[CRIU] [PATCH 2/4] s390: Add guarded-storage support
Alice Frosi
alice at linux.vnet.ibm.com
Thu Sep 28 14:50:10 MSK 2017
Dump and restore tasks with GS control blocks. Guarded-storage is a new
s390 feature to improve garbage collecting languages like Java.
There are two control blocks in the CPU:
- GS control block
- GS broadcast control block
Both control blocks have to be dumped and restored for all threads.
Signed-off-by: Alice Frosi <alice at linux.vnet.ibm.com>
Reviewed-by: Michael Holzheu <holzheu at linux.vnet.ibm.com>
---
.../s390/src/lib/include/uapi/asm/infect-types.h | 6 +
compel/arch/s390/src/lib/infect.c | 102 +++++++-
criu/arch/s390/crtools.c | 267 ++++++++++++++++++---
images/core-s390.proto | 6 +
4 files changed, 340 insertions(+), 41 deletions(-)
diff --git a/compel/arch/s390/src/lib/include/uapi/asm/infect-types.h b/compel/arch/s390/src/lib/include/uapi/asm/infect-types.h
index 84edea5..038c4a0 100644
--- a/compel/arch/s390/src/lib/include/uapi/asm/infect-types.h
+++ b/compel/arch/s390/src/lib/include/uapi/asm/infect-types.h
@@ -38,12 +38,18 @@ struct prfpreg {
};
#define USER_FPREGS_VXRS 0x000000001
+/* Guarded-storage control block */
+#define USER_GS_CB 0x000000002
+/* Guarded-storage broadcast control block */
+#define USER_GS_BC 0x000000004
typedef struct {
uint32_t flags;
struct prfpreg prfpreg;
uint64_t vxrs_low[16];
vector128_t vxrs_high[16];
+ uint64_t gs_cb[4];
+ uint64_t gs_bc[4];
} user_fpregs_struct_t;
typedef struct {
diff --git a/compel/arch/s390/src/lib/infect.c b/compel/arch/s390/src/lib/infect.c
index 3c1fff6..897b66f 100644
--- a/compel/arch/s390/src/lib/infect.c
+++ b/compel/arch/s390/src/lib/infect.c
@@ -19,6 +19,8 @@
#define NT_PRFPREG 2
#define NT_S390_VXRS_LOW 0x309
#define NT_S390_VXRS_HIGH 0x30a
+#define NT_S390_GS_CB 0x30b
+#define NT_S390_GS_BC 0x30c
/*
* Print general purpose and access registers
@@ -40,17 +42,12 @@ static void print_user_regs_struct(const char *msg, int pid,
}
/*
- * Print floating point and vector registers
+ * Print vector registers
*/
-static void print_user_fpregs_struct(const char *msg, int pid,
- user_fpregs_struct_t *fpregs)
+static void print_vxrs(user_fpregs_struct_t *fpregs)
{
int i;
- pr_debug("%s: FP registers for pid=%d\n", msg, pid);
- pr_debug(" fpc %08x\n", fpregs->prfpreg.fpc);
- for (i = 0; i < 16; i++)
- pr_debug(" f%02d %016lx\n", i, fpregs->prfpreg.fprs[i]);
if (!(fpregs->flags & USER_FPREGS_VXRS)) {
pr_debug(" No VXRS\n");
return;
@@ -63,6 +60,53 @@ static void print_user_fpregs_struct(const char *msg, int pid,
fpregs->vxrs_high[i].part2);
}
+/*
+ * Print guarded-storage control block
+ */
+static void print_gs_cb(user_fpregs_struct_t *fpregs)
