From 45ec8860be2f681b2e3e521a3fb4554ec29fbaac Mon Sep 17 00:00:00 2001 From: Max Filippov Date: Sun, 19 Jan 2014 20:00:48 +0400 Subject: xtensa: export ccount_freq Now that ccount_freq is used in udelay and ndelay it needs to be exported in order to be available to modules. Signed-off-by: Max Filippov --- arch/xtensa/kernel/time.c | 1 + 1 file changed, 1 insertion(+) (limited to 'arch/xtensa/kernel') diff --git a/arch/xtensa/kernel/time.c b/arch/xtensa/kernel/time.c index 08b769d3b3a1..2a1823de69cc 100644 --- a/arch/xtensa/kernel/time.c +++ b/arch/xtensa/kernel/time.c @@ -30,6 +30,7 @@ #include unsigned long ccount_freq; /* ccount Hz */ +EXPORT_SYMBOL(ccount_freq); static cycle_t ccount_read(struct clocksource *cs) { -- cgit v1.2.3 From e2fd1374c705abe4661df3fb6fadb3879c7c1846 Mon Sep 17 00:00:00 2001 From: Max Filippov Date: Wed, 22 Jan 2014 08:04:43 +0400 Subject: xtensa: introduce spill_registers_kernel macro Most in-kernel users want registers spilled on the kernel stack and don't require PS.EXCM to be set. That means that they don't need fixup routine and could reuse regular window overflow mechanism for that, which makes spill routine very simple. Cc: stable@vger.kernel.org Suggested-by: Chris Zankel Signed-off-by: Max Filippov --- arch/xtensa/kernel/entry.S | 60 ++++++++++++++++++++++++++++++++++++---------- 1 file changed, 48 insertions(+), 12 deletions(-) (limited to 'arch/xtensa/kernel') diff --git a/arch/xtensa/kernel/entry.S b/arch/xtensa/kernel/entry.S index 21dbe6bdb8ed..0489918e40d2 100644 --- a/arch/xtensa/kernel/entry.S +++ b/arch/xtensa/kernel/entry.S @@ -1794,6 +1794,43 @@ ENTRY(system_call) ENDPROC(system_call) +/* + * Spill live registers on the kernel stack macro. + * + * Entry condition: ps.woe is set, ps.excm is cleared + * Exit condition: windowstart has single bit set + * May clobber: a12, a13 + */ + .macro spill_registers_kernel + +#if XCHAL_NUM_AREGS > 16 + call12 1f + _j 2f + retw + .align 4 +1: + _entry a1, 48 + addi a12, a0, 3 +#if XCHAL_NUM_AREGS > 32 + .rept (XCHAL_NUM_AREGS - 32) / 12 + _entry a1, 48 + mov a12, a0 + .endr +#endif + _entry a1, 48 +#if XCHAL_NUM_AREGS % 12 == 0 + mov a8, a8 +#elif XCHAL_NUM_AREGS % 12 == 4 + mov a12, a12 +#elif XCHAL_NUM_AREGS % 12 == 8 + mov a4, a4 +#endif + retw +2: +#else + mov a12, a12 +#endif + .endm /* * Task switch. @@ -1806,21 +1843,20 @@ ENTRY(_switch_to) entry a1, 16 - mov a12, a2 # preserve 'prev' (a2) - mov a13, a3 # and 'next' (a3) + mov a10, a2 # preserve 'prev' (a2) + mov a11, a3 # and 'next' (a3) l32i a4, a2, TASK_THREAD_INFO l32i a5, a3, TASK_THREAD_INFO - save_xtregs_user a4 a6 a8 a9 a10 a11 THREAD_XTREGS_USER + save_xtregs_user a4 a6 a8 a9 a12 a13 THREAD_XTREGS_USER - s32i a0, a12, THREAD_RA # save return address - s32i a1, a12, THREAD_SP # save stack pointer + s32i a0, a10, THREAD_RA # save return address + s32i a1, a10, THREAD_SP # save stack pointer /* Disable ints while we manipulate the stack pointer. */ - movi a14, (1 << PS_EXCM_BIT) | LOCKLEVEL - xsr a14, ps + rsil a14, LOCKLEVEL rsr a3, excsave1 rsync s32i a3, a3, EXC_TABLE_FIXUP /* enter critical section */ @@ -1835,7 +1871,7 @@ ENTRY(_switch_to) /* Flush register file. */ - call0 _spill_registers # destroys a3, a4, and SAR + spill_registers_kernel /* Set kernel stack (and leave critical section) * Note: It's save to set it here. The stack will not be overwritten @@ -1851,13 +1887,13 @@ ENTRY(_switch_to) /* restore context of the task 'next' */ - l32i a0, a13, THREAD_RA # restore return address - l32i a1, a13, THREAD_SP # restore stack pointer + l32i a0, a11, THREAD_RA # restore return address + l32i a1, a11, THREAD_SP # restore stack pointer - load_xtregs_user a5 a6 a8 a9 a10 a11 THREAD_XTREGS_USER + load_xtregs_user a5 a6 a8 a9 a12 a13 THREAD_XTREGS_USER wsr a14, ps - mov a2, a12 # return 'prev' + mov a2, a10 # return 'prev' rsync retw -- cgit v1.2.3 From 3251f1e27a5a17f0efd436cfd1e7b9896cfab0a0 Mon Sep 17 00:00:00 2001 From: Max Filippov Date: Wed, 30 Oct 2013 16:18:25 +0400 Subject: xtensa: save current register frame in fast_syscall_spill_registers_fixup We need it saved because it contains a3 where we track which register windows we still need to spill, and fixup handler may call C exception handlers. Also fix comments. Cc: stable@vger.kernel.org Signed-off-by: Max Filippov --- arch/xtensa/kernel/entry.S | 12 ++++++++++++ 1 file changed, 12 insertions(+) (limited to 'arch/xtensa/kernel') diff --git a/arch/xtensa/kernel/entry.S b/arch/xtensa/kernel/entry.S index 0489918e40d2..b61e25146a2f 100644 --- a/arch/xtensa/kernel/entry.S +++ b/arch/xtensa/kernel/entry.S @@ -1117,6 +1117,13 @@ ENDPROC(fast_syscall_spill_registers) * We basically restore WINDOWBASE and WINDOWSTART to the condition when * we entered the spill routine and jump to the user exception handler. * + * Note that we only need to restore the bits in windowstart that have not + * been spilled yet by the _spill_register routine. Luckily, a3 contains a + * rotated windowstart with only those bits set for frames that haven't been + * spilled yet. Because a3 is rotated such that bit 0 represents the register + * frame for the current windowbase - 1, we need to rotate a3 left by the + * value of the current windowbase + 1 and move it to windowstart. + * * a0: value of depc, original value in depc * a2: trashed, original value in EXC_TABLE_DOUBLE_SAVE * a3: exctable, original value in excsave1 @@ -1131,10 +1138,15 @@ ENTRY(fast_syscall_spill_registers_fixup) /* We need to make sure the current registers (a0-a3) are preserved. * To do this, we simply set the bit for the current window frame * in WS, so that the exception handlers save them to the task stack. + * + * Note: we use a3 to set the windowbase, so we take a special care + * of it, saving it in the original _spill_registers frame across + * the exception handler call. */ xsr a3, excsave1 # get spill-mask slli a3, a3, 1 # shift left by one + addi a3, a3, 1 # set the bit for the current window frame slli a2, a3, 32-WSBITS src a2, a3, a2 # a2 = xxwww1yyxxxwww1yy...... -- cgit v1.2.3 From 6b5a1f74e50170e64104135490dc32b657483594 Mon Sep 17 00:00:00 2001 From: Chris Zankel Date: Tue, 28 Jan 2014 22:09:51 -0800 Subject: xtensa: fix fast_syscall_spill_registers