From: He Sheng hesheng@wxiat.com

Sunway inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I56OLG

--------------------------------

These header files have became unused for a long time.

Signed-off-by: He Sheng hesheng@wxiat.com

Signed-off-by: Gu Zitao guzitao@wxiat.com --- arch/sw_64/include/asm/agp.h | 19 ---- arch/sw_64/include/asm/asm-prototypes.h | 1 - arch/sw_64/include/asm/compiler.h | 7 -- arch/sw_64/include/asm/console.h | 11 --- arch/sw_64/include/asm/elf.h | 6 +- arch/sw_64/include/asm/floppy.h | 116 ------------------------ arch/sw_64/include/asm/special_insns.h | 20 ---- arch/sw_64/include/uapi/asm/console.h | 51 ----------- 8 files changed, 2 insertions(+), 229 deletions(-) delete mode 100644 arch/sw_64/include/asm/agp.h delete mode 100644 arch/sw_64/include/asm/compiler.h delete mode 100644 arch/sw_64/include/asm/console.h delete mode 100644 arch/sw_64/include/asm/floppy.h delete mode 100644 arch/sw_64/include/asm/special_insns.h delete mode 100644 arch/sw_64/include/uapi/asm/console.h

diff --git a/arch/sw_64/include/asm/agp.h b/arch/sw_64/include/asm/agp.h deleted file mode 100644 index e9d16888910e..000000000000 --- a/arch/sw_64/include/asm/agp.h +++ /dev/null @@ -1,19 +0,0 @@ -/* SPDX-License-Identifier: GPL-2.0 */ -#ifndef _ASM_SW64_AGP_H -#define _ASM_SW64_AGP_H 1 - -#include <asm/io.h> - -/* dummy for now */ - -#define map_page_into_agp(page) -#define unmap_page_from_agp(page) -#define flush_agp_cache() mb() - -/* GATT allocation. Returns/accepts GATT kernel virtual address. */ -#define alloc_gatt_pages(order) \ - ((char *)__get_free_pages(GFP_KERNEL, (order))) -#define free_gatt_pages(table, order) \ - free_pages((unsigned long)(table), (order)) - -#endif diff --git a/arch/sw_64/include/asm/asm-prototypes.h b/arch/sw_64/include/asm/asm-prototypes.h index 21f4f494d74d..15bad8ef6883 100644 --- a/arch/sw_64/include/asm/asm-prototypes.h +++ b/arch/sw_64/include/asm/asm-prototypes.h @@ -4,7 +4,6 @@

#include <linux/spinlock.h> #include <asm/checksum.h> -#include <asm/console.h> #include <asm/page.h> #include <asm/string.h> #include <linux/uaccess.h> diff --git a/arch/sw_64/include/asm/compiler.h b/arch/sw_64/include/asm/compiler.h deleted file mode 100644 index 9a80aa6a0ba8..000000000000 --- a/arch/sw_64/include/asm/compiler.h +++ /dev/null @@ -1,7 +0,0 @@ -/* SPDX-License-Identifier: GPL-2.0 */ -#ifndef _ASM_SW64_COMPILER_H -#define _ASM_SW64_COMPILER_H - -#include <uapi/asm/compiler.h> - -#endif /* _ASM_SW64_COMPILER_H */ diff --git a/arch/sw_64/include/asm/console.h b/arch/sw_64/include/asm/console.h deleted file mode 100644 index 0c01cb740bce..000000000000 --- a/arch/sw_64/include/asm/console.h +++ /dev/null @@ -1,11 +0,0 @@ -/* SPDX-License-Identifier: GPL-2.0 */ -#ifndef _ASM_SW64_CONSOLE_H -#define _ASM_SW64_CONSOLE_H - -#include <uapi/asm/console.h> -#ifndef __ASSEMBLY__ -struct crb_struct; -extern int callback_init_done; -extern void callback_init(void); -#endif /* __ASSEMBLY__ */ -#endif /* _ASM_SW64_CONSOLE_H */ diff --git a/arch/sw_64/include/asm/elf.h b/arch/sw_64/include/asm/elf.h index 150629b0b615..abecc5653eba 100644 --- a/arch/sw_64/include/asm/elf.h +++ b/arch/sw_64/include/asm/elf.h @@ -3,7 +3,6 @@ #define _ASM_SW64_ELF_H #ifdef __KERNEL__ #include <asm/auxvec.h> -#include <asm/special_insns.h> #endif /* Special values for the st_other field in the symbol table. */

@@ -141,11 +140,10 @@ extern int dump_elf_task_fp(elf_fpreg_t *dest, struct task_struct *task);

/* * This yields a mask that user programs can use to figure out what - * instruction set this CPU supports. This is trivial on SW-64, - * but not so on other machines. + * instruction set this CPU supports. */

-#define ELF_HWCAP (~amask(-1)) +#define ELF_HWCAP 0

/* * This yields a string that ld.so will use to load implementation diff --git a/arch/sw_64/include/asm/floppy.h b/arch/sw_64/include/asm/floppy.h deleted file mode 100644 index f4646d99d80c..000000000000 --- a/arch/sw_64/include/asm/floppy.h +++ /dev/null @@ -1,116 +0,0 @@ -/* SPDX-License-Identifier: GPL-2.0 */ -/* - * Architecture specific parts of the Floppy driver - * - * This file is subject to the terms and conditions of the GNU General Public - * License. See the file "COPYING" in the main directory of this archive - * for more details. - * - * Copyright (C) 1995 - */ -#ifndef _ASM_SW64_FLOPPY_H -#define _ASM_SW64_FLOPPY_H - -#define fd_inb(port) inb_p(port) -#define fd_outb(value, port) outb_p(value, port) - -#define fd_enable_dma() enable_dma(FLOPPY_DMA) -#define fd_disable_dma() disable_dma(FLOPPY_DMA) -#define fd_request_dma() request_dma(FLOPPY_DMA, "floppy") -#define fd_free_dma() free_dma(FLOPPY_DMA) -#define fd_clear_dma_ff() clear_dma_ff(FLOPPY_DMA) -#define fd_set_dma_mode(mode) set_dma_mode(FLOPPY_DMA, mode) -#define fd_set_dma_addr(addr) set_dma_addr(FLOPPY_DMA, virt_to_bus(addr)) -#define fd_set_dma_count(count) set_dma_count(FLOPPY_DMA, count) -#define fd_enable_irq() enable_irq(FLOPPY_IRQ) -#define fd_disable_irq() disable_irq(FLOPPY_IRQ) -#define fd_cacheflush(addr, size) /* nothing */ -#define fd_request_irq() \ - request_irq(FLOPPY_IRQ, floppy_interrupt, 0, "floppy", NULL) -#define fd_free_irq() free_irq(FLOPPY_IRQ, NULL) - -#ifdef CONFIG_PCI - -#include <linux/pci.h> - -#define fd_dma_setup(addr, size, mode, io) \ - sw64_fd_dma_setup(addr, size, mode, io) - -static inline int -sw64_fd_dma_setup(char *addr, unsigned long size, int mode, int io) -{ - static unsigned long prev_size; - static dma_addr_t bus_addr; - static char *prev_addr; - static int prev_dir; - int dir; - - dir = (mode != DMA_MODE_READ) ? PCI_DMA_FROMDEVICE : PCI_DMA_TODEVICE; - - if (bus_addr - && (addr != prev_addr || size != prev_size || dir != prev_dir)) { - /* different from last time -- unmap prev */ - bus_addr = 0; - } - - if (!bus_addr) /* need to map it */ - bus_addr = virt_to_bus(addr); - - /* remember this one as prev */ - prev_addr = addr; - prev_size = size; - prev_dir = dir; - - fd_clear_dma_ff(); - fd_cacheflush(addr, size); - fd_set_dma_mode(mode); - set_dma_addr(FLOPPY_DMA, bus_addr); - fd_set_dma_count(size); - virtual_dma_port = io; - fd_enable_dma(); - - return 0; -} - -#endif /* CONFIG_PCI */ - -inline void virtual_dma_init(void) -{ - /* Nothing to do on an sw64 */ -} - -static int FDC1 = 0x3f0; -static int FDC2 = -1; - -/* - * Again, the CMOS information doesn't work on the sw64.. - */ -#define FLOPPY0_TYPE 6 -#define FLOPPY1_TYPE 0 - -#define N_FDC 2 -#define N_DRIVE 8 - -/* - * Most sw64s have no problems with floppy DMA crossing 64k borders, - * except for certain ones, like XL and RUFFIAN. - * - * However, the test is simple and fast, and this *is* floppy, after all, - * so we do it for all platforms, just to make sure. - * - * This is advantageous in other circumstances as well, as in moving - * about the PCI DMA windows and forcing the floppy to start doing - * scatter-gather when it never had before, and there *is* a problem - * on that platform... ;-} - */ - -static inline unsigned long CROSS_64KB(void *a, unsigned long s) -{ - unsigned long p = (unsigned long)a; - - return ((p + s - 1) ^ p) & ~0xffffUL; -} - -#define EXTRA_FLOPPY_PARAMS - -#endif /* __ASM_SW64_FLOPPY_H */ diff --git a/arch/sw_64/include/asm/special_insns.h b/arch/sw_64/include/asm/special_insns.h deleted file mode 100644 index 7f5a52b20444..000000000000 --- a/arch/sw_64/include/asm/special_insns.h +++ /dev/null @@ -1,20 +0,0 @@ -/* SPDX-License-Identifier: GPL-2.0 */ -#ifndef _ASM_SW64_SPECIAL_INSNS_H -#define _ASM_SW64_SPECIAL_INSNS_H - -enum amask_enum { - AMASK_BWX = (1UL << 0), - AMASK_FIX = (1UL << 1), - AMASK_CIX = (1UL << 2), - AMASK_MAX = (1UL << 8), - AMASK_PRECISE_TRAP = (1UL << 9), -}; - -#define amask(mask) \ -({ \ - unsigned long __amask, __input = (mask); \ - __asm__ ("mov %1, %0" : "=r"(__amask) : "rI"(__input)); \ - __amask; \ -}) - -#endif /* _ASM_SW64_SPECIAL_INSNS_H */ diff --git a/arch/sw_64/include/uapi/asm/console.h b/arch/sw_64/include/uapi/asm/console.h deleted file mode 100644 index a40cd7aeb31f..000000000000 --- a/arch/sw_64/include/uapi/asm/console.h +++ /dev/null @@ -1,51 +0,0 @@ -/* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */ -#ifndef _UAPI_ASM_SW64_CONSOLE_H -#define _UAPI_ASM_SW64_CONSOLE_H - -/* - * Console callback routine numbers - */ -#define CCB_GETC 0x01 -#define CCB_PUTS 0x02 -#define CCB_RESET_TERM 0x03 -#define CCB_SET_TERM_INT 0x04 -#define CCB_SET_TERM_CTL 0x05 -#define CCB_PROCESS_KEYCODE 0x06 -#define CCB_OPEN_CONSOLE 0x07 -#define CCB_CLOSE_CONSOLE 0x08 - -#define CCB_OPEN 0x10 -#define CCB_CLOSE 0x11 -#define CCB_IOCTL 0x12 -#define CCB_READ 0x13 -#define CCB_WRITE 0x14 - -#define CCB_SET_ENV 0x20 -#define CCB_RESET_ENV 0x21 -#define CCB_GET_ENV 0x22 -#define CCB_SAVE_ENV 0x23 - -#define CCB_PSWITCH 0x30 -#define CCB_BIOS_EMUL 0x32 - -/* - * Environment variable numbers - */ -#define ENV_AUTO_ACTION 0x01 -#define ENV_BOOT_DEV 0x02 -#define ENV_BOOTDEF_DEV 0x03 -#define ENV_BOOTED_DEV 0x04 -#define ENV_BOOT_FILE 0x05 -#define ENV_BOOTED_FILE 0x06 -#define ENV_BOOT_OSFLAGS 0x07 -#define ENV_BOOTED_OSFLAGS 0x08 -#define ENV_BOOT_RESET 0x09 -#define ENV_DUMP_DEV 0x0A -#define ENV_ENABLE_AUDIT 0x0B -#define ENV_LICENSE 0x0C -#define ENV_CHAR_SET 0x0D -#define ENV_LANGUAGE 0x0E -#define ENV_TTY_DEV 0x0F - - -#endif /* _UAPI_ASM_SW64_CONSOLE_H */

From: He Sheng hesheng@wxiat.com

Sunway inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I56OLG

--------------------------------

Most macro definitions are the same with generic version.

Signed-off-by: He Sheng hesheng@wxiat.com

Signed-off-by: Gu Zitao guzitao@wxiat.com --- arch/sw_64/include/uapi/asm/param.h | 9 +-------- 1 file changed, 1 insertion(+), 8 deletions(-)

diff --git a/arch/sw_64/include/uapi/asm/param.h b/arch/sw_64/include/uapi/asm/param.h index 16c4934c937e..d38e8202dd97 100644 --- a/arch/sw_64/include/uapi/asm/param.h +++ b/arch/sw_64/include/uapi/asm/param.h @@ -2,15 +2,8 @@ #ifndef _UAPI_ASM_SW64_PARAM_H #define _UAPI_ASM_SW64_PARAM_H

-#define HZ 100 - #define EXEC_PAGESIZE 8192

-#ifndef NOGROUP -#define NOGROUP (-1) -#endif - -#define MAXHOSTNAMELEN 64 /* max length of hostname */ - +#include <asm-generic/param.h>

#endif /* _UAPI_ASM_SW64_PARAM_H */

From: He Sheng hesheng@wxiat.com

Sunway inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I56OLG

--------------------------------

These kapi headers are proved to be the same with generic version.

Signed-off-by: He Sheng hesheng@wxiat.com

Signed-off-by: Gu Zitao guzitao@wxiat.com --- arch/sw_64/include/asm/Kbuild | 7 +- arch/sw_64/include/asm/div64.h | 7 -- arch/sw_64/include/asm/emergency-restart.h | 7 -- arch/sw_64/include/asm/exec.h | 7 -- arch/sw_64/include/asm/irq_regs.h | 7 -- arch/sw_64/include/asm/kmap_types.h | 15 --- arch/sw_64/include/asm/local.h | 125 --------------------- arch/sw_64/include/asm/local64.h | 7 -- arch/sw_64/include/asm/param.h | 11 -- arch/sw_64/include/asm/parport.h | 19 ---- arch/sw_64/include/asm/preempt.h | 7 -- arch/sw_64/include/asm/seccomp.h | 15 --- arch/sw_64/include/asm/sections.h | 8 -- arch/sw_64/include/asm/segment.h | 7 -- arch/sw_64/include/asm/serial.h | 16 --- arch/sw_64/include/asm/shmparam.h | 7 -- arch/sw_64/include/asm/trace_clock.h | 10 -- arch/sw_64/include/asm/types.h | 7 -- arch/sw_64/include/asm/unaligned.h | 12 -- 19 files changed, 6 insertions(+), 295 deletions(-) delete mode 100644 arch/sw_64/include/asm/div64.h delete mode 100644 arch/sw_64/include/asm/emergency-restart.h delete mode 100644 arch/sw_64/include/asm/exec.h delete mode 100644 arch/sw_64/include/asm/irq_regs.h delete mode 100644 arch/sw_64/include/asm/kmap_types.h delete mode 100644 arch/sw_64/include/asm/local.h delete mode 100644 arch/sw_64/include/asm/local64.h delete mode 100644 arch/sw_64/include/asm/param.h delete mode 100644 arch/sw_64/include/asm/parport.h delete mode 100644 arch/sw_64/include/asm/preempt.h delete mode 100644 arch/sw_64/include/asm/seccomp.h delete mode 100644 arch/sw_64/include/asm/sections.h delete mode 100644 arch/sw_64/include/asm/segment.h delete mode 100644 arch/sw_64/include/asm/serial.h delete mode 100644 arch/sw_64/include/asm/shmparam.h delete mode 100644 arch/sw_64/include/asm/trace_clock.h delete mode 100644 arch/sw_64/include/asm/types.h delete mode 100644 arch/sw_64/include/asm/unaligned.h

