* Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
*/
-#include <linux/kernel.h>
-#include <linux/module.h>
-#include <linux/init.h>
-#include <linux/rwsem.h>
-#include <linux/proc_fs.h>
-#include <linux/file.h>
-#include <linux/smp.h>
-#include <linux/ctype.h>
-#include <asm/uaccess.h>
-#ifdef HAVE_MM_INLINE
-#include <linux/mm_inline.h>
-#endif
-
-#define DEBUG_SUBSYSTEM S_PORTALS
-#include <linux/kp30.h>
-#include <linux/portals_compat25.h>
-#include <linux/libcfs.h>
+#define DEBUG_SUBSYSTEM S_LNET
+#define LUSTRE_TRACEFILE_PRIVATE
+#include "tracefile.h"
-#define TCD_MAX_PAGES (5 << (20 - PAGE_SHIFT))
+#include <libcfs/kp30.h>
+#include <libcfs/libcfs.h>
/* XXX move things up to the top, comment */
+union trace_data_union (*trace_data[TCD_MAX_TYPES])[NR_CPUS] __cacheline_aligned;
-static union {
- struct trace_cpu_data {
- struct list_head tcd_pages;
- unsigned long tcd_cur_pages;
-
- struct list_head tcd_daemon_pages;
- unsigned long tcd_cur_daemon_pages;
-
- unsigned long tcd_max_pages;
- int tcd_shutting_down;
- } tcd;
- char __pad[SMP_CACHE_BYTES];
-} trace_data[NR_CPUS] __cacheline_aligned;
-
-struct page_collection {
- struct list_head pc_pages;
- spinlock_t pc_lock;
- int pc_want_daemon_pages;
-};
-
-struct tracefiled_ctl {
- struct completion tctl_start;
- struct completion tctl_stop;
- wait_queue_head_t tctl_waitq;
- pid_t tctl_pid;
- atomic_t tctl_shutdown;
-};
-
-#define TRACEFILE_SIZE (500 << 20)
-static DECLARE_RWSEM(tracefile_sem);
-static char *tracefile = NULL;
-static long long tracefile_size = TRACEFILE_SIZE;
+char tracefile[TRACEFILE_NAME_SIZE];
+long long tracefile_size = TRACEFILE_SIZE;
static struct tracefiled_ctl trace_tctl;
-static DECLARE_MUTEX(trace_thread_sem);
+struct semaphore trace_thread_sem;
static int thread_running = 0;
-#ifndef get_cpu
-#define get_cpu() smp_processor_id()
-#define put_cpu() do { } while (0)
-#endif
+atomic_t tage_allocated = ATOMIC_INIT(0);
-#define trace_get_tcd(FLAGS) ({ \
- struct trace_cpu_data *__ret; \
- int __cpu = get_cpu(); \
- local_irq_save(FLAGS); \
- __ret = &trace_data[__cpu].tcd; \
- __ret; \
-})
+static void put_pages_on_tcd_daemon_list(struct page_collection *pc,
+ struct trace_cpu_data *tcd);
-#define trace_put_tcd(TCD, FLAGS) do { \
- local_irq_restore(FLAGS); \
- put_cpu(); \
-} while (0)
-
-static void put_pages_on_daemon_list_on_cpu(void *info);
+static inline struct trace_page *tage_from_list(struct list_head *list)
+{
+ return list_entry(list, struct trace_page, linkage);
+}
-/* return a page that has 'len' bytes left at the end */
-static struct page *trace_get_page(struct trace_cpu_data *tcd,
- unsigned long len)
+static struct trace_page *tage_alloc(int gfp)
{
- struct page *page = NULL;
+ cfs_page_t *page;
+ struct trace_page *tage;
+
+ /*
+ * Don't spam console with allocation failures: they will be reported
+ * by upper layer anyway.
+ */
+ gfp |= CFS_ALLOC_NOWARN;
+ page = cfs_alloc_page(gfp);
+ if (page == NULL)
+ return NULL;
- if (len > PAGE_SIZE) {
- printk(KERN_ERR "cowardly refusing to write %lu bytes in a "
- "page\n", len);
+ tage = cfs_alloc(sizeof(*tage), gfp);
+ if (tage == NULL) {
+ cfs_free_page(page);
return NULL;
}
- if (!list_empty(&tcd->tcd_pages)) {
- page = list_entry(tcd->tcd_pages.prev, struct page,
- PAGE_LIST_ENTRY);
- if (page->index + len <= PAGE_SIZE)
- return page;
+ tage->page = page;
+ atomic_inc(&tage_allocated);
+ return tage;
+}
+
+static void tage_free(struct trace_page *tage)
+{
+ __LASSERT(tage != NULL);
+ __LASSERT(tage->page != NULL);
+
+ cfs_free_page(tage->page);
+ cfs_free(tage);
+ atomic_dec(&tage_allocated);
+}
+
+static void tage_to_tail(struct trace_page *tage, struct list_head *queue)
+{
+ __LASSERT(tage != NULL);
+ __LASSERT(queue != NULL);
+
+ list_move_tail(&tage->linkage, queue);
+}
+
+int trace_refill_stock(struct trace_cpu_data *tcd, int gfp,
+ struct list_head *stock)
+{
+ int i;
+
+ /*
+ * XXX nikita: do NOT call portals_debug_msg() (CDEBUG/ENTRY/EXIT)
+ * from here: this will lead to infinite recursion.
+ */
+
+ for (i = 0; i + tcd->tcd_cur_stock_pages < TCD_STOCK_PAGES ; ++ i) {
+ struct trace_page *tage;
+
+ tage = tage_alloc(gfp);
+ if (tage == NULL)
+ break;
+ list_add_tail(&tage->linkage, stock);
+ }
+ return i;
+}
+
+/* return a page that has 'len' bytes left at the end */
+static struct trace_page *trace_get_tage_try(struct trace_cpu_data *tcd,
+ unsigned long len)
+{
+ struct trace_page *tage;
+
+ if (tcd->tcd_cur_pages > 0) {
+ __LASSERT(!list_empty(&tcd->tcd_pages));
+ tage = tage_from_list(tcd->tcd_pages.prev);
+ if (tage->used + len <= CFS_PAGE_SIZE)
+ return tage;
}
if (tcd->tcd_cur_pages < tcd->tcd_max_pages) {
- page = alloc_page(GFP_ATOMIC);
- if (page == NULL) {
- /* the kernel should print a message for us. fall back
- * to using the last page in the ring buffer. */
- goto ring_buffer;
+ if (tcd->tcd_cur_stock_pages > 0) {
+ tage = tage_from_list(tcd->tcd_stock_pages.prev);
+ -- tcd->tcd_cur_stock_pages;
+ list_del_init(&tage->linkage);
+ } else {
+ tage = tage_alloc(CFS_ALLOC_ATOMIC);
+ if (tage == NULL) {
+ printk(KERN_WARNING
+ "failure to allocate a tage (%ld)\n",
+ tcd->tcd_cur_pages);
+ return NULL;
+ }
}
- page->index = 0;
- page->mapping = (void *)(long)smp_processor_id();
- list_add_tail(&PAGE_LIST(page), &tcd->tcd_pages);
+
+ tage->used = 0;
+ tage->cpu = smp_processor_id();
+ tage->type = tcd->tcd_type;
+ list_add_tail(&tage->linkage, &tcd->tcd_pages);
tcd->tcd_cur_pages++;
if (tcd->tcd_cur_pages > 8 && thread_running) {
struct tracefiled_ctl *tctl = &trace_tctl;
- wake_up(&tctl->tctl_waitq);
+ /*
+ * wake up tracefiled to process some pages.
