1// SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause
2/*
3 * Copyright (C) 2017 Intel Deutschland GmbH
4 * Copyright (C) 2018-2023 Intel Corporation
5 */
6#include "iwl-trans.h"
7#include "iwl-fh.h"
8#include "iwl-context-info.h"
9#include "internal.h"
10#include "iwl-prph.h"
11
12static void *_iwl_pcie_ctxt_info_dma_alloc_coherent(struct iwl_trans *trans,
13						    size_t size,
14						    dma_addr_t *phys,
15						    int depth)
16{
17	void *result;
18
19	if (WARN(depth > 2,
20		 "failed to allocate DMA memory not crossing 2^32 boundary"))
21		return NULL;
22
23	result = dma_alloc_coherent(trans->dev, size, phys, GFP_KERNEL);
24
25	if (!result)
26		return NULL;
27
28	if (unlikely(iwl_txq_crosses_4g_boundary(*phys, size))) {
29		void *old = result;
30		dma_addr_t oldphys = *phys;
31
32		result = _iwl_pcie_ctxt_info_dma_alloc_coherent(trans, size,
33								phys,
34								depth + 1);
35		dma_free_coherent(trans->dev, size, old, oldphys);
36	}
37
38	return result;
39}
40
41void *iwl_pcie_ctxt_info_dma_alloc_coherent(struct iwl_trans *trans,
42					    size_t size,
43					    dma_addr_t *phys)
44{
45	return _iwl_pcie_ctxt_info_dma_alloc_coherent(trans, size, phys, 0);
46}
47
48int iwl_pcie_ctxt_info_alloc_dma(struct iwl_trans *trans,
49				 const void *data, u32 len,
50				 struct iwl_dram_data *dram)
51{
52	dram->block = iwl_pcie_ctxt_info_dma_alloc_coherent(trans, len,
53							    &dram->physical);
54	if (!dram->block)
55		return -ENOMEM;
56
57	dram->size = len;
58	memcpy(dram->block, data, len);
59
60	return 0;
61}
62
63void iwl_pcie_ctxt_info_free_paging(struct iwl_trans *trans)
64{
65	struct iwl_self_init_dram *dram = &trans->init_dram;
66	int i;
67
68	if (!dram->paging) {
69		WARN_ON(dram->paging_cnt);
70		return;
71	}
72
73	/* free paging*/
74	for (i = 0; i < dram->paging_cnt; i++)
75		dma_free_coherent(trans->dev, dram->paging[i].size,
76				  dram->paging[i].block,
77				  dram->paging[i].physical);
78
79	kfree(dram->paging);
80	dram->paging_cnt = 0;
81	dram->paging = NULL;
82}
83
84int iwl_pcie_init_fw_sec(struct iwl_trans *trans,
85			 const struct fw_img *fw,
86			 struct iwl_context_info_dram *ctxt_dram)
87{
88	struct iwl_self_init_dram *dram = &trans->init_dram;
89	int i, ret, lmac_cnt, umac_cnt, paging_cnt;
90
91	if (WARN(dram->paging,
92		 "paging shouldn't already be initialized (%d pages)\n",
93		 dram->paging_cnt))
94		iwl_pcie_ctxt_info_free_paging(trans);
95
96	lmac_cnt = iwl_pcie_get_num_sections(fw, 0);
97	/* add 1 due to separator */
98	umac_cnt = iwl_pcie_get_num_sections(fw, lmac_cnt + 1);
99	/* add 2 due to separators */
100	paging_cnt = iwl_pcie_get_num_sections(fw, lmac_cnt + umac_cnt + 2);
101
102	dram->fw = kcalloc(umac_cnt + lmac_cnt, sizeof(*dram->fw), GFP_KERNEL);
103	if (!dram->fw)
104		return -ENOMEM;
105	dram->paging = kcalloc(paging_cnt, sizeof(*dram->paging), GFP_KERNEL);
106	if (!dram->paging)
107		return -ENOMEM;
108
109	/* initialize lmac sections */
110	for (i = 0; i < lmac_cnt; i++) {
111		ret = iwl_pcie_ctxt_info_alloc_dma(trans, fw->sec[i].data,
112						   fw->sec[i].len,
113						   &dram->fw[dram->fw_cnt]);
114		if (ret)
115			return ret;
116		ctxt_dram->lmac_img[i] =
117			cpu_to_le64(dram->fw[dram->fw_cnt].physical);
118		dram->fw_cnt++;
119	}
120
121	/* initialize umac sections */
122	for (i = 0; i < umac_cnt; i++) {
123		/* access FW with +1 to make up for lmac separator */
124		ret = iwl_pcie_ctxt_info_alloc_dma(trans,
125						   fw->sec[dram->fw_cnt + 1].data,
126						   fw->sec[dram->fw_cnt + 1].len,
127						   &dram->fw[dram->fw_cnt]);
128		if (ret)
129			return ret;
130		ctxt_dram->umac_img[i] =
131			cpu_to_le64(dram->fw[dram->fw_cnt].physical);
132		dram->fw_cnt++;
133	}
134
135	/*
136	 * Initialize paging.
137	 * Paging memory isn't stored in dram->fw as the umac and lmac - it is
138	 * stored separately.
139	 * This is since the timing of its release is different -
140	 * while fw memory can be released on alive, the paging memory can be
141	 * freed only when the device goes down.