diff --git a/arch/sw_64/include/asm/Kbuild b/arch/sw_64/include/asm/Kbuild index a94a978edf3e..d08f0b08918e 100644 --- a/arch/sw_64/include/asm/Kbuild +++ b/arch/sw_64/include/asm/Kbuild @@ -3,10 +3,15 @@ generic-y += clkdev.h generic-y += export.h generic-y += kvm_types.h +generic-y += local64.h generic-y += mcs_spinlock.h +generic-y += param.h generic-y += qrwlock.h generic-y += qspinlock.h generic-y += rwsem.h -generic-y += scatterlist.h +generic-y += seccomp.h +generic-y += segment.h +generic-y += types.h generic-y += user.h + generated-y += syscall_table.h diff --git a/arch/sw_64/include/asm/div64.h b/arch/sw_64/include/asm/div64.h deleted file mode 100644 index 306581407ba5..000000000000 --- a/arch/sw_64/include/asm/div64.h +++ /dev/null @@ -1,7 +0,0 @@ -/* SPDX-License-Identifier: GPL-2.0 */ -#ifndef _ASM_SW64_DIV64_H -#define _ASM_SW64_DIV64_H - -#include <asm-generic/div64.h> - -#endif diff --git a/arch/sw_64/include/asm/emergency-restart.h b/arch/sw_64/include/asm/emergency-restart.h deleted file mode 100644 index fabb33ebf0eb..000000000000 --- a/arch/sw_64/include/asm/emergency-restart.h +++ /dev/null @@ -1,7 +0,0 @@ -/* SPDX-License-Identifier: GPL-2.0 */ -#ifndef _ASM_SW64_EMERGENCY_RESTART_H -#define _ASM_SW64_EMERGENCY_RESTART_H - -#include <asm-generic/emergency-restart.h> - -#endif /* _ASM_SW64_EMERGENCY_RESTART_H */ diff --git a/arch/sw_64/include/asm/exec.h b/arch/sw_64/include/asm/exec.h deleted file mode 100644 index 4a9cb71c5c47..000000000000 --- a/arch/sw_64/include/asm/exec.h +++ /dev/null @@ -1,7 +0,0 @@ -/* SPDX-License-Identifier: GPL-2.0 */ -#ifndef _ASM_SW64_EXEC_H -#define _ASM_SW64_EXEC_H - -#define arch_align_stack(x) (x) - -#endif /* _ASM_SW64_EXEC_H */ diff --git a/arch/sw_64/include/asm/irq_regs.h b/arch/sw_64/include/asm/irq_regs.h deleted file mode 100644 index bba48f36a40f..000000000000 --- a/arch/sw_64/include/asm/irq_regs.h +++ /dev/null @@ -1,7 +0,0 @@ -/* SPDX-License-Identifier: GPL-2.0 */ -#ifndef _ASM_SW64_IRQ_REGS_H -#define _ASM_SW64_IRQ_REGS_H - -#include <asm-generic/irq_regs.h> - -#endif diff --git a/arch/sw_64/include/asm/kmap_types.h b/arch/sw_64/include/asm/kmap_types.h deleted file mode 100644 index 8e86b08dee94..000000000000 --- a/arch/sw_64/include/asm/kmap_types.h +++ /dev/null @@ -1,15 +0,0 @@ -/* SPDX-License-Identifier: GPL-2.0 */ -#ifndef _ASM_SW64_KMAP_TYPES_H -#define _ASM_SW64_KMAP_TYPES_H - -/* Dummy header just to define km_type. */ - -#ifdef CONFIG_DEBUG_HIGHMEM -#define __WITH_KM_FENCE -#endif - -#include <asm-generic/kmap_types.h> - -#undef __WITH_KM_FENCE - -#endif diff --git a/arch/sw_64/include/asm/local.h b/arch/sw_64/include/asm/local.h deleted file mode 100644 index 9144600f641d..000000000000 --- a/arch/sw_64/include/asm/local.h +++ /dev/null @@ -1,125 +0,0 @@ -/* SPDX-License-Identifier: GPL-2.0 */ -#ifndef _ASM_SW64_LOCAL_H -#define _ASM_SW64_LOCAL_H - -#include <linux/percpu.h> -#include <linux/atomic.h> - -typedef struct { - atomic_long_t a; -} local_t; - -#define LOCAL_INIT(i) { ATOMIC_LONG_INIT(i) } -#define local_read(l) atomic_long_read(&(l)->a) -#define local_set(l, i) atomic_long_set(&(l)->a, (i)) -#define local_inc(l) atomic_long_inc(&(l)->a) -#define local_dec(l) atomic_long_dec(&(l)->a) -#define local_add(i, l) atomic_long_add((i), (&(l)->a)) -#define local_sub(i, l) atomic_long_sub((i), (&(l)->a)) - -static inline long local_add_return(long i, local_t *l) -{ - long temp1, temp2, result, addr; - - __asm__ __volatile__( -#ifdef CONFIG_LOCK_MEMB - " memb\n" -#endif - " ldi %4, %2\n" - "1: lldl %0, 0(%4)\n" - " ldi %1, 1\n" - " wr_f %1\n" - " addl %0, %5, %3\n" - " addl %0, %5, %0\n" -#ifdef CONFIG_LOCK_FIXUP - " memb\n" -#endif - " lstl %0, 0(%4)\n" - " rd_f %0\n" - " beq %0, 2f\n" - ".subsection 2\n" - "2: br 1b\n" - ".previous" - : "=&r" (temp1), "=&r" (temp2), "=m" (l->a.counter), - "=&r" (result), "=&r" (addr) - : "Ir" (i), "m" (l->a.counter) : "memory"); - return result; -} - -static inline long local_sub_return(long i, local_t *l) -{ - long temp1, temp2, result, addr; - - __asm__ __volatile__( -#ifdef CONFIG_LOCK_MEMB - " memb\n" -#endif - " ldi %4, %2\n" - "1: lldl %0, 0(%4)\n" - " ldi %1, 1\n" - " wr_f %1\n" - " subl %0, %5, %3\n" - " subl %0, %5, %0\n" -#ifdef CONFIG_LOCK_FIXUP - " memb\n" -#endif - " lstl %0, 0(%4)\n" - " rd_f %0\n" - " beq %0, 2f\n" - ".subsection 2\n" - "2: br 1b\n" - ".previous" - : "=&r" (temp1), "=&r" (temp2), "=m" (l->a.counter), - "=&r" (result), "=&r" (addr) - : "Ir" (i), "m" (l->a.counter) : "memory"); - return result; -} - -#define local_cmpxchg(l, o, n) \ - (cmpxchg_local(&((l)->a.counter), (o), (n))) -#define local_xchg(l, n) (xchg_local(&((l)->a.counter), (n))) - -/** - * local_add_unless - add unless the number is a given value - * @l: pointer of type local_t - * @a: the amount to add to l... - * @u: ...unless l is equal to u. - * - * Atomically adds @a to @l, so long as it was not @u. - * Returns non-zero if @l was not @u, and zero otherwise. - */ -#define local_add_unless(l, a, u) \ -({ \ - long c, old; \ - c = local_read(l); \ - for (;;) { \ - if (unlikely(c == (u))) \ - break; \ - old = local_cmpxchg((l), c, c + (a)); \ - if (likely(old == c)) \ - break; \ - c = old; \ - } \ - c != (u); \ -}) -#define local_inc_not_zero(l) local_add_unless((l), 1, 0) - -#define local_add_negative(a, l) (local_add_return((a), (l)) < 0) - -#define local_dec_return(l) local_sub_return(1, (l)) - -#define local_inc_return(l) local_add_return(1, (l)) - -#define local_sub_and_test(i, l) (local_sub_return((i), (l)) == 0) - -#define local_inc_and_test(l) (local_add_return(1, (l)) == 0) - -#define local_dec_and_test(l) (local_sub_return(1, (l)) == 0) - -/* Verify if faster than atomic ops */ -#define __local_inc(l) ((l)->a.counter++) -#define __local_dec(l) ((l)->a.counter++) -#define __local_add(i, l) ((l)->a.counter += (i)) -#define __local_sub(i, l) ((l)->a.counter -= (i)) - -#endif /* _ASM_SW64_LOCAL_H */ diff --git a/arch/sw_64/include/asm/local64.h b/arch/sw_64/include/asm/local64.h deleted file mode 100644 index 4278133cd8fa..000000000000 --- a/arch/sw_64/include/asm/local64.h +++ /dev/null @@ -1,7 +0,0 @@ -/* SPDX-License-Identifier: GPL-2.0 */ -#ifndef _ASM_SW64_LOCAL64_H -#define _ASM_SW64_LOCAL64_H - -#include <asm-generic/local64.h> - -#endif diff --git a/arch/sw_64/include/asm/param.h b/arch/sw_64/include/asm/param.h deleted file mode 100644 index 49c5d03a3370..000000000000 --- a/arch/sw_64/include/asm/param.h +++ /dev/null @@ -1,11 +0,0 @@ -/* SPDX-License-Identifier: GPL-2.0 */ -#ifndef _ASM_SW64_PARAM_H -#define _ASM_SW64_PARAM_H - -#include <uapi/asm/param.h> - -#undef HZ -#define HZ CONFIG_HZ -#define USER_HZ 100 -#define CLOCKS_PER_SEC USER_HZ /* frequency at which times() counts */ -#endif /* _ASM_SW64_PARAM_H */ diff --git a/arch/sw_64/include/asm/parport.h b/arch/sw_64/include/asm/parport.h deleted file mode 100644 index 82b9a219b797..000000000000 --- a/arch/sw_64/include/asm/parport.h +++ /dev/null @@ -1,19 +0,0 @@ -/* SPDX-License-Identifier: GPL-2.0 */ -/* - * parport.h: platform-specific PC-style parport initialisation - * - * Copyright (C) 1999, 2000 Tim Waugh tim@cyberelk.demon.co.uk - * - * This file should only be included by drivers/parport/parport_pc.c. - */ - -#ifndef _ASM_SW64_PARPORT_H -#define _ASM_SW64_PARPORT_H - -static int parport_pc_find_isa_ports(int autoirq, int autodma); -static int parport_pc_find_nonpci_ports(int autoirq, int autodma) -{ - return parport_pc_find_isa_ports(autoirq, autodma); -} - -#endif /* !(_ASM_SW64_PARPORT_H) */ diff --git a/arch/sw_64/include/asm/preempt.h b/arch/sw_64/include/asm/preempt.h deleted file mode 100644 index dc6643a43766..000000000000 --- a/arch/sw_64/include/asm/preempt.h +++ /dev/null @@ -1,7 +0,0 @@ -/* SPDX-License-Identifier: GPL-2.0 */ -#ifndef _ASM_SW64_PREEMPT_H -#define _ASM_SW64_PREEMPT_H - -#include <asm-generic/preempt.h> - -#endif /* _ASM_SW64_PREEMPT_H */ diff --git a/arch/sw_64/include/asm/seccomp.h b/arch/sw_64/include/asm/seccomp.h deleted file mode 100644 index db2f298862c3..000000000000 --- a/arch/sw_64/include/asm/seccomp.h +++ /dev/null @@ -1,15 +0,0 @@ -/* SPDX-License-Identifier: GPL-2.0 */ -/* - * arch/sw_64/include/asm/seccomp.h - * - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License version 2 as - * published by the Free Software Foundation. - */ -#ifndef _ASM_SW64_SECCOMP_H -#define _ASM_SW64_SECCOMP_H - -#include <asm/unistd.h> -#include <asm-generic/seccomp.h> - -#endif /* _ASM_SW64_SECCOMP_H */ diff --git a/arch/sw_64/include/asm/sections.h b/arch/sw_64/include/asm/sections.h deleted file mode 100644 index 37dab4fde720..000000000000 --- a/arch/sw_64/include/asm/sections.h +++ /dev/null @@ -1,8 +0,0 @@ -/* SPDX-License-Identifier: GPL-2.0 */ -#ifndef _ASM_SW64_SECTIONS_H -#define _ASM_SW64_SECTIONS_H - -/* nothing to see, move along */ -#include <asm-generic/sections.h> - -#endif diff --git a/arch/sw_64/include/asm/segment.h b/arch/sw_64/include/asm/segment.h deleted file mode 100644 index dc90357765e5..000000000000 --- a/arch/sw_64/include/asm/segment.h +++ /dev/null @@ -1,7 +0,0 @@ -/* SPDX-License-Identifier: GPL-2.0 */ -#ifndef _ASM_SW64_SEGMENT_H -#define _ASM_SW64_SEGMENT_H - -/* Only here because we have some old header files that expect it.. */ - -#endif diff --git a/arch/sw_64/include/asm/serial.h b/arch/sw_64/include/asm/serial.h deleted file mode 100644 index 059e603642b9..000000000000 --- a/arch/sw_64/include/asm/serial.h +++ /dev/null @@ -1,16 +0,0 @@ -/* SPDX-License-Identifier: GPL-2.0 */ -#ifndef _ASM_SW64_SERIAL_H -#define _ASM_SW64_SERIAL_H - -#define BASE_BAUD (1843200 / 16) - -/* Standard COM flags (except for COM4, because of the 8514 problem) */ -#ifdef CONFIG_SERIAL_8250_DETECT_IRQ -#define STD_COM_FLAGS (UPF_BOOT_AUTOCONF | UPF_SKIP_TEST | UPF_AUTO_IRQ) -#define STD_COM4_FLAGS (UPF_BOOT_AUTOCONF | UPF_AUTO_IRQ) -#else -#define STD_COM_FLAGS (UPF_BOOT_AUTOCONF | UPF_SKIP_TEST) -#define STD_COM4_FLAGS UPF_BOOT_AUTOCONF -#endif - -#endif /* _ASM_SW64_SERIAL_H */ diff --git a/arch/sw_64/include/asm/shmparam.h b/arch/sw_64/include/asm/shmparam.h deleted file mode 100644 index 15f71533b1ed..000000000000 --- a/arch/sw_64/include/asm/shmparam.h +++ /dev/null @@ -1,7 +0,0 @@ -/* SPDX-License-Identifier: GPL-2.0 */ -#ifndef _ASM_SW64_SHMPARAM_H -#define _ASM_SW64_SHMPARAM_H - -#define SHMLBA PAGE_SIZE /* attach addr a multiple of this */ - -#endif /* _ASM_SW64_SHMPARAM_H */ diff --git a/arch/sw_64/include/asm/trace_clock.h b/arch/sw_64/include/asm/trace_clock.h deleted file mode 100644 index 57324215a837..000000000000 --- a/arch/sw_64/include/asm/trace_clock.h +++ /dev/null @@ -1,10 +0,0 @@ -/* SPDX-License-Identifier: GPL-2.0 */ -#ifndef _ASM_SW64_TRACE_CLOCK_H -#define _ASM_SW64_TRACE_CLOCK_H - -#include <linux/compiler.h> -#include <linux/types.h> - -#define ARCH_TRACE_CLOCKS - -#endif /* _ASM_SW64_TRACE_CLOCK_H */ diff --git a/arch/sw_64/include/asm/types.h b/arch/sw_64/include/asm/types.h deleted file mode 100644 index 37d626269a02..000000000000 --- a/arch/sw_64/include/asm/types.h +++ /dev/null @@ -1,7 +0,0 @@ -/* SPDX-License-Identifier: GPL-2.0 */ -#ifndef _ASM_SW64_TYPES_H -#define _ASM_SW64_TYPES_H - -#include <asm-generic/int-ll64.h> - -#endif /* _ASM_SW64_TYPES_H */ diff --git a/arch/sw_64/include/asm/unaligned.h b/arch/sw_64/include/asm/unaligned.h deleted file mode 100644 index 91fdff923ce5..000000000000 --- a/arch/sw_64/include/asm/unaligned.h +++ /dev/null @@ -1,12 +0,0 @@ -/* SPDX-License-Identifier: GPL-2.0 */ -#ifndef _ASM_SW64_UNALIGNED_H -#define _ASM_SW64_UNALIGNED_H - -#include <linux/unaligned/le_struct.h> -#include <linux/unaligned/be_byteshift.h> -#include <linux/unaligned/generic.h> - -#define get_unaligned __get_unaligned_le -#define put_unaligned __put_unaligned_le - -#endif /* _ASM_SW64_UNALIGNED_H */

From: He Sheng hesheng@wxiat.com

Sunway inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I56OLG

--------------------------------

SMP_CACHE_BYTES is also defined in linux/cache.h, so remove the replicated one.

Signed-off-by: He Sheng hesheng@wxiat.com

Signed-off-by: Gu Zitao guzitao@wxiat.com --- arch/sw_64/include/asm/cache.h | 4 +--- 1 file changed, 1 insertion(+), 3 deletions(-)

diff --git a/arch/sw_64/include/asm/cache.h b/arch/sw_64/include/asm/cache.h index a59a74110884..1dca2e2e04a4 100644 --- a/arch/sw_64/include/asm/cache.h +++ b/arch/sw_64/include/asm/cache.h @@ -5,9 +5,7 @@ #ifndef _ASM_SW64_CACHE_H #define _ASM_SW64_CACHE_H

-#define L1_CACHE_BYTES 128 #define L1_CACHE_SHIFT 7 - -#define SMP_CACHE_BYTES L1_CACHE_BYTES +#define L1_CACHE_BYTES (1 << L1_CACHE_SHIFT)

#endif

From: He Sheng hesheng@wxiat.com

Sunway inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I56OLG

--------------------------------

This patch removes replicated macros which have been defined in generic version.

Signed-off-by: He Sheng hesheng@wxiat.com

Signed-off-by: Gu Zitao guzitao@wxiat.com --- arch/sw_64/include/asm/module.h | 10 ++-------- 1 file changed, 2 insertions(+), 8 deletions(-)

diff --git a/arch/sw_64/include/asm/module.h b/arch/sw_64/include/asm/module.h index 55e6e333585f..d1663aab4097 100644 --- a/arch/sw_64/include/asm/module.h +++ b/arch/sw_64/include/asm/module.h @@ -2,18 +2,12 @@ #ifndef _ASM_SW64_MODULE_H #define _ASM_SW64_MODULE_H

+#include <asm-generic/module.h> + struct mod_arch_specific { unsigned int gotsecindex; };

-#define Elf_Sym Elf64_Sym -#define Elf_Shdr Elf64_Shdr -#define Elf_Ehdr Elf64_Ehdr -#define Elf_Phdr Elf64_Phdr -#define Elf_Dyn Elf64_Dyn -#define Elf_Rel Elf64_Rel -#define Elf_Rela Elf64_Rela - #define ARCH_SHF_SMALL SHF_SW64_GPREL

#ifdef MODULE

From: He Sheng hesheng@wxiat.com

Sunway inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I56OLG

--------------------------------

For sw64, IO address is mapped by direct mapping without pages. It doesn't have to do anything to unmap PCI MMIO. The result is that __is_mmio helper can be removed.

Signed-off-by: He Sheng hesheng@wxiat.com

Signed-off-by: Gu Zitao guzitao@wxiat.com --- arch/sw_64/include/asm/io.h | 8 -------- arch/sw_64/kernel/pci.c | 4 ---- 2 files changed, 12 deletions(-)

diff --git a/arch/sw_64/include/asm/io.h b/arch/sw_64/include/asm/io.h index 6796c64f94ae..cb6c7c1f4a6b 100644 --- a/arch/sw_64/include/asm/io.h +++ b/arch/sw_64/include/asm/io.h @@ -218,14 +218,6 @@ static inline int __is_ioaddr(unsigned long addr)

#define __is_ioaddr(a) __is_ioaddr((unsigned long)(a))

-static inline int __is_mmio(const volatile void __iomem *xaddr) -{ - unsigned long addr = (unsigned long)xaddr; - - return (addr & 0x100000000UL) == 0; -} - -

#define ioread16be(p) be16_to_cpu(ioread16(p)) #define ioread32be(p) be32_to_cpu(ioread32(p)) diff --git a/arch/sw_64/kernel/pci.c b/arch/sw_64/kernel/pci.c index 44264e3da18f..77bbd1b68176 100644 --- a/arch/sw_64/kernel/pci.c +++ b/arch/sw_64/kernel/pci.c @@ -358,12 +358,8 @@ asmlinkage long sys_pciconfig_iobase(long which, unsigned long bus, unsigned lon return -EOPNOTSUPP; }

-/* Destroy an __iomem token. Not copied from lib/iomap.c. */ - void pci_iounmap(struct pci_dev *dev, void __iomem *addr) { - if (__is_mmio(addr)) - iounmap(addr); } EXPORT_SYMBOL(pci_iounmap);

From: He Sheng hesheng@wxiat.com

Sunway inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I56OLG

--------------------------------

IO interface of chipset is wrapped as a hook in sw64_platform_ops.

Signed-off-by: He Sheng hesheng@wxiat.com

Signed-off-by: Gu Zitao guzitao@wxiat.com --- arch/sw_64/include/asm/io.h | 7 ------- 1 file changed, 7 deletions(-)

diff --git a/arch/sw_64/include/asm/io.h b/arch/sw_64/include/asm/io.h index e8e80c761467..3138665133e1 100644 --- a/arch/sw_64/include/asm/io.h +++ b/arch/sw_64/include/asm/io.h @@ -64,13 +64,6 @@ static inline void * __deprecated bus_to_virt(unsigned long address) } #define isa_bus_to_virt bus_to_virt

-/* - * There are different chipsets to interface the sw64 CPUs to the world. - */ - -#define IO_CONCAT(a, b) _IO_CONCAT(a, b) -#define _IO_CONCAT(a, b) a ## _ ## b - #include <asm/sw64io.h>

/*

From: Zhou Qihang zhouqihang@wxiat.com

Sunway inclusion category: bugfix bugzilla: https://gitee.com/openeuler/kernel/issues/I5G5T3

--------------------------------

Since the RCs of SW64 chipset do not support Peer-to-Peer DMA, we used to return -EINVAL error on iova mampping on PCI BARs. However, it will cause driver loading errors when QEMU calls vfio module to passthrough PCI devices with this kind of BARs, shown as follows:

qemu-system-sw64: VFIO_MAP_DMA: -22 qemu-system-sw64: vfio_dma_map(0x121c28020, 0x8800c0000000, 0x40000,

0x4165f21b2000) = -22 (Invalid argument)

qemu: hardware error: vfio: DMA mapping failed, unable to continue

To fix this problem, this patch will return success on the iova mapping of PCI BARs that pretends to support this feature.

Signed-off-by: Zhou Qihang zhouqihang@wxiat.com

Signed-off-by: Gu Zitao guzitao@wxiat.com --- drivers/iommu/sw64/sunway_iommu.c | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/drivers/iommu/sw64/sunway_iommu.c b/drivers/iommu/sw64/sunway_iommu.c index 8b851e0a0c20..5797adf8fbc5 100644 --- a/drivers/iommu/sw64/sunway_iommu.c +++ b/drivers/iommu/sw64/sunway_iommu.c @@ -1509,6 +1509,9 @@ sunway_iommu_iova_to_phys(struct iommu_domain *dom, dma_addr_t iova) struct sunway_iommu_domain *sdomain = to_sunway_domain(dom); unsigned long paddr, grn;

+ if (iova > SW64_BAR_ADDRESS) + return iova; + paddr = fetch_pte(sdomain, iova, PTE_LEVEL2_VAL);

if ((paddr & SW64_IOMMU_ENTRY_VALID) == 0) @@ -1544,7 +1547,7 @@ sunway_iommu_map(struct iommu_domain *dom, unsigned long iova, * to avoid VFIO trying to map pci config space. */ if (iova > SW64_BAR_ADDRESS) - return -EINVAL; + return 0;

mutex_lock(&sdomain->api_lock); ret = sunway_iommu_map_page(sdomain, iova, paddr, page_size);

From: He Sheng hesheng@wxiat.com

Sunway inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I56OLG

--------------------------------

The hmcall rdio64 can access IOR directly with its PA.

Signed-off-by: He Sheng hesheng@wxiat.com

Signed-off-by: Gu Zitao guzitao@wxiat.com --- arch/sw_64/chip/chip3/chip.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/arch/sw_64/chip/chip3/chip.c b/arch/sw_64/chip/chip3/chip.c index 4d2f99cc6402..704588d7c3e1 100644 --- a/arch/sw_64/chip/chip3/chip.c +++ b/arch/sw_64/chip/chip3/chip.c @@ -54,7 +54,7 @@ static struct clocksource clocksource_longtime = { static u64 read_vtime(struct clocksource *cs) { u64 result; - unsigned long vtime_addr = PAGE_OFFSET | IO_BASE | LONG_TIME; + unsigned long vtime_addr = IO_BASE | LONG_TIME;

result = rdio64(vtime_addr); return result;

From: He Sheng hesheng@wxiat.com

Sunway inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I5G5YB

--------------------------------

It's more standard to access IO registers by calling readX/writeX interfaces. In addition, the code is easier to maintain.