+ */
+ cfs_waitq_signal(&tctl->tctl_waitq);
}
- return page;
+ return tage;
}
+ return NULL;
+}
- ring_buffer:
- if (thread_running) {
- int pgcount = tcd->tcd_cur_pages / 10;
- struct page_collection pc;
- struct list_head *pos, *tmp;
- printk(KERN_WARNING "debug daemon buffer overflowed; discarding"
- " 10%% of pages (%d)\n", pgcount + 1);
-
- INIT_LIST_HEAD(&pc.pc_pages);
- spin_lock_init(&pc.pc_lock);
+static void tcd_shrink(struct trace_cpu_data *tcd)
+{
+ int pgcount = tcd->tcd_cur_pages / 10;
+ struct page_collection pc;
+ struct trace_page *tage;
+ struct trace_page *tmp;
- list_for_each_safe(pos, tmp, &tcd->tcd_pages) {
- struct page *page;
+ /*
+ * XXX nikita: do NOT call portals_debug_msg() (CDEBUG/ENTRY/EXIT)
+ * from here: this will lead to infinite recursion.
+ */
- if (pgcount-- == 0)
- break;
+ printk(KERN_WARNING "debug daemon buffer overflowed; discarding"
+ " 10%% of pages (%d of %ld)\n", pgcount + 1, tcd->tcd_cur_pages);
- page = list_entry(pos, struct page, PAGE_LIST_ENTRY);
- list_del(&PAGE_LIST(page));
- list_add_tail(&PAGE_LIST(page), &pc.pc_pages);
- tcd->tcd_cur_pages--;
- }
- put_pages_on_daemon_list_on_cpu(&pc);
- }
- LASSERT(!list_empty(&tcd->tcd_pages));
+ CFS_INIT_LIST_HEAD(&pc.pc_pages);
+ spin_lock_init(&pc.pc_lock);
- page = list_entry(tcd->tcd_pages.next, struct page, PAGE_LIST_ENTRY);
- page->index = 0;
+ list_for_each_entry_safe(tage, tmp, &tcd->tcd_pages, linkage) {
+ if (pgcount-- == 0)
+ break;
- list_del(&PAGE_LIST(page));
- list_add_tail(&PAGE_LIST(page), &tcd->tcd_pages);
- return page;
+ list_move_tail(&tage->linkage, &pc.pc_pages);
+ tcd->tcd_cur_pages--;
+ }
+ put_pages_on_tcd_daemon_list(&pc, tcd);
}
-static void print_to_console(struct ptldebug_header *hdr, int mask, char *buf,
- int len, char *file, const char *fn)
+/* return a page that has 'len' bytes left at the end */
+static struct trace_page *trace_get_tage(struct trace_cpu_data *tcd,
+ unsigned long len)
{
- char *prefix = NULL, *ptype = NULL;
+ struct trace_page *tage;
+
+ /*
+ * XXX nikita: do NOT call portals_debug_msg() (CDEBUG/ENTRY/EXIT)
+ * from here: this will lead to infinite recursion.
+ */
- if ((mask & D_EMERG) != 0) {
- prefix = "LustreError";
- ptype = KERN_EMERG;
- } else if ((mask & D_ERROR) != 0) {
- prefix = "LustreError";
- ptype = KERN_ERR;
- } else if ((mask & D_WARNING) != 0) {
- prefix = "Lustre";
- ptype = KERN_WARNING;
- } else if (portal_printk) {
- prefix = "Lustre";
- ptype = KERN_INFO;
+ if (len > CFS_PAGE_SIZE) {
+ printk(KERN_ERR
+ "cowardly refusing to write %lu bytes in a page\n", len);
+ return NULL;
}
- printk("%s%s: %d:%d:(%s:%d:%s()) %.*s", ptype, prefix, hdr->ph_pid,
- hdr->ph_extern_pid, file, hdr->ph_line_num, fn, len, buf);
+ tage = trace_get_tage_try(tcd, len);
+ if (tage != NULL)
+ return tage;
+ if (thread_running)
+ tcd_shrink(tcd);
+ if (tcd->tcd_cur_pages > 0) {
+ tage = tage_from_list(tcd->tcd_pages.next);
+ tage->used = 0;
+ tage_to_tail(tage, &tcd->tcd_pages);
+ }
+ return tage;
}
-void portals_debug_msg(int subsys, int mask, char *file, const char *fn,
- const int line, unsigned long stack, char *format, ...)
+int libcfs_debug_vmsg2(cfs_debug_limit_state_t *cdls, int subsys, int mask,
+ const char *file, const char *fn, const int line,
+ const char *format1, va_list args,
+ const char *format2, ...)
{
- struct trace_cpu_data *tcd;
- struct ptldebug_header header;
- struct page *page;
- char *debug_buf = format;
- int known_size, needed = 85 /* average message length */, max_nob;
- va_list ap;
- unsigned long flags;
- struct timeval tv;
-
- if (*(format + strlen(format) - 1) != '\n')
+ struct trace_cpu_data *tcd = NULL;
+ struct ptldebug_header header;
+ struct trace_page *tage;
+ /* string_buf is used only if tcd != NULL, and is always set then */
+ char *string_buf = NULL;
+ char *debug_buf;
+ int known_size;
+ int needed = 85; /* average message length */
+ int max_nob;
+ va_list ap;
+ int depth;
+ int i;
+ int remain;
+
+ if (strchr(file, '/'))
+ file = strrchr(file, '/') + 1;
+
+
+ set_ptldebug_header(&header, subsys, mask, line, CDEBUG_STACK());
+
+ tcd = trace_get_tcd();
+ if (tcd == NULL) /* arch may not log in IRQ context */
+ goto console;
+
+ if (tcd->tcd_shutting_down) {
+ trace_put_tcd(tcd);
+ tcd = NULL;
+ goto console;
+ }
+
+ depth = __current_nesting_level();
+ known_size = strlen(file) + 1 + depth;
+ if (fn)
+ known_size += strlen(fn) + 1;
+
+ if (libcfs_debug_binary)
+ known_size += sizeof(header);
+
+ /*/
+ * '2' used because vsnprintf return real size required for output
+ * _without_ terminating NULL.
+ * if needed is to small for this format.
