[S390] boot from NSS support

Add support to boot from a named saved segment (NSS).

Signed-off-by: Hongjie Yang <hongjie@us.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

5 files changed, 275 insertions, 64 deletions

diff --git a/arch/s390/kernel/head31.S b/arch/s390/kernel/head31.S
index eca507050e47..b3dcdcdc80c0 100644
--- a/arch/s390/kernel/head31.S
+++ b/arch/s390/kernel/head31.S
@@ -51,20 +51,12 @@ startup_continue:
         st      %r15,__LC_KERNEL_STACK  # set end of kernel stack
         ahi     %r15,-96
         xc      __SF_BACKCHAIN(4,%r15),__SF_BACKCHAIN(%r15) # clear backchain
-
-        l       %r14,.Lipl_save_parameters-.LPG1(%r13)
-        basr    %r14,%r14
 #
-# clear bss memory
+# Save ipl parameters, clear bss memory, initialize storage key for kernel pages,
+# and create a kernel NSS if the SAVESYS= parm is defined
 #
-        l       %r2,.Lbss_bgn-.LPG1(%r13) # start of bss
-        l       %r3,.Lbss_end-.LPG1(%r13) # end of bss
-        sr      %r3,%r2                 # length of bss
-        sr      %r4,%r4
-        sr      %r5,%r5                 # set src,length and pad to zero
-        sr      %r0,%r0
-        mvcle   %r2,%r4,0               # clear mem
-        jo      .-4                     # branch back, if not finish
+        l       %r14,.Lstartup_init-.LPG1(%r13)
+        basr    %r14,%r14
 
         l       %r2,.Lrcp-.LPG1(%r13)   # Read SCP forced command word
 .Lservicecall:
@@ -125,10 +117,10 @@ startup_continue:
         b       .Lfchunk-.LPG1(%r13)
 
         .align 4
-.Lipl_save_parameters:
-        .long   ipl_save_parameters
 .Linittu:
         .long   init_thread_union
+.Lstartup_init:
+        .long   startup_init
 .Lpmask:
         .byte   0
         .align  8
@@ -207,20 +199,6 @@ startup_continue:
 .Ldonemem:
         l       %r12,.Lmflags-.LPG1(%r13) # get address of machine_flags
 #
-# find out if we are running under VM
-#
-        stidp   __LC_CPUID              # store cpuid
-        tm      __LC_CPUID,0xff         # running under VM ?
-        bno     .Lnovm-.LPG1(%r13)
-        oi      3(%r12),1               # set VM flag
-.Lnovm:
-        lh      %r0,__LC_CPUID+4        # get cpu version
-        chi     %r0,0x7490              # running on a P/390 ?
-        bne     .Lnop390-.LPG1(%r13)
-        oi      3(%r12),4               # set P/390 flag
-.Lnop390:
-
-#
 # find out if we have an IEEE fpu
 #
         mvc     __LC_PGM_NEW_PSW(8),.Lpcfpu-.LPG1(%r13)
diff --git a/arch/s390/kernel/head64.S b/arch/s390/kernel/head64.S
index e940e802cb40..030a1c95f47c 100644
--- a/arch/s390/kernel/head64.S
+++ b/arch/s390/kernel/head64.S
@@ -58,18 +58,11 @@ startup_continue:
         stg     %r15,__LC_KERNEL_STACK  # set end of kernel stack
         aghi    %r15,-160
         xc      __SF_BACKCHAIN(4,%r15),__SF_BACKCHAIN(%r15) # clear backchain
-
-        brasl   %r14,ipl_save_parameters
 #
-# clear bss memory
+# Save ipl parameters, clear bss memory, initialize storage key for kernel pages,
+# and create a kernel NSS if the SAVESYS= parm is defined
 #
-        larl    %r2,__bss_start         # start of bss segment
-        larl    %r3,_end                # end of bss segment
-        sgr     %r3,%r2                 # length of bss
-        sgr     %r4,%r4                 #
-        sgr     %r5,%r5                 # set src,length and pad to zero
-        mvcle   %r2,%r4,0               # clear mem
-        jo      .-4                     # branch back, if not finish
+        brasl   %r14,startup_init
                                         # set program check new psw mask
         mvc     __LC_PGM_NEW_PSW(8),.Lpcmsk-.LPG1(%r13)
         larl    %r1,.Lslowmemdetect     # set program check address
@@ -78,6 +71,10 @@ startup_continue:
         diag    %r0,%r1,0x260           # get memory size of virtual machine
         cgr     %r0,%r1                 # different? -> old detection routine
         jne     .Lslowmemdetect
+        larl    %r3,ipl_flags
+        llgt    %r3,0(%r3)
+        chi     %r3,4                   # ipled from an kernel NSS
+        je      .Lslowmemdetect
         aghi    %r1,1                   # size is one more than end
         larl    %r2,memory_chunk
         stg     %r1,8(%r2)              # store size of chunk
@@ -226,19 +223,6 @@ startup_continue:
 