+{
+ int i;
+
+ if (!(fpregs->flags & USER_GS_CB)) {
+ pr_debug(" No GS_CB\n");
+ return;
+ }
+ for (i = 0; i < 4; i++)
+ pr_debug(" gs_cb%02d %016lx\n", i, fpregs->gs_cb[i]);
+}
+
+/*
+ * Print guarded-storage broadcast control block
+ */
+static void print_gs_bc(user_fpregs_struct_t *fpregs)
+{
+ int i;
+
+ if (!(fpregs->flags & USER_GS_BC)) {
+ pr_debug(" No GS_BC\n");
+ return;
+ }
+ for (i = 0; i < 4; i++)
+ pr_debug(" gs_bc%02d %016lx\n", i, fpregs->gs_bc[i]);
+}
+
+/*
+ * Print FP registers, VX registers, and guarded storage
+ */
+static void print_user_fpregs_struct(const char *msg, int pid,
+ user_fpregs_struct_t *fpregs)
+{
+ int i;
+
+ pr_debug("%s: FP registers for pid=%d\n", msg, pid);
+ pr_debug(" fpc %08x\n", fpregs->prfpreg.fpc);
+ for (i = 0; i < 16; i++)
+ pr_debug(" f%02d %016lx\n", i, fpregs->prfpreg.fprs[i]);
+ print_vxrs(fpregs);
+ print_gs_cb(fpregs);
+ print_gs_bc(fpregs);
+}
+
int sigreturn_prep_regs_plain(struct rt_sigframe *sigframe,
user_regs_struct_t *regs,
user_fpregs_struct_t *fpregs)
@@ -148,7 +192,47 @@ int get_vx_regs(pid_t pid, user_fpregs_struct_t *fpregs)
return 0;
}
+/*
+ * Get guarded-storage control block
+ */
+int get_gs_cb(pid_t pid, user_fpregs_struct_t *fpregs)
+{
+ struct iovec iov;
+ fpregs->flags &= ~(USER_GS_CB | USER_GS_BC);
+ iov.iov_base = &fpregs->gs_cb;
+ iov.iov_len = sizeof(fpregs->gs_cb);
+ if (ptrace(PTRACE_GETREGSET, pid, NT_S390_GS_CB, &iov) < 0) {
+ switch (errno) {
+ case EINVAL:
+ case ENODEV:
+ memset(&fpregs->gs_cb, 0, sizeof(fpregs->gs_cb));
+ memset(&fpregs->gs_bc, 0, sizeof(fpregs->gs_bc));
+ pr_debug("GS_CB not supported\n");
+ return 0;
+ case ENODATA:
+ pr_debug("GS_CB not set\n");
+ break;
+ default:
+ return -1;
+ }
+ } else {
+ fpregs->flags |= USER_GS_CB;
+ }
+ iov.iov_base = &fpregs->gs_bc;
+ iov.iov_len = sizeof(fpregs->gs_bc);
+ if (ptrace(PTRACE_GETREGSET, pid, NT_S390_GS_BC, &iov) < 0) {
+ if (errno == ENODATA) {
+ pr_debug("GS_BC not set\n");
+ return 0;
+ }
+ pr_perror("Couldn't get GS_BC\n");
+ return -1;
+ }
+ fpregs->flags |= USER_GS_BC;
+
+ return 0;
+}
/*
* Prepare task registers for restart
*/
@@ -172,6 +256,10 @@ int get_task_regs(pid_t pid, user_regs_struct_t *regs, save_regs_t save,
pr_perror("Couldn't get vector registers");
return -1;
}
+ if (get_gs_cb(pid, &fpregs)) {
+ pr_perror("Couldn't get guarded-storage");
+ return -1;
+ }
print_user_fpregs_struct("get_task_regs", pid, &fpregs);
/* Check for system call restarting. */
if (regs->system_call) {
diff --git a/criu/arch/s390/crtools.c b/criu/arch/s390/crtools.c
index cfab508..dd801fa 100644
--- a/criu/arch/s390/crtools.c
+++ b/criu/arch/s390/crtools.c
@@ -24,10 +24,13 @@
