diff options
author | patacongo <patacongo@42af7a65-404d-4744-a932-0658087f49c3> | 2012-02-24 18:24:35 +0000 |
---|---|---|
committer | patacongo <patacongo@42af7a65-404d-4744-a932-0658087f49c3> | 2012-02-24 18:24:35 +0000 |
commit | 1b4c92672cfd43d05dbee2db143af08ec84d90af (patch) | |
tree | 334ac14c141647e11b2ca6f79d362f15dfd727fa | |
parent | 200fa47c3ca70ddbb91057823425c252f3985942 (diff) | |
download | nuttx-1b4c92672cfd43d05dbee2db143af08ec84d90af.tar.gz nuttx-1b4c92672cfd43d05dbee2db143af08ec84d90af.tar.bz2 nuttx-1b4c92672cfd43d05dbee2db143af08ec84d90af.zip |
select() fix to handl POLLHUP; STM32 FPU saving in context switches seems to be functional
git-svn-id: svn://svn.code.sf.net/p/nuttx/code/trunk@4420 42af7a65-404d-4744-a932-0658087f49c3
-rw-r--r-- | nuttx/ChangeLog | 11 | ||||
-rw-r--r-- | nuttx/arch/arm/src/armv7-m/up_fpu.S | 29 | ||||
-rw-r--r-- | nuttx/arch/arm/src/stm32/stm32_qencoder.c | 2 | ||||
-rw-r--r-- | nuttx/arch/arm/src/stm32/stm32_vectors.S | 82 | ||||
-rwxr-xr-x | nuttx/configs/stm3240g-eval/README.txt | 34 | ||||
-rwxr-xr-x | nuttx/configs/stm32f4discovery/README.txt | 36 | ||||
-rw-r--r-- | nuttx/fs/fs_select.c | 35 |
7 files changed, 170 insertions, 59 deletions
diff --git a/nuttx/ChangeLog b/nuttx/ChangeLog index dde876459..fa85503ef 100644 --- a/nuttx/ChangeLog +++ b/nuttx/ChangeLog @@ -2477,6 +2477,9 @@ * arch/arm/src/stm32_sdio.c: STM32 F4 SDIO DMA is now supported * configs/stm3240g-eval/nsh/defconfig: This configuration now supports SDIO with DMA (see configs/stm3240g-eval/README.txt for some issues). + * arch/arm/src/armv7-m/up_vectors.S and arch/arm/src/armv7-m/up_vectors.S: New, + streamlined Cortex-M exception handling (with FPU supported). Contributed byh + Mike Smith * net/accept.c, connect.c,and net_monitor.c: Correct an error in the accept logic. After a new connection is made via accept(), monitoring for losses of TCP connection must be set up (just as with connect()). The new file @@ -2495,3 +2498,11 @@ accepted. * configs/stm3240g-eval, configs/stm32f40discovery, and arch/*/src/Makefile: Add changes to support building with the Atollic "Lite" toolchain. + * fs/fs_select.c: Correct select(), in the case of loss of network + connection (POLLHUP), select() must report a read-ready event. This + is how the standard select() interface is supposed to work: In the case + of loss-of-connection, select() reports read-ready. The next time you + read from the socket, you detect the end-of-connection event. Change + submitted by Max Nekludov. + * arch/arm/src/armv7-m/up_fpu.S and arch/arm/src/stm32/stm32_vectors.S: Fix + lazy FPU register saving with CONFIG_ARCH_FPU is set in the configuration. diff --git a/nuttx/arch/arm/src/armv7-m/up_fpu.S b/nuttx/arch/arm/src/armv7-m/up_fpu.S index ee3644594..fd6449c2f 100644 --- a/nuttx/arch/arm/src/armv7-m/up_fpu.S +++ b/nuttx/arch/arm/src/armv7-m/up_fpu.S @@ -95,11 +95,15 @@ up_savefpu: /* Some older GNU assemblers don't support all the newer UAL mnemonics. */ #if 1 /* Use UAL mnemonics */ - /* Store all floating point registers */ + /* Store all floating point registers. Registers are stored in numeric order, + * s0, s1, ... in increasing address order. + */ vstmia r1!, {s0-s31} /* Save the full FP context */ - /* Store the floating point control and status register */ + /* Store the floating point control and status register. At the end of the + * vstmia, r1 will point to the FPCSR storage location. + */ vmrs r2, fpscr /* Fetch the FPCSR */ str r2, [r1], #4 /* Save the floating point control and status register */ @@ -180,10 +184,11 @@ up_savefpu: * * Input Parameters: * regs - A pointer to the register save area containing the floating point - * registers + * registers. * * Returned Value: - * None + * This function does not return anything explicitly. However, it