The original implementation could clobber registers under certain conditions. The Xtensa processor architecture uses windowed registers and the original implementation was using a4 as a temporary register, which under certain conditions could be register a0 of the oldest window frame, and didn't always restore the content correctly. By moving the _spill_registers routine inside the fast system call, it frees up one more register (the return address is not required anymore) for the spill routine. Signed-off-by: Chris Zankel --- arch/xtensa/kernel/entry.S | 383 ++++++++++++++++++++------------------------- 1 file changed, 174 insertions(+), 209 deletions(-) (limited to 'arch/xtensa/kernel') diff --git a/arch/xtensa/kernel/entry.S b/arch/xtensa/kernel/entry.S index b61e25146a2f..ef7f4990722b 100644 --- a/arch/xtensa/kernel/entry.S +++ b/arch/xtensa/kernel/entry.S @@ -1081,34 +1081,202 @@ ENTRY(fast_syscall_spill_registers) rsr a0, sar s32i a3, a2, PT_AREG3 - s32i a4, a2, PT_AREG4 - s32i a0, a2, PT_AREG5 # store SAR to PT_AREG5 + s32i a0, a2, PT_SAR - /* The spill routine might clobber a7, a11, and a15. */ + /* The spill routine might clobber a4, a7, a8, a11, a12, and a15. */ + s32i a4, a2, PT_AREG4 s32i a7, a2, PT_AREG7 + s32i a8, a2, PT_AREG8 s32i a11, a2, PT_AREG11 + s32i a12, a2, PT_AREG12 s32i a15, a2, PT_AREG15 - call0 _spill_registers # destroys a3, a4, and SAR + /* + * Rotate ws so that the current windowbase is at bit 0. + * Assume ws = xxxwww1yy (www1 current window frame). + * Rotate ws right so that a4 = yyxxxwww1. + */ + + rsr a0, windowbase + rsr a3, windowstart # a3 = xxxwww1yy + ssr a0 # holds WB + slli a0, a3, WSBITS + or a3, a3, a0 # a3 = xxxwww1yyxxxwww1yy + srl a3, a3 # a3 = 00xxxwww1yyxxxwww1 + + /* We are done if there are no more than the current register frame. */ + + extui a3, a3, 1, WSBITS-1 # a3 = 0yyxxxwww + movi a0, (1 << (WSBITS-1)) + _beqz a3, .Lnospill # only one active frame? jump + + /* We want 1 at the top, so that we return to the current windowbase */ + + or a3, a3, a0 # 1yyxxxwww + + /* Skip empty frames - get 'oldest' WINDOWSTART-bit. */ + + wsr a3, windowstart # save shifted windowstart + neg a0, a3 + and a3, a0, a3 # first bit set from right: 000010000 + + ffs_ws a0, a3 # a0: shifts to skip empty frames + movi a3, WSBITS + sub a0, a3, a0 # WSBITS-a0:number of 0-bits from right + ssr a0 # save in SAR for later. + + rsr a3, windowbase + add a3, a3, a0 + wsr a3, windowbase + rsync + + rsr a3, windowstart + srl a3, a3 # shift windowstart + + /* WB is now just one frame below the oldest frame in the register + window. WS is shifted so the oldest frame is in bit 0, thus, WB + and WS differ by one 4-register frame. */ + + /* Save frames. Depending what call was used (call4, call8, call12), + * we have to save 4,8. or 12 registers. + */ + + +.Lloop: _bbsi.l a3, 1, .Lc4 + _bbci.l a3, 2, .Lc12 + +.Lc8: s32e a4, a13, -16 + l32e a4, a5, -12 + s32e a8, a4, -32 + s32e a5, a13, -12 + s32e a6, a13, -8 + s32e a7, a13, -4 + s32e a9, a4, -28 + s32e a10, a4, -24 + s32e a11, a4, -20 + srli a11, a3, 2 # shift windowbase by 2 + rotw 2 + _bnei