Signed-off-by: He Sheng hesheng@wxiat.com

Signed-off-by: Gu Zitao guzitao@wxiat.com --- arch/sw_64/chip/chip3/i2c-lib.c | 15 +++++-------- arch/sw_64/include/asm/sw64io.h | 37 ++++++++------------------------- 2 files changed, 14 insertions(+), 38 deletions(-)

diff --git a/arch/sw_64/chip/chip3/i2c-lib.c b/arch/sw_64/chip/chip3/i2c-lib.c index b3dcdcd32735..e70f0f0c9a56 100644 --- a/arch/sw_64/chip/chip3/i2c-lib.c +++ b/arch/sw_64/chip/chip3/i2c-lib.c @@ -118,19 +118,14 @@ void __iomem *get_i2c_bar_addr(uint8_t i2c_controller_index) } }

-void write_cpu_i2c_controller(uint64_t offset, uint32_t data) +static inline void write_cpu_i2c_controller(uint64_t offset, uint32_t data) { - mb(); - *(volatile uint32_t *)(m_i2c_base_address + offset) = data; + writel(data, m_i2c_base_address + offset); }

-uint32_t read_cpu_i2c_controller(uint64_t offset) +static inline uint32_t read_cpu_i2c_controller(uint64_t offset) { - uint32_t data; - - data = *(volatile uint32_t *)(m_i2c_base_address + offset); - mb(); - return data; + return readl(m_i2c_base_address + offset); }

static int poll_for_status_set0(uint16_t status_bit) @@ -241,7 +236,7 @@ static int i2c_read(uint8_t reg_offset, uint8_t *buffer, uint32_t length) write_cpu_i2c_controller(DW_IC_DATA_CMD, DW_IC_CMD);

if (poll_for_status_set0(DW_IC_STATUS_RFNE) == 0) - buffer[i] = *(uint8_t *) (m_i2c_base_address + DW_IC_DATA_CMD); + buffer[i] = readb(m_i2c_base_address + DW_IC_DATA_CMD); else pr_err("Read timeout line %d.\n", __LINE__); } diff --git a/arch/sw_64/include/asm/sw64io.h b/arch/sw_64/include/asm/sw64io.h index e66aba66932b..7d79a5b75090 100644 --- a/arch/sw_64/include/asm/sw64io.h +++ b/arch/sw_64/include/asm/sw64io.h @@ -2,6 +2,7 @@ #ifndef _ASM_SW64_SW64IO_H #define _ASM_SW64_SW64IO_H

+#include <asm/io.h> #include <asm/page.h>

extern void setup_chip_clocksource(void); @@ -22,13 +23,9 @@ read_rc_conf(unsigned long node, unsigned long rc, unsigned int offset) { void __iomem *addr; - unsigned int value;

addr = __va(MK_RC_CFG(node, rc) | offset); - value = *(volatile unsigned int *)addr; - mb(); - - return value; + return readl(addr); }

static inline void @@ -38,8 +35,7 @@ write_rc_conf(unsigned long node, unsigned long rc, void __iomem *addr;

addr = __va(MK_RC_CFG(node, rc) | offset); - *(unsigned int *)addr = data; - mb(); + writel(data, addr); }

static inline unsigned long @@ -47,13 +43,9 @@ read_piu_ior0(unsigned long node, unsigned long rc, unsigned int reg) { void __iomem *addr; - unsigned long value;

addr = __va(MK_PIU_IOR0(node, rc) + reg); - value = *(volatile unsigned long __iomem *)addr; - mb(); - - return value; + return readq(addr); }

static inline void @@ -63,8 +55,7 @@ write_piu_ior0(unsigned long node, unsigned long rc, void __iomem *addr;

addr = __va(MK_PIU_IOR0(node, rc) + reg); - *(unsigned long __iomem *)addr = data; - mb(); + writeq(data, addr); }

static inline unsigned long @@ -72,13 +63,9 @@ read_piu_ior1(unsigned long node, unsigned long rc, unsigned int reg) { void __iomem *addr; - unsigned long value;

addr = __va(MK_PIU_IOR1(node, rc) + reg); - value = *(volatile unsigned long __iomem *)addr; - mb(); - - return value; + return readq(addr); }

static inline void @@ -88,21 +75,16 @@ write_piu_ior1(unsigned long node, unsigned long rc, void __iomem *addr;

addr = __va(MK_PIU_IOR1(node, rc) + reg); - *(volatile unsigned long __iomem *)addr = data; - mb(); + writeq(data, addr); }

static inline unsigned long sw64_io_read(unsigned long node, unsigned long reg) { void __iomem *addr; - unsigned long value;

addr = __va(SW64_IO_BASE(node) | reg); - value = *(volatile unsigned long __iomem *)addr; - mb(); - - return value; + return readq(addr); }

static inline void @@ -111,7 +93,6 @@ sw64_io_write(unsigned long node, unsigned long reg, unsigned long data) void __iomem *addr;

addr = __va(SW64_IO_BASE(node) | reg); - *(volatile unsigned long __iomem *)addr = data; - mb(); + writeq(data, addr); } #endif

From: He Sheng hesheng@wxiat.com

Sunway inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I5GF7A

--------------------------------

We put thread's callee-saved registers in thread_struct, and put exception callee-saved registers in pt_regs. For context switch, it switches fpu state first, and then switches integer registers in __switch_to.

Because callee-saved registers are always saved and restored, it will cause a little overhead.

As a result, the special sw64 fork like system calls are removed.

Signed-off-by: He Sheng hesheng@wxiat.com

Signed-off-by: Gu Zitao guzitao@wxiat.com --- arch/sw_64/include/asm/processor.h | 3 + arch/sw_64/include/asm/switch_to.h | 37 +++++- arch/sw_64/include/uapi/asm/ptrace.h | 15 ++- arch/sw_64/kernel/asm-offsets.c | 16 +++ arch/sw_64/kernel/entry.S | 173 +++++++++++++++---------- arch/sw_64/kernel/process.c | 21 ++- arch/sw_64/kernel/syscalls/syscall.tbl | 6 +- 7 files changed, 175 insertions(+), 96 deletions(-)

diff --git a/arch/sw_64/include/asm/processor.h b/arch/sw_64/include/asm/processor.h index 645c33a596ff..67d3b4368987 100644 --- a/arch/sw_64/include/asm/processor.h +++ b/arch/sw_64/include/asm/processor.h @@ -78,6 +78,9 @@ struct context_fpregs { struct thread_struct { struct context_fpregs ctx_fp; unsigned long fpcr; + /* Callee-saved registers */ + unsigned long ra; + unsigned long s[7]; /* s0 ~ s6 */ }; #define INIT_THREAD { }

diff --git a/arch/sw_64/include/asm/switch_to.h b/arch/sw_64/include/asm/switch_to.h index 22045b247557..d503fc59390f 100644 --- a/arch/sw_64/include/asm/switch_to.h +++ b/arch/sw_64/include/asm/switch_to.h @@ -2,12 +2,41 @@ #ifndef _ASM_SW64_SWITCH_TO_H #define _ASM_SW64_SWITCH_TO_H

-struct task_struct; -extern struct task_struct *__switch_to(unsigned long, struct task_struct *); +#include<linux/sched.h> + +extern void __fpstate_save(struct task_struct *save_to); +extern void __fpstate_restore(struct task_struct *restore_from); +extern struct task_struct *__switch_to(unsigned long pcb, + struct task_struct *prev, struct task_struct *next); extern void restore_da_match_after_sched(void); -#define switch_to(P, N, L) \ + +static inline void fpstate_save(struct task_struct *task) +{ + if (likely(!(task->flags & PF_KTHREAD))) + __fpstate_save(task); +} + +static inline void fpstate_restore(struct task_struct *task) +{ + if (likely(!(task->flags & PF_KTHREAD))) + __fpstate_restore(task); +} + +static inline void __switch_to_aux(struct task_struct *prev, + struct task_struct *next) +{ + fpstate_save(prev); + fpstate_restore(next); +} + + +#define switch_to(prev, next, last) \ do { \ - (L) = __switch_to(virt_to_phys(&task_thread_info(N)->pcb), (P));\ + struct task_struct *__prev = (prev); \ + struct task_struct *__next = (next); \ + __u64 __nextpcb = virt_to_phys(&task_thread_info(__next)->pcb); \ + __switch_to_aux(__prev, __next); \ + (last) = __switch_to(__nextpcb, __prev, __next); \ check_mmu_context(); \ } while (0)

diff --git a/arch/sw_64/include/uapi/asm/ptrace.h b/arch/sw_64/include/uapi/asm/ptrace.h index 96cb10891bea..4549e8d4bf57 100644 --- a/arch/sw_64/include/uapi/asm/ptrace.h +++ b/arch/sw_64/include/uapi/asm/ptrace.h @@ -9,12 +9,7 @@ * * NOTE! I want to minimize the overhead of system calls, so this * struct has as little information as possible. I does not have - * - * - floating point regs: the kernel doesn't change those - * - r9-15: saved by the C compiler - * - * This makes "fork()" and "exec()" a bit more complex, but should - * give us low system call latency. + * floating point regs, because the kernel doesn't change those */

struct pt_regs { @@ -27,6 +22,14 @@ struct pt_regs { unsigned long r6; unsigned long r7; unsigned long r8; + unsigned long r9; + unsigned long r10; + unsigned long r11; + unsigned long r12; + unsigned long r13; + unsigned long r14; + unsigned long r15; + /* r16 ~ r18 saved by hmcode */ unsigned long r19; unsigned long r20; unsigned long r21; diff --git a/arch/sw_64/kernel/asm-offsets.c b/arch/sw_64/kernel/asm-offsets.c index bea12d2d96fe..349bc5c4e2f1 100644 --- a/arch/sw_64/kernel/asm-offsets.c +++ b/arch/sw_64/kernel/asm-offsets.c @@ -72,6 +72,13 @@ void foo(void) DEFINE(PT_REGS_R6, offsetof(struct pt_regs, r6)); DEFINE(PT_REGS_R7, offsetof(struct pt_regs, r7)); DEFINE(PT_REGS_R8, offsetof(struct pt_regs, r8)); + DEFINE(PT_REGS_R9, offsetof(struct pt_regs, r9)); + DEFINE(PT_REGS_R10, offsetof(struct pt_regs, r10)); + DEFINE(PT_REGS_R11, offsetof(struct pt_regs, r11)); + DEFINE(PT_REGS_R12, offsetof(struct pt_regs, r12)); + DEFINE(PT_REGS_R13, offsetof(struct pt_regs, r13)); + DEFINE(PT_REGS_R14, offsetof(struct pt_regs, r14)); + DEFINE(PT_REGS_R15, offsetof(struct pt_regs, r15)); DEFINE(PT_REGS_R19, offsetof(struct pt_regs, r19)); DEFINE(PT_REGS_R20, offsetof(struct pt_regs, r20)); DEFINE(PT_REGS_R21, offsetof(struct pt_regs, r21)); @@ -258,4 +265,13 @@ void foo(void) DEFINE(CTX_FP_F29, offsetof(struct context_fpregs, f29)); DEFINE(CTX_FP_F30, offsetof(struct context_fpregs, f30)); BLANK(); + OFFSET(TASK_THREAD_RA, task_struct, thread.ra); + OFFSET(TASK_THREAD_S0, task_struct, thread.s[0]); + OFFSET(TASK_THREAD_S1, task_struct, thread.s[1]); + OFFSET(TASK_THREAD_S2, task_struct, thread.s[2]); + OFFSET(TASK_THREAD_S3, task_struct, thread.s[3]); + OFFSET(TASK_THREAD_S4, task_struct, thread.s[4]); + OFFSET(TASK_THREAD_S5, task_struct, thread.s[5]); + OFFSET(TASK_THREAD_S6, task_struct, thread.s[6]); + BLANK(); } diff --git a/arch/sw_64/kernel/entry.S b/arch/sw_64/kernel/entry.S index 6c40d2015439..ac62461aa9dc 100644 --- a/arch/sw_64/kernel/entry.S +++ b/arch/sw_64/kernel/entry.S @@ -21,52 +21,84 @@ * the hmcode-provided values are available to the signal handler. */

-#define SAVE_ALL \ - ldi $sp, -PT_REGS_PS($sp); \ - stl $0, PT_REGS_R0($sp); \ - stl $1, PT_REGS_R1($sp); \ - stl $2, PT_REGS_R2($sp); \ - stl $3, PT_REGS_R3($sp); \ - stl $4, PT_REGS_R4($sp); \ - stl $28, PT_REGS_R28($sp); \ - stl $5, PT_REGS_R5($sp); \ - stl $6, PT_REGS_R6($sp); \ - stl $7, PT_REGS_R7($sp); \ - stl $8, PT_REGS_R8($sp); \ - stl $19, PT_REGS_R19($sp); \ - stl $20, PT_REGS_R20($sp); \ - stl $21, PT_REGS_R21($sp); \ - stl $22, PT_REGS_R22($sp); \ - stl $23, PT_REGS_R23($sp); \ - stl $24, PT_REGS_R24($sp); \ - stl $25, PT_REGS_R25($sp); \ - stl $26, PT_REGS_R26($sp); \ - stl $27, PT_REGS_R27($sp); \ - stl $16, PT_REGS_TRAP_A0($sp); \ - stl $17, PT_REGS_TRAP_A1($sp); \ + .macro SAVE_COMMON_REGS + ldi $sp, -PT_REGS_PS($sp) + stl $0, PT_REGS_R0($sp) + stl $1, PT_REGS_R1($sp) + stl $2, PT_REGS_R2($sp) + stl $3, PT_REGS_R3($sp) + stl $4, PT_REGS_R4($sp) + stl $28, PT_REGS_R28($sp) + stl $5, PT_REGS_R5($sp) + stl $6, PT_REGS_R6($sp) + stl $7, PT_REGS_R7($sp) + stl $8, PT_REGS_R8($sp) + stl $19, PT_REGS_R19($sp) + stl $20, PT_REGS_R20($sp) + stl $21, PT_REGS_R21($sp) + stl $22, PT_REGS_R22($sp) + stl $23, PT_REGS_R23($sp) + stl $24, PT_REGS_R24($sp) + stl $25, PT_REGS_R25($sp) + stl $26, PT_REGS_R26($sp) + stl $27, PT_REGS_R27($sp) + stl $16, PT_REGS_TRAP_A0($sp) + stl $17, PT_REGS_TRAP_A1($sp) stl $18, PT_REGS_TRAP_A2($sp) + .endm

-#define RESTORE_ALL \ - ldl $0, PT_REGS_R0($sp); \ - ldl $1, PT_REGS_R1($sp); \ - ldl $2, PT_REGS_R2($sp); \ - ldl $3, PT_REGS_R3($sp); \ - ldl $4, PT_REGS_R4($sp); \ - ldl $5, PT_REGS_R5($sp); \ - ldl $6, PT_REGS_R6($sp); \ - ldl $7, PT_REGS_R7($sp); \ - ldl $8, PT_REGS_R8($sp); \ - ldl $19, PT_REGS_R19($sp); \ - ldl $20, PT_REGS_R20($sp); \ - ldl $21, PT_REGS_R21($sp); \ - ldl $22, PT_REGS_R22($sp); \ - ldl $23, PT_REGS_R23($sp); \ - ldl $24, PT_REGS_R24($sp); \ - ldl $25, PT_REGS_R25($sp); \ - ldl $26, PT_REGS_R26($sp); \ - ldl $27, PT_REGS_R27($sp); \ - ldl $28, PT_REGS_R28($sp); \ + .macro RESTORE_COMMON_REGS + ldl $0, PT_REGS_R0($sp) + ldl $1, PT_REGS_R1($sp) + ldl $2, PT_REGS_R2($sp) + ldl $3, PT_REGS_R3($sp) + ldl $4, PT_REGS_R4($sp) + ldl $5, PT_REGS_R5($sp) + ldl $6, PT_REGS_R6($sp) + ldl $7, PT_REGS_R7($sp) + ldl $8, PT_REGS_R8($sp) + ldl $19, PT_REGS_R19($sp) + ldl $20, PT_REGS_R20($sp) + ldl $21, PT_REGS_R21($sp) + ldl $22, PT_REGS_R22($sp) + ldl $23, PT_REGS_R23($sp) + ldl $24, PT_REGS_R24($sp) + ldl $25, PT_REGS_R25($sp) + ldl $26, PT_REGS_R26($sp) + ldl $27, PT_REGS_R27($sp) + ldl $28, PT_REGS_R28($sp) ldi $sp, PT_REGS_PS($sp) + .endm + + .macro SAVE_CALLEE_REGS + stl $9, PT_REGS_R9($sp) + stl $10, PT_REGS_R10($sp) + stl $11, PT_REGS_R11($sp) + stl $12, PT_REGS_R12($sp) + stl $13, PT_REGS_R13($sp) + stl $14, PT_REGS_R14($sp) + stl $15, PT_REGS_R15($sp) + .endm + + .macro RESTORE_CALLEE_REGS + ldl $9, PT_REGS_R9($sp) + ldl $10, PT_REGS_R10($sp) + ldl $11, PT_REGS_R11($sp) + ldl $12, PT_REGS_R12($sp) + ldl $13, PT_REGS_R13($sp) + ldl $14, PT_REGS_R14($sp) + ldl $15, PT_REGS_R15($sp) + .endm + + .macro SAVE_ALL + SAVE_COMMON_REGS + SAVE_CALLEE_REGS + .endm + + .macro RESTORE_ALL + RESTORE_CALLEE_REGS + RESTORE_COMMON_REGS + .endm

/* * Non-syscall kernel entry points. @@ -591,19 +623,42 @@ $setfpec_over: .end undo_switch_stack

/* - * The meat of the context switch code. + * Integer register context switch + * The callee-saved registers must be saved and restored. + * + * a0: physical address of next task's pcb, used by hmcode + * a1: previous task_struct (must be preserved across the switch) + * a2: next task_struct + * + * The value of a1 must be preserved by this function, as that's how + * arguments are passed to schedule_tail. */ - .align 4 .globl __switch_to .ent __switch_to __switch_to: .prologue 0 - bsr $1, do_switch_stack + /* Save context into prev->thread */ + stl $26, TASK_THREAD_RA($17) + stl $9, TASK_THREAD_S0($17) + stl $10, TASK_THREAD_S1($17) + stl $11, TASK_THREAD_S2($17) + stl $12, TASK_THREAD_S3($17) + stl $13, TASK_THREAD_S4($17) + stl $14, TASK_THREAD_S5($17) + stl $15, TASK_THREAD_S6($17) + /* Restore context from next->thread */ + ldl $26, TASK_THREAD_RA($18) + ldl $9, TASK_THREAD_S0($18) + ldl $10, TASK_THREAD_S1($18) + ldl $11, TASK_THREAD_S2($18) + ldl $12, TASK_THREAD_S3($18) + ldl $13, TASK_THREAD_S4($18) + ldl $14, TASK_THREAD_S5($18) + ldl $15, TASK_THREAD_S6($18) sys_call HMC_swpctx ldi $8, 0x3fff bic $sp, $8, $8 - bsr $1, undo_switch_stack mov $17, $0 ret .end __switch_to @@ -637,30 +692,6 @@ ret_from_kernel_thread: br $31, ret_to_user .end ret_from_kernel_thread

-/* - * Special system calls. Most of these are special in that they either - * have to play switch_stack games or in some way use the pt_regs struct. - */ - -.macro fork_like name - .align 4 - .globl sw64_\name - .ent sw64_\name -sw64_\name: - .prologue 0 - bsr $1, do_switch_stack - call $26, sys_\name - ldl $26, SWITCH_STACK_RA($sp) - ldi $sp, SWITCH_STACK_SIZE($sp) - ret - .end sw64_\name - .endm - -fork_like fork -fork_like vfork -fork_like clone -fork_like clone3 - .align 4 .globl sys_sigreturn .ent sys_sigreturn diff --git a/arch/sw_64/kernel/process.c b/arch/sw_64/kernel/process.c index 4192d50f5b0e..c6e87874af19 100644 --- a/arch/sw_64/kernel/process.c +++ b/arch/sw_64/kernel/process.c @@ -11,6 +11,7 @@ #include <linux/random.h>

#include <asm/fpu.h> +#include <asm/switch_to.h>

#include "proto.h"

@@ -158,19 +159,18 @@ copy_thread(unsigned long clone_flags, unsigned long usp, struct thread_info *childti = task_thread_info(p); struct pt_regs *childregs = task_pt_regs(p); struct pt_regs *regs = current_pt_regs(); - struct switch_stack *childstack, *stack;

- childstack = ((struct switch_stack *) childregs) - 1; - childti->pcb.ksp = (unsigned long) childstack; + childti->pcb.ksp = (unsigned long) childregs; childti->pcb.flags = 7; /* set FEN, clear everything else */ + __fpstate_save(current); + p->thread = current->thread;

if (unlikely(p->flags & PF_KTHREAD)) { /* kernel thread */ - memset(childstack, 0, - sizeof(struct switch_stack) + sizeof(struct pt_regs)); - childstack->r26 = (unsigned long) ret_from_kernel_thread; - childstack->r9 = usp; /* function */ - childstack->r10 = kthread_arg; + memset(childregs, 0, sizeof(struct pt_regs)); + p->thread.ra = (unsigned long) ret_from_kernel_thread; + p->thread.s[0] = usp; /* function */ + p->thread.s[1] = kthread_arg; childti->pcb.usp = 0; return 0; } @@ -189,10 +189,7 @@ copy_thread(unsigned long clone_flags, unsigned long usp, *childregs = *regs; childregs->r0 = 0; childregs->r19 = 0; - stack = ((struct switch_stack *) regs) - 1; - *childstack = *stack; - p->thread = current->thread; - childstack->r26 = (unsigned long) ret_from_fork; + p->thread.ra = (unsigned long) ret_from_fork; return 0; }

diff --git a/arch/sw_64/kernel/syscalls/syscall.tbl b/arch/sw_64/kernel/syscalls/syscall.tbl index b9b93d70124d..42a179422b6b 100644 --- a/arch/sw_64/kernel/syscalls/syscall.tbl +++ b/arch/sw_64/kernel/syscalls/syscall.tbl @@ -73,7 +73,7 @@ 63 common getpgrp sys_getpgrp #64 is unused #65 is unused -66 common vfork sw64_vfork +66 common vfork sys_vfork 67 common stat sys_newstat 68 common lstat sys_newlstat #69 is unused @@ -289,7 +289,7 @@ 279 common fsmount sys_fsmount 280 common fspick sys_fspick 281 common pidfd_open sys_pidfd_open -282 common clone3 sw64_clone3 +282 common clone3 sys_clone3 283 common close_range sys_close_range 284 common openat2 sys_openat2 285 common pidfd_getfd sys_pidfd_getfd @@ -319,7 +319,7 @@ 309 common get_kernel_syms sys_ni_syscall 310 common syslog sys_syslog 311 common reboot sys_reboot -312 common clone sw64_clone +312 common clone sys_clone 313 common uselib sys_uselib 314 common mlock sys_mlock 315 common munlock sys_munlock

From: He Sheng hesheng@wxiat.com

Sunway inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I5GF7A

--------------------------------

This patch removes switch_stack and its usages in mm fault entry and system call entry. Note that ptrace accesses registers with struct pt_regs only since strace gave up a switch_stack trick.

