From 5116d8f6b977970ebefc1932c0f313163a6ec91f Mon Sep 17 00:00:00 2001 From: "Michael S. Tsirkin" Date: Sun, 26 Jul 2009 17:10:01 +0300 Subject: KVM: fix ack not being delivered when msi present kvm_notify_acked_irq does not check irq type, so that it sometimes interprets msi vector as irq. As a result, ack notifiers are not called, which typially hangs the guest. The fix is to track and check irq type. Signed-off-by: Michael S. Tsirkin Signed-off-by: Avi Kivity --- include/linux/kvm_host.h | 1 + 1 file changed, 1 insertion(+) (limited to 'include') diff --git a/include/linux/kvm_host.h b/include/linux/kvm_host.h index 16713dc672e4..3060bdc35ffe 100644 --- a/include/linux/kvm_host.h +++ b/include/linux/kvm_host.h @@ -110,6 +110,7 @@ struct kvm_memory_slot { struct kvm_kernel_irq_routing_entry { u32 gsi; + u32 type; int (*set)(struct kvm_kernel_irq_routing_entry *e, struct kvm *kvm, int level); union { -- cgit v1.2.3 From a044560c3a1f0ad75ce685c1ed7604820b9ed319 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 10 Aug 2009 11:16:52 +0200 Subject: perf_counter: Correct PERF_SAMPLE_RAW output PERF_SAMPLE_* output switches should unconditionally output the correct format, as they are the only way to unambiguously parse the PERF_EVENT_SAMPLE data. Signed-off-by: Peter Zijlstra Acked-by: Frederic Weisbecker Cc: Arnaldo Carvalho de Melo Cc: Mike Galbraith Cc: Paul Mackerras LKML-Reference: <1249896447.17467.74.camel@twins> Signed-off-by: Ingo Molnar --- include/linux/perf_counter.h | 2 ++ include/trace/ftrace.h | 3 ++- kernel/perf_counter.c | 30 ++++++++++++++++++++++++------ 3 files changed, 28 insertions(+), 7 deletions(-) (limited to 'include') diff --git a/include/linux/perf_counter.h b/include/linux/perf_counter.h index 2aabe43c1d04..a9d823a93fe8 100644 --- a/include/linux/perf_counter.h +++ b/include/linux/perf_counter.h @@ -369,6 +369,8 @@ enum perf_event_type { * * { u64 nr, * u64 ips[nr]; } && PERF_SAMPLE_CALLCHAIN + * { u32 size; + * char data[size];}&& PERF_SAMPLE_RAW * }; */ PERF_EVENT_SAMPLE = 9, diff --git a/include/trace/ftrace.h b/include/trace/ftrace.h index 7fb16d90e7b1..7167b9b97da2 100644 --- a/include/trace/ftrace.h +++ b/include/trace/ftrace.h @@ -685,7 +685,8 @@ static void ftrace_profile_##call(proto) \ pc = preempt_count(); \ \ __data_size = ftrace_get_offsets_##call(&__data_offsets, args); \ - __entry_size = ALIGN(__data_size + sizeof(*entry), sizeof(u64));\ + __entry_size = ALIGN(__data_size + sizeof(*entry) + sizeof(u32),\ + sizeof(u64)); \ \ do { \ char raw_data[__entry_size]; \ diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 546e62d62941..5229d1666fa5 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -2646,7 +2646,6 @@ static void perf_counter_output(struct perf_counter *counter, int nmi, u64 counter; } group_entry; struct perf_callchain_entry *callchain = NULL; - struct perf_raw_record *raw = NULL; int callchain_size = 0; u64 time; struct { @@ -2716,9 +2715,15 @@ static void perf_counter_output(struct perf_counter *counter, int nmi, } if (sample_type & PERF_SAMPLE_RAW) { - raw = data->raw; - if (raw) - header.size += raw->size; + int size = sizeof(u32); + + if (data->raw) + size += data->raw->size; + else + size += sizeof(u32); + + WARN_ON_ONCE(size & (sizeof(u64)-1)); + header.size += size; } ret = perf_output_begin(&handle, counter, header.size, nmi, 1); @@ -2784,8 +2789,21 @@ static void perf_counter_output(struct perf_counter *counter, int nmi, } } - if ((sample_type & PERF_SAMPLE_RAW) && raw) - perf_output_copy(&handle, raw->data, raw->size); + if (sample_type & PERF_SAMPLE_RAW) { + if (data->raw) { + perf_output_put(&handle, data->raw->size); + perf_output_copy(&handle, data->raw->data, data->raw->size); + } else { + struct { + u32 size; + u32 data; + } raw = { + .size = sizeof(u32), + .data = 0, + }; + perf_output_put(&handle, raw); + } + } perf_output_end(&handle); } -- cgit v1.2.3 From 2fc391112fb6f3424435a3aa2fda887497b5f807 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 10 Aug 2009 12:33:05 +0100 Subject: locking, sched: Give waitqueue spinlocks their own lockdep classes Give waitqueue spinlocks their own lockdep classes when they are initialised from init_waitqueue_head(). This means that struct wait_queue::func