On 02/23/2016 10:16 AM, Minoru Usui wrote:
Hello ZhouI'm sorry for late reply, too.-----Original Message----- From: "Zhou, Wenjian/周文剑" [mailto:[email protected]] Sent: Monday, February 15, 2016 11:15 AM To: Usui Minoru(碓井 成) <[email protected]>; [email protected] Subject: Re: [PATCH v1] Improve the performance of --num-threads -d 31 Hello Usui, Thanks very much for your comments. And sorry for the late reply. See below. On 02/08/2016 01:00 PM, Minoru Usui wrote:Hello,-----Original Message----- From: kexec [mailto:[email protected]] On Behalf Of Minoru Usui Sent: Thursday, February 04, 2016 8:52 AM To: Zhou Wenjian <[email protected]>; [email protected] Subject: Re: [PATCH v1] Improve the performance of --num-threads -d 31 Hi, Zhou I have some comments. I'm sorry if I have misunderstood your code.-----Original Message----- From: kexec [mailto:[email protected]] On Behalf Of Zhou Wenjian Sent: Monday, February 01, 2016 3:22 PM To: [email protected] Subject: [PATCH v1] Improve the performance of --num-threads -d 31 v1: 1. change page_flag.ready's value to enum 2. change the patch description 3. cleanup some codes 4. fix a bug in cyclic mode multi-threads implementation will introduce extra cost when handling each page. The origin implementation will also do the extra work for filtered pages. So there is a big performance degradation in --num-threads -d 31. The new implementation won't do the extra work for filtered pages any more. So the performance of -d 31 is close to that of serial processing. The new implementation is just like the following: * The basic idea is producer producing page and consumer writing page. * Each producer have a page_flag_buf list which is used for storing page's description. * The size of page_flag_buf is little so it won't take too much memory. * And all producers will share a page_data_buf array which is used for storing page's compressed data. * The main thread is the consumer. It will find the next pfn and write it into file. * The next pfn is smallest pfn in all page_flag_buf. Signed-off-by: Zhou Wenjian <[email protected]> --- makedumpfile.c | 258 ++++++++++++++++++++++++++++++++++++++------------------- makedumpfile.h | 31 ++++--- 2 files changed, 193 insertions(+), 96 deletions(-) diff --git a/makedumpfile.c b/makedumpfile.c index fa0b779..0ecd065 100644 --- a/makedumpfile.c +++ b/makedumpfile.c @@ -3483,7 +3483,8 @@ initial_for_parallel() unsigned long page_data_buf_size; unsigned long limit_size; int page_data_num; - int i; + struct page_flag *current; + int i, j; len_buf_out = calculate_len_buf_out(info->page_size); @@ -3562,8 +3563,10 @@ initial_for_parallel() - MAP_REGION * info->num_threads) * 0.6; page_data_num = limit_size / page_data_buf_size; + info->num_buffers = 3 * info->num_threads; - info->num_buffers = MIN(NUM_BUFFERS, page_data_num); + info->num_buffers = MAX(info->num_buffers, NUM_BUFFERS); + info->num_buffers = MIN(info->num_buffers, page_data_num); DEBUG_MSG("Number of struct page_data for produce/consume: %d\n", info->num_buffers); @@ -3588,6 +3591,36 @@ initial_for_parallel() } /* + * initial page_flag for each thread + */ + if ((info->page_flag_buf = malloc(sizeof(void *) * info->num_threads)) + == NULL) { + MSG("Can't allocate memory for page_flag_buf. %s\n", + strerror(errno)); + return FALSE; + } + memset(info->page_flag_buf, 0, sizeof(void *) * info->num_threads); + + for (i = 0; i < info->num_threads; i++) { + if ((info->page_flag_buf[i] = malloc(sizeof(struct page_flag))) == NULL) {Fist element of struct page_flag in circular list is allocated by malloc(), but other elements are allocated by calloc().(see below) I think both elements should be allocated by calloc().Yes, you are right. I have made a mistake.