
+#define PERF(enb, cnt, ff) \
+	{\
+		struct timespec64 startT = {0};\
+		u64 *counter = cnt;\
+		if (enb && counter) {\
+			ktime_get_boottime_ts64(&startT);\
+		} \
+		ff ;\
+		if (enb && counter) {\
+			*counter += getnstimediff(&startT);\
+		} \
+	}
+
+#define GET_COUNTER(perf_ptr, offset)  \
+	(perf_ptr != NULL ?\
+		(((offset >= 0) && (offset < PERF_KEY_MAX)) ?\
+			(u64 *)(perf_ptr + offset)\
+				: (u64 *)NULL) : (u64 *)NULL)
+
+static inline s64 getnstimediff(struct timespec64 *start)
+{
+	s64 ns;
+	struct timespec64 ts, b;
+
+	ktime_get_boottime_ts64(&ts);
+	b = timespec64_sub(ts, *start);
+	ns = timespec64_to_ns(&b);
+	return ns;
+}
+
 static void fastrpc_free_map(struct kref *ref)
 {
 	struct fastrpc_map *map;
@@ -497,6 +562,7 @@ static void fastrpc_context_free(struct kref *ref)
 	idr_remove(&cctx->ctx_idr, ctx->ctxid >> 4);
 	spin_unlock_irqrestore(&cctx->lock, flags);
 
+	kfree(ctx->perf);
 	kfree(ctx->maps);
 	kfree(ctx->olaps);
 	kfree(ctx);
@@ -614,6 +680,13 @@ static struct fastrpc_invoke_ctx *fastrpc_context_alloc(
 	fastrpc_channel_ctx_get(cctx);
 
 	ctx->crc = (u32 *)(uintptr_t)inv2->crc;
+	ctx->perf_dsp = (u64 *)(uintptr_t)inv2->perf_dsp;
+	ctx->perf_kernel = (u64 *)(uintptr_t)inv2->perf_kernel;
+	if (ctx->perf_kernel) {
+		ctx->perf = kzalloc(sizeof(*(ctx->perf)), GFP_KERNEL);
+		if (!ctx->perf)
+			return ERR_PTR(-ENOMEM);
+	}
 	ctx->sc = sc;
 	ctx->retval = -1;
 	ctx->pid = current->pid;
@@ -877,7 +950,8 @@ static int fastrpc_get_meta_size(struct fastrpc_invoke_ctx *ctx)
 		sizeof(struct fastrpc_invoke_buf) +
 		sizeof(struct fastrpc_phy_page)) * ctx->nscalars +
 		sizeof(u64) * FASTRPC_MAX_FDLIST +
-		sizeof(u32) * FASTRPC_MAX_CRCLIST;
+		sizeof(u32) * FASTRPC_MAX_CRCLIST +
+		sizeof(u32) + sizeof(u64) * FASTRPC_DSP_PERF_LIST;
 
 	return size;
 }
@@ -951,7 +1025,9 @@ static int fastrpc_get_args(u32 kernel, struct fastrpc_invoke_ctx *ctx)
 	metalen = fastrpc_get_meta_size(ctx);
 	pkt_size = fastrpc_get_payload_size(ctx, metalen);
 
+	PERF(ctx->perf_kernel, GET_COUNTER((u64 *)ctx->perf, PERF_MAP),
 	err = fastrpc_create_maps(ctx);
+	PERF_END);
 	if (err)
 		return err;
 
@@ -987,6 +1063,7 @@ static int fastrpc_get_args(u32 kernel, struct fastrpc_invoke_ctx *ctx)
 			continue;
 
 		if (ctx->maps[i]) {
+			PERF(ctx->perf_kernel, GET_COUNTER((u64 *)ctx->perf, PERF_MAP),
 			struct vm_area_struct *vma = NULL;
 