functions can operate other waitqueues. This is used by CacheFiles to catch the page from a backing fs being unlocked and to wake up another thread to take a copy of it. Signed-off-by: Peter Zijlstra Signed-off-by: David Howells Tested-by: Takashi Iwai Cc: linux-cachefs@redhat.com Cc: torvalds@osdl.org Cc: akpm@linux-foundation.org LKML-Reference: <20090810113305.17284.81508.stgit@warthog.procyon.org.uk> Signed-off-by: Ingo Molnar --- include/linux/wait.h | 9 ++++++++- kernel/wait.c | 5 +++-- 2 files changed, 11 insertions(+), 3 deletions(-) (limited to 'include') diff --git a/include/linux/wait.h b/include/linux/wait.h index 6788e1a4d4ca..cf3c2f5dba51 100644 --- a/include/linux/wait.h +++ b/include/linux/wait.h @@ -77,7 +77,14 @@ struct task_struct; #define __WAIT_BIT_KEY_INITIALIZER(word, bit) \ { .flags = word, .bit_nr = bit, } -extern void init_waitqueue_head(wait_queue_head_t *q); +extern void __init_waitqueue_head(wait_queue_head_t *q, struct lock_class_key *); + +#define init_waitqueue_head(q) \ + do { \ + static struct lock_class_key __key; \ + \ + __init_waitqueue_head((q), &__key); \ + } while (0) #ifdef CONFIG_LOCKDEP # define __WAIT_QUEUE_HEAD_INIT_ONSTACK(name) \ diff --git a/kernel/wait.c b/kernel/wait.c index ea7c3b4275cf..c4bd3d825f35 100644 --- a/kernel/wait.c +++ b/kernel/wait.c @@ -10,13 +10,14 @@ #include #include -void init_waitqueue_head(wait_queue_head_t *q) +void __init_waitqueue_head(wait_queue_head_t *q, struct lock_class_key *key) { spin_lock_init(&q->lock); + lockdep_set_class(&q->lock, key); INIT_LIST_HEAD(&q->task_list); } -EXPORT_SYMBOL(init_waitqueue_head); +EXPORT_SYMBOL(__init_waitqueue_head); void add_wait_queue(wait_queue_head_t *q, wait_queue_t *wait) { -- cgit v1.2.3 From 304703aba31a87903b8c0db8f5e6890cac2d596d Mon Sep 17 00:00:00 2001 From: Frederic Weisbecker Date: Mon, 10 Aug 2009 16:11:32 +0200 Subject: perf_counter: Subtract the buffer size field from the event record size We compute the perf raw sample size by aligning the raw ftrace event size plus the buffer size field itself. We do that instead of aligning only the perf raw sample size, so that we might economize some in some cases. But this buffer size field is not stored in the perf raw sample, we must then substract its size from the buffer once we computed the alignment unless we may get a useless u32 field in the buffer. Signed-off-by: Frederic Weisbecker Acked-by: Peter Zijlstra Cc: Arnaldo Carvalho de Melo Cc: Mike Galbraith Cc: Paul Mackerras LKML-Reference: <20090810141129.GA5124@nowhere> Signed-off-by: Ingo Molnar --- include/trace/ftrace.h | 8 +++++++- 1 file changed, 7 insertions(+), 1 deletion(-) (limited to 'include') diff --git a/include/trace/ftrace.h b/include/trace/ftrace.h index 7167b9b97da2..a05524fa245d 100644 --- a/include/trace/ftrace.h +++ b/include/trace/ftrace.h @@ -637,7 +637,12 @@ __attribute__((section("_ftrace_events"))) event_##call = { \ * pc = preempt_count(); * * __data_size = ftrace_get_offsets_(&__data_offsets, args); - * __entry_size = __data_size + sizeof(*entry); + * + * // Below we want to get the aligned size by taking into account + * // the u32 field that will later store the buffer size + * __entry_size = ALIGN(__data_size + sizeof(*entry) + sizeof(u32), + * sizeof(u64)); + * __entry_size -= sizeof(u32); * * do { * char raw_data[__entry_size]; <- allocate our sample in the stack @@ -687,6 +692,7 @@ static void ftrace_profile_##call(proto) \ __data_size = ftrace_get_offsets_##call(&__data_offsets, args); \ __entry_size = ALIGN(__data_size + sizeof(*entry) + sizeof(u32),\ sizeof(u64)); \ + __entry_size -= sizeof(u32); \ \ do { \ char raw_data[__entry_size]; \ -- cgit v1.2.3 From 1853db0e02ae4088f102b0d8e59e83dc98f93f03 Mon Sep 17 00:00:00 2001 From: Frederic Weisbecker Date: Mon, 10 Aug 2009 16:38:36 +0200 Subject: perf_counter: Zero dead bytes from ftrace raw samples size alignment After aligning the ftrace raw samples, there are dead bytes storing random data from the stack. We don't want to leak these to userspace, then zero these out. Before: 0x2de88 [0x50]: event: 9 . . ... raw event: size 80 bytes . 0000: 09 00 00 00 01 00 50 00 d0 c7 00 81 ff ff ff ff ......P........ . 0010: 68 01 00 00 68 01 00 00 2c 00 00 00 00 00 00 00 h...h...,...... . 