From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (qmail 86860 invoked by alias); 10 Apr 2019 15:58:13 -0000 Mailing-List: contact newlib-help@sourceware.org; run by ezmlm Precedence: bulk List-Id: List-Subscribe: List-Archive: List-Post: List-Help: , Sender: newlib-owner@sourceware.org Received: (qmail 86829 invoked by uid 89); 10 Apr 2019 15:58:12 -0000 Authentication-Results: sourceware.org; auth=none X-Spam-SWARE-Status: No, score=-14.2 required=5.0 tests=AWL,BAYES_00,FREEMAIL_FROM,GIT_PATCH_0,GIT_PATCH_1,GIT_PATCH_2,GIT_PATCH_3,RCVD_IN_DNSWL_NONE,SPF_PASS autolearn=ham version=3.3.1 spammy=Back, 48, 4.8, H*f:sk:0e86fe9 X-HELO: mail-io1-f44.google.com Received: from mail-io1-f44.google.com (HELO mail-io1-f44.google.com) (209.85.166.44) by sourceware.org (qpsmtpd/0.93/v0.84-503-g423c35a) with ESMTP; Wed, 10 Apr 2019 15:58:09 +0000 Received: by mail-io1-f44.google.com with SMTP id x7so2607601ioh.4 for ; Wed, 10 Apr 2019 08:58:09 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20161025; h=mime-version:references:in-reply-to:from:date:message-id:subject:to :cc; bh=wQJlMH6EZj2F9H4KCnxmC/7/zm7JWv44VgtM6tuafLc=; b=qhIVQlqlxdvYJcr2iagWfcQ7MBk+5qvn8vP7/CxijSsG0AUL/Wv938kb5Sa2ts27rX pNk6Ha//lculiG6nf0VTrskq0BE662JRSjRdpEysy5x9S01M5Q+JA0O4nnrZxb97i1BM +2AAqvUAL7Wbo6HumJ6CZ5tYMW4Gl7kggtxFq+l8u21Er5NnWcM5ueFXer0ZgxsmM3t/ AfAFUZQ4fQSJyRkWYggQdsPIWLeVxpVQCwaNMjfIpbkzOw35pats3g/W5Yut/dsfB62S FiSQ2C5oI2O3+2cgi8LSWcF9PTSBnp7uBH7zWUqJLhgAODL1FJhvEV+FUfkqIerm5eK+ 46Mw== MIME-Version: 1.0 References: <20190410142353.GI4248@calimero.vinschen.de> <0e86fe93-50b9-82e2-c57d-8ef731080d2e@arm.com> In-Reply-To: <0e86fe93-50b9-82e2-c57d-8ef731080d2e@arm.com> From: Alexander Fedotov Date: Wed, 10 Apr 2019 15:58:00 -0000 Message-ID: Subject: Re: [Arm] SP and SL initialization refactored To: "Richard Earnshaw (lists)" Cc: Tamar Christina , Newlib , nd Content-Type: text/plain; charset="UTF-8" X-SW-Source: 2019/txt/msg00156.txt.bz2 Hi Richard New patch is below (only for libgloss). If it's fine with you I will do the same changes in newlib/libc/sys/arm as well. >This code is clearly ARM/Thumb2 only. Have you tested that this builds >for, say ARM7TDMI in thumb state? If you use a build of gcc configured >with --with-multilib-list=aprofile or --with-multilib-list=rmprofile >you'll get pretty comprehensive checking of all the variants that need >to be supported (beware, it can take quite a long time, even on a fast >machine). At the very least, this code needs to be checked on: I have built gcc with --with-multilib-list=rmprofile Unfortunately I have no ability to check on all of them. At least this code is verified on Cortex-r52 that was the aim. >This implies dropping support for Armv3 as system mode was added in >Armv4. That's probably not an issue now as GCC has also dropped such >support, but it should be added to the News. I don't see News file for newlib. But I have preserved original version in the patch. I suggest to drop Armv3 by another commit. I can't get why description is corrupted for _stack_init function in the patch below while lines are less than 80 symbols :( >From c8ff1ecf005dba2a4a69c5754b1911c0d351aba5 Mon Sep 17 00:00:00 2001 From: Alexander Fedotov Date: Wed, 23 Jan 2019 13:33:14 +0300 Subject: [PATCH] SP initialization changes: 1. set default value in semihosting case as well 2. moved existing SP & SL init