On Fri, 14 Feb 2025 16:13:32 +0900 Masami Hiramatsu (Google) <[email protected]> wrote:
> On Thu, 13 Feb 2025 21:21:47 -0500 > Steven Rostedt <[email protected]> wrote: > > > On Fri, 14 Feb 2025 11:07:22 +0900 > > Masami Hiramatsu (Google) <[email protected]> wrote: > > > > > This may be good for fixing crash short term but may not allow us to read > > > the > > > subbufs which is read right before crash. Can we also add an option to > > > reset > > > the read pointer in the previous boot for the persistent ring buffer? > > > > I'm not sure what you mean here. > > > > Note, this is just for mmapping the buffers. Currently we never tried it as > > if we did, it would have crashed. But this does not affect normal reads. > > But reading buffer via mmap() is already supported. Can we read all pages, > which had been read by trace_pipe_raw in previous boot, again on this boot? It's not supported. If you try it, it will crash. This prevents reading via mmap() on a boot buffer. I don't know what you are asking. Once this patch is applied, mmap() will always fail on the boot buffer before or after you start it. > > Anyway, I made another patch to fix it to allow mmap() on persistent ring > buffer here. I thought I can get the phys_addr from struct vm_area, but > it could not for vmap()'d area. So this stores the phys_addr in trace_buffer. Which is way too complex and intrusive. If you want to allow mapping, all it needs is this: diff --git a/kernel/trace/ring_buffer.c b/kernel/trace/ring_buffer.c index 07b421115692..9339adc88ad5 100644 --- a/kernel/trace/ring_buffer.c +++ b/kernel/trace/ring_buffer.c @@ -5927,12 +5941,18 @@ static void rb_clear_buffer_page(struct buffer_page *page) static void rb_update_meta_page(struct ring_buffer_per_cpu *cpu_buffer) { struct trace_buffer_meta *meta = cpu_buffer->meta_page; + struct page *page; if (!meta) return; meta->reader.read = cpu_buffer->reader_page->read; - meta->reader.id = cpu_buffer->reader_page->id; + /* For boot buffers, the id is the index */ + if (cpu_buffer->ring_meta) + meta->reader.id = rb_meta_subbuf_idx(cpu_buffer->ring_meta, + cpu_buffer->reader_page->page); + else + meta->reader.id = cpu_buffer->reader_page->id; meta->reader.lost_events = cpu_buffer->lost_events; meta->entries = local_read(&cpu_buffer->entries); @@ -5940,7 +5960,12 @@ static void rb_update_meta_page(struct ring_buffer_per_cpu *cpu_buffer) meta->read = cpu_buffer->read; /* Some archs do not have data cache coherency between kernel and user-space */ - flush_dcache_folio(virt_to_folio(cpu_buffer->meta_page)); + if (virt_addr_valid(cpu_buffer->meta_page)) + page = virt_to_page(cpu_buffer->meta_page); + else + page = vmalloc_to_page(cpu_buffer->meta_page); + + flush_dcache_folio(page_folio(page)); } static void @@ -7041,7 +7066,10 @@ static int __rb_map_vma(struct ring_buffer_per_cpu *cpu_buffer, goto out; } - page = virt_to_page((void *)cpu_buffer->subbuf_ids[s]); + if (virt_addr_valid(cpu_buffer->subbuf_ids[s])) + page = virt_to_page((void *)cpu_buffer->subbuf_ids[s]); + else + page = vmalloc_to_page((void *)cpu_buffer->subbuf_ids[s]); for (; off < (1 << (subbuf_order)); off++, page++) { if (p >= nr_pages) @@ -7187,6 +7215,7 @@ int ring_buffer_map_get_reader(struct trace_buffer *buffer, int cpu) unsigned long missed_events; unsigned long reader_size; unsigned long flags; + struct page *page; cpu_buffer = rb_get_mapped_buffer(buffer, cpu); if (IS_ERR(cpu_buffer)) @@ -7255,7 +7291,12 @@ int ring_buffer_map_get_reader(struct trace_buffer *buffer, int cpu) out: /* Some archs do not have data cache coherency between kernel and user-space */ - flush_dcache_folio(virt_to_folio(cpu_buffer->reader_page->page)); + if (virt_addr_valid(cpu_buffer->meta_page)) + page = virt_to_page(cpu_buffer->meta_page); + else + page = vmalloc_to_page(cpu_buffer->meta_page); + + flush_dcache_folio(page_folio(page)); rb_update_meta_page(cpu_buffer); But this still doesn't work, as the index is still off (I'm still fixing it). The persistent ring buffer uses the page->id for one thing but the user space mmap() uses it for something else. -- Steve
