Skip to contentSkip to conversation

Various PCL fixes

PATCH In review Anton Blanchard 3 participants 21 messages Jun 30, 2009 – Jul 1, 2009

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

cover Various PCL fixes

Here are a few fixes for PCL I found when testing on ppc64.

#1 lore raw

PATCH 1/8 perf report: Fix -z option

Fix a copy and paste error, -z was setting the group option.

---

tools/perf/builtin-top.c +1 -1
Index: linux-2.6-tip/tools/perf/builtin-top.c
===================================================================
@@ -675,7 +675,7 @@
 			    "put the counters into a counter group"),
 	OPT_STRING('s', "sym-filter", &sym_filter, "pattern",
 		    "only display symbols matchig this pattern"),
-	OPT_BOOLEAN('z', "zero", &group,
+	OPT_BOOLEAN('z', "zero", &zero,
 		    "zero history across updates"),
 	OPT_INTEGER('F', "freq", &freq,
 		    "profile at this frequency"),
#2 lore raw

[tip:perfcounters/urgent] perf report: Fix -z option

Commit-ID:  1f208ea67821703fd4de056ea6f0baa81f4ad4a5
Gitweb:     http://git.kernel.org/tip/1f208ea67821703fd4de056ea6f0baa81f4ad4a5
Author:     Anton Blanchard <[email protected]>
AuthorDate: Wed, 1 Jul 2009 09:00:44 +1000
Committer:  Ingo Molnar <[email protected]>
CommitDate: Wed, 1 Jul 2009 01:25:18 +0200

perf report: Fix -z option

Fix a copy and paste error, -z was setting the group option.

LKML-Reference: <[email protected]>

---
 tools/perf/builtin-top.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
tools/perf/builtin-top.c +1 -1
index cf0d21f..5c29655 100644
@@ -675,7 +675,7 @@ static const struct option options[] = {
 			    "put the counters into a counter group"),
 	OPT_STRING('s', "sym-filter", &sym_filter, "pattern",
 		    "only display symbols matchig this pattern"),
-	OPT_BOOLEAN('z', "zero", &group,
+	OPT_BOOLEAN('z', "zero", &zero,
 		    "zero history across updates"),
 	OPT_INTEGER('F', "freq", &freq,
 		    "profile at this frequency"),
#3 lore raw

PATCH 2/8 perf_counter tools: Remove zlib dependency

In reply to Anton Blanchard

The zlib devel libraries may not be installed and since we aren't using zlib we may as well remove it.

---

tools/perf/Makefile +0 -6
Index: linux-2.6-tip/tools/perf/Makefile
===================================================================
@@ -379,12 +379,6 @@
 	endif
 endif
 
-ifdef ZLIB_PATH
-	BASIC_CFLAGS += -I$(ZLIB_PATH)/include
-	EXTLIBS += -L$(ZLIB_PATH)/$(lib) $(CC_LD_DYNPATH)$(ZLIB_PATH)/$(lib)
-endif
-EXTLIBS += -lz
-
 ifdef NEEDS_SOCKET
 	EXTLIBS += -lsocket
 endif
#4 lore raw

[tip:perfcounters/urgent] perf_counter tools: Remove zlib dependency

Commit-ID:  6717534ddc328ae5cdf89f1ef802db83fc451f19
Gitweb:     http://git.kernel.org/tip/6717534ddc328ae5cdf89f1ef802db83fc451f19
Author:     Anton Blanchard <[email protected]>
AuthorDate: Wed, 1 Jul 2009 09:00:45 +1000
Committer:  Ingo Molnar <[email protected]>
CommitDate: Wed, 1 Jul 2009 01:25:18 +0200

perf_counter tools: Remove zlib dependency

The zlib devel libraries may not be installed and since we aren't using zlib we may as well remove it.

LKML-Reference: <[email protected]>

---
 tools/perf/Makefile |    6 ------
 1 files changed, 0 insertions(+), 6 deletions(-)
tools/perf/Makefile +0 -6
index 9c6d0ae..f572c90 100644
@@ -381,12 +381,6 @@ ifndef CC_LD_DYNPATH
 	endif
 endif
 
-ifdef ZLIB_PATH
-	BASIC_CFLAGS += -I$(ZLIB_PATH)/include
-	EXTLIBS += -L$(ZLIB_PATH)/$(lib) $(CC_LD_DYNPATH)$(ZLIB_PATH)/$(lib)
-endif
-EXTLIBS += -lz
-
 ifdef NEEDS_SOCKET
 	EXTLIBS += -lsocket
 endif
#5 lore raw

PATCH 3/8 perf top: Move skip symbols to an array

In reply to Anton Blanchard

Move the list of symbols we skip into an array, making it easier to add new ones.

