forked from ~ljy/RK356X_SDK_RELEASE
blame | history | patch | commit | commitdiff
hc
2023-12-06 08f87f769b595151be1afeff53e144f543faa614 
 kernel/drivers/gpu/drm/omapdrm/omap_dmm_tiler.c
....@@ -18,6 +18,7 @@
1818 #include <linux/completion.h>
1919 #include <linux/delay.h>
2020 #include <linux/dma-mapping.h>
21
+#include <linux/dmaengine.h>
2122 #include <linux/errno.h>
2223 #include <linux/init.h>
2324 #include <linux/interrupt.h>
....@@ -79,14 +80,136 @@
7980 			DMM_PAT_DESCR__2, DMM_PAT_DESCR__3},
8081 };
8182 
83
+static int dmm_dma_copy(struct dmm *dmm, dma_addr_t src, dma_addr_t dst)
84
+{
85
+	struct dma_async_tx_descriptor *tx;
86
+	enum dma_status status;
87
+	dma_cookie_t cookie;
88
+
89
+	tx = dmaengine_prep_dma_memcpy(dmm->wa_dma_chan, dst, src, 4, 0);
90
+	if (!tx) {
91
+		dev_err(dmm->dev, "Failed to prepare DMA memcpy\n");
92
+		return -EIO;
93
+	}
94
+
95
+	cookie = tx->tx_submit(tx);
96
+	if (dma_submit_error(cookie)) {
97
+		dev_err(dmm->dev, "Failed to do DMA tx_submit\n");
98
+		return -EIO;
99
+	}
100
+
101
+	status = dma_sync_wait(dmm->wa_dma_chan, cookie);
102
+	if (status != DMA_COMPLETE)
103
+		dev_err(dmm->dev, "i878 wa DMA copy failure\n");
104
+
105
+	dmaengine_terminate_all(dmm->wa_dma_chan);
106
+	return 0;
107
+}
108
+
109
+static u32 dmm_read_wa(struct dmm *dmm, u32 reg)
110
+{
111
+	dma_addr_t src, dst;
112
+	int r;
113
+
114
+	src = dmm->phys_base + reg;
115
+	dst = dmm->wa_dma_handle;
116
+
117
+	r = dmm_dma_copy(dmm, src, dst);
118
+	if (r) {
119
+		dev_err(dmm->dev, "sDMA read transfer timeout\n");
120
+		return readl(dmm->base + reg);
121
+	}
122
+
123
+	/*
124
+	 * As per i878 workaround, the DMA is used to access the DMM registers.
125
+	 * Make sure that the readl is not moved by the compiler or the CPU
126
+	 * earlier than the DMA finished writing the value to memory.
127
+	 */
128
+	rmb();
129
+	return readl(dmm->wa_dma_data);
130
+}
131
+
132
+static void dmm_write_wa(struct dmm *dmm, u32 val, u32 reg)
133
+{
134
+	dma_addr_t src, dst;
135
+	int r;
136
+
137
+	writel(val, dmm->wa_dma_data);
138
+	/*
139
+	 * As per i878 workaround, the DMA is used to access the DMM registers.
140
+	 * Make sure that the writel is not moved by the compiler or the CPU, so
141
+	 * the data will be in place before we start the DMA to do the actual
142
+	 * register write.
143
+	 */
144
+	wmb();
145
+
146
+	src = dmm->wa_dma_handle;
147
+	dst = dmm->phys_base + reg;
148
+
149
+	r = dmm_dma_copy(dmm, src, dst);
150
+	if (r) {
151
+		dev_err(dmm->dev, "sDMA write transfer timeout\n");
152
+		writel(val, dmm->base + reg);
153
+	}
154
+}
155
+
82156 static u32 dmm_read(struct dmm *dmm, u32 reg)
83157 {
84
-	return readl(dmm->base + reg);
158
+	if (dmm->dmm_workaround) {
159
+		u32 v;
160
+		unsigned long flags;
161
+
162
+		spin_lock_irqsave(&dmm->wa_lock, flags);
163
+		v = dmm_read_wa(dmm, reg);
164
+		spin_unlock_irqrestore(&dmm->wa_lock, flags);
165
+
166
+		return v;
167
+	} else {
168
+		return readl(dmm->base + reg);
169
+	}
85170 }
86171 
87172 static void dmm_write(struct dmm *dmm, u32 val, u32 reg)
88173 {
89
-	writel(val, dmm->base + reg);
174
+	if (dmm->dmm_workaround) {
175
+		unsigned long flags;
176
+
177
+		spin_lock_irqsave(&dmm->wa_lock, flags);
178
+		dmm_write_wa(dmm, val, reg);
179
+		spin_unlock_irqrestore(&dmm->wa_lock, flags);
180
+	} else {
181
+		writel(val, dmm->base + reg);
182
+	}
183
+}
184
+
185
+static int dmm_workaround_init(struct dmm *dmm)
186
+{
187
+	dma_cap_mask_t mask;
188
+
189
+	spin_lock_init(&dmm->wa_lock);
190
+
191
+	dmm->wa_dma_data = dma_alloc_coherent(dmm->dev,  sizeof(u32),
192
+					      &dmm->wa_dma_handle, GFP_KERNEL);
193
+	if (!dmm->wa_dma_data)
194
+		return -ENOMEM;
195
+
196
+	dma_cap_zero(mask);
197
+	dma_cap_set(DMA_MEMCPY, mask);
198
+
199
+	dmm->wa_dma_chan = dma_request_channel(mask, NULL, NULL);
200
+	if (!dmm->wa_dma_chan) {
201
+		dma_free_coherent(dmm->dev, 4, dmm->wa_dma_data, dmm->wa_dma_handle);
