X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=memory.c;h=af4d5ef8ee371701f02806369a3581ecfdc42628;hp=d9cfe9bae2c0b35f24ec3e301ef74d459b4d8407;hb=dd117eb614043fa01480e6a903b8aac6df5bbf80;hpb=e1161c325f7866bae879e686d1c673ca32ab09ae

diff --git a/memory.c b/memory.c
index d9cfe9ba..af4d5ef8 100644
--- a/memory.c
+++ b/memory.c
@@ -1,138 +1,283 @@
 /*
  * Memory helpers
  */
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <fcntl.h>
 #include <unistd.h>
-#include <sys/shm.h>
 #include <sys/mman.h>
 
 #include "fio.h"
-#include "os.h"
-
-static void *pinned_mem;
+#ifndef FIO_NO_HAVE_SHM_H
+#include <sys/shm.h>
+#endif
 
-void fio_unpin_memory(void)
+void fio_unpin_memory(struct thread_data *td)
 {
-	if (pinned_mem) {
-		if (munlock(pinned_mem, mlock_size) < 0)
+	if (td->pinned_mem) {
+		dprint(FD_MEM, "unpinning %llu bytes\n", td->o.lockmem);
+		if (munlock(td->pinned_mem, td->o.lockmem) < 0)
 			perror("munlock");
-		munmap(pinned_mem, mlock_size);
-		pinned_mem = NULL;
+		munmap(td->pinned_mem, td->o.lockmem);
+		td->pinned_mem = NULL;
 	}
 }
 
-int fio_pin_memory(void)
+int fio_pin_memory(struct thread_data *td)
 {
 	unsigned long long phys_mem;
 
-	if (!mlock_size)
+	if (!td->o.lockmem)
 		return 0;
 
+	dprint(FD_MEM, "pinning %llu bytes\n", td->o.lockmem);
+
 	/*
 	 * Don't allow mlock of more than real_mem-128MB
 	 */
 	phys_mem = os_phys_mem();
 	if (phys_mem) {
-		if ((mlock_size + 128 * 1024 * 1024) > phys_mem) {
-			mlock_size = phys_mem - 128 * 1024 * 1024;
-			fprintf(f_out, "fio: limiting mlocked memory to %lluMiB\n", mlock_size >> 20);
+		if ((td->o.lockmem + 128 * 1024 * 1024) > phys_mem) {
+			td->o.lockmem = phys_mem - 128 * 1024 * 1024;
+			log_info("fio: limiting mlocked memory to %lluMB\n",
+							td->o.lockmem >> 20);
 		}
 	}
 
-	pinned_mem = mmap(NULL, mlock_size, PROT_READ | PROT_WRITE, MAP_PRIVATE | OS_MAP_ANON, 0, 0);
-	if (pinned_mem == MAP_FAILED) {
+	td->pinned_mem = mmap(NULL, td->o.lockmem, PROT_READ | PROT_WRITE,
+				MAP_PRIVATE | OS_MAP_ANON, -1, 0);
+	if (td->pinned_mem == MAP_FAILED) {
 		perror("malloc locked mem");
-		pinned_mem = NULL;
+		td->pinned_mem = NULL;
 		return 1;
 	}
-	if (mlock(pinned_mem, mlock_size) < 0) {
+	if (mlock(td->pinned_mem, td->o.lockmem) < 0) {
 		perror("mlock");
-		munmap(pinned_mem, mlock_size);
-		pinned_mem = NULL;
+		munmap(td->pinned_mem, td->o.lockmem);
+		td->pinned_mem = NULL;
 		return 1;
 	}
 
 	return 0;
 }
 
-/*
- * Setup the buffer area we need for io.
- */
-int allocate_io_mem(struct thread_data *td)
+static int alloc_mem_shm(struct thread_data *td, unsigned int total_mem)
 {
-	if (td->mem_type == MEM_MALLOC)
-		td->orig_buffer = malloc(td->orig_buffer_size);
-	else if (td->mem_type == MEM_SHM || td->mem_type == MEM_SHMHUGE) {
-		int flags = IPC_CREAT | SHM_R | SHM_W;
-
-		if (td->mem_type == MEM_SHMHUGE)
-			flags |= SHM_HUGETLB;
-
-		td->shm_id = shmget(IPC_PRIVATE, td->orig_buffer_size, flags);
-		if (td->shm_id < 0) {
-			td_verror(td, errno, "shmget");
-			perror("shmget");
-			return 1;
-		}
+#ifndef CONFIG_NO_SHM
+	int flags = IPC_CREAT | S_IRUSR | S_IWUSR;
 
