forked from ~ljy/RK356X_SDK_RELEASE
blame | history | patch | commit | commitdiff
hc
2023-12-11 072de836f53be56a70cecf70b43ae43b7ce17376 
 kernel/arch/powerpc/mm/init_64.c
....@@ -1,3 +1,4 @@
1
+// SPDX-License-Identifier: GPL-2.0-or-later
12 /*
23  *  PowerPC version
34  *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
....@@ -11,12 +12,6 @@
1112  *
1213  *  Dave Engebretsen <engebret@us.ibm.com>
1314  *      Rework for PPC64 port.
14
- *
15
- *  This program is free software; you can redistribute it and/or
16
- *  modify it under the terms of the GNU General Public License
17
- *  as published by the Free Software Foundation; either version
18
- *  2 of the License, or (at your option) any later version.
19
- *
2015  */
2116 
2217 #undef DEBUG
....@@ -52,7 +47,6 @@
5247 #include <asm/rtas.h>
5348 #include <asm/io.h>
5449 #include <asm/mmu_context.h>
55
-#include <asm/pgtable.h>
5650 #include <asm/mmu.h>
5751 #include <linux/uaccess.h>
5852 #include <asm/smp.h>
....@@ -66,40 +60,50 @@
6660 #include <asm/iommu.h>
6761 #include <asm/vdso.h>
6862 
69
-#include "mmu_decl.h"
70
-
71
-phys_addr_t memstart_addr = ~0;
72
-EXPORT_SYMBOL_GPL(memstart_addr);
73
-phys_addr_t kernstart_addr;
74
-EXPORT_SYMBOL_GPL(kernstart_addr);
63
+#include <mm/mmu_decl.h>
7564 
7665 #ifdef CONFIG_SPARSEMEM_VMEMMAP
7766 /*
78
- * Given an address within the vmemmap, determine the pfn of the page that
79
- * represents the start of the section it is within.  Note that we have to
67
+ * Given an address within the vmemmap, determine the page that
68
+ * represents the start of the subsection it is within.  Note that we have to
8069  * do this by hand as the proffered address may not be correctly aligned.
8170  * Subtraction of non-aligned pointers produces undefined results.
8271  */
83
-static unsigned long __meminit vmemmap_section_start(unsigned long page)
72
+static struct page * __meminit vmemmap_subsection_start(unsigned long vmemmap_addr)
8473 {
85
-	unsigned long offset = page - ((unsigned long)(vmemmap));
74
+	unsigned long start_pfn;
75
+	unsigned long offset = vmemmap_addr - ((unsigned long)(vmemmap));
8676 
8777 	/* Return the pfn of the start of the section. */
88
-	return (offset / sizeof(struct page)) & PAGE_SECTION_MASK;
78
+	start_pfn = (offset / sizeof(struct page)) & PAGE_SUBSECTION_MASK;
79
+	return pfn_to_page(start_pfn);
8980 }
9081 
9182 /*
92
- * Check if this vmemmap page is already initialised.  If any section
93
- * which overlaps this vmemmap page is initialised then this page is
94
- * initialised already.
83
+ * Since memory is added in sub-section chunks, before creating a new vmemmap
84
+ * mapping, the kernel should check whether there is an existing memmap mapping
85
+ * covering the new subsection added. This is needed because kernel can map
86
+ * vmemmap area using 16MB pages which will cover a memory range of 16G. Such
87
+ * a range covers multiple subsections (2M)
88
+ *
89
+ * If any subsection in the 16G range mapped by vmemmap is valid we consider the
90
+ * vmemmap populated (There is a page table entry already present). We can't do
91
+ * a page table lookup here because with the hash translation we don't keep
92
+ * vmemmap details in linux page table.
9593  */
96
-static int __meminit vmemmap_populated(unsigned long start, int page_size)
94
+static int __meminit vmemmap_populated(unsigned long vmemmap_addr, int vmemmap_map_size)
9795 {
98
-	unsigned long end = start + page_size;
99
-	start = (unsigned long)(pfn_to_page(vmemmap_section_start(start)));
96
+	struct page *start;
97
+	unsigned long vmemmap_end = vmemmap_addr + vmemmap_map_size;
98
+	start = vmemmap_subsection_start(vmemmap_addr);
10099 
101
-	for (; start < end; start += (PAGES_PER_SECTION * sizeof(struct page)))
102
-		if (pfn_valid(page_to_pfn((struct page *)start)))
100
+	for (; (unsigned long)start < vmemmap_end; start += PAGES_PER_SUBSECTION)
101
+		/*
102
+		 * pfn valid check here is intended to really check
103
+		 * whether we have any subsection already initialized
104
+		 * in this range.