a3, 1, .Lloop + j .Lexit + +.Lc4: s32e a4, a9, -16 + s32e a5, a9, -12 + s32e a6, a9, -8 + s32e a7, a9, -4 + + srli a7, a3, 1 + rotw 1 + _bnei a3, 1, .Lloop + j .Lexit + +.Lc12: _bbci.l a3, 3, .Linvalid_mask # bit 2 shouldn't be zero! + + /* 12-register frame (call12) */ + + l32e a0, a5, -12 + s32e a8, a0, -48 + mov a8, a0 + + s32e a9, a8, -44 + s32e a10, a8, -40 + s32e a11, a8, -36 + s32e a12, a8, -32 + s32e a13, a8, -28 + s32e a14, a8, -24 + s32e a15, a8, -20 + srli a15, a3, 3 + + /* The stack pointer for a4..a7 is out of reach, so we rotate the + * window, grab the stackpointer, and rotate back. + * Alternatively, we could also use the following approach, but that + * makes the fixup routine much more complicated: + * rotw 1 + * s32e a0, a13, -16 + * ... + * rotw 2 + */ + + rotw 1 + mov a4, a13 + rotw -1 + + s32e a4, a8, -16 + s32e a5, a8, -12 + s32e a6, a8, -8 + s32e a7, a8, -4 + + rotw 3 + + _beqi a3, 1, .Lexit + j .Lloop + +.Lexit: + + /* Done. Do the final rotation and set WS */ + + rotw 1 + rsr a3, windowbase + ssl a3 + movi a3, 1 + sll a3, a3 + wsr a3, windowstart +.Lnospill: /* Advance PC, restore registers and SAR, and return from exception. */ - l32i a3, a2, PT_AREG5 - l32i a4, a2, PT_AREG4 + l32i a3, a2, PT_SAR l32i a0, a2, PT_AREG0 wsr a3, sar l32i a3, a2, PT_AREG3 /* Restore clobbered registers. */ + l32i a4, a2, PT_AREG4 l32i a7, a2, PT_AREG7 + l32i a8, a2, PT_AREG8 l32i a11, a2, PT_AREG11 + l32i a12, a2, PT_AREG12 l32i a15, a2, PT_AREG15 movi a2, 0 rfe +.Linvalid_mask: + + /* We get here because of an unrecoverable error in the window + * registers, so set up a dummy frame and kill the user application. + * Note: We assume EXC_TABLE_KSTK contains a valid stack pointer. + */ + + movi a0, 1 + movi a1, 0 + + wsr a0, windowstart + wsr a1, windowbase + rsync + + movi a0, 0 + + rsr a3, excsave1 + l32i a1, a3, EXC_TABLE_KSTK + + movi a4, (1 << PS_WOE_BIT) | LOCKLEVEL + wsr a4, ps + rsync + + movi a6, SIGSEGV + movi a4, do_exit + callx4 a4 + + /* shouldn't return, so panic */ + + wsr a0, excsave1 + movi a0, unrecoverable_exception + callx0 a0 # should not return +1: j 1b + + ENDPROC(fast_syscall_spill_registers) /* Fixup handler. @@ -1232,209 +1400,6 @@ ENTRY(fast_syscall_spill_registers_fixup_return) ENDPROC(fast_syscall_spill_registers_fixup_return) -/* - * spill all registers. - * - * This is not a real function. The following conditions must be met: - * - * - must be called with call0. - * - uses a3, a4 and SAR. - * - the last 'valid' register of each frame are clobbered. - * - the caller must have registered a fixup handler - * (or be inside a critical section) - * - PS_EXCM must be set (PS_WOE cleared?) - */ - -ENTRY(_spill_registers) - - /* - * Rotate ws so that the current windowbase is at bit 0. - * Assume ws = xxxwww1yy (www1 current window frame). - * Rotate ws right so that a4 = yyxxxwww1. - */ - - rsr a4, windowbase - rsr a3, windowstart # a3 = xxxwww1yy - ssr a4 # holds WB - slli a4, a3, WSBITS - or a3, a3, a4 # a3 = xxxwww1yyxxxwww1yy - srl a3, a3 # a3 = 00xxxwww1yyxxxwww1 - - /* We are done if there are no more than the current