-		td->orig_buffer = shmat(td->shm_id, NULL, 0);
-		if (td->orig_buffer == (void *) -1) {
-			td_verror(td, errno, "shmat");
-			perror("shmat");
-			td->orig_buffer = NULL;
-			return 1;
+	if (td->o.mem_type == MEM_SHMHUGE) {
+		unsigned long mask = td->o.hugepage_size - 1;
+
+		flags |= SHM_HUGETLB;
+		total_mem = (total_mem + mask) & ~mask;
+	}
+
+	td->shm_id = shmget(IPC_PRIVATE, total_mem, flags);
+	dprint(FD_MEM, "shmget %u, %d\n", total_mem, td->shm_id);
+	if (td->shm_id < 0) {
+		td_verror(td, errno, "shmget");
+		if (geteuid() != 0 && (errno == ENOMEM || errno == EPERM))
+			log_err("fio: you may need to run this job as root\n");
+		if (td->o.mem_type == MEM_SHMHUGE) {
+			if (errno == EINVAL) {
+				log_err("fio: check that you have free huge"
+					" pages and that hugepage-size is"
+					" correct.\n");
+			} else if (errno == ENOSYS) {
+				log_err("fio: your system does not appear to"
+					" support huge pages.\n");
+			} else if (errno == ENOMEM) {
+				log_err("fio: no huge pages available, do you"
+					" need to allocate some? See HOWTO.\n");
+			}
 		}
-	} else if (td->mem_type == MEM_MMAP || td->mem_type == MEM_MMAPHUGE) {
-		int flags = MAP_PRIVATE;
 
-		td->mmapfd = 0;
+		return 1;
+	}
 
-		if (td->mmapfile) {
-			td->mmapfd = open(td->mmapfile, O_RDWR|O_CREAT, 0644);
+	td->orig_buffer = shmat(td->shm_id, NULL, 0);
+	dprint(FD_MEM, "shmat %d, %p\n", td->shm_id, td->orig_buffer);
+	if (td->orig_buffer == (void *) -1) {
+		td_verror(td, errno, "shmat");
+		td->orig_buffer = NULL;
+		return 1;
+	}
 
-			if (td->mmapfd < 0) {
-				td_verror(td, errno, "open");
-				perror("open mmap file");
-				td->orig_buffer = NULL;
-				return 1;
-			}
-		} else
-			flags |= OS_MAP_ANON;
+	return 0;
+#else
+	log_err("fio: shm not supported\n");
+	return 1;
+#endif
+}
+
+static void free_mem_shm(struct thread_data *td)
+{
+#ifndef CONFIG_NO_SHM
+	struct shmid_ds sbuf;
+
+	dprint(FD_MEM, "shmdt/ctl %d %p\n", td->shm_id, td->orig_buffer);
+	shmdt(td->orig_buffer);
+	shmctl(td->shm_id, IPC_RMID, &sbuf);
+#endif
+}
+
+static int alloc_mem_mmap(struct thread_data *td, size_t total_mem)
+{
+	int flags = 0;
+
+	td->mmapfd = -1;
 
-		td->orig_buffer = mmap(NULL, td->orig_buffer_size, PROT_READ | PROT_WRITE, flags, td->mmapfd, 0);
-		if (td->orig_buffer == MAP_FAILED) {
-			td_verror(td, errno, "mmap");
-			perror("mmap");
+	if (td->o.mem_type == MEM_MMAPHUGE) {
+		unsigned long mask = td->o.hugepage_size - 1;
+
+		/* TODO: make sure the file is a real hugetlbfs file */
+		if (!td->o.mmapfile)
+			flags |= MAP_HUGETLB;
+		total_mem = (total_mem + mask) & ~mask;
+	}
+
+	if (td->o.mmapfile) {
+		td->mmapfd = open(td->o.mmapfile, O_RDWR|O_CREAT, 0644);
+
+		if (td->mmapfd < 0) {
+			td_verror(td, errno, "open mmap file");
 			td->orig_buffer = NULL;
-			if (td->mmapfd) {
-				close(td->mmapfd);
-				unlink(td->mmapfile);
-			}
-				
 			return 1;
 		}
+		if (td->o.mem_type != MEM_MMAPHUGE &&
+		    td->o.mem_type != MEM_MMAPSHARED &&
+		    ftruncate(td->mmapfd, total_mem) < 0) {
+			td_verror(td, errno, "truncate mmap file");
+			td->orig_buffer = NULL;
+			return 1;
+		}
+		if (td->o.mem_type == MEM_MMAPHUGE ||
+		    td->o.mem_type == MEM_MMAPSHARED)
+			flags |= MAP_SHARED;
+		else
+			flags |= MAP_PRIVATE;
+	} else
+		flags |= OS_MAP_ANON | MAP_PRIVATE;
+
+	td->orig_buffer = mmap(NULL, total_mem, PROT_READ | PROT_WRITE, flags,
+				td->mmapfd, 0);
+	dprint(FD_MEM, "mmap %llu/%d %p\n", (unsigned long long) total_mem,
+						td->mmapfd, td->orig_buffer);
+	if (td->orig_buffer == MAP_FAILED) {
+		td_verror(td, errno, "mmap");
+		td->orig_buffer = NULL;
+		if (td->mmapfd != 1 && td->mmapfd != -1) {
+			close(td->mmapfd);
+			if (td->o.mmapfile)
+				unlink(td->o.mmapfile);
+		}
+
+		return 1;
 	}
 