202
+		return -ENODEV;
203
+	}
204
+
205
+	return 0;
206
+}
207
+
208
+static void dmm_workaround_uninit(struct dmm *dmm)
209
+{
210
+	dma_release_channel(dmm->wa_dma_chan);
211
+
212
+	dma_free_coherent(dmm->dev, 4, dmm->wa_dma_data, dmm->wa_dma_handle);
90213 }
91214 
92215 /* simple allocator to grab next 16 byte aligned memory from txn */
....@@ -614,6 +737,10 @@
614737 	unsigned long flags;
615738 
616739 	if (omap_dmm) {
740
+		/* Disable all enabled interrupts */
741
+		dmm_write(omap_dmm, 0x7e7e7e7e, DMM_PAT_IRQENABLE_CLR);
742
+		free_irq(omap_dmm->irq, omap_dmm);
743
+
617744 		/* free all area regions */
618745 		spin_lock_irqsave(&list_lock, flags);
619746 		list_for_each_entry_safe(block, _block, &omap_dmm->alloc_head,
....@@ -636,8 +763,8 @@
636763 		if (omap_dmm->dummy_page)
637764 			__free_page(omap_dmm->dummy_page);
638765 
639
-		if (omap_dmm->irq > 0)
640
-			free_irq(omap_dmm->irq, omap_dmm);
766
+		if (omap_dmm->dmm_workaround)
767
+			dmm_workaround_uninit(omap_dmm);
641768 
642769 		iounmap(omap_dmm->base);
643770 		kfree(omap_dmm);
....@@ -684,6 +811,7 @@
684811 		goto fail;
685812 	}
686813 
814
+	omap_dmm->phys_base = mem->start;
687815 	omap_dmm->base = ioremap(mem->start, SZ_2K);
688816 
689817 	if (!omap_dmm->base) {
....@@ -698,6 +826,22 @@
698826 	}
699827 
700828 	omap_dmm->dev = &dev->dev;
829
+
830
+	if (of_machine_is_compatible("ti,dra7")) {
831
+		/*
832
+		 * DRA7 Errata i878 says that MPU should not be used to access
833
+		 * RAM and DMM at the same time. As it's not possible to prevent
834
+		 * MPU accessing RAM, we need to access DMM via a proxy.
835
+		 */
836
+		if (!dmm_workaround_init(omap_dmm)) {
837
+			omap_dmm->dmm_workaround = true;
838
+			dev_info(&dev->dev,
839
+				"workaround for errata i878 in use\n");
840
+		} else {
841
+			dev_warn(&dev->dev,
842
+				 "failed to initialize work-around for i878\n");
843
+		}
844
+	}
701845 
702846 	hwinfo = dmm_read(omap_dmm, DMM_PAT_HWINFO);
703847 	omap_dmm->num_engines = (hwinfo >> 24) & 0x1F;
....@@ -724,24 +868,6 @@
724868 	dmm_write(omap_dmm, 0x80000000, DMM_PAT_VIEW_MAP_BASE);
725869 	dmm_write(omap_dmm, 0x88888888, DMM_TILER_OR__0);
726870 	dmm_write(omap_dmm, 0x88888888, DMM_TILER_OR__1);
727
-
728
-	ret = request_irq(omap_dmm->irq, omap_dmm_irq_handler, IRQF_SHARED,
729
-				"omap_dmm_irq_handler", omap_dmm);
730
-
731
-	if (ret) {
732
-		dev_err(&dev->dev, "couldn't register IRQ %d, error %d\n",
733
-			omap_dmm->irq, ret);
734
-		omap_dmm->irq = -1;
735
-		goto fail;
736
-	}
737
-
738
-	/* Enable all interrupts for each refill engine except
739
-	 * ERR_LUT_MISS<n> (which is just advisory, and we don't care
740
-	 * about because we want to be able to refill live scanout
741
-	 * buffers for accelerated pan/scroll) and FILL_DSC<n> which
742
-	 * we just generally don't care about.
743
-	 */
744
-	dmm_write(omap_dmm, 0x7e7e7e7e, DMM_PAT_IRQENABLE_SET);
745871 
746872 	omap_dmm->dummy_page = alloc_page(GFP_KERNEL | __GFP_DMA32);
747873 	if (!omap_dmm->dummy_page) {
....@@ -835,6 +961,24 @@
835961 		.p1.y = omap_dmm->container_height - 1,
836962 	};
837963 
964
+	ret = request_irq(omap_dmm->irq, omap_dmm_irq_handler, IRQF_SHARED,
965
+				"omap_dmm_irq_handler", omap_dmm);
966
+
967
+	if (ret) {
968
+		dev_err(&dev->dev, "couldn't register IRQ %d, error %d\n",
969
+			omap_dmm->irq, ret);
970
+		omap_dmm->irq = -1;
971
+		goto fail;
972
+	}
973
+
974
+	/* Enable all interrupts for each refill engine except
975
+	 * ERR_LUT_MISS<n> (which is just advisory, and we don't care
976
+	 * about because we want to be able to refill live scanout
977
+	 * buffers for accelerated pan/scroll) and FILL_DSC<n> which
978
+	 * we just generally don't care about.
979
+	 */
980
+	dmm_write(omap_dmm, 0x7e7e7e7e, DMM_PAT_IRQENABLE_SET);
981
+
838982 	/* initialize all LUTs to dummy page entries */
839983 	for (i = 0; i < omap_dmm->num_lut; i++) {
840984 		area.tcm = omap_dmm->tcm[i];