Messages in this thread | | | Date | Tue, 30 Sep 2008 13:56:36 -0400 (EDT) | From | Steven Rostedt <> | Subject | Re: [PATCH v10 Golden] Unified trace buffer |
| |
On Tue, 30 Sep 2008, Peter Zijlstra wrote: > > The problem with storing the page link information inside the page is > that it doesnt transfer to another address space, so if you do indeed > mmap these pages, then the link information is bogus. > > Of course, in such a situation you could ignore these headers, but > somehow that doesn't sound too apealing.
No that's not what I'm proposing. I'm proposing to allocate a page_header structure for every page we alloc, and make a link list of them. In other words:
struct ring_buffer_per_cpu { [...] struct list_head pages; [...] };
struct buffer_page { [...]; void *page; struct list_head list; [...]; };
In ring_buffer_allocate_cpu:
struct buffer_page *bpage; struct unsigned long addr;
[...]
for every page() { bpage = kzalloc(sizeof(*bpage), GFP_KERNEL); addr = get_free_page(); bpage->page = (void *)addr; list_add(&bpage->list, &cpu_buffer->pages); }
Obviously need to add the error checking, but you get the idea. Here I do not need to change any of the later logic, because we are still dealing with the buffer_page. I only need to update way to index the page which is already encapsulated in its own function.
-- Steve
| |