+ */
+ for (i=0;i<2;i++) {
+ tage = trace_get_tage(tcd, needed + known_size + 1);
+ if (tage == NULL) {
+ if (needed + known_size > CFS_PAGE_SIZE)
+ mask |= D_ERROR;
+
+ trace_put_tcd(tcd);
+ tcd = NULL;
+ goto console;
+ }
+
+ string_buf = (char *)cfs_page_address(tage->page)+tage->used+known_size;
+
+ max_nob = CFS_PAGE_SIZE - tage->used - known_size;
+ if (max_nob <= 0) {
+ printk(KERN_EMERG "negative max_nob: %i\n", max_nob);
+ mask |= D_ERROR;
+ trace_put_tcd(tcd);
+ tcd = NULL;
+ goto console;
+ }
+
+ needed = 0;
+ if (format1) {
+ va_copy(ap, args);
+ needed = vsnprintf(string_buf, max_nob, format1, ap);
+ va_end(ap);
+ }
+
+
+ if (format2) {
+ remain = max_nob - needed;
+ if (remain < 0)
+ remain = 0;
+
+ va_start(ap, format2);
+ needed += vsnprintf(string_buf+needed, remain, format2, ap);
+ va_end(ap);
+ }
+
+ if (needed < max_nob) /* well. printing ok.. */
+ break;
+ }
+
+ if (*(string_buf+needed-1) != '\n')
printk(KERN_INFO "format at %s:%d:%s doesn't end in newline\n",
file, line, fn);
+
+ header.ph_len = known_size + needed;
+ debug_buf = (char *)cfs_page_address(tage->page) + tage->used;
- tcd = trace_get_tcd(flags);
- if (tcd->tcd_shutting_down)
- goto out;
+ if (libcfs_debug_binary) {
+ memcpy(debug_buf, &header, sizeof(header));
+ tage->used += sizeof(header);
+ debug_buf += sizeof(header);
+ }
- do_gettimeofday(&tv);
-
- header.ph_subsys = subsys;
- header.ph_mask = mask;
- header.ph_cpu_id = smp_processor_id();
- header.ph_sec = (__u32)tv.tv_sec;
- header.ph_usec = tv.tv_usec;
- header.ph_stack = stack;
- header.ph_pid = current->pid;
- header.ph_line_num = line;
-
-#if defined(__arch_um__) && (LINUX_VERSION_CODE < KERNEL_VERSION(2,4,20))
- header.ph_extern_pid = current->thread.extern_pid;
-#elif defined(__arch_um__) && (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
- header.ph_extern_pid = current->thread.mode.tt.extern_pid;
-#else
- header.ph_extern_pid = 0;
-#endif
+ /* indent message according to the nesting level */
+ while (depth-- > 0) {
+ *(debug_buf++) = '.';
+ ++ tage->used;
+ }
- known_size = sizeof(header) + strlen(file) + strlen(fn) + 2; // nulls
+ strcpy(debug_buf, file);
+ tage->used += strlen(file) + 1;
+ debug_buf += strlen(file) + 1;
- retry:
- page = trace_get_page(tcd, needed + known_size);
- if (page == NULL) {
- debug_buf = format;
- if (needed + known_size > PAGE_SIZE)
- mask |= D_ERROR;
- needed = strlen(format);
- goto out;
+ if (fn) {
+ strcpy(debug_buf, fn);
+ tage->used += strlen(fn) + 1;
+ debug_buf += strlen(fn) + 1;
}
- debug_buf = page_address(page) + page->index + known_size;
+ __LASSERT(debug_buf == string_buf);
- max_nob = PAGE_SIZE - page->index - known_size;
- LASSERT(max_nob > 0);
- va_start(ap, format);
- needed = vsnprintf(debug_buf, max_nob, format, ap);
- va_end(ap);
+ tage->used += needed;
+ __LASSERT (tage->used <= CFS_PAGE_SIZE);
- if (needed > max_nob) /* overflow. oh poop. */
- goto retry;
+console:
+ if ((mask & libcfs_printk) == 0) {
+ /* no console output requested */
+ if (tcd != NULL)
+ trace_put_tcd(tcd);
+ return 1;
+ }
- header.ph_len = known_size + needed;
- debug_buf = page_address(page) + page->index;
+ if (cdls != NULL) {
+ if (libcfs_console_ratelimit &&
+ cdls->cdls_next != 0 && /* not first time ever */
+ !cfs_time_after(cfs_time_current(), cdls->cdls_next)) {
+ /* skipping a console message */
+ cdls->cdls_count++;
+ if (tcd != NULL)
+ trace_put_tcd(tcd);
+ return 1;
+ }
- memcpy(debug_buf, &header, sizeof(header));
- page->index += sizeof(header);
- debug_buf += sizeof(header);
+ if (cfs_time_after(cfs_time_current(), cdls->cdls_next +
+ libcfs_console_max_delay
+ + cfs_time_seconds(10))) {
+ /* last timeout was a long time ago */
+ cdls->cdls_delay /= libcfs_console_backoff * 4;
+ } else {
+ cdls->cdls_delay *= libcfs_console_backoff;
+
+ if (cdls->cdls_delay < libcfs_console_min_delay)
+ cdls->cdls_delay = libcfs_console_min_delay;
+ else if (cdls->cdls_delay > libcfs_console_max_delay)
+ cdls->cdls_delay = libcfs_console_max_delay;
+ }
- strcpy(debug_buf, file);
- page->index += strlen(file) + 1;
- debug_buf += strlen(file) + 1;
+ /* ensure cdls_next is never zero after it's been seen */
+ cdls->cdls_next = (cfs_time_current() + cdls->cdls_delay) | 1;
+ }
+
+ if (tcd != NULL) {
+ print_to_console(&header, mask, string_buf, needed, file, fn);
+ trace_put_tcd(tcd);
+ } else {
+ string_buf = trace_get_console_buffer();
+
+ needed = 0;
+ if (format1 != NULL) {
+ va_copy(ap, args);
+ needed = vsnprintf(string_buf, TRACE_CONSOLE_BUFFER_SIZE, format1, ap);
+ va_end(ap);
+ }
+ if (format2 != NULL) {
+ remain = TRACE_CONSOLE_BUFFER_SIZE - needed;
+ if (remain > 0) {
+ va_start(ap, format2);
+ needed += vsnprintf(string_buf+needed, remain, format2, ap);
+ va_end(ap);
+ }
+ }
+ print_to_console(&header, mask,
+ string_buf, needed, file, fn);
- strcpy(debug_buf, fn);
- page->index += strlen(fn) + 1;
- debug_buf += strlen(fn) + 1;