#include "images/creds.pb-c.h"
#include "ptrace.h"
#include "pstree.h"
+#include "image.h"
#define NT_PRFPREG 2
#define NT_S390_VXRS_LOW 0x309
#define NT_S390_VXRS_HIGH 0x30a
+#define NT_S390_GS_CB 0x30b
+#define NT_S390_GS_BC 0x30c
/*
* Print general purpose and access registers
@@ -47,24 +50,18 @@ static void print_core_gpregs(const char *msg, UserS390RegsEntry *gpregs)
}
/*
- * Print floating point and vector registers
+ * Print vector registers
*/
-static void print_core_fp_regs(const char *msg, CoreEntry *core)
+static void print_core_vx_regs(CoreEntry *core)
{
UserS390VxrsHighEntry *vxrs_high;
UserS390VxrsLowEntry *vxrs_low;
- UserS390FpregsEntry *fpregs;
int i;
vxrs_high = CORE_THREAD_ARCH_INFO(core)->vxrs_high;
vxrs_low = CORE_THREAD_ARCH_INFO(core)->vxrs_low;
- fpregs = CORE_THREAD_ARCH_INFO(core)->fpregs;
- pr_debug("%s: Floating point registers\n", msg);
- pr_debug(" fpc %08x\n", fpregs->fpc);
- for (i = 0; i < 16; i++)
- pr_debug(" f%02d %016lx\n", i, fpregs->fprs[i]);
- if (!vxrs_low) {
+ if (vxrs_low == NULL) {
pr_debug(" No VXRS\n");
return;
}
@@ -76,6 +73,60 @@ static void print_core_fp_regs(const char *msg, CoreEntry *core)
}
/*
+ * Print guarded-storage control block
+ */
+static void print_core_gs_cb(CoreEntry *core)
+{
+ UserS390GsCbEntry *gs_cb;
+ int i;
+
+ gs_cb = CORE_THREAD_ARCH_INFO(core)->gs_cb;
+ if (!gs_cb) {
+ pr_debug(" No GS_CB\n");
+ return;
+ }
+ for (i = 0; i < 4; i++)
+ pr_debug(" gs_cb%d %lx\n", i, gs_cb->regs[i]);
+}
+
+/*
+ * Print guarded-storage broadcast control block
+ */
+static void print_core_gs_bc(CoreEntry *core)
+{
+ UserS390GsCbEntry *gs_bc;
+ int i;
+
+ gs_bc = CORE_THREAD_ARCH_INFO(core)->gs_bc;
+
+ if (!gs_bc) {
+ pr_debug(" No GS_BC\n");
+ return;
+ }
+ for (i = 0; i < 4; i++)
+ pr_debug(" gs_bc%d %lx\n", i, gs_bc->regs[i]);
+}
+
+/*
+ * Print architecture registers
+ */
+static void print_core_fp_regs(const char *msg, CoreEntry *core)
+{
+ UserS390FpregsEntry *fpregs;
+ int i;
+
+ fpregs = CORE_THREAD_ARCH_INFO(core)->fpregs;
+
+ pr_debug("%s: Floating point registers\n", msg);
+ pr_debug(" fpc %08x\n", fpregs->fpc);
+ for (i = 0; i < 16; i++)
+ pr_debug(" f%02d %016lx\n", i, fpregs->fprs[i]);
+ print_core_vx_regs(core);
+ print_core_gs_cb(core);
+ print_core_gs_bc(core);
+}
+
+/*
* Allocate VxrsLow registers
*/
static UserS390VxrsLowEntry *allocate_vxrs_low_regs(void)
@@ -144,14 +195,49 @@ static void free_vxrs_high_regs(UserS390VxrsHighEntry *vxrs_high)
}
/*
+ * Allocate guarded-storage control block (GS_CB and GS_BC)
+ */
+static UserS390GsCbEntry *allocate_gs_cb(void)
+{
+ UserS390GsCbEntry *gs_cb;
+
+ gs_cb = xmalloc(sizeof(*gs_cb));
+ if (!gs_cb)
+ return NULL;
