hwpmc: hwpmc: record page size to fix analysis in some case

This fixes a bug where a binary linked using max-page-size=0x200000
can result in a bogus relocation offset when running on a system
with a smaller page size.  This causes samples  to fall outside
the image mapping or be translated to the wrong address (resulting
in symbol resolution, or incorrect symbol resolution).  We noticed
this at Netflix because we run a patchset enabling 16k pages on
amd64 and have been compiling userspace with a 2MB page size.
Since we started doing this profiling userspace binaries has been
wonky.

Reviewed by: ali_mashtizadeh.com
Differential Revision: https://reviews.freebsd.org/D59771
Sponsored by: Netflix

lib/libpmc/pmclog.c

@@ -338,7 +338,7 @@ pmclog_get_event(void *cookie, char **data, ssize_t *len,
 	case PMCLOG_TYPE_MAP_IN:
 		PMCLOG_GET_PATHLEN(pathlen,evlen,pmclog_map_in);
 		PMCLOG_READ32(le,ev->pl_u.pl_mi.pl_pid);
-		PMCLOG_SKIP32(le);
+		PMCLOG_READ32(le,ev->pl_u.pl_mi.pl_u);
 		PMCLOG_READADDR(le,ev->pl_u.pl_mi.pl_start);
 		PMCLOG_READSTRING(le, ev->pl_u.pl_mi.pl_pathname, pathlen);
 		break;

lib/libpmc/pmclog.h

@@ -69,6 +69,13 @@ struct pmclog_ev_initialize {
 
 struct pmclog_ev_map_in {
 	pid_t		pl_pid;
+	union {
+		uint32_t	pl_u;
+		struct {
+			uint8_t	pl_pageshift;
+			uint8_t	pl_pad[3];
+		};
+	};
 	uintfptr_t	pl_start;
 	char		pl_pathname[PATH_MAX];
 };

lib/libpmcstat/libpmcstat.h

@@ -132,6 +132,7 @@ struct pmcstat_args {
 	int	pa_mergepmc;		/* merge PMC with same name */
 	double	pa_duration;		/* time duration */
 	uint32_t pa_tid;
+	uint8_t	pa_pageshift;		/* page shift recorded in log */
 	int	pa_argc;
 	char	**pa_argv;
 	STAILQ_HEAD(, pmcstat_ev) pa_events;

lib/libpmcstat/libpmcstat_image.c

@@ -312,7 +312,7 @@ pmcstat_image_get_elf_params(struct pmcstat_image *image,
 	const char *path, *elfbase;
 	char *p, *endp;
 	bool first_exec_segment;
-	uintfptr_t minva, maxva;
+	uintfptr_t minva, maxva, pagesize;
 	Elf *e;
 	Elf_Scn *scn;
 	GElf_Ehdr eh;
@@ -330,6 +330,11 @@ pmcstat_image_get_elf_params(struct pmcstat_image *image,
 	image->pi_dynlinkerpath = NULL;
 	image->pi_vaddr = 0;
 
+	if (args->pa_pageshift != 0)
+		pagesize = 1ULL << args->pa_pageshift;
+	else
+		pagesize = getpagesize();
+
 	path = pmcstat_string_unintern(image->pi_execpath);
 	assert(path != NULL);
 
@@ -421,7 +426,17 @@ pmcstat_image_get_elf_params(struct pmcstat_image *image,
 			case PT_LOAD:
 				if ((ph.p_flags & PF_X) != 0 &&
 				    first_exec_segment) {
-					image->pi_vaddr = ph.p_vaddr & (-ph.p_align);
+					if (image->pi_iskernelmodule) {
+						image->pi_vaddr = ph.p_vaddr &
+						    (-ph.p_align);
+					} else {
+						/*
+						 * User MAP_IN records depend
+						 * on page size.
+						 */
+						image->pi_vaddr = rounddown2(
+						    ph.p_vaddr, pagesize);
+					}
 					first_exec_segment = false;
 				}
 				break;

lib/libpmcstat/libpmcstat_logging.c

@@ -235,6 +235,7 @@ pmcstat_analyze_log(struct pmcstat_args *args,
 				    PMCSTAT_ALLOCATE);
 
 			assert(pp != NULL);
+			args->pa_pageshift = ev.pl_u.pl_mi.pl_pageshift;
 
 			image_path = pmcstat_string_intern(ev.pl_u.pl_mi.
 			    pl_pathname);

sys/dev/hwpmc/hwpmc_logging.c

@@ -947,6 +947,7 @@ void
 pmclog_process_map_in(struct pmc_owner *po, pid_t pid, uintfptr_t start,
     const char *path)
 {
+	struct pmclog_map_in *mi;
 	int pathlen, recordlen;
 