Signed-off-by: He Sheng hesheng@wxiat.com

Signed-off-by: Gu Zitao guzitao@wxiat.com --- arch/sw_64/kernel/entry.S | 42 +++----------------------------------- arch/sw_64/kernel/ptrace.c | 12 ++--------- 2 files changed, 5 insertions(+), 49 deletions(-)

diff --git a/arch/sw_64/kernel/entry.S b/arch/sw_64/kernel/entry.S index ac62461aa9dc..f843acd88249 100644 --- a/arch/sw_64/kernel/entry.S +++ b/arch/sw_64/kernel/entry.S @@ -133,31 +133,11 @@ entArith: .ent entMM entMM: SAVE_ALL -/* save $9 - $15 so the inline exception code can manipulate them. */ - subl $sp, SWITCH_STACK_RA, $sp - stl $9, SWITCH_STACK_R9($sp) - stl $10, SWITCH_STACK_R10($sp) - stl $11, SWITCH_STACK_R11($sp) - stl $12, SWITCH_STACK_R12($sp) - stl $13, SWITCH_STACK_R13($sp) - stl $14, SWITCH_STACK_R14($sp) - stl $15, SWITCH_STACK_R15($sp) - addl $sp, SWITCH_STACK_RA, $19 -/* handle the fault */ ldi $8, 0x3fff + ldi $26, ret_from_sys_call bic $sp, $8, $8 - call $26, do_page_fault -/* reload the registers after the exception code played. */ - ldl $9, SWITCH_STACK_R9($sp) - ldl $10, SWITCH_STACK_R10($sp) - ldl $11, SWITCH_STACK_R11($sp) - ldl $12, SWITCH_STACK_R12($sp) - ldl $13, SWITCH_STACK_R13($sp) - ldl $14, SWITCH_STACK_R14($sp) - ldl $15, SWITCH_STACK_R15($sp) - addl $sp, SWITCH_STACK_RA, $sp -/* finish up the syscall as normal. */ - br ret_from_sys_call + mov $sp, $19 + call $31, do_page_fault .end entMM

.align 4 @@ -400,9 +380,7 @@ $work_resched:

$work_notifysig: mov $sp, $16 - bsr $1, do_switch_stack call $26, do_work_pending - bsr $1, undo_switch_stack br restore_all .end work_pending

@@ -416,14 +394,9 @@ $work_notifysig: .ent strace strace: /* set up signal stack, call syscall_trace */ - bsr $1, do_switch_stack mov $0, $9 mov $19, $10 call $26, syscall_trace_enter - mov $9, $18 - mov $10, $19 - bsr $1, undo_switch_stack - blt $0, $syscall_trace_failed

/* get the system call number and the arguments back.. */ @@ -452,10 +425,7 @@ ret_from_straced: stl $31, PT_REGS_R19($sp) /* a3=0 => no error */ $strace_success: stl $0, PT_REGS_R0($sp) /* save return value */ - - bsr $1, do_switch_stack call $26, syscall_trace_leave - bsr $1, undo_switch_stack br $31, ret_from_sys_call

.align 3 @@ -470,25 +440,19 @@ $strace_error: stl $0, PT_REGS_R0($sp) stl $1, PT_REGS_R19($sp) /* a3 for return */

- bsr $1, do_switch_stack mov $18, $9 /* save old syscall number */ mov $19, $10 /* save old a3 */ call $26, syscall_trace_leave mov $9, $18 mov $10, $19 - bsr $1, undo_switch_stack

mov $31, $26 /* tell "ret_from_sys_call" we can restart */ br ret_from_sys_call

$syscall_trace_failed: - bsr $1, do_switch_stack - mov $18, $9 - mov $19, $10 call $26, syscall_trace_leave mov $9, $18 mov $10, $19 - bsr $1, undo_switch_stack mov $31, $26 /* tell "ret_from_sys_call" we can restart */ br ret_from_sys_call .end strace diff --git a/arch/sw_64/kernel/ptrace.c b/arch/sw_64/kernel/ptrace.c index 3b29ff7114f7..097590b22a5a 100644 --- a/arch/sw_64/kernel/ptrace.c +++ b/arch/sw_64/kernel/ptrace.c @@ -33,10 +33,6 @@ * | frame generated by SAVE_ALL | | * | | v * +================================+ - * | | ^ - * | frame saved by do_switch_stack | | struct switch_stack - * | | v - * +================================+ */

/* @@ -59,18 +55,14 @@ enum { #define PT_REG(reg) \ (PAGE_SIZE * 2 - sizeof(struct pt_regs) + offsetof(struct pt_regs, reg))

-#define SW_REG(reg) \ - (PAGE_SIZE * 2 - sizeof(struct pt_regs) - sizeof(struct switch_stack) \ - + offsetof(struct switch_stack, reg)) - #define FP_REG(fp_regno, vector_regno) \ (fp_regno * 32 + vector_regno * 8)

static int regoff[] = { PT_REG(r0), PT_REG(r1), PT_REG(r2), PT_REG(r3), PT_REG(r4), PT_REG(r5), PT_REG(r6), PT_REG(r7), - PT_REG(r8), SW_REG(r9), SW_REG(r10), SW_REG(r11), - SW_REG(r12), SW_REG(r13), SW_REG(r14), SW_REG(r15), + PT_REG(r8), PT_REG(r9), PT_REG(r10), PT_REG(r11), + PT_REG(r12), PT_REG(r13), PT_REG(r14), PT_REG(r15), PT_REG(r16), PT_REG(r17), PT_REG(r18), PT_REG(r19), PT_REG(r20), PT_REG(r21), PT_REG(r22), PT_REG(r23), PT_REG(r24), PT_REG(r25), PT_REG(r26), PT_REG(r27),

From: He Sheng hesheng@wxiat.com

Sunway inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I5GF7A

--------------------------------

Signed-off-by: He Sheng hesheng@wxiat.com

Signed-off-by: Gu Zitao guzitao@wxiat.com --- arch/sw_64/kernel/process.c | 17 +++++++---------- 1 file changed, 7 insertions(+), 10 deletions(-)

diff --git a/arch/sw_64/kernel/process.c b/arch/sw_64/kernel/process.c index 9a6812851b15..4e7a1418af30 100644 --- a/arch/sw_64/kernel/process.c +++ b/arch/sw_64/kernel/process.c @@ -199,9 +199,6 @@ copy_thread(unsigned long clone_flags, unsigned long usp, void dump_elf_thread(elf_greg_t *dest, struct pt_regs *pt, struct thread_info *ti) { - /* switch stack follows right below pt_regs: */ - struct switch_stack *sw = ((struct switch_stack *) pt) - 1; - dest[0] = pt->r0; dest[1] = pt->r1; dest[2] = pt->r2; @@ -211,13 +208,13 @@ dump_elf_thread(elf_greg_t *dest, struct pt_regs *pt, struct thread_info *ti) dest[6] = pt->r6; dest[7] = pt->r7; dest[8] = pt->r8; - dest[9] = sw->r9; - dest[10] = sw->r10; - dest[11] = sw->r11; - dest[12] = sw->r12; - dest[13] = sw->r13; - dest[14] = sw->r14; - dest[15] = sw->r15; + dest[9] = pt->r9; + dest[10] = pt->r10; + dest[11] = pt->r11; + dest[12] = pt->r12; + dest[13] = pt->r13; + dest[14] = pt->r14; + dest[15] = pt->r15; dest[16] = pt->r16; dest[17] = pt->r17; dest[18] = pt->r18;

From: He Sheng hesheng@wxiat.com

Sunway inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I5GF7A

--------------------------------

It's simple to maintain unified register layout for user-mode and kernel-mode unaligned accesses. With the new struct pt_regs, we are able to make the exception handler cleaner.

After that, the unused struct allregs can be cleaned up.

Signed-off-by: He Sheng hesheng@wxiat.com

Signed-off-by: Gu Zitao guzitao@wxiat.com --- arch/sw_64/kernel/asm-offsets.c | 41 ------------ arch/sw_64/kernel/entry.S | 109 +++++--------------------------- arch/sw_64/kernel/traps.c | 76 +++++----------------- 3 files changed, 33 insertions(+), 193 deletions(-)

diff --git a/arch/sw_64/kernel/asm-offsets.c b/arch/sw_64/kernel/asm-offsets.c index 349bc5c4e2f1..0e8a814f47ce 100644 --- a/arch/sw_64/kernel/asm-offsets.c +++ b/arch/sw_64/kernel/asm-offsets.c @@ -111,47 +111,6 @@ void foo(void) DEFINE(SWITCH_STACK_RA, offsetof(struct switch_stack, r26)); BLANK();

- DEFINE(ALLREGS_SIZE, sizeof(struct allregs)); - DEFINE(ALLREGS_R0, offsetof(struct allregs, regs[0])); - DEFINE(ALLREGS_R1, offsetof(struct allregs, regs[1])); - DEFINE(ALLREGS_R2, offsetof(struct allregs, regs[2])); - DEFINE(ALLREGS_R3, offsetof(struct allregs, regs[3])); - DEFINE(ALLREGS_R4, offsetof(struct allregs, regs[4])); - DEFINE(ALLREGS_R5, offsetof(struct allregs, regs[5])); - DEFINE(ALLREGS_R6, offsetof(struct allregs, regs[6])); - DEFINE(ALLREGS_R7, offsetof(struct allregs, regs[7])); - DEFINE(ALLREGS_R8, offsetof(struct allregs, regs[8])); - DEFINE(ALLREGS_R9, offsetof(struct allregs, regs[9])); - DEFINE(ALLREGS_R10, offsetof(struct allregs, regs[10])); - DEFINE(ALLREGS_R11, offsetof(struct allregs, regs[11])); - DEFINE(ALLREGS_R12, offsetof(struct allregs, regs[12])); - DEFINE(ALLREGS_R13, offsetof(struct allregs, regs[13])); - DEFINE(ALLREGS_R14, offsetof(struct allregs, regs[14])); - DEFINE(ALLREGS_R15, offsetof(struct allregs, regs[15])); - DEFINE(ALLREGS_R16, offsetof(struct allregs, regs[16])); - DEFINE(ALLREGS_R17, offsetof(struct allregs, regs[17])); - DEFINE(ALLREGS_R18, offsetof(struct allregs, regs[18])); - DEFINE(ALLREGS_R19, offsetof(struct allregs, regs[19])); - DEFINE(ALLREGS_R20, offsetof(struct allregs, regs[20])); - DEFINE(ALLREGS_R21, offsetof(struct allregs, regs[21])); - DEFINE(ALLREGS_R22, offsetof(struct allregs, regs[22])); - DEFINE(ALLREGS_R23, offsetof(struct allregs, regs[23])); - DEFINE(ALLREGS_R24, offsetof(struct allregs, regs[24])); - DEFINE(ALLREGS_R25, offsetof(struct allregs, regs[25])); - DEFINE(ALLREGS_R26, offsetof(struct allregs, regs[26])); - DEFINE(ALLREGS_R27, offsetof(struct allregs, regs[27])); - DEFINE(ALLREGS_R28, offsetof(struct allregs, regs[28])); - DEFINE(ALLREGS_R29, offsetof(struct allregs, regs[29])); - DEFINE(ALLREGS_R30, offsetof(struct allregs, regs[30])); - DEFINE(ALLREGS_R31, offsetof(struct allregs, regs[31])); - DEFINE(ALLREGS_PS, offsetof(struct allregs, ps)); - DEFINE(ALLREGS_PC, offsetof(struct allregs, pc)); - DEFINE(ALLREGS_GP, offsetof(struct allregs, gp)); - DEFINE(ALLREGS_A0, offsetof(struct allregs, a0)); - DEFINE(ALLREGS_A1, offsetof(struct allregs, a1)); - DEFINE(ALLREGS_A2, offsetof(struct allregs, a2)); - BLANK(); - DEFINE(KVM_REGS_SIZE, sizeof(struct kvm_regs)); DEFINE(KVM_REGS_R0, offsetof(struct kvm_regs, r0)); DEFINE(KVM_REGS_R1, offsetof(struct kvm_regs, r1)); diff --git a/arch/sw_64/kernel/entry.S b/arch/sw_64/kernel/entry.S index 730346754ecc..f54b02c1b3a9 100644 --- a/arch/sw_64/kernel/entry.S +++ b/arch/sw_64/kernel/entry.S @@ -152,109 +152,32 @@ entIF: call $31, do_entIF .end entIF

+/* + * Handle unalignment exception. + * We don't handle the "gp" register correctly, but if we fault on a + * gp-register unaligned load/store, something is _very_ wrong in the + * kernel anyway. + */ .align 4 .globl entUna .ent entUna entUna: - ldi $sp, -ALLREGS_PS($sp) - stl $0, ALLREGS_R0($sp) - ldl $0, ALLREGS_PS($sp) /* get PS */ - stl $1, ALLREGS_R1($sp) - stl $2, ALLREGS_R2($sp) - stl $3, ALLREGS_R3($sp) - and $0, 8, $0 /* user mode? */ - stl $4, ALLREGS_R4($sp) - bne $0, entUnaUser /* yup -> do user-level unaligned fault */ - stl $5, ALLREGS_R5($sp) - stl $6, ALLREGS_R6($sp) - stl $7, ALLREGS_R7($sp) - stl $8, ALLREGS_R8($sp) - stl $9, ALLREGS_R9($sp) - stl $10, ALLREGS_R10($sp) - stl $11, ALLREGS_R11($sp) - stl $12, ALLREGS_R12($sp) - stl $13, ALLREGS_R13($sp) - stl $14, ALLREGS_R14($sp) - stl $15, ALLREGS_R15($sp) - /* 16-18 HMCODE-saved */ - stl $19, ALLREGS_R19($sp) - stl $20, ALLREGS_R20($sp) - stl $21, ALLREGS_R21($sp) - stl $22, ALLREGS_R22($sp) - stl $23, ALLREGS_R23($sp) - stl $24, ALLREGS_R24($sp) - stl $25, ALLREGS_R25($sp) - stl $26, ALLREGS_R26($sp) - stl $27, ALLREGS_R27($sp) - stl $28, ALLREGS_R28($sp) - mov $sp, $19 - stl $gp, ALLREGS_R29($sp) + SAVE_ALL ldi $8, 0x3fff - stl $31, ALLREGS_R31($sp) bic $sp, $8, $8 + mov $sp, $19 + ldl $0, PT_REGS_PS($sp) + and $0, 8, $0 /* user mode ? */ + beq $0, 1f + ldi $26, ret_from_sys_call + call $31, do_entUnaUser /* return to ret_from_syscall */ +1: ldl $9, PT_REGS_GP($sp) call $26, do_entUna - ldl $0, ALLREGS_R0($sp) - ldl $1, ALLREGS_R1($sp) - ldl $2, ALLREGS_R2($sp) - ldl $3, ALLREGS_R3($sp) - ldl $4, ALLREGS_R4($sp) - ldl $5, ALLREGS_R5($sp) - ldl $6, ALLREGS_R6($sp) - ldl $7, ALLREGS_R7($sp) - ldl $8, ALLREGS_R8($sp) - ldl $9, ALLREGS_R9($sp) - ldl $10, ALLREGS_R10($sp) - ldl $11, ALLREGS_R11($sp) - ldl $12, ALLREGS_R12($sp) - ldl $13, ALLREGS_R13($sp) - ldl $14, ALLREGS_R14($sp) - ldl $15, ALLREGS_R15($sp) - /* 16-18 HMCODE-saved */ - ldl $19, ALLREGS_R19($sp) - ldl $20, ALLREGS_R20($sp) - ldl $21, ALLREGS_R21($sp) - ldl $22, ALLREGS_R22($sp) - ldl $23, ALLREGS_R23($sp) - ldl $24, ALLREGS_R24($sp) - ldl $25, ALLREGS_R25($sp) - ldl $26, ALLREGS_R26($sp) - ldl $27, ALLREGS_R27($sp) - ldl $28, ALLREGS_R28($sp) - ldl $gp, ALLREGS_R29($sp) - ldi $sp, ALLREGS_PS($sp) + stl $9, PT_REGS_GP($sp) + RESTORE_ALL sys_call HMC_rti .end entUna

- .align 4 - .ent entUnaUser -entUnaUser: - ldl $0, ALLREGS_R0($sp) /* restore original $0 */ - ldi $sp, ALLREGS_PS($sp) /* pop entUna's stack frame */ - SAVE_ALL /* setup normal kernel stack */ - ldi $sp, -SWITCH_STACK_RA($sp) - stl $9, SWITCH_STACK_R9($sp) - stl $10, SWITCH_STACK_R10($sp) - stl $11, SWITCH_STACK_R11($sp) - stl $12, SWITCH_STACK_R12($sp) - stl $13, SWITCH_STACK_R13($sp) - stl $14, SWITCH_STACK_R14($sp) - stl $15, SWITCH_STACK_R15($sp) - ldi $8, 0x3fff - addl $sp, SWITCH_STACK_RA, $19 - bic $sp, $8, $8 - call $26, do_entUnaUser - ldl $9, SWITCH_STACK_R9($sp) - ldl $10, SWITCH_STACK_R10($sp) - ldl $11, SWITCH_STACK_R11($sp) - ldl $12, SWITCH_STACK_R12($sp) - ldl $13, SWITCH_STACK_R13($sp) - ldl $14, SWITCH_STACK_R14($sp) - ldl $15, SWITCH_STACK_R15($sp) - ldi $sp, SWITCH_STACK_RA($sp) - br ret_from_sys_call - .end entUnaUser - - /* * The system call entry point is special. Most importantly, it looks * like a function call to userspace as far as clobbered registers. We diff --git a/arch/sw_64/kernel/traps.c b/arch/sw_64/kernel/traps.c index b7c4e45df87b..6cd3ade905d1 100644 --- a/arch/sw_64/kernel/traps.c +++ b/arch/sw_64/kernel/traps.c @@ -321,41 +321,35 @@ do_entIF(unsigned long inst_type, struct pt_regs *regs) force_sig_fault(SIGILL, ILL_ILLOPC, (void __user *)regs->pc, 0); }

