/* * Copyright 2013-2020 Software Radio Systems Limited * * This file is part of srsLTE. * * srsLTE is free software: you can redistribute it and/or modify * it under the terms of the GNU Affero General Public License as * published by the Free Software Foundation, either version 3 of * the License, or (at your option) any later version. * * srsLTE is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the * GNU Affero General Public License for more details. * * A copy of the GNU Affero General Public License can be found in * the LICENSE file in the top-level directory of this distribution * and at http://www.gnu.org/licenses/. * */ #include "scheduler_test_common.h" #include "srsenb/hdr/stack/mac/scheduler.h" #include "srsenb/hdr/stack/upper/common_enb.h" #include "srslte/mac/pdu.h" #include "srslte/common/test_common.h" #include using namespace srsenb; /*************************** * Random Utils **************************/ std::default_random_engine rand_gen; float ::srsenb::randf() { static std::uniform_real_distribution unif_dist(0, 1.0); return unif_dist(rand_gen); } void ::srsenb::set_randseed(uint64_t seed) { rand_gen = std::default_random_engine(seed); } std::default_random_engine& ::srsenb::get_rand_gen() { return rand_gen; } /*************************** * Sched Testers **************************/ int output_sched_tester::test_pusch_collisions(const tti_params_t& tti_params, const sched_interface::ul_sched_res_t& ul_result, prbmask_t& ul_allocs) const { uint32_t nof_prb = cell_params.nof_prb(); ul_allocs.resize(nof_prb); ul_allocs.reset(); auto try_ul_fill = [&](prb_interval alloc, const char* ch_str, bool strict = true) { CONDERROR(alloc.stop() > nof_prb, "Allocated RBs %s out-of-bounds\n", alloc.to_string().c_str()); CONDERROR(alloc.empty(), "Allocations must have at least one PRB\n"); if (strict and ul_allocs.any(alloc.start(), alloc.stop())) { TESTERROR("Collision Detected of %s alloc=%s and cumulative_mask=0x%s\n", ch_str, alloc.to_string().c_str(), ul_allocs.to_hex().c_str()); } ul_allocs.fill(alloc.start(), alloc.stop(), true); return SRSLTE_SUCCESS; }; /* TEST: Check if there is space for PRACH */ bool is_prach_tti_tx_ul = srslte_prach_tti_opportunity_config_fdd(cell_params.cfg.prach_config, tti_params.tti_tx_ul, -1); if (is_prach_tti_tx_ul) { try_ul_fill({cell_params.cfg.prach_freq_offset, cell_params.cfg.prach_freq_offset + 6}, "PRACH"); } /* TEST: check collisions in PUCCH */ bool strict = nof_prb != 6 or (not is_prach_tti_tx_ul); // and not tti_data.ul_pending_msg3_present); try_ul_fill({0, (uint32_t)cell_params.cfg.nrb_pucch}, "PUCCH", strict); try_ul_fill({cell_params.cfg.cell.nof_prb - cell_params.cfg.nrb_pucch, (uint32_t)cell_params.cfg.cell.nof_prb}, "PUCCH", strict); /* TEST: check collisions in the UL PUSCH */ for (uint32_t i = 0; i < ul_result.nof_dci_elems; ++i) { uint32_t L, RBstart; srslte_ra_type2_from_riv(ul_result.pusch[i].dci.type2_alloc.riv, &L, &RBstart, nof_prb, nof_prb); strict = ul_result.pusch[i].needs_pdcch or nof_prb != 6; // Msg3 may collide with PUCCH at PRB==6 try_ul_fill({RBstart, RBstart + L}, "PUSCH", strict); // ue_stats[ul_result.pusch[i].dci.rnti].nof_ul_rbs += L; } return SRSLTE_SUCCESS; } int output_sched_tester::test_pdsch_collisions(const tti_params_t& tti_params, const sched_interface::dl_sched_res_t& dl_result, rbgmask_t& rbgmask) const { srslte::bounded_bitset<100, true> dl_allocs(cell_params.cfg.cell.nof_prb), alloc_mask(cell_params.cfg.cell.nof_prb); auto try_dl_mask_fill = [&](const srslte_dci_dl_t& dci, const char* channel) { if (extract_dl_prbmask(cell_params.cfg.cell, dci, &alloc_mask) != SRSLTE_SUCCESS) { return SRSLTE_ERROR; } if ((dl_allocs & alloc_mask).any()) { TESTERROR("Detected collision in the DL %s allocation (%s intersects %s)\n", channel, dl_allocs.to_string().c_str(), alloc_mask.to_string().c_str()); } dl_allocs |= alloc_mask; return SRSLTE_SUCCESS; }; // Decode BC allocations, check collisions, and fill cumulative mask for (uint32_t i = 0; i < dl_result.nof_bc_elems; ++i) { TESTASSERT(try_dl_mask_fill(dl_result.bc[i].dci, "BC") == SRSLTE_SUCCESS); } // Decode RAR allocations, check collisions, and fill cumulative mask for (uint32_t i = 0; i < dl_result.nof_rar_elems; ++i) { TESTASSERT(try_dl_mask_fill(dl_result.rar[i].dci, "RAR") == SRSLTE_SUCCESS); } // forbid Data in DL if it conflicts with PRACH for PRB==6 if (cell_params.cfg.cell.nof_prb == 6) { uint32_t tti_rx_ack = tti_params.tti_rx_ack_dl(); if (srslte_prach_tti_opportunity_config_fdd(cell_params.cfg.prach_config, tti_rx_ack, -1)) { dl_allocs.fill(0, dl_allocs.size()); } } // Decode