 			rpra[i].buf.pv = (u64) ctx->args[i].ptr;
@@ -1003,9 +1080,9 @@ static int fastrpc_get_args(u32 kernel, struct fastrpc_invoke_ctx *ctx)
 			pg_end = ((ctx->args[i].ptr + len - 1) & PAGE_MASK) >>
 				  PAGE_SHIFT;
 			pages[i].size = (pg_end - pg_start + 1) * PAGE_SIZE;
-
+			PERF_END);
 		} else {
-
+			PERF(ctx->perf_kernel, GET_COUNTER((u64 *)ctx->perf, PERF_COPY),
 			if (ctx->olaps[oix].offset == 0) {
 				rlen -= ALIGN(args, FASTRPC_ALIGN) - args;
 				args = ALIGN(args, FASTRPC_ALIGN);
@@ -1027,12 +1104,14 @@ static int fastrpc_get_args(u32 kernel, struct fastrpc_invoke_ctx *ctx)
 			pages[i].size = (pg_end - pg_start + 1) * PAGE_SIZE;
 			args = args + mlen;
 			rlen -= mlen;
+			PERF_END);
 		}
 
 		if (i < inbufs && !ctx->maps[i]) {
 			void *dst = (void *)(uintptr_t)rpra[i].buf.pv;
 			void *src = (void *)(uintptr_t)ctx->args[i].ptr;
 
+			PERF(ctx->perf_kernel, GET_COUNTER((u64 *)ctx->perf, PERF_COPY),
 			if (!kernel) {
 				if (copy_from_user(dst, (void __user *)src,
 						   len)) {
@@ -1042,6 +1121,7 @@ static int fastrpc_get_args(u32 kernel, struct fastrpc_invoke_ctx *ctx)
 			} else {
 				memcpy(dst, src, len);
 			}
+			PERF_END);
 		}
 	}
 
@@ -1072,9 +1152,9 @@ static int fastrpc_put_args(struct fastrpc_invoke_ctx *ctx,
 	struct fastrpc_map *mmap = NULL;
 	struct fastrpc_invoke_buf *list;
 	struct fastrpc_phy_page *pages;
-	u64 *fdlist;
-	u32 *crclist;
-	int i, inbufs, outbufs, handles;
+	u64 *fdlist, *perf_dsp_list;
+	u32 *crclist, *poll;
+	int i, inbufs, outbufs, handles, perferr;
 
 	inbufs = REMOTE_SCALARS_INBUFS(ctx->sc);
 	outbufs = REMOTE_SCALARS_OUTBUFS(ctx->sc);
@@ -1083,6 +1163,8 @@ static int fastrpc_put_args(struct fastrpc_invoke_ctx *ctx,
 	pages = fastrpc_phy_page_start(list, ctx->nscalars);
 	fdlist = (uint64_t *)(pages + inbufs + outbufs + handles);
 	crclist = (u32 *)(fdlist + FASTRPC_MAX_FDLIST);
+	poll = (u32 *)(crclist + FASTRPC_MAX_CRCLIST);
+	perf_dsp_list = (u64 *)(poll + 1);
 
 	for (i = inbufs; i < ctx->nbufs; ++i) {
 		if (!ctx->maps[i]) {
@@ -1113,6 +1195,14 @@ static int fastrpc_put_args(struct fastrpc_invoke_ctx *ctx,
 			return -EFAULT;
 	}
 
+	if (ctx->perf_dsp && perf_dsp_list) {
+		perferr = copy_to_user((void __user *)ctx->perf_dsp,
+				      perf_dsp_list,
+				      FASTRPC_DSP_PERF_LIST * sizeof(u64));
+		if (perferr)
+			dev_dbg(fl->sctx->dev, "Warning: failed to copy perf data %d\n", perferr);
+	}
+
 	return 0;
 }
 
@@ -1148,13 +1238,35 @@ static int fastrpc_invoke_send(struct fastrpc_session_ctx *sctx,
 
