perf: Address range filtering fixes
Alexander Shishkin
maintainercover perf: Address range filtering fixes
Hi Peter,
These are bugfixes for the address range filtering. One had been posted before [1], but fell through the cracks. One was discovered and originally patched by Adrian recently. Both can be theoretically backported all the way to v4.7, but that may require manual conflict resolution and the benefits are not clear.
[1] https://marc.info/?l=linux-kernel&m=153900881208142
Alexander Shishkin (2):
perf: Copy parent's address filter offsets on clone
perf, pt, coresight: Fix address filters for vmas with non-zero offset
arch/x86/events/intel/pt.c | 9 +- .../hwtracing/coresight/coresight-etm-perf.c | 7 +- include/linux/perf_event.h | 7 +- kernel/events/core.c | 90 ++++++++++++------- 4 files changed, 74 insertions(+), 39 deletions(-)
Signature · 1 line
2.20.1
Alexander Shishkin
maintainerPATCH 1/2 perf: Copy parent's address filter offsets on clone
When a child event is allocated in the inherit_event() path, the VMA based filter offsets are not copied from the parent, even though the address space mapping of the new task remains the same, which leads to no trace for the new task until exec.
--- kernel/events/core.c | 15 +++++++++++++++ 1 file changed, 15 insertions(+)
kernel/events/core.c +15 -0
@@ -1255,6 +1255,7 @@ static void put_ctx(struct perf_event_context *ctx)* perf_event_context::lock * perf_event::mmap_mutex * mmap_sem +* perf_addr_filters_head::lock * * cpu_hotplug_lock * pmus_lock
@@ -10312,6 +10313,20 @@ perf_event_alloc(struct perf_event_attr *attr, int cpu, goto err_per_task; } +/* +* Clone the parent's vma offsets: they are valid until exec() +* even if the mm is not shared with the parent. +*/ +if (event->parent) { +struct perf_addr_filters_head *ifh =perf_event_addr_filters (event); + +raw_spin_lock_irq (&ifh->lock); +memcpy (event->addr_filters_offs, + event->parent->addr_filters_offs, + pmu->nr_addr_filters *sizeof (unsigned long )); +raw_spin_unlock_irq (&ifh->lock); + } +/* force hw sync on the address filters */ event->addr_filters_gen =1 ; }
Signature · 1 line
2.20.1
Re: [PATCH v0 1/2] perf: Copy parent's address filter offsets on clone
Em Fri, Feb 15, 2019 at 01:56:54PM +0200, Alexander Shishkin escreveu:
Quoted 4 lines
When a child event is allocated in the
inherit_event()path, the VMA based filter offsets are not copied from the parent, even though the address space mapping of the new task remains the same, which leads to no trace for the new task until exec.
Peter, I'm processing this one, ok? Ack?
- Arnaldo
Quoted 42 lines
Signed-off-by Alexander Shishkin <[email protected]>Fixes 375637bc5249 ("perf/core: Introduce address range filtering")Reported-by Mansour Alharthi <[email protected]>--- kernel/events/core.c | 15 +++++++++++++++ 1 file changed, 15 insertions(+)kernel/events/core.c +15
-0@@ -1255,6 +1255,7 @@ static void put_ctx(struct perf_event_context *ctx)* perf_event_context::lock * perf_event::mmap_mutex * mmap_sem +* perf_addr_filters_head::lock * * cpu_hotplug_lock * pmus_lock @@ -10312,6 +10313,20 @@ perf_event_alloc(struct perf_event_attr *attr, int cpu, goto err_per_task; } +/* +* Clone the parent's vma offsets: they are valid until exec() +* even if the mm is not shared with the parent. +*/ +if (event->parent) { +struct perf_addr_filters_head *ifh =perf_event_addr_filters (event); + +raw_spin_lock_irq (&ifh->lock); +memcpy (event->addr_filters_offs, + event->parent->addr_filters_offs, + pmu->nr_addr_filters *sizeof (unsigned long )); +raw_spin_unlock_irq (&ifh->lock); + } +/* force hw sync on the address filters */ event->addr_filters_gen =1 ; }Signature · 1 line
2.20.1
Alexander Shishkin
maintainerRe: [PATCH v0 1/2] perf: Copy parent's address filter offsets on clone
Quoted 7 lines · Arnaldo Carvalho de Melo
Em Fri, Feb 15, 2019 at 01:56:54PM +0200, Alexander Shishkin escreveu:
Quoted 4 lines
When a child event is allocated in the
inherit_event()path, the VMA based filter offsets are not copied from the parent, even though the address space mapping of the new task remains the same, which leads to no trace for the new task until exec.Peter, I'm processing this one, ok? Ack?
