s390/kvm: avoid global config of vm.alloc_pgste=1

The system control vm.alloc_pgste is used to control the size of the
page tables, either 2K or 4K. The idea is that a KVM host sets the
vm.alloc_pgste control to 1 which causes *all* new processes to run
with 4K page tables. For a non-kvm system the control should stay off
to save on memory used for page tables.

Trouble is that distributions choose to set the control globally to
be able to run KVM guests. This wastes memory on non-KVM systems.

Introduce the PT_S390_PGSTE ELF segment type to "mark" the qemu
executable with it. All executables with this (empty) segment in
its ELF phdr array will be started with 4K page tables. Any executable
without PT_S390_PGSTE will run with the default 2K page tables.

This removes the need to set vm.alloc_pgste=1 for a KVM host and
minimizes the waste of memory for page tables.

Signed-off-by: Martin Schwidefsky <[e-mail]>

arch/s390/Kconfig

···
 
 config S390
 	def_bool y
+	select ARCH_BINFMT_ELF_STATE
 	select ARCH_HAS_DEVMEM_IS_ALLOWED
 	select ARCH_HAS_ELF_RANDOMIZE
 	select ARCH_HAS_GCOV_PROFILE_ALL

arch/s390/include/asm/elf.h

···
 #define ELF_DATA	ELFDATA2MSB
 #define ELF_ARCH	EM_S390
 
+/* s390 specific phdr types */
+#define PT_S390_PGSTE	0x70000000
+
 /*
  * ELF register definitions..
  */
···
 	(((x)->e_machine == EM_S390 || (x)->e_machine == EM_S390_OLD) \
 	 && (x)->e_ident[EI_CLASS] == ELF_CLASS)
 #define compat_start_thread	start_thread31
+
+struct arch_elf_state {
+	int rc;
+};
+
+#define INIT_ARCH_ELF_STATE { .rc = 0 }
+
+#define arch_check_elf(ehdr, interp, interp_ehdr, state) (0)
+#ifdef CONFIG_PGSTE
+#define arch_elf_pt_proc(ehdr, phdr, elf, interp, state)	\
+({								\
+	struct arch_elf_state *_state = state;			\
+	if ((phdr)->p_type == PT_S390_PGSTE &&			\
+	    !page_table_allocate_pgste &&			\
+	    !test_thread_flag(TIF_PGSTE) &&			\
+	    !current->mm->context.alloc_pgste) {		\
+		set_thread_flag(TIF_PGSTE);			\
+		set_pt_regs_flag(task_pt_regs(current),		\
+				 PIF_SYSCALL_RESTART);		\
+		_state->rc = -EAGAIN;				\
+	}							\
+	_state->rc;						\
+})
+#else
+#define arch_elf_pt_proc(ehdr, phdr, elf, interp, state)	\
+({								\
+	(state)->rc;						\
+})
+#endif
 
 /* For SVR4/S390 the function pointer to be registered with `atexit` is
    passed in R14. */

arch/s390/include/asm/mmu_context.h

···
 	mm->context.gmap_asce = 0;
 	mm->context.flush_mm = 0;
 #ifdef CONFIG_PGSTE
-	mm->context.alloc_pgste = page_table_allocate_pgste;
+	mm->context.alloc_pgste = page_table_allocate_pgste ||
+		test_thread_flag(TIF_PGSTE) ||
+		current->mm->context.alloc_pgste;
 	mm->context.has_pgste = 0;
 	mm->context.use_skey = 0;
 	mm->context.use_cmma = 0;

arch/s390/include/asm/ptrace.h

···
 
 #define PIF_SYSCALL		0	/* inside a system call */
 #define PIF_PER_TRAP		1	/* deliver sigtrap on return to user */
+#define PIF_SYSCALL_RESTART	2	/* restart the current system call */
 
 #define _PIF_SYSCALL		_BITUL(PIF_SYSCALL)
 #define _PIF_PER_TRAP		_BITUL(PIF_PER_TRAP)
+#define _PIF_SYSCALL_RESTART	_BITUL(PIF_SYSCALL_RESTART)
 
 #ifndef __ASSEMBLY__
 

arch/s390/include/asm/thread_info.h

···
 #define TIF_UPROBE		3	/* breakpointed or single-stepping */
 #define TIF_GUARDED_STORAGE	4	/* load guarded storage control block */
 #define TIF_PATCH_PENDING	5	/* pending live patching update */
+#define TIF_PGSTE		6	/* New mm's will use 4K page tables */
 
 #define TIF_31BIT		16	/* 32bit process */
 #define TIF_MEMDIE		17	/* is terminating due to OOM killer */

arch/s390/kernel/entry.S

···
 		   _TIF_SYSCALL_TRACEPOINT)
 _CIF_WORK	= (_CIF_MCCK_PENDING | _CIF_ASCE_PRIMARY | \
 		   _CIF_ASCE_SECONDARY | _CIF_FPU)
-_PIF_WORK	= (_PIF_PER_TRAP)
+_PIF_WORK	= (_PIF_PER_TRAP | _PIF_SYSCALL_RESTART)
 
 #define BASED(name) name-cleanup_critical(%r13)
 
···
 	jo	.Lsysc_mcck_pending
 	TSTMSK	__TI_flags(%r12),_TIF_NEED_RESCHED
 	jo	.Lsysc_reschedule
+	TSTMSK	__PT_FLAGS(%r11),_PIF_SYSCALL_RESTART
+	jo	.Lsysc_syscall_restart
 #ifdef CONFIG_UPROBES
 	TSTMSK	__TI_flags(%r12),_TIF_UPROBE
 	jo	.Lsysc_uprobe_notify
···
 	jo	.Lsysc_patch_pending	# handle live patching just before
 					# signals and possible syscall restart
 #endif
+	TSTMSK	__PT_FLAGS(%r11),_PIF_SYSCALL_RESTART
+	jo	.Lsysc_syscall_restart
 	TSTMSK	__TI_flags(%r12),_TIF_SIGPENDING
 	jo	.Lsysc_sigpending
 	TSTMSK	__TI_flags(%r12),_TIF_NOTIFY_RESUME
···
 	lgr	%r2,%r11		# pass pointer to pt_regs
 	larl	%r14,.Lsysc_return
 	jg	do_per_trap
+
+#
+# _PIF_SYSCALL_RESTART is set, repeat the current system call
+#
+.Lsysc_syscall_restart:
+	ni	__PT_FLAGS+7(%r11),255-_PIF_SYSCALL_RESTART
+	lmg	%r1,%r7,__PT_R1(%r11)	# load svc arguments
+	lg	%r2,__PT_ORIG_GPR2(%r11)
+	j	.Lsysc_do_svc
 
 #
 # call tracehook_report_syscall_entry/tracehook_report_syscall_exit before