         larl    %r12,machine_flags
 #
-# find out if we are running under VM
-#
-        stidp   __LC_CPUID              # store cpuid
-        tm      __LC_CPUID,0xff         # running under VM ?
-        bno     0f-.LPG1(%r13)
-        oi      7(%r12),1               # set VM flag
-0:      lh      %r0,__LC_CPUID+4        # get cpu version
-        chi     %r0,0x7490              # running on a P/390 ?
-        bne     1f-.LPG1(%r13)
-        oi      7(%r12),4               # set P/390 flag
-1:
-
-#
 # find out if we have the MVPG instruction
 #
         la      %r1,0f-.LPG1(%r13)      # set program check address
diff --git a/arch/s390/kernel/ipl.c b/arch/s390/kernel/ipl.c
index 2c91226e1d40..13eacce62011 100644
--- a/arch/s390/kernel/ipl.c
+++ b/arch/s390/kernel/ipl.c
@@ -34,12 +34,14 @@ enum ipl_type {
         IPL_TYPE_UNKNOWN = 2,
         IPL_TYPE_CCW     = 4,
         IPL_TYPE_FCP     = 8,
+        IPL_TYPE_NSS     = 16,
 };
 
 #define IPL_NONE_STR     "none"
 #define IPL_UNKNOWN_STR  "unknown"
 #define IPL_CCW_STR      "ccw"
 #define IPL_FCP_STR      "fcp"
+#define IPL_NSS_STR      "nss"
 
 static char *ipl_type_str(enum ipl_type type)
 {
@@ -50,6 +52,8 @@ static char *ipl_type_str(enum ipl_type type)
                 return IPL_CCW_STR;
         case IPL_TYPE_FCP:
                 return IPL_FCP_STR;
+        case IPL_TYPE_NSS:
+                return IPL_NSS_STR;
         case IPL_TYPE_UNKNOWN:
         default:
                 return IPL_UNKNOWN_STR;
@@ -64,6 +68,7 @@ enum ipl_method {
         IPL_METHOD_FCP_RO_DIAG,
         IPL_METHOD_FCP_RW_DIAG,
         IPL_METHOD_FCP_RO_VM,
+        IPL_METHOD_NSS,
 };
 
 enum shutdown_action {
@@ -114,11 +119,14 @@ enum diag308_rc {
 static int diag308_set_works = 0;
 
 static int reipl_capabilities = IPL_TYPE_UNKNOWN;
+
 static enum ipl_type reipl_type = IPL_TYPE_UNKNOWN;
 static enum ipl_method reipl_method = IPL_METHOD_NONE;
 static struct ipl_parameter_block *reipl_block_fcp;
 static struct ipl_parameter_block *reipl_block_ccw;
 
+static char reipl_nss_name[NSS_NAME_SIZE + 1];
+
 static int dump_capabilities = IPL_TYPE_NONE;
 static enum ipl_type dump_type = IPL_TYPE_NONE;
 static enum ipl_method dump_method = IPL_METHOD_NONE;
@@ -173,6 +181,24 @@ static struct subsys_attribute sys_##_prefix##_##_name##_attr =		\
                         sys_##_prefix##_##_name##_show,                 \
                         sys_##_prefix##_##_name##_store);
 