Any news on this one?
Thanks,
--
Alex
Re: [PATCH v0 1/2] perf: Copy parent's address filter offsets on clone
Em Fri, Feb 22, 2019 at 02:56:26PM +0200, Alexander Shishkin escreveu:
Quoted 11 lines
Quoted 7 lines · Arnaldo Carvalho de Melo
Em Fri, Feb 15, 2019 at 01:56:54PM +0200, Alexander Shishkin escreveu:
Quoted 4 lines
When a child event is allocated in the
inherit_event()path, the VMA based filter offsets are not copied from the parent, even though the address space mapping of the new task remains the same, which leads to no trace for the new task until exec.Peter, I'm processing this one, ok? Ack?
Any news on this one?
Will process it now, was waiting for some ack from Peter.
- Arnaldo
tip-bot for Alexander Shishkin
merge bot[tip:perf/core] perf: Copy parent's address filter offsets on clone
In reply to Alexander Shishkinperf: Copy parent's address filter offsets on clone
Commit-ID: 18736eef12137c59f60cc9f56dc5bea05c92e0eb Gitweb: https://git.kernel.org/tip/18736eef12137c59f60cc9f56dc5bea05c92e0eb Author: Alexander Shishkin <[email protected]> AuthorDate: Fri, 15 Feb 2019 13:56:54 +0200 Committer: Arnaldo Carvalho de Melo <[email protected]> CommitDate: Fri, 22 Feb 2019 16:52:07 -0300
perf: Copy parent's address filter offsets on clone
When a child event is allocated in the inherit_event() path, the VMA based filter offsets are not copied from the parent, even though the address space mapping of the new task remains the same, which leads to no trace for the new task until exec.
--- kernel/events/core.c | 15 +++++++++++++++ 1 file changed, 15 insertions(+)
kernel/events/core.c +15 -0
@@ -1255,6 +1255,7 @@ static void put_ctx(struct perf_event_context *ctx)* perf_event_context::lock * perf_event::mmap_mutex * mmap_sem +* perf_addr_filters_head::lock * * cpu_hotplug_lock * pmus_lock
@@ -10312,6 +10313,20 @@ perf_event_alloc(struct perf_event_attr *attr, int cpu, goto err_per_task; } +/* +* Clone the parent's vma offsets: they are valid until exec() +* even if the mm is not shared with the parent. +*/ +if (event->parent) { +struct perf_addr_filters_head *ifh =perf_event_addr_filters (event); + +raw_spin_lock_irq (&ifh->lock); +memcpy (event->addr_filters_offs, + event->parent->addr_filters_offs, + pmu->nr_addr_filters *sizeof (unsigned long )); +raw_spin_unlock_irq (&ifh->lock); + } +/* force hw sync on the address filters */ event->addr_filters_gen =1 ; }
Alexander Shishkin
maintainerPATCH 2/2 perf, pt, coresight: Fix address filters for vmas with non-zero offset
In reply to Alexander ShishkinCurrently, the address range calculation for file-based filters works as long as the vma that maps the matching part of the object file starts from offset zero into the file (vm_pgoff==0). Otherwise, the resulting filter range would be off by vm_pgoff pages. Another related problem is that in case of a partially matching vma, that is, a vma that matches part of a filter region, the filter range size wouldn't be adjusted.
Fix the arithmetics around address filter range calculations, taking into account vma offset, so that the entire calculation is done before the filter configuration is passed to the PMU drivers instead of having those drivers do the final bit of arithmetics.
Based on the patch by Adrian Hunter <adrian.hunter.intel.com>.