register frame. */ - - extui a3, a3, 1, WSBITS-1 # a3 = 0yyxxxwww - movi a4, (1 << (WSBITS-1)) - _beqz a3, .Lnospill # only one active frame? jump - - /* We want 1 at the top, so that we return to the current windowbase */ - - or a3, a3, a4 # 1yyxxxwww - - /* Skip empty frames - get 'oldest' WINDOWSTART-bit. */ - - wsr a3, windowstart # save shifted windowstart - neg a4, a3 - and a3, a4, a3 # first bit set from right: 000010000 - - ffs_ws a4, a3 # a4: shifts to skip empty frames - movi a3, WSBITS - sub a4, a3, a4 # WSBITS-a4:number of 0-bits from right - ssr a4 # save in SAR for later. - - rsr a3, windowbase - add a3, a3, a4 - wsr a3, windowbase - rsync - - rsr a3, windowstart - srl a3, a3 # shift windowstart - - /* WB is now just one frame below the oldest frame in the register - window. WS is shifted so the oldest frame is in bit 0, thus, WB - and WS differ by one 4-register frame. */ - - /* Save frames. Depending what call was used (call4, call8, call12), - * we have to save 4,8. or 12 registers. - */ - - _bbsi.l a3, 1, .Lc4 - _bbsi.l a3, 2, .Lc8 - - /* Special case: we have a call12-frame starting at a4. */ - - _bbci.l a3, 3, .Lc12 # bit 3 shouldn't be zero! (Jump to Lc12 first) - - s32e a4, a1, -16 # a1 is valid with an empty spill area - l32e a4, a5, -12 - s32e a8, a4, -48 - mov a8, a4 - l32e a4, a1, -16 - j .Lc12c - -.Lnospill: - ret - -.Lloop: _bbsi.l a3, 1, .Lc4 - _bbci.l a3, 2, .Lc12 - -.Lc8: s32e a4, a13, -16 - l32e a4, a5, -12 - s32e a8, a4, -32 - s32e a5, a13, -12 - s32e a6, a13, -8 - s32e a7, a13, -4 - s32e a9, a4, -28 - s32e a10, a4, -24 - s32e a11, a4, -20 - - srli a11, a3, 2 # shift windowbase by 2 - rotw 2 - _bnei a3, 1, .Lloop - -.Lexit: /* Done. Do the final rotation, set WS, and return. */ - - rotw 1 - rsr a3, windowbase - ssl a3 - movi a3, 1 - sll a3, a3 - wsr a3, windowstart - ret - -.Lc4: s32e a4, a9, -16 - s32e a5, a9, -12 - s32e a6, a9, -8 - s32e a7, a9, -4 - - srli a7, a3, 1 - rotw 1 - _bnei a3, 1, .Lloop - j .Lexit - -.Lc12: _bbci.l a3, 3, .Linvalid_mask # bit 2 shouldn't be zero! - - /* 12-register frame (call12) */ - - l32e a2, a5, -12 - s32e a8, a2, -48 - mov a8, a2 - -.Lc12c: s32e a9, a8, -44 - s32e a10, a8, -40 - s32e a11, a8, -36 - s32e a12, a8, -32 - s32e a13, a8, -28 - s32e a14, a8, -24 - s32e a15, a8, -20 - srli a15, a3, 3 - - /* The stack pointer for a4..a7 is out of reach, so we rotate the - * window, grab the stackpointer, and rotate back. - * Alternatively, we could also use the following approach, but that - * makes the fixup routine much more complicated: - * rotw 1 - * s32e a0, a13, -16 - * ... - * rotw 2 - */ - - rotw 1 - mov a5, a13 - rotw -1 - - s32e a4, a9, -16 - s32e a5, a9, -12 - s32e a6, a9, -8 - s32e a7, a9, -4 - - rotw 3 - - _beqi a3, 1, .Lexit - j .Lloop - -.Linvalid_mask: - - /* We get here because of an unrecoverable error in the window - * registers. If we are in user space, we kill the application, - * however, this condition is unrecoverable in kernel space. - */ - - rsr a0, ps - _bbci.l a0, PS_UM_BIT, 1f - - /* User space: Setup a dummy frame and kill application. - * Note: We assume EXC_TABLE_KSTK contains