+ user_s390_gs_cb_entry__init(gs_cb);
+
+ gs_cb->n_regs = 4;
+ gs_cb->regs = xzalloc(4 * sizeof(uint64_t));
+ if (!gs_cb->regs)
+ goto fail_free_gs_cb;
+ return gs_cb;
+
+fail_free_gs_cb:
+ xfree(gs_cb);
+ return NULL;
+}
+
+/*
+ * Free Guareded Storage control blocks
+ */
+static void free_gs_cb(UserS390GsCbEntry *gs_cb)
+{
+ if (gs_cb) {
+ xfree(gs_cb->regs);
+ xfree(gs_cb);
+ }
+}
+/*
* Copy internal structures into Google Protocol Buffers
*/
int save_task_regs(void *arg, user_regs_struct_t *u, user_fpregs_struct_t *f)
{
- UserS390VxrsHighEntry *vxrs_high;
- UserS390VxrsLowEntry *vxrs_low;
- UserS390FpregsEntry *fpregs;
- UserS390RegsEntry *gpregs;
+ UserS390VxrsHighEntry *vxrs_high = NULL;
+ UserS390VxrsLowEntry *vxrs_low = NULL;
+ UserS390FpregsEntry *fpregs = NULL;
+ UserS390RegsEntry *gpregs = NULL;
+ UserS390GsCbEntry *gs_cb = NULL;
+ UserS390GsCbEntry *gs_bc = NULL;
CoreEntry *core = arg;
gpregs = CORE_THREAD_ARCH_INFO(core)->gpregs;
@@ -163,15 +249,29 @@ int save_task_regs(void *arg, user_regs_struct_t *u, user_fpregs_struct_t *f)
if (!vxrs_low)
return -1;
vxrs_high = allocate_vxrs_high_regs();
- if (!vxrs_high) {
- free_vxrs_low_regs(vxrs_low);
- return -1;
- }
+ if (!vxrs_high)
+ goto fail_free_vxrs_low;
memcpy(vxrs_low->regs, &f->vxrs_low, sizeof(f->vxrs_low));
memcpy(vxrs_high->regs, &f->vxrs_high, sizeof(f->vxrs_high));
CORE_THREAD_ARCH_INFO(core)->vxrs_low = vxrs_low;
CORE_THREAD_ARCH_INFO(core)->vxrs_high = vxrs_high;
}
+ /* Guarded-storage control block */
+ if (f->flags & USER_GS_CB) {
+ gs_cb = allocate_gs_cb();
+ if (!gs_cb)
+ goto fail_free_gs_cb;
+ memcpy(gs_cb->regs, &f->gs_cb, sizeof(f->gs_cb));
+ CORE_THREAD_ARCH_INFO(core)->gs_cb = gs_cb;
+ }
+ /* Guarded-storage broadcast control block */
+ if (f->flags & USER_GS_BC) {
+ gs_bc = allocate_gs_cb();
+ if (!gs_bc)
+ goto fail_free_gs_bc;
+ memcpy(gs_bc->regs, &f->gs_bc, sizeof(f->gs_bc));
+ CORE_THREAD_ARCH_INFO(core)->gs_bc = gs_bc;
+ }
/* General purpose registers */
memcpy(gpregs->gprs, u->prstatus.gprs, sizeof(u->prstatus.gprs));
gpregs->psw_mask = u->prstatus.psw.mask;
@@ -184,6 +284,13 @@ int save_task_regs(void *arg, user_regs_struct_t *u, user_fpregs_struct_t *f)
fpregs->fpc = f->prfpreg.fpc;
memcpy(fpregs->fprs, f->prfpreg.fprs, sizeof(f->prfpreg.fprs));
return 0;
+fail_free_gs_cb:
+ free_gs_cb(gs_cb);
+fail_free_gs_bc:
+ free_gs_cb(gs_bc);
+fail_free_vxrs_low:
+ free_vxrs_low_regs(vxrs_low);
+ return -1;
}
/*
@@ -225,7 +332,6 @@ int restore_fpu(struct rt_sigframe *f, CoreEntry *core)
memcpy(&dst_ext->vxrs_high, vxrs_high->regs,
sizeof(dst_ext->vxrs_high));
}
- print_core_fp_regs("restore_fp_regs", core);
return 0;
}
@@ -342,6 +448,8 @@ void arch_free_thread_info(CoreEntry *core)
free_fp_regs(CORE_THREAD_ARCH_INFO(core)->fpregs);
free_vxrs_low_regs(CORE_THREAD_ARCH_INFO(core)->vxrs_low);
free_vxrs_high_regs(CORE_THREAD_ARCH_INFO(core)->vxrs_high);
+ free_gs_cb(CORE_THREAD_ARCH_INFO(core)->gs_cb);
+ free_gs_cb(CORE_THREAD_ARCH_INFO(core)->gs_bc);
xfree(CORE_THREAD_ARCH_INFO(core));
CORE_THREAD_ARCH_INFO(core) = NULL;
}
@@ -390,6 +498,52 @@ static int set_vx_regs(pid_t pid, user_fpregs_struct_t *fpregs)
}
/*
+ * Set guarded-storage control block
+ */
+static int set_gs_cb(pid_t pid, user_fpregs_struct_t *fpregs)
+{
+ struct iovec iov;
+
+ if (fpregs->flags & USER_GS_CB) {
+ iov.iov_base = &fpregs->gs_cb;
+ iov.iov_len = sizeof(fpregs->gs_cb);
+ if (setregset(pid, NT_S390_GS_CB, "S390_GS_CB", &iov))
+ return -1;
+ }
+
+ if (!(fpregs->flags & USER_GS_BC))
+ return 0;
+ iov.iov_base = &fpregs->gs_bc;
+ iov.iov_len = sizeof(fpregs->gs_bc);
+ return setregset(pid, NT_S390_GS_BC, "S390_GS_BC", &iov);
+}
+
+/*
+ * Restore registers not present in sigreturn signal frame
+ */
+static int set_task_regs_nosigrt(pid_t pid, CoreEntry *core)
+{
+ user_fpregs_struct_t fpregs;
+ UserS390GsCbEntry *cgs_cb;
+ UserS390GsCbEntry *cgs_bc;
+
+ memset(&fpregs, 0, sizeof(fpregs));
+ /* Guarded-storage control block (optional) */
+ cgs_cb = CORE_THREAD_ARCH_INFO(core)->gs_cb;
+ if (cgs_cb != NULL) {
+ fpregs.flags |= USER_GS_CB;
+ memcpy(&fpregs.gs_cb, cgs_cb->regs, sizeof(fpregs.gs_cb));
+ }
+ /* Guarded-storage broadcast control block (optional) */
+ cgs_bc = CORE_THREAD_ARCH_INFO(core)->gs_bc;
+ if (cgs_bc != NULL) {
+ fpregs.flags |= USER_GS_BC;
+ memcpy(&fpregs.gs_bc, cgs_bc->regs, sizeof(fpregs.gs_bc));
+ }
+ return set_gs_cb(pid, &fpregs);
+}
+
+/*
* Restore registers for pid from core
*/
static int set_task_regs(pid_t pid, CoreEntry *core)
@@ -410,34 +564,39 @@ static int set_task_regs(pid_t pid, CoreEntry *core)
return -1;
/* Vector registers (optional) */
cvxrs_low = CORE_THREAD_ARCH_INFO(core)->vxrs_low;
- if (!cvxrs_low)
- return 0;
- cvxrs_high = CORE_THREAD_ARCH_INFO(core)->vxrs_high;
- if (!cvxrs_high)
- return -1;
- fpregs.flags |= USER_FPREGS_VXRS;
- memcpy(&fpregs.vxrs_low, cvxrs_low->regs, sizeof(fpregs.vxrs_low));
- memcpy(&fpregs.vxrs_high, cvxrs_high->regs, sizeof(fpregs.vxrs_high));
-
- return set_vx_regs(pid, &fpregs);
+ if (cvxrs_low != NULL) {
+ cvxrs_high = CORE_THREAD_ARCH_INFO(core)->vxrs_high;
+ if (!cvxrs_high)
+ return -1;
+ fpregs.flags |= USER_FPREGS_VXRS;
+ memcpy(&fpregs.vxrs_low, cvxrs_low->regs,
+ sizeof(fpregs.vxrs_low));
+ memcpy(&fpregs.vxrs_high, cvxrs_high->regs,
+ sizeof(fpregs.vxrs_high));
+ if (set_vx_regs(pid, &fpregs) < 0)
+ return -1;
+ }
+ return set_task_regs_nosigrt(pid, core);
}
/*
- * Restore vector and floating point registers for all threads
+ * Restore registers for all threads:
+ * - Floating point registers
+ * - Vector registers
+ * - Guarded-storage control block
+ * - Guarded-storage broadcast control block
*/
-int arch_set_thread_regs(struct pstree_item *item)
+int arch_set_thread_regs(struct pstree_item *item, bool with_threads)
{
int i;
for_each_pstree_item(item) {
- if (item->pid->state == TASK_DEAD ||
- item->pid->state == TASK_ZOMBIE ||
- item->pid->state == TASK_HELPER)
- continue;
for (i = 0; i < item->nr_threads; i++) {
if (item->threads[i]->state == TASK_DEAD ||
item->threads[i]->state == TASK_ZOMBIE)
continue;
+ if (!with_threads && i > 0)
+ continue;
if (set_task_regs(item->threads[i]->real,
item->core[i])) {
pr_perror("Not set registers for task %d",
@@ -448,3 +607,43 @@ int arch_set_thread_regs(struct pstree_item *item)
}
return 0;
}
+
+static int open_core(int pid, CoreEntry **pcore)
+{
+ struct cr_img *img;
+ int ret;
+
+ img = open_image(CR_FD_CORE, O_RSTR, pid);
+ if (!img) {
+ pr_err("Can't open core data for %d\n", pid);
+ return -1;
+ }
+ ret = pb_read_one(img, pcore, PB_CORE);
+ close_image(img);
+
+ return ret <= 0 ? -1 : 0;
+}
+
+/*
+ * Restore all registers not present in sigreturn signal frame
+ *
+ * - Guarded-storage control block
+ * - Guarded-storage broadcast control block
+ */
+int arch_set_thread_regs_nosigrt(struct pid *pid)
+{
+ CoreEntry *core;
+
+ core = xmalloc(sizeof(*core));
+ if (open_core(pid->ns[0].virt, &core) < 0) {
+ pr_perror("Cannot open core for virt pid %d", pid->ns[0].virt);
+ return -1;
+ }
+
+ if (set_task_regs_nosigrt(pid->real, core) < 0) {
+ pr_perror("Set register for pid %d", pid->real);
+ return -1;
+ }
+ print_core_fp_regs("restore_fp_regs", core);
+ return 0;
+}
diff --git a/images/core-s390.proto b/images/core-s390.proto
index 78d3e14..3b587fb 100644
--- a/images/core-s390.proto
+++ b/images/core-s390.proto
@@ -30,10 +30,16 @@ message user_s390_fpregs_entry {
repeated uint64 fprs = 2;
}
+message user_s390_gs_cb_entry {
+ repeated uint64 regs = 1;
+}
+
message thread_info_s390 {
required uint64 clear_tid_addr = 1[(criu).hex = true];
required user_s390_regs_entry gpregs = 2[(criu).hex = true];
required user_s390_fpregs_entry fpregs = 3[(criu).hex = true];
optional user_s390_vxrs_low_entry vxrs_low = 4[(criu).hex = true];
optional user_s390_vxrs_high_entry vxrs_high = 5[(criu).hex = true];
+ optional user_s390_gs_cb_entry gs_cb = 6[(criu).hex = true];
+ optional user_s390_gs_cb_entry gs_bc = 7[(criu).hex = true];
}
--
2.9.3
More information about the CRIU
mailing list