--- arch/x86/events/intel/pt.c | 9 ++- .../hwtracing/coresight/coresight-etm-perf.c | 7 +- include/linux/perf_event.h | 7 +- kernel/events/core.c | 81 +++++++++++-------- 4 files changed, 62 insertions(+), 42 deletions(-)
arch/x86/events/intel/pt.c +5 -4
@@ -1223,7 +1223,8 @@ static int pt_event_addr_filters_validate(struct list_head *filters)static void pt_event_addr_filters_sync (struct perf_event *event) {struct perf_addr_filters_head *head =perf_event_addr_filters (event); -unsigned long msr_a, msr_b, *offs = event->addr_filters_offs; +unsigned long msr_a, msr_b; +struct perf_addr_filter_range *fr = event->addr_filter_ranges;struct pt_filters *filters = event->hw.addr_filters;struct perf_addr_filter *filter;int range =0 ;
@@ -1232,12 +1233,12 @@ static void pt_event_addr_filters_sync(struct perf_event *event)return ;list_for_each_entry (filter, &head->list, entry) { -if (filter->path.dentry && !offs[range]) { +if (filter->path.dentry && !fr[range].start) { msr_a = msr_b =0 ; }else {/* apply the offset */ - msr_a = filter->offset + offs[range]; - msr_b = filter->size + msr_a -1 ; + msr_a = fr[range].start; + msr_b = msr_a + fr[range].size -1 ; } filters->filter[range].msr_a = msr_a;
drivers/hwtracing/coresight/coresight-etm-perf.c +4 -3
@@ -433,15 +433,16 @@ static int etm_addr_filters_validate(struct list_head *filters)static void etm_addr_filters_sync (struct perf_event *event) {struct perf_addr_filters_head *head =perf_event_addr_filters (event); -unsigned long start, stop, *offs = event->addr_filters_offs; +unsigned long start, stop; +struct perf_addr_filter_range *fr = event->addr_filter_ranges;struct etm_filters *filters = event->hw.addr_filters;struct etm_filter *etm_filter;struct perf_addr_filter *filter;int i =0 ;list_for_each_entry (filter, &head->list, entry) { - start = filter->offset + offs[i]; - stop = start + filter->size; + start = fr[i].start; + stop = start + fr[i].size; etm_filter = &filters->etm_filter[i];switch (filter->action) {
include/linux/perf_event.h +6 -1
@@ -490,6 +490,11 @@ struct perf_addr_filters_head {unsigned int nr_file_filters; }; +struct perf_addr_filter_range { +unsigned long start; +unsigned long size; +}; +/** * enum perf_event_state - the states of an event: */
@@ -666,7 +671,7 @@ struct perf_event {/* address range filters */ struct perf_addr_filters_head addr_filters;/* vma address array for file-based filders */ -unsigned long *addr_filters_offs; +struct perf_addr_filter_range *addr_filter_ranges;unsigned long addr_filters_gen;void (*destroy)(struct perf_event *);
kernel/events/core.c +47 -34
163 lines not on this page · raw message
Signature · 1 line
2.20.1
Re: [PATCH v0 2/2] perf, pt, coresight: Fix address filters for vmas with non-zero offset
Adding Mathieu to the CC list.
Em Fri, Feb 15, 2019 at 01:56:55PM +0200, Alexander Shishkin escreveu:
Quoted 281 lines
Currently, the address range calculation for file-based filters works as long as the vma that maps the matching part of the object file starts from offset zero into the file (vm_pgoff==0). Otherwise, the resulting filter range would be off by
vm_pgoffpages. Another related problem is that in case of a partially matching vma, that is, a vma that matches part of a filter region, the filter range size wouldn't be adjusted.Fix the arithmetics around address filter range calculations, taking into account vma offset, so that the entire calculation is done before the filter configuration is passed to the PMU drivers instead of having those drivers do the final bit of arithmetics.
Based on the patch by Adrian Hunter <adrian.hunter.intel.com>.
Signed-off-by Alexander Shishkin <[email protected]>Fixes 375637bc5249 ("perf/core: Introduce address range filtering")Reported-by Adrian Hunter <[email protected]>--- arch/x86/events/intel/pt.c | 9 ++- .../hwtracing/coresight/coresight-etm-perf.c | 7 +- include/linux/perf_event.h | 7 +- kernel/events/core.c | 81 +++++++++++-------- 4 files changed, 62 insertions(+), 42 deletions(-)arch/x86/events/intel/pt.c +5
