Re: [PATCH v2 03/13] btrace, linux: add perf event buffer abstraction

[Pedro Alves <palves at redhat dot com>]

On 11/20/2014 10:47 AM, Markus Metzger wrote:

> -  volatile struct perf_event_mmap_page *header;
> +  struct perf_event_buffer *pevent;
>    const uint8_t *begin, *end, *start;
> -  unsigned long data_head, data_tail, retries = 5;
> -  size_t buffer_size, size;
> +  unsigned long long data_head, data_tail, buffer_size, size;
> +  size_t retries = 5;

I notice that this is changing types from long to long long.
Also, a bit odd that retries is now size_t, given it's just a counter.

> +#if HAVE_LINUX_PERF_EVENT_H
> +/* A Linux perf event buffer.  */
> +struct perf_event_buffer
> +{
> +  /* The mapped memory.  */
> +  const uint8_t *mem;
> +
> +  /* The size of the mapped memory in bytes.  */
> +  unsigned long long size;
> +
> +  /* A pointer to the data_head field for this buffer. */
> +  volatile unsigned long long *data_head;
> +
> +  /* The data_head value from the last read.  */
> +  unsigned long long last_head;
> +};

Isn't there a better type to use here instead of "long long"?
Why not size_t, if a host buffer size; uint64_t, if it's a fixed
buffer format, assuming 64-bit here; or ULONGEST, the widest
integer we support?

> +#endif /* HAVE_LINUX_PERF_EVENT_H */


Thanks,
Pedro Alves