Hangup on => client key exchange record

[Dmitry_Shabrov]

When working with MBETLS, it hangs on calling the static int ecdh_make_public_internal function (ecdh.c, 485) (=> client key exchange record)

I wrote about it in my post earlier

Https://test.mosquitto.org/

I observe incorrect work only when communicating with my server. Incorrect work is to switch to HardFault when issuing a message => write client key exchange. As far as I could step through the steps in the debugger, I determined that the stack was being destroyed along the chain: ssl_cli.c(2936) → ecdh.c(542) → ecdh.c(514) → ecdh.c(102) → ecdh.c(84) → ecp.c(2372) → ecp. c(1934) → ecp.c(1823) → ecp.c(1237) Return back from subroutine ecp.c(1237) → ecp.c(1823) → ecp.c(1934) and then HardFault. …

Printout and data analysis show incorrect data being passed in the function:

ctx->grp.nbits ← 0x209
ctx->grp.pbits ← 0x209

This hangup happens in low-level functions:
MBEDTLS_MPI_CHK(ecp_mul_comb( grp, R, m, P, f_rng, p_rng, rs_ctx));
MBEDTLS_MPI_CHK(ecp_mul_mxz( grp, R, m, P, f_rng, p_rng));
MBEDTLS_MPI_CHK(mbedtls_ecp_check_privkey(grp, m));
MBEDTLS_MPI_CHK(mbedtls_ecp_check_pubkey(grp, P));

spontaneous return from one function processor go to HardFault in next function.
If I correct the data as follows:

ctx->grp.nbits ← 0x100
ctx->grp.pbits ← 0x100

in static int ecdh_make_public_internal (ecdh.c, 485), this effect disappears completely. I believe that no input data should lead to a similar effect.