+ trace_put_console_buffer(string_buf);
+ }
- page->index += needed;
- if (page->index > PAGE_SIZE)
- printk(KERN_EMERG "page->index == %lu in portals_debug_msg\n",
- page->index);
+ if (cdls != NULL && cdls->cdls_count != 0) {
+ string_buf = trace_get_console_buffer();
- out:
- if ((mask & (D_EMERG | D_ERROR | D_WARNING)) || portal_printk)
- print_to_console(&header, mask, debug_buf, needed, file, fn);
+ needed = snprintf(string_buf, TRACE_CONSOLE_BUFFER_SIZE,
+ "Skipped %d previous similar message%s\n",
+ cdls->cdls_count, (cdls->cdls_count > 1) ? "s" : "");
+
+ print_to_console(&header, mask,
+ string_buf, needed, file, fn);
- trace_put_tcd(tcd, flags);
+ trace_put_console_buffer(string_buf);
+ cdls->cdls_count = 0;
+ }
+
+ return 0;
+}
+EXPORT_SYMBOL(libcfs_debug_vmsg2);
+
+void
+libcfs_assertion_failed(const char *expr, const char *file,
+ const char *func, const int line)
+{
+ libcfs_debug_msg(NULL, 0, D_EMERG, file, func, line,
+ "ASSERTION(%s) failed\n", expr);
+ LBUG();
+}
+EXPORT_SYMBOL(libcfs_assertion_failed);
+
+void
+trace_assertion_failed(const char *str,
+ const char *fn, const char *file, int line)
+{
+ struct ptldebug_header hdr;
+
+ libcfs_panic_in_progress = 1;
+ libcfs_catastrophe = 1;
+ mb();
+
+ set_ptldebug_header(&hdr, DEBUG_SUBSYSTEM, D_EMERG, line,
+ CDEBUG_STACK());
+
+ print_to_console(&hdr, D_EMERG, str, strlen(str), file, fn);
+
+ LIBCFS_PANIC("Lustre debug assertion failure\n");
+
+ /* not reached */
+}
+
+static void
+panic_collect_pages(struct page_collection *pc)
+{
+ /* Do the collect_pages job on a single CPU: assumes that all other
+ * CPUs have been stopped during a panic. If this isn't true for some
+ * arch, this will have to be implemented separately in each arch. */
+ int i;
+ int j;
+ struct trace_cpu_data *tcd;
+
+ CFS_INIT_LIST_HEAD(&pc->pc_pages);
+
+ tcd_for_each(tcd, i, j) {
+ list_splice_init(&tcd->tcd_pages, &pc->pc_pages);
+ tcd->tcd_cur_pages = 0;
+
+ if (pc->pc_want_daemon_pages) {
+ list_splice_init(&tcd->tcd_daemon_pages, &pc->pc_pages);
+ tcd->tcd_cur_daemon_pages = 0;
+ }
+ }
}
-EXPORT_SYMBOL(portals_debug_msg);
static void collect_pages_on_cpu(void *info)
{
struct trace_cpu_data *tcd;
- unsigned long flags;
struct page_collection *pc = info;
-
- tcd = trace_get_tcd(flags);
+ int i;
spin_lock(&pc->pc_lock);
- list_splice(&tcd->tcd_pages, &pc->pc_pages);
- INIT_LIST_HEAD(&tcd->tcd_pages);
- tcd->tcd_cur_pages = 0;
- if (pc->pc_want_daemon_pages) {
- list_splice(&tcd->tcd_daemon_pages, &pc->pc_pages);
- INIT_LIST_HEAD(&tcd->tcd_daemon_pages);
- tcd->tcd_cur_daemon_pages = 0;
+ tcd_for_each_type_lock(tcd, i) {
+ list_splice_init(&tcd->tcd_pages, &pc->pc_pages);
+ tcd->tcd_cur_pages = 0;
+ if (pc->pc_want_daemon_pages) {
+ list_splice_init(&tcd->tcd_daemon_pages, &pc->pc_pages);
+ tcd->tcd_cur_daemon_pages = 0;
+ }
}
spin_unlock(&pc->pc_lock);
-
- trace_put_tcd(tcd, flags);
}
static void collect_pages(struct page_collection *pc)
{
- /* needs to be fixed up for preempt */
- INIT_LIST_HEAD(&pc->pc_pages);
- collect_pages_on_cpu(pc);
- smp_call_function(collect_pages_on_cpu, pc, 0, 1);
+ CFS_INIT_LIST_HEAD(&pc->pc_pages);
+
+ if (libcfs_panic_in_progress)
+ panic_collect_pages(pc);
+ else
+ trace_call_on_all_cpus(collect_pages_on_cpu, pc);
}
static void put_pages_back_on_cpu(void *info)
{
struct page_collection *pc = info;
struct trace_cpu_data *tcd;
- struct list_head *pos, *tmp, *cur_head;
- unsigned long flags;
-
- tcd = trace_get_tcd(flags);
-
- cur_head = tcd->tcd_pages.next;
+ struct list_head *cur_head;
+ struct trace_page *tage;
+ struct trace_page *tmp;
+ int i;
spin_lock(&pc->pc_lock);
- list_for_each_safe(pos, tmp, &pc->pc_pages) {
- struct page *page;
+ tcd_for_each_type_lock(tcd, i) {
+ cur_head = tcd->tcd_pages.next;
- page = list_entry(pos, struct page, PAGE_LIST_ENTRY);
- LASSERT(page->index <= PAGE_SIZE);
- LASSERT(page_count(page) > 0);
+ list_for_each_entry_safe(tage, tmp, &pc->pc_pages, linkage) {
- if ((unsigned long)page->mapping != smp_processor_id())
- continue;
+ __LASSERT_TAGE_INVARIANT(tage);
- list_del(&PAGE_LIST(page));
- list_add_tail(&PAGE_LIST(page), cur_head);
- tcd->tcd_cur_pages++;
+ if (tage->cpu != smp_processor_id() || tage->type != i)
+ continue;
+
+ tage_to_tail(tage, cur_head);
+ tcd->tcd_cur_pages++;
+ }
}
spin_unlock(&pc->pc_lock);
-
- trace_put_tcd(tcd, flags);
}
static void put_pages_back(struct page_collection *pc)
{
- /* needs to be fixed up for preempt */
- put_pages_back_on_cpu(pc);
- smp_call_function(put_pages_back_on_cpu, pc, 0, 1);
+ if (!libcfs_panic_in_progress)
+ trace_call_on_all_cpus(put_pages_back_on_cpu, pc);
}
/* Add pages to a per-cpu debug daemon ringbuffer. This buffer makes sure that
* we have a good amount of data at all times for dumping during an LBUG, even
* if we have been steadily writing (and otherwise discarding) pages via the
* debug daemon. */
-static void put_pages_on_daemon_list_on_cpu(void *info)
