* For buddy system initialization, the holes between banks adjacent will be released to buddy allocator?
@ 2012-08-26 10:39 Li Haifeng
2012-08-27 2:42 ` Li Haifeng
0 siblings, 1 reply; 5+ messages in thread
From: Li Haifeng @ 2012-08-26 10:39 UTC (permalink / raw)
To: linux-arm-kernel
Sorry to disturb you again.
I am still confused about holes between banks adjacent.
The lines 180~181 get the low memory's range, and the range includes holes.
But while releasing the memory to buddy allocator, the code doesn't
prevent the holes from releasing to the buddy allocator.
172 static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata)
173 {
174 struct page *page;
175 unsigned long start, end, pages, count = 0;
176
177 if (!bdata->node_bootmem_map)
178 return 0;
179
180 start = bdata->node_min_pfn;
181 end = bdata->node_low_pfn;
182
183 bdebug("nid=%td start=%lx end=%lx\n",
184 bdata - bootmem_node_data, start, end);
185
186 while (start < end) {
187 unsigned long *map, idx, vec;
188
189 map = bdata->node_bootmem_map;
190 idx = start - bdata->node_min_pfn;
191 vec = ~map[idx / BITS_PER_LONG];
192 /*
193 * If we have a properly aligned and fully unreserved
194 * BITS_PER_LONG block of pages in front of us, free
195 * it in one go.
196 */
197 if (IS_ALIGNED(start, BITS_PER_LONG) && vec == ~0UL) {
198 int order = ilog2(BITS_PER_LONG);
199
200 __free_pages_bootmem(pfn_to_page(start), order);
201 count += BITS_PER_LONG;
202 start += BITS_PER_LONG;
203 } else {
204 unsigned long off = 0;
205
206 while (vec && off < BITS_PER_LONG) {
207 if (vec & 1) {
208 page = pfn_to_page(start + off);
209 __free_pages_bootmem(page, 0);
210 count++;
211 }
212 vec >>= 1;
213 off++;
214 }
215 start = ALIGN(start + 1, BITS_PER_LONG);
216 }
217 }
218
219 page = virt_to_page(bdata->node_bootmem_map);
220 pages = bdata->node_low_pfn - bdata->node_min_pfn;
221 pages = bootmem_bootmap_pages(pages);
222 count += pages;
223 while (pages--)
224 __free_pages_bootmem(page++, 0);
225
226 bdebug("nid=%td released=%lx\n", bdata - bootmem_node_data, count);
227
228 return count;
229 }
^ permalink raw reply [flat|nested] 5+ messages in thread* For buddy system initialization, the holes between banks adjacent will be released to buddy allocator?
2012-08-26 10:39 For buddy system initialization, the holes between banks adjacent will be released to buddy allocator? Li Haifeng
@ 2012-08-27 2:42 ` Li Haifeng
2012-08-27 8:09 ` Russell King - ARM Linux
0 siblings, 1 reply; 5+ messages in thread
From: Li Haifeng @ 2012-08-27 2:42 UTC (permalink / raw)
To: linux-arm-kernel
Just now, I tested it on my ARM development board.
The holes between banks does be released to buddy allocator. And the
buddy system will allocate holes for thread, which will cause fault.
Is it right?
2012/8/26 Li Haifeng <omycle@gmail.com>:
> Sorry to disturb you again.
>
> I am still confused about holes between banks adjacent.
>
> The lines 180~181 get the low memory's range, and the range includes holes.
> But while releasing the memory to buddy allocator, the code doesn't
> prevent the holes from releasing to the buddy allocator.