105
+		 */
106
+		if (pfn_valid(page_to_pfn(start)))
103107 			return 1;
104108 
105109 	return 0;
....@@ -158,16 +162,16 @@
158162 	return next++;
159163 }
160164 
161
-static __meminit void vmemmap_list_populate(unsigned long phys,
162
-					    unsigned long start,
163
-					    int node)
165
+static __meminit int vmemmap_list_populate(unsigned long phys,
166
+					   unsigned long start,
167
+					   int node)
164168 {
165169 	struct vmemmap_backing *vmem_back;
166170 
167171 	vmem_back = vmemmap_list_alloc(node);
168172 	if (unlikely(!vmem_back)) {
169
-		WARN_ON(1);
170
-		return;
173
+		pr_debug("vmemap list allocation failed\n");
174
+		return -ENOMEM;
171175 	}
172176 
173177 	vmem_back->phys = phys;
....@@ -175,15 +179,32 @@
175179 	vmem_back->list = vmemmap_list;
176180 
177181 	vmemmap_list = vmem_back;
182
+	return 0;
183
+}
184
+
185
+static bool altmap_cross_boundary(struct vmem_altmap *altmap, unsigned long start,
186
+				unsigned long page_size)
187
+{
188
+	unsigned long nr_pfn = page_size / sizeof(struct page);
189
+	unsigned long start_pfn = page_to_pfn((struct page *)start);
190
+
191
+	if ((start_pfn + nr_pfn) > altmap->end_pfn)
192
+		return true;
193
+
194
+	if (start_pfn < altmap->base_pfn)
195
+		return true;
196
+
197
+	return false;
178198 }
179199 
180200 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
181201 		struct vmem_altmap *altmap)
182202 {
203
+	bool altmap_alloc;
183204 	unsigned long page_size = 1 << mmu_psize_defs[mmu_vmemmap_psize].shift;
184205 
185206 	/* Align to the page size of the linear mapping. */
186
-	start = _ALIGN_DOWN(start, page_size);
207
+	start = ALIGN_DOWN(start, page_size);
187208 
188209 	pr_debug("vmemmap_populate %lx..%lx, node %d\n", start, end, node);
189210 
....@@ -191,6 +212,12 @@
191212 		void *p = NULL;
192213 		int rc;
193214 
215
+		/*
216
+		 * This vmemmap range is backing different subsections. If any
217
+		 * of that subsection is marked valid, that means we already
218
+		 * have initialized a page table covering this range and hence
219
+		 * the vmemmap range is populated.
220
+		 */
194221 		if (vmemmap_populated(start, page_size))
195222 			continue;
196223 
....@@ -199,14 +226,36 @@
199226 		 * fail due to alignment issues when using 16MB hugepages, so
200227 		 * fall back to system memory if the altmap allocation fail.
201228 		 */
202
-		if (altmap)
203
-			p = altmap_alloc_block_buf(page_size, altmap);
204
-		if (!p)
205
-			p = vmemmap_alloc_block_buf(page_size, node);
229
+		if (altmap && !altmap_cross_boundary(altmap, start, page_size)) {
230
+			p = vmemmap_alloc_block_buf(page_size, node, altmap);
231
+			if (!p)
232
+				pr_debug("altmap block allocation failed, falling back to system memory");
233
+			else
234
+				altmap_alloc = true;
235
+		}
236
+		if (!p) {
237
+			p = vmemmap_alloc_block_buf(page_size, node, NULL);
238
+			altmap_alloc = false;
239
+		}
206240 		if (!p)
207241 			return -ENOMEM;
208242 
209
-		vmemmap_list_populate(__pa(p), start, node);
243
+		if (vmemmap_list_populate(__pa(p), start, node)) {
244
+			/*
245
+			 * If we don't populate vmemap list, we don't have
246
+			 * the ability to free the allocated vmemmap
247
+			 * pages in section_deactivate. Hence free them
248
+			 * here.