---

tools/perf/builtin-top.c +17 -7
Index: linux-2.6-tip/tools/perf/builtin-top.c
===================================================================
@@ -286,11 +286,22 @@
 	return NULL;
 }
 
+/* Tag samples to be skipped. */
+char *skip_symbols[] = {
+	"default_idle",
+	"cpu_idle",
+	"enter_idle",
+	"exit_idle",
+	"mwait_idle",
+	NULL
+};
+
 static int symbol_filter(struct dso *self, struct symbol *sym)
 {
 	static int filter_match;
 	struct sym_entry *syme;
 	const char *name = sym->name;
+	int i;
 
 	if (!strcmp(name, "_text") ||
 	    !strcmp(name, "_etext") ||
@@ -302,13 +313,12 @@
 		return 1;
 
 	syme = dso__sym_priv(self, sym);
-	/* Tag samples to be skipped. */
-	if (!strcmp("default_idle", name) ||
-	    !strcmp("cpu_idle", name) ||
-	    !strcmp("enter_idle", name) ||
-	    !strcmp("exit_idle", name) ||
-	    !strcmp("mwait_idle", name))
-		syme->skip = 1;
+	for (i = 0; skip_symbols[i]; i++) {
+		if (!strcmp(skip_symbols[i], name)) {
+			syme->skip = 1;
+			break;
+		}
+	}
 
 	if (filter_match == 1) {
 		filter_end = sym->start;
#6 lore raw

[tip:perfcounters/urgent] perf top: Move skip symbols to an array

Commit-ID:  2ab52083ffc057014e502cf3473adc41436922fa
Gitweb:     http://git.kernel.org/tip/2ab52083ffc057014e502cf3473adc41436922fa
Author:     Anton Blanchard <[email protected]>
AuthorDate: Wed, 1 Jul 2009 09:00:46 +1000
Committer:  Ingo Molnar <[email protected]>
CommitDate: Wed, 1 Jul 2009 01:25:19 +0200

perf top: Move skip symbols to an array

Move the list of symbols we skip into an array, making it easier to add new ones.

LKML-Reference: <[email protected]>

---
 tools/perf/builtin-top.c |   24 +++++++++++++++++-------
 1 files changed, 17 insertions(+), 7 deletions(-)
tools/perf/builtin-top.c +17 -7
index 5c29655..731ec6d 100644
@@ -286,11 +286,22 @@ static void *display_thread(void *arg)
 	return NULL;
 }
 
+/* Tag samples to be skipped. */
+char *skip_symbols[] = {
+	"default_idle",
+	"cpu_idle",
+	"enter_idle",
+	"exit_idle",
+	"mwait_idle",
+	NULL
+};
+
 static int symbol_filter(struct dso *self, struct symbol *sym)
 {
 	static int filter_match;
 	struct sym_entry *syme;
 	const char *name = sym->name;
+	int i;
 
 	if (!strcmp(name, "_text") ||
 	    !strcmp(name, "_etext") ||
@@ -302,13 +313,12 @@ static int symbol_filter(struct dso *self, struct symbol *sym)
 		return 1;
 
 	syme = dso__sym_priv(self, sym);
-	/* Tag samples to be skipped. */
-	if (!strcmp("default_idle", name) ||
-	    !strcmp("cpu_idle", name) ||
-	    !strcmp("enter_idle", name) ||
-	    !strcmp("exit_idle", name) ||
-	    !strcmp("mwait_idle", name))
-		syme->skip = 1;
+	for (i = 0; skip_symbols[i]; i++) {
+		if (!strcmp(skip_symbols[i], name)) {
+			syme->skip = 1;
+			break;
+		}
+	}
 
 	if (filter_match == 1) {
 		filter_end = sym->start;
#7 lore raw

Re: [patch 3/8] perf top: Move skip symbols to an array

In reply to Anton Blanchard

HI Anton,

tools/perf/builtin-top.c line 290

Quoted 3 lines · On Wed, 01 Jul 2009 09:00:46 +1000 Anton Blanchard <[email protected]> wrote:
tools/perf/builtin-top.c +2 -0
+/* Tag samples to be skipped. */
+char *skip_symbols[] = {

"static const" ?