0020: 2c 00 00 00 2b 00 01 02 68 01 00 00 68 01 00 00 ,...+...h...h.. . 0030: 6b 6f 6e 64 65 6d 61 6e 64 2f 30 00 00 00 00 00 kondemand/0.... . 0040: 68 01 00 00 40 7f 46 81 ff ff ff ff 00 10 1b 7f h...@.F........ ^ ^ ^ ^ Leak After: 0x2d318 [0x50]: event: 9 . . ... raw event: size 80 bytes . 0000: 09 00 00 00 01 00 50 00 d0 c7 00 81 ff ff ff ff ......P........ . 0010: 68 01 00 00 68 01 00 00 68 14 00 00 00 00 00 00 h...h...h...... . 0020: 2c 00 00 00 2b 00 01 02 68 01 00 00 68 01 00 00 ,...+...h...h.. . 0030: 6b 6f 6e 64 65 6d 61 6e 64 2f 30 00 00 00 00 00 kondemand/0.... . 0040: 68 01 00 00 a0 80 46 81 ff ff ff ff 00 00 00 00 h.....F........ ^ ^ ^ ^ Fixed Reported-by: Peter Zijlstra Signed-off-by: Frederic Weisbecker Cc: Frederic Weisbecker Cc: Peter Zijlstra Cc: Arnaldo Carvalho de Melo Cc: Mike Galbraith LKML-Reference: <1249915116-5210-1-git-send-email-fweisbec@gmail.com> Signed-off-by: Ingo Molnar Cc: Peter Zijlstra Cc: Arnaldo Carvalho de Melo Cc: Mike Galbraith --- include/trace/ftrace.h | 4 ++++ 1 file changed, 4 insertions(+) (limited to 'include') diff --git a/include/trace/ftrace.h b/include/trace/ftrace.h index a05524fa245d..f64fbaae781a 100644 --- a/include/trace/ftrace.h +++ b/include/trace/ftrace.h @@ -648,6 +648,9 @@ __attribute__((section("_ftrace_events"))) event_##call = { \ * char raw_data[__entry_size]; <- allocate our sample in the stack * struct trace_entry *ent; * + * zero dead bytes from alignment to avoid stack leak to userspace: + * + * *(u64 *)(&raw_data[__entry_size - sizeof(u64)]) = 0ULL; * entry = (struct ftrace_raw_ *)raw_data; * ent = &entry->ent; * tracing_generic_entry_update(ent, irq_flags, pc); @@ -698,6 +701,7 @@ static void ftrace_profile_##call(proto) \ char raw_data[__entry_size]; \ struct trace_entry *ent; \ \ + *(u64 *)(&raw_data[__entry_size - sizeof(u64)]) = 0ULL; \ entry = (struct ftrace_raw_##call *)raw_data; \ ent = &entry->ent; \ tracing_generic_entry_update(ent, irq_flags, pc); \ -- cgit v1.2.3 From 1ae88b2e446261c038f2c0c3150ffae142b227a2 Mon Sep 17 00:00:00 2001 From: Trond Myklebust Date: Wed, 12 Aug 2009 09:12:30 -0400 Subject: NFS: Fix an O_DIRECT Oops... We can't call nfs_readdata_release()/nfs_writedata_release() without first initialising and referencing args.context. Doing so inside nfs_direct_read_schedule_segment()/nfs_direct_write_schedule_segment() causes an Oops. We should rather be calling nfs_readdata_free()/nfs_writedata_free() in those cases. Looking at the O_DIRECT code, the "struct nfs_direct_req" is already referencing the nfs_open_context for us. Since the readdata and writedata structures carry a reference to that, we can simplify things by getting rid of the extra nfs_open_context references, so that we can replace all instances of nfs_readdata_release()/nfs_writedata_release(). Reported-by: Catalin Marinas Signed-off-by: Trond Myklebust Tested-by: Catalin Marinas Cc: stable@kernel.org Signed-off-by: Linus Torvalds --- fs/nfs/direct.c | 20 ++++++++++---------- fs/nfs/read.c | 6 ++---- fs/nfs/write.c | 6 ++---- include/linux/nfs_fs.h | 5 ++--- 4 files changed, 16 insertions(+), 21 deletions(-) (limited to 'include') diff --git a/fs/nfs/direct.c b/fs/nfs/direct.c index 489fc01a3204..e4e089a8f294 100644 --- a/fs/nfs/direct.c +++ b/fs/nfs/direct.c @@ -255,7 +255,7 @@ static void nfs_direct_read_release(void *calldata) if (put_dreq(dreq)) nfs_direct_complete(dreq); - nfs_readdata_release(calldata); + nfs_readdata_free(data); } static const struct rpc_call_ops nfs_read_direct_ops = { @@ -314,14 +314,14 @@ static ssize_t nfs_direct_read_schedule_segment(struct nfs_direct_req *dreq, data->npages, 1, 0, data->pagevec, NULL); up_read(¤t->mm->mmap_sem); if (result < 0) { - nfs_readdata_release(data); + nfs_readdata_free(data); break; } if ((unsigned)result < data->npages) { bytes = result * PAGE_SIZE; if (bytes <= pgbase) { nfs_direct_release_pages(data->pagevec, result); - nfs_readdata_release(data); + nfs_readdata_free(data); break; } bytes -= pgbase; @@ -334,7 +334,7 @@ static ssize_t nfs_direct_read_schedule_segment(struct nfs_direct_req *dreq, data->inode = inode; data->cred = msg.rpc_cred; data->args.fh = NFS_FH(inode); - data->args.context = get_nfs_open_context(ctx); + data->args.context = ctx; data->args.offset = pos; data->args.pgbase = pgbase; data->args.pages = data->pagevec; @@ -441,7 +441,7 @@ static void nfs_direct_free_writedata(struct nfs_direct_req *dreq) struct nfs_write_data *data = list_entry(dreq->rewrite_list.next, struct nfs_write_data, pages); list_del(&data->pages); nfs_direct_release_pages(data->pagevec, data->npages); - nfs_writedata_release(data); + nfs_writedata_free(data); } } @@ -534,7 +534,7 @@ static void nfs_direct_commit_release(void *calldata) dprintk("NFS: %5u commit returned %d\n", data->task.tk_pid, status); nfs_direct_write_complete(dreq, data->inode); - nfs_commitdata_release(calldata); + nfs_commit_free(data); } static const struct rpc_call_ops nfs_commit_direct_ops = { @@ -570,7 +570,7 @@ static void nfs_direct_commit_schedule(struct nfs_direct_req *dreq) data->args.fh = NFS_FH(data->inode); data->args.offset = 0; data->args.count = 0; - data->args.context = get_nfs_open_context(dreq->ctx); + data->args.context = dreq->ctx; data->res.count = 0; data->res.fattr = &data->fattr; data->res.verf = &data->verf; @@ -734,14 +734,14 @@ static ssize_t nfs_direct_write_schedule_segment(struct nfs_direct_req *dreq, data->npages, 0, 0, data->pagevec, NULL); up_read(¤t->mm->mmap_sem); if (result < 0) { - nfs_writedata_release(data); + nfs_writedata_free(data); break; } if ((unsigned)result < data->npages) { bytes = result * PAGE_SIZE; if (bytes <= pgbase) { nfs_direct_release_pages(data->pagevec, result); - nfs_writedata_release(data); + nfs_writedata_free(data); break; } bytes -= pgbase; @@ -756,7 +756,7 @@ static ssize_t nfs_direct_write_schedule_segment(struct nfs_direct_req *dreq, data->inode = inode; data->cred = msg.rpc_cred; data->args.fh = NFS_FH(inode); - data->args.context = get_nfs_open_context(ctx); + data->args.context = ctx; data->args.offset = pos; data->args.pgbase = pgbase; data->args.pages = data->pagevec; diff --git a/fs/nfs/read.c b/fs/nfs/read.c index 73ea5e8d66ce..12c9e66d3f1d 100644 --- a/fs/nfs/read.c +++ b/fs/nfs/read.c @@ -60,17 +60,15 @@ struct nfs_read_data *nfs_readdata_alloc(unsigned int pagecount) return p; } -static void nfs_readdata_free(struct nfs_read_data *p) +void nfs_readdata_free(struct nfs_read_data *p) { if (p && (p->pagevec != &p->page_array[0])) kfree(p->pagevec); mempool_free(p, nfs_rdata_mempool); } -void nfs_readdata_release(void *data) +static void nfs_readdata_release(struct nfs_read_data *rdata) { - struct nfs_read_data *rdata = data; - put_nfs_open_context(rdata->args.context); nfs_readdata_free(rdata); } diff --git a/fs/nfs/write.c b/fs/nfs/write.c index 0a0a2ff767c3..a34fae21fe10 100644 --- a/fs/nfs/write.c +++ b/fs/nfs/write.c @@ -87,17 +87,15 @@ struct nfs_write_data *nfs_writedata_alloc(unsigned int pagecount) return p; } -static void nfs_writedata_free(struct nfs_write_data *p) +void nfs_writedata_free(struct nfs_write_data *p) { if (p && (p->pagevec != &p->page_array[0])) kfree(p->pagevec); mempool_free(p, nfs_wdata_mempool); } -void nfs_writedata_release(void *data) +static void nfs_writedata_release(struct nfs_write_data *wdata) { - struct nfs_write_data *wdata = data; - put_nfs_open_context(wdata->args.context); nfs_writedata_free(wdata); } diff --git a/include/linux/nfs_fs.h b/include/linux/nfs_fs.h index fdffb413b192..f6b90240dd41 100644 --- a/include/linux/nfs_fs.h +++ b/include/linux/nfs_fs.h @@ -473,7 +473,6 @@ extern int nfs_writepages(struct address_space *, struct writeback_control *); extern int nfs_flush_incompatible(struct file *file, struct page *page); extern int nfs_updatepage(struct file *, struct page *, unsigned int, unsigned int); extern int nfs_writeback_done(struct rpc_task *, struct nfs_write_data *); -extern void nfs_writedata_release(void *); /* * Try to write back everything synchronously (but check the @@ -488,7 +487,6 @@ extern int nfs_wb_page_cancel(struct inode *inode, struct page* page); extern int nfs_commit_inode(struct inode *, int); extern struct nfs_write_data *nfs_commitdata_alloc(void); extern void nfs_commit_free(struct nfs_write_data *wdata); -extern void nfs_commitdata_release(void *wdata); #else static inline int nfs_commit_inode(struct inode *inode, int how) @@ -507,6 +505,7 @@ nfs_have_writebacks(struct inode *inode) * Allocate