 	KASSERT(path != NULL, ("[pmclog,%d] map-in, null path", __LINE__));
@@ -956,8 +957,11 @@ pmclog_process_map_in(struct pmc_owner *po, pid_t pid, uintfptr_t start,
 	    pathlen;
 
 	PMCLOG_RESERVE(po, PMCLOG_TYPE_MAP_IN, recordlen);
+	mi = (struct pmclog_map_in *)ph;
 	PMCLOG_EMIT32(pid);
-	PMCLOG_EMIT32(0);
+	mi->pl_u = 0;
+	mi->pl_pageshift = PAGE_SHIFT;
+	PMCLOG_EMIT32(mi->pl_u);
 	PMCLOG_EMITADDR(start);
 	PMCLOG_EMITSTRING(path,pathlen);
 	PMCLOG_DESPATCH_SYNC(po);

sys/sys/pmclog.h

@@ -159,7 +159,13 @@ struct pmclog_initialize {
 struct pmclog_map_in {
 	PMCLOG_ENTRY_HEADER
 	uint32_t		pl_pid;
-	uint32_t		pl_pad;
+	union {
+		uint32_t	pl_u;
+		struct {
+			uint8_t	pl_pageshift;
+			uint8_t	pl_pad[3];
+		};
+	};
 	uintfptr_t		pl_start;	/* 8 byte aligned */
 	char			pl_pathname[PATH_MAX];
 } __packed;

usr.sbin/pmc/view.cc

@@ -89,8 +89,8 @@ syminfo::to_string(bool show_line)
 	return ss.str();
 }
 
-pmcview::pmcview() : tscfreq(0), pmcid(), pmcinfo(), procs(), tidtopid(),
-    images(), sysroot(""), filter()
+pmcview::pmcview() : tscfreq(0), pageshift(0), pmcid(), pmcinfo(), procs(),
+    tidtopid(), images(), sysroot(""), filter()
 {
 	char *root;
 
@@ -436,7 +436,7 @@ pmcview::process(struct pmclog_ev_threadexit &p)
  * if the dwarf symbols are available.
  */
 image
-pmcview::loadimage(const std::string &path)
+pmcview::loadimage(const std::string &path, bool iskernel)
 {
 	std::string fullpath;
 	image im;
@@ -519,7 +519,12 @@ pmcview::loadimage(const std::string &path)
 
 			if (ph.p_type == PT_LOAD) {
 				if ((ph.p_flags & PF_X) != 0 && !foundexec) {
-					im.vaddr = ph.p_vaddr & ~(ph.p_align - 1);
+					uint64_t alignment;
+
+					alignment = iskernel ? ph.p_align :
+					    (pageshift != 0 ? 1ULL << pageshift :
+					    (uint64_t)getpagesize());
+					im.vaddr = rounddown2(ph.p_vaddr, alignment);
 					foundexec = true;
 				}
 			}
@@ -727,8 +732,9 @@ pmcview::process(struct pmclog_ev_map_in &p)
 {
 	// Kernel map-in events should be mapped to pid 0
 	pid_t pid = (p.pl_pid == -1) ? 0 : p.pl_pid;
+	pageshift = p.pl_pageshift;
 
-	image im = loadimage(p.pl_pathname);
+	image im = loadimage(p.pl_pathname, pid == 0);
 
 	mapimage(pid, im, p.pl_start);
 }
@@ -958,4 +964,3 @@ pmcview::printvm(pid_t pid)
 		    i.second.highpc, i.second.image.c_str());
 	}
 }
-

usr.sbin/pmc/view.hh

@@ -389,6 +389,7 @@ protected:
 	void printvm(pid_t pid);
 	// Fields available to views
 	uint64_t				tscfreq;
+	uint8_t					pageshift;
 	std::unordered_map<uint32_t, uint32_t>	pmcid;
 	std::unordered_map<uint32_t, struct pmcinfo> pmcinfo;
 	std::unordered_map<pid_t, struct procinfo> procs;
@@ -404,7 +405,7 @@ protected:
 	std::vector<struct pmcinfox>		extpmcinfo;
 	std::map<uint32_t, struct cpuidleaf>	cpuid; // x86 Only
 private:
-	image loadimage(const std::string &path);
+	image loadimage(const std::string &path, bool iskernel = false);
 	void mapimage(pid_t pid, const image &im, uint64_t linkaddr);
 	void loadsymboltable(image *im, Elf *e, Elf_Scn *scn, GElf_Shdr *sh);
 	void loadsymbols(image *im);