-/* - * entUna has a different register layout to be reasonably simple. It - * needs access to all the integer registers (the kernel doesn't use - * fp-regs), and it needs to have them in order for simpler access. - * - * Due to the non-standard register layout (and because we don't want - * to handle floating-point regs), user-mode unaligned accesses are - * handled separately by do_entUnaUser below. - * - * Oh, btw, we don't handle the "gp" register correctly, but if we fault - * on a gp-register unaligned load/store, something is _very_ wrong - * in the kernel anyway.. - */ -struct allregs { - unsigned long regs[32]; - unsigned long ps, pc, gp, a0, a1, a2; -}; - struct unaligned_stat { unsigned long count, va, pc; } unaligned[2];

/* Macro for exception fixup code to access integer registers. */ -#define una_reg(r) (_regs[(r) >= 16 && (r) <= 18 ? (r) + 19 : (r)]) +#define R(x) ((size_t) &((struct pt_regs *)0)->x) + +static int regoffsets[32] = { + R(r0), R(r1), R(r2), R(r3), R(r4), R(r5), R(r6), R(r7), R(r8), + R(r9), R(r10), R(r11), R(r12), R(r13), R(r14), R(r15), + R(r16), R(r17), R(r18), + R(r19), R(r20), R(r21), R(r22), R(r23), R(r24), R(r25), R(r26), + R(r27), R(r28), R(gp), + 0, 0 +}; + +#undef R + +#define una_reg(r) (*(unsigned long *)((char *)regs + regoffsets[r]))

asmlinkage void do_entUna(void *va, unsigned long opcode, unsigned long reg, - struct allregs *regs) + struct pt_regs *regs) { long error; unsigned long tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp8; unsigned long pc = regs->pc - 4; - unsigned long *_regs = regs->regs; const struct exception_table_entry *fixup;

unaligned[0].count++; @@ -566,29 +560,7 @@ do_entUna(void *va, unsigned long opcode, unsigned long reg, printk("%s(%d): unhandled unaligned exception\n", current->comm, task_pid_nr(current));

- printk("pc = [<%016lx>] ra = [<%016lx>] ps = %04lx\n", - pc, una_reg(26), regs->ps); - printk("r0 = %016lx r1 = %016lx r2 = %016lx\n", - una_reg(0), una_reg(1), una_reg(2)); - printk("r3 = %016lx r4 = %016lx r5 = %016lx\n", - una_reg(3), una_reg(4), una_reg(5)); - printk("r6 = %016lx r7 = %016lx r8 = %016lx\n", - una_reg(6), una_reg(7), una_reg(8)); - printk("r9 = %016lx r10= %016lx r11= %016lx\n", - una_reg(9), una_reg(10), una_reg(11)); - printk("r12= %016lx r13= %016lx r14= %016lx\n", - una_reg(12), una_reg(13), una_reg(14)); - printk("r15= %016lx\n", una_reg(15)); - printk("r16= %016lx r17= %016lx r18= %016lx\n", - una_reg(16), una_reg(17), una_reg(18)); - printk("r19= %016lx r20= %016lx r21= %016lx\n", - una_reg(19), una_reg(20), una_reg(21)); - printk("r22= %016lx r23= %016lx r24= %016lx\n", - una_reg(22), una_reg(23), una_reg(24)); - printk("r25= %016lx r27= %016lx r28= %016lx\n", - una_reg(25), una_reg(27), una_reg(28)); - printk("gp = %016lx sp = %p\n", regs->gp, regs+1); - + dik_show_regs(regs); dik_show_code((unsigned int *)pc); dik_show_trace((unsigned long *)(regs+1), KERN_DEFAULT);

@@ -668,20 +640,6 @@ s_reg_to_mem(unsigned long s_reg) 1L << 0x2c | 1L << 0x2d | /* stw stl */ \ 1L << 0x0d | 1L << 0x0e) /* sth stb */

-#define R(x) ((size_t) &((struct pt_regs *)0)->x) - -static int unauser_reg_offsets[32] = { - R(r0), R(r1), R(r2), R(r3), R(r4), R(r5), R(r6), R(r7), R(r8), - /* r9 ... r15 are stored in front of regs. */ - -56, -48, -40, -32, -24, -16, -8, - R(r16), R(r17), R(r18), - R(r19), R(r20), R(r21), R(r22), R(r23), R(r24), R(r25), R(r26), - R(r27), R(r28), R(gp), - 0, 0 -}; - -#undef R - asmlinkage void do_entUnaUser(void __user *va, unsigned long opcode, unsigned long reg, struct pt_regs *regs) @@ -734,7 +692,7 @@ do_entUnaUser(void __user *va, unsigned long opcode, /* it's an integer load/store */ if (reg < 30) { reg_addr = (unsigned long *) - ((char *)regs + unauser_reg_offsets[reg]); + ((char *)regs + regoffsets[reg]); } else if (reg == 30) { /* usp in HMCODE regs */ fake_reg = rdusp();

From: He Sheng hesheng@wxiat.com

Sunway inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I5GF7A

--------------------------------

Based on the previous series of patches, the do_switch_stack, undo_switch_stack and struct switch_stack become unused. Herein they are cleaned up.

Signed-off-by: He Sheng hesheng@wxiat.com

Signed-off-by: Gu Zitao guzitao@wxiat.com --- arch/sw_64/include/uapi/asm/ptrace.h | 15 ---- arch/sw_64/kernel/asm-offsets.c | 11 --- arch/sw_64/kernel/entry.S | 129 --------------------------- arch/sw_64/kernel/traps.c | 5 +- 4 files changed, 2 insertions(+), 158 deletions(-)

diff --git a/arch/sw_64/include/uapi/asm/ptrace.h b/arch/sw_64/include/uapi/asm/ptrace.h index 4549e8d4bf57..deb7e4096c93 100644 --- a/arch/sw_64/include/uapi/asm/ptrace.h +++ b/arch/sw_64/include/uapi/asm/ptrace.h @@ -54,21 +54,6 @@ struct pt_regs { unsigned long r18; };

-/* - * This is the extended stack used by signal handlers and the context - * switcher: it's pushed after the normal "struct pt_regs". - */ -struct switch_stack { - unsigned long r9; - unsigned long r10; - unsigned long r11; - unsigned long r12; - unsigned long r13; - unsigned long r14; - unsigned long r15; - unsigned long r26; -}; - #define PTRACE_GETREGS 12 /* get general purpose registers */ #define PTRACE_SETREGS 13 /* set general purpose registers */ #define PTRACE_GETFPREGS 14 /* get floating-point registers */ diff --git a/arch/sw_64/kernel/asm-offsets.c b/arch/sw_64/kernel/asm-offsets.c index 0e8a814f47ce..5dc59346996f 100644 --- a/arch/sw_64/kernel/asm-offsets.c +++ b/arch/sw_64/kernel/asm-offsets.c @@ -100,17 +100,6 @@ void foo(void) DEFINE(PT_REGS_R18, offsetof(struct pt_regs, r18)); BLANK();

- DEFINE(SWITCH_STACK_SIZE, sizeof(struct switch_stack)); - DEFINE(SWITCH_STACK_R9, offsetof(struct switch_stack, r9)); - DEFINE(SWITCH_STACK_R10, offsetof(struct switch_stack, r10)); - DEFINE(SWITCH_STACK_R11, offsetof(struct switch_stack, r11)); - DEFINE(SWITCH_STACK_R12, offsetof(struct switch_stack, r12)); - DEFINE(SWITCH_STACK_R13, offsetof(struct switch_stack, r13)); - DEFINE(SWITCH_STACK_R14, offsetof(struct switch_stack, r14)); - DEFINE(SWITCH_STACK_R15, offsetof(struct switch_stack, r15)); - DEFINE(SWITCH_STACK_RA, offsetof(struct switch_stack, r26)); - BLANK(); - DEFINE(KVM_REGS_SIZE, sizeof(struct kvm_regs)); DEFINE(KVM_REGS_R0, offsetof(struct kvm_regs, r0)); DEFINE(KVM_REGS_R1, offsetof(struct kvm_regs, r1)); diff --git a/arch/sw_64/kernel/entry.S b/arch/sw_64/kernel/entry.S index f54b02c1b3a9..977c774ad799 100644 --- a/arch/sw_64/kernel/entry.S +++ b/arch/sw_64/kernel/entry.S @@ -380,135 +380,6 @@ $syscall_trace_failed: br ret_from_sys_call .end strace

- .align 4 - .ent do_switch_stack -do_switch_stack: - ldi $sp, -SWITCH_STACK_SIZE($sp) - flds $f31, 0($sp) /* fillde hint */ - stl $9, SWITCH_STACK_R9($sp) - stl $10, SWITCH_STACK_R10($sp) - stl $11, SWITCH_STACK_R11($sp) - stl $12, SWITCH_STACK_R12($sp) - stl $13, SWITCH_STACK_R13($sp) - stl $14, SWITCH_STACK_R14($sp) - stl $15, SWITCH_STACK_R15($sp) - stl $26, SWITCH_STACK_RA($sp) - // SIMD-FP - ldl $9, TI_TASK($8) - ldi $9, TASK_THREAD($9) - ldi $10, THREAD_CTX_FP($9) - vstd $f0, CTX_FP_F0($10) - vstd $f1, CTX_FP_F1($10) - vstd $f2, CTX_FP_F2($10) - vstd $f3, CTX_FP_F3($10) - vstd $f4, CTX_FP_F4($10) - vstd $f5, CTX_FP_F5($10) - vstd $f6, CTX_FP_F6($10) - vstd $f7, CTX_FP_F7($10) - vstd $f8, CTX_FP_F8($10) - vstd $f9, CTX_FP_F9($10) - vstd $f10, CTX_FP_F10($10) - vstd $f11, CTX_FP_F11($10) - vstd $f12, CTX_FP_F12($10) - vstd $f13, CTX_FP_F13($10) - vstd $f14, CTX_FP_F14($10) - vstd $f15, CTX_FP_F15($10) - vstd $f16, CTX_FP_F16($10) - vstd $f17, CTX_FP_F17($10) - vstd $f18, CTX_FP_F18($10) - vstd $f19, CTX_FP_F19($10) - vstd $f20, CTX_FP_F20($10) - vstd $f21, CTX_FP_F21($10) - vstd $f22, CTX_FP_F22($10) - vstd $f23, CTX_FP_F23($10) - vstd $f24, CTX_FP_F24($10) - vstd $f25, CTX_FP_F25($10) - vstd $f26, CTX_FP_F26($10) - vstd $f27, CTX_FP_F27($10) - rfpcr $f0 - vstd $f28, CTX_FP_F28($10) - vstd $f29, CTX_FP_F29($10) - vstd $f30, CTX_FP_F30($10) - fstd $f0, THREAD_FPCR($9) - vldd $f0, CTX_FP_F0($10) - ldl $9, SWITCH_STACK_R9($sp) - ldl $10, SWITCH_STACK_R10($sp) - ret $31, ($1), 1 - .end do_switch_stack - - .align 4 - .ent undo_switch_stack -undo_switch_stack: -#ifdef CONFIG_SUBARCH_C3B - fillcs 0($sp) /* prefetch */ -#endif - ldl $11, SWITCH_STACK_R11($sp) - ldl $12, SWITCH_STACK_R12($sp) - ldl $13, SWITCH_STACK_R13($sp) - ldl $14, SWITCH_STACK_R14($sp) - ldl $15, SWITCH_STACK_R15($sp) - ldl $26, SWITCH_STACK_RA($sp) - // SIMD-FP - ldl $9, TI_TASK($8) - ldi $9, TASK_THREAD($9) - fldd $f0, THREAD_FPCR($9) - wfpcr $f0 - fimovd $f0, $10 - and $10, 0x3, $10 - beq $10, $setfpec_0 - subl $10, 0x1, $10 - beq $10, $setfpec_1 - subl $10, 0x1, $10 - beq $10, $setfpec_2 - setfpec3 - br $setfpec_over -$setfpec_0: - setfpec0 - br $setfpec_over -$setfpec_1: - setfpec1 - br $setfpec_over -$setfpec_2: - setfpec2 -$setfpec_over: - ldi $10, THREAD_CTX_FP($9) - vldd $f0, CTX_FP_F0($10) - vldd $f1, CTX_FP_F1($10) - vldd $f2, CTX_FP_F2($10) - vldd $f3, CTX_FP_F3($10) - vldd $f4, CTX_FP_F4($10) - vldd $f5, CTX_FP_F5($10) - vldd $f6, CTX_FP_F6($10) - vldd $f7, CTX_FP_F7($10) - vldd $f8, CTX_FP_F8($10) - vldd $f9, CTX_FP_F9($10) - vldd $f10, CTX_FP_F10($10) - vldd $f11, CTX_FP_F11($10) - vldd $f12, CTX_FP_F12($10) - vldd $f13, CTX_FP_F13($10) - vldd $f14, CTX_FP_F14($10) - vldd $f15, CTX_FP_F15($10) - vldd $f16, CTX_FP_F16($10) - vldd $f17, CTX_FP_F17($10) - vldd $f18, CTX_FP_F18($10) - vldd $f19, CTX_FP_F19($10) - vldd $f20, CTX_FP_F20($10) - vldd $f21, CTX_FP_F21($10) - vldd $f22, CTX_FP_F22($10) - vldd $f23, CTX_FP_F23($10) - vldd $f24, CTX_FP_F24($10) - vldd $f25, CTX_FP_F25($10) - vldd $f26, CTX_FP_F26($10) - vldd $f27, CTX_FP_F27($10) - vldd $f28, CTX_FP_F28($10) - vldd $f29, CTX_FP_F29($10) - vldd $f30, CTX_FP_F30($10) - ldl $9, SWITCH_STACK_R9($sp) - ldl $10, SWITCH_STACK_R10($sp) - ldi $sp, SWITCH_STACK_SIZE($sp) - ret $31, ($1), 1 - .end undo_switch_stack - /* * Integer register context switch * The callee-saved registers must be saved and restored. diff --git a/arch/sw_64/kernel/traps.c b/arch/sw_64/kernel/traps.c index 6cd3ade905d1..19fbf50b6ebc 100644 --- a/arch/sw_64/kernel/traps.c +++ b/arch/sw_64/kernel/traps.c @@ -300,9 +300,8 @@ do_entIF(unsigned long inst_type, struct pt_regs *regs) case 3: /* FEN fault */ /* * Irritating users can call HMC_clrfen to disable the - * FPU for the process. The kernel will then trap in - * do_switch_stack and undo_switch_stack when we try - * to save and restore the FP registers. + * FPU for the process. The kernel will then trap to + * save and restore the FP registers.

* Given that GCC by default generates code that uses the * FP registers, HMC_clrfen is not useful except for DoS

From: He Sheng hesheng@wxiat.com

Sunway inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I5GF7A

--------------------------------

This struct should be invisible to user space, and we will define new regs struct for user in uapi later.

Signed-off-by: He Sheng hesheng@wxiat.com

Signed-off-by: Gu Zitao guzitao@wxiat.com --- arch/sw_64/include/asm/ptrace.h | 47 +++++++++++++++++++++++++ arch/sw_64/include/uapi/asm/ptrace.h | 52 ---------------------------- 2 files changed, 47 insertions(+), 52 deletions(-)

diff --git a/arch/sw_64/include/asm/ptrace.h b/arch/sw_64/include/asm/ptrace.h index 0f0de9a3f8ce..85f7a47fdee3 100644 --- a/arch/sw_64/include/asm/ptrace.h +++ b/arch/sw_64/include/asm/ptrace.h @@ -9,6 +9,53 @@ #include <asm/processor.h> #include <asm/page.h>

+/* + * This struct defines the way the registers are stored on the + * kernel stack during a system call or other kernel entry + */ + +struct pt_regs { + unsigned long r0; + unsigned long r1; + unsigned long r2; + unsigned long r3; + unsigned long r4; + unsigned long r5; + unsigned long r6; + unsigned long r7; + unsigned long r8; + unsigned long r9; + unsigned long r10; + unsigned long r11; + unsigned long r12; + unsigned long r13; + unsigned long r14; + unsigned long r15; + /* r16 ~ r18 saved by hmcode */ + unsigned long r19; + unsigned long r20; + unsigned long r21; + unsigned long r22; + unsigned long r23; + unsigned long r24; + unsigned long r25; + unsigned long r26; + unsigned long r27; + unsigned long r28; + unsigned long hae; +/* JRP - These are the values provided to a0-a2 by HMcode */ + unsigned long trap_a0; + unsigned long trap_a1; + unsigned long trap_a2; +/* These are saved by HMcode: */ + unsigned long ps; + unsigned long pc; + unsigned long gp; + unsigned long r16; + unsigned long r17; + unsigned long r18; +}; + #define arch_has_single_step() (1) #define user_mode(regs) (((regs)->ps & 8) != 0) #define instruction_pointer(regs) ((regs)->pc) diff --git a/arch/sw_64/include/uapi/asm/ptrace.h b/arch/sw_64/include/uapi/asm/ptrace.h index deb7e4096c93..ac7ce6b6510b 100644 --- a/arch/sw_64/include/uapi/asm/ptrace.h +++ b/arch/sw_64/include/uapi/asm/ptrace.h @@ -2,58 +2,6 @@ #ifndef _UAPI_ASM_SW64_PTRACE_H #define _UAPI_ASM_SW64_PTRACE_H

- -/* - * This struct defines the way the registers are stored on the - * kernel stack during a system call or other kernel entry - * - * NOTE! I want to minimize the overhead of system calls, so this - * struct has as little information as possible. I does not have - * floating point regs, because the kernel doesn't change those - */ - -struct pt_regs { - unsigned long r0; - unsigned long r1; - unsigned long r2; - unsigned long r3; - unsigned long r4; - unsigned long r5; - unsigned long r6; - unsigned long r7; - unsigned long r8; - unsigned long r9; - unsigned long r10; - unsigned long r11; - unsigned long r12; - unsigned long r13; - unsigned long r14; - unsigned long r15; - /* r16 ~ r18 saved by hmcode */ - unsigned long r19; - unsigned long r20; - unsigned long r21; - unsigned long r22; - unsigned long r23; - unsigned long r24; - unsigned long r25; - unsigned long r26; - unsigned long r27; - unsigned long r28; - unsigned long hae; -/* JRP - These are the values provided to a0-a2 by HMcode */ - unsigned long trap_a0; - unsigned long trap_a1; - unsigned long trap_a2; -/* These are saved by HMcode: */ - unsigned long ps; - unsigned long pc; - unsigned long gp; - unsigned long r16; - unsigned long r17; - unsigned long r18; -}; - #define PTRACE_GETREGS 12 /* get general purpose registers */ #define PTRACE_SETREGS 13 /* set general purpose registers */ #define PTRACE_GETFPREGS 14 /* get floating-point registers */

From: He Chuyue hechuyue@wxiat.com

Sunway inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I56OLG

--------------------------------

In these methods, the local variable `y` is redundant and can be removed.

Signed-off-by: He Chuyue hechuyue@wxiat.com

Signed-off-by: Gu Zitao guzitao@wxiat.com --- arch/sw_64/mm/physaddr.c | 30 +++++++++++++----------------- 1 file changed, 13 insertions(+), 17 deletions(-)

diff --git a/arch/sw_64/mm/physaddr.c b/arch/sw_64/mm/physaddr.c index fbb489ae4db5..26769f0bf7bf 100644 --- a/arch/sw_64/mm/physaddr.c +++ b/arch/sw_64/mm/physaddr.c @@ -5,34 +5,30 @@

unsigned long __phys_addr(unsigned long x) { - unsigned long y = x; - - if (y >= __START_KERNEL_map) { - y -= __START_KERNEL_map; - VIRTUAL_BUG_ON(y >= KERNEL_IMAGE_SIZE); + if (x >= __START_KERNEL_map) { + x -= __START_KERNEL_map; + VIRTUAL_BUG_ON(x >= KERNEL_IMAGE_SIZE); } else { - VIRTUAL_BUG_ON(y < PAGE_OFFSET); - y -= PAGE_OFFSET; - VIRTUAL_BUG_ON(!phys_addr_valid(y)); + VIRTUAL_BUG_ON(x < PAGE_OFFSET); + x -= PAGE_OFFSET; + VIRTUAL_BUG_ON(!phys_addr_valid(x)); } - return y; + return x; } EXPORT_SYMBOL(__phys_addr);

bool __virt_addr_valid(unsigned long x) { - unsigned long y = x; - - if (y >= __START_KERNEL_map) { - y -= __START_KERNEL_map; - if (y >= KERNEL_IMAGE_SIZE) + if (x >= __START_KERNEL_map) { + x -= __START_KERNEL_map; + if (x >= KERNEL_IMAGE_SIZE) return false; } else { - if (y < PAGE_OFFSET) + if (x < PAGE_OFFSET) return false; - y -= PAGE_OFFSET; + x -= PAGE_OFFSET; }

- return pfn_valid(y >> PAGE_SHIFT); + return pfn_valid(x >> PAGE_SHIFT); } EXPORT_SYMBOL(__virt_addr_valid);

From: He Chuyue hechuyue@wxiat.com

Sunway inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I56OLG

--------------------------------

If kernel image is so large that it corrupts dtb, system will break down early to give a message.