+static void put_pages_on_tcd_daemon_list(struct page_collection *pc,
+ struct trace_cpu_data *tcd)
{
- struct page_collection *pc = info;
- struct trace_cpu_data *tcd;
- struct list_head *pos, *tmp;
- unsigned long flags;
-
- tcd = trace_get_tcd(flags);
+ struct trace_page *tage;
+ struct trace_page *tmp;
spin_lock(&pc->pc_lock);
- list_for_each_safe(pos, tmp, &pc->pc_pages) {
- struct page *page;
+ list_for_each_entry_safe(tage, tmp, &pc->pc_pages, linkage) {
- page = list_entry(pos, struct page, PAGE_LIST_ENTRY);
- LASSERT(page->index <= PAGE_SIZE);
- LASSERT(page_count(page) > 0);
- if ((unsigned long)page->mapping != smp_processor_id())
+ __LASSERT_TAGE_INVARIANT(tage);
+
+ if (tage->cpu != smp_processor_id() ||
+ tage->type != tcd->tcd_type)
continue;
- list_del(&PAGE_LIST(page));
- list_add_tail(&PAGE_LIST(page), &tcd->tcd_daemon_pages);
+ tage_to_tail(tage, &tcd->tcd_daemon_pages);
tcd->tcd_cur_daemon_pages++;
if (tcd->tcd_cur_daemon_pages > tcd->tcd_max_pages) {
- LASSERT(!list_empty(&tcd->tcd_daemon_pages));
- page = list_entry(tcd->tcd_daemon_pages.next,
- struct page, PAGE_LIST_ENTRY);
+ struct trace_page *victim;
+
+ __LASSERT(!list_empty(&tcd->tcd_daemon_pages));
+ victim = tage_from_list(tcd->tcd_daemon_pages.next);
- LASSERT(page->index <= PAGE_SIZE);
- LASSERT(page_count(page) > 0);
+ __LASSERT_TAGE_INVARIANT(victim);
- page->index = 0;
- list_del(&PAGE_LIST(page));
- page->mapping = NULL;
- __free_page(page);
+ list_del(&victim->linkage);
+ tage_free(victim);
tcd->tcd_cur_daemon_pages--;
}
}
spin_unlock(&pc->pc_lock);
+}
- trace_put_tcd(tcd, flags);
+static void put_pages_on_daemon_list_on_cpu(void *info)
+{
+ struct trace_cpu_data *tcd;
+ int i;
+
+ tcd_for_each_type_lock(tcd, i)
+ put_pages_on_tcd_daemon_list(info, tcd);
}
static void put_pages_on_daemon_list(struct page_collection *pc)
{
- put_pages_on_daemon_list_on_cpu(pc);
- smp_call_function(put_pages_on_daemon_list_on_cpu, pc, 0, 1);
+ trace_call_on_all_cpus(put_pages_on_daemon_list_on_cpu, pc);
}
void trace_debug_print(void)
{
struct page_collection pc;
- struct list_head *pos, *tmp;
+ struct trace_page *tage;
+ struct trace_page *tmp;
spin_lock_init(&pc.pc_lock);
+ pc.pc_want_daemon_pages = 1;
collect_pages(&pc);
- list_for_each_safe(pos, tmp, &pc.pc_pages) {
- struct page *page;
+ list_for_each_entry_safe(tage, tmp, &pc.pc_pages, linkage) {
char *p, *file, *fn;
+ cfs_page_t *page;
- page = list_entry(pos, struct page, PAGE_LIST_ENTRY);
- LASSERT(page->index <= PAGE_SIZE);
- LASSERT(page_count(page) > 0);
+ __LASSERT_TAGE_INVARIANT(tage);
- p = page_address(page);
- while (p < ((char *)page_address(page) + PAGE_SIZE)) {
+ page = tage->page;
+ p = cfs_page_address(page);
+ while (p < ((char *)cfs_page_address(page) + tage->used)) {
struct ptldebug_header *hdr;
int len;
hdr = (void *)p;
len = hdr->ph_len - (p - (char *)hdr);
print_to_console(hdr, D_EMERG, p, len, file, fn);
+
+ p += len;
}
- list_del(&PAGE_LIST(page));
- page->mapping = NULL;
- __free_page(page);
+ list_del(&tage->linkage);
+ tage_free(tage);
}
}
int tracefile_dump_all_pages(char *filename)
{
struct page_collection pc;
- struct file *filp;
- struct list_head *pos, *tmp;
- mm_segment_t oldfs;
+ cfs_file_t *filp;
+ struct trace_page *tage;
+ struct trace_page *tmp;
int rc;
- down_write(&tracefile_sem);
+ CFS_DECL_MMSPACE;
- filp = filp_open(filename, O_CREAT|O_EXCL|O_WRONLY|O_LARGEFILE, 0600);
- if (IS_ERR(filp)) {
- rc = PTR_ERR(filp);
- printk(KERN_ERR "LustreError: can't open %s for dump: rc %d\n",
- filename, rc);
+ tracefile_write_lock();
+
+ filp = cfs_filp_open(filename,
+ O_CREAT|O_EXCL|O_WRONLY|O_LARGEFILE, 0600, &rc);
+ if (!filp) {
+ if (rc != -EEXIST)
+ printk(KERN_ERR "LustreError: can't open %s for dump: rc %d\n",
+ filename, rc);
goto out;
}
/* ok, for now, just write the pages. in the future we'll be building
* iobufs with the pages and calling generic_direct_IO */
- oldfs = get_fs();
- set_fs(get_ds());
- list_for_each_safe(pos, tmp, &pc.pc_pages) {
- struct page *page;
-
- page = list_entry(pos, struct page, PAGE_LIST_ENTRY);
- LASSERT(page->index <= PAGE_SIZE);
- LASSERT(page_count(page) > 0);
-
- rc = filp->f_op->write(filp, page_address(page), page->index,
- &filp->f_pos);
- if (rc != page->index) {
- printk(KERN_WARNING "wanted to write %lu but wrote "
- "%d\n", page->index, rc);
+ CFS_MMSPACE_OPEN;
+ list_for_each_entry_safe(tage, tmp, &pc.pc_pages, linkage) {
+
+ __LASSERT_TAGE_INVARIANT(tage);
+
+ rc = cfs_filp_write(filp, cfs_page_address(tage->page),
+ tage->used, cfs_filp_poff(filp));
+ if (rc != (int)tage->used) {
+ printk(KERN_WARNING "wanted to write %u but wrote "
+ "%d\n", tage->used, rc);
put_pages_back(&pc);
+ __LASSERT(list_empty(&pc.pc_pages));
break;
}
- list_del(&PAGE_LIST(page));
- page->mapping = NULL;
- __free_page(page);
+ list_del(&tage->linkage);
+ tage_free(tage);
}
- set_fs(oldfs);
- rc = filp->f_op->fsync(filp, filp->f_dentry, 1);
+ CFS_MMSPACE_CLOSE;
+ rc = cfs_filp_fsync(filp);