 	return 0;
 }
 
-void free_io_mem(struct thread_data *td)
+static void free_mem_mmap(struct thread_data *td, size_t total_mem)
 {
-	if (td->mem_type == MEM_MALLOC)
-		free(td->orig_buffer);
-	else if (td->mem_type == MEM_SHM || td->mem_type == MEM_SHMHUGE) {
-		struct shmid_ds sbuf;
-
-		shmdt(td->orig_buffer);
-		shmctl(td->shm_id, IPC_RMID, &sbuf);
-	} else if (td->mem_type == MEM_MMAP || td->mem_type == MEM_MMAPHUGE) {
-		munmap(td->orig_buffer, td->orig_buffer_size);
-		if (td->mmapfile) {
+	dprint(FD_MEM, "munmap %llu %p\n", (unsigned long long) total_mem,
+						td->orig_buffer);
+	munmap(td->orig_buffer, td->orig_buffer_size);
+	if (td->o.mmapfile) {
+		if (td->mmapfd != -1)
 			close(td->mmapfd);
-			unlink(td->mmapfile);
-			free(td->mmapfile);
-		}
-	} else
-		log_err("Bad memory type %u\n", td->mem_type);
+		unlink(td->o.mmapfile);
+		free(td->o.mmapfile);
+	}
+}
+
+static int alloc_mem_malloc(struct thread_data *td, size_t total_mem)
+{
+	td->orig_buffer = malloc(total_mem);
+	dprint(FD_MEM, "malloc %llu %p\n", (unsigned long long) total_mem,
+							td->orig_buffer);
+
+	return td->orig_buffer == NULL;
+}
+
+static void free_mem_malloc(struct thread_data *td)
+{
+	dprint(FD_MEM, "free malloc mem %p\n", td->orig_buffer);
+	free(td->orig_buffer);
+}
+
+/*
+ * Set up the buffer area we need for io.
+ */
+int allocate_io_mem(struct thread_data *td)
+{
+	size_t total_mem;
+	int ret = 0;
+
+	if (td->io_ops->flags & FIO_NOIO)
+		return 0;
+
+	total_mem = td->orig_buffer_size;
+
+	if (td->o.odirect || td->o.mem_align || td->o.oatomic ||
+	    (td->io_ops->flags & FIO_MEMALIGN)) {
+		total_mem += page_mask;
+		if (td->o.mem_align && td->o.mem_align > page_size)
+			total_mem += td->o.mem_align - page_size;
+	}
+
+	dprint(FD_MEM, "Alloc %llu for buffers\n", (unsigned long long) total_mem);
+
+	/*
+	 * If the IO engine has hooks to allocate/free memory, use those. But
+	 * error out if the user explicitly asked for something else.
+	 */
+	if (td->io_ops->iomem_alloc) {
+		if (fio_option_is_set(&td->o, mem_type)) {
+			log_err("fio: option 'mem/iomem' conflicts with specified IO engine\n");
+			ret = 1;
+		} else
+			ret = td->io_ops->iomem_alloc(td, total_mem);
+	} else if (td->o.mem_type == MEM_MALLOC)
+		ret = alloc_mem_malloc(td, total_mem);
+	else if (td->o.mem_type == MEM_SHM || td->o.mem_type == MEM_SHMHUGE)
+		ret = alloc_mem_shm(td, total_mem);
+	else if (td->o.mem_type == MEM_MMAP || td->o.mem_type == MEM_MMAPHUGE ||
+		 td->o.mem_type == MEM_MMAPSHARED)
+		ret = alloc_mem_mmap(td, total_mem);
+	else {
+		log_err("fio: bad mem type: %d\n", td->o.mem_type);
+		ret = 1;
+	}
+
+	if (ret)
+		td_verror(td, ENOMEM, "iomem allocation");
+
+	return ret;
+}
+
+void free_io_mem(struct thread_data *td)
+{
+	unsigned int total_mem;
+
+	total_mem = td->orig_buffer_size;
+	if (td->o.odirect || td->o.oatomic)
+		total_mem += page_mask;
+
+	if (td->io_ops->iomem_alloc) {
+		if (td->io_ops->iomem_free)
+			td->io_ops->iomem_free(td);
+	} else if (td->o.mem_type == MEM_MALLOC)
+		free_mem_malloc(td);
+	else if (td->o.mem_type == MEM_SHM || td->o.mem_type == MEM_SHMHUGE)
+		free_mem_shm(td);
+	else if (td->o.mem_type == MEM_MMAP || td->o.mem_type == MEM_MMAPHUGE ||
+		 td->o.mem_type == MEM_MMAPSHARED)
+		free_mem_mmap(td, total_mem);
+	else
+		log_err("Bad memory type %u\n", td->o.mem_type);
 
 	td->orig_buffer = NULL;
+	td->orig_buffer_size = 0;
 }