nfs_write_data structures */ extern struct nfs_write_data *nfs_writedata_alloc(unsigned int npages); +extern void nfs_writedata_free(struct nfs_write_data *); /* * linux/fs/nfs/read.c @@ -515,7 +514,6 @@ extern int nfs_readpage(struct file *, struct page *); extern int nfs_readpages(struct file *, struct address_space *, struct list_head *, unsigned); extern int nfs_readpage_result(struct rpc_task *, struct nfs_read_data *); -extern void nfs_readdata_release(void *data); extern int nfs_readpage_async(struct nfs_open_context *, struct inode *, struct page *); @@ -523,6 +521,7 @@ extern int nfs_readpage_async(struct nfs_open_context *, struct inode *, * Allocate nfs_read_data structures */ extern struct nfs_read_data *nfs_readdata_alloc(unsigned int npages); +extern void nfs_readdata_free(struct nfs_read_data *); /* * linux/fs/nfs3proc.c -- cgit v1.2.3 From 28402971d869e26271b25301011f667d3a5640c3 Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Thu, 13 Aug 2009 10:13:22 +0200 Subject: perf_counter: Provide hw_perf_counter_setup_online() APIs Provide weak aliases for hw_perf_counter_setup_online(). This is used by the BTS patches (for v2.6.32), but it interacts with fixes so propagate this upstream. (it has no effect as of yet) Also export perf_counter_output() to architecture code. Cc: Peter Zijlstra Cc: Paul Mackerras Cc: Frederic Weisbecker LKML-Reference: Signed-off-by: Ingo Molnar --- include/linux/perf_counter.h | 2 ++ kernel/perf_counter.c | 10 +++++++++- 2 files changed, 11 insertions(+), 1 deletion(-) (limited to 'include') diff --git a/include/linux/perf_counter.h b/include/linux/perf_counter.h index a9d823a93fe8..2b36afe6ae0f 100644 --- a/include/linux/perf_counter.h +++ b/include/linux/perf_counter.h @@ -694,6 +694,8 @@ struct perf_sample_data { extern int perf_counter_overflow(struct perf_counter *counter, int nmi, struct perf_sample_data *data); +extern void perf_counter_output(struct perf_counter *counter, int nmi, + struct perf_sample_data *data); /* * Return 1 for a software counter, 0 for a hardware counter diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index b0b20a07f394..e26d2fcfa320 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -88,6 +88,7 @@ void __weak hw_perf_disable(void) { barrier(); } void __weak hw_perf_enable(void) { barrier(); } void __weak hw_perf_counter_setup(int cpu) { barrier(); } +void __weak hw_perf_counter_setup_online(int cpu) { barrier(); } int __weak hw_perf_group_sched_in(struct perf_counter *group_leader, @@ -2630,7 +2631,7 @@ static u32 perf_counter_tid(struct perf_counter *counter, struct task_struct *p) return task_pid_nr_ns(p, counter->ns); } -static void perf_counter_output(struct perf_counter *counter, int nmi, +void perf_counter_output(struct perf_counter *counter, int nmi, struct perf_sample_data *data) { int ret; @@ -4592,6 +4593,11 @@ perf_cpu_notify(struct notifier_block *self, unsigned long action, void *hcpu) perf_counter_init_cpu(cpu); break; + case CPU_ONLINE: + case CPU_ONLINE_FROZEN: + hw_perf_counter_setup_online(cpu); + break; + case CPU_DOWN_PREPARE: case CPU_DOWN_PREPARE_FROZEN: perf_counter_exit_cpu(cpu); @@ -4616,6 +4622,8 @@ void __init perf_counter_init(void) { perf_cpu_notify(&perf_cpu_nb, (unsigned long)CPU_UP_PREPARE, (void *)(long)smp_processor_id()); + perf_cpu_notify(&perf_cpu_nb, (unsigned long)CPU_ONLINE, + (void *)(long)smp_processor_id()); register_cpu_notifier(&perf_cpu_nb); } -- cgit v1.2.3 From 3dab77fb1bf89664bb1c9544607159dcab6f7d57 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Thu, 13 Aug 2009 11:47:53 +0200 Subject: perf: Rework/fix the whole read vs group stuff Replace PERF_SAMPLE_GROUP with PERF_SAMPLE_READ and introduce PERF_FORMAT_GROUP to deal with group reads in a more generic way. This allows you to get group reads out of read() as well. Signed-off-by: Peter Zijlstra Cc: Corey J Ashford Cc: Paul Mackerras Cc: stephane eranian LKML-Reference: <20090813103655.117411814@chello.nl> Signed-off-by: Ingo Molnar --- include/linux/perf_counter.h | 47 ++++++-- kernel/perf_counter.c | 274 +++++++++++++++++++++++++++++++------------ 2 files changed, 238 insertions(+), 83 deletions(-) (limited to 'include') diff --git a/include/linux/perf_counter.h b/include/linux/perf_counter.h