if (rc)
printk(KERN_ERR "sync returns %d\n", rc);
close:
- filp_close(filp, 0);
+ cfs_filp_close(filp);
out:
- up_write(&tracefile_sem);
+ tracefile_write_unlock();
return rc;
}
void trace_flush_pages(void)
{
struct page_collection pc;
- struct list_head *pos, *tmp;
+ struct trace_page *tage;
+ struct trace_page *tmp;
spin_lock_init(&pc.pc_lock);
+ pc.pc_want_daemon_pages = 1;
collect_pages(&pc);
- list_for_each_safe(pos, tmp, &pc.pc_pages) {
- struct page *page;
+ list_for_each_entry_safe(tage, tmp, &pc.pc_pages, linkage) {
- page = list_entry(pos, struct page, PAGE_LIST_ENTRY);
- LASSERT(page->index <= PAGE_SIZE);
- LASSERT(page_count(page) > 0);
+ __LASSERT_TAGE_INVARIANT(tage);
- list_del(&PAGE_LIST(page));
- page->mapping = NULL;
- __free_page(page);
+ list_del(&tage->linkage);
+ tage_free(tage);
}
}
-int trace_dk(struct file *file, const char *buffer, unsigned long count,
- void *data)
+int trace_copyin_string(char *knl_buffer, int knl_buffer_nob,
+ const char *usr_buffer, int usr_buffer_nob)
{
- char *name;
- unsigned long off;
- int rc;
+ int nob;
+
+ if (usr_buffer_nob > knl_buffer_nob)
+ return -EOVERFLOW;
+
+ if (copy_from_user((void *)knl_buffer,
+ (void *)usr_buffer, usr_buffer_nob))
+ return -EFAULT;
+
+ nob = strnlen(knl_buffer, usr_buffer_nob);
+ while (nob-- >= 0) /* strip trailing whitespace */
+ if (!isspace(knl_buffer[nob]))
+ break;
+
+ if (nob < 0) /* empty string */
+ return -EINVAL;
+
+ if (nob == knl_buffer_nob) /* no space to terminate */
+ return -EOVERFLOW;
+
+ knl_buffer[nob + 1] = 0; /* terminate */
+ return 0;
+}
+
+int trace_copyout_string(char *usr_buffer, int usr_buffer_nob,
+ const char *knl_buffer, char *append)
+{
+ /* NB if 'append' != NULL, it's a single character to append to the
+ * copied out string - usually "\n", for /proc entries and "" (i.e. a
+ * terminating zero byte) for sysctl entries */
+ int nob = strlen(knl_buffer);
+
+ if (nob > usr_buffer_nob)
+ nob = usr_buffer_nob;
+
+ if (copy_to_user(usr_buffer, knl_buffer, nob))
+ return -EFAULT;
+
+ if (append != NULL && nob < usr_buffer_nob) {
+ if (copy_to_user(usr_buffer + nob, append, 1))
+ return -EFAULT;
+
+ nob++;
+ }
- name = kmalloc(count + 1, GFP_KERNEL);
- if (name == NULL)
+ return nob;
+}
+
+int trace_allocate_string_buffer(char **str, int nob)
+{
+ if (nob > 2 * CFS_PAGE_SIZE) /* string must be "sensible" */
+ return -EINVAL;
+
+ *str = cfs_alloc(nob, CFS_ALLOC_STD | CFS_ALLOC_ZERO);
+ if (*str == NULL)
return -ENOMEM;
- if (copy_from_user(name, buffer, count)) {
- rc = -EFAULT;
+ return 0;
+}
+
+void trace_free_string_buffer(char *str, int nob)
+{
+ cfs_free(str);
+}
+
+int trace_dump_debug_buffer_usrstr(void *usr_str, int usr_str_nob)
+{
+ char *str;
+ int rc;
+
+ rc = trace_allocate_string_buffer(&str, usr_str_nob + 1);
+ if (rc != 0)
+ return rc;
+
+ rc = trace_copyin_string(str, usr_str_nob + 1,
+ usr_str, usr_str_nob);
+ if (rc != 0)
goto out;
- }
- if (name[0] != '/') {
+#if !defined(__WINNT__)
+ if (str[0] != '/') {
rc = -EINVAL;
goto out;
}
+#endif
+ rc = tracefile_dump_all_pages(str);
+out:
+ trace_free_string_buffer(str, usr_str_nob + 1);
+ return rc;
+}
- /* be nice and strip out trailing '\n' */
- for (off = count ; off > 2 && isspace(name[off - 1]); off--)
- ;
+int trace_daemon_command(char *str)
+{
+ int rc = 0;
+
+ tracefile_write_lock();
+
+ if (strcmp(str, "stop") == 0) {
+ trace_stop_thread();
+ memset(tracefile, 0, sizeof(tracefile));
+
+ } else if (strncmp(str, "size=", 5) == 0) {
+ tracefile_size = simple_strtoul(str + 5, NULL, 0);
+ if (tracefile_size < 10 || tracefile_size > 20480)
+ tracefile_size = TRACEFILE_SIZE;
+ else
+ tracefile_size <<= 20;
+
+ } else if (strlen(str) >= sizeof(tracefile)) {
+ rc = -ENAMETOOLONG;
+#ifndef __WINNT__
+ } else if (str[0] != '/') {
+ rc = -EINVAL;
+#endif
+ } else {
+ strcpy(tracefile, str);
- name[off] = '\0';
- rc = tracefile_dump_all_pages(name);
-out:
- if (name)
- kfree(name);
- return count;
+ printk(KERN_INFO "Lustre: debug daemon will attempt to start writing "
+ "to %s (%lukB max)\n", tracefile,
+ (long)(tracefile_size >> 10));
+
+ trace_start_thread();
+ }
+
+ tracefile_write_unlock();
+ return rc;
+}
+
+int trace_daemon_command_usrstr(void *usr_str, int usr_str_nob)
+{
+ char *str;
+ int rc;
+
+ rc = trace_allocate_string_buffer(&str, usr_str_nob + 1);
+ if (rc != 0)
+ return rc;
+
+ rc = trace_copyin_string(str, usr_str_nob + 1,
+ usr_str, usr_str_nob);
+ if (rc == 0)
+ rc = trace_daemon_command(str);
+
+ trace_free_string_buffer(str, usr_str_nob + 1);
+ return rc;
+}
+
+int trace_set_debug_mb(int mb)
+{
+ int i;
+ int j;
+ int pages;
+ int limit = trace_max_debug_mb();
+ struct trace_cpu_data *tcd;
+
+ if (mb < num_possible_cpus())
+ return -EINVAL;
+
+ if (mb > limit) {
+ printk(KERN_ERR "Lustre: Refusing to set debug buffer size to "
+ "%dMB - limit is %d\n", mb, limit);
+ return -EINVAL;
+ }
+
+ mb /= num_possible_cpus();