Data allocations, check collisions and fill cumulative mask for (uint32_t i = 0; i < dl_result.nof_data_elems; ++i) { TESTASSERT(try_dl_mask_fill(dl_result.data[i].dci, "data") == SRSLTE_SUCCESS); } // TEST: check for holes in the PRB mask (RBGs not fully filled) rbgmask.resize(cell_params.nof_rbgs); rbgmask.reset(); srslte::bounded_bitset<100, true> rev_alloc = ~dl_allocs; for (uint32_t i = 0; i < cell_params.nof_rbgs; ++i) { uint32_t lim = SRSLTE_MIN((i + 1) * cell_params.P, dl_allocs.size()); bool val = dl_allocs.any(i * cell_params.P, lim); CONDERROR(rev_alloc.any(i * cell_params.P, lim) and val, "No holes can be left in an RBG\n"); if (val) { rbgmask.set(i); } } return SRSLTE_SUCCESS; } int output_sched_tester::test_sib_scheduling(const tti_params_t& tti_params, const sched_interface::dl_sched_res_t& dl_result) const { uint32_t sfn = tti_params.sfn_tx_dl; uint32_t sf_idx = tti_params.sf_idx_tx_dl; bool sib1_present = ((sfn % 2) == 0) and sf_idx == 5; using bc_elem = const sched_interface::dl_sched_bc_t; bc_elem* bc_begin = &dl_result.bc[0]; bc_elem* bc_end = &dl_result.bc[dl_result.nof_bc_elems]; /* Test if SIB1 was correctly scheduled */ if (sib1_present) { auto it = std::find_if(bc_begin, bc_end, [](bc_elem& elem) { return elem.index == 0; }); CONDERROR(it == bc_end, "Failed to allocate SIB1 in even sfn, sf_idx==5\n"); } /* Test if any SIB was scheduled with wrong index, tbs, or outside of its window */ for (bc_elem* bc = bc_begin; bc != bc_end; ++bc) { if (bc->index == 0) { continue; } CONDERROR(bc->index >= sched_interface::MAX_SIBS, "Invalid SIB idx=%d\n", bc->index + 1); CONDERROR(bc->tbs < cell_params.cfg.sibs[bc->index].len, "Allocated BC process with TBS=%d < sib_len=%d\n", bc->tbs, cell_params.cfg.sibs[bc->index].len); uint32_t x = (bc->index - 1) * cell_params.cfg.si_window_ms; uint32_t sf = x % 10; uint32_t sfn_start = sfn; while ((sfn_start % cell_params.cfg.sibs[bc->index].period_rf) != x / 10) { sfn_start--; } uint32_t win_start = sfn_start * 10 + sf; uint32_t win_end = win_start + cell_params.cfg.si_window_ms; CONDERROR(tti_params.tti_tx_dl < win_start or tti_params.tti_tx_dl > win_end, "Scheduled SIB is outside of its SIB window\n"); } return SRSLTE_SUCCESS; } int output_sched_tester::test_pdcch_collisions(const sched_interface::dl_sched_res_t& dl_result, const sched_interface::ul_sched_res_t& ul_result, srslte::bounded_bitset<128, true>* used_cce) const { used_cce->resize(srslte_regs_pdcch_ncce(cell_params.regs.get(), dl_result.cfi)); used_cce->reset(); // Helper Function: checks if there is any collision. If not, fills the PDCCH mask auto try_cce_fill = [&](const srslte_dci_location_t& dci_loc, const char* ch) { uint32_t cce_start = dci_loc.ncce, cce_stop = dci_loc.ncce + (1u << dci_loc.L); if (used_cce->any(cce_start, cce_stop)) { TESTERROR("%s DCI collision between CCE positions (%u, %u)\n", ch, cce_start, cce_stop); } used_cce->fill(cce_start, cce_stop); return SRSLTE_SUCCESS; }; /* TEST: verify there are no dci collisions for UL, DL data, BC, RAR */ for (uint32_t i = 0; i < ul_result.nof_dci_elems; ++i) { const auto& pusch = ul_result.pusch[i]; if (not pusch.needs_pdcch) { // In case of non-adaptive retx or Msg3 continue; } try_cce_fill(pusch.dci.location, "UL"); } for (uint32_t i = 0; i < dl_result.nof_data_elems; ++i) { try_cce_fill(dl_result.data[i].dci.location, "DL data"); } for (uint32_t i = 0; i < dl_result.nof_bc_elems; ++i) { try_cce_fill(dl_result.bc[i].dci.location, "DL BC"); } for (uint32_t i = 0; i < dl_result.nof_rar_elems; ++i) { try_cce_fill(dl_result.rar[i].dci.location, "DL RAR"); } return SRSLTE_SUCCESS; } int output_sched_tester::test_dci_values_consistency(const sched_interface::dl_sched_res_t& dl_result, const sched_interface::ul_sched_res_t& ul_result) const { for (uint32_t i = 0; i < ul_result.nof_dci_elems; ++i) { const auto& pusch = ul_result.pusch[i]; CONDERROR(pusch.tbs == 0, "Allocated PUSCH with invalid TBS=%d\n", pusch.tbs); // CONDERROR(ue_db.count(pusch.dci.rnti) == 0, "The allocated rnti=0x%x does not exist\n", pusch.dci.rnti); if (not pusch.needs_pdcch) { // In case of non-adaptive retx or Msg3 continue; } CONDERROR(pusch.dci.location.L == 0, "Invalid aggregation level %d\n", pusch.dci.location.L); // TODO: Extend this test } for (uint32_t i = 0; i < dl_result.nof_data_elems; ++i) { auto& data = dl_result.data[i]; CONDERROR(data.tbs[0] == 0, "Allocated DL data has empty TBS\n"); } for (uint32_t i = 0; i < dl_result.nof_bc_elems; ++i) { auto& bc = dl_result.bc[i]; if (bc.type == sched_interface::dl_sched_bc_t::BCCH) { CONDERROR(bc.tbs < cell_params.cfg.sibs[bc.index].len, "Allocated BC process with TBS=%d < sib_len=%d\n", bc.tbs, cell_params.cfg.sibs[bc.index].len); } else if (bc.type == sched_interface::dl_sched_bc_t::PCCH) { CONDERROR(bc.tbs == 0, "Allocated paging process with invalid TBS=%d\n", bc.tbs); } else { TESTERROR("Invalid broadcast process id=%d\n", (int)bc.type); } } for (uint32_t i = 0; i < dl_result.nof_rar_elems; ++i) { const auto& rar = dl_result.rar[i]; CONDERROR(rar.tbs == 0, "Allocated RAR process with invalid TBS=%d\n", rar.tbs); } return SRSLTE_SUCCESS; } int output_sched_tester::test_all(const tti_params_t& tti_params, const sched_interface::dl_sched_res_t& dl_result, const sched_interface::ul_sched_res_t& ul_result) const { prbmask_t ul_allocs; TESTASSERT(test_pusch_collisions(tti_params, ul_result, ul_allocs) == SRSLTE_SUCCESS); rbgmask_t dl_mask; TESTASSERT(test_pdsch_collisions(tti_params, dl_result, dl_mask) == SRSLTE_SUCCESS); TESTASSERT(test_sib_scheduling(tti_params, dl_result) == SRSLTE_SUCCESS); srslte::bounded_bitset<128, true> used_cce; TESTASSERT(test_pdcch_collisions(dl_result, ul_result, &used_cce) == SRSLTE_SUCCESS); return SRSLTE_SUCCESS; } /*********************** * User State Tester ***********************/ ue_ctxt_test::ue_ctxt_test(uint16_t rnti_, uint32_t preamble_idx_, srslte::tti_point prach_tti_, const ue_ctxt_test_cfg& cfg_, const std::vector& cell_params_) : sim_cfg(cfg_), rnti(rnti_), prach_tti(prach_tti_), preamble_idx(preamble_idx_), cell_params(cell_params_), current_tti_rx(prach_tti_) { set_cfg(cfg_.ue_cfg); } int ue_ctxt_test::set_cfg(const sched::ue_cfg_t& ue_cfg_) { for (uint32_t ue_cc_idx = 0; ue_cc_idx < ue_cfg_.supported_cc_list.size(); ++ue_cc_idx) { const auto& cc = ue_cfg_.supported_cc_list[ue_cc_idx]; if (ue_cc_idx >= active_ccs.size()) { // new cell active_ccs.emplace_back(); active_ccs.back().ue_cc_idx = active_ccs.size() - 1; active_ccs.back().enb_cc_idx = cc.enb_cc_idx; for (size_t i = 0; i < active_ccs.back().dl_harqs.size(); ++i) { active_ccs.back().dl_harqs[i].pid = i; active_ccs.back().ul_harqs[i].pid = i; } } else { CONDERROR(cc.enb_cc_idx != active_ccs[ue_cc_idx].enb_cc_idx, "changing ccs not supported\n"); } } user_cfg = ue_cfg_; return SRSLTE_SUCCESS; } ue_ctxt_test::cc_ue_ctxt_test* ue_ctxt_test::get_cc_state(uint32_t enb_cc_idx) { auto it = std::find_if(active_ccs.begin(), active_ccs.end(), [enb_cc_idx](const cc_ue_ctxt_test& c) { return c.enb_cc_idx == enb_cc_idx; }); return it == active_ccs.end() ? nullptr : &(*it); } int ue_ctxt_test::new_tti(sched* sched_ptr, srslte::tti_point tti_rx) { current_tti_rx = tti_rx; TESTASSERT(fwd_pending_acks(sched_ptr) == SRSLTE_SUCCESS); if (sim_cfg.periodic_cqi and (tti_rx.to_uint() % sim_cfg.cqi_Npd) == sim_cfg.cqi_Noffset) { for (auto& cc : active_ccs) { sched_ptr->dl_cqi_info( tti_rx.to_uint(), rnti, cc.enb_cc_idx, std::uniform_int_distribution{5, 24}(get_rand_gen())); sched_ptr->ul_cqi_info( tti_rx.to_uint(), rnti, cc.enb_cc_idx, std::uniform_int_distribution{5, 24}(get_rand_gen()), 0); } } return SRSLTE_SUCCESS; } int ue_ctxt_test::fwd_pending_acks(sched* sched_ptr) { /* Ack DL HARQs */ // Checks: // - Pending DL ACK {cc_idx,rnti,tb} exist in scheduler harqs // - Pending DL ACK tti_ack correspond to the expected based on tti_tx_dl while (not pending_dl_acks.empty()) { auto& p = pending_dl_acks.top(); if (p.tti_ack > current_tti_rx) { break; } auto& h = active_ccs[p.ue_cc_idx].dl_harqs[p.pid]; CONDERROR(not h.active, "The ACKed DL Harq pid=%d is not active\n", h.pid); CONDERROR(h.tti_tx + FDD_HARQ_DELAY_UL_MS != p.tti_ack, "dl ack hasn't arrived when expected\n"); CONDERROR(sched_ptr->dl_ack_info(current_tti_rx.to_uint(), rnti, p.cc_idx, p.tb, p.ack) <= 0, "The ACKed DL Harq pid=%d does not exist.\n", p.pid); if (p.ack) { h.active = false; log_h->info("DL ACK tti=%u rnti=0x%x pid=%d\n", current_tti_rx.to_uint(), rnti, p.pid); } pending_dl_acks.pop(); } /* Ack UL HARQs */ while (not pending_ul_acks.empty()) { auto& p = pending_ul_acks.top(); if (p.tti_ack > current_tti_rx) { break; } auto& h = active_ccs[p.ue_cc_idx].ul_harqs[p.pid]; CONDERROR(not h.active, "The ACKed UL Harq pid=%d is not active\n", h.pid); CONDERROR(h.tti_tx != p.tti_ack, "UL CRC wasn't set when expected\n"); CONDERROR(sched_ptr->ul_crc_info(current_tti_rx.to_uint(), rnti, p.cc_idx, p.ack) != SRSLTE_SUCCESS, "Failed UL ACK\n"); if (p.ack) { h.active = false; log_h->info("UL ACK tti=%u rnti=0x%x