Re: [PATCH 1/2] soc: qcom: qmi: Fix "invalid data length" in encoder

On Wed, Feb 18, 2026 at 10:00:22AM -0800, Jeff Johnson wrote:
> On 2/16/2026 7:25 AM, Bjorn Andersson wrote:
> > It might very well be that the underlying bug is my expectation that
> > elem_size should be reflected in the struct and not only in the encoded
> > message, and hence what I wrote in https://github.com/linux-msm/qmic.
> > Perhaps the length-specifier of an array should always be u32?
> > 
> > @Chris, what does the downstream generator produce here?
> 
> Is this behavior just constrained to QMI_DATA_LEN TLVs?
> 
> I'm looking at downstream Android WLAN code and it has the same discrepancy,
> so it appears the code generator is always producing a u32 member in the host
> struct to hold a QMI_DATA_LEN member even though the actual element size as
> defined in the qmi_elem_info array is either sizeof(u8) or sizeof(u16).
> 
> Does this issue get fixed if we change the member in the host struct, i.e. for
> the issue mentioned (that I chopped off) modify:
> struct qmi_wlanfw_request_mem_ind_msg_v01 {
> -	u32 mem_seg_len;
> +	u8 mem_seg_len;
> 

Yes, that would fix the problem.

But if your elem_info arrays are coming from the downstream tool, then I
think the correct way to fix this would be to move in the other
direction and turn some those u8/u16 members elsewhere into u32.

I have the downstream code generator source here somewhere...

Regards,
Bjorn