is called from + * interrupt level assembly logic that assumes that r0 is preserved. * ************************************************************************************/ @@ -196,16 +201,22 @@ up_restorefpu: /* Some older GNU assemblers don't support all the newer UAL mnemonics. */ #if 1 /* Use UAL mnemonics */ - /* Load all floating point registers */ + /* Load all floating point registers. Registers are loaded in numeric order, + * s0, s1, ... in increasing address order. + */ vldmia r1!, {s0-s31} /* Restore the full FP context */ - /* Load the floating point control and status register */ + /* Load the floating point control and status register. At the end of the + * vstmia, r1 will point to the FPCSR storage location. + */ ldr r2, [r1], #4 /* Fetch the floating point control and status register */ vmsr fpscr, r2 /* Restore the FPCSR */ #else - /* Load all floating point registers */ + /* Load all floating point registers Registers are loaded in numeric order, + * s0, s1, ... in increasing address order. + */ #if 1 /* Use load multiple */ fldmias r1!, {s0-s31} /* Restore the full FP context */ @@ -260,7 +271,9 @@ up_restorefpu: vmov d15, r2, r3 /* Save as d15 */ #endif - /* Load the floating point control and status register */ + /* Load the floating point control and status register. r1 points t + * the address of the FPCSR register. + */ ldr r2, [r1], #4 /* Fetch the floating point control and status register */ fmxr fpscr, r2 /* Restore the FPCSR */ diff --git a/nuttx/arch/arm/src/stm32/stm32_qencoder.c b/nuttx/arch/arm/src/stm32/stm32_qencoder.c index 67ba1eabb..82c9e361b 100644 --- a/nuttx/arch/arm/src/stm32/stm32_qencoder.c +++ b/nuttx/arch/arm/src/stm32/stm32_qencoder.c @@ -499,7 +499,7 @@ static FAR struct stm32_lowerhalf_s *stm32_tim2lower(int tim) } /************************************************************************************ - * Name: stm32_setup + * Name: stm32_interrupt * * Description: * Common timer interrupt handling diff --git a/nuttx/arch/arm/src/stm32/stm32_vectors.S b/nuttx/arch/arm/src/stm32/stm32_vectors.S index 91f530f05..1629dfc9f 100644 --- a/nuttx/arch/arm/src/stm32/stm32_vectors.S +++ b/nuttx/arch/arm/src/stm32/stm32_vectors.S @@ -211,16 +211,37 @@ stm32_common: */ adds r2, r14, #3 /* If R14=0xfffffffd, then r2 == 0 */ - ite ne /* Next two instructions are condition */ + ite ne /* Next two instructions are conditional */ mrsne r1, msp /* R1=The main stack pointer */ mrseq r1, psp /* R1=The process stack pointer */ #else mrs r1, msp /* R1=The main stack pointer */ #endif + /* r1 holds the value of the stack pointer AFTER the excption handling logic + * pushed the various registers onto the stack. Get r2 = the value of the + * stack pointer BEFORE the interrupt modified it. + */ + mov r2, r1 /* R2=Copy of the main/process stack pointer */ add r2, #HW_XCPT_SIZE /* R2=MSP/PSP before the interrupt was taken */ mrs r3, primask /* R3=Current PRIMASK setting */ + +#ifdef CONFIG_ARCH_FPU + /* Skip over the block of memory reserved for floating pointer register save. + * Lazy FPU register saving is used. FPU registers will be saved in this + * block only if a context switch occurs (this means, of course, that the FPU + * cannot be used in interrupt processing). + */ + + sub r1, #(4*SW_FPU_REGS) +#endif + + /* Save the the remaining registers on the stack after the registers pushed + * by the exception handling logic. r2=SP and r3=primask, r4-r11,r14=register + * values. + */ + #ifdef CONFIG_NUTTX_KERNEL stmdb r1!, {r2-r11,r14} /* Save the remaining registers plus the SP value */ #else @@ -257,36 +278,47 @@ stm32_common: cmp r0, r1 /* Context switch? */ beq 1f /* Branch if no context switch */ - /* We are returning with a pending context switch. This case is different - * because in this case, the