pid=%d\n", current_tti_rx.to_uint(), rnti, p.pid); } pending_ul_acks.pop(); } return SRSLTE_SUCCESS; } int ue_ctxt_test::test_sched_result(uint32_t enb_cc_idx, const sched::dl_sched_res_t& dl_result, const sched::ul_sched_res_t& ul_result) { cc_result result{enb_cc_idx, &dl_result, &ul_result}; TESTASSERT(test_harqs(result) == SRSLTE_SUCCESS); TESTASSERT(test_ra(result) == SRSLTE_SUCCESS); TESTASSERT(test_scell_activation(result) == SRSLTE_SUCCESS); TESTASSERT(schedule_acks(result) == SRSLTE_SUCCESS); return SRSLTE_SUCCESS; } /** * Tests whether the RAR and Msg3 were scheduled within the expected windows. Individual tests: * - No UL allocs before Msg3 * - No DL data allocs before Msg3 is correctly ACKed * - RAR alloc falls within RAR window and is unique per user * - Msg3 is allocated in expected TTI, without PDCCH, and correct rnti * - First Data allocation happens after Msg3, and contains a ConRes * - No RARs are allocated with wrong enb_cc_idx, preamble_idx or wrong user * TODO: * - check Msg3 PRBs match the ones advertised in the RAR */ int ue_ctxt_test::test_ra(cc_result result) { if (result.enb_cc_idx != active_ccs[0].enb_cc_idx) { // only check for RAR/Msg3 presence for a UE's PCell return SRSLTE_SUCCESS; } /* TEST: RAR allocation */ uint32_t rar_win_size = cell_params[result.enb_cc_idx].prach_rar_window; std::array rar_window = {prach_tti + 3, prach_tti + 3 + rar_win_size}; srslte::tti_point tti_tx_dl = srslte::to_tx_dl(current_tti_rx); srslte::tti_point tti_tx_ul = srslte::to_tx_ul(current_tti_rx); bool is_in_rar_window = tti_tx_dl >= rar_window[0] and tti_tx_dl <= rar_window[1]; if (not is_in_rar_window) { CONDERROR(not rar_tti.is_valid() and tti_tx_dl > rar_window[1], "rnti=0x%x RAR not scheduled within the RAR Window\n", rnti); for (uint32_t i = 0; i < result.dl_result->nof_rar_elems; ++i) { CONDERROR(result.dl_result->rar[i].dci.rnti == rnti, "No RAR allocations allowed outside of user RAR window\n"); } } else { // Inside RAR window for (uint32_t i = 0; i < result.dl_result->nof_rar_elems; ++i) { for (uint32_t j = 0; j < result.dl_result->rar[i].nof_grants; ++j) { const auto& data = result.dl_result->rar[i].msg3_grant[j].data; if (data.prach_tti == (uint32_t)prach_tti.to_uint() and data.preamble_idx == preamble_idx) { CONDERROR(rar_tti.is_valid(), "There was more than one RAR for the same user\n"); CONDERROR(rnti != data.temp_crnti, "RAR grant C-RNTI does not match the expected.\n"); msg3_riv = result.dl_result->rar[i].msg3_grant[j].grant.rba; rar_tti = tti_tx_dl; } } } } /* TEST: Check Msg3 */ if (rar_tti.is_valid() and not msg3_tti.is_valid()) { // RAR scheduled, Msg3 not yet scheduled srslte::tti_point expected_msg3_tti = rar_tti + FDD_HARQ_DELAY_DL_MS + MSG3_DELAY_MS; CONDERROR(expected_msg3_tti < tti_tx_ul and not msg3_tti.is_valid(), "No UL msg3 alloc was made\n"); if (expected_msg3_tti == tti_tx_ul) { // Msg3 should exist for (uint32_t i = 0; i < result.ul_result->nof_dci_elems; ++i) { if (result.ul_result->pusch[i].dci.rnti == rnti) { CONDERROR(msg3_tti.is_valid(), "Only one Msg3 allowed per user\n"); CONDERROR(result.ul_result->pusch[i].needs_pdcch, "Msg3 allocations do not require PDCCH\n"); CONDERROR(msg3_riv != result.ul_result->pusch[i].dci.type2_alloc.riv, "The Msg3 was not allocated in the expected PRBs.\n"); msg3_tti = tti_tx_ul; } } } } /* TEST: Check Msg4 */ if (msg3_tti.is_valid() and not msg4_tti.is_valid()) { // Msg3 scheduled, but Msg4 not yet scheduled for (uint32_t i = 0; i < result.dl_result->nof_data_elems; ++i) { if (result.dl_result->data[i].dci.rnti == rnti) { CONDERROR(current_tti_rx < msg3_tti, "Msg4 cannot be scheduled without Msg3 being tx\n"); for (uint32_t j = 0; j < result.dl_result->data[i].nof_pdu_elems[0]; ++j) { if (result.dl_result->data[i].pdu[0][j].lcid == (uint32_t)srslte::dl_sch_lcid::CON_RES_ID) { // ConRes found CONDERROR(result.dl_result->data[i].dci.format != SRSLTE_DCI_FORMAT1, "ConRes must be format1\n"); CONDERROR(msg4_tti.is_valid(), "Duplicate ConRes CE for the same rnti\n"); msg4_tti = tti_tx_dl; } } } } } /* TEST: Txs out of place */ if (not msg4_tti.is_valid()) { // Msg4 not yet received by user for (uint32_t i = 0; i < result.dl_result->nof_data_elems; ++i) { CONDERROR(result.dl_result->data[i].dci.rnti == rnti, "No DL data allocs allowed before Msg4 is scheduled\n"); } if (msg3_tti.is_valid() and msg3_tti != tti_tx_ul) { // Msg3 scheduled. No UL alloc allowed unless it is a newtx (the Msg3 itself) for (uint32_t i = 0; i < result.ul_result->nof_dci_elems; ++i) { // Needs PDCCH - filters out UL retxs bool msg3_retx = ((tti_tx_ul - msg3_tti) % (FDD_HARQ_DELAY_UL_MS + FDD_HARQ_DELAY_DL_MS)) == 0; CONDERROR(result.ul_result->pusch[i].dci.rnti == rnti and not msg3_retx, "No UL txs allowed except for Msg3 before user received Msg4\n"); } } else if (not msg3_tti.is_valid()) { // No Msg3 sched TTI for (uint32_t i = 0; i < result.ul_result->nof_dci_elems; ++i) { CONDERROR(result.ul_result->pusch[i].dci.rnti == rnti, "No UL newtxs allowed before user received Msg4\n"); } } } return SRSLTE_SUCCESS; } /** * Tests whether the SCells are correctly activated. Individual tests: * - no DL and UL allocations in inactive carriers */ int ue_ctxt_test::test_scell_activation(cc_result result) { auto cc_it = std::find_if(user_cfg.supported_cc_list.begin(), user_cfg.supported_cc_list.end(), [&result](const sched::ue_cfg_t::cc_cfg_t& cc) { return cc.enb_cc_idx == result.enb_cc_idx; }); if (cc_it == user_cfg.supported_cc_list.end() or not cc_it->active) { // cell not active. Ensure data allocations are not made for (uint32_t i = 0; i < result.dl_result->nof_data_elems; ++i) { CONDERROR(result.dl_result->data[i].dci.rnti == rnti, "Allocated user in inactive carrier\n"); } for (uint32_t i = 0; i < result.ul_result->nof_dci_elems; ++i) { CONDERROR(result.ul_result->pusch[i].dci.rnti == rnti, "Allocated user in inactive carrier\n"); } } else { uint32_t ue_cc_idx = std::distance(user_cfg.supported_cc_list.begin(), cc_it); for (uint32_t i = 0; i < result.dl_result->nof_data_elems; ++i) { if (result.dl_result->data[i].dci.rnti == rnti) { CONDERROR(result.dl_result->data[i].dci.ue_cc_idx != ue_cc_idx, "User cell index was incorrectly set\n"); } } for (uint32_t i = 0; i < result.ul_result->nof_dci_elems; ++i) { if (result.ul_result->pusch[i].dci.rnti == rnti) { CONDERROR(result.ul_result->pusch[i].dci.ue_cc_idx != ue_cc_idx, "The user cell index was incorrectly set\n"); } } } return SRSLTE_SUCCESS; } /** * Sanity checks of the DCI values in the scheduling result for a given user. Current checks: * - invalid ue_cc_idx<->enb_cc_idx matching in dl_result * - reusing same pid too early (ACK still didn't arrive yet) * - invalid rv value (nof retxs is incorrect) and ndi value */ int ue_ctxt_test::test_harqs(cc_result result) { cc_ue_ctxt_test* cc = get_cc_state(result.enb_cc_idx); if (cc == nullptr) { // unsupported carrier return SRSLTE_SUCCESS; } /* Test DL Harqs */ for (uint32_t i = 0; i < result.dl_result->nof_data_elems; ++i) { const auto& data = result.dl_result->data[i]; if (data.dci.rnti != rnti) { continue; } CONDERROR(data.dci.ue_cc_idx != cc->ue_cc_idx, "invalid ue_cc_idx=%d in sched result\n", data.dci.ue_cc_idx); CONDERROR(data.dci.pid >= cc->dl_harqs.size(), "invalid pid\n"); auto& h = cc->dl_harqs[data.dci.pid]; uint32_t nof_retx = sched_utils::get_nof_retx(data.dci.tb[0].rv); // 0..3 if (h.nof_txs == 0 or h.ndi != data.dci.tb[0].ndi) { // It is newtx CONDERROR(nof_retx != 0, "Invalid rv index for new tx\n"); h.active = true; h.nof_retxs = 0; h.ndi = data.dci.tb[0].ndi; h.tti_tx = srslte::to_tx_dl(current_tti_rx); } else { // it is retx CONDERROR(sched_utils::get_rvidx(h.nof_retxs + 1) != (uint32_t)data.dci.tb[0].rv, "Invalid rv index for retx\n"); CONDERROR(h.ndi != data.dci.tb[0].ndi, "Invalid ndi for retx\n"); CONDERROR(not h.active, "retx for inactive dl harq pid=%d\n", h.pid); CONDERROR(h.tti_tx > current_tti_rx, "harq pid=%d reused too soon\n", h.pid); CONDERROR(h.nof_retxs + 1 > sim_cfg.ue_cfg.maxharq_tx, "The number of retx=%d exceeded its max=%d\n", h.nof_retxs + 1, sim_cfg.ue_cfg.maxharq_tx); h.nof_retxs++; h.tti_tx = srslte::to_tx_dl(current_tti_rx); } h.nof_txs++; } /* Test UL Harqs */ for (uint32_t i = 0; i < result.ul_result->nof_dci_elems; ++i) { const auto& pusch = result.ul_result->pusch[i]; if (pusch.dci.rnti != rnti) { continue; } CONDERROR(pusch.dci.ue_cc_idx != cc->ue_cc_idx, "invalid ue_cc_idx=%d in sched result\n", pusch.dci.ue_cc_idx); auto& h = cc->ul_harqs[srslte::to_tx_ul(current_tti_rx).to_uint() % cc->ul_harqs.size()]; uint32_t nof_retx = sched_utils::get_nof_retx(pusch.dci.tb.rv); // 0..3 if (h.nof_txs == 0 or h.ndi != pusch.dci.tb.ndi) { // newtx CONDERROR(nof_retx != 0, "Invalid rv index for new tx\n"); h.active = true; h.nof_retxs = 0; h.ndi = pusch.dci.tb.ndi; } else { if (pusch.needs_pdcch) { // adaptive retx } else { // non-adaptive retx CONDERROR(pusch.dci.type2_alloc.riv != h.riv, "Non-adaptive retx must keep the same riv\n"); } if (pusch.tbs > 0) { CONDERROR(sched_utils::get_rvidx(h.nof_retxs + 1) != (uint32_t)pusch.dci.tb.rv, "Invalid rv index for retx\n"); } CONDERROR(h.ndi != pusch.dci.tb.ndi, "Invalid ndi for retx\n"); CONDERROR(not h.active, "retx for inactive UL harq pid=%d\n", h.pid); CONDERROR(h.tti_tx > current_tti_rx, "UL harq pid=%d was reused too soon\n", h.pid); h.nof_retxs++; } h.tti_tx = srslte::to_tx_ul(current_tti_rx); h.riv = pusch.dci.type2_alloc.riv; h.nof_txs++; } return SRSLTE_SUCCESS; } int ue_ctxt_test::schedule_acks(cc_result result) { auto* cc = get_cc_state(result.enb_cc_idx); if (cc == nullptr) { return SRSLTE_SUCCESS; } /* Schedule DL ACKs */ for (uint32_t i = 0; i < result.dl_result->nof_data_elems; ++i) { const auto& data = result.dl_result->data[i]; if (data.dci.rnti != rnti) { continue; } pending_ack_t ack_data; ack_data.tti_ack = srslte::to_tx_dl_ack(current_tti_rx); ack_data.cc_idx = result.enb_cc_idx; ack_data.tb = 0; ack_data.pid = data.dci.pid; ack_data.ue_cc_idx = data.dci.ue_cc_idx; uint32_t nof_retx = sched_utils::get_nof_retx(data.dci.tb[0].rv); // 0..3 ack_data.ack = randf() < sim_cfg.prob_dl_ack_mask[nof_retx % sim_cfg.prob_dl_ack_mask.size()]; pending_dl_acks.push(ack_data); } /* Schedule UL ACKs */ for (uint32_t i = 0; i < result.ul_result->nof_dci_elems; ++i) { const auto& pusch = result.ul_result->pusch[i]; if (pusch.dci.rnti != rnti) { continue; } pending_ack_t ack_data; ack_data.tti_ack = srslte::to_tx_ul(current_tti_rx); ack_data.cc_idx = result.enb_cc_idx; ack_data.ue_cc_idx = pusch.dci.ue_cc_idx; ack_data.tb = 0; ack_data.pid = srslte::to_tx_ul(current_tti_rx).to_uint() % cc->ul_harqs.size(); uint32_t nof_retx = sched_utils::get_nof_retx(pusch.dci.tb.rv); // 0..3 ack_data.ack = randf() < sim_cfg.prob_ul_ack_mask[nof_retx % sim_cfg.prob_ul_ack_mask.size()]; pending_ul_acks.push(ack_data); } return SRSLTE_SUCCESS; } int srsenb::extract_dl_prbmask(const srslte_cell_t& cell, const srslte_dci_dl_t& dci, srslte::bounded_bitset<100, true>* alloc_mask) { srslte_pdsch_grant_t grant; srslte_dl_sf_cfg_t dl_sf = {}; alloc_mask->resize(cell.nof_prb); alloc_mask->reset(); CONDERROR(srslte_ra_dl_dci_to_grant(&cell, &dl_sf, SRSLTE_TM1, false, &dci, &grant) == SRSLTE_ERROR, "Failed to decode PDSCH grant\n"); for (uint32_t j = 0; j < alloc_mask->size(); ++j) { if (grant.prb_idx[0][j]) { alloc_mask->set(j); } } return SRSLTE_SUCCESS; } void user_state_sched_tester::new_tti(sched* sched_ptr, uint32_t tti_rx) { tic++; for (auto& u : users) { u.second.new_tti(sched_ptr, srslte::tti_point{tti_rx}); } } int user_state_sched_tester::add_user(uint16_t rnti, uint32_t preamble_idx, const ue_ctxt_test_cfg& cfg_) { CONDERROR(!srslte_prach_tti_opportunity_config_fdd( cell_params[cfg_.ue_cfg.supported_cc_list[0].enb_cc_idx].prach_config, tic.to_uint(), -1), "New user added in a non-PRACH TTI\n"); TESTASSERT(users.count(rnti) == 0); ue_ctxt_test ue{rnti, preamble_idx, srslte::tti_point{tic.to_uint()}, cfg_, cell_params}; users.insert(std::make_pair(rnti, ue)); return SRSLTE_SUCCESS; } int user_state_sched_tester::user_reconf(uint16_t rnti, const srsenb::sched_interface::ue_cfg_t& ue_cfg) { TESTASSERT(users.count(rnti) > 0); users.at(rnti).set_cfg(ue_cfg); return SRSLTE_SUCCESS; } int user_state_sched_tester::bearer_cfg(uint16_t rnti, uint32_t lcid, const srsenb::sched_interface::ue_bearer_cfg_t& bearer_cfg) { auto it = users.find(rnti); TESTASSERT(it != users.end()); it->second.user_cfg.ue_bearers[lcid] = bearer_cfg; users.at(rnti).drb_cfg_flag = false; for (uint32_t i = 2; i < it->second.user_cfg.ue_bearers.size(); ++i) { if (it->second.user_cfg.ue_bearers[i].direction != sched_interface::ue_bearer_cfg_t::IDLE) { users.at(rnti).drb_cfg_flag = true; } } return SRSLTE_SUCCESS; } void user_state_sched_tester::rem_user(uint16_t rnti) { users.erase(rnti); } /** * Individual tests: * - All RARs belong to a user that just PRACHed * - All DL/UL data allocs have a valid RNTI */ int user_state_sched_tester::test_ctrl_info(uint32_t enb_cc_idx, const sched_interface::dl_sched_res_t& dl_result, const sched_interface::ul_sched_res_t& ul_result) { /* TEST: Ensure there are no spurious RARs that do not belong to any user */ for (uint32_t i = 0; i < dl_result.nof_rar_elems; ++i) { for (uint32_t j = 0; j < dl_result.rar[i].nof_grants; ++j) { uint32_t prach_tti = dl_result.rar[i].msg3_grant[j].data.prach_tti; uint32_t preamble_idx = dl_result.rar[i].msg3_grant[j].data.preamble_idx; auto it = std::find_if(users.begin(), users.end(), [&](const std::pair& u) { return u.second.preamble_idx == preamble_idx and ((uint32_t)u.second.prach_tti.to_uint() == prach_tti); }); CONDERROR(it == users.end(), "There was a RAR allocation with no associated