 }
 
+static void fastrpc_update_invoke_count(u32 handle, u64 *perf_counter,
+					struct timespec64 *invoket)
+{
+	/* update invoke count for dynamic handles */
+	u64 *invcount, *count;
+
+	if (handle <= FASTRPC_MAX_STATIC_HANDLE)
+		return;
+
+	invcount = GET_COUNTER(perf_counter, PERF_INVOKE);
+	if (invcount)
+		*invcount += getnstimediff(invoket);
+
+	count = GET_COUNTER(perf_counter, PERF_COUNT);
+	if (count)
+		*count++;
+}
+
 static int fastrpc_internal_invoke(struct fastrpc_user *fl,  u32 kernel, struct fastrpc_invoke_v2 *inv2)
 {
 	struct fastrpc_invoke_ctx *ctx = NULL;
 	struct fastrpc_buf *buf, *b;
 	struct fastrpc_invoke inv;
 	u32 handle, sc;
-	int err = 0;
+	int err = 0, perferr = 0;
+	struct timespec64 invoket = {0};
+
+	if (inv2->perf_kernel)
+		ktime_get_boottime_ts64(&invoket);
 
 	if (!fl->sctx)
 		return -EINVAL;
@@ -1174,14 +1286,18 @@ static int fastrpc_internal_invoke(struct fastrpc_user *fl,  u32 kernel, struct
 	if (IS_ERR(ctx))
 		return PTR_ERR(ctx);
 
+	PERF(ctx->perf_kernel, GET_COUNTER((u64 *)ctx->perf, PERF_GETARGS),
 	err = fastrpc_get_args(kernel, ctx);
+	PERF_END);
 	if (err)
 		goto bail;
 
 	/* make sure that all CPU memory writes are seen by DSP */
 	dma_wmb();
+	PERF(ctx->perf_kernel, GET_COUNTER((u64 *)ctx->perf, PERF_LINK),
 	/* Send invoke buffer to remote dsp */
 	err = fastrpc_invoke_send(fl->sctx, ctx, kernel, handle);
+	PERF_END);
 	if (err)
 		goto bail;
 
@@ -1197,8 +1313,10 @@ static int fastrpc_internal_invoke(struct fastrpc_user *fl,  u32 kernel, struct
 
 	/* make sure that all memory writes by DSP are seen by CPU */
 	dma_rmb();
+	PERF(ctx->perf_kernel, GET_COUNTER((u64 *)ctx->perf, PERF_PUTARGS),
 	/* populate all the output buffers with results */
 	err = fastrpc_put_args(ctx, kernel);
+	PERF_END);
 	if (err)
 		goto bail;
 
@@ -1209,6 +1327,17 @@ static int fastrpc_internal_invoke(struct fastrpc_user *fl,  u32 kernel, struct
 
 bail:
 	if (err != -ERESTARTSYS && err != -ETIMEDOUT) {
+		if (ctx) {
+			if (ctx->perf_kernel && !err)
+				fastrpc_update_invoke_count(handle, (u64 *)ctx->perf, &invoket);
+			if (ctx->perf_kernel && ctx->perf && ctx->perf_kernel) {
+				perferr = copy_to_user((void __user *)ctx->perf_kernel,
+							ctx->perf,
+							FASTRPC_KERNEL_PERF_LIST * sizeof(u64));
+				if (perferr)
+					dev_dbg(fl->sctx->dev, "Warning: failed to copy perf data %d\n", perferr);
+			}
+		}
 		/* We are done with this compute context */
 		spin_lock(&fl->lock);
 		list_del(&ctx->node);
diff --git a/include/uapi/misc/fastrpc.h b/include/uapi/misc/fastrpc.h
index 406b80555d41..1edc7c04b171 100644
--- a/include/uapi/misc/fastrpc.h
+++ b/include/uapi/misc/fastrpc.h
@@ -84,7 +84,9 @@ struct fastrpc_invoke {
 struct fastrpc_invoke_v2 {
 	struct fastrpc_invoke inv;
 	__u64 crc;
-	__u32 reserved[16];
+	__u64 perf_kernel;
+	__u64 perf_dsp;
+	__u32 reserved[12];
 };
 
 struct fastrpc_init_create {
-- 
2.34.1

On Mon, Oct 07, 2024 at 02:15:16PM +0530, Ekansh Gupta wrote:
> Add support to capture kernel performance counters for different
> kernel level operations. These counters collects the information
> for remote call and copies the information to a buffer shared
> by user.
> 
> Collection of DSP performance counters is also added as part of
> this change. DSP updates the performance information in the
> metadata which is then copied to a buffer passed by the users.

"also" usually means "should be broken up into a new patch", right?

Also, where is any of these new "performance counters" documented?  Why
not use the normal kernel apis for performance counters?

thanks,

greg k-h