Signature · 3 lines
Cheers,
Stephen Rothwell                    [email protected]
http://www.canb.auug.org.au/~sfr/
#8 lore raw

Re: [patch 3/8] perf top: Move skip symbols to an array

Quoted 8 lines · Stephen Rothwell

HI Anton,

tools/perf/builtin-top.c line 290

Quoted 3 lines · On Wed, 01 Jul 2009 09:00:46 +1000 Anton Blanchard <[email protected]> wrote:
tools/perf/builtin-top.c +2 -0
+/* Tag samples to be skipped. */
+char *skip_symbols[] = {

"static const" ?

I fixed this. Too bad there's no GCC warning for that. (Sparse catches is but it is not enabled yet for perf.)

Ingo

#9 lore raw

PATCH 4/8 perf top: Add ppc64 specific skip symbols and strip ppc64 . prefix

In reply to Anton Blanchard

Filter out some ppc64 specific idle loop functions and remove leading '.' on ppc64 text symbols.

---

Do we want to wrap them with #ifdef __PPC__ ?

tools/perf/builtin-top.c +9 -0
Index: linux-2.6-tip/tools/perf/builtin-top.c
===================================================================
@@ -293,6 +293,8 @@
 	"enter_idle",
 	"exit_idle",
 	"mwait_idle",
+	"ppc64_runlatch_off",
+	"pseries_dedicated_idle_sleep",
 	NULL
 };
 
@@ -303,6 +305,13 @@
 	const char *name = sym->name;
 	int i;
 
+	/*
+	 * ppc64 uses function descriptors and appends a '.' to the
+	 * start of every instruction address. Remove it.
+	 */
+	if (name[0] == '.')
+		name++;
+
 	if (!strcmp(name, "_text") ||
 	    !strcmp(name, "_etext") ||
 	    !strcmp(name, "_sinittext") ||
#10 lore raw

Re: [patch 4/8] perf top: Add ppc64 specific skip symbols and strip ppc64 . prefix

tools/perf/builtin-top.c line 297

Quoted 13 lines · Anton Blanchard

Filter out some ppc64 specific idle loop functions and remove leading '.' on ppc64 text symbols.

---

Do we want to wrap them with #ifdef __PPC__ ?

tools/perf/builtin-top.c +2 -0
 	"enter_idle",
 	"exit_idle",
 	"mwait_idle",
+	"ppc64_runlatch_off",
+	"pseries_dedicated_idle_sleep",

We only want to wrap them into #ifdef __PPC__ if we do so for the x86 ones too.

I think it's fine as-is - the names should be pretty unique, right?

Ingo

#11 lore raw

Re: [patch 4/8] perf top: Add ppc64 specific skip symbols and strip ppc64 . prefix

Hi Ingo,

Quoted 11 lines

tools/perf/builtin-top.c line 297

Quoted 7 lines

Do we want to wrap them with #ifdef __PPC__ ?

tools/perf/builtin-top.c +2 -0
 	"enter_idle",
 	"exit_idle",
 	"mwait_idle",
+	"ppc64_runlatch_off",
+	"pseries_dedicated_idle_sleep",

We only want to wrap them into #ifdef __PPC__ if we do so for the x86 ones too.

Quoted 1 line

I think it's fine as-is - the names should be pretty unique, right?

I agree, we can start wrapping them if things get out of hand.

Anton

#12 lore raw

[tip:perfcounters/urgent] perf top: Add ppc64 specific skip symbols and strip ppc64 . prefix

In reply to Anton Blanchard
Commit-ID:  3a3393ef75a14ae259a82f3f38624efa17884168
Gitweb:     http://git.kernel.org/tip/3a3393ef75a14ae259a82f3f38624efa17884168
Author:     Anton Blanchard <[email protected]>
AuthorDate: Wed, 1 Jul 2009 09:00:47 +1000
Committer:  Ingo Molnar <[email protected]>
CommitDate: Wed, 1 Jul 2009 01:25:19 +0200

perf top: Add ppc64 specific skip symbols and strip ppc64 . prefix

Filter out some ppc64 specific idle loop functions and remove leading '.' on ppc64 text symbols.