user"); CONDERROR(it->second.user_cfg.supported_cc_list[0].enb_cc_idx != enb_cc_idx, "The allocated RAR is in the wrong cc\n"); } } /* TEST: All DL allocs have a correct rnti */ std::set alloc_rntis; for (uint32_t i = 0; i < dl_result.nof_data_elems; ++i) { uint16_t rnti = dl_result.data[i].dci.rnti; CONDERROR(alloc_rntis.count(rnti) > 0, "The user rnti=0x%x got allocated multiple times in DL\n", rnti); CONDERROR(users.count(rnti) == 0, "The user rnti=0x%x allocated in DL does not exist\n", rnti); alloc_rntis.insert(rnti); } /* TEST: All UL allocs have a correct rnti */ alloc_rntis.clear(); for (uint32_t i = 0; i < ul_result.nof_dci_elems; ++i) { uint16_t rnti = ul_result.pusch[i].dci.rnti; CONDERROR(alloc_rntis.count(rnti) > 0, "The user rnti=0x%x got allocated multiple times in UL\n", rnti); CONDERROR(users.count(rnti) == 0, "The user rnti=0x%x allocated in UL does not exist\n", rnti); alloc_rntis.insert(rnti); } return SRSLTE_SUCCESS; } int user_state_sched_tester::test_all(uint32_t enb_cc_idx, const sched_interface::dl_sched_res_t& dl_result, const sched_interface::ul_sched_res_t& ul_result) { TESTASSERT(test_ctrl_info(enb_cc_idx, dl_result, ul_result) == SRSLTE_SUCCESS); for (auto& u : users) { TESTASSERT(u.second.test_sched_result(enb_cc_idx, dl_result, ul_result) == SRSLTE_SUCCESS); } return SRSLTE_SUCCESS; } /*********************** * Sim Stats Storage **********************/ void sched_result_stats::process_results(const tti_params_t& tti_params, const std::vector& dl_result, const std::vector& ul_result) { for (uint32_t ccidx = 0; ccidx < dl_result.size(); ++ccidx) { for (uint32_t i = 0; i < dl_result[ccidx].nof_data_elems; ++i) { user_stats* user = get_user(dl_result[ccidx].data[i].dci.rnti); user->tot_dl_sched_data[ccidx] += dl_result[ccidx].data[i].tbs[0]; user->tot_dl_sched_data[ccidx] += dl_result[ccidx].data[i].tbs[1]; } for (uint32_t i = 0; i < ul_result[ccidx].nof_dci_elems; ++i) { user_stats* user = get_user(ul_result[ccidx].pusch[i].dci.rnti); user->tot_ul_sched_data[ccidx] += ul_result[ccidx].pusch[i].tbs; } } } sched_result_stats::user_stats* sched_result_stats::get_user(uint16_t rnti) { if (users.count(rnti) != 0) { return &users[rnti]; } users[rnti].rnti = rnti; users[rnti].tot_dl_sched_data.resize(cell_params.size(), 0); users[rnti].tot_ul_sched_data.resize(cell_params.size(), 0); return &users[rnti]; } /*********************** * Common Sched Tester **********************/ const sched::ue_cfg_t* common_sched_tester::get_current_ue_cfg(uint16_t rnti) const { return ue_tester->get_user_cfg(rnti); } int common_sched_tester::sim_cfg(sim_sched_args args) { sim_args0 = std::move(args); sched::cell_cfg(sim_args0.cell_cfg); // call parent cfg sched::set_sched_cfg(&sim_args0.sched_args); ue_tester.reset(new user_state_sched_tester{sim_args0.cell_cfg}); output_tester.clear(); output_tester.reserve(sim_args0.cell_cfg.size()); for (uint32_t i = 0; i < sim_args0.cell_cfg.size(); ++i) { output_tester.emplace_back(sched_cell_params[i]); } sched_stats.reset(new sched_result_stats{sim_args0.cell_cfg}); tester_log = sim_args0.sim_log; return SRSLTE_SUCCESS; } int common_sched_tester::add_user(uint16_t rnti, const ue_ctxt_test_cfg& ue_cfg_) { CONDERROR(ue_cfg(rnti, generate_rach_ue_cfg(ue_cfg_.ue_cfg)) != SRSLTE_SUCCESS, "Configuring new user rnti=0x%x to sched\n", rnti); // CONDERROR(!srslte_prach_tti_opportunity_config_fdd( // sched_cell_params[CARRIER_IDX].cfg.prach_config, tti_info.tti_params.tti_rx, -1), // "New user added in a non-PRACH TTI\n"); dl_sched_rar_info_t rar_info = {}; rar_info.prach_tti = tti_info.tti_params.tti_rx; rar_info.temp_crnti = rnti; rar_info.msg3_size = 7; rar_info.preamble_idx = tti_info.nof_prachs++; uint32_t pcell_idx = ue_cfg_.ue_cfg.supported_cc_list[0].enb_cc_idx; dl_rach_info(pcell_idx, rar_info); ue_tester->add_user(rnti, rar_info.preamble_idx, ue_cfg_); tester_log->info("Adding user rnti=0x%x\n", rnti); return SRSLTE_SUCCESS; } int common_sched_tester::reconf_user(uint16_t rnti, const sched_interface::ue_cfg_t& ue_cfg_) { CONDERROR(not ue_tester->user_exists(rnti), "User must already exist to be configured\n"); CONDERROR(ue_cfg(rnti, ue_cfg_) != SRSLTE_SUCCESS, "Configuring new user rnti=0x%x to sched\n", rnti); ue_tester->user_reconf(rnti, ue_cfg_); return SRSLTE_SUCCESS; } void common_sched_tester::rem_user(uint16_t rnti) { tester_log->info("Removing user rnti=0x%x\n", rnti); sched::ue_rem(rnti); ue_tester->rem_user(rnti); } void common_sched_tester::new_test_tti() { if (not tic.is_valid()) { tic = srslte::tti_point{sim_args0.start_tti}; } else { tic++; } tti_info.tti_params = tti_params_t{tic.to_uint()}; tti_info.nof_prachs = 0; tti_info.dl_sched_result.clear(); tti_info.ul_sched_result.clear(); tti_info.dl_sched_result.resize(sched_cell_params.size()); tti_info.ul_sched_result.resize(sched_cell_params.size()); tester_log->step(tti_info.tti_params.tti_rx); } int common_sched_tester::process_results() { for (uint32_t i = 0; i < sched_cell_params.size(); ++i) { TESTASSERT(ue_tester->test_all(i, tti_info.dl_sched_result[i], tti_info.ul_sched_result[i]) == SRSLTE_SUCCESS); TESTASSERT(output_tester[i].test_all( tti_info.tti_params, tti_info.dl_sched_result[i], tti_info.ul_sched_result[i]) == SRSLTE_SUCCESS); } sched_stats->process_results(tti_info.tti_params, tti_info.dl_sched_result, tti_info.ul_sched_result); return SRSLTE_SUCCESS; } int common_sched_tester::process_tti_events(const tti_ev& tti_ev) { for (const tti_ev::user_cfg_ev& ue_ev : tti_ev.user_updates) { // There is a new configuration if (ue_ev.ue_sim_cfg != nullptr) { if (not ue_tester->user_exists(ue_ev.rnti)) { // new user TESTASSERT(add_user(ue_ev.rnti, *ue_ev.ue_sim_cfg) == SRSLTE_SUCCESS); } else { // reconfiguration TESTASSERT(reconf_user(ue_ev.rnti, ue_ev.ue_sim_cfg->ue_cfg) == SRSLTE_SUCCESS); } } // There is a user to remove if (ue_ev.rem_user) { rem_user(ue_ev.rnti); } // configure bearers if (ue_ev.bearer_cfg != nullptr) { CONDERROR(not ue_tester->user_exists(ue_ev.rnti), "User rnti=0x%x does not exist\n", ue_ev.rnti); // TODO: Instantiate more bearers bearer_ue_cfg(ue_ev.rnti, 0, ue_ev.bearer_cfg.get()); } const ue_ctxt_test* user = ue_tester->get_user_ctxt(ue_ev.rnti); if (user != nullptr and not user->msg4_tti.is_valid() and user->msg3_tti.is_valid() and user->msg3_tti <= tic) { // Msg3 has been received but Msg4 has not been yet transmitted // Setup default UE config reconf_user(user->rnti, generate_setup_ue_cfg(sim_args0.default_ue_sim_cfg.ue_cfg)); // Schedule RRC Setup and ConRes CE uint32_t pending_dl_new_data = ue_db[ue_ev.rnti].get_pending_dl_new_data(); if (pending_dl_new_data == 0) { uint32_t lcid = RB_ID_SRB0; // Use SRB0 to schedule Msg4 dl_rlc_buffer_state(ue_ev.rnti, lcid, 50, 0); dl_mac_buffer_state(ue_ev.rnti, (uint32_t)srslte::dl_sch_lcid::CON_RES_ID); } else { // Let SRB0 Msg4 get fully transmitted } } // push UL SRs and DL packets if (ue_ev.buffer_ev != nullptr) { CONDERROR(user == nullptr, "TESTER ERROR: Trying to schedule data for user that does not exist\n"); if (ue_ev.buffer_ev->dl_data > 0 and user->msg4_tti.is_valid()) { // If Msg4 has already been tx and there DL data to transmit uint32_t lcid = RB_ID_DRB1; uint32_t pending_dl_new_data = ue_db[ue_ev.rnti].get_pending_dl_new_data(); if (user->drb_cfg_flag or pending_dl_new_data == 0) { // If RRCSetup finished if (not user->drb_cfg_flag) { reconf_user(user->rnti, sim_args0.default_ue_sim_cfg.ue_cfg); // setup lcid==drb1 bearer sched::ue_bearer_cfg_t cfg = {}; cfg.direction = ue_bearer_cfg_t::BOTH; cfg.group = 1; ue_tester->bearer_cfg(ue_ev.rnti, lcid, cfg); bearer_ue_cfg(ue_ev.rnti, lcid, &cfg); } // DRB is set. Update DL buffer uint32_t tot_dl_data = pending_dl_new_data + ue_ev.buffer_ev->dl_data; // TODO: derive pending based on rx dl_rlc_buffer_state(ue_ev.rnti, lcid, tot_dl_data, 0); // TODO: Check retx_queue } else { // Let SRB0 get emptied } } if (ue_ev.buffer_ev->sr_data > 0 and user->drb_cfg_flag) { uint32_t tot_ul_data = ue_db[ue_ev.rnti].get_pending_ul_new_data(tti_info.tti_params.tti_tx_ul) + ue_ev.buffer_ev->sr_data; uint32_t lcg = 1; ul_bsr(ue_ev.rnti, lcg, tot_ul_data); } } } return SRSLTE_SUCCESS; } int common_sched_tester::run_tti(const tti_ev& tti_events) { new_test_tti(); tester_log->info("---- tti=%u | nof_ues=%zd ----\n", tic.to_uint(), ue_db.size()); ue_tester->new_tti(this, tti_info.tti_params.tti_rx); process_tti_events(tti_events); before_sched(); // Call scheduler for all carriers tti_info.dl_sched_result.resize(sched_cell_params.size()); for (uint32_t i = 0; i < sched_cell_params.size(); ++i) { dl_sched(tti_info.tti_params.tti_tx_dl, i, tti_info.dl_sched_result[i]); } tti_info.ul_sched_result.resize(sched_cell_params.size()); for (uint32_t i = 0; i < sched_cell_params.size(); ++i) { ul_sched(tti_info.tti_params.tti_tx_ul, i, tti_info.ul_sched_result[i]); } process_results(); tti_count++; return SRSLTE_SUCCESS; } int common_sched_tester::test_next_ttis(const std::vector& tti_events) { while (tti_count < tti_events.size()) { TESTASSERT(run_tti(tti_events[tti_count]) == SRSLTE_SUCCESS); } return SRSLTE_SUCCESS; }