Signed-off-by: He Chuyue hechuyue@wxiat.com

Signed-off-by: Gu Zitao guzitao@wxiat.com --- arch/sw_64/kernel/setup.c | 8 +++++++- 1 file changed, 7 insertions(+), 1 deletion(-)

diff --git a/arch/sw_64/kernel/setup.c b/arch/sw_64/kernel/setup.c index ca19445ac883..faae7cb3c5a1 100644 --- a/arch/sw_64/kernel/setup.c +++ b/arch/sw_64/kernel/setup.c @@ -565,8 +565,14 @@ static void __init setup_machine_fdt(void) /* Give a chance to select kernel builtin DTB firstly */ if (IS_ENABLED(CONFIG_SW64_BUILTIN_DTB)) dt_virt = (void *)__dtb_start; - else + else { dt_virt = (void *)sunway_boot_params->dtb_start; + if (dt_virt < (void *)__bss_stop) { + pr_emerg("BUG: DTB has been corrupted by kernel image!\n"); + while (true) + cpu_relax(); + } + }

phys_addr = __phys_addr((unsigned long)dt_virt); if (!phys_addr_valid(phys_addr) ||

From: Zhu Donghong zhudonghong@wxiat.com

Sunway inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I5G7AK

--------------------------------

This adds an irqchip driver for the secondary interrupt controllers based on sw64 chip3 cpu builtin LPC.

Signed-off-by: Zhu Donghong zhudonghong@wxiat.com

Signed-off-by: Gu Zitao guzitao@wxiat.com --- drivers/irqchip/Kconfig | 7 ++ drivers/irqchip/Makefile | 1 + drivers/irqchip/irq-lpc.c | 137 +++++++++++++++++++++++++++++++++ drivers/mfd/lpc_sunway_chip3.c | 130 ------------------------------- 4 files changed, 145 insertions(+), 130 deletions(-) create mode 100644 drivers/irqchip/irq-lpc.c

diff --git a/drivers/irqchip/Kconfig b/drivers/irqchip/Kconfig index 375ba3b30b72..5bf6cf60999b 100644 --- a/drivers/irqchip/Kconfig +++ b/drivers/irqchip/Kconfig @@ -9,6 +9,13 @@ config SW64_INTC The INTC controls devices interrupts and connects them to each core's local interrupt controller.

+config SW64_CHIP3_LPC + bool "SW64 Chip3 Buildin LPC Interrupt Controller" + depends on SW64_CHIP3 + help + This enables support for the LPC interrupt controller bultin in + on chip3 series. + config IRQCHIP def_bool y depends on OF_IRQ diff --git a/drivers/irqchip/Makefile b/drivers/irqchip/Makefile index 4c78b0f64e6c..78eb12ab4d4c 100644 --- a/drivers/irqchip/Makefile +++ b/drivers/irqchip/Makefile @@ -28,6 +28,7 @@ obj-$(CONFIG_ARCH_SUNXI) += irq-sunxi-nmi.o obj-$(CONFIG_ARCH_SPEAR3XX) += spear-shirq.o obj-$(CONFIG_ARM_GIC) += irq-gic.o irq-gic-common.o obj-$(CONFIG_SW64_INTC) += irq-intc-v1.o +obj-$(CONFIG_SW64_CHIP3_LPC) += irq-lpc.o obj-$(CONFIG_ARM_GIC_PM) += irq-gic-pm.o obj-$(CONFIG_ARCH_REALVIEW) += irq-gic-realview.o obj-$(CONFIG_ARM_GIC_V2M) += irq-gic-v2m.o diff --git a/drivers/irqchip/irq-lpc.c b/drivers/irqchip/irq-lpc.c new file mode 100644 index 000000000000..1cbf87478242 --- /dev/null +++ b/drivers/irqchip/irq-lpc.c @@ -0,0 +1,137 @@ +// SPDX-License-Identifier: GPL-2.0 +#include <linux/bitops.h> +#include <linux/irq.h> +#include <linux/of_address.h> +#include <linux/of_irq.h> +#include <linux/irqchip.h> +#include <linux/irqchip/chained_irq.h> +#include <linux/interrupt.h> + +#define LPC_NR_IRQS 16 +#define LPC_IRQ 0x4 +#define LPC_IRQ_MASK 0x8 + +struct lpc_intc_data { + struct irq_domain *domain; + struct irq_chip_generic *gc; +}; + +static void lpc_irq_mask_ack(struct irq_data *data) +{ + struct irq_chip_generic *gc = irq_data_get_irq_chip_data(data); + struct irq_chip_type *ct = irq_data_get_chip_type(data); + unsigned int mask = data->mask; + + irq_gc_lock(gc); + *ct->mask_cache |= mask; + irq_reg_writel(gc, *ct->mask_cache, ct->regs.mask); + irq_reg_writel(gc, mask, ct->regs.ack); + irq_gc_unlock(gc); +} + +static void lpc_irq_handler(struct irq_desc *desc) +{ + struct lpc_intc_data *b = irq_desc_get_handler_data(desc); + struct irq_chip *chip = irq_desc_get_chip(desc); + unsigned int irq; + u32 status; + + chained_irq_enter(chip, desc); + + status = irq_reg_readl(b->gc, LPC_IRQ); + + if (status == 0) { + raw_spin_lock(&desc->lock); + handle_bad_irq(desc); + raw_spin_unlock(&desc->lock); + goto out; + } + + while (status) { + irq = __ffs(status); + status &= ~BIT(irq); + generic_handle_irq(irq_find_mapping(b->domain, irq)); + } + +out: + chained_irq_exit(chip, desc); +} + +static int __init lpc_intc_of_init(struct device_node *np, + struct device_node *parent) +{ + unsigned int set = IRQ_NOPROBE | IRQ_LEVEL; + struct lpc_intc_data *data; + struct irq_chip_type *ct; + int parent_irq, ret; + void __iomem *base; + int hwirq = 0; + + data = kzalloc(sizeof(*data), GFP_KERNEL); + if (!data) + return -ENOMEM; + + base = of_iomap(np, 0); + if (!base) { + pr_err("failed to remap lpc intc registers\n"); + ret = -ENOMEM; + goto out_free; + } + + parent_irq = irq_of_parse_and_map(np, 0); + if (!parent_irq) { + pr_err("failed to find parent interrupt\n"); + ret = -EINVAL; + goto out_unmap; + } + + data->domain = irq_domain_add_linear(np, LPC_NR_IRQS, + &irq_generic_chip_ops, NULL); + if (!data->domain) { + ret = -ENOMEM; + goto out_unmap; + } + + /* Allocate a single Generic IRQ chip for this node */ + ret = irq_alloc_domain_generic_chips(data->domain, 16, 1, np->name, + handle_level_irq, 0, set, + IRQ_GC_INIT_MASK_CACHE); + if (ret) { + pr_err("failed to allocate generic irq chip\n"); + goto out_free_domain; + } + + /* Set the IRQ chaining logic */ + irq_set_chained_handler_and_data(parent_irq, + lpc_irq_handler, data); + + data->gc = irq_get_domain_generic_chip(data->domain, 0); + data->gc->reg_base = base; + data->gc->private = data; + + ct = data->gc->chip_types; + + ct->regs.ack = LPC_IRQ; + ct->regs.mask = LPC_IRQ_MASK; + ct->chip.irq_mask = irq_gc_mask_set_bit; + ct->chip.irq_unmask = irq_gc_mask_clr_bit; + ct->chip.irq_ack = irq_gc_ack_set_bit; + ct->chip.irq_mask_ack = lpc_irq_mask_ack; + + for (hwirq = 0 ; hwirq < 16 ; hwirq++) + irq_create_mapping(data->domain, hwirq); + + /* Enable LPC interrupts */ + writel(0xffffebdd, base + LPC_IRQ_MASK); + + return 0; + +out_free_domain: + irq_domain_remove(data->domain); +out_unmap: + iounmap(base); +out_free: + kfree(data); + return ret; +} +IRQCHIP_DECLARE(sw_lpc_intc, "sw64,lpc_intc", lpc_intc_of_init); diff --git a/drivers/mfd/lpc_sunway_chip3.c b/drivers/mfd/lpc_sunway_chip3.c index 793b557195c2..878aff87c992 100644 --- a/drivers/mfd/lpc_sunway_chip3.c +++ b/drivers/mfd/lpc_sunway_chip3.c @@ -75,53 +75,16 @@ enum { LPC_DMA_SWRST = 0x70, };

-enum { - LPC_IRQ0 = 0, /* 8254 Timer */ - LPC_IRQ1, /* Keyboard */ - LPC_IRQ2, /* Reserved */ - LPC_IRQ3, /* UART */ - LPC_IRQ4, /* UART */ - LPC_IRQ5, /* LPC Parallel Port2 */ - LPC_IRQ6, /* FDC-Floppy Disk Controller */ - LPC_IRQ7, /* LPT-Parallel Port1 */ - LPC_NR_IRQS, - LPC_IRQ8, /* RTC */ - LPC_IRQ9, /* Undefined */ - LPC_IRQ10, /* Undefined */ - LPC_IRQ11, /* Undefined */ - LPC_IRQ12, /* Mouse */ - LPC_IRQ13, /* Undefined */ - LPC_IRQ14, /* Undefined */ - LPC_IRQ15, /* Undefined */ -}; - struct lpc_chip3_adapter { void __iomem *hst_regs; struct device *dev; int irq; - struct irq_chip_generic *gc; unsigned int features; };

static struct resource superio_chip3_resources[] = { { .flags = IORESOURCE_IO, - }, { - .start = LPC_IRQ1, - .flags = IORESOURCE_IRQ, - .name = "i8042_kbd_irq", - }, { - .start = LPC_IRQ12, - .flags = IORESOURCE_IRQ, - .name = "i8042_aux_irq", - }, { - .start = LPC_IRQ5, - .flags = IORESOURCE_IRQ, - .name = "uart0_irq", - }, { - .start = LPC_IRQ4, - .flags = IORESOURCE_IRQ, - .name = "uart1_irq", } };

@@ -218,75 +181,9 @@ static void lpc_fw_flash_init(struct platform_device *pdev,

}

-static u32 lpc_do_irq(struct lpc_chip3_adapter *lpc_adapter) -{ - u32 irq_status = readl_relaxed(lpc_adapter->hst_regs + LPC_IRQ); - u32 ret = irq_status; - - DBG_LPC("%s irq_status=%#x\n", __func__, irq_status); - while (irq_status) { - int hwirq = fls(irq_status) - 1; - - generic_handle_irq(hwirq); - irq_status &= ~BIT(hwirq); - } - - lpc_writel(lpc_adapter->hst_regs, LPC_IRQ, ret); - return 1; -} - -static void lpc_irq_handler_mfd(struct irq_desc *desc) -{ - unsigned int irq = irq_desc_get_irq(desc); - struct lpc_chip3_adapter *lpc_adapter = irq_get_handler_data(irq); - u32 worked = 0; - - DBG_LPC("enter %s line:%d\n", __func__, __LINE__); - - worked = lpc_do_irq(lpc_adapter); - if (worked == IRQ_HANDLED) - dev_dbg(lpc_adapter->dev, "LPC irq handled.\n"); - - DBG_LPC("leave %s line:%d\n", __func__, __LINE__); -} - -static void lpc_unmask_interrupt_all(struct lpc_chip3_adapter *lpc_adapter) -{ - lpc_writel(lpc_adapter->hst_regs, LPC_IRQ_MASK, 0); -} - -static void lpc_irq_mask_ack(struct irq_data *d) -{ - struct irq_chip_generic *gc = irq_data_get_irq_chip_data(d); - struct irq_chip_type *ct = irq_data_get_chip_type(d); - u32 mask = d->mask; - - irq_gc_lock(gc); - *ct->mask_cache |= mask; - irq_reg_writel(gc, *ct->mask_cache, ct->regs.mask); - irq_reg_writel(gc, mask, ct->regs.ack); - irq_gc_unlock(gc); -} - -static void lpc_enable_irqs(struct lpc_chip3_adapter *lpc_adapter) -{ - int interrupt = 0; - - lpc_unmask_interrupt_all(lpc_adapter); - - interrupt = lpc_readl(lpc_adapter->hst_regs, LPC_IRQ); - - lpc_writel(lpc_adapter->hst_regs, LPC_CTL, 0x1600); - interrupt = lpc_readl(lpc_adapter->hst_regs, LPC_IRQ); -} - static int lpc_chip3_probe(struct platform_device *pdev) { int ret; - int num_ct = 1; - int irq_base; - struct irq_chip_generic *gc; - struct irq_chip_type *ct; struct lpc_chip3_adapter *lpc_adapter; struct resource *mem;

@@ -312,32 +209,6 @@ static int lpc_chip3_probe(struct platform_device *pdev) lpc_adapter->dev = &pdev->dev; lpc_adapter->features = 0;

- lpc_adapter->irq = platform_get_irq(pdev, 0); - if (lpc_adapter->irq < 0) { - dev_err(&pdev->dev, "no irq resource?\n"); - return lpc_adapter->irq; /* -ENXIO */ - } - - irq_base = LPC_IRQ0; - gc = irq_alloc_generic_chip("LPC_CHIP3", num_ct, irq_base, - lpc_adapter->hst_regs, handle_level_irq); - - ct = gc->chip_types; - ct->regs.mask = LPC_IRQ_MASK; - ct->regs.ack = LPC_IRQ; - ct->chip.irq_mask = irq_gc_mask_set_bit; - ct->chip.irq_unmask = irq_gc_mask_clr_bit; - ct->chip.irq_ack = irq_gc_ack_set_bit; - ct->chip.irq_mask_ack = lpc_irq_mask_ack; - irq_setup_generic_chip(gc, IRQ_MSK(LPC_NR_IRQS), 0, 0, - IRQ_NOPROBE | IRQ_LEVEL); - - lpc_adapter->gc = gc; - - irq_set_handler_data(lpc_adapter->irq, lpc_adapter); - irq_set_chained_handler(lpc_adapter->irq, - (irq_flow_handler_t) lpc_irq_handler_mfd); - lpc_enable(lpc_adapter);

lpc_mem_flash_init(pdev, lpc_adapter); @@ -350,7 +221,6 @@ static int lpc_chip3_probe(struct platform_device *pdev) goto out_dev;

dev_info(lpc_adapter->dev, "probe succeed !\n"); - lpc_enable_irqs(lpc_adapter);

return ret;

From: He Chuyue hechuyue@wxiat.com

Sunway inclusion category: bugfix bugzilla: https://gitee.com/openeuler/kernel/issues/I5GF7A

--------------------------------

Because of previous modifications to struct pt_regs, this patch fixes struct pt_regs_offset, dbg_reg_def and perf_event_sw64_regs to keep them consistent.

Signed-off-by: He Chuyue hechuyue@wxiat.com

Signed-off-by: Gu Zitao guzitao@wxiat.com --- arch/sw_64/include/uapi/asm/perf_regs.h | 7 +++++++ arch/sw_64/kernel/kgdb.c | 14 +++++++------- arch/sw_64/kernel/ptrace.c | 7 +++++++ 3 files changed, 21 insertions(+), 7 deletions(-)

diff --git a/arch/sw_64/include/uapi/asm/perf_regs.h b/arch/sw_64/include/uapi/asm/perf_regs.h index 426ae642fcc8..1378a7397951 100644 --- a/arch/sw_64/include/uapi/asm/perf_regs.h +++ b/arch/sw_64/include/uapi/asm/perf_regs.h @@ -13,6 +13,13 @@ enum perf_event_sw64_regs { PERF_REG_SW64_R6, PERF_REG_SW64_R7, PERF_REG_SW64_R8, + PERF_REG_SW64_R9, + PERF_REG_SW64_R10, + PERF_REG_SW64_R11, + PERF_REG_SW64_R12, + PERF_REG_SW64_R13, + PERF_REG_SW64_R14, + PERF_REG_SW64_R15, PERF_REG_SW64_R19, PERF_REG_SW64_R20, PERF_REG_SW64_R21, diff --git a/arch/sw_64/kernel/kgdb.c b/arch/sw_64/kernel/kgdb.c index 491f287eede9..ac2f397f1609 100644 --- a/arch/sw_64/kernel/kgdb.c +++ b/arch/sw_64/kernel/kgdb.c @@ -34,13 +34,13 @@ struct dbg_reg_def_t dbg_reg_def[DBG_MAX_REG_NUM] = { { "r7", 8, offsetof(struct pt_regs, r7)}, { "r8", 8, offsetof(struct pt_regs, r8)},

- { "r9", 8, -1 }, - { "r10", 8, -1 }, - { "r11", 8, -1 }, - { "r12", 8, -1 }, - { "r13", 8, -1 }, - { "r14", 8, -1 }, - { "r15", 8, -1 }, + { "r9", 8, offsetof(struct pt_regs, r9)}, + { "r10", 8, offsetof(struct pt_regs, r10)}, + { "r11", 8, offsetof(struct pt_regs, r11)}, + { "r12", 8, offsetof(struct pt_regs, r12)}, + { "r13", 8, offsetof(struct pt_regs, r13)}, + { "r14", 8, offsetof(struct pt_regs, r14)}, + { "r15", 8, offsetof(struct pt_regs, r15)},

{ "r16", 8, offsetof(struct pt_regs, r16)}, { "r17", 8, offsetof(struct pt_regs, r17)}, diff --git a/arch/sw_64/kernel/ptrace.c b/arch/sw_64/kernel/ptrace.c index ce41d89a54cb..94809804e23e 100644 --- a/arch/sw_64/kernel/ptrace.c +++ b/arch/sw_64/kernel/ptrace.c @@ -619,6 +619,13 @@ static const struct pt_regs_offset regoffset_table[] = { REG_OFFSET_NAME(r6), REG_OFFSET_NAME(r7), REG_OFFSET_NAME(r8), + REG_OFFSET_NAME(r9), + REG_OFFSET_NAME(r10), + REG_OFFSET_NAME(r11), + REG_OFFSET_NAME(r12), + REG_OFFSET_NAME(r13), + REG_OFFSET_NAME(r14), + REG_OFFSET_NAME(r15), REG_OFFSET_NAME(r19), REG_OFFSET_NAME(r20), REG_OFFSET_NAME(r21),

From: Hang Xiaoqian hangxiaoqian@wxiat.com

Sunway inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I56OLG

--------------------------------

The value of physical_id in /proc/cpuinfo should be the socket_id of the processor.

Signed-off-by: Hang Xiaoqian hangxiaoqian@wxiat.com

Signed-off-by: Gu Zitao guzitao@wxiat.com --- arch/sw_64/kernel/setup.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/arch/sw_64/kernel/setup.c b/arch/sw_64/kernel/setup.c index faae7cb3c5a1..4d39db5fb1ef 100644 --- a/arch/sw_64/kernel/setup.c +++ b/arch/sw_64/kernel/setup.c @@ -907,7 +907,7 @@ show_cpuinfo(struct seq_file *f, void *slot) "physical id\t: %d\n" "bogomips\t: %lu.%02lu\n", cpu_freq, cpu_data[i].tcache.size >> 10, - cpu_to_rcid(i), + cpu_topology[i].package_id, loops_per_jiffy / (500000/HZ), (loops_per_jiffy / (5000/HZ)) % 100);

From: He Sheng hesheng@wxiat.com

Sunway inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I5G7NL

--------------------------------

The user_fpsimd_state struct has been defined to hold fpu state which is also held in thread_struct. To reuse this struct, this patch makes a little change to user_fpsimd_state, then replace struct context_fpregs and fpcr of thread_struct.

This patch also moves definition of task_pt_regs and mm_segment_t to appropriate headers to avoid compile errors.

Signed-off-by: He Sheng hesheng@wxiat.com

Signed-off-by: Gu Zitao guzitao@wxiat.com --- arch/sw_64/include/asm/processor.h | 45 +-------- arch/sw_64/include/asm/ptrace.h | 4 - arch/sw_64/include/asm/thread_info.h | 5 + arch/sw_64/include/uapi/asm/ptrace.h | 8 +- arch/sw_64/kernel/asm-offsets.c | 67 +++++++------- arch/sw_64/kernel/fpu.S | 134 +++++++++++++-------------- arch/sw_64/kernel/process.c | 2 +- arch/sw_64/kernel/ptrace.c | 51 +++------- arch/sw_64/kernel/signal.c | 13 +-- arch/sw_64/kvm/entry.S | 40 ++++---- 10 files changed, 155 insertions(+), 214 deletions(-)

diff --git a/arch/sw_64/include/asm/processor.h b/arch/sw_64/include/asm/processor.h index 67d3b4368987..08e8cc8e5428 100644 --- a/arch/sw_64/include/asm/processor.h +++ b/arch/sw_64/include/asm/processor.h @@ -9,6 +9,10 @@ #define _ASM_SW64_PROCESSOR_H

#include <linux/personality.h> /* for ADDR_LIMIT_32BIT */ +#include <asm/ptrace.h> + +#define task_pt_regs(task) \ + ((struct pt_regs *) (task_stack_page(task) + 2 * PAGE_SIZE) - 1)

/* * Returns current instruction pointer ("program counter"). @@ -37,47 +41,8 @@ #define TASK_UNMAPPED_BASE \ ((current->personality & ADDR_LIMIT_32BIT) ? 0x40000000 : UNMAPPED_BASE)

-typedef struct { - unsigned long seg; -} mm_segment_t; - -struct context_fpregs { - unsigned long f0[4]; - unsigned long f1[4]; - unsigned long f2[4]; - unsigned long f3[4]; - unsigned long f4[4]; - unsigned long f5[4]; - unsigned long f6[4]; - unsigned long f7[4]; - unsigned long f8[4]; - unsigned long f9[4]; - unsigned long f10[4]; - unsigned long f11[4]; - unsigned long f12[4]; - unsigned long f13[4]; - unsigned long f14[4]; - unsigned long f15[4]; - unsigned long f16[4]; - unsigned long f17[4]; - unsigned long f18[4]; - unsigned long f19[4]; - unsigned long f20[4]; - unsigned long f21[4]; - unsigned long f22[4]; - unsigned long f23[4]; - unsigned long f24[4]; - unsigned long f25[4]; - unsigned long f26[4]; - unsigned long f27[4]; - unsigned long f28[4]; - unsigned long f29[4]; - unsigned long f30[4]; -} __aligned(32); /* 256 bits aligned for simd */ - struct thread_struct { - struct context_fpregs ctx_fp; - unsigned long fpcr; + struct user_fpsimd_state fpstate; /* Callee-saved registers */ unsigned long ra; unsigned long s[7]; /* s0 ~ s6 */ diff --git a/arch/sw_64/include/asm/ptrace.h b/arch/sw_64/include/asm/ptrace.h index 85f7a47fdee3..ac9943015663 100644 --- a/arch/sw_64/include/asm/ptrace.h +++ b/arch/sw_64/include/asm/ptrace.h @@ -3,10 +3,8 @@ #define _ASM_SW64_PTRACE_H

#include <uapi/asm/ptrace.h> -#include <linux/sched/task_stack.h> #include <asm/hmcall.h> #include <asm/thread_info.h> -#include <asm/processor.h> #include <asm/page.h>

/* @@ -65,8 +63,6 @@ struct pt_regs { #define kernel_stack_pointer(regs) (((regs->ps) >> 4) & (TASK_SIZE - 1)) #define instruction_pointer_set(regs, val) ((regs)->pc = val)

-#define task_pt_regs(task) \ - ((struct pt_regs *) (task_stack_page(task) + 2 * PAGE_SIZE) - 1)

#define current_pt_regs() \ ((struct pt_regs *) ((char *)current_thread_info() + 2 * PAGE_SIZE) - 1) diff --git a/arch/sw_64/include/asm/thread_info.h b/arch/sw_64/include/asm/thread_info.h index cffb09fc6262..33b95f815448 100644 --- a/arch/sw_64/include/asm/thread_info.h +++ b/arch/sw_64/include/asm/thread_info.h @@ -9,6 +9,11 @@ #include <asm/types.h> #include <asm/sysinfo.h>

+typedef struct { + unsigned long seg; +} mm_segment_t; + + struct pcb_struct { unsigned long ksp; unsigned long usp; diff --git a/arch/sw_64/include/uapi/asm/ptrace.h b/arch/sw_64/include/uapi/asm/ptrace.h index 1169d87c4b9c..a45b10b8914a 100644 --- a/arch/sw_64/include/uapi/asm/ptrace.h +++ b/arch/sw_64/include/uapi/asm/ptrace.h @@ -11,9 +11,15 @@ struct user_pt_regs { __u64 pstate; };

+/* 256 bits aligned for simd */ +struct fpreg { + __u64 v[4] __attribute__((aligned(32))); +}; + struct user_fpsimd_state { - __u64 vregs[124]; + struct fpreg fp[31]; __u64 fpcr; + __u64 __reserved[3]; };

/* PTRACE_ATTACH is 16 */ diff --git a/arch/sw_64/kernel/asm-offsets.c b/arch/sw_64/kernel/asm-offsets.c index 5dc59346996f..0c7e1e26eb05 100644 --- a/arch/sw_64/kernel/asm-offsets.c +++ b/arch/sw_64/kernel/asm-offsets.c @@ -178,40 +178,39 @@ void foo(void) DEFINE(HOST_INT_R16, offsetof(struct host_int_args, r16)); BLANK();

- DEFINE(TASK_THREAD, offsetof(struct task_struct, thread)); - DEFINE(THREAD_CTX_FP, offsetof(struct thread_struct, ctx_fp)); - DEFINE(THREAD_FPCR, offsetof(struct thread_struct, fpcr)); - DEFINE(CTX_FP_F0, offsetof(struct context_fpregs, f0)); - DEFINE(CTX_FP_F1, offsetof(struct context_fpregs, f1)); - DEFINE(CTX_FP_F2, offsetof(struct context_fpregs, f2)); - DEFINE(CTX_FP_F3, offsetof(struct context_fpregs, f3)); - DEFINE(CTX_FP_F4, offsetof(struct context_fpregs, f4)); - DEFINE(CTX_FP_F5, offsetof(struct context_fpregs, f5)); - DEFINE(CTX_FP_F6, offsetof(struct context_fpregs, f6)); - DEFINE(CTX_FP_F7, offsetof(struct context_fpregs, f7)); - DEFINE(CTX_FP_F8, offsetof(struct context_fpregs, f8)); - DEFINE(CTX_FP_F9, offsetof(struct context_fpregs, f9)); - DEFINE(CTX_FP_F10, offsetof(struct context_fpregs, f10)); - DEFINE(CTX_FP_F11, offsetof(struct context_fpregs, f11)); - DEFINE(CTX_FP_F12, offsetof(struct context_fpregs, f12)); - DEFINE(CTX_FP_F13, offsetof(struct context_fpregs, f13)); - DEFINE(CTX_FP_F14, offsetof(struct context_fpregs, f14)); - DEFINE(CTX_FP_F15, offsetof(struct context_fpregs, f15)); - DEFINE(CTX_FP_F16, offsetof(struct context_fpregs, f16)); - DEFINE(CTX_FP_F17, offsetof(struct context_fpregs, f17)); - DEFINE(CTX_FP_F18, offsetof(struct context_fpregs, f18)); - DEFINE(CTX_FP_F19, offsetof(struct context_fpregs, f19)); - DEFINE(CTX_FP_F20, offsetof(struct context_fpregs, f20)); - DEFINE(CTX_FP_F21, offsetof(struct context_fpregs, f21)); - DEFINE(CTX_FP_F22, offsetof(struct context_fpregs, f22)); - DEFINE(CTX_FP_F23, offsetof(struct context_fpregs, f23)); - DEFINE(CTX_FP_F24, offsetof(struct context_fpregs, f24)); - DEFINE(CTX_FP_F25, offsetof(struct context_fpregs, f25)); - DEFINE(CTX_FP_F26, offsetof(struct context_fpregs, f26)); - DEFINE(CTX_FP_F27, offsetof(struct context_fpregs, f27)); - DEFINE(CTX_FP_F28, offsetof(struct context_fpregs, f28)); - DEFINE(CTX_FP_F29, offsetof(struct context_fpregs, f29)); - DEFINE(CTX_FP_F30, offsetof(struct context_fpregs, f30)); + OFFSET(TASK_THREAD, task_struct, thread); + OFFSET(TASK_THREAD_F0, task_struct, thread.fpstate.fp[0]); + OFFSET(TASK_THREAD_F1, task_struct, thread.fpstate.fp[1]); + OFFSET(TASK_THREAD_F2, task_struct, thread.fpstate.fp[2]); + OFFSET(TASK_THREAD_F3, task_struct, thread.fpstate.fp[3]); + OFFSET(TASK_THREAD_F4, task_struct, thread.fpstate.fp[4]); + OFFSET(TASK_THREAD_F5, task_struct, thread.fpstate.fp[5]); + OFFSET(TASK_THREAD_F6, task_struct, thread.fpstate.fp[6]); + OFFSET(TASK_THREAD_F7, task_struct, thread.fpstate.fp[7]); + OFFSET(TASK_THREAD_F8, task_struct, thread.fpstate.fp[8]); + OFFSET(TASK_THREAD_F9, task_struct, thread.fpstate.fp[9]); + OFFSET(TASK_THREAD_F10, task_struct, thread.fpstate.fp[10]); + OFFSET(TASK_THREAD_F11, task_struct, thread.fpstate.fp[11]); + OFFSET(TASK_THREAD_F12, task_struct, thread.fpstate.fp[12]); + OFFSET(TASK_THREAD_F13, task_struct, thread.fpstate.fp[13]); + OFFSET(TASK_THREAD_F14, task_struct, thread.fpstate.fp[14]); + OFFSET(TASK_THREAD_F15, task_struct, thread.fpstate.fp[15]); + OFFSET(TASK_THREAD_F16, task_struct, thread.fpstate.fp[16]); + OFFSET(TASK_THREAD_F17, task_struct, thread.fpstate.fp[17]); + OFFSET(TASK_THREAD_F18, task_struct, thread.fpstate.fp[18]); + OFFSET(TASK_THREAD_F19, task_struct, thread.fpstate.fp[19]); + OFFSET(TASK_THREAD_F20, task_struct, thread.fpstate.fp[20]); + OFFSET(TASK_THREAD_F21, task_struct, thread.fpstate.fp[21]); + OFFSET(TASK_THREAD_F22, task_struct, thread.fpstate.fp[22]); + OFFSET(TASK_THREAD_F23, task_struct, thread.fpstate.fp[23]); + OFFSET(TASK_THREAD_F24, task_struct, thread.fpstate.fp[24]); + OFFSET(TASK_THREAD_F25, task_struct, thread.fpstate.fp[25]); + OFFSET(TASK_THREAD_F26, task_struct, thread.fpstate.fp[26]); + OFFSET(TASK_THREAD_F27, task_struct, thread.fpstate.fp[27]); + OFFSET(TASK_THREAD_F28, task_struct, thread.fpstate.fp[28]); + OFFSET(TASK_THREAD_F29, task_struct, thread.fpstate.fp[29]); + OFFSET(TASK_THREAD_F30, task_struct, thread.fpstate.fp[30]); + OFFSET(TASK_THREAD_FPCR, task_struct, thread.fpstate.fpcr); BLANK(); OFFSET(TASK_THREAD_RA, task_struct, thread.ra); OFFSET(TASK_THREAD_S0, task_struct, thread.s[0]); diff --git a/arch/sw_64/kernel/fpu.S b/arch/sw_64/kernel/fpu.S index 25dcf4740525..3cb3bfab08e8 100644 --- a/arch/sw_64/kernel/fpu.S +++ b/arch/sw_64/kernel/fpu.S @@ -8,49 +8,46 @@ .set noat ENTRY(__fpstate_save) /* a0: prev task */ - ldi a0, TASK_THREAD(a0) - ldi t0, THREAD_CTX_FP(a0) - vstd $f0, CTX_FP_F0(t0) - vstd $f1, CTX_FP_F1(t0) - vstd $f2, CTX_FP_F2(t0) - vstd $f3, CTX_FP_F3(t0) - vstd $f4, CTX_FP_F4(t0) - vstd $f5, CTX_FP_F5(t0) - vstd $f6, CTX_FP_F6(t0) - vstd $f7, CTX_FP_F7(t0) - vstd $f8, CTX_FP_F8(t0) - vstd $f9, CTX_FP_F9(t0) - vstd $f10, CTX_FP_F10(t0) - vstd $f11, CTX_FP_F11(t0) - vstd $f12, CTX_FP_F12(t0) - vstd $f13, CTX_FP_F13(t0) - vstd $f14, CTX_FP_F14(t0) - vstd $f15, CTX_FP_F15(t0) - vstd $f16, CTX_FP_F16(t0) - vstd $f17, CTX_FP_F17(t0) - vstd $f18, CTX_FP_F18(t0) - vstd $f19, CTX_FP_F19(t0) - vstd $f20, CTX_FP_F20(t0) - vstd $f21, CTX_FP_F21(t0) - vstd $f22, CTX_FP_F22(t0) - vstd $f23, CTX_FP_F23(t0) - vstd $f24, CTX_FP_F24(t0) - vstd $f25, CTX_FP_F25(t0) - vstd $f26, CTX_FP_F26(t0) - vstd $f27, CTX_FP_F27(t0) + vstd $f0, TASK_THREAD_F0(a0) + vstd $f1, TASK_THREAD_F1(a0) + vstd $f2, TASK_THREAD_F2(a0) + vstd $f3, TASK_THREAD_F3(a0) + vstd $f4, TASK_THREAD_F4(a0) + vstd $f5, TASK_THREAD_F5(a0) + vstd $f6, TASK_THREAD_F6(a0) + vstd $f7, TASK_THREAD_F7(a0) + vstd $f8, TASK_THREAD_F8(a0) + vstd $f9, TASK_THREAD_F9(a0) + vstd $f10, TASK_THREAD_F10(a0) + vstd $f11, TASK_THREAD_F11(a0) + vstd $f12, TASK_THREAD_F12(a0) + vstd $f13, TASK_THREAD_F13(a0) + vstd $f14, TASK_THREAD_F14(a0) + vstd $f15, TASK_THREAD_F15(a0) + vstd $f16, TASK_THREAD_F16(a0) + vstd $f17, TASK_THREAD_F17(a0) + vstd $f18, TASK_THREAD_F18(a0) + vstd $f19, TASK_THREAD_F19(a0) + vstd $f20, TASK_THREAD_F20(a0) + vstd $f21, TASK_THREAD_F21(a0) + vstd $f22, TASK_THREAD_F22(a0) + vstd $f23, TASK_THREAD_F23(a0) + vstd $f24, TASK_THREAD_F24(a0) + vstd $f25, TASK_THREAD_F25(a0) + vstd $f26, TASK_THREAD_F26(a0) + vstd $f27, TASK_THREAD_F27(a0) rfpcr $f0 - vstd $f28, CTX_FP_F28(t0) - vstd $f29, CTX_FP_F29(t0) - vstd $f30, CTX_FP_F30(t0) - fstd $f0, THREAD_FPCR(a0) - vldd $f0, CTX_FP_F0(t0) + vstd $f28, TASK_THREAD_F28(a0) + vstd $f29, TASK_THREAD_F29(a0) + vstd $f30, TASK_THREAD_F30(a0) + fstd $f0, TASK_THREAD_FPCR(a0) + vldd $f0, TASK_THREAD_F0(a0) ret END(__fpstate_save)

ENTRY(__fpstate_restore) /* a0: next task */ - ldi a0, TASK_THREAD(a0) - fldd $f0, THREAD_FPCR(a0) + fldd $f0, TASK_THREAD_FPCR(a0) wfpcr $f0 fimovd $f0, t1 and t1, 0x3, t1 @@ -70,37 +67,36 @@ $setfpec_1: $setfpec_2: setfpec2 $setfpec_over: - ldi t0, THREAD_CTX_FP(a0) - vldd $f0, CTX_FP_F0(t0) - vldd $f1, CTX_FP_F1(t0) - vldd $f2, CTX_FP_F2(t0) - vldd $f3, CTX_FP_F3(t0) - vldd $f4, CTX_FP_F4(t0) - vldd $f5, CTX_FP_F5(t0) - vldd $f6, CTX_FP_F6(t0) - vldd $f7, CTX_FP_F7(t0) - vldd $f8, CTX_FP_F8(t0) - vldd $f9, CTX_FP_F9(t0) - vldd $f10, CTX_FP_F10(t0) - vldd $f11, CTX_FP_F11(t0) - vldd $f12, CTX_FP_F12(t0) - vldd $f13, CTX_FP_F13(t0) - vldd $f14, CTX_FP_F14(t0) - vldd $f15, CTX_FP_F15(t0) - vldd $f16, CTX_FP_F16(t0) - vldd $f17, CTX_FP_F17(t0) - vldd $f18, CTX_FP_F18(t0) - vldd $f19, CTX_FP_F19(t0) - vldd $f20, CTX_FP_F20(t0) - vldd $f21, CTX_FP_F21(t0) - vldd $f22, CTX_FP_F22(t0) - vldd $f23, CTX_FP_F23(t0) - vldd $f24, CTX_FP_F24(t0) - vldd $f25, CTX_FP_F25(t0) - vldd $f26, CTX_FP_F26(t0) - vldd $f27, CTX_FP_F27(t0) - vldd $f28, CTX_FP_F28(t0) - vldd $f29, CTX_FP_F29(t0) - vldd $f30, CTX_FP_F30(t0) + vldd $f0, TASK_THREAD_F0(a0) + vldd $f1, TASK_THREAD_F1(a0) + vldd $f2, TASK_THREAD_F2(a0) + vldd $f3, TASK_THREAD_F3(a0) + vldd $f4, TASK_THREAD_F4(a0) + vldd $f5, TASK_THREAD_F5(a0) + vldd $f6, TASK_THREAD_F6(a0) + vldd $f7, TASK_THREAD_F7(a0) + vldd $f8, TASK_THREAD_F8(a0) + vldd $f9, TASK_THREAD_F9(a0) + vldd $f10, TASK_THREAD_F10(a0) + vldd $f11, TASK_THREAD_F11(a0) + vldd $f12, TASK_THREAD_F12(a0) + vldd $f13, TASK_THREAD_F13(a0) + vldd $f14, TASK_THREAD_F14(a0) + vldd $f15, TASK_THREAD_F15(a0) + vldd $f16, TASK_THREAD_F16(a0) + vldd $f17, TASK_THREAD_F17(a0) + vldd $f18, TASK_THREAD_F18(a0) + vldd $f19, TASK_THREAD_F19(a0) + vldd $f20, TASK_THREAD_F20(a0) + vldd $f21, TASK_THREAD_F21(a0) + vldd $f22, TASK_THREAD_F22(a0) + vldd $f23, TASK_THREAD_F23(a0) + vldd $f24, TASK_THREAD_F24(a0) + vldd $f25, TASK_THREAD_F25(a0) + vldd $f26, TASK_THREAD_F26(a0) + vldd $f27, TASK_THREAD_F27(a0) + vldd $f28, TASK_THREAD_F28(a0) + vldd $f29, TASK_THREAD_F29(a0) + vldd $f30, TASK_THREAD_F30(a0) ret END(__fpstate_restore) diff --git a/arch/sw_64/kernel/process.c b/arch/sw_64/kernel/process.c index 6b5f0f3a8345..308675d29c04 100644 --- a/arch/sw_64/kernel/process.c +++ b/arch/sw_64/kernel/process.c @@ -256,7 +256,7 @@ EXPORT_SYMBOL(dump_elf_task); int dump_elf_task_fp(elf_fpreg_t *dest, struct task_struct *task) { - memcpy(dest, &task->thread.ctx_fp, 32 * 8); + memcpy(dest, &task->thread.fpstate, 32 * 8); return 1; } EXPORT_SYMBOL(dump_elf_task_fp); diff --git a/arch/sw_64/kernel/ptrace.c b/arch/sw_64/kernel/ptrace.c index 94fba3569781..bdbd0d97a130 100644 --- a/arch/sw_64/kernel/ptrace.c +++ b/arch/sw_64/kernel/ptrace.c @@ -9,6 +9,7 @@ #include <linux/audit.h> #include <linux/regset.h> #include <linux/elf.h> +#include <linux/sched/task_stack.h>

#include <asm/reg.h> #include <asm/asm-offsets.h> @@ -55,9 +56,6 @@ enum { REG_GP = 29 };