LKML-Reference: <[email protected]>

---
 tools/perf/builtin-top.c |    9 +++++++++
 1 files changed, 9 insertions(+), 0 deletions(-)
tools/perf/builtin-top.c +9 -0
index 731ec6d..0506cd6 100644
@@ -293,6 +293,8 @@ char *skip_symbols[] = {
 	"enter_idle",
 	"exit_idle",
 	"mwait_idle",
+	"ppc64_runlatch_off",
+	"pseries_dedicated_idle_sleep",
 	NULL
 };
 
@@ -303,6 +305,13 @@ static int symbol_filter(struct dso *self, struct symbol *sym)
 	const char *name = sym->name;
 	int i;
 
+	/*
+	 * ppc64 uses function descriptors and appends a '.' to the
+	 * start of every instruction address. Remove it.
+	 */
+	if (name[0] == '.')
+		name++;
+
 	if (!strcmp(name, "_text") ||
 	    !strcmp(name, "_etext") ||
 	    !strcmp(name, "_sinittext") ||
#13 lore raw

PATCH 5/8 perf report: Fix reporting of hypervisor

In reply to Anton Blanchard

PERF_EVENT_MISC_* is not a bitmask, so we have to mask and compare.

---

tools/perf/builtin-report.c +5 -2
Index: linux-2.6-tip/tools/perf/builtin-report.c
===================================================================
@@ -1210,6 +1210,7 @@
 	struct map *map = NULL;
 	void *more_data = event->ip.__more_data;
 	struct ip_callchain *chain = NULL;
+	int cpumode;
 
 	if (sample_type & PERF_SAMPLE_PERIOD) {
 		period = *(u64 *)more_data;
@@ -1250,7 +1251,9 @@
 		return -1;
 	}
 
-	if (event->header.misc & PERF_EVENT_MISC_KERNEL) {
+	cpumode = event->header.misc & PERF_EVENT_MISC_CPUMODE_MASK;
+
+	if (cpumode == PERF_EVENT_MISC_KERNEL) {
 		show = SHOW_KERNEL;
 		level = 'k';
 
@@ -1258,7 +1261,7 @@
 
 		dprintf(" ...... dso: %s\n", dso->name);
 
-	} else if (event->header.misc & PERF_EVENT_MISC_USER) {
+	} else if (cpumode == PERF_EVENT_MISC_USER) {
 
 		show = SHOW_USER;
 		level = '.';
#14 lore raw

[tip:perfcounters/urgent] perf report: Fix reporting of hypervisor

Commit-ID:  d8db1b57d31a6b30ea2f0df318eab50fc92b38d6
Gitweb:     http://git.kernel.org/tip/d8db1b57d31a6b30ea2f0df318eab50fc92b38d6
Author:     Anton Blanchard <[email protected]>
AuthorDate: Wed, 1 Jul 2009 09:00:48 +1000
Committer:  Ingo Molnar <[email protected]>
CommitDate: Wed, 1 Jul 2009 01:25:20 +0200

perf report: Fix reporting of hypervisor

PERF_EVENT_MISC_* is not a bitmask, so we have to mask and compare.

LKML-Reference: <[email protected]>

---
 tools/perf/builtin-report.c |    7 +++++--
 1 files changed, 5 insertions(+), 2 deletions(-)
tools/perf/builtin-report.c +5 -2
index 135b783..88e88c5 100644
@@ -1213,6 +1213,7 @@ process_sample_event(event_t *event, unsigned long offset, unsigned long head)
 	struct map *map = NULL;
 	void *more_data = event->ip.__more_data;
 	struct ip_callchain *chain = NULL;
+	int cpumode;
 
 	if (sample_type & PERF_SAMPLE_PERIOD) {
 		period = *(u64 *)more_data;
@@ -1256,7 +1257,9 @@ process_sample_event(event_t *event, unsigned long offset, unsigned long head)
 	if (comm_list && !strlist__has_entry(comm_list, thread->comm))
 		return 0;
 
-	if (event->header.misc & PERF_EVENT_MISC_KERNEL) {
+	cpumode = event->header.misc & PERF_EVENT_MISC_CPUMODE_MASK;
+
+	if (cpumode == PERF_EVENT_MISC_KERNEL) {
 		show = SHOW_KERNEL;
 		level = 'k';
 
@@ -1264,7 +1267,7 @@ process_sample_event(event_t *event, unsigned long offset, unsigned long head)
 
 		dprintf(" ...... dso: %s\n", dso->name);
 
-	} else if (event->header.misc & PERF_EVENT_MISC_USER) {
+	} else if (cpumode == PERF_EVENT_MISC_USER) {
 
 		show = SHOW_USER;
 		level = '.';
#15 lore raw

PATCH 6/8 perf report: Add hypervisor dso

In reply to Anton Blanchard

Add a dso for hypervisor samples. We don't get any symbol information on the ppc64 hypervisor but this at least gives us a high level summary of the time spent in there.

