RFC: Do we want a new igt_write() function?

Peter Senna Tschudin peter.senna at linux.intel.com
Mon Jan 20 13:39:16 UTC 2025


Dear List,

I find myself wanting to repeat some small variation of a write function
that retries on recoverable errors and that aborts on unrecoverable errors.
Is the idea of an igt_write() function a welcome addition to lib/igt_core?

Something like:

#define MAX_WRITE_RETRIES 5
/**
 * igt_write - Writes the buffer to the file descriptor retrying when possible.
 * @fd: The file descriptor to write to.
 * @buf: Pointer to the data to write.
 * @count: Total number of bytes to write.
 *
 * Returns the total number of bytes written on success, or -1 on failure.
 */
ssize_t igt_write(int fd, const void *buf, size_t count) {
	const char *ptr = buf;
	size_t remaining = count;
	ssize_t written;
	int retries = 0;

	while (remaining > 0) {
		written = write(fd, ptr, remaining);
		if (written > 0) {
			ptr += written;
			remaining -= written;
		} else if (written == -1) {
			if (errno == EINTR || errno == EAGAIN) {
				/* Retry for recoverable errors */
				if (++retries > MAX_WRITE_RETRIES) {
					igt_err("igt_write: Exceeded retry limit\n");
					return -1;
				}
				continue;
			} else {
				/* Log unrecoverable error */
				igt_err("igt_write: unrecoverable write error");
				return -1;
			}
		}
	}
	return count;
}

Thanks,

Peter


More information about the igt-dev mailing list