249
+			 */
250
+			int nr_pfns = page_size >> PAGE_SHIFT;
251
+			unsigned long page_order = get_order(page_size);
252
+
253
+			if (altmap_alloc)
254
+				vmem_altmap_free(altmap, nr_pfns);
255
+			else
256
+				free_pages((unsigned long)p, page_order);
257
+			return -ENOMEM;
258
+		}
210259 
211260 		pr_debug("      * %016lx..%016lx allocated at %p\n",
212261 			 start, start + page_size, p);
....@@ -236,10 +285,8 @@
236285 		vmem_back_prev = vmem_back;
237286 	}
238287 
239
-	if (unlikely(!vmem_back)) {
240
-		WARN_ON(1);
288
+	if (unlikely(!vmem_back))
241289 		return 0;
242
-	}
243290 
244291 	/* remove it from vmemmap_list */
245292 	if (vmem_back == vmemmap_list) /* remove head */
....@@ -263,7 +310,7 @@
263310 	unsigned long alt_start = ~0, alt_end = ~0;
264311 	unsigned long base_pfn;
265312 
266
-	start = _ALIGN_DOWN(start, page_size);
313
+	start = ALIGN_DOWN(start, page_size);
267314 	if (altmap) {
268315 		alt_start = altmap->base_pfn;
269316 		alt_end = altmap->base_pfn + altmap->reserve +
....@@ -274,13 +321,13 @@
274321 
275322 	for (; start < end; start += page_size) {
276323 		unsigned long nr_pages, addr;
277
-		struct page *section_base;
278324 		struct page *page;
279325 
280326 		/*
281
-		 * the section has already be marked as invalid, so
282
-		 * vmemmap_populated() true means some other sections still
283
-		 * in this page, so skip it.
327
+		 * We have already marked the subsection we are trying to remove
328
+		 * invalid. So if we want to remove the vmemmap range, we
329
+		 * need to make sure there is no subsection marked valid
330
+		 * in this range.
284331 		 */
285332 		if (vmemmap_populated(start, page_size))
286333 			continue;
....@@ -290,7 +337,6 @@
290337 			continue;
291338 
292339 		page = pfn_to_page(addr >> PAGE_SHIFT);
293
-		section_base = pfn_to_page(vmemmap_section_start(start));
294340 		nr_pages = 1 << page_order;
295341 		base_pfn = PHYS_PFN(addr);
296342 
....@@ -379,13 +425,15 @@
379425 		}
380426 		if (!(vec5[OV5_INDX(OV5_RADIX_GTSE)] &
381427 						OV5_FEAT(OV5_RADIX_GTSE))) {
382
-			pr_warn("WARNING: Hypervisor doesn't support RADIX with GTSE\n");
383
-		}
428
+			cur_cpu_spec->mmu_features &= ~MMU_FTR_GTSE;
429
+		} else
430
+			cur_cpu_spec->mmu_features |= MMU_FTR_GTSE;
384431 		/* Do radix anyway - the hypervisor said we had to */
385432 		cur_cpu_spec->mmu_features |= MMU_FTR_TYPE_RADIX;
386433 	} else if (mmu_supported == OV5_FEAT(OV5_MMU_HASH)) {
387434 		/* Hypervisor only supports hash - disable radix */
388435 		cur_cpu_spec->mmu_features &= ~MMU_FTR_TYPE_RADIX;
436
+		cur_cpu_spec->mmu_features &= ~MMU_FTR_GTSE;
389437 	}
390438 }
391439 
....@@ -404,9 +452,16 @@
404452 	if (!(mfmsr() & MSR_HV))
405453 		early_check_vec5();
406454 
407
-	if (early_radix_enabled())
455
+	if (early_radix_enabled()) {
408456 		radix__early_init_devtree();
409
-	else
457
+		/*
458
+		 * We have finalized the translation we are going to use by now.
459
+		 * Radix mode is not limited by RMA / VRMA addressing.
460
+		 * Hence don't limit memblock allocations.
461
+		 */
462
+		ppc64_rma_size = ULONG_MAX;
463
+		memblock_set_current_limit(MEMBLOCK_ALLOC_ANYWHERE);
464
+	} else
410465 		hash__early_init_devtree();
411466 }
412467 #endif /* CONFIG_PPC_BOOK3S_64 */