+ pages = mb << (20 - CFS_PAGE_SHIFT);
+
+ tracefile_write_lock();
+
+ tcd_for_each(tcd, i, j)
+ tcd->tcd_max_pages = (pages * tcd->tcd_pages_factor) / 100;
+
+ tracefile_write_unlock();
+
+ return 0;
+}
+
+int trace_set_debug_mb_usrstr(void *usr_str, int usr_str_nob)
+{
+ char str[32];
+ int rc;
+
+ rc = trace_copyin_string(str, sizeof(str), usr_str, usr_str_nob);
+ if (rc < 0)
+ return rc;
+
+ return trace_set_debug_mb(simple_strtoul(str, NULL, 0));
+}
+
+int trace_get_debug_mb(void)
+{
+ int i;
+ int j;
+ struct trace_cpu_data *tcd;
+ int total_pages = 0;
+
+ tracefile_read_lock();
+
+ tcd_for_each(tcd, i, j)
+ total_pages += tcd->tcd_max_pages;
+
+ tracefile_read_unlock();
+
+ return (total_pages >> (20 - CFS_PAGE_SHIFT)) + 1;
}
-EXPORT_SYMBOL(trace_dk);
static int tracefiled(void *arg)
{
struct page_collection pc;
struct tracefiled_ctl *tctl = arg;
- struct list_head *pos, *tmp;
+ struct trace_page *tage;
+ struct trace_page *tmp;
struct ptldebug_header *hdr;
- struct file *filp;
- struct page *page;
- mm_segment_t oldfs;
+ cfs_file_t *filp;
int rc;
+ CFS_DECL_MMSPACE;
+
/* we're started late enough that we pick up init's fs context */
/* this is so broken in uml? what on earth is going on? */
- kportal_daemonize("ktracefiled");
- reparent_to_init();
+ cfs_daemonize("ktracefiled");
spin_lock_init(&pc.pc_lock);
complete(&tctl->tctl_start);
while (1) {
- wait_queue_t __wait;
+ cfs_waitlink_t __wait;
- init_waitqueue_entry(&__wait, current);
- add_wait_queue(&tctl->tctl_waitq, &__wait);
+ cfs_waitlink_init(&__wait);
+ cfs_waitq_add(&tctl->tctl_waitq, &__wait);
set_current_state(TASK_INTERRUPTIBLE);
- schedule_timeout(HZ);
- remove_wait_queue(&tctl->tctl_waitq, &__wait);
+ cfs_waitq_timedwait(&__wait, CFS_TASK_INTERRUPTIBLE,
+ cfs_time_seconds(1));
+ cfs_waitq_del(&tctl->tctl_waitq, &__wait);
if (atomic_read(&tctl->tctl_shutdown))
break;
continue;
filp = NULL;
- down_read(&tracefile_sem);
- if (tracefile != NULL) {
- filp = filp_open(tracefile, O_CREAT|O_RDWR|O_LARGEFILE,
- 0600);
- if (IS_ERR(filp)) {
- printk("couldn't open %s: %ld\n", tracefile,
- PTR_ERR(filp));
- filp = NULL;
- }
+ tracefile_read_lock();
+ if (tracefile[0] != 0) {
+ filp = cfs_filp_open(tracefile,
+ O_CREAT | O_RDWR | O_LARGEFILE,
+ 0600, &rc);
+ if (!(filp))
+ printk(KERN_WARNING "couldn't open %s: %d\n",
+ tracefile, rc);
}
- up_read(&tracefile_sem);
+ tracefile_read_unlock();
if (filp == NULL) {
put_pages_on_daemon_list(&pc);
+ __LASSERT(list_empty(&pc.pc_pages));
continue;
}
- oldfs = get_fs();
- set_fs(get_ds());
+ CFS_MMSPACE_OPEN;
/* mark the first header, so we can sort in chunks */
- page = list_entry(pc.pc_pages.next, struct page,
- PAGE_LIST_ENTRY);
- LASSERT(page->index <= PAGE_SIZE);
- LASSERT(page_count(page) > 0);
+ tage = tage_from_list(pc.pc_pages.next);
+ __LASSERT_TAGE_INVARIANT(tage);
- hdr = page_address(page);
+ hdr = cfs_page_address(tage->page);
hdr->ph_flags |= PH_FLAG_FIRST_RECORD;
- list_for_each_safe(pos, tmp, &pc.pc_pages) {
+ list_for_each_entry_safe(tage, tmp, &pc.pc_pages, linkage) {
static loff_t f_pos;
- page = list_entry(pos, struct page, PAGE_LIST_ENTRY);
- LASSERT(page->index <= PAGE_SIZE);
- LASSERT(page_count(page) > 0);
- if (f_pos >= tracefile_size)
+ __LASSERT_TAGE_INVARIANT(tage);
+
+ if (f_pos >= (off_t)tracefile_size)
f_pos = 0;
- else if (f_pos > filp->f_dentry->d_inode->i_size)
- f_pos = filp->f_dentry->d_inode->i_size;
-
- rc = filp->f_op->write(filp, page_address(page),
- page->index, &f_pos);
- if (rc != page->index) {
- printk(KERN_WARNING "wanted to write %lu but "
- "wrote %d\n", page->index, rc);
+ else if (f_pos > cfs_filp_size(filp))
+ f_pos = cfs_filp_size(filp);
+
+ rc = cfs_filp_write(filp, cfs_page_address(tage->page),
+ tage->used, &f_pos);
+ if (rc != (int)tage->used) {
+ printk(KERN_WARNING "wanted to write %u but "
+ "wrote %d\n", tage->used, rc);
put_pages_back(&pc);
+ __LASSERT(list_empty(&pc.pc_pages));
}
}
- set_fs(oldfs);
- filp_close(filp, 0);
+ CFS_MMSPACE_CLOSE;
+ cfs_filp_close(filp);
put_pages_on_daemon_list(&pc);
+ __LASSERT(list_empty(&pc.pc_pages));
}
complete(&tctl->tctl_stop);
return 0;
struct tracefiled_ctl *tctl = &trace_tctl;
int rc = 0;
- down(&trace_thread_sem);
+ mutex_down(&trace_thread_sem);
if (thread_running)
goto out;
init_completion(&tctl->tctl_start);
init_completion(&tctl->tctl_stop);
- init_waitqueue_head(&tctl->tctl_waitq);
+ cfs_waitq_init(&tctl->tctl_waitq);
atomic_set(&tctl->tctl_shutdown, 0);
- if (kernel_thread(tracefiled, tctl, 0) < 0) {
+ if (cfs_kernel_thread(tracefiled, tctl, 0) < 0) {
rc = -ECHILD;
goto out;
}
wait_for_completion(&tctl->tctl_start);
thread_running = 1;
out:
- up(&trace_thread_sem);
+ mutex_up(&trace_thread_sem);
return rc;
}
{
struct tracefiled_ctl *tctl = &trace_tctl;
- down(&trace_thread_sem);
+ mutex_down(&trace_thread_sem);
if (thread_running) {
- printk(KERN_INFO "Shutting down debug daemon thread...\n");
+ printk(KERN_INFO "Lustre: shutting down debug daemon thread...\n");