+#define DEFINE_IPL_ATTR_STR_RW(_prefix, _name, _fmt_out, _fmt_in, _value)\
+static ssize_t sys_##_prefix##_##_name##_show(struct subsystem *subsys, \
+                char *page)                                             \
+{                                                                       \
+        return sprintf(page, _fmt_out, _value);                         \
+}                                                                       \
+static ssize_t sys_##_prefix##_##_name##_store(struct subsystem *subsys,\
+                const char *buf, size_t len)                            \
+{                                                                       \
+        if (sscanf(buf, _fmt_in, _value) != 1)                          \
+                return -EINVAL;                                         \
+        return len;                                                     \
+}                                                                       \
+static struct subsys_attribute sys_##_prefix##_##_name##_attr =         \
+        __ATTR(_name,(S_IRUGO | S_IWUSR),                               \
+                        sys_##_prefix##_##_name##_show,                 \
+                        sys_##_prefix##_##_name##_store);
+
 static void make_attrs_ro(struct attribute **attrs)
 {
         while (*attrs) {
@@ -189,6 +215,8 @@ static enum ipl_type ipl_get_type(void)
 {
         struct ipl_parameter_block *ipl = IPL_PARMBLOCK_START;
 
+        if (ipl_flags & IPL_NSS_VALID)
+                return IPL_TYPE_NSS;
         if (!(ipl_flags & IPL_DEVNO_VALID))
                 return IPL_TYPE_UNKNOWN;
         if (!(ipl_flags & IPL_PARMBLOCK_VALID))
@@ -324,6 +352,20 @@ static struct attribute_group ipl_ccw_attr_group = {
         .attrs = ipl_ccw_attrs,
 };
 
+/* NSS ipl device attributes */
+
+DEFINE_IPL_ATTR_RO(ipl_nss, name, "%s\n", kernel_nss_name);
+
+static struct attribute *ipl_nss_attrs[] = {
+        &sys_ipl_type_attr.attr,
+        &sys_ipl_nss_name_attr.attr,
+        NULL,
+};
+
+static struct attribute_group ipl_nss_attr_group = {
+        .attrs = ipl_nss_attrs,
+};
+
 /* UNKNOWN ipl device attributes */
 
 static struct attribute *ipl_unknown_attrs[] = {
@@ -432,6 +474,21 @@ static struct attribute_group reipl_ccw_attr_group = {
         .attrs = reipl_ccw_attrs,
 };
 
+
+/* NSS reipl device attributes */
+
+DEFINE_IPL_ATTR_STR_RW(reipl_nss, name, "%s\n", "%s\n", reipl_nss_name);
+
+static struct attribute *reipl_nss_attrs[] = {
+        &sys_reipl_nss_name_attr.attr,
+        NULL,
+};
+
+static struct attribute_group reipl_nss_attr_group = {
+        .name  = IPL_NSS_STR,
+        .attrs = reipl_nss_attrs,
+};
+
 /* reipl type */
 
 static int reipl_set_type(enum ipl_type type)
@@ -454,6 +511,9 @@ static int reipl_set_type(enum ipl_type type)
                 else
                         reipl_method = IPL_METHOD_FCP_RO_DIAG;
                 break;
+        case IPL_TYPE_NSS:
+                reipl_method = IPL_METHOD_NSS;
+                break;
         default:
                 reipl_method = IPL_METHOD_NONE;
         }
@@ -475,6 +535,8 @@ static ssize_t reipl_type_store(struct subsystem *subsys, const char *buf,
                 rc = reipl_set_type(IPL_TYPE_CCW);
         else if (strncmp(buf, IPL_FCP_STR, strlen(IPL_FCP_STR)) == 0)
                 rc = reipl_set_type(IPL_TYPE_FCP);
+        else if (strncmp(buf, IPL_NSS_STR, strlen(IPL_NSS_STR)) == 0)
+                rc = reipl_set_type(IPL_TYPE_NSS);
         return (rc != 0) ? rc : len;
 }
 
@@ -647,6 +709,10 @@ void do_reipl(void)
         case IPL_METHOD_FCP_RO_VM:
                 __cpcmd("IPL", NULL, 0, NULL);
                 break;
+        case IPL_METHOD_NSS:
+                sprintf(buf, "IPL %s", reipl_nss_name);
+                __cpcmd(buf, NULL, 0, NULL);
+                break;
         case IPL_METHOD_NONE:
         default:
                 if (MACHINE_IS_VM)
@@ -733,6 +799,10 @@ static int __init ipl_init(void)
         case IPL_TYPE_FCP:
                 rc = ipl_register_fcp_files();
                 break;
+        case IPL_TYPE_NSS:
+                rc = sysfs_create_group(&ipl_subsys.kset.kobj,
+                                        &ipl_nss_attr_group);
+                break;
         default:
                 rc = sysfs_create_group(&ipl_subsys.kset.kobj,
                                         &ipl_unknown_attr_group);
@@ -755,6 +825,20 @@ static void __init reipl_probe(void)
         free_page((unsigned long)buffer);
 }
 
+static int __init reipl_nss_init(void)
+{
+        int rc;
+
+        if (!MACHINE_IS_VM)
+                return 0;
+        rc = sysfs_create_group(&reipl_subsys.kset.kobj, &reipl_nss_attr_group);
+        if (rc)
+                return rc;
+        strncpy(reipl_nss_name, kernel_nss_name, NSS_NAME_SIZE + 1);
+        reipl_capabilities |= IPL_TYPE_NSS;
+        return 0;
+}
+
 static int __init reipl_ccw_init(void)
 {
         int rc;
@@ -837,6 +921,9 @@ static int __init reipl_init(void)
         rc = reipl_fcp_init();
         if (rc)
                 return rc;
+        rc = reipl_nss_init();
+        if (rc)
+                return rc;
         rc = reipl_set_type(ipl_get_type());
         if (rc)
                 return rc;
diff --git a/arch/s390/kernel/setup.c b/arch/s390/kernel/setup.c
index b1b9a931237d..2569aafcc543 100644
--- a/arch/s390/kernel/setup.c
+++ b/arch/s390/kernel/setup.c
@@ -38,6 +38,7 @@
 #include <linux/device.h>
 #include <linux/notifier.h>
 #include <linux/pfn.h>
+#include <linux/ctype.h>
 #include <linux/reboot.h>
 
 #include <asm/uaccess.h>
@@ -50,6 +51,7 @@
 #include <asm/page.h>
 #include <asm/ptrace.h>
 #include <asm/sections.h>
+#include <asm/ebcdic.h>
 #include <asm/compat.h>
 
 long psw_kernel_bits    = (PSW_BASE_BITS | PSW_MASK_DAT | PSW_ASC_PRIMARY |
@@ -282,6 +284,140 @@ static void __init conmode_default(void)
         }
 }
 