code for processor modes in separate routine and made it as "hook" 3. init SP for processor modes in Thumb mode as well Add new macro FN_RETURN, FN_EH_START and FN_EH_END. --- libgloss/arm/arm.h | 26 +++++ libgloss/arm/crt0.S | 279 +++++++++++++++++++++++++++++++------------- 2 files changed, 224 insertions(+), 81 deletions(-) diff --git a/libgloss/arm/arm.h b/libgloss/arm/arm.h index 0489f2d92..04b3a5ac7 100644 --- a/libgloss/arm/arm.h +++ b/libgloss/arm/arm.h @@ -61,4 +61,30 @@ # define HAVE_CALL_INDIRECT #endif +/* A and R profiles (and legacy Arm). + Current Program Status Register (CPSR) + M[4:0] Mode bits. M[4] is always 1 for 32-bit modes. + T[5] 1: Thumb, 0: ARM instruction set + F[6] 1: disables FIQ + I[7] 1: disables IRQ + A[8] 1: disables imprecise aborts + E[9] 0: Little-endian, 1: Big-endian + J[24] 1: Jazelle instruction set + */ +#define CPSR_M_USR 0x00 /* User mode */ +#define CPSR_M_FIQ 0x01 /* Fast Interrupt mode */ +#define CPSR_M_IRQ 0x02 /* Interrupt mode */ +#define CPSR_M_SVR 0x03 /* Supervisor mode */ +#define CPSR_M_MON 0x06 /* Monitor mode */ +#define CPSR_M_ABT 0x07 /* Abort mode */ +#define CPSR_M_HYP 0x0A /* Hypervisor mode */ +#define CPSR_M_UND 0x0B /* Undefined mode */ +#define CPSR_M_SYS 0x0F /* System mode */ +#define CPSR_M_32BIT 0x10 /* 32-bit mode */ +#define CPSR_T_BIT 0x20 /* Thumb bit */ +#define CPSR_F_MASK 0x40 /* FIQ bit */ +#define CPSR_I_MASK 0x80 /* IRQ bit */ + +#define CPSR_M_MASK 0x0F /* Mode mask except M[4] */ + #endif /* _LIBGLOSS_ARM_H */ diff --git a/libgloss/arm/crt0.S b/libgloss/arm/crt0.S index c708f63d8..89256d5e8 100644 --- a/libgloss/arm/crt0.S +++ b/libgloss/arm/crt0.S @@ -59,6 +59,21 @@ .endm #endif +/* Annotation for EABI unwinding tables. */ +.macro FN_EH_START +#if defined(__ELF__) && !defined(__USING_SJLJ_EXCEPTIONS__) + .fnstart +#endif +.endm + +.macro FN_EH_END +#if defined(__ELF__) && !defined(__USING_SJLJ_EXCEPTIONS__) + /* Protect against unhandled exceptions. */ + .cantunwind + .fnend +#endif +.endm + .macro indirect_call reg #ifdef HAVE_CALL_INDIRECT blx \reg @@ -68,14 +83,170 @@ #endif .endm - .align 0 +/* For armv4t and newer, toolchains will transparently convert + 'bx lr' to 'mov pc, lr' if needed. GCC has deprecated support + for anything older than armv4t, but this should handle that + corner case in case anyone needs it anyway */ +.macro FN_RETURN +#if __ARM_ARCH <= 4 && __ARM_ARCH_ISA_THUMB == 0 + mov pc, lr +#else + bx lr +#endif +.endm + + + +/****************************************************************************** +* User mode only: This routine makes default target specific Stack +* +-----+ <- SL_sys, Pointer initialization for different processor modes: +* | | SL_usr FIQ, Abort, IRQ, Undefined, Supervisor, System (User) +* | SYS | and setups a default Stack Limit in-case the code has +* | USR | -=0x10000 been compiled with "-mapcs-stack-check" for FIQ and +* | | System (User) modes. +* | | +* +-----+ <- initial SP, +* becomes SP_sys Hard-wiring SL value is not ideal, since there is +* and SL_usr currently no support for checking that the heap and +* stack have not collided, or that this default 64k is +* All modes: is enough for the program being executed. However, +* +-----+ <- SL_sys, it ensures that this simple crt0 world will not +* | | SL_usr immediately cause an overflow event. +* | SYS | +* | USR | -=0x10000 