142	 * Given that, the logic here in accessing the fw image is a bit
143	 * different - fw_cnt isn't changing so loop counter is added to it.
144	 */
145	for (i = 0; i < paging_cnt; i++) {
146		/* access FW with +2 to make up for lmac & umac separators */
147		int fw_idx = dram->fw_cnt + i + 2;
148
149		ret = iwl_pcie_ctxt_info_alloc_dma(trans, fw->sec[fw_idx].data,
150						   fw->sec[fw_idx].len,
151						   &dram->paging[i]);
152		if (ret)
153			return ret;
154
155		ctxt_dram->virtual_img[i] =
156			cpu_to_le64(dram->paging[i].physical);
157		dram->paging_cnt++;
158	}
159
160	return 0;
161}
162
163int iwl_pcie_ctxt_info_init(struct iwl_trans *trans,
164			    const struct fw_img *fw)
165{
166	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
167	struct iwl_context_info *ctxt_info;
168	struct iwl_context_info_rbd_cfg *rx_cfg;
169	u32 control_flags = 0, rb_size;
170	dma_addr_t phys;
171	int ret;
172
173	ctxt_info = iwl_pcie_ctxt_info_dma_alloc_coherent(trans,
174							  sizeof(*ctxt_info),
175							  &phys);
176	if (!ctxt_info)
177		return -ENOMEM;
178
179	trans_pcie->ctxt_info_dma_addr = phys;
180
181	ctxt_info->version.version = 0;
182	ctxt_info->version.mac_id =
183		cpu_to_le16((u16)trans->hw_rev);
184	/* size is in DWs */
185	ctxt_info->version.size = cpu_to_le16(sizeof(*ctxt_info) / 4);
186
187	switch (trans_pcie->rx_buf_size) {
188	case IWL_AMSDU_2K:
189		rb_size = IWL_CTXT_INFO_RB_SIZE_2K;
190		break;
191	case IWL_AMSDU_4K:
192		rb_size = IWL_CTXT_INFO_RB_SIZE_4K;
193		break;
194	case IWL_AMSDU_8K:
195		rb_size = IWL_CTXT_INFO_RB_SIZE_8K;
196		break;
197	case IWL_AMSDU_12K:
198		rb_size = IWL_CTXT_INFO_RB_SIZE_16K;
199		break;
200	default:
201		WARN_ON(1);
202		rb_size = IWL_CTXT_INFO_RB_SIZE_4K;
203	}
204
205	WARN_ON(RX_QUEUE_CB_SIZE(trans->cfg->num_rbds) > 12);
206	control_flags = IWL_CTXT_INFO_TFD_FORMAT_LONG;
207	control_flags |=
208		u32_encode_bits(RX_QUEUE_CB_SIZE(trans->cfg->num_rbds),
209				IWL_CTXT_INFO_RB_CB_SIZE);
210	control_flags |= u32_encode_bits(rb_size, IWL_CTXT_INFO_RB_SIZE);
211	ctxt_info->control.control_flags = cpu_to_le32(control_flags);
212
213	/* initialize RX default queue */
214	rx_cfg = &ctxt_info->rbd_cfg;
215	rx_cfg->free_rbd_addr = cpu_to_le64(trans_pcie->rxq->bd_dma);
216	rx_cfg->used_rbd_addr = cpu_to_le64(trans_pcie->rxq->used_bd_dma);
217	rx_cfg->status_wr_ptr = cpu_to_le64(trans_pcie->rxq->rb_stts_dma);
218
219	/* initialize TX command queue */
220	ctxt_info->hcmd_cfg.cmd_queue_addr =
221		cpu_to_le64(trans->txqs.txq[trans->txqs.cmd.q_id]->dma_addr);
222	ctxt_info->hcmd_cfg.cmd_queue_size =
223		TFD_QUEUE_CB_SIZE(IWL_CMD_QUEUE_SIZE);
224
225	/* allocate ucode sections in dram and set addresses */
226	ret = iwl_pcie_init_fw_sec(trans, fw, &ctxt_info->dram);
227	if (ret) {
228		dma_free_coherent(trans->dev, sizeof(*trans_pcie->ctxt_info),
229				  ctxt_info, trans_pcie->ctxt_info_dma_addr);
230		return ret;
231	}
232
233	trans_pcie->ctxt_info = ctxt_info;
234
235	iwl_enable_fw_load_int_ctx_info(trans);
236
237	/* Configure debug, if exists */
238	if (iwl_pcie_dbg_on(trans))
239		iwl_pcie_apply_destination(trans);
240
241	/* kick FW self load */
242	iwl_write64(trans, CSR_CTXT_INFO_BA, trans_pcie->ctxt_info_dma_addr);
243
244	/* Context info will be released upon alive or failure to get one */
245
246	return 0;
247}
248
249void iwl_pcie_ctxt_info_free(struct iwl_trans *trans)
250{
251	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
252
253	if (!trans_pcie->ctxt_info)
254		return;
255
256	dma_free_coherent(trans->dev, sizeof(*trans_pcie->ctxt_info),
257			  trans_pcie->ctxt_info,
258			  trans_pcie->ctxt_info_dma_addr);
259	trans_pcie->ctxt_info_dma_addr = 0;
260	trans_pcie->ctxt_info = NULL;
261
262	iwl_pcie_ctxt_info_free_fw_img(trans);
263}
264