+/*
+ * Create a Kernel NSS if the SAVESYS= parameter is defined
+*/
+#define DEFSYS_CMD_SIZE 96
+#define SAVESYS_CMD_SIZE        32
+
+extern int _eshared;
+char kernel_nss_name[NSS_NAME_SIZE + 1];
+
+#ifdef CONFIG_SHARED_KERNEL
+static __init void create_kernel_nss(void)
+{
+        unsigned int i, stext_pfn, eshared_pfn, end_pfn, min_size;
+#ifdef CONFIG_BLK_DEV_INITRD
+        unsigned int sinitrd_pfn, einitrd_pfn;
+#endif
+        int response;
+        char *savesys_ptr;
+        char upper_command_line[COMMAND_LINE_SIZE];
+        char defsys_cmd[DEFSYS_CMD_SIZE];
+        char savesys_cmd[SAVESYS_CMD_SIZE];
+
+        /* Do nothing if we are not running under VM */
+        if (!MACHINE_IS_VM)
+                return;
+
+        /* Convert COMMAND_LINE to upper case */
+        for (i = 0; i < strlen(COMMAND_LINE); i++)
+                upper_command_line[i] = toupper(COMMAND_LINE[i]);
+
+        savesys_ptr = strstr(upper_command_line, "SAVESYS=");
+
+        if (!savesys_ptr)
+                return;
+
+        savesys_ptr += 8;    /* Point to the beginning of the NSS name */
+        for (i = 0; i < NSS_NAME_SIZE; i++) {
+                if (savesys_ptr[i] == ' ' || savesys_ptr[i] == '\0')
+                        break;
+                kernel_nss_name[i] = savesys_ptr[i];
+        }
+
+        stext_pfn = PFN_DOWN(__pa(&_stext));
+        eshared_pfn = PFN_DOWN(__pa(&_eshared));
+        end_pfn = PFN_UP(__pa(&_end));
+        min_size = end_pfn << 2;
+
+        sprintf(defsys_cmd, "DEFSYS %s 00000-%.5X EW %.5X-%.5X SR %.5X-%.5X",
+                kernel_nss_name, stext_pfn - 1, stext_pfn, eshared_pfn - 1,
+                eshared_pfn, end_pfn);
+
+#ifdef CONFIG_BLK_DEV_INITRD
+        if (INITRD_START && INITRD_SIZE) {
+                sinitrd_pfn = PFN_DOWN(__pa(INITRD_START));
+                einitrd_pfn = PFN_UP(__pa(INITRD_START + INITRD_SIZE));
+                min_size = einitrd_pfn << 2;
+                sprintf(defsys_cmd, "%s EW %.5X-%.5X", defsys_cmd,
+                sinitrd_pfn, einitrd_pfn);
+        }
+#endif
+
+        sprintf(defsys_cmd, "%s EW MINSIZE=%.7iK", defsys_cmd, min_size);
+        sprintf(savesys_cmd, "SAVESYS %s \n IPL %s",
+                kernel_nss_name, kernel_nss_name);
+
+        __cpcmd(defsys_cmd, NULL, 0, &response);
+
+        if (response != 0)
+                return;
+
+        __cpcmd(savesys_cmd, NULL, 0, &response);
+
+        if (response != strlen(savesys_cmd))
+                return;
+
+        ipl_flags = IPL_NSS_VALID;
+}
+
+#else /* CONFIG_SHARED_KERNEL */
+
+static inline void create_kernel_nss(void) { }
+
+#endif /* CONFIG_SHARED_KERNEL */
+
+/*
+ * Clear bss memory
+ */
+static __init void clear_bss_section(void)
+{
+        memset(__bss_start, 0, _end - __bss_start);
+}
+
+/*
+ * Initialize storage key for kernel pages
+ */
+static __init void init_kernel_storage_key(void)
+{
+        unsigned long end_pfn, init_pfn;
+
+        end_pfn = PFN_UP(__pa(&_end));
+
+        for (init_pfn = 0 ; init_pfn < end_pfn; init_pfn++)
+                page_set_storage_key(init_pfn << PAGE_SHIFT, PAGE_DEFAULT_KEY);
+}
+
+static __init void detect_machine_type(void)
+{
+        struct cpuinfo_S390 *cpuinfo = &S390_lowcore.cpu_data;
+
+        asm volatile("stidp %0" : "=m" (S390_lowcore.cpu_data.cpu_id));
+
+        /* Running under z/VM ? */
+        if (cpuinfo->cpu_id.version == 0xff)
+                machine_flags |= 1;
+
+        /* Running on a P/390 ? */
+        if (cpuinfo->cpu_id.machine == 0x7490)
+                machine_flags |= 4;
+}
+
+/*
+ * Save ipl parameters, clear bss memory, initialize storage keys
+ * and create a kernel NSS at startup if the SAVESYS= parm is defined
+ */
+void __init startup_init(void)
+{
+        ipl_save_parameters();
+        clear_bss_section();
+        init_kernel_storage_key();
+        lockdep_init();
+        detect_machine_type();
+        create_kernel_nss();
+}
+
 #ifdef CONFIG_SMP
 void (*_machine_restart)(char *command) = machine_restart_smp;
 void (*_machine_halt)(void) = machine_halt_smp;
@@ -523,7 +659,7 @@ setup_lowcore(void)
 static void __init
 setup_resources(void)
 {
-        struct resource *res;
+        struct resource *res, *sub_res;
         int i;
 
