On Tue, 27 Jun 2017 10:57:55 +0200, Takashi Sakamoto wrote:
Hi,
On Jun 27 2017 17:15, Takashi Iwai wrote:
On Tue, 27 Jun 2017 10:09:37 +0200, Takashi Sakamoto wrote:
Hi,
On Jun 27 2017 00:09, Takashi Iwai wrote:
On Sun, 25 Jun 2017 06:41:19 +0200, Takashi Sakamoto wrote:
Handling mapping operation for status/control data includes some supplemental operations for fallback mode. It's better to have helper functions for this purpose.
This commit adds the helper functions.
src/pcm/pcm_hw.c | 48 ++++++++++++++++++++++++++++++++++-------------- 1 file changed, 34 insertions(+), 14 deletions(-)
diff --git a/src/pcm/pcm_hw.c b/src/pcm/pcm_hw.c index a648d12c..abf4afe0 100644 --- a/src/pcm/pcm_hw.c +++ b/src/pcm/pcm_hw.c @@ -31,6 +31,7 @@ #include <stdlib.h> #include <stddef.h> #include <unistd.h> +#include <stdbool.h> #include <signal.h> #include <string.h> #include <fcntl.h> @@ -866,7 +867,7 @@ static snd_pcm_sframes_t snd_pcm_hw_readn(snd_pcm_t *pcm, void **bufs, snd_pcm_u return xfern.result; } -static int snd_pcm_hw_mmap_status(snd_pcm_t *pcm) +static int map_status_data(snd_pcm_t *pcm) { snd_pcm_hw_t *hw = pcm->private_data; struct snd_pcm_sync_ptr sync_ptr; @@ -900,7 +901,7 @@ static int snd_pcm_hw_mmap_status(snd_pcm_t *pcm) return 0; } -static int snd_pcm_hw_mmap_control(snd_pcm_t *pcm) +static int map_control_data(snd_pcm_t *pcm) { snd_pcm_hw_t *hw = pcm->private_data; void *ptr; @@ -922,10 +923,28 @@ static int snd_pcm_hw_mmap_control(snd_pcm_t *pcm) return 0; } -static int snd_pcm_hw_munmap_status(snd_pcm_t *pcm) +static int map_status_and_control_data(snd_pcm_t *pcm, bool force_fallback) { snd_pcm_hw_t *hw = pcm->private_data; int err;
- hw->sync_ptr_ioctl = (int)force_fallback;
- err = map_status_data(pcm);
- if (err < 0)
return err;- err = map_control_data(pcm);
- if (err < 0)
return err;This would leave the status mmap in the error path?
In the error path, snd_pcm_hw_open_fd() calls snd_pcm_close(), then The status data is going to be unmapped.
snd_pcm_hw_open_fd() ->snd_pcm_close() ->snd_pcm_hw_close() ->unmap_status_and_control_data() ->unmap_status_data() ->munmap(2) or free(3)
Ah then there was already a bug there. It free/munmap unconditionally for both status and control even for an error path of the mmap of control (while status was mmapped).
Yep. These codes were originally written with a loose manner. But it would be allowed because both 'mmap(NULL, non-zero)' and 'free(NULL)' generates no error and 'sync_ptr_ioctl' works fine, and 'sync_ptr' is assigned to NULL after deallocated.
Yes, that's a bit fragile way, but it works.
BTW, the whole patchset seems missing your sign-off. I can add it manually, though.
Takashi