+ MSG("Can't allocate memory for page_flag_buf. %s\n", + strerror(errno)); + return FALSE; + } + current = info->page_flag_buf[i]; + + for (j = 1; j < NUM_BUFFERS; j++) { + if ((current->next = calloc(0, sizeof(struct page_flag))) == NULL) { + MSG("Can't allocate memory for data of page_data_buf. %s\n", + strerror(errno)); + return FALSE; + }First argument of calloc() should be 1, not 0. And there is typo in error message. Allocated element is not page_data_buf.I agree.+ current = current->next; + } + current->next = info->page_flag_buf[i]; + } + + /* * initial fd_memory for threads */ for (i = 0; i < info->num_threads; i++) { @@ -3612,7 +3645,8 @@ initial_for_parallel() void free_for_parallel() { - int i; + int i, j; + struct page_flag *current; if (info->threads != NULL) { for (i = 0; i < info->num_threads; i++) { @@ -3655,6 +3689,19 @@ free_for_parallel() free(info->page_data_buf); } + if (info->page_flag_buf != NULL) { + for (i = 0; i < info->num_threads; i++) { + for (j = 0; j < NUM_BUFFERS; j++) { + if (info->page_flag_buf[i] != NULL) { + current = info->page_flag_buf[i]; + info->page_flag_buf[i] = current->next; + free(current); + } + } + } + free(info->page_flag_buf); + } + if (info->parallel_info == NULL) return; @@ -7076,10 +7123,10 @@ kdump_thread_function_cyclic(void *arg) { void *retval = PTHREAD_FAIL; struct thread_args *kdump_thread_args = (struct thread_args *)arg; struct page_data *page_data_buf = kdump_thread_args->page_data_buf; + struct page_flag *page_flag_buf = kdump_thread_args->page_flag_buf; struct cycle *cycle = kdump_thread_args->cycle; - int page_data_num = kdump_thread_args->page_data_num; mdf_pfn_t pfn; - int index; + int index = kdump_thread_args->thread_num; int buf_ready; int dumpable; int fd_memory = 0; @@ -7125,47 +7172,47 @@ kdump_thread_function_cyclic(void *arg) { kdump_thread_args->thread_num); } - while (1) { - /* get next pfn */ - pthread_mutex_lock(&info->current_pfn_mutex); - pfn = info->current_pfn; - info->current_pfn++; - pthread_mutex_unlock(&info->current_pfn_mutex); + /* + * filtered page won't take anything + * unfiltered zero page will only take a page_flag_buf + * unfiltered non-zero page will take a page_flag_buf and a page_data_buf + */ + while (page_flag_buf->pfn < kdump_thread_args->end_pfn) {At first, page_flag_buf->pfn is not initialized. I think this block should be replaced with the following code. === do { : } while(page_flag_buf->pfn < kdump_thread_args->end_pfn) ===I'm sorry, above suggestion is meaningless in terms of page_flag_buf->pfn is uninitialized. It should be replaced like following. === while (1) { : while (buf_ready == FALSE) { : if (pfn >= kdump_thread_args->end_pfn) { : goto finish; } : } : } finish: ===page_flag_buf is allocated by calloc(). The page_flag_buf->pfn's value is 0. So I think it is not necessary to modify the code.Thanks, Minoru Usui+ buf_ready = FALSE; - if (pfn >= kdump_thread_args->end_pfn) - break; + while (page_data_buf[index].used != 0 || + pthread_mutex_trylock(&page_data_buf[index].mutex) != 0) + index = (index + 1) % info->num_buffers; - index = -1; - buf_ready = FALSE; + page_data_buf[index].used = 1;"1" is a magic number. It should be changed TRUE or FALSE.I see.while (buf_ready == FALSE) { pthread_testcancel(); - - index = pfn % page_data_num; - - if (pfn - info->consumed_pfn > info->num_buffers) + if (page_flag_buf->ready == FLAG_READY) continue;At first, page_flag_buf->ready is uninitialized, too. Should it be initialized in head part of this function, even if FLAG_UNUSED is defined 0?The same topic as the page_flag_buf is allocated by calloc().