>
> 172 static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata)
> 173 {
> 174 struct page *page;
> 175 unsigned long start, end, pages, count = 0;
> 176
> 177 if (!bdata->node_bootmem_map)
> 178 return 0;
> 179
> 180 start = bdata->node_min_pfn;
> 181 end = bdata->node_low_pfn;
> 182
> 183 bdebug("nid=%td start=%lx end=%lx\n",
> 184 bdata - bootmem_node_data, start, end);
> 185
> 186 while (start < end) {
> 187 unsigned long *map, idx, vec;
> 188
> 189 map = bdata->node_bootmem_map;
> 190 idx = start - bdata->node_min_pfn;
> 191 vec = ~map[idx / BITS_PER_LONG];
> 192 /*
> 193 * If we have a properly aligned and fully unreserved
> 194 * BITS_PER_LONG block of pages in front of us, free
> 195 * it in one go.
> 196 */
> 197 if (IS_ALIGNED(start, BITS_PER_LONG) && vec == ~0UL) {
> 198 int order = ilog2(BITS_PER_LONG);
> 199
> 200 __free_pages_bootmem(pfn_to_page(start), order);
> 201 count += BITS_PER_LONG;
> 202 start += BITS_PER_LONG;
> 203 } else {
> 204 unsigned long off = 0;
> 205
> 206 while (vec && off < BITS_PER_LONG) {
> 207 if (vec & 1) {
> 208 page = pfn_to_page(start + off);
> 209 __free_pages_bootmem(page, 0);
> 210 count++;
> 211 }
> 212 vec >>= 1;
> 213 off++;
> 214 }
> 215 start = ALIGN(start + 1, BITS_PER_LONG);
> 216 }
> 217 }
> 218
> 219 page = virt_to_page(bdata->node_bootmem_map);
> 220 pages = bdata->node_low_pfn - bdata->node_min_pfn;
> 221 pages = bootmem_bootmap_pages(pages);
> 222 count += pages;
> 223 while (pages--)
> 224 __free_pages_bootmem(page++, 0);
> 225
> 226 bdebug("nid=%td released=%lx\n", bdata - bootmem_node_data, count);
> 227
> 228 return count;
> 229 }
^ permalink raw reply [flat|nested] 5+ messages in thread* For buddy system initialization, the holes between banks adjacent will be released to buddy allocator?
2012-08-27 2:42 ` Li Haifeng
@ 2012-08-27 8:09 ` Russell King - ARM Linux
2012-08-27 10:30 ` Li Haifeng
0 siblings, 1 reply; 5+ messages in thread
From: Russell King - ARM Linux @ 2012-08-27 8:09 UTC (permalink / raw)
To: linux-arm-kernel
On Mon, Aug 27, 2012 at 10:42:40AM +0800, Li Haifeng wrote:
> Just now, I tested it on my ARM development board.
>
> The holes between banks does be released to buddy allocator. And the
> buddy system will allocate holes for thread, which will cause fault.
> Is it right?
Could you rephrase your question please?
^ permalink raw reply [flat|nested] 5+ messages in thread
* For buddy system initialization, the holes between banks adjacent will be released to buddy allocator?
2012-08-27 8:09 ` Russell King - ARM Linux
@ 2012-08-27 10:30 ` Li Haifeng
2012-08-27 10:54 ` Russell King - ARM Linux
0 siblings, 1 reply; 5+ messages in thread
From: Li Haifeng @ 2012-08-27 10:30 UTC (permalink / raw)
To: linux-arm-kernel
2012/8/27 Russell King - ARM Linux <linux@arm.linux.org.uk>:
> On Mon, Aug 27, 2012 at 10:42:40AM +0800, Li Haifeng wrote:
>> Just now, I tested it on my ARM development board.
>>
>> The holes between banks does be released to buddy allocator. And the
>> buddy system will allocate holes for thread, which will cause fault.
>> Is it right?
>
> Could you rephrase your question please?
The code below is to release free memory to buddy allocator, according
to node_bootmem_map.
The lines 180~181 get the low memory's range, and the range includes holes.
But while releasing the memory to buddy allocator, the code doesn't
prevent the holes from releasing to the buddy allocator. So, if the
holes are allocated to thread, fault will happen.