---

tools/perf/builtin-report.c +14 -1
Index: linux-2.6-tip/tools/perf/builtin-report.c
===================================================================
@@ -118,6 +118,7 @@
 static LIST_HEAD(dsos);
 static struct dso *kernel_dso;
 static struct dso *vdso;
+static struct dso *hypervisor_dso;
 
 static void dsos__add(struct dso *dso)
 {
@@ -199,6 +200,11 @@
 
 	dsos__add(vdso);
 
+	hypervisor_dso = dso__new("[hypervisor]", 0);
+	if (!hypervisor_dso)
+		return -1;
+	dsos__add(hypervisor_dso);
+
 	return err;
 }
 
@@ -637,7 +643,8 @@
 
 	if (self->sym) {
 		ret += fprintf(fp, "[%c] %s",
-			self->dso == kernel_dso ? 'k' : '.', self->sym->name);
+			self->dso == kernel_dso ? 'k' :
+			self->dso == hypervisor_dso ? 'h' : '.', self->sym->name);
 	} else {
 		ret += fprintf(fp, "%#016llx", (u64)self->ip);
 	}
@@ -960,6 +967,9 @@
 			}
 
 			switch (context) {
+			case PERF_CONTEXT_HV:
+				dso = hypervisor_dso;
+				break;
 			case PERF_CONTEXT_KERNEL:
 				dso = kernel_dso;
 				break;
@@ -1269,6 +1279,9 @@
 	} else {
 		show = SHOW_HV;
 		level = 'H';
+
+		dso = hypervisor_dso;
+
 		dprintf(" ...... dso: [hypervisor]\n");
 	}
 
#16 lore raw

[tip:perfcounters/urgent] perf report: Add hypervisor dso

Commit-ID:  fb9c818873a788c5c01c9868cc6050df96e2c7df
Gitweb:     http://git.kernel.org/tip/fb9c818873a788c5c01c9868cc6050df96e2c7df
Author:     Anton Blanchard <[email protected]>
AuthorDate: Wed, 1 Jul 2009 09:00:49 +1000
Committer:  Ingo Molnar <[email protected]>
CommitDate: Wed, 1 Jul 2009 01:25:20 +0200

perf report: Add hypervisor dso

Add a dso for hypervisor samples. We don't get any symbol information on the ppc64 hypervisor but this at least gives us a high level summary of the time spent in there.

LKML-Reference: <[email protected]>

---
 tools/perf/builtin-report.c |   15 ++++++++++++++-
 1 files changed, 14 insertions(+), 1 deletions(-)
tools/perf/builtin-report.c +14 -1
index 88e88c5..3f5d8ea 100644
@@ -121,6 +121,7 @@ typedef union event_union {
 static LIST_HEAD(dsos);
 static struct dso *kernel_dso;
 static struct dso *vdso;
+static struct dso *hypervisor_dso;
 
 static void dsos__add(struct dso *dso)
 {
@@ -202,6 +203,11 @@ static int load_kernel(void)
 
 	dsos__add(vdso);
 
+	hypervisor_dso = dso__new("[hypervisor]", 0);
+	if (!hypervisor_dso)
+		return -1;
+	dsos__add(hypervisor_dso);
+
 	return err;
 }
 
@@ -640,7 +646,8 @@ sort__sym_print(FILE *fp, struct hist_entry *self)
 
 	if (self->sym) {
 		ret += fprintf(fp, "[%c] %s",
-			self->dso == kernel_dso ? 'k' : '.', self->sym->name);
+			self->dso == kernel_dso ? 'k' :
+			self->dso == hypervisor_dso ? 'h' : '.', self->sym->name);
 	} else {
 		ret += fprintf(fp, "%#016llx", (u64)self->ip);
 	}
@@ -963,6 +970,9 @@ hist_entry__add(struct thread *thread, struct map *map, struct dso *dso,
 			}
 
 			switch (context) {
+			case PERF_CONTEXT_HV:
+				dso = hypervisor_dso;
+				break;
 			case PERF_CONTEXT_KERNEL:
 				dso = kernel_dso;
 				break;
@@ -1275,6 +1285,9 @@ process_sample_event(event_t *event, unsigned long offset, unsigned long head)
 	} else {
 		show = SHOW_HV;
 		level = 'H';
+
+		dso = hypervisor_dso;
+
 		dprintf(" ...... dso: [hypervisor]\n");
 	}
 
#17 lore raw

PATCH 7/8 powerpc: Cleanup PCL output by hiding and adding symbols

In reply to Anton Blanchard

A lot of hits in "setup" doesn't make much sense, so hide this symbol and allow all the hits to end up in copy_4k_page.