-4@@ -1223,7 +1223,8 @@ static int pt_event_addr_filters_validate(struct list_head *filters)static void pt_event_addr_filters_sync (struct perf_event *event) {struct perf_addr_filters_head *head =perf_event_addr_filters (event); -unsigned long msr_a, msr_b, *offs = event->addr_filters_offs; +unsigned long msr_a, msr_b; +struct perf_addr_filter_range *fr = event->addr_filter_ranges;struct pt_filters *filters = event->hw.addr_filters;struct perf_addr_filter *filter;int range =0 ;@@ -1232,12 +1233,12 @@ static void pt_event_addr_filters_sync(struct perf_event *event)return ;list_for_each_entry (filter, &head->list, entry) { -if (filter->path.dentry && !offs[range]) { +if (filter->path.dentry && !fr[range].start) { msr_a = msr_b =0 ; }else {/* apply the offset */ - msr_a = filter->offset + offs[range]; - msr_b = filter->size + msr_a -1 ; + msr_a = fr[range].start; + msr_b = msr_a + fr[range].size -1 ; } filters->filter[range].msr_a = msr_a;drivers/hwtracing/coresight/coresight-etm-perf.c +4
-3@@ -433,15 +433,16 @@ static int etm_addr_filters_validate(struct list_head *filters)static void etm_addr_filters_sync (struct perf_event *event) {struct perf_addr_filters_head *head =perf_event_addr_filters (event); -unsigned long start, stop, *offs = event->addr_filters_offs; +unsigned long start, stop; +struct perf_addr_filter_range *fr = event->addr_filter_ranges;struct etm_filters *filters = event->hw.addr_filters;struct etm_filter *etm_filter;struct perf_addr_filter *filter;int i =0 ;list_for_each_entry (filter, &head->list, entry) { - start = filter->offset + offs[i]; - stop = start + filter->size; + start = fr[i].start; + stop = start + fr[i].size; etm_filter = &filters->etm_filter[i];switch (filter->action) {include/linux/perf_event.h +6
-1@@ -490,6 +490,11 @@ struct perf_addr_filters_head {unsigned int nr_file_filters; }; +struct perf_addr_filter_range { +unsigned long start; +unsigned long size; +}; +/** * enum perf_event_state - the states of an event: */ @@ -666,7 +671,7 @@ struct perf_event {/* address range filters */ struct perf_addr_filters_head addr_filters;/* vma address array for file-based filders */ -unsigned long *addr_filters_offs; +struct perf_addr_filter_range *addr_filter_ranges;unsigned long addr_filters_gen;void (*destroy)(struct perf_event *);kernel/events/core.c +47
-34@@ -2799,7 +2799,7 @@ static int perf_event_stop(struct perf_event *event, int restart)* * (p1) when userspace mappings change as a result of (1) or (2) or (3) below, * we update the addresses of corresponding vmas in -* event::addr_filters_offs array and bump the event::addr_filters_gen; +* event::addr_filter_ranges array and bump the event::addr_filters_gen; * (p2) when an event is scheduled in (pmu::add), it calls * perf_event_addr_filters_sync() which calls pmu::addr_filters_sync() * if the generation has changed since the previous call. @@ -4446,7 +4446,7 @@ static void _free_event(struct perf_event *event)perf_event_free_bpf_prog (event);perf_addr_filters_splice (event,NULL ); -kfree (event->addr_filters_offs); +kfree (event->addr_filter_ranges);if (event->destroy) event->destroy (event);@@ -6687,7 +6687,8 @@ static void perf_event_addr_filters_exec(struct perf_event *event, void *data)raw_spin_lock_irqsave (&ifh->lock, flags);list_for_each_entry (filter, &ifh->list, entry) {if (filter->path.dentry) { - event->addr_filters_offs[count] =0 ; + event->addr_filter_ranges[count].start =0 ; + event->addr_filter_ranges[count].size =0 ; restart++; }@@ -7367,28 +7368,47 @@ static bool perf_addr_filter_match(struct perf_addr_filter *filter,return true; } +static bool perf_addr_filter_vma_adjust (struct perf_addr_filter *filter, +struct vm_area_struct *vma, +struct perf_addr_filter_range *fr) +{ +unsigned long vma_size = vma->vm_end - vma->vm_start; +unsigned long off = vma->vm_pgoff <<PAGE_SHIFT ; +struct file *file = vma->vm_file; + +if (!perf_addr_filter_match (filter, file, off, vma_size)) +return false; + +if (filter->offset < off) { + fr->start = vma->vm_start; + fr->size =min (vma_size, filter->size - (off - filter->offset)); + }else { + fr->start = vma->vm_start + filter->offset - off; + fr->size =min (vma->vm_end - fr->start, filter->size); + } + +return true; +} +static void __perf_addr_filters_adjust (struct perf_event *event,void *data) {struct perf_addr_filters_head *ifh =perf_event_addr_filters (event);struct vm_area_struct *vma = data; -unsigned