index 2b36afe6ae0f..b53f7006cc4e 100644 --- a/include/linux/perf_counter.h +++ b/include/linux/perf_counter.h @@ -115,7 +115,7 @@ enum perf_counter_sample_format { PERF_SAMPLE_TID = 1U << 1, PERF_SAMPLE_TIME = 1U << 2, PERF_SAMPLE_ADDR = 1U << 3, - PERF_SAMPLE_GROUP = 1U << 4, + PERF_SAMPLE_READ = 1U << 4, PERF_SAMPLE_CALLCHAIN = 1U << 5, PERF_SAMPLE_ID = 1U << 6, PERF_SAMPLE_CPU = 1U << 7, @@ -127,16 +127,32 @@ enum perf_counter_sample_format { }; /* - * Bits that can be set in attr.read_format to request that - * reads on the counter should return the indicated quantities, - * in increasing order of bit value, after the counter value. + * The format of the data returned by read() on a perf counter fd, + * as specified by attr.read_format: + * + * struct read_format { + * { u64 value; + * { u64 time_enabled; } && PERF_FORMAT_ENABLED + * { u64 time_running; } && PERF_FORMAT_RUNNING + * { u64 id; } && PERF_FORMAT_ID + * } && !PERF_FORMAT_GROUP + * + * { u64 nr; + * { u64 time_enabled; } && PERF_FORMAT_ENABLED + * { u64 time_running; } && PERF_FORMAT_RUNNING + * { u64 value; + * { u64 id; } && PERF_FORMAT_ID + * } cntr[nr]; + * } && PERF_FORMAT_GROUP + * }; */ enum perf_counter_read_format { PERF_FORMAT_TOTAL_TIME_ENABLED = 1U << 0, PERF_FORMAT_TOTAL_TIME_RUNNING = 1U << 1, PERF_FORMAT_ID = 1U << 2, + PERF_FORMAT_GROUP = 1U << 3, - PERF_FORMAT_MAX = 1U << 3, /* non-ABI */ + PERF_FORMAT_MAX = 1U << 4, /* non-ABI */ }; #define PERF_ATTR_SIZE_VER0 64 /* sizeof first published struct */ @@ -343,10 +359,8 @@ enum perf_event_type { * struct { * struct perf_event_header header; * u32 pid, tid; - * u64 value; - * { u64 time_enabled; } && PERF_FORMAT_ENABLED - * { u64 time_running; } && PERF_FORMAT_RUNNING - * { u64 parent_id; } && PERF_FORMAT_ID + * + * struct read_format values; * }; */ PERF_EVENT_READ = 8, @@ -364,11 +378,22 @@ enum perf_event_type { * { u32 cpu, res; } && PERF_SAMPLE_CPU * { u64 period; } && PERF_SAMPLE_PERIOD * - * { u64 nr; - * { u64 id, val; } cnt[nr]; } && PERF_SAMPLE_GROUP + * { struct read_format values; } && PERF_SAMPLE_READ * * { u64 nr, * u64 ips[nr]; } && PERF_SAMPLE_CALLCHAIN + * + * # + * # The RAW record below is opaque data wrt the ABI + * # + * # That is, the ABI doesn't make any promises wrt to + * # the stability of its content, it may vary depending + * # on event, hardware, kernel version and phase of + * # the moon. + * # + * # In other words, PERF_SAMPLE_RAW contents are not an ABI. + * # + * * { u32 size; * char data[size];}&& PERF_SAMPLE_RAW * }; diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 3dd4339589a0..b8c6b97a20a3 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1692,7 +1692,32 @@ static int perf_release(struct inode *inode, struct file *file) return 0; } -static u64 perf_counter_read_tree(struct perf_counter *counter) +static int perf_counter_read_size(struct perf_counter *counter) +{ + int entry = sizeof(u64); /* value */ + int size = 0; + int nr = 1; + + if (counter->attr.read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) + size += sizeof(u64); + + if (counter->attr.read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) + size += sizeof(u64); + + if (counter->attr.read_format & PERF_FORMAT_ID) + entry += sizeof(u64); + + if (counter->attr.read_format & PERF_FORMAT_GROUP) { + nr += counter->group_leader->nr_siblings; + size += sizeof(u64); + } + + size += entry * nr; + + return size; +} + +static u64 perf_counter_read_value(struct perf_counter *counter) { struct perf_counter *child; u64 total = 0; @@ -1704,14 +1729,96 @@ static u64 perf_counter_read_tree(struct perf_counter *counter) return total; } +static int perf_counter_read_entry(struct perf_counter *counter, + u64 read_format, char __user *buf) +{ + int n = 0, count = 0; + u64 values[2]; + + values[n++] = perf_counter_read_value(counter); + if (read_format & PERF_FORMAT_ID) + values[n++] = primary_counter_id(counter); + + count = n * sizeof(u64); + + if (copy_to_user(buf, values, count)) + return -EFAULT; + + return count; +} + +static int perf_counter_read_group(struct perf_counter *counter, + u64 read_format, char __user *buf) +{ + struct perf_counter *leader = counter->group_leader, *sub; + int n = 0, size = 0, err = -EFAULT; + u64 