atomic_set(&tctl->tctl_shutdown, 1);
wait_for_completion(&tctl->tctl_stop);
thread_running = 0;
}
- up(&trace_thread_sem);
-}
-
-int trace_write_daemon_file(struct file *file, const char *buffer,
- unsigned long count, void *data)
-{
- char *name;
- unsigned long off;
- int rc;
-
- name = kmalloc(count + 1, GFP_KERNEL);
- if (name == NULL)
- return -ENOMEM;
-
- if (copy_from_user(name, buffer, count)) {
- rc = -EFAULT;
- goto out;
- }
-
- /* be nice and strip out trailing '\n' */
- for (off = count ; off > 2 && isspace(name[off - 1]); off--)
- ;
-
- name[off] = '\0';
-
- down_write(&tracefile_sem);
- if (strcmp(name, "stop") == 0) {
- tracefile = NULL;
- trace_stop_thread();
- goto out_sem;
- } else if (strncmp(name, "size=", 5) == 0) {
- tracefile_size = simple_strtoul(name + 5, NULL, 0);
- if (tracefile_size < 10 || tracefile_size > 20480)
- tracefile_size = TRACEFILE_SIZE;
- else
- tracefile_size <<= 20;
- goto out_sem;
- }
-
- if (name[0] != '/') {
- rc = -EINVAL;
- goto out_sem;
- }
-
- if (tracefile != NULL)
- kfree(tracefile);
-
- tracefile = name;
- name = NULL;
-
- printk(KERN_INFO "Lustre: debug daemon will attempt to start writing "
- "to %s (%lukB max)\n", tracefile, (long)(tracefile_size >> 10));
-
- trace_start_thread();
-
- out_sem:
- up_write(&tracefile_sem);
-
- out:
- kfree(name);
- return count;
-}
-
-int trace_read_daemon_file(char *page, char **start, off_t off, int count,
- int *eof, void *data)
-{
- int rc;
-
- down_read(&tracefile_sem);
- rc = snprintf(page, count, "%s", tracefile);
- up_read(&tracefile_sem);
-
- return rc;
+ mutex_up(&trace_thread_sem);
}
-int trace_write_debug_mb(struct file *file, const char *buffer,
- unsigned long count, void *data)
-{
- char string[32];
- int i;
- unsigned max;
-
- if (count >= sizeof(string)) {
- printk(KERN_ERR "Lustre: value too large (length %lu bytes)\n",
- count);
- return -EOVERFLOW;
- }
-
- if (copy_from_user(string, buffer, count))
- return -EFAULT;
-
- max = simple_strtoul(string, NULL, 0);
- if (max == 0)
- return -EINVAL;
- max /= smp_num_cpus;
-
- if (max * smp_num_cpus > (num_physpages >> (20 - 2 - PAGE_SHIFT)) / 5) {
- printk(KERN_ERR "Lustre: Refusing to set debug buffer size to "
- "%d MB, which is more than 80%% of physical RAM "
- "(%lu).\n", max * smp_num_cpus,
- (num_physpages >> (20 - 2 - PAGE_SHIFT)) / 5);
- return -EINVAL;
- }
-
- for (i = 0; i < NR_CPUS; i++) {
- struct trace_cpu_data *tcd;
- tcd = &trace_data[i].tcd;
- tcd->tcd_max_pages = max << (20 - PAGE_SHIFT);
- }
- return count;
-}
-
-int trace_read_debug_mb(char *page, char **start, off_t off, int count,
- int *eof, void *data)
-{
- struct trace_cpu_data *tcd;
- unsigned long flags;
- int rc;
-
- tcd = trace_get_tcd(flags);
- rc = snprintf(page, count, "%lu\n",
- (tcd->tcd_max_pages >> (20 - PAGE_SHIFT)) * smp_num_cpus);
- trace_put_tcd(tcd, flags);
-
- return rc;
-}
-
-int tracefile_init(void)
+int tracefile_init(int max_pages)
{
struct trace_cpu_data *tcd;
- int i;
-
- for (i = 0; i < NR_CPUS; i++) {
- tcd = &trace_data[i].tcd;
- INIT_LIST_HEAD(&tcd->tcd_pages);
- INIT_LIST_HEAD(&tcd->tcd_daemon_pages);
+ int i;
+ int j;
+ int rc;
+ int factor;
+
+ rc = tracefile_init_arch();
+ if (rc != 0)
+ return rc;
+
+ tcd_for_each(tcd, i, j) {
+ /* tcd_pages_factor is initialized int tracefile_init_arch. */
+ factor = tcd->tcd_pages_factor;
+ CFS_INIT_LIST_HEAD(&tcd->tcd_pages);
+ CFS_INIT_LIST_HEAD(&tcd->tcd_stock_pages);
+ CFS_INIT_LIST_HEAD(&tcd->tcd_daemon_pages);
tcd->tcd_cur_pages = 0;
+ tcd->tcd_cur_stock_pages = 0;
tcd->tcd_cur_daemon_pages = 0;
- tcd->tcd_max_pages = TCD_MAX_PAGES;
+ tcd->tcd_max_pages = (max_pages * factor) / 100;
+ LASSERT(tcd->tcd_max_pages > 0);
tcd->tcd_shutting_down = 0;
}
+
return 0;
}
static void trace_cleanup_on_cpu(void *info)
{
struct trace_cpu_data *tcd;
- struct list_head *pos, *tmp;
- unsigned long flags;
-
- tcd = trace_get_tcd(flags);
-
- tcd->tcd_shutting_down = 1;
+ struct trace_page *tage;
+ struct trace_page *tmp;
+ int i;
- list_for_each_safe(pos, tmp, &tcd->tcd_pages) {
- struct page *page;
+ tcd_for_each_type_lock(tcd, i) {
+ tcd->tcd_shutting_down = 1;
- page = list_entry(pos, struct page, PAGE_LIST_ENTRY);
- LASSERT(page->index <= PAGE_SIZE);
- LASSERT(page_count(page) > 0);
+ list_for_each_entry_safe(tage, tmp, &tcd->tcd_pages, linkage) {
+ __LASSERT_TAGE_INVARIANT(tage);
- list_del(&PAGE_LIST(page));
- page->mapping = NULL;
- __free_page(page);
+ list_del(&tage->linkage);
+ tage_free(tage);
+ }
+ tcd->tcd_cur_pages = 0;
}
- tcd->tcd_cur_pages = 0;
-
- trace_put_tcd(tcd, flags);
}
static void trace_cleanup(void)
{
struct page_collection pc;
- INIT_LIST_HEAD(&pc.pc_pages);
+ CFS_INIT_LIST_HEAD(&pc.pc_pages);
spin_lock_init(&pc.pc_lock);
- trace_cleanup_on_cpu(&pc);
- smp_call_function(trace_cleanup_on_cpu, &pc, 0, 1);
+ trace_call_on_all_cpus(trace_cleanup_on_cpu, &pc);
+
+ tracefile_fini_arch();
}
void tracefile_exit(void)