long off = vma->vm_pgoff <<PAGE_SHIFT , flags; -struct file *file = vma->vm_file;struct perf_addr_filter *filter;unsigned int restart =0 , count =0 ; +unsigned long flags;if (!has_addr_filter (event))return ; -if (!file) +if (!vma->vm_file)return ;raw_spin_lock_irqsave (&ifh->lock, flags);list_for_each_entry (filter, &ifh->list, entry) { -if (perf_addr_filter_match (filter, file, off, - vma->vm_end - vma->vm_start)) { - event->addr_filters_offs[count] = vma->vm_start; +if (perf_addr_filter_vma_adjust (filter, vma, + &event->addr_filter_ranges[count])) restart++; - } count++; }@@ -8978,26 +8998,19 @@ static void perf_addr_filters_splice(struct perf_event *event,* @filter; if so, adjust filter's address range. * Called with mm::mmap_sem down for reading. */ -static unsigned long perf_addr_filter_apply (struct perf_addr_filter *filter, -struct mm_struct *mm) +static void perf_addr_filter_apply (struct perf_addr_filter *filter, +struct mm_struct *mm, +struct perf_addr_filter_range *fr) {struct vm_area_struct *vma;for (vma = mm->mmap; vma; vma = vma->vm_next) { -struct file *file = vma->vm_file; -unsigned long off = vma->vm_pgoff <<PAGE_SHIFT ; -unsigned long vma_size = vma->vm_end - vma->vm_start; - -if (!file) +if (!vma->vm_file)continue ; -if (!perf_addr_filter_match (filter, file, off, vma_size)) -continue ; - -return vma->vm_start; +if (perf_addr_filter_vma_adjust (filter, vma, fr)) +return ; } - -return 0 ; }/* @@ -9031,15 +9044,15 @@ static void perf_event_addr_filters_apply(struct perf_event *event)raw_spin_lock_irqsave (&ifh->lock, flags);list_for_each_entry (filter, &ifh->list, entry) { - event->addr_filters_offs[count] =0 ; + event->addr_filter_ranges[count].start =0 ; + event->addr_filter_ranges[count].size =0 ;/* * Adjust base offset if the filter is associated to a binary * that needs to be mapped: */ if (filter->path.dentry) - event->addr_filters_offs[count] = -perf_addr_filter_apply (filter, mm); +perf_addr_filter_apply (filter, mm, &event->addr_filter_ranges[count]); count++; }@@ -10305,10 +10318,10 @@ perf_event_alloc(struct perf_event_attr *attr, int cpu, goto err_pmu;if (has_addr_filter (event)) { - event->addr_filters_offs =kcalloc (pmu->nr_addr_filters, -sizeof (unsigned long ), -GFP_KERNEL ); -if (!event->addr_filters_offs) { + event->addr_filter_ranges =kcalloc (pmu->nr_addr_filters, +sizeof (struct perf_addr_filter_range ), +GFP_KERNEL ); +if (!event->addr_filter_ranges) { err = -ENOMEM ; goto err_per_task; }@@ -10321,9 +10334,9 @@ perf_event_alloc(struct perf_event_attr *attr, int cpu,struct perf_addr_filters_head *ifh =perf_event_addr_filters (event);raw_spin_lock_irq (&ifh->lock); -memcpy (event->addr_filters_offs, - event->parent->addr_filters_offs, - pmu->nr_addr_filters *sizeof (unsigned long )); +memcpy (event->addr_filter_ranges, + event->parent->addr_filter_ranges, + pmu->nr_addr_filters *sizeof (struct perf_addr_filter_range ));raw_spin_unlock_irq (&ifh->lock); }@@ -10345,7 +10358,7 @@ perf_event_alloc(struct perf_event_attr *attr, int cpu,return event; err_addr_filters: -kfree (event->addr_filters_offs); +kfree (event->addr_filter_ranges); err_per_task:exclusive_event_destroy (event);Signature · 1 line
2.20.1
Mathieu Poirier
Tested-byRe: [PATCH v0 2/2] perf, pt, coresight: Fix address filters for vmas with non-zero offset
Quoted 274 lines · Arnaldo Carvalho de Melo
Adding Mathieu to the CC list.
Em Fri, Feb 15, 2019 at 01:56:55PM +0200, Alexander Shishkin escreveu:
Quoted 270 lines
Currently, the address range calculation for file-based filters works as long as the vma that maps the matching part of the object file starts from offset zero into the file (vm_pgoff==0). Otherwise, the resulting filter range would be off by
vm_pgoffpages. Another related problem is that in case of a partially matching vma, that is, a vma that matches part of a filter region, the filter range size wouldn't be adjusted.Fix the arithmetics around address filter range calculations, taking into account vma offset, so that the entire calculation is done before the filter configuration is passed to the PMU drivers instead of having those drivers do the final bit of arithmetics.
Based on the patch by Adrian Hunter <adrian.hunter.intel.com>.