IMO, it is a bug. Is it right?
172 static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata)
173 {
174 struct page *page;
175 unsigned long start, end, pages, count = 0;
176
177 if (!bdata->node_bootmem_map)
178 return 0;
179
180 start = bdata->node_min_pfn;
181 end = bdata->node_low_pfn;
182
183 bdebug("nid=%td start=%lx end=%lx\n",
184 bdata - bootmem_node_data, start, end);
185
186 while (start < end) {
187 unsigned long *map, idx, vec;
188
189 map = bdata->node_bootmem_map;
190 idx = start - bdata->node_min_pfn;
191 vec = ~map[idx / BITS_PER_LONG];
192 /*
193 * If we have a properly aligned and fully unreserved
194 * BITS_PER_LONG block of pages in front of us, free
195 * it in one go.
196 */
197 if (IS_ALIGNED(start, BITS_PER_LONG) && vec == ~0UL) {
198 int order = ilog2(BITS_PER_LONG);
199
200 __free_pages_bootmem(pfn_to_page(start), order);
201 count += BITS_PER_LONG;
202 start += BITS_PER_LONG;
203 } else {
204 unsigned long off = 0;
205
206 while (vec && off < BITS_PER_LONG) {
207 if (vec & 1) {
208 page = pfn_to_page(start + off);
209 __free_pages_bootmem(page, 0);
210 count++;
211 }
212 vec >>= 1;
213 off++;
214 }
215 start = ALIGN(start + 1, BITS_PER_LONG);
216 }
217 }
218
219 page = virt_to_page(bdata->node_bootmem_map);
220 pages = bdata->node_low_pfn - bdata->node_min_pfn;
221 pages = bootmem_bootmap_pages(pages);
222 count += pages;
223 while (pages--)
224 __free_pages_bootmem(page++, 0);
225
226 bdebug("nid=%td released=%lx\n", bdata - bootmem_node_data, count);
227
228 return count;
229 }
^ permalink raw reply [flat|nested] 5+ messages in thread* For buddy system initialization, the holes between banks adjacent will be released to buddy allocator?
2012-08-27 10:30 ` Li Haifeng
@ 2012-08-27 10:54 ` Russell King - ARM Linux
0 siblings, 0 replies; 5+ messages in thread
From: Russell King - ARM Linux @ 2012-08-27 10:54 UTC (permalink / raw)
To: linux-arm-kernel
On Mon, Aug 27, 2012 at 06:30:42PM +0800, Li Haifeng wrote:
> 2012/8/27 Russell King - ARM Linux <linux@arm.linux.org.uk>:
> > On Mon, Aug 27, 2012 at 10:42:40AM +0800, Li Haifeng wrote:
> >> Just now, I tested it on my ARM development board.
> >>
> >> The holes between banks does be released to buddy allocator. And the
> >> buddy system will allocate holes for thread, which will cause fault.
> >> Is it right?
> >
> > Could you rephrase your question please?
>
> The code below is to release free memory to buddy allocator, according
> to node_bootmem_map.
>
> The lines 180~181 get the low memory's range, and the range includes holes.
> But while releasing the memory to buddy allocator, the code doesn't
> prevent the holes from releasing to the buddy allocator. So, if the
> holes are allocated to thread, fault will happen.
The whole purpose _is_ to allow these pages to be re-used, rather than
being tied up uselessly in the map where they're not being used.
^ permalink raw reply [flat|nested] 5+ messages in thread
end of thread, other threads:[~2012-08-27 10:54 UTC | newest]
Thread overview: 5+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2012-08-26 10:39 For buddy system initialization, the holes between banks adjacent will be released to buddy allocator? Li Haifeng
2012-08-27 2:42 ` Li Haifeng
2012-08-27 8:09 ` Russell King - ARM Linux
2012-08-27 10:30 ` Li Haifeng
2012-08-27 10:54 ` Russell King - ARM Linux
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).