register save structure does not lie on the - * stack but, rather, are within a TCB structure. We'll have to copy some - * values to the stack. + /* We are returning with a pending context switch. + * + * If the FPU is enabled, then we will need to restore FPU registers. + * This is not done in normal interrupt save/restore because the cost + * is prohibitive. This is only done when switching contexts. A + * consequence of this is that floating point operations may not be + * performed in interrupt handling logic. + * + * Here: + * r0 = Address of the register save area + + * NOTE: It is a requirement that up_restorefpu() preserve the value of + * r0! + */ + +#ifdef CONFIG_ARCH_FPU + bl up_restorefpu /* Restore the FPU registers */ +#endif + + /* Returning with a pending context switch is different from the normal + * return because in this case, the register save structure does not lie + * on the stack but, rather, are within a TCB structure. We'll have to + * copy somevalues to the new stack. */ add r1, r0, #SW_XCPT_SIZE /* R1=Address of HW save area in reg array */ ldmia r1, {r4-r11} /* Fetch eight registers in HW save area */ - ldr r1, [r0, #(4*REG_SP)] /* R1=Value of SP before interrupt */ + ldr r1, [r0, #(4*REG_SP)] /* R1=Value of SP before interrupt */ stmdb r1!, {r4-r11} /* Store eight registers in HW save area */ #ifdef CONFIG_NUTTX_KERNEL ldmia r0, {r2-r11,r14} /* Recover R4-R11, r14 + 2 temp values */ #else ldmia r0, {r2-r11} /* Recover R4-R11 + 2 temp values */ #endif - - /* We may also need to restore FPU registers. This is not done in - * normal interrupt save/restore because the cost is prohibitive. This - * is only done when switching contexts. A consequence of this is that - * floating point operations may not be performed in interrupt handling - * logic. - */ - -#ifdef CONFIG_ARCH_FPU - bl up_restorefpu /* Restore the FPU registers */ -#endif b 2f /* Re-join common logic */ /* We are returning with no context switch. We simply need to "unwind" * the same stack frame that we created + * + * Here: + * r1 = Address of the return stack (same as r0) */ 1: #ifdef CONFIG_NUTTX_KERNEL @@ -294,6 +326,22 @@ stm32_common: #else ldmia r1!, {r2-r11} /* Recover R4-R11 + 2 temp values */ #endif +#ifdef CONFIG_ARCH_FPU + /* Skip over the block of memory reserved for floating pointer register + * save. Then R1 is the address of the HW save area + */ + + add r1, #(4*SW_FPU_REGS) +#endif + + /* Set up to return from the exception + * + * Here: + * r1 = Address on the target thread's stack position at the start of + * the registers saved by hardware + * r3 = primask + * r4-r11 = restored register values + */ 2: #ifdef CONFIG_NUTTX_KERNEL /* The EXC_RETURN value will be 0xfffffff9 (privileged thread) or 0xfffffff1 diff --git a/nuttx/configs/stm3240g-eval/README.txt b/nuttx/configs/stm3240g-eval/README.txt index 2c7c09bdf..34c6a09f3 100755 --- a/nuttx/configs/stm3240g-eval/README.txt +++ b/nuttx/configs/stm3240g-eval/README.txt @@ -32,6 +32,8 @@ Development Environment GNU Toolchain Options ===================== + Toolchain Configurations + ------------------------ The NuttX make system has been modified to support the following different toolchain options. @@ -91,12 +93,16 @@ GNU Toolchain Options If you have problems with the dependency build (for example, if you are not building on C:), then you may need to modify tools/mkdeps.sh - NOTE 1: The CodeSourcery toolchain (2009q1) does not work with default optimization + The CodeSourcery Toolchain (2009q1) + ----------------------------------- + The CodeSourcery toolchain (2009q1) does not work with default optimization level of -Os (See Make.defs). It will work with -O0, -O1, or -O2, but not with -Os. - NOTE 2: The free, "Lite" version of the Atollic toolchain does not support C++ - nor