s390/vmcp: make use of contiguous memory allocator

If memory is fragmented it is unlikely that large order memory
allocations succeed. This has been an issue with the vmcp device
driver since a long time, since it requires large physical contiguous
memory ares for large responses.

To hopefully resolve this issue make use of the contiguous memory
allocator (cma). This patch adds a vmcp specific vmcp cma area with a
default size of 4MB. The size can be changed either via the
VMCP_CMA_SIZE config option at compile time or with the "vmcp_cma"
kernel parameter (e.g. "vmcp_cma=16m").

For any vmcp response buffers larger than 16k memory from the cma area
will be allocated. If such an allocation fails, there is a fallback to
the buddy allocator.

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

3 files changed, 79 insertions, 9 deletions

diff --git a/drivers/s390/char/Kconfig b/drivers/s390/char/Kconfig
index b3f1c45..97c4c9f 100644
--- a/drivers/s390/char/Kconfig
+++ b/drivers/s390/char/Kconfig
@@ -169,10 +169,21 @@ config VMCP
 	def_bool y
 	prompt "Support for the z/VM CP interface"
 	depends on S390
+	select CMA
 	help
 	  Select this option if you want to be able to interact with the control
 	  program on z/VM
 
+config VMCP_CMA_SIZE
+	int "Memory in MiB reserved for z/VM CP interface"
+	default "4"
+	depends on VMCP
+	help
+	  Specify the default amount of memory in MiB reserved for the z/VM CP
+	  interface. If needed this memory is used for large contiguous memory
+	  allocations. The default can be changed with the kernel command line
+	  parameter "vmcp_cma".
+
 config MONREADER
 	def_tristate m
 	prompt "API for reading z/VM monitor service records"
diff --git a/drivers/s390/char/vmcp.c b/drivers/s390/char/vmcp.c
index b5e3a49..c202b40 100644
--- a/drivers/s390/char/vmcp.c
+++ b/drivers/s390/char/vmcp.c
@@ -17,15 +17,77 @@
 #include <linux/kernel.h>
 #include <linux/miscdevice.h>
 #include <linux/slab.h>
+#include <linux/uaccess.h>
 #include <linux/export.h>
+#include <linux/mutex.h>
+#include <linux/cma.h>
+#include <linux/mm.h>
 #include <asm/compat.h>
 #include <asm/cpcmd.h>
 #include <asm/debug.h>
-#include <linux/uaccess.h>
 #include "vmcp.h"
 
 static debug_info_t *vmcp_debug;
 
+static unsigned long vmcp_cma_size __initdata = CONFIG_VMCP_CMA_SIZE * 1024 * 1024;
+static struct cma *vmcp_cma;
+
+static int __init early_parse_vmcp_cma(char *p)
+{
+	vmcp_cma_size = ALIGN(memparse(p, NULL), PAGE_SIZE);
+	return 0;
+}
+early_param("vmcp_cma", early_parse_vmcp_cma);
+
+void __init vmcp_cma_reserve(void)
+{
+	if (!MACHINE_IS_VM)
+		return;
+	cma_declare_contiguous(0, vmcp_cma_size, 0, 0, 0, false, "vmcp", &vmcp_cma);
+}
+
+static void vmcp_response_alloc(struct vmcp_session *session)
+{
+	struct page *page = NULL;
+	int nr_pages, order;
+
+	order = get_order(session->bufsize);
+	nr_pages = ALIGN(session->bufsize, PAGE_SIZE) >> PAGE_SHIFT;
+	/*
+	 * For anything below order 3 allocations rely on the buddy
+	 * allocator. If such low-order allocations can't be handled
+	 * anymore the system won't work anyway.
+	 */
+	if (order > 2)
+		page = cma_alloc(vmcp_cma, nr_pages, 0, GFP_KERNEL);
+	if (page) {
+		session->response = (char *)page_to_phys(page);
+		session->cma_alloc = 1;
+		return;
+	}
+	session->response = (char *)__get_free_pages(GFP_KERNEL | __GFP_RETRY_MAYFAIL, order);
+}
+
+static void vmcp_response_free(struct vmcp_session *session)
+{
+	int nr_pages, order;
+	struct page *page;
+
+	if (!session->response)
+		return;
+	order = get_order(session->bufsize);
+	nr_pages = ALIGN(session->bufsize, PAGE_SIZE) >> PAGE_SHIFT;
+	if (session->cma_alloc) {
+		page = phys_to_page((unsigned long)session->response);
+		cma_release(vmcp_cma, page, nr_pages);
+		session->cma_alloc = 0;
+		goto out;
+	}
+	free_pages((unsigned long)session->response, order);
+out:
+	session->response = NULL;
+}
+
 static int vmcp_open(struct inode *inode, struct file *file)
 {
 	struct vmcp_session *session;
@@ -51,7 +113,7 @@ static int vmcp_release(struct inode *inode, struct file *file)
 
 	session = file->private_data;
 	file->private_data = NULL;
-	free_pages((unsigned long)session->response, get_order(session->bufsize));
+	vmcp_response_free(session);
 	kfree(session);
 	return 0;
 }
@@ -97,9 +159,7 @@ vmcp_write(struct file *file, const char __user *buff, size_t count,
 		return -ERESTARTSYS;
 	}
 	if (!session->response)
-		session->response = (char *)__get_free_pages(GFP_KERNEL
-						| __GFP_RETRY_MAYFAIL,
-						get_order(session->bufsize));
+		vmcp_response_alloc(session);
 	if (!session->response) {
 		mutex_unlock(&session->mutex);
 		kfree(cmd);
@@ -146,9 +206,7 @@ static long vmcp_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 		mutex_unlock(&session->mutex);
 		return put_user(temp, argp);
 	case VMCP_SETBUF:
-		free_pages((unsigned long)session->response,
-				get_order(session->bufsize));
-		session->response=NULL;
+		vmcp_response_free(session);
 		temp = get_user(session->bufsize, argp);
 		if (temp)
 			session->bufsize = PAGE_SIZE;
diff --git a/drivers/s390/char/vmcp.h b/drivers/s390/char/vmcp.h
index 1e29b04..4e725ed 100644
--- a/drivers/s390/char/vmcp.h
+++ b/drivers/s390/char/vmcp.h
@@ -20,8 +20,9 @@
 #define VMCP_GETSIZE _IOR(0x10, 3, int)
 
 struct vmcp_session {
-	unsigned int bufsize;
 	char *response;
+	unsigned int bufsize;
+	unsigned int cma_alloc : 1;
 	int resp_size;
 	int resp_code;
 	/* As we use copy_from/to_user, which might     *