a valid stack pointer. - */ - - movi a0, 1 - movi a1, 0 - - wsr a0, windowstart - wsr a1, windowbase - rsync - - movi a0, 0 - - rsr a3, excsave1 - l32i a1, a3, EXC_TABLE_KSTK - - movi a4, (1 << PS_WOE_BIT) | LOCKLEVEL - wsr a4, ps - rsync - - movi a6, SIGSEGV - movi a4, do_exit - callx4 a4 - -1: /* Kernel space: PANIC! */ - - wsr a0, excsave1 - movi a0, unrecoverable_exception - callx0 a0 # should not return -1: j 1b - -ENDPROC(_spill_registers) - #ifdef CONFIG_MMU /* * We should never get here. Bail out! -- cgit v1.2.3 From c0e50d41126e4786d9cf1105bdf783e55c99f915 Mon Sep 17 00:00:00 2001 From: Chris Zankel Date: Tue, 28 Jan 2014 22:09:51 -0800 Subject: xtensa: fix fast_syscall_spill_registers The original implementation could clobber registers under certain conditions. The Xtensa processor architecture uses windowed registers and the original implementation was using a4 as a temporary register, which under certain conditions could be register a0 of the oldest window frame, and didn't always restore the content correctly. By moving the _spill_registers routine inside the fast system call, it frees up one more register (the return address is not required anymore) for the spill routine. Signed-off-by: Chris Zankel --- arch/xtensa/kernel/vectors.S | 2 +- arch/xtensa/kernel/xtensa_ksyms.c | 2 -- 2 files changed, 1 insertion(+), 3 deletions(-) (limited to 'arch/xtensa/kernel') diff --git a/arch/xtensa/kernel/vectors.S b/arch/xtensa/kernel/vectors.S index cb8fd44caabc..f9e1ec346e35 100644 --- a/arch/xtensa/kernel/vectors.S +++ b/arch/xtensa/kernel/vectors.S @@ -235,7 +235,7 @@ ENTRY(_DoubleExceptionVector) /* Check for overflow/underflow exception, jump if overflow. */ - _bbci.l a0, 6, _DoubleExceptionVector_WindowOverflow + bbci.l a0, 6, _DoubleExceptionVector_WindowOverflow /* * Restart window underflow exception. diff --git a/arch/xtensa/kernel/xtensa_ksyms.c b/arch/xtensa/kernel/xtensa_ksyms.c index 74a60c7e085e..80b33ed51f31 100644 --- a/arch/xtensa/kernel/xtensa_ksyms.c +++ b/arch/xtensa/kernel/xtensa_ksyms.c @@ -122,9 +122,7 @@ EXPORT_SYMBOL(insw); EXPORT_SYMBOL(insl); extern long common_exception_return; -extern long _spill_registers; EXPORT_SYMBOL(common_exception_return); -EXPORT_SYMBOL(_spill_registers); #ifdef CONFIG_FUNCTION_TRACER EXPORT_SYMBOL(_mcount); -- cgit v1.2.3 From bda8932d234aeaee870ac666e776a5ba03bb13a4 Mon Sep 17 00:00:00 2001 From: Max Filippov Date: Wed, 29 Jan 2014 06:20:46 +0400 Subject: xtensa: support common clock framework Signed-off-by: Max Filippov --- arch/xtensa/kernel/setup.c | 2 ++ 1 file changed, 2 insertions(+) (limited to 'arch/xtensa/kernel') diff --git a/arch/xtensa/kernel/setup.c b/arch/xtensa/kernel/setup.c index 7d12af1317f1..84fe931bb60e 100644 --- a/arch/xtensa/kernel/setup.c +++ b/arch/xtensa/kernel/setup.c @@ -22,6 +22,7 @@ #include #include #include +#include #include #include #include @@ -276,6 +277,7 @@ void __init early_init_devtree(void *params) static int __init xtensa_device_probe(void) { + of_clk_init(NULL); of_platform_populate(NULL, of_default_bus_match_table, NULL, NULL); return 0; } -- cgit v1.2.3