Also add some dummy symbols for the branches at 0xf00, 0xf20 and 0xf40, otherwise hits end up in trap_0e which is confusing.

---

arch/powerpc/lib/copypage_64.S +2 -2
Index: linux-2.6-tip/arch/powerpc/lib/copypage_64.S
===================================================================
@@ -26,11 +26,11 @@
 	srd	r8,r5,r11
 
 	mtctr	r8
-setup:
+.Lsetup:
 	dcbt	r9,r4
 	dcbz	r9,r3
 	add	r9,r9,r12
-	bdnz	setup
+	bdnz	.Lsetup
 END_FTR_SECTION_IFSET(CPU_FTR_CP_USE_DCBTZ)
 	addi	r3,r3,-8
 	srdi    r8,r5,7		/* page is copied in 128 byte strides */
arch/powerpc/kernel/exceptions-64s.S +3 -0
Index: linux-2.6-tip/arch/powerpc/kernel/exceptions-64s.S
===================================================================
@@ -185,12 +185,15 @@
 	 * prolog code of the PerformanceMonitor one. A little
 	 * trickery is thus necessary
 	 */
+performance_monitor_pSeries_1:
 	. = 0xf00
 	b	performance_monitor_pSeries
 
+altivec_unavailable_pSeries_1:
 	. = 0xf20
 	b	altivec_unavailable_pSeries
 
+vsx_unavailable_pSeries_1:
 	. = 0xf40
 	b	vsx_unavailable_pSeries
 
#18 lore raw

Re: [patch 7/8] powerpc: Cleanup PCL output by hiding and adding symbols

Quoted 7 lines · Anton Blanchard

A lot of hits in "setup" doesn't make much sense, so hide this symbol and allow all the hits to end up in copy_4k_page.

Also add some dummy symbols for the branches at 0xf00, 0xf20 and 0xf40, otherwise hits end up in trap_0e which is confusing.

looks good (we have similar fixes for x86 too), but i think this one should go via BenH (Cc:-ed) as it affects generic PowerPC code.

Ingo

#19 lore raw

PATCH 8/8 powerpc: Fix PCL vdso detection

In reply to Anton Blanchard

PCL uses arch_vma_name() to detect a vdso region which in turn uses current->mm->context.vdso_base. We need to initialise this before doing the mmap or else we fail to detect the vdso.

---

arch/powerpc/kernel/vdso.c +10 -4
Index: linux-2.6-tip/arch/powerpc/kernel/vdso.c
===================================================================
@@ -235,6 +235,13 @@
 	}
 
 	/*
+	 * Put vDSO base into mm struct. We need to do this before calling
+	 * install_special_mapping or the perf counter mmap tracking code
+	 * will fail to recognise it as a vDSO (since arch_vma_name fails).
+	 */
+	current->mm->context.vdso_base = vdso_base;
+
+	/*
 	 * our vma flags don't have VM_WRITE so by default, the process isn't
 	 * allowed to write those pages.
 	 * gdb can break that with ptrace interface, and thus trigger COW on
@@ -254,11 +261,10 @@
 				     VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC|
 				     VM_ALWAYSDUMP,
 				     vdso_pagelist);
-	if (rc)
+	if (rc) {
+		current->mm->context.vdso_base = 0;
 		goto fail_mmapsem;
-
-	/* Put vDSO base into mm struct */
-	current->mm->context.vdso_base = vdso_base;
+	}
 
 	up_write(&mm->mmap_sem);
 	return 0;
#20 lore raw

Re: [patch 8/8] powerpc: Fix PCL vdso detection

Quoted 3 lines · Anton Blanchard

PCL uses arch_vma_name() to detect a vdso region which in turn uses current->mm->context.vdso_base. We need to initialise this before doing the mmap or else we fail to detect the vdso.

This too should go via the PowerPC tree i think.

I've applied the rest - nice stuff!!

Ingo