         code_resource.start = (unsigned long) &_text;
@@ -548,8 +684,38 @@ setup_resources(void)
                 res->start = memory_chunk[i].addr;
                 res->end = memory_chunk[i].addr +  memory_chunk[i].size - 1;
                 request_resource(&iomem_resource, res);
-                request_resource(res, &code_resource);
-                request_resource(res, &data_resource);
+
+                if (code_resource.start >= res->start  &&
+                        code_resource.start <= res->end &&
+                        code_resource.end > res->end) {
+                        sub_res = alloc_bootmem_low(sizeof(struct resource));
+                        memcpy(sub_res, &code_resource,
+                                sizeof(struct resource));
+                        sub_res->end = res->end;
+                        code_resource.start = res->end + 1;
+                        request_resource(res, sub_res);
+                }
+
+                if (code_resource.start >= res->start &&
+                        code_resource.start <= res->end &&
+                        code_resource.end <= res->end)
+                        request_resource(res, &code_resource);
+
+                if (data_resource.start >= res->start &&
+                        data_resource.start <= res->end &&
+                        data_resource.end > res->end) {
+                        sub_res = alloc_bootmem_low(sizeof(struct resource));
+                        memcpy(sub_res, &data_resource,
+                                sizeof(struct resource));
+                        sub_res->end = res->end;
+                        data_resource.start = res->end + 1;
+                        request_resource(res, sub_res);
+                }
+
+                if (data_resource.start >= res->start &&
+                        data_resource.start <= res->end &&
+                        data_resource.end <= res->end)
+                        request_resource(res, &data_resource);
         }
 }
 
@@ -585,7 +751,7 @@ static void __init
 setup_memory(void)
 {
         unsigned long bootmap_size;
-        unsigned long start_pfn, end_pfn, init_pfn;
+        unsigned long start_pfn, end_pfn;
         int i;
 
         /*
@@ -595,10 +761,6 @@ setup_memory(void)
         start_pfn = PFN_UP(__pa(&_end));
         end_pfn = max_pfn = PFN_DOWN(memory_end);
 
-        /* Initialize storage key for kernel pages */
-        for (init_pfn = 0 ; init_pfn < start_pfn; init_pfn++)
-                page_set_storage_key(init_pfn << PAGE_SHIFT, PAGE_DEFAULT_KEY);
-
 #ifdef CONFIG_BLK_DEV_INITRD
         /*
          * Move the initrd in case the bitmap of the bootmem allocater
diff --git a/arch/s390/kernel/vmlinux.lds.S b/arch/s390/kernel/vmlinux.lds.S
index fe0f2e97ba7b..8fedb1f9fc97 100644
--- a/arch/s390/kernel/vmlinux.lds.S
+++ b/arch/s390/kernel/vmlinux.lds.S
@@ -31,11 +31,6 @@ SECTIONS
 
   _etext = .;                   /* End of text section */
 
-  . = ALIGN(16);                /* Exception table */
-  __start___ex_table = .;
-  __ex_table : { *(__ex_table) }
-  __stop___ex_table = .;
-
   RODATA
 
 #ifdef CONFIG_SHARED_KERNEL
@@ -44,6 +39,11 @@ SECTIONS
   _eshared = .;                 /* End of shareable data */
 #endif
 
+  . = ALIGN(16);                /* Exception table */
+  __start___ex_table = .;
+  __ex_table : { *(__ex_table) }
+  __stop___ex_table = .;
+
   .data : {                     /* Data */
         *(.data)
         CONSTRUCTORS