does it support ar, nm, objdump, or objdcopy. If you use the Atollic "Lite" + The Atollic "Lite" Toolchain + ---------------------------- + The free, "Lite" version of the Atollic toolchain does not support C++ nor + does it support ar, nm, objdump, or objdcopy. If you use the Atollic "Lite" toolchain, you will have to set: CONFIG_HAVE_CXX=n @@ -122,22 +128,24 @@ GNU Toolchain Options errors build some host binaries in tools/ when you first make. Here is my workaround kludge. - 1. Edit the setenv.sh to put the Atollic toolchain at the beginning of the PATH - 2. Source the setenv.sh file: . ./setenv.sh. A side effect of this is that it - will set an environment variable called PATH_ORIG. - 3. Then go back to the original patch: export PATH=$PATH_ORIG - 4. Then make. The make will build all of the host executable but will fail - when it gets to the first ARM binary. - 5. Then source setenv.sh again: . ./setenv.sh. That will correct the PATH - again. When you do make again, the host executables are already made and - now the correct PATH is in place for the ARM build. + 1. Edit the setenv.sh to put the Atollic toolchain at the beginning of the PATH + 2. Source the setenv.sh file: . ./setenv.sh. A side effect of this is that it + will set an environment variable called PATH_ORIG. + 3. Then go back to the original patch: export PATH=$PATH_ORIG + 4. Then make. The make will build all of the host executable but will fail + when it gets to the first ARM binary. + 5. Then source setenv.sh again: . ./setenv.sh. That will correct the PATH + again. When you do make again, the host executables are already made and + now the correct PATH is in place for the ARM build. Also, the Atollic toolchain is the only toolchain that has built-in support for the FPU in these configurations. If you plan to use the Cortex-M4 FPU, you will need to use the Atollic toolchain for now. See the FPU section below for more information. - NOTE 3: The devkitARM toolchain includes a version of MSYS make. Make sure that + devkitARM + --------- + The devkitARM toolchain includes a version of MSYS make. Make sure that the the paths to Cygwin's /bin and /usr/bin directories appear BEFORE the devkitARM path or will get the wrong version of make. diff --git a/nuttx/configs/stm32f4discovery/README.txt b/nuttx/configs/stm32f4discovery/README.txt index 23e45c742..06b835b07 100755 --- a/nuttx/configs/stm32f4discovery/README.txt +++ b/nuttx/configs/stm32f4discovery/README.txt @@ -31,6 +31,8 @@ Development Environment GNU Toolchain Options ===================== + Toolchain Configurations + ------------------------ The NuttX make system has been modified to support the following different toolchain options. @@ -90,12 +92,16 @@ GNU Toolchain Options If you have problems with the dependency build (for example, if you are not building on C:), then you may need to modify tools/mkdeps.sh - NOTE 1: The CodeSourcery toolchain (2009q1) does not work with default optimization + The CodeSourcery Toolchain (2009q1) + ----------------------------------- + The CodeSourcery toolchain (2009q1) does not work with default optimization level of -Os (See Make.defs). It will work with -O0, -O1, or -O2, but not with -Os. - NOTE 2: The free, "Lite" version of the Atollic toolchain does not support C++ - nor does it support ar, nm, objdump, or objdcopy. If you use the Atollic "Lite" + The Atollic "Lite" Toolchain + ---------------------------- + The free, "Lite" version of the Atollic toolchain does not support C++ nor + does it support ar, nm, objdump, or objdcopy. If you use the Atollic "Lite" toolchain, you will have to set: CONFIG_HAVE_CXX=n @@ -118,25 +124,27 @@ GNU Toolchain Options and g++.exe in the same bin/ file as their ARM binaries. If the Atollic bin/ path appears in your PATH variable before /usr/bin, then you will get the wrong gcc when you try to build host executables. This will cause to strange, uninterpretable - errors build some host binaries in tools/ when you first make. Here is my + errors build some host binaries in tools/ when you first make. Here is my workaround kludge. - 1. Edit the setenv.sh to put the Atollic toolchain at the beginning of the PATH - 2. Source the setenv.sh file: . ./setenv.sh. A side effect of this is that it - will set an environment variable called PATH_ORIG. - 3. Then go back to the original patch: export PATH=$PATH_ORIG - 4. Then make. The make will build all of the host executable but will fail - when it gets to the first ARM binary. - 5. Then source setenv.sh again: . ./setenv.sh. That will correct the PATH - again. When you do make again, the host executables are already made and - now the correct PATH is in place for the ARM build. + 1. Edit the setenv.sh to put the Atollic toolchain at the beginning of the PATH + 2. Source the setenv.sh file: . ./setenv.sh. A side effect of this is that it + will set an environment variable called PATH_ORIG. + 3. Then go back to the original patch: export PATH=$PATH_ORIG + 4. Then make. The make will build all of the host executable but will fail + when it gets to the first ARM binary. + 5. Then source setenv.sh again: . ./setenv.sh. That will correct the PATH + again. When you do make again, the host executables are already made and + now the correct PATH is in place for the ARM build. Also, the Atollic toolchain is the only toolchain that has built-in support for the FPU in these configurations. If you plan to use the Cortex-M4 FPU, you will need to use the Atollic toolchain for now. See the FPU section below for more information. - NOTE 3: The devkitARM toolchain includes a version of MSYS make. Make sure that + devkitARM + --------- + The devkitARM toolchain includes a version of MSYS make. Make sure that the the paths to Cygwin's /bin and /usr/bin directories appear BEFORE the devkitARM path or will get the wrong version of make. diff --git a/nuttx/fs/fs_select.c b/nuttx/fs/fs_select.c index 213530e44..14ccfe287 100644 --- a/nuttx/fs/fs_select.c +++ b/nuttx/fs/fs_select.c @@ -1,8 +1,8 @@ /**************************************************************************** * fs/fs_select.c * - * Copyright (C) 2008-2009 Gregory Nutt. All rights reserved. - * Author: Gregory Nutt <spudmonkey@racsa.co.cr> + * Copyright (C) 2008-2009, 2012 Gregory Nutt. All rights reserved. + * Author: Gregory Nutt <gnutt@nuttx.org> * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions @@ -124,24 +124,37 @@ int select(int nfds, FAR fd_set *readfds, FAR fd_set *writefds, { int incr = 0; + /* The readfs set holds the set of FDs that the caller can be assured + * of reading from without blocking. Note that POLLHUP is included as + * a read-able condition. POLLHUP will be reported at the end-of-file + * or when a connection is lost. In either case, the read() can then + * be performed without blocking. + */ + if (readfds && FD_ISSET(fd, readfds)) { pollset[npfds].fd = fd; pollset[npfds].events |= POLLIN; - incr = 1; + incr = 1; } + /* The writefds set holds the set of FDs that the caller can be assured + * of writing to without blocking. + */ + if (writefds && FD_ISSET(fd, writefds)) { pollset[npfds].fd = fd; pollset[npfds].events |= POLLOUT; - incr = 1; + incr = 1; } + /* The exceptfds set holds the set of FDs that are watched for exceptions */ + if (exceptfds && FD_ISSET(fd, exceptfds)) { pollset[npfds].fd = fd; - incr = 1; + incr = 1; } npfds += incr; @@ -179,15 +192,23 @@ int select(int nfds, FAR fd_set *readfds, FAR fd_set *writefds, ret = 0; for (ndx = 0; ndx < npfds; ndx++) { + /* Check for read conditions. Note that POLLHUP is included as a + * read condition. