[PATCH] MCA/INIT: use per cpu stacks

The bulk of the change.  Use per cpu MCA/INIT stacks.  Change the SAL
to OS state (sos) to be per process.  Do all the assembler work on the
MCA/INIT stacks, leaving the original stack alone.  Pass per cpu state
data to the C handlers for MCA and INIT, which also means changing the
mca_drv interfaces slightly.  Lots of verification on whether the
original stack is usable before converting it to a sleeping process.
Signed-off-by: Keith Owens <kaos@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

arch/ia64/kernel/asm-offsets.c

@@ -211,17 +211,41 @@ void foo(void)
 #endif
 
 	BLANK();
-	DEFINE(IA64_MCA_CPU_PROC_STATE_DUMP_OFFSET,
-	       offsetof (struct ia64_mca_cpu, proc_state_dump));
-	DEFINE(IA64_MCA_CPU_STACK_OFFSET,
-	       offsetof (struct ia64_mca_cpu, stack));
-	DEFINE(IA64_MCA_CPU_STACKFRAME_OFFSET,
-	       offsetof (struct ia64_mca_cpu, stackframe));
-	DEFINE(IA64_MCA_CPU_RBSTORE_OFFSET,
-	       offsetof (struct ia64_mca_cpu, rbstore));
+	DEFINE(IA64_MCA_CPU_MCA_STACK_OFFSET,
+	       offsetof (struct ia64_mca_cpu, mca_stack));
 	DEFINE(IA64_MCA_CPU_INIT_STACK_OFFSET,
 	       offsetof (struct ia64_mca_cpu, init_stack));
 	BLANK();
+	DEFINE(IA64_SAL_OS_STATE_COMMON_OFFSET,
+	       offsetof (struct ia64_sal_os_state, sal_ra));
+	DEFINE(IA64_SAL_OS_STATE_OS_GP_OFFSET,
+	       offsetof (struct ia64_sal_os_state, os_gp));
+	DEFINE(IA64_SAL_OS_STATE_PAL_MIN_STATE_OFFSET,
+	       offsetof (struct ia64_sal_os_state, pal_min_state));
+	DEFINE(IA64_SAL_OS_STATE_PROC_STATE_PARAM_OFFSET,
+	       offsetof (struct ia64_sal_os_state, proc_state_param));
+	DEFINE(IA64_SAL_OS_STATE_SIZE,
+	       sizeof (struct ia64_sal_os_state));
+	DEFINE(IA64_PMSA_GR_OFFSET,
+	       offsetof (struct pal_min_state_area_s, pmsa_gr));
+	DEFINE(IA64_PMSA_BANK1_GR_OFFSET,
+	       offsetof (struct pal_min_state_area_s, pmsa_bank1_gr));
+	DEFINE(IA64_PMSA_PR_OFFSET,
+	       offsetof (struct pal_min_state_area_s, pmsa_pr));
+	DEFINE(IA64_PMSA_BR0_OFFSET,
+	       offsetof (struct pal_min_state_area_s, pmsa_br0));
+	DEFINE(IA64_PMSA_RSC_OFFSET,
+	       offsetof (struct pal_min_state_area_s, pmsa_rsc));
+	DEFINE(IA64_PMSA_IIP_OFFSET,
+	       offsetof (struct pal_min_state_area_s, pmsa_iip));
+	DEFINE(IA64_PMSA_IPSR_OFFSET,
+	       offsetof (struct pal_min_state_area_s, pmsa_ipsr));
+	DEFINE(IA64_PMSA_IFS_OFFSET,
+	       offsetof (struct pal_min_state_area_s, pmsa_ifs));
+	DEFINE(IA64_PMSA_XIP_OFFSET,
+	       offsetof (struct pal_min_state_area_s, pmsa_xip));
+	BLANK();
+
 	/* used by fsys_gettimeofday in arch/ia64/kernel/fsys.S */
 	DEFINE(IA64_TIME_INTERPOLATOR_ADDRESS_OFFSET, offsetof (struct time_interpolator, addr));
 	DEFINE(IA64_TIME_INTERPOLATOR_SOURCE_OFFSET, offsetof (struct time_interpolator, source));

arch/ia64/kernel/mca_drv.c

@@ -4,6 +4,8 @@
  *
  * Copyright (C) 2004 FUJITSU LIMITED
  * Copyright (C) Hidetoshi Seto (seto.hidetoshi@jp.fujitsu.com)
+ * Copyright (C) 2005 Silicon Graphics, Inc
+ * Copyright (C) 2005 Keith Owens <kaos@sgi.com>
  */
 #include <linux/config.h>
 #include <linux/types.h>
@@ -38,10 +40,6 @@
 /* max size of SAL error record (default) */
 static int sal_rec_max = 10000;
 
-/* from mca.c */
-static ia64_mca_sal_to_os_state_t *sal_to_os_handoff_state;
-static ia64_mca_os_to_sal_state_t *os_to_sal_handoff_state;
-
 /* from mca_drv_asm.S */
 extern void *mca_handler_bhhook(void);
 
@@ -316,7 +314,8 @@ init_record_index_pools(void)
  */
 
 static mca_type_t
-is_mca_global(peidx_table_t *peidx, pal_bus_check_info_t *pbci)
+is_mca_global(peidx_table_t *peidx, pal_bus_check_info_t *pbci,
+	      struct ia64_sal_os_state *sos)
 {
 	pal_processor_state_info_t *psp = (pal_processor_state_info_t*)peidx_psp(peidx);
 
@@ -327,7 +326,7 @@ is_mca_global(peidx_table_t *peidx, pal_bus_check_info_t *pbci)
 	 * Therefore it is local MCA when rendezvous has not been requested.
 	 * Failed to rendezvous, the system must be down.
 	 */
-	switch (sal_to_os_handoff_state->imsto_rendez_state) {
+	switch (sos->rv_rc) {
 		case -1: /* SAL rendezvous unsuccessful */
 			return MCA_IS_GLOBAL;
 		case  0: /* SAL rendezvous not required */
@@ -388,7 +387,8 @@ is_mca_global(peidx_table_t *peidx, pal_bus_check_info_t *pbci)
  */
 
 static int
-recover_from_read_error(slidx_table_t *slidx, peidx_table_t *peidx, pal_bus_check_info_t *pbci)
+recover_from_read_error(slidx_table_t *slidx, peidx_table_t *peidx, pal_bus_check_info_t *pbci,
+			struct ia64_sal_os_state *sos)
 {
 	sal_log_mod_error_info_t *smei;
 	pal_min_state_area_t *pmsa;
@@ -426,7 +426,7 @@ recover_from_read_error(slidx_table_t *slidx, peidx_table_t *peidx, pal_bus_chec
 			 *  setup for resume to bottom half of MCA,
 			 * "mca_handler_bhhook"
 			 */
-			pmsa = (pal_min_state_area_t *)(sal_to_os_handoff_state->pal_min_state | (6ul<<61));
+			pmsa = sos->pal_min_state;
 			/* pass to bhhook as 1st argument (gr8) */
 			pmsa->pmsa_gr[8-1] = smei->target_identifier;
 			/* set interrupted return address (but no use) */
@@ -459,7 +459,8 @@ recover_from_read_error(slidx_table_t *slidx, peidx_table_t *peidx, pal_bus_chec
  */
 
 static int
-recover_from_platform_error(slidx_table_t *slidx, peidx_table_t *peidx, pal_bus_check_info_t *pbci)
+recover_from_platform_error(slidx_table_t *slidx, peidx_table_t *peidx, pal_bus_check_info_t *pbci,
+			    struct ia64_sal_os_state *sos)
 {
 	int status = 0;
 	pal_processor_state_info_t *psp = (pal_processor_state_info_t*)peidx_psp(peidx);
@@ -469,7 +470,7 @@ recover_from_platform_error(slidx_table_t *slidx, peidx_table_t *peidx, pal_bus_
 		case 1: /* partial read */
 		case 3: /* full line(cpu) read */
 		case 9: /* I/O space read */
-			status = recover_from_read_error(slidx, peidx, pbci);
+			status = recover_from_read_error(slidx, peidx, pbci, sos);
 			break;
 		case 0: /* unknown */
 		case 2: /* partial write */
@@ -508,7 +509,8 @@ recover_from_platform_error(slidx_table_t *slidx, peidx_table_t *peidx, pal_bus_
  */
 
 static int
-recover_from_processor_error(int platform, slidx_table_t *slidx, peidx_table_t *peidx, pal_bus_check_info_t *pbci)
+recover_from_processor_error(int platform, slidx_table_t *slidx, peidx_table_t *peidx, pal_bus_check_info_t *pbci,
+			     struct ia64_sal_os_state *sos)
 {
 	pal_processor_state_info_t *psp = (pal_processor_state_info_t*)peidx_psp(peidx);
 
@@ -545,7 +547,7 @@ recover_from_processor_error(int platform, slidx_table_t *slidx, peidx_table_t *
 	 * This means "there are some platform errors".
 	 */
 	if (platform) 
-		return recover_from_platform_error(slidx, peidx, pbci);
+		return recover_from_platform_error(slidx, peidx, pbci, sos);
 	/* 
 	 * On account of strange SAL error record, we cannot recover. 
 	 */
@@ -562,8 +564,7 @@ recover_from_processor_error(int platform, slidx_table_t *slidx, peidx_table_t *
 
 static int
 mca_try_to_recover(void *rec, 
-	ia64_mca_sal_to_os_state_t *sal_to_os_state,
-	ia64_mca_os_to_sal_state_t *os_to_sal_state)
+	struct ia64_sal_os_state *sos)
 {
 	int platform_err;
 	int n_proc_err;
@@ -571,10 +572,6 @@ mca_try_to_recover(void *rec,
 	peidx_table_t peidx;
 	pal_bus_check_info_t pbci;
 
-	/* handoff state from/to mca.c */
-	sal_to_os_handoff_state = sal_to_os_state;
-	os_to_sal_handoff_state = os_to_sal_state;
-
 	/* Make index of SAL error record */
 	platform_err = mca_make_slidx(rec, &slidx);
 
@@ -597,11 +594,11 @@ mca_try_to_recover(void *rec,
 	*((u64*)&pbci) = peidx_check_info(&peidx, bus_check, 0);
 
 	/* Check whether MCA is global or not */
-	if (is_mca_global(&peidx, &pbci))
+	if (is_mca_global(&peidx, &pbci, sos))
 		return 0;
 	
 	/* Try to recover a processor error */
-	return recover_from_processor_error(platform_err, &slidx, &peidx, &pbci);
+	return recover_from_processor_error(platform_err, &slidx, &peidx, &pbci, sos);
 }
 
 /*

include/asm-ia64/mca.h

@@ -11,8 +11,6 @@
 #ifndef _ASM_IA64_MCA_H
 #define _ASM_IA64_MCA_H
 
-#define IA64_MCA_STACK_SIZE	8192
-
 #if !defined(__ASSEMBLY__)
 
 #include <linux/interrupt.h>
@@ -48,7 +46,8 @@ typedef union cmcv_reg_u {
 
 enum {
 	IA64_MCA_RENDEZ_CHECKIN_NOTDONE	=	0x0,
-	IA64_MCA_RENDEZ_CHECKIN_DONE	=	0x1
+	IA64_MCA_RENDEZ_CHECKIN_DONE	=	0x1,
+	IA64_MCA_RENDEZ_CHECKIN_INIT	=	0x2,
 };
 
 /* Information maintained by the MC infrastructure */
@@ -63,18 +62,42 @@ typedef struct ia64_mc_info_s {
 
 } ia64_mc_info_t;
 
-typedef struct ia64_mca_sal_to_os_state_s {
-	u64		imsto_os_gp;		/* GP of the os registered with the SAL */
-	u64		imsto_pal_proc;		/* PAL_PROC entry point - physical addr */
-	u64		imsto_sal_proc;		/* SAL_PROC entry point - physical addr */
-	u64		imsto_sal_gp;		/* GP of the SAL - physical */
-	u64		imsto_rendez_state;	/* Rendez state information */
-	u64		imsto_sal_check_ra;	/* Return address in SAL_CHECK while going
-						 * back to SAL from OS after MCA handling.
-						 */
-	u64		pal_min_state;		/* from PAL in r17 */
-	u64		proc_state_param;	/* from PAL in r18. See SDV 2:268 11.3.2.1 */
-} ia64_mca_sal_to_os_state_t;
+/* Handover state from SAL to OS and vice versa, for both MCA and INIT events.
+ * Besides the handover state, it also contains some saved registers from the
+ * time of the event.
+ * Note: mca_asm.S depends on the precise layout of this structure.
+ */
+
+struct ia64_sal_os_state {
+	/* SAL to OS, must be at offset 0 */
+	u64			os_gp;			/* GP of the os registered with the SAL, physical */
+	u64			pal_proc;		/* PAL_PROC entry point, physical */
+	u64			sal_proc;		/* SAL_PROC entry point, physical */
+	u64			rv_rc;			/* MCA - Rendezvous state, INIT - reason code */
+	u64			proc_state_param;	/* from R18 */
+	u64			monarch;		/* 1 for a monarch event, 0 for a slave */
+	/* common, must follow SAL to OS */
+	u64			sal_ra;			/* Return address in SAL, physical */
+	u64			sal_gp;			/* GP of the SAL - physical */
+	pal_min_state_area_t	*pal_min_state;		/* from R17.  physical in asm, virtual in C */
+	u64			prev_IA64_KR_CURRENT;	/* previous value of IA64_KR(CURRENT) */
+	struct task_struct	*prev_task;		/* previous task, NULL if it is not useful */
+	/* Some interrupt registers are not saved in minstate, pt_regs or
+	 * switch_stack.  Because MCA/INIT can occur when interrupts are
+	 * disabled, we need to save the additional interrupt registers over
+	 * MCA/INIT and resume.
+	 */
+	u64			isr;
+	u64			ifa;
+	u64			itir;
+	u64			iipa;
+	u64			iim;
+	u64			iha;
+	/* OS to SAL, must follow common */
+	u64			os_status;		/* OS status to SAL, enum below */
+	u64			context;		/* 0 if return to same context
+							   1 if return to new context */
+};
 
 enum {
 	IA64_MCA_CORRECTED	=	0x0,	/* Error has been corrected by OS_MCA */
@@ -83,36 +106,22 @@ enum {
 	IA64_MCA_HALT		=	-3	/* System to be halted by SAL */
 };
 
+enum {
+	IA64_INIT_RESUME	=	0x0,	/* Resume after return from INIT */
+	IA64_INIT_WARM_BOOT	=	-1,	/* Warm boot of the system need from SAL */
+};
+
 enum {
 	IA64_MCA_SAME_CONTEXT	=	0x0,	/* SAL to return to same context */
 	IA64_MCA_NEW_CONTEXT	=	-1	/* SAL to return to new context */
 };
 
-typedef struct ia64_mca_os_to_sal_state_s {
-	u64		imots_os_status;	/*   OS status to SAL as to what happened
-						 *   with the MCA handling.
-						 */
-	u64		imots_sal_gp;		/* GP of the SAL - physical */
-	u64		imots_context;		/* 0 if return to same context
-						   1 if return to new context */
-	u64		*imots_new_min_state;	/* Pointer to structure containing
-						 * new values of registers in the min state
-						 * save area.
-						 */
-	u64		imots_sal_check_ra;	/* Return address in SAL_CHECK while going
-						 * back to SAL from OS after MCA handling.
-						 */
-} ia64_mca_os_to_sal_state_t;
-
 /* Per-CPU MCA state that is too big for normal per-CPU variables.  */
 
 struct ia64_mca_cpu {
-	u64 stack[IA64_MCA_STACK_SIZE/8];	/* MCA memory-stack */
-	u64 proc_state_dump[512];
-	u64 stackframe[32];
-	u64 rbstore[IA64_MCA_STACK_SIZE/8];	/* MCA reg.-backing store */
+	u64 mca_stack[KERNEL_STACK_SIZE/8];
 	u64 init_stack[KERNEL_STACK_SIZE/8];
-} __attribute__ ((aligned(16)));
+};
 
 /* Array of physical addresses of each CPU's MCA area.  */
 extern unsigned long __per_cpu_mca[NR_CPUS];
@@ -121,12 +130,29 @@ extern void ia64_mca_init(void);
 extern void ia64_mca_cpu_init(void *);
 extern void ia64_os_mca_dispatch(void);
 extern void ia64_os_mca_dispatch_end(void);
-extern void ia64_mca_ucmc_handler(void);
+extern void ia64_mca_ucmc_handler(struct pt_regs *, struct ia64_sal_os_state *);
+extern void ia64_init_handler(struct pt_regs *,
+			      struct switch_stack *,
+			      struct ia64_sal_os_state *);
 extern void ia64_monarch_init_handler(void);
 extern void ia64_slave_init_handler(void);
 extern void ia64_mca_cmc_vector_setup(void);
-extern int  ia64_reg_MCA_extension(void*);
+extern int  ia64_reg_MCA_extension(int (*fn)(void *, struct ia64_sal_os_state *));
 extern void ia64_unreg_MCA_extension(void);
+extern u64 ia64_get_rnat(u64 *);
+
+#else	/* __ASSEMBLY__ */
+
+#define IA64_MCA_CORRECTED	0x0	/* Error has been corrected by OS_MCA */
+#define IA64_MCA_WARM_BOOT	-1	/* Warm boot of the system need from SAL */
+#define IA64_MCA_COLD_BOOT	-2	/* Cold boot of the system need from SAL */
+#define IA64_MCA_HALT		-3	/* System to be halted by SAL */
+
+#define IA64_INIT_RESUME	0x0	/* Resume after return from INIT */
+#define IA64_INIT_WARM_BOOT	-1	/* Warm boot of the system need from SAL */
+
+#define IA64_MCA_SAME_CONTEXT	0x0	/* SAL to return to same context */
+#define IA64_MCA_NEW_CONTEXT	-1	/* SAL to return to new context */
 
 #endif /* !__ASSEMBLY__ */
 #endif /* _ASM_IA64_MCA_H */

include/asm-ia64/mca_asm.h

@@ -8,6 +8,8 @@
  * Copyright (C) 2000 David Mosberger-Tang <davidm@hpl.hp.com>
  * Copyright (C) 2002 Intel Corp.
  * Copyright (C) 2002 Jenna Hall <jenna.s.hall@intel.com>
+ * Copyright (C) 2005 Silicon Graphics, Inc
+ * Copyright (C) 2005 Keith Owens <kaos@sgi.com>
  */
 #ifndef _ASM_IA64_MCA_ASM_H
 #define _ASM_IA64_MCA_ASM_H
@@ -207,106 +209,33 @@
 	;;
 
 /*
- * The following offsets capture the order in which the
- * RSE related registers from the old context are
- * saved onto the new stack frame.
+ * The MCA and INIT stacks in struct ia64_mca_cpu look like normal kernel
+ * stacks, except that the SAL/OS state and a switch_stack are stored near the
+ * top of the MCA/INIT stack.  To support concurrent entry to MCA or INIT, as
+ * well as MCA over INIT, each event needs its own SAL/OS state.  All entries
+ * are 16 byte aligned.
  *
- *	+-----------------------+
- *	|NDIRTY [BSP - BSPSTORE]|
- *	+-----------------------+
- *	|	RNAT		|
- *	+-----------------------+
- *	|	BSPSTORE	|
- *	+-----------------------+
- *	|	IFS		|
- *	+-----------------------+
- *	|	PFS		|
- *	+-----------------------+
- *	|	RSC		|
- *	+-----------------------+ <-------- Bottom of new stack frame
+ *      +---------------------------+
+ *      |          pt_regs          |
+ *      +---------------------------+
+ *      |        switch_stack       |
+ *      +---------------------------+
+ *      |        SAL/OS state       |
+ *      +---------------------------+
+ *      |    16 byte scratch area   |
+ *      +---------------------------+ <-------- SP at start of C MCA handler
+ *      |           .....           |
+ *      +---------------------------+
+ *      | RBS for MCA/INIT handler  |
+ *      +---------------------------+
+ *      | struct task for MCA/INIT  |
+ *      +---------------------------+ <-------- Bottom of MCA/INIT stack
  */
-#define  rse_rsc_offset		0
-#define  rse_pfs_offset		(rse_rsc_offset+0x08)
-#define  rse_ifs_offset		(rse_pfs_offset+0x08)
-#define  rse_bspstore_offset	(rse_ifs_offset+0x08)
-#define  rse_rnat_offset	(rse_bspstore_offset+0x08)
-#define  rse_ndirty_offset	(rse_rnat_offset+0x08)
 