values[3]; + + values[n++] = 1 + leader->nr_siblings; + if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) { + values[n++] = leader->total_time_enabled + + atomic64_read(&leader->child_total_time_enabled); + } + if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) { + values[n++] = leader->total_time_running + + atomic64_read(&leader->child_total_time_running); + } + + size = n * sizeof(u64); + + if (copy_to_user(buf, values, size)) + return -EFAULT; + + err = perf_counter_read_entry(leader, read_format, buf + size); + if (err < 0) + return err; + + size += err; + + list_for_each_entry(sub, &leader->sibling_list, list_entry) { + err = perf_counter_read_entry(counter, read_format, + buf + size); + if (err < 0) + return err; + + size += err; + } + + return size; +} + +static int perf_counter_read_one(struct perf_counter *counter, + u64 read_format, char __user *buf) +{ + u64 values[4]; + int n = 0; + + values[n++] = perf_counter_read_value(counter); + if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) { + values[n++] = counter->total_time_enabled + + atomic64_read(&counter->child_total_time_enabled); + } + if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) { + values[n++] = counter->total_time_running + + atomic64_read(&counter->child_total_time_running); + } + if (read_format & PERF_FORMAT_ID) + values[n++] = primary_counter_id(counter); + + if (copy_to_user(buf, values, n * sizeof(u64))) + return -EFAULT; + + return n * sizeof(u64); +} + /* * Read the performance counter - simple non blocking version for now */ static ssize_t perf_read_hw(struct perf_counter *counter, char __user *buf, size_t count) { - u64 values[4]; - int n; + u64 read_format = counter->attr.read_format; + int ret; /* * Return end-of-file for a read on a counter that is in @@ -1721,28 +1828,18 @@ perf_read_hw(struct perf_counter *counter, char __user *buf, size_t count) if (counter->state == PERF_COUNTER_STATE_ERROR) return 0; + if (count < perf_counter_read_size(counter)) + return -ENOSPC; + WARN_ON_ONCE(counter->ctx->parent_ctx); mutex_lock(&counter->child_mutex); - values[0] = perf_counter_read_tree(counter); - n = 1; - if (counter->attr.read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) - values[n++] = counter->total_time_enabled + - atomic64_read(&counter->child_total_time_enabled); - if (counter->attr.read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) - values[n++] = counter->total_time_running + - atomic64_read(&counter->child_total_time_running); - if (counter->attr.read_format & PERF_FORMAT_ID) - values[n++] = primary_counter_id(counter); + if (read_format & PERF_FORMAT_GROUP) + ret = perf_counter_read_group(counter, read_format, buf); + else + ret = perf_counter_read_one(counter, read_format, buf); mutex_unlock(&counter->child_mutex); - if (count < n * sizeof(u64)) - return -EINVAL; - count = n * sizeof(u64); - - if (copy_to_user(buf, values, count)) - return -EFAULT; - - return count; + return ret; } static ssize_t @@ -2631,6 +2728,79 @@ static u32 perf_counter_tid(struct perf_counter *counter, struct task_struct *p) return task_pid_nr_ns(p, counter->ns); } +static void perf_output_read_one(struct perf_output_handle *handle, + struct perf_counter *counter) +{ + u64 read_format = counter->attr.read_format; + u64 values[4]; + int n = 0; + + values[n++] = atomic64_read(&counter->count); + if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) { + values[n++] = counter->total_time_enabled + + atomic64_read(&counter->child_total_time_enabled); + } + if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) { + values[n++] = counter->total_time_running + + atomic64_read(&counter->child_total_time_running); + } + if (read_format & PERF_FORMAT_ID) + values[n++] = primary_counter_id(counter); + + perf_output_copy(handle, values, n * sizeof(u64)); +} + +/* + * XXX PERF_FORMAT_GROUP vs inherited counters seems difficult. + */ +static void perf_output_read_group(struct perf_output_handle *handle, + struct perf_counter *counter) +{ + struct perf_counter *leader = counter->group_leader, *sub; + u64 read_format = counter->attr.read_format; + u64 values[5]; + int n = 0; + + values[n++] = 1 + leader->nr_siblings; + + if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) + values[n++] = leader->total_time_enabled; + + if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) + values[n++] = leader->total_time_running; + + if (leader != counter) + leader->pmu->read(leader); + + values[n++] = atomic64_read(&leader->count); + if (read_format & PERF_FORMAT_ID) + values[n++] = primary_counter_id(leader); + + perf_output_copy(handle, values, n * sizeof(u64)); + + list_for_each_entry(sub, &leader->sibling_list, list_entry) { + n = 0; + + if (sub != counter) + sub->pmu->read(sub); + + values[n++] = atomic64_read(&sub->count); + if (read_format & PERF_FORMAT_ID) + values[n++] = primary_counter_id(sub); + + perf_output_copy(handle, values, n * sizeof(u64)); + } +} + +static void perf_output_read(struct perf_output_handle *handle, + struct perf_counter *counter) +{ + if (counter->attr.read_format & PERF_FORMAT_GROUP) + perf_output_read_group(handle, counter); + else + perf_output_read_one(handle, counter); +} + void perf_counter_output(struct perf_counter *counter, int nmi, struct perf_sample_data *data) { @@ -2642,10 +2812,6 @@ void perf_counter_output(struct perf_counter *counter, int nmi, struct { u32 pid, tid; } tid_entry; - struct { - u64 id; - u64 counter; - } group_entry; struct perf_callchain_entry *callchain = NULL; int callchain_size = 0; u64 time; @@ -2700,10 +2866,8 @@ void perf_counter_output(struct perf_counter *counter, int nmi, if (sample_type & PERF_SAMPLE_PERIOD) header.size += sizeof(u64); - if (sample_type & PERF_SAMPLE_GROUP) { - header.size += sizeof(u64) + - counter->nr_siblings * sizeof(group_entry); - } + if (sample_type & PERF_SAMPLE_READ) + header.size += perf_counter_read_size(counter); if (sample_type & PERF_SAMPLE_CALLCHAIN) { callchain = perf_callchain(data->regs); @@ -2760,26 +2924,8 @@ void perf_counter_output(struct perf_counter *counter, int nmi, if (sample_type & PERF_SAMPLE_PERIOD) perf_output_put(&handle, data->period); - /* - * XXX PERF_SAMPLE_GROUP vs inherited counters seems difficult. - */ - if (sample_type & PERF_SAMPLE_GROUP) { - struct perf_counter *leader, *sub; - u64 nr = counter->nr_siblings; - - perf_output_put(&handle, nr); - - leader = counter->group_leader; - list_for_each_entry(sub, &leader->sibling_list, list_entry) { - if (sub != counter) - sub->pmu->read(sub); - - group_entry.id = primary_counter_id(sub); - group_entry.counter = atomic64_read(&sub->count); - - perf_output_put(&handle, group_entry); - } - } + if (sample_type & PERF_SAMPLE_READ) + perf_output_read(&handle, counter); if (sample_type & PERF_SAMPLE_CALLCHAIN) { if (callchain) @@ -2818,8 +2964,6 @@ struct perf_read_event { u32 pid; u32 tid; - u64 value; - u64 format[3]; }; static void @@ -2831,34 +2975,20 @@ perf_counter_read_event(struct perf_counter *counter, .header = { .type = PERF_EVENT_READ, .misc = 0, - .size = sizeof(event) - sizeof(event.format), + .size = sizeof(event) + perf_counter_read_size(counter), }, .pid = perf_counter_pid(counter, task), .tid = perf_counter_tid(counter, task), - .value = atomic64_read(&counter->count), }; - int ret, i = 0; - - if (counter->attr.read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) { - event.header.size += sizeof(u64); - event.format[i++] = counter->total_time_enabled; - } - - if (counter->attr.read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) { - event.header.size += sizeof(u64); - event.format[i++] = counter->total_time_running; - } - - if (counter->attr.read_format & PERF_FORMAT_ID) { - event.header.size += sizeof(u64); - event.format[i++] = primary_counter_id(counter); - } + int ret; ret = perf_output_begin(&handle, counter, event.header.size, 0, 0); if (ret) return; - perf_output_copy(&handle, &event, event.header.size); + perf_output_put(&handle, event); + perf_output_read(&handle, counter); + perf_output_end(&handle); } @@ -3921,9 +4051,9 @@ perf_counter_alloc(struct perf_counter_attr *attr, atomic64_set(&hwc->period_left, hwc->sample_period); /* - * we currently do not support PERF_SAMPLE_GROUP on inherited counters + * we currently do not support PERF_FORMAT_GROUP on inherited counters */ - if (attr->inherit && (attr->sample_type & PERF_SAMPLE_GROUP)) + if (attr->inherit && (attr->read_format & PERF_FORMAT_GROUP)) goto done; switch (attr->type) { -- cgit v1.2.3