Jérôme Pouiller <jerome.pouiller@xxxxxxxxxx> writes:
> On Thursday 7 October 2021 12:49:47 CEST Kalle Valo wrote:
>> CAUTION: This email originated from outside of the organization. Do
>> not click links or open attachments unless you recognize the sender
>> and know the content is safe.
>>
>>
>> Kalle Valo <kvalo@xxxxxxxxxxxxxx> writes:
>>
>> > Jérôme Pouiller <jerome.pouiller@xxxxxxxxxx> writes:
>> >
>> >>> >> >> I'm not really fond of having this kind of ASCII based parser in the
>> >>> >> >> kernel. Do you have an example compressed file somewhere?
>> >>> >> >
>> >>> >> > An example of uncompressed configuration file can be found here[1]. Once
>> >>> >> > compressed with [2], you get:
>> >>> >> >
>> >>> >> > {a:{a:4,b:1},b:{a:{a:4,b:0,c:0,d:0,e:A},b:{a:4,b:0,c:0,d:0,e:B},c:{a:4,b:0,c:0,d:0,e:C},d:{a:4,b:0,c:0,d:0,e:D},e:{a:4,b:0,c:0,d:0,e:E},f:{a:4,b:0,c:0,d:0,e:F},g:{a:4,b:0,c:0,d:0,e:G},h:{a:4,b:0,c:0,d:0,e:H},i:{a:4,b:0,c:0,d:0,e:I},j:{a:4,b:0,c:0,d:0,e:J},k:{a:4,b:0,c:0,d:0,e:K},l:{a:4,b:0,c:0,d:1,e:L},m:{a:4,b:0,c:0,d:1,e:M}},c:{a:{a:4},b:{a:6},c:{a:6,c:0},d:{a:6},e:{a:6},f:{a:6}},e:{b:0,c:1},h:{e:0,a:50,b:0,d:0,c:[{a:1,b:[0,0,0,0,0,0]},{a:2,b:[0,0,0,0,0,0]},{a:[3,9],b:[0,0,0,0,0,0]},{a:A,b:[0,0,0,0,0,0]},{a:B,b:[0,0,0,0,0,0]},{a:[C,D],b:[0,0,0,0,0,0]},{a:E,b:[0,0,0,0,0,0]}]},j:{a:0,b:0}}
>> >>> >>
>> >>> >> So what's the grand idea with this braces format? I'm not getting it.
>> >>> >
>> >>> > - It allows to describe a tree structure
>> >>> > - It is ascii (easy to dump, easy to copy-paste)
>> >>> > - It is small (as I explain below, size matters)
>> >>> > - Since it is similar to JSON, the structure is obvious to many people
>> >>> >
>> >>> > Anyway, I am not the author of that and I have to deal with it.
>> >>>
>> >>> I'm a supported for JSON like formats, flexibility and all that. But
>> >>> they belong to user space, not kernel.
>> >>>
>> >>> >> Usually the drivers just consider this kind of firmware configuration
>> >>> >> data as a binary blob and dump it to the firmware, without knowing what
>> >>> >> the data contains. Can't you do the same?
>> >>> >
>> >>> > [I didn't had received this mail :( ]
>> >>> >
>> >>> > The idea was also to send it as a binary blob. However, the firmware use
>> >>> > a limited buffer (1500 bytes) to parse it. In most of case the PDS exceeds
>> >>> > this size. So, we have to split the PDS before to send it.
>> >>> >
>> >>> > Unfortunately, we can't split it anywhere. The PDS is a tree structure and
>> >>> > the firmware expects to receive a well formatted tree.
>> >>> >
>> >>> > So, the easiest way to send it to the firmware is to split the tree
>> >>> > between each root nodes and send each subtree separately (see also the
>> >>> > comment above wfx_send_pds()).
>> >>> >
>> >>> > Anyway, someone has to cook this configuration before to send it to the
>> >>> > firmware. This could be done by a script outside of the kernel. Then we
>> >>> > could change the input format to simplify a bit the processing in the
>> >>> > kernel.
>> >>>
>> >>> I think a binary file with TLV format would be much better, but I'm sure
>> >>> there also other good choises.
>> >>>
>> >>> > However, the driver has already some users and I worry that changing
>> >>> > the input format would lead to a mess.
>> >>>
>> >>> You can implement a script which converts the old format to the new
>> >>> format. And you can use different naming scheme in the new format so
>> >>> that we don't accidentally load the old format. And even better if you
>> >>> add a some kind of signature in the new format and give a proper error
>> >>> from the driver if it doesn't match.
>> >>
>> >> Ok. I am going to change the input format. I think the new function is
>> >> going to look like:
>> >>
>> >> int wfx_send_pds(struct wfx_dev *wdev, u8 *buf, size_t buf_len)
>> >> {
>> >> int ret;
>> >> int start = 0;
>> >>
>> >> if (buf[start] != '{') {
>> >> dev_err(wdev->dev, "valid PDS start with '{'. Did you forget to compress it?\n");
>> >> return -EINVAL;
>> >> }
>> >> while (start < buf_len) {
>> >> len = strnlen(buf + start, buf_len - start);
>> >> if (len > WFX_PDS_MAX_SIZE) {
>> >> dev_err(wdev->dev, "PDS chunk is too big (legacy format?)\n");
>> >> return -EINVAL;
>> >> }
>> >> dev_dbg(wdev->dev, "send PDS '%s'\n", buf + start);
>> >> ret = wfx_hif_configuration(wdev, buf + start, len);
>> >> /* FIXME: Add error handling here */
>> >> start += len;
>> >> }
>> >> return 0;
>> >
>> > Did you read at all what I wrote above? Please ditch the ASCII format
>> > completely.
>>
>> Sorry, I read this too hastily. I just saw "buf[start] != '{'" and
>> assumed this is the same ASCII format, but not sure anymore. Can you
>> explain what changes you made now?
>
> The script I am going to write will compute where the PDS have to be split
> (this work is currently done by the driver). The script will add a
> separating character (let's say '\0') between each chunk.
>
> The driver will just have to find the separating character, send the
> chunk and repeat.
I would forget ASCII altogether and implement a proper binary format
like TLV. For example, ath10k uses TLV with board-2.bin files (grep for
enum ath10k_bd_ie_type).
Also I recommend changing the file "signature" ('{') to something else
so that the driver detects incorrect formats. And maybe even use suffix
.pds2 or something like that to make it more obvious and avoid
confusion?
--
https://patchwork.kernel.org/project/linux-wireless/list/
https://wireless.wiki.kernel.org/en/developers/documentation/submittingpatches