-/*
- * rse_switch_context
- *
- *	1. Save old RSC onto the new stack frame
- *	2. Save PFS onto new stack frame
- *	3. Cover the old frame and start a new frame.
- *	4. Save IFS onto new stack frame
- *	5. Save the old BSPSTORE on the new stack frame
- *	6. Save the old RNAT on the new stack frame
- *	7. Write BSPSTORE with the new backing store pointer
- *	8. Read and save the new BSP to calculate the #dirty registers
- * NOTE: Look at pages 11-10, 11-11 in PRM Vol 2
- */
-#define rse_switch_context(temp,p_stackframe,p_bspstore)			\
-	;;									\
-	mov     temp=ar.rsc;;							\
-	st8     [p_stackframe]=temp,8;;					\
-	mov     temp=ar.pfs;;							\
-	st8     [p_stackframe]=temp,8;						\
-	cover ;;								\
-	mov     temp=cr.ifs;;							\
-	st8     [p_stackframe]=temp,8;;						\
-	mov     temp=ar.bspstore;;						\
-	st8     [p_stackframe]=temp,8;;					\
-	mov     temp=ar.rnat;;							\
-	st8     [p_stackframe]=temp,8;						\
-	mov     ar.bspstore=p_bspstore;;					\
-	mov     temp=ar.bsp;;							\
-	sub     temp=temp,p_bspstore;;						\
-	st8     [p_stackframe]=temp,8;;
-
-/*
- * rse_return_context
- *	1. Allocate a zero-sized frame
- *	2. Store the number of dirty registers RSC.loadrs field
- *	3. Issue a loadrs to insure that any registers from the interrupted
- *	   context which were saved on the new stack frame have been loaded
- *	   back into the stacked registers
- *	4. Restore BSPSTORE
- *	5. Restore RNAT
- *	6. Restore PFS
- *	7. Restore IFS
- *	8. Restore RSC
- *	9. Issue an RFI
- */
-#define rse_return_context(psr_mask_reg,temp,p_stackframe)			\
-	;;									\
-	alloc   temp=ar.pfs,0,0,0,0;						\
-	add     p_stackframe=rse_ndirty_offset,p_stackframe;;			\
-	ld8     temp=[p_stackframe];;						\
-	shl     temp=temp,16;;							\
-	mov     ar.rsc=temp;;							\
-	loadrs;;								\
-	add     p_stackframe=-rse_ndirty_offset+rse_bspstore_offset,p_stackframe;;\
-	ld8     temp=[p_stackframe];;						\
-	mov     ar.bspstore=temp;;						\
-	add     p_stackframe=-rse_bspstore_offset+rse_rnat_offset,p_stackframe;;\
-	ld8     temp=[p_stackframe];;						\
-	mov     ar.rnat=temp;;							\
-	add     p_stackframe=-rse_rnat_offset+rse_pfs_offset,p_stackframe;;	\
-	ld8     temp=[p_stackframe];;						\
-	mov     ar.pfs=temp;;							\
-	add     p_stackframe=-rse_pfs_offset+rse_ifs_offset,p_stackframe;;	\
-	ld8     temp=[p_stackframe];;						\
-	mov     cr.ifs=temp;;							\
-	add     p_stackframe=-rse_ifs_offset+rse_rsc_offset,p_stackframe;;	\
-	ld8     temp=[p_stackframe];;						\
-	mov     ar.rsc=temp ;							\
-	mov     temp=psr;;							\
-	or      temp=temp,psr_mask_reg;;					\
-	mov     cr.ipsr=temp;;							\
-	mov     temp=ip;;							\
-	add     temp=0x30,temp;;						\
-	mov     cr.iip=temp;;							\
-	srlz.i;;								\
-	rfi;;
+#define ALIGN16(x)			((x)&~15)
+#define MCA_PT_REGS_OFFSET		ALIGN16(KERNEL_STACK_SIZE-IA64_PT_REGS_SIZE)
+#define MCA_SWITCH_STACK_OFFSET		ALIGN16(MCA_PT_REGS_OFFSET-IA64_SWITCH_STACK_SIZE)
+#define MCA_SOS_OFFSET			ALIGN16(MCA_SWITCH_STACK_OFFSET-IA64_SAL_OS_STATE_SIZE)
+#define MCA_SP_OFFSET			ALIGN16(MCA_SOS_OFFSET-16)
 
 #endif /* _ASM_IA64_MCA_ASM_H */