- if (page_data_buf[index].ready != 0) - continue; - - pthread_mutex_lock(&page_data_buf[index].mutex); + /* get next pfn */ + pthread_mutex_lock(&info->current_pfn_mutex); + pfn = info->current_pfn; + info->current_pfn++; + page_flag_buf->ready = FLAG_FILLING; + pthread_mutex_unlock(&info->current_pfn_mutex); - if (page_data_buf[index].ready != 0) - goto unlock; + page_flag_buf->pfn = pfn;It set FLAG_FILLING to page_flag_buf->ready before setting pfn to page_flag_buf->pfn. But consumer gets page_flag_buf->pfn after checking page_flag_buf->ready != FLAG_UNUSED in getting minimum pfn of each thread block. Should it set page_flag_buf->pfn first?Have you noticed the following code in the consumer? <cut> if (current_pfn == info->page_flag_buf[consuming]->pfn) break; <cut>No, I pointed following code. This part accesses info->page_flag_buf[i]->ready, then it accesses info->page_flag_buf[i]->pfn immediately. So, temp_pfn may be wrong pfn at this moment. --- for (i = 0; i < info->num_threads; i++) { if (info->page_flag_buf[i]->ready == FLAG_UNUSED) continue; temp_pfn = info->page_flag_buf[i]->pfn; ---The consumer will check if the pfn is changed after the page_flag_buf->ready turns to be FLAG_READY. So it's not important whether setting page_flag_buf->pfn first or not.As you said, consumer checks pfn which is changed. So it works well.In the other hand, even setting page_flag_buf->pfn first, if the pfn is not dumpable, the producer will also reset the page_flag_buf->pfn.Thank you for your explanation. I didn't notice that pfn can be undumpable.- buf_ready = TRUE; - - page_data_buf[index].pfn = pfn; - page_data_buf[index].ready = 1; + if (pfn >= kdump_thread_args->end_pfn) { + page_data_buf[index].used = 0; + page_flag_buf->ready = FLAG_READY; + info->current_pfn--; + break; + }This block decrements info->current_pfn without info->current_pfn_mutex. I think this block should be moved into previous pthread_mutex_lock(info->current_pfn_mutex) block, so it can remove.Why do you think it should have current_pfn_mutex? If pfn >= kdump_thread_args->end_pfn, info->current_pfn will always larger than kdump_thread_args->end_pfn. info->current_pfn-- won't affect anything. The decrement operation is for cyclic mode.dumpable = is_dumpable( info->fd_bitmap ? &bitmap_parallel : info->bitmap2, pfn, cycle); - page_data_buf[index].dumpable = dumpable; if (!dumpable) - goto unlock; + continue; if (!read_pfn_parallel(fd_memory, pfn, buf, &bitmap_memory_parallel, @@ -7178,11 +7225,11 @@ kdump_thread_function_cyclic(void *arg) { if ((info->dump_level & DL_EXCLUDE_ZERO) && is_zero_page(buf, info->page_size)) { - page_data_buf[index].zero = TRUE; - goto unlock; + page_flag_buf->zero = TRUE; + goto next; }First, this code gets page_data_buf, then it gets page_flag_buf. However, if processed pfn is zero page, it processes next pfn while keeping page_data_buf. I think it should get page_flag_buf, then get page_data_buf in order to shorten the holding period of the page_data_buf[index].mutex.Do you mean the following logic? 1. get the page_flag_buf first 2. if the pfn is not zero page, then get the page_data_buf.Yes.Think about the following case. A producer get the page_flag_buf, and the pfn is not zero page. It wants to get a page_data_buf, but there is no more page_data_buf. Then ...It's not a problem. In not zero page case, this logic needs both page_flag_buf and page_data_buf, so waiting buffer is obvious when it isn't able to get page_flag_buf or page_data_buf.
Of course, waiting is not a problem. But if other page_data_bufs are all used by later pfns, it will wait forever. That's the problem. -- Thanks Zhou
Since there are several page_data_bufs, it's not a problem that each producer will always hold a page_data_buf.It depends on the speed of consumer and producer. It's not possible to predict it. In zero page case, I think each producer executes more parallel theoretically if page_data_buf doesn't get. Thanks, Minoru UsuiThanks again for your comments. And I will post the next version later. -- Thanks ZhouThanks, Minoru Usui- page_data_buf[index].zero = FALSE; + page_flag_buf->zero = FALSE; /* * Compress the page data. @@ -7232,12 +7279,16 @@ kdump_thread_function_cyclic(void *arg) { page_data_buf[index].size = info->page_size; memcpy(page_data_buf[index].buf, buf, info->page_size); } -unlock: - pthread_mutex_unlock(&page_data_buf[index].mutex); + page_flag_buf->index = index; + buf_ready = TRUE; +next: + page_flag_buf->ready = FLAG_READY; + page_flag_buf = page_flag_buf->next; } - } + pthread_mutex_unlock(&page_data_buf[index].mutex); + } retval = NULL; fail: @@ -7265,14 +7316,15 @@ write_kdump_pages_parallel_cyclic(struct cache_data *cd_header, struct page_desc pd; struct timeval tv_start; struct timeval last, new; - unsigned long long consuming_pfn; pthread_t **threads = NULL; struct thread_args *kdump_thread_args = NULL; void *thread_result; - int page_data_num; + int page_buf_num; struct page_data *page_data_buf = NULL; int i; int index; + int end_count, consuming, check_count; + mdf_pfn_t current_pfn, temp_pfn; if (info->flag_elf_dumpfile) return FALSE; @@ -7319,16 +7371,11 @@ write_kdump_pages_parallel_cyclic(struct cache_data *cd_header, threads = info->threads; kdump_thread_args = info->kdump_thread_args; - page_data_num = info->num_buffers; + page_buf_num = info->num_buffers; page_data_buf = info->page_data_buf; - for (i = 0; i < page_data_num; i++) { - /* - * producer will use pfn in page_data_buf to decide the - * consumed pfn - */ - page_data_buf[i].pfn = start_pfn - 1; - page_data_buf[i].ready = 0; + for (i = 0; i < page_buf_num; i++) { + page_data_buf[i].used = 0; res = pthread_mutex_init(&page_data_buf[i].mutex, NULL); if (res != 0) { ERRMSG("Can't initialize mutex of page_data_buf. %s\n", @@ -7342,8 +7389,9 @@ write_kdump_pages_parallel_cyclic(struct cache_data *cd_header, kdump_thread_args[i].len_buf_out = len_buf_out; kdump_thread_args[i].start_pfn = start_pfn; kdump_thread_args[i].end_pfn = end_pfn; - kdump_thread_args[i].page_data_num = page_data_num; + kdump_thread_args[i].page_buf_num = page_buf_num; kdump_thread_args[i].page_data_buf = page_data_buf; + kdump_thread_args[i].page_flag_buf = info->page_flag_buf[i]; kdump_thread_args[i].cycle = cycle; res = pthread_create(threads[i], NULL, @@ -7356,55 +7404,94 @@ write_kdump_pages_parallel_cyclic(struct cache_data *cd_header, } } - consuming_pfn = start_pfn; - index = -1; + while (1) { + consuming = 0; + check_count = 0; + end_count = 0; - gettimeofday(&last, NULL); + /* + * The basic idea is producer producing page and consumer writing page. + * Each producer have a page_flag_buf list which is used for storing page's description. + * The size of page_flag_buf is little so it won't take too much memory. + * And all producers will share a page_data_buf array which is used for storing page's compressed data. + * The main thread is the consumer. It will find the next pfn and write it into file. + * The next pfn is smallest pfn in all page_flag_buf. + */ + while (1) { + current_pfn = end_pfn; - while (consuming_pfn < end_pfn) { - index = consuming_pfn % page_data_num; + /* + * page_flag_buf is in circular linked list. + * The array info->page_flag_buf[] records the current page_flag_buf in each thread's + * page_flag_buf list. + * consuming is used for recording in which thread the pfn is the smallest. + * current_pfn is used for recording the value of pfn when checking the pfn. + */ + for (i = 0; i < info->num_threads; i++) { + if (info->page_flag_buf[i]->ready == FLAG_UNUSED) + continue; + temp_pfn = info->page_flag_buf[i]->pfn; - gettimeofday(&new, NULL); - if (new.tv_sec - last.tv_sec > WAIT_TIME) { - ERRMSG("Can't get data of pfn %llx.\n", consuming_pfn); - goto out; - } + /* + * count how many threads have reached the end. + */ + if (temp_pfn >= end_pfn) { + end_count++; + info->page_flag_buf[i]->ready = FLAG_UNUSED; + continue; + } - /* - * check pfn first without mutex locked to reduce the time - * trying to lock the mutex - */ - if (page_data_buf[index].pfn != consuming_pfn) - continue; + if (current_pfn < temp_pfn) + continue; - if (pthread_mutex_trylock(&page_data_buf[index].mutex) != 0) - continue; + check_count++; + consuming = i; + current_pfn = temp_pfn; + } + + /* + * If all the threads have reached the end, we will finish writing. + */ + if (end_count >= info->num_threads) + goto finish; + + /* + * Since it has the probabilty that there is no page_flag_buf being ready, + * we should recheck if it happens. + */ + if (check_count == 0) + continue; + + /* + * When we check the pfn in page_flag_buf, it may be being produced. + * So we should wait until it is ready to use. And if the pfn is + * different from the value when we check, we should rechoose the buf. + */ + gettimeofday(&last, NULL); + while (info->page_flag_buf[consuming]->ready != FLAG_READY) { + gettimeofday(&new, NULL); + if (new.tv_sec - last.tv_sec > WAIT_TIME) { + ERRMSG("Can't get data of pfn.