-#define FP_REG(fp_regno, vector_regno) \ - (fp_regno * 32 + vector_regno * 8) - #define R(x) ((size_t) &((struct pt_regs *)0)->x)

short regoffsets[32] = { @@ -91,8 +89,8 @@ static unsigned long zero; static unsigned long * get_reg_addr(struct task_struct *task, unsigned long regno) { - unsigned long *addr; - int fp_regno, vector_regno; + void *addr; + int fno, vno;

switch (regno) { case USP: @@ -108,9 +106,8 @@ get_reg_addr(struct task_struct *task, unsigned long regno) addr = (void *)task_pt_regs(task) + regoffsets[regno]; break; case FPREG_BASE ... FPREG_END: - fp_regno = regno - FPREG_BASE; - vector_regno = 0; - addr = (void *)((unsigned long)&task->thread.ctx_fp + FP_REG(fp_regno, vector_regno)); + fno = regno - FPREG_BASE; + addr = &task->thread.fpstate.fp[fno].v[0]; break; case VECREG_BASE ... VECREG_END: /* @@ -121,12 +118,12 @@ get_reg_addr(struct task_struct *task, unsigned long regno) addr = &zero; break; } - fp_regno = (regno - VECREG_BASE) & 0x1f; - vector_regno = 1 + ((regno - VECREG_BASE) >> 5); - addr = (void *)((unsigned long)&task->thread.ctx_fp + FP_REG(fp_regno, vector_regno)); + fno = (regno - VECREG_BASE) & 0x1f; + vno = 1 + ((regno - VECREG_BASE) >> 5); + addr = &task->thread.fpstate.fp[fno].v[vno]; break; case FPCR: - addr = (void *)&task->thread.fpcr; + addr = &task->thread.fpstate.fpcr; break; case PC: addr = (void *)task_pt_regs(task) + PT_REGS_PC; @@ -322,17 +319,9 @@ static int fpr_get(struct task_struct *target, const struct user_regset *regset, struct membuf to) { - int ret; - struct user_fpsimd_state uregs; - - memcpy(uregs.vregs, &target->thread.ctx_fp, - sizeof(struct context_fpregs)); - - uregs.fpcr = target->thread.fpcr; - - ret = membuf_write(&to, &uregs, sizeof(uregs));

- return ret; + return membuf_write(&to, &target->thread.fpstate, + sizeof(struct user_fpsimd_state)); }

static int fpr_set(struct task_struct *target, @@ -340,21 +329,9 @@ static int fpr_set(struct task_struct *target, unsigned int pos, unsigned int count, const void *kbuf, const void __user *ubuf) { - int ret; - struct user_fpsimd_state uregs; - - ret = user_regset_copyin(&pos, &count, &kbuf, &ubuf, - &uregs, 0, sizeof(uregs)); - - if (ret) - return ret; - - memcpy(&target->thread.ctx_fp, uregs.vregs, - sizeof(struct context_fpregs)); - - target->thread.fpcr = uregs.fpcr; - - return ret; + return user_regset_copyin(&pos, &count, &kbuf, &ubuf, + &target->thread.fpstate, 0, + sizeof(struct user_fpsimd_state)); }

enum sw64_regset { diff --git a/arch/sw_64/kernel/signal.c b/arch/sw_64/kernel/signal.c index 887326812624..6a6203ccb04f 100644 --- a/arch/sw_64/kernel/signal.c +++ b/arch/sw_64/kernel/signal.c @@ -100,9 +100,10 @@ restore_sigcontext(struct sigcontext __user *sc, struct pt_regs *regs) err |= __get_user(usp, sc->sc_regs+30); wrusp(usp); /* simd-fp */ - err |= __copy_from_user(&current->thread.ctx_fp, - &sc->sc_fpregs, sizeof(struct context_fpregs)); - err |= __get_user(current->thread.fpcr, &sc->sc_fpcr); + err |= __copy_from_user(&current->thread.fpstate, &sc->sc_fpregs, + offsetof(struct user_fpsimd_state, fpcr)); + err |= __get_user(current->thread.fpstate.fpcr, &sc->sc_fpcr); + if (likely(!err)) __fpstate_restore(current);

@@ -230,9 +231,9 @@ setup_sigcontext(struct sigcontext __user *sc, struct pt_regs *regs, err |= __put_user(0, sc->sc_regs+31); /* simd-fp */ __fpstate_save(current); - err |= __copy_to_user(&sc->sc_fpregs, - &current->thread.ctx_fp, sizeof(struct context_fpregs)); - err |= __put_user(current->thread.fpcr, &sc->sc_fpcr); + err |= __copy_to_user(&sc->sc_fpregs, &current->thread.fpstate, + offsetof(struct user_fpsimd_state, fpcr)); + err |= __put_user(current->thread.fpstate.fpcr, &sc->sc_fpcr);

err |= __put_user(regs->trap_a0, &sc->sc_traparg_a0); err |= __put_user(regs->trap_a1, &sc->sc_traparg_a1); diff --git a/arch/sw_64/kvm/entry.S b/arch/sw_64/kvm/entry.S index 1e089da5378e..0c02b68ee7d0 100644 --- a/arch/sw_64/kvm/entry.S +++ b/arch/sw_64/kvm/entry.S @@ -15,18 +15,16 @@ ENTRY(__sw64_vcpu_run)

/* save host fpregs */ ldl $1, TI_TASK($8) - ldi $1, TASK_THREAD($1) rfpcr $f0 - fstd $f0, THREAD_FPCR($1) - ldi $1, THREAD_CTX_FP($1) - vstd $f2, CTX_FP_F2($1) - vstd $f3, CTX_FP_F3($1) - vstd $f4, CTX_FP_F4($1) - vstd $f5, CTX_FP_F5($1) - vstd $f6, CTX_FP_F6($1) - vstd $f7, CTX_FP_F7($1) - vstd $f8, CTX_FP_F8($1) - vstd $f9, CTX_FP_F9($1) + fstd $f0, TASK_THREAD_FPCR($1) + vstd $f2, TASK_THREAD_F2($1) + vstd $f3, TASK_THREAD_F3($1) + vstd $f4, TASK_THREAD_F4($1) + vstd $f5, TASK_THREAD_F5($1) + vstd $f6, TASK_THREAD_F6($1) + vstd $f7, TASK_THREAD_F7($1) + vstd $f8, TASK_THREAD_F8($1) + vstd $f9, TASK_THREAD_F9($1)

ldi sp, -VCPU_RET_SIZE(sp) /* r16 = guest kvm_vcpu_arch.vcb struct pointer */ @@ -213,8 +211,7 @@ $g_setfpec_over: bic sp, $8, $8 /* restore host fpregs */ ldl $1, TI_TASK($8) - ldi $1, TASK_THREAD($1) - fldd $f0, THREAD_FPCR($1) + fldd $f0, TASK_THREAD_FPCR($1) wfpcr $f0 fimovd $f0, $2 and $2, 0x3, $2 @@ -234,15 +231,14 @@ $setfpec_1: $setfpec_2: setfpec2 $setfpec_over: - ldi $1, THREAD_CTX_FP($1) - vldd $f2, CTX_FP_F2($1) - vldd $f3, CTX_FP_F3($1) - vldd $f4, CTX_FP_F4($1) - vldd $f5, CTX_FP_F5($1) - vldd $f6, CTX_FP_F6($1) - vldd $f7, CTX_FP_F7($1) - vldd $f8, CTX_FP_F8($1) - vldd $f9, CTX_FP_F9($1) + vldd $f2, TASK_THREAD_F2($1) + vldd $f3, TASK_THREAD_F3($1) + vldd $f4, TASK_THREAD_F4($1) + vldd $f5, TASK_THREAD_F5($1) + vldd $f6, TASK_THREAD_F6($1) + vldd $f7, TASK_THREAD_F7($1) + vldd $f8, TASK_THREAD_F8($1) + vldd $f9, TASK_THREAD_F9($1)

/* if $0 > 0, handle hcall */ bgt $0, $ret_to

From: He Sheng hesheng@wxiat.com

Sunway inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I56OLG

--------------------------------

For gpreg, simplify dump method to make it easier to maintain.

For fpreg, redefine elf_fpregset_t with struct user_fpsimd_state and add a arch-specific dump_fpu() method.

Signed-off-by: He Sheng hesheng@wxiat.com

Signed-off-by: Gu Zitao guzitao@wxiat.com --- arch/sw_64/include/asm/elf.h | 34 +++++-------------- arch/sw_64/kernel/process.c | 66 ++++++++---------------------------- 2 files changed, 23 insertions(+), 77 deletions(-)

diff --git a/arch/sw_64/include/asm/elf.h b/arch/sw_64/include/asm/elf.h index abecc5653eba..8c858cff5573 100644 --- a/arch/sw_64/include/asm/elf.h +++ b/arch/sw_64/include/asm/elf.h @@ -55,23 +55,18 @@ #define EF_SW64_32BIT 1 /* All addresses are below 2GB */

/* - * ELF register definitions.. - */ - -/* - * The legacy version of <sys/procfs.h> makes gregset_t 46 entries long. - * I have no idea why that is so. For now, we just leave it at 33 - * (32 general regs + processor status word). + * ELF register definitions. + * + * For now, we just leave it at 33 (32 general regs + processor status word). */ #define ELF_NGREG 33 -#define ELF_NFPREG 32 -

typedef unsigned long elf_greg_t; typedef elf_greg_t elf_gregset_t[ELF_NGREG];

-typedef double elf_fpreg_t; -typedef elf_fpreg_t elf_fpregset_t[ELF_NFPREG]; +/* Same with user_fpsimd_state */ +#include <uapi/asm/ptrace.h> +typedef struct user_fpsimd_state elf_fpregset_t;

/* * This is used to ensure we don't load something for the wrong architecture. @@ -121,22 +116,9 @@ extern int arch_setup_additional_pages(struct linux_binprm *bprm,

#ifdef __KERNEL__ struct pt_regs; -struct thread_info; struct task_struct; -extern void dump_elf_thread(elf_greg_t *dest, struct pt_regs *pt, - struct thread_info *ti); -#define ELF_CORE_COPY_REGS(DEST, REGS) \ - dump_elf_thread(DEST, REGS, current_thread_info()); - -/* Similar, but for a thread other than current. */ - -extern int dump_elf_task(elf_greg_t *dest, struct task_struct *task); -#define ELF_CORE_COPY_TASK_REGS(TASK, DEST) dump_elf_task(*(DEST), TASK) - -/* Similar, but for the FP registers. */ - -extern int dump_elf_task_fp(elf_fpreg_t *dest, struct task_struct *task); -#define ELF_CORE_COPY_FPREGS(TASK, DEST) dump_elf_task_fp(*(DEST), TASK) +extern void sw64_elf_core_copy_regs(elf_greg_t *dest, struct pt_regs *pt); +#define ELF_CORE_COPY_REGS(DEST, REGS) sw64_elf_core_copy_regs(DEST, REGS);

/* * This yields a mask that user programs can use to figure out what diff --git a/arch/sw_64/kernel/process.c b/arch/sw_64/kernel/process.c index 308675d29c04..7a7578d530c6 100644 --- a/arch/sw_64/kernel/process.c +++ b/arch/sw_64/kernel/process.c @@ -200,66 +200,30 @@ copy_thread(unsigned long clone_flags, unsigned long usp,

/* * Fill in the user structure for a ELF core dump. + * @regs: should be signal_pt_regs() or task_pt_reg(task) */ -void -dump_elf_thread(elf_greg_t *dest, struct pt_regs *pt, struct thread_info *ti) +void sw64_elf_core_copy_regs(elf_greg_t *dest, struct pt_regs *regs) { - dest[0] = pt->r0; - dest[1] = pt->r1; - dest[2] = pt->r2; - dest[3] = pt->r3; - dest[4] = pt->r4; - dest[5] = pt->r5; - dest[6] = pt->r6; - dest[7] = pt->r7; - dest[8] = pt->r8; - dest[9] = pt->r9; - dest[10] = pt->r10; - dest[11] = pt->r11; - dest[12] = pt->r12; - dest[13] = pt->r13; - dest[14] = pt->r14; - dest[15] = pt->r15; - dest[16] = pt->r16; - dest[17] = pt->r17; - dest[18] = pt->r18; - dest[19] = pt->r19; - dest[20] = pt->r20; - dest[21] = pt->r21; - dest[22] = pt->r22; - dest[23] = pt->r23; - dest[24] = pt->r24; - dest[25] = pt->r25; - dest[26] = pt->r26; - dest[27] = pt->r27; - dest[28] = pt->r28; - dest[29] = pt->gp; - dest[30] = ti == current_thread_info() ? rdusp() : ti->pcb.usp; - dest[31] = pt->pc; + int i; + struct thread_info *ti;

- /* Once upon a time this was the PS value. Which is stupid - * since that is always 8 for usermode. Usurped for the more - * useful value of the thread's UNIQUE field. - */ - dest[32] = ti->pcb.unique; -} -EXPORT_SYMBOL(dump_elf_thread); + ti = (void *)((__u64)regs & ~(THREAD_SIZE - 1));

-int -dump_elf_task(elf_greg_t *dest, struct task_struct *task) -{ - dump_elf_thread(dest, task_pt_regs(task), task_thread_info(task)); - return 1; + for (i = 0; i < 30; i++) + dest[i] = *(__u64 *)((void *)regs + regoffsets[i]); + dest[30] = ti == current_thread_info() ? rdusp() : ti->pcb.usp; + dest[31] = regs->pc; + dest[32] = ti->pcb.unique; } -EXPORT_SYMBOL(dump_elf_task); +EXPORT_SYMBOL(sw64_elf_core_copy_regs);

-int -dump_elf_task_fp(elf_fpreg_t *dest, struct task_struct *task) +/* Fill in the fpu structure for a core dump. */ +int dump_fpu(struct pt_regs *regs, elf_fpregset_t *fpu) { - memcpy(dest, &task->thread.fpstate, 32 * 8); + memcpy(fpu, &current->thread.fpstate, sizeof(*fpu)); return 1; } -EXPORT_SYMBOL(dump_elf_task_fp); +EXPORT_SYMBOL(dump_fpu);

/* * Under heavy swap load I've seen this lose in an ugly way. So do

From: Xu Chenjiao xuchenjiao@wxiat.com

Sunway inclusion category: bugfix bugzilla: https://gitee.com/openeuler/kernel/issues/I56QAM

--------------------------------

IO register DEVINT_WKEN, DEVINTWK_INTEN, PME_ENABLE_INTD_CORE0 and AER_ENABLE_INTD_CORE0 are only defined in chip3. This will lead to compile errors if other chip is selected. To avoid these errors, we define default set_devint_wken() and set_pcieport_service_irq() as weak symbols.

Signed-off-by: Xu Chenjiao xuchenjiao@wxiat.com

Signed-off-by: Gu Zitao guzitao@wxiat.com --- arch/sw_64/chip/chip3/chip.c | 16 ++++++++++++++++ arch/sw_64/kernel/pci.c | 15 ++++----------- 2 files changed, 20 insertions(+), 11 deletions(-)

diff --git a/arch/sw_64/chip/chip3/chip.c b/arch/sw_64/chip/chip3/chip.c index acd25fe99669..b73afc9c62ac 100644 --- a/arch/sw_64/chip/chip3/chip.c +++ b/arch/sw_64/chip/chip3/chip.c @@ -90,6 +90,22 @@ void setup_chip_clocksource(void) #endif }

+void set_devint_wken(int node) +{ + unsigned long val; + + /* enable INTD wakeup */ + val = 0x80; + sw64_io_write(node, DEVINT_WKEN, val); + sw64_io_write(node, DEVINTWK_INTEN, val); +} + +void set_pcieport_service_irq(int node, int index) +{ + write_piu_ior0(node, index, PMEINTCONFIG, PME_ENABLE_INTD_CORE0); + write_piu_ior0(node, index, AERERRINTCONFIG, AER_ENABLE_INTD_CORE0); +} + static int chip3_get_cpu_nums(void) { unsigned long trkmode; diff --git a/arch/sw_64/kernel/pci.c b/arch/sw_64/kernel/pci.c index 401ee0b781be..a004ea9fde7f 100644 --- a/arch/sw_64/kernel/pci.c +++ b/arch/sw_64/kernel/pci.c @@ -600,15 +600,7 @@ sw64_init_host(unsigned long node, unsigned long index) } }

-static void set_devint_wken(int node) -{ - unsigned long val; - - /* enable INTD wakeup */ - val = 0x80; - sw64_io_write(node, DEVINT_WKEN, val); - sw64_io_write(node, DEVINTWK_INTEN, val); -} +void __weak set_devint_wken(int node) {}

void __init sw64_init_arch(void) { @@ -651,6 +643,8 @@ void __init sw64_init_arch(void) } }

+void __weak set_pcieport_service_irq(int node, int index) {} + static void __init sw64_init_intx(struct pci_controller *hose) { unsigned long int_conf, node, val_node; @@ -679,8 +673,7 @@ static void __init sw64_init_intx(struct pci_controller *hose) if (sw64_chip_init->pci_init.set_intx) sw64_chip_init->pci_init.set_intx(node, index, int_conf);

- write_piu_ior0(node, index, PMEINTCONFIG, PME_ENABLE_INTD_CORE0); - write_piu_ior0(node, index, AERERRINTCONFIG, AER_ENABLE_INTD_CORE0); + set_pcieport_service_irq(node, index); }

void __init sw64_init_irq(void)

From: Xu Chenjiao xuchenjiao@wxiat.com

Sunway inclusion category: bugfix bugzilla: https://gitee.com/openeuler/kernel/issues/I56XUF

--------------------------------

On SW64 platform, we used to enable PME and AER interrupt by default, however, there is a risk that no body cares the interrupt routine when CONFIG_PCIE_PME=n or CONFIG_PCIEAER=n. So let's fix this problem.

Signed-off-by: Xu Chenjiao xuchenjiao@wxiat.com

Signed-off-by: Gu Zitao guzitao@wxiat.com --- arch/sw_64/chip/chip3/chip.c | 30 +++++++++++++++++++----------- 1 file changed, 19 insertions(+), 11 deletions(-)

diff --git a/arch/sw_64/chip/chip3/chip.c b/arch/sw_64/chip/chip3/chip.c index b73afc9c62ac..2103d93a53a2 100644 --- a/arch/sw_64/chip/chip3/chip.c +++ b/arch/sw_64/chip/chip3/chip.c @@ -102,8 +102,11 @@ void set_devint_wken(int node)

void set_pcieport_service_irq(int node, int index) { - write_piu_ior0(node, index, PMEINTCONFIG, PME_ENABLE_INTD_CORE0); - write_piu_ior0(node, index, AERERRINTCONFIG, AER_ENABLE_INTD_CORE0); + if (IS_ENABLED(CONFIG_PCIE_PME)) + write_piu_ior0(node, index, PMEINTCONFIG, PME_ENABLE_INTD_CORE0); + + if (IS_ENABLED(CONFIG_PCIEAER)) + write_piu_ior0(node, index, AERERRINTCONFIG, AER_ENABLE_INTD_CORE0); }

static int chip3_get_cpu_nums(void) @@ -438,7 +441,7 @@ extern struct pci_controller *hose_head, **hose_tail; static void sw6_handle_intx(unsigned int offset) { struct pci_controller *hose; - unsigned long value, pme_value, aer_value; + unsigned long value;

hose = hose_head; for (hose = hose_head; hose; hose = hose->next) { @@ -451,15 +454,20 @@ static void sw6_handle_intx(unsigned int offset) write_piu_ior0(hose->node, hose->index, INTACONFIG + (offset << 7), value); }

- pme_value = read_piu_ior0(hose->node, hose->index, PMEINTCONFIG); - aer_value = read_piu_ior0(hose->node, hose->index, AERERRINTCONFIG); - if ((pme_value >> 63) || (aer_value >> 63)) { - handle_irq(hose->service_irq); + if (IS_ENABLED(CONFIG_PCIE_PME)) { + value = read_piu_ior0(hose->node, hose->index, PMEINTCONFIG); + if (value >> 63) { + handle_irq(hose->service_irq); + write_piu_ior0(hose->node, hose->index, PMEINTCONFIG, value); + } + }

- if (pme_value >> 63) - write_piu_ior0(hose->node, hose->index, PMEINTCONFIG, pme_value); - if (aer_value >> 63) - write_piu_ior0(hose->node, hose->index, AERERRINTCONFIG, aer_value); + if (IS_ENABLED(CONFIG_PCIEAER)) { + value = read_piu_ior0(hose->node, hose->index, AERERRINTCONFIG); + if (value >> 63) { + handle_irq(hose->service_irq); + write_piu_ior0(hose->node, hose->index, AERERRINTCONFIG, value); + } }

if (hose->iommu_enable) {