We go through all execution modes and set up SP +* | | for each of them. +* +-----+ <- SP_sys, +* | | SP_usr Note: +* | SVC | -= 0x8000 Mode switch via CPSR is not allowed once in +* | | non-privileged mode, so we take care not to enter +* +-----+ <- SP_svc "User" to set up its sp, and also skip most +* | | operations if already in that mode. +* | IRQ | -= 0x2000 +* | | Input parameters: +* ^ +-----+ <- SP_und - sp - Initialized SP +* s | | - r2 - May contain SL value from semihosting +* t | UND | -= 0x1000 SYS_HEAPINFO call +* a | | Scratch registers: +* c +-----+ <- SP_und - r1 - new value of CPSR +* k | | - r2 - intermediate value (in standalone mode) +* | ABT | -= 0x1000 - r3 - new SP value +* g | | - r4 - save/restore CPSR on entry/exit +* r +-----+ <- SP_abt, +* o | | SL_fiq Declared as "weak" so that user can write and use +* w | FIQ | -= 0x1000 his own implementation if current doesn't fit. +* t | | +* h +-----+ <- initial SP, +* becomes SP_fiq +* +******************************************************************************/ + .align 0 + FUNC_START _stack_init + .weak FUNCTION (_stack_init) + FN_EH_START + + /* M profile doesn't have CPSR register. */ +#if (__ARM_ARCH_PROFILE != 'M') + /* Following code is compatible for both ARM and Thumb ISA */ + mrs r4, CPSR + /* Test mode bits - in User of all are 0 */ + tst r4, #(CPSR_M_MASK) + /* "eq" means r4 AND #0x0F is 0 */ + beq .Lskip_cpu_modes + + mov r3, sp /* save input SP value */ + + /* FIQ mode, interrupts disabled */ + mov r1, #(CPSR_M_FIQ|CPSR_M_32BIT|CPSR_I_MASK|CPSR_F_MASK) + msr CPSR_c, r1 + mov sp, r3 + sub sl, sp, #0x1000 /* FIQ mode has its own SL */ + + /* Abort mode, interrupts disabled */ + mov r3, sl + mov r1, #(CPSR_M_ABT|CPSR_M_32BIT|CPSR_I_MASK|CPSR_F_MASK) + msr CPSR_c, r1 + mov sp, r3 + sub r3, r3, #0x1000 + + /* Undefined mode, interrupts disabled */ + mov r1, #(CPSR_M_UND|CPSR_M_32BIT|CPSR_I_MASK|CPSR_F_MASK) + msr CPSR_c, r1 + mov sp, r3 + sub r3, r3, #0x1000 + + /* IRQ mode, interrupts disabled */ + mov r1, #(CPSR_M_IRQ|CPSR_M_32BIT|CPSR_I_MASK|CPSR_F_MASK) + msr CPSR_c, r1 + mov sp, r3 + sub r3, r3, #0x2000 + + /* Supervisory mode, interrupts disabled */ + mov r1, #(CPSR_M_SVR|CPSR_M_32BIT|CPSR_I_MASK|CPSR_F_MASK) + msr CPSR_c, r1 + mov sp, r3 + + sub r3, r3, #0x8000 /* Min size 32k */ + bic r3, r3, #0x00FF /* Align with current 64k block */ + bic r3, r3, #0xFF00 + +# if __ARM_ARCH > 3 + /* System (shares regs with User) mode, interrupts disabled */ + mov r1, #(CPSR_M_SYS|CPSR_M_32BIT|CPSR_I_MASK|CPSR_F_MASK) + msr CPSR_c, r1 + mov sp, r3 +# else + /* Keep this for ARMv3, but GCC actually dropped it. */ + /* Move value into user mode sp without changing modes, */ + /* via '^' form of ldm */ + str r3, [r3, #-4] + ldmdb r3, {sp}^ +# endif + + /* Back to original mode, presumably SVC, with diabled FIQ/IRQ */ + orr r4, r4, #(CPSR_I_MASK|CPSR_F_MASK) + msr CPSR_c, r4 + +.Lskip_cpu_modes: +#endif + + /* Set SL register */ +#if defined (ARM_RDI_MONITOR) /* semihosting */ + cmp r2, #0 + beq .Lsl_forced_zero + /* allow slop for stack overflow handling and small frames */ +# ifdef THUMB1_ONLY + adds r2, #128 + adds r2, #128 + mov sl, r2 +# else + add sl, r2, #256 +# endif +.Lsl_forced_zero: + +#else /* standalone */ + /* r3 contains SP for System/User mode. Set SL = SP - 0x10000 */ + #ifdef THUMB1_ONLY + movs r2, #64 + lsls r2, r2, #10 + subs r2, r3, r2 + mov sl, r2 + #else + /* Still assumes 256bytes below sl */ + sub sl, r3, #64 << 10 + #endif +#endif + + FN_RETURN + FN_EH_END + +/******************************************************************************* +* Main library startup code. +*******************************************************************************/ + .align 0 FUNC_START _mainCRTStartup FUNC_START _start -#if defined(__ELF__) && !defined(__USING_SJLJ_EXCEPTIONS__) - /* Annotation for EABI unwinding tables. */ - .fnstart -#endif + FN_EH_START /* __ARM_ARCH_PROFILE is defined from GCC 4.8 onwards, however __ARM_ARCH_7A has been defined since 4.2 onwards, which is when v7-a support was added @@ -148,34 +319,27 @@ to skip setting sp/sl to 0 here. - Considering M-profile processors, We might want to initialize sp by the first entry of vector table and return 0 to SYS_HEAPINFO - semihosting call, which will be skipped here. */ + semihosting call, which will be skipped here. + - Considering R-profile processors there is no automatic SP init by hardware + so we need to initialize it by default value. */ + ldr r3, .Lstack cmp r1, #0 beq .LC26 - mov sp, r1 + mov r3, r1 .LC26: - cmp r2, #0 - beq .LC27 - /* allow slop for stack overflow handling and small frames */ -#ifdef THUMB1_ONLY - adds r2, #128 - adds r2, #128 - mov sl, r2 -#else - add sl, r2, #256 -#endif -.LC27: -#else + mov sp, r3 + + /* r2 (SL value) will be used in _stack_init */ + bl FUNCTION (_stack_init) + + +#else /* standalone */ /* Set up the stack pointer to a fixed value */ /* Changes by toralf: - Allow linker script to provide stack via __stack symbol - see defintion of .Lstack - Provide "hooks" that may be used by the application to add - custom init code - see .Lhwinit and .Lswinit - - Go through all execution modes and set up stack for each of them. - Loosely based on init.s from ARM/Motorola example code. - Note: Mode switch via CPSR is not allowed once in non-privileged - mode, so we take care not to enter "User" to set up its sp, - and also skip most operations if already in that mode. */ + custom init code - see .Lhwinit and .Lswinit */ ldr r3, .Lstack cmp r3, #0 @@ -194,57 +358,11 @@ have somehow missed it below (in which case it gets the same value as FIQ - not ideal, but better than nothing.) */ mov sp, r3 -#ifdef PREFER_THUMB - /* XXX Fill in stack assignments for interrupt modes. */ -#else - mrs r2, CPSR - tst r2, #0x0F /* Test mode bits - in User of all are 0 */ - beq .LC23 /* "eq" means r2 AND #0x0F is 0 */ - msr CPSR_c, #0xD1 /* FIRQ mode, interrupts disabled */ - mov sp, r3 - sub sl, sp, #0x1000 /* This mode also has its own sl (see below) */ - - mov r3, sl - msr CPSR_c, #0xD7 /* Abort mode, interrupts disabled */ - mov sp, r3 - sub r3, r3, #0x1000 - - msr CPSR_c, #0xDB /* Undefined mode, interrupts disabled */ - mov sp, r3 - sub r3, r3, #0x1000 - msr CPSR_c, #0xD2 /* IRQ mode, interrupts disabled */ - mov sp, r3 - sub r3, r3, #0x2000 - - msr CPSR_c, #0xD3 /* Supervisory mode, interrupts disabled */ + /* we don't care of r2 value in standalone */ + bl FUNCTION (_stack_init) - mov sp, r3 - sub r3, r3, #0x8000 /* Min size 32k */ - bic r3, r3, #0x00FF /* Align with current 64k block */ - bic r3, r3, #0xFF00 - str r3, [r3, #-4] /* Move value into user mode sp without */ - ldmdb r3, {sp}^ /* changing modes, via '^' form of ldm */ - orr r2, r2, #0xC0 /* Back to original mode, presumably SVC, */ - msr CPSR_c, r2 /* with FIQ/IRQ disable bits forced to 1 */ -#endif -.LC23: - /* Setup a default stack-limit in-case the code has been - compiled with "-mapcs-stack-check". Hard-wiring this value - is not ideal, since there is currently no support for - checking that the heap and stack have not collided, or that - this default 64k is enough for the program being executed. - However, it ensures that this simple crt0 world will not - immediately cause an overflow event: */ -#ifdef THUMB1_ONLY - movs r2, #64 - lsls r2, r2, #10 - subs r2, r3, r2 - mov sl, r2 -#else - sub sl, r3, #64 << 10 /* Still assumes 256bytes below sl */ -#endif #endif #endif /* Zero the memory in the .bss section. */ @@ -445,6 +563,8 @@ change_back: swi SWI_Exit #endif + FN_EH_END + /* For Thumb, constants must be after the code since only positive offsets are supported for PC relative addresses. */ @@ -464,8 +584,6 @@ change_back: #else .word 0x80000 /* Top of RAM on the PIE board. */ #endif -.Lstack: - .word __stack .Lhwinit: .word FUNCTION (hardware_init_hook) .Lswinit: @@ -478,17 +596,16 @@ change_back: and only if, a normal version of the same symbol isn't provided e.g. by a linker script or another object file.) */ - .weak __stack .weak FUNCTION (hardware_init_hook) .weak FUNCTION (software_init_hook) #endif #endif -#if defined(__ELF__) && !defined(__USING_SJLJ_EXCEPTIONS__) - /* Protect against unhandled exceptions. */ - .cantunwind - .fnend -#endif + +.Lstack: + .word __stack + .weak __stack + .LC1: .word __bss_start__ .LC2: -- 2.20.1.windows.1 On Wed, Apr 10, 2019 at 5:55 PM Richard Earnshaw (lists) wrote: > > On 10/04/2019 15:23, Corinna Vinschen wrote: > > Hi Richard, > > > > On Apr 10 13:26, Richard Earnshaw (lists) wrote: > >> On 19/03/2019 13:54, Alexander Fedotov wrote: > >>> Hi all > >>> > >>> In continue of threads > >>> https://sourceware.org/ml/newlib/2019/msg00117.html and > >>> https://sourceware.org/ml/newlib/2019/msg00102.html > >>> > >>> I have reworked stack pointer initialization and stack limit as well. > >>> General idea behind of all these changes is to provide hook for > >>> flexibility (including semihosting when debug tap is used) and improve > >>> readability of existing crt0 code. > >>> And now code makes SP initialization in Thumb mode :) > >>> > >>> Any comments are very appreciated. > >>> > >>> Alex > >>> > >> > >> Apologies once again for the delayed review. > >> > >> A small plea: when posting patches, please can you ensure that your > >> attachments are text/patch, or something similar, so that replies will > >> include the contents (and I can view them directly in the email rather > >> than having to mess about extracting the attachment). > > > > Ideally the patches are inline patches sent with `git send-email' or > > equivalent. I added some wording in terms of how to provide patches to > > https://sourceware.org/newlib/ lately. Do you have a suggestion for > > improving this? > > > > > > Thanks, > > Corinna > > > > Not really; When I use git format-patch with --attach I get > text/x-patch as the MIME type, which is fine. Problems only start if > users have poorly configured mail client that attach as octet-stream or > other binary types. Then mail clients can't display the attachment > inline as they don't know the format is really displayable. > > R.