\n"); + goto out; + } + } - /* check whether the found one is ready to be consumed */ - if (page_data_buf[index].pfn != consuming_pfn || - page_data_buf[index].ready != 1) { - goto unlock; + if (current_pfn == info->page_flag_buf[consuming]->pfn) + break; } if ((num_dumped % per) == 0) print_progress(PROGRESS_COPY, num_dumped, info->num_dumpable); - /* next pfn is found, refresh last here */ - last = new; - consuming_pfn++; - info->consumed_pfn++; - page_data_buf[index].ready = 0; - - if (page_data_buf[index].dumpable == FALSE) - goto unlock; - num_dumped++; - if (page_data_buf[index].zero == TRUE) { + + if (info->page_flag_buf[consuming]->zero == TRUE) { if (!write_cache(cd_header, pd_zero, sizeof(page_desc_t))) goto out; pfn_zero++; } else { + index = info->page_flag_buf[consuming]->index; pd.flags = page_data_buf[index].flags; pd.size = page_data_buf[index].size; pd.page_flags = 0; @@ -7420,12 +7507,12 @@ write_kdump_pages_parallel_cyclic(struct cache_data *cd_header, */ if (!write_cache(cd_page, page_data_buf[index].buf, pd.size)) goto out; - + page_data_buf[index].used = 0; } -unlock: - pthread_mutex_unlock(&page_data_buf[index].mutex); + info->page_flag_buf[consuming]->ready = FLAG_UNUSED; + info->page_flag_buf[consuming] = info->page_flag_buf[consuming]->next; } - +finish: ret = TRUE; /* * print [100 %] @@ -7464,7 +7551,7 @@ out: } if (page_data_buf != NULL) { - for (i = 0; i < page_data_num; i++) { + for (i = 0; i < page_buf_num; i++) { pthread_mutex_destroy(&page_data_buf[i].mutex); } } @@ -7564,6 +7651,7 @@ write_kdump_pages_cyclic(struct cache_data *cd_header, struct cache_data *cd_pag num_dumped++; if (!read_pfn(pfn, buf)) goto out; + filter_data_buffer(buf, pfn_to_paddr(pfn), info->page_size); /* diff --git a/makedumpfile.h b/makedumpfile.h index e0b5bbf..8a9a5b2 100644 --- a/makedumpfile.h +++ b/makedumpfile.h @@ -977,7 +977,7 @@ typedef unsigned long long int ulonglong; #define PAGE_DATA_NUM (50) #define WAIT_TIME (60 * 10) #define PTHREAD_FAIL ((void *)-2) -#define NUM_BUFFERS (50) +#define NUM_BUFFERS (20) struct mmap_cache { char *mmap_buf; @@ -985,28 +985,36 @@ struct mmap_cache { off_t mmap_end_offset; }; +enum { + FLAG_UNUSED, + FLAG_READY, + FLAG_FILLING +}; +struct page_flag { + mdf_pfn_t pfn; + char zero; + char ready; + short index; + struct page_flag *next; +}; + struct page_data { - mdf_pfn_t pfn; - int dumpable; - int zero; - unsigned int flags; + pthread_mutex_t mutex; long size; unsigned char *buf; - pthread_mutex_t mutex; - /* - * whether the page_data is ready to be consumed - */ - int ready; + int flags; + int used; }; struct thread_args { int thread_num; unsigned long len_buf_out; mdf_pfn_t start_pfn, end_pfn; - int page_data_num; + int page_buf_num; struct cycle *cycle; struct page_data *page_data_buf; + struct page_flag *page_flag_buf; }; /* @@ -1295,6 +1303,7 @@ struct DumpInfo { pthread_t **threads; struct thread_args *kdump_thread_args; struct page_data *page_data_buf; + struct page_flag **page_flag_buf; pthread_rwlock_t usemmap_rwlock; mdf_pfn_t current_pfn; pthread_mutex_t current_pfn_mutex; -- 1.8.3.1 _______________________________________________ kexec mailing list [email protected] http://lists.infradead.org/mailman/listinfo/kexec_______________________________________________ kexec mailing list [email protected] http://lists.infradead.org/mailman/listinfo/kexec
_______________________________________________ kexec mailing list [email protected] http://lists.infradead.org/mailman/listinfo/kexec
