On Tue, 1 Apr 2025 18:16:40 -0400
Steven Rostedt <[email protected]> wrote:
> Note, your patch to free the persistent ring buffer wasn't fully
> functional, as it only did the "vunmap()". That doesn't return the buffer
> back to the buddy allocator. What you saw was just the freeing of all the
> other descriptors that make up a trace instance.
>
> Before this patch:
>
> ~# free
> total used free shared buff/cache
> available
> Mem: 8185908 297404 7825896 916 162288
> 7888504
> Swap: 7812092 0 7812092
> ~# rmdir /sys/kernel/tracing/instances/boot_mapped
> ~# free
> total used free shared buff/cache
> available
> Mem: 8206384 297956 7845904 916 162260
> 7908428
> Swap: 7812092 0 7812092
>
> Amount freed: 7845904 - 7825896 = 20008 (20M)
>
> After this patch:
>
> ~# free
> total used free shared buff/cache
> available
> Mem: 8185912 301808 7820696 920 162860
> 7884104
> Swap: 7812092 0 7812092
> ~# rmdir /sys/kernel/tracing/instances/boot_mapped
> ~# free
> total used free shared buff/cache
> available
> Mem: 8226868 295968 7867644 920 162836
> 7930900
> Swap: 7812092 0 7812092
>
> Amount freed: 7867644 - 7820696 = 46948 (46M)!
Bah! My patch is buggy. Yeah, your code released the memory properly, as you
had:
reserve_mem_release_by_name()
which does do free_reserve_area() on the memory mapping.
Which means this patch is buggy and freed the same memory twice.
OK, time for v5 to fix that :-p
-- Steve