Signed-off-by Alexander Shishkin <[email protected]>Fixes 375637bc5249 ("perf/core: Introduce address range filtering")Reported-by Adrian Hunter <[email protected]>--- arch/x86/events/intel/pt.c | 9 ++- .../hwtracing/coresight/coresight-etm-perf.c | 7 +- include/linux/perf_event.h | 7 +- kernel/events/core.c | 81 +++++++++++-------- 4 files changed, 62 insertions(+), 42 deletions(-)arch/x86/events/intel/pt.c +5
-4@@ -1223,7 +1223,8 @@ static int pt_event_addr_filters_validate(struct list_head *filters)static void pt_event_addr_filters_sync (struct perf_event *event) {struct perf_addr_filters_head *head =perf_event_addr_filters (event); -unsigned long msr_a, msr_b, *offs = event->addr_filters_offs; +unsigned long msr_a, msr_b; +struct perf_addr_filter_range *fr = event->addr_filter_ranges;struct pt_filters *filters = event->hw.addr_filters;struct perf_addr_filter *filter;int range =0 ;@@ -1232,12 +1233,12 @@ static void pt_event_addr_filters_sync(struct perf_event *event)return ;list_for_each_entry (filter, &head->list, entry) { -if (filter->path.dentry && !offs[range]) { +if (filter->path.dentry && !fr[range].start) { msr_a = msr_b =0 ; }else {/* apply the offset */ - msr_a = filter->offset + offs[range]; - msr_b = filter->size + msr_a -1 ; + msr_a = fr[range].start; + msr_b = msr_a + fr[range].size -1 ; } filters->filter[range].msr_a = msr_a;drivers/hwtracing/coresight/coresight-etm-perf.c +4
-3@@ -433,15 +433,16 @@ static int etm_addr_filters_validate(struct list_head *filters)static void etm_addr_filters_sync (struct perf_event *event) {struct perf_addr_filters_head *head =perf_event_addr_filters (event); -unsigned long start, stop, *offs = event->addr_filters_offs; +unsigned long start, stop; +struct perf_addr_filter_range *fr = event->addr_filter_ranges;struct etm_filters *filters = event->hw.addr_filters;struct etm_filter *etm_filter;struct perf_addr_filter *filter;int i =0 ;list_for_each_entry (filter, &head->list, entry) { - start = filter->offset + offs[i]; - stop = start + filter->size; + start = fr[i].start; + stop = start + fr[i].size; etm_filter = &filters->etm_filter[i];switch (filter->action) {include/linux/perf_event.h +6
-1@@ -490,6 +490,11 @@ struct perf_addr_filters_head {unsigned int nr_file_filters; }; +struct perf_addr_filter_range { +unsigned long start; +unsigned long size; +}; +/** * enum perf_event_state - the states of an event: */ @@ -666,7 +671,7 @@ struct perf_event {/* address range filters */ struct perf_addr_filters_head addr_filters;/* vma address array for file-based filders */ -unsigned long *addr_filters_offs; +struct perf_addr_filter_range *addr_filter_ranges;unsigned long addr_filters_gen;void (*destroy)(struct perf_event *);kernel/events/core.c +46
-33@@ -2799,7 +2799,7 @@ static int perf_event_stop(struct perf_event *event, int restart)* * (p1) when userspace mappings change as a result of (1) or (2) or (3) below, * we update the addresses of corresponding vmas in -* event::addr_filters_offs array and bump the event::addr_filters_gen; +* event::addr_filter_ranges array and bump the event::addr_filters_gen; * (p2) when an event is scheduled in (pmu::add), it calls * perf_event_addr_filters_sync() which calls pmu::addr_filters_sync() * if the generation has changed since the previous call. @@ -4446,7 +4446,7 @@ static void _free_event(struct perf_event *event)perf_event_free_bpf_prog (event);perf_addr_filters_splice (event,NULL ); -kfree (event->addr_filters_offs); +kfree (event->addr_filter_ranges);if (event->destroy) event->destroy (event);@@ -6687,7 +6687,8 @@ static void perf_event_addr_filters_exec(struct perf_event *event, void *data)raw_spin_lock_irqsave (&ifh->lock, flags);list_for_each_entry (filter, &ifh->list, entry) {if (filter->path.dentry) { - event->addr_filters_offs[count] =0 ; + event->addr_filter_ranges[count].start =0 ; + event->addr_filter_ranges[count].size =0 ; restart++; }@@ -7367,28 +7368,47 @@ static bool perf_addr_filter_match(struct perf_addr_filter *filter,return true; } +static bool perf_addr_filter_vma_adjust (struct perf_addr_filter *filter, +struct vm_area_struct *vma, +struct perf_addr_filter_range *fr) +{ +unsigned long vma_size = vma->vm_end - vma->vm_start; +unsigned long off = vma->vm_pgoff <<PAGE_SHIFT ; +struct file *file = vma->vm_file; + +if (!perf_addr_filter_match (filter, file, off, vma_size)) +return false; + +if (filter->offset < off) { + fr->start = vma->vm_start; + fr->size =min (vma_size, filter->size - (off - filter->offset)); + }else { + fr->start = vma->vm_start + filter->offset - off; + fr->size =min (vma->vm_end - fr->start, filter->size); + } + +return true; +} +static void __perf_addr_filters_adjust (struct perf_event *event,void *data) {struct perf_addr_filters_head *ifh =perf_event_addr_filters (event);struct vm_area_struct *vma = data; -unsigned long off = vma->vm_pgoff <<PAGE_SHIFT , flags; -struct file *file = vma->vm_file;struct perf_addr_filter *filter;unsigned int restart =0 , count =0 ; +unsigned long flags;if (!has_addr_filter (event))return ; -if (!file) +if (!vma->vm_file)return ;raw_spin_lock_irqsave (&ifh->lock, flags);list_for_each_entry (filter, &ifh->list, entry) { -if (perf_addr_filter_match (filter, file, off, - vma->vm_end - vma->vm_start)) { - event->addr_filters_offs[count] = vma->vm_start; +if (perf_addr_filter_vma_adjust (filter, vma, + &event->addr_filter_ranges[count])) restart++; - } count++; }@@ -8978,26 +8998,19 @@ static void perf_addr_filters_splice(struct perf_event *event,* @filter; if so, adjust filter's address range. * Called with mm::mmap_sem down for reading. */ -static unsigned long perf_addr_filter_apply (struct perf_addr_filter *filter, -struct mm_struct *mm) +static void perf_addr_filter_apply (struct perf_addr_filter *filter, +struct mm_struct *mm, +struct perf_addr_filter_range *fr) {struct vm_area_struct *vma;for (vma = mm->mmap; vma; vma = vma->vm_next) { -struct file *file = vma->vm_file; -unsigned long off = vma->vm_pgoff <<PAGE_SHIFT ; -unsigned long vma_size = vma->vm_end - vma->vm_start; - -if (!file) +if (!vma->vm_file)continue ; -if (!perf_addr_filter_match (filter, file, off, vma_size)) -continue ; - -return vma->vm_start; +if (perf_addr_filter_vma_adjust (filter, vma, fr)) +return ; } - -return 0 ; }/* @@ -9031,15 +9044,15 @@ static void perf_event_addr_filters_apply(struct perf_event *event)raw_spin_lock_irqsave (&ifh->lock, flags);list_for_each_entry (filter, &ifh->list, entry) { - event->addr_filters_offs[count] =0 ; + event->addr_filter_ranges[count].start =0 ; + event->addr_filter_ranges[count].size =0 ;/* * Adjust base offset if the filter is associated to a binary * that needs to be mapped: */ if (filter->path.dentry) - event->addr_filters_offs[count] = -perf_addr_filter_apply (filter, mm); +perf_addr_filter_apply (filter, mm, &event->addr_filter_ranges[count]); count++; }@@ -10305,10 +10318,10 @@ perf_event_alloc(struct perf_event_attr *attr, int cpu, goto err_pmu;if (has_addr_filter (event)) { - event->addr_filters_offs =kcalloc (pmu->nr_addr_filters, -sizeof (unsigned long ), -GFP_KERNEL ); -if (!event->addr_filters_offs) { + event->addr_filter_ranges =kcalloc (pmu->nr_addr_filters, +sizeof (struct perf_addr_filter_range ), +GFP_KERNEL ); +if (!event->addr_filter_ranges) { err = -ENOMEM ; goto err_per_task; }@@ -10321,9 +10334,9 @@ perf_event_alloc(struct perf_event_attr *attr, int cpu,struct perf_addr_filters_head *ifh =perf_event_addr_filters (event);raw_spin_lock_irq (&ifh->lock); -memcpy (event->addr_filters_offs, - event->parent->addr_filters_offs, - pmu->nr_addr_filters *sizeof (unsigned long )); +memcpy (event->addr_filter_ranges, + event->parent->addr_filter_ranges, + pmu->nr_addr_filters *sizeof (struct perf_addr_filter_range ));raw_spin_unlock_irq (&ifh->lock); }
Quoted 7 lines
kernel/events/core.c perf_event_alloc(struct perf_event_attr *attr, int cpu, · line 10362
Quoted 7 lines
kernel/events/core.c +1
-1@@ -10345,7 +10358,7 @@ perf_event_alloc(struct perf_event_attr *attr, int cpu,return event; err_addr_filters: -kfree (event->addr_filters_offs); +kfree (event->addr_filter_ranges);
Thanks for CC'ing me on this Arnaldo.
Quoted 4 lines
Quoted 4 lines
kernel/events/core.c +0
-1err_per_task:exclusive_event_destroy (event); --2.20.1
tip-bot for Alexander Shishkin
merge bot[tip:perf/core] perf, pt, coresight: Fix address filters for vmas with non-zero offset
In reply to Alexander Shishkinperf, pt, coresight: Fix address filters for vmas with non-zero offset
Commit-ID: c60f83b813e5b25ccd5de7e8c8925c31b3aebcc1 Gitweb: https://git.kernel.org/tip/c60f83b813e5b25ccd5de7e8c8925c31b3aebcc1 Author: Alexander Shishkin <[email protected]> AuthorDate: Fri, 15 Feb 2019 13:56:55 +0200 Committer: Arnaldo Carvalho de Melo <[email protected]> CommitDate: Fri, 22 Feb 2019 16:52:07 -0300
perf, pt, coresight: Fix address filters for vmas with non-zero offset
Currently, the address range calculation for file-based filters works as long as the vma that maps the matching part of the object file starts from offset zero into the file (vm_pgoff==0). Otherwise, the resulting filter range would be off by vm_pgoff pages. Another related problem is that in case of a partially matching vma, that is, a vma that matches part of a filter region, the filter range size wouldn't be adjusted.
Fix the arithmetics around address filter range calculations, taking into account vma offset, so that the entire calculation is done before the filter configuration is passed to the PMU drivers instead of having those drivers do the final bit of arithmetics.
Based on the patch by Adrian Hunter <adrian.hunter.intel.com>.
--- arch/x86/events/intel/pt.c | 9 +-- drivers/hwtracing/coresight/coresight-etm-perf.c | 7 +- include/linux/perf_event.h | 7 +- kernel/events/core.c | 81 ++++++++++++++---------- 4 files changed, 62 insertions(+), 42 deletions(-)
arch/x86/events/intel/pt.c +5 -4
@@ -1223,7 +1223,8 @@ static int pt_event_addr_filters_validate(struct list_head *filters)static void pt_event_addr_filters_sync (struct perf_event *event) {struct perf_addr_filters_head *head =perf_event_addr_filters (event); -unsigned long msr_a, msr_b, *offs = event->addr_filters_offs; +unsigned long msr_a, msr_b; +struct perf_addr_filter_range *fr = event->addr_filter_ranges;struct pt_filters *filters = event->hw.addr_filters;struct perf_addr_filter *filter;int range =0 ;
@@ -1232,12 +1233,12 @@ static void pt_event_addr_filters_sync(struct perf_event *event)return ;list_for_each_entry (filter, &head->list, entry) { -if (filter->path.dentry && !offs[range]) { +if (filter->path.dentry && !fr[range].start) { msr_a = msr_b =0 ; }else {/* apply the offset */ - msr_a = filter->offset + offs[range]; - msr_b = filter->size + msr_a -1 ; + msr_a = fr[range].start; + msr_b = msr_a + fr[range].size -1 ; } filters->filter[range].msr_a = msr_a;
drivers/hwtracing/coresight/coresight-etm-perf.c +4 -3
@@ -433,15 +433,16 @@ static int etm_addr_filters_validate(struct list_head *filters)static void etm_addr_filters_sync (struct perf_event *event) {struct perf_addr_filters_head *head =perf_event_addr_filters (event); -unsigned long start, stop, *offs = event->addr_filters_offs; +unsigned long start, stop; +struct perf_addr_filter_range *fr = event->addr_filter_ranges;struct etm_filters *filters = event->hw.addr_filters;struct etm_filter *etm_filter;struct perf_addr_filter *filter;int i =0 ;list_for_each_entry (filter, &head->list, entry) { - start = filter->offset + offs[i]; - stop = start + filter->size; + start = fr[i].start; + stop = start + fr[i].size; etm_filter = &filters->etm_filter[i];switch (filter->action) {
include/linux/perf_event.h +6 -1
@@ -490,6 +490,11 @@ struct perf_addr_filters_head {unsigned int nr_file_filters; }; +struct perf_addr_filter_range { +unsigned long start; +unsigned long size; +}; +/** * enum perf_event_state - the states of an event: */
@@ -666,7 +671,7 @@ struct perf_event {/* address range filters */ struct perf_addr_filters_head addr_filters;/* vma address array for file-based filders */ -unsigned long *addr_filters_offs; +struct perf_addr_filter_range *addr_filter_ranges;unsigned long addr_filters_gen;void (*destroy)(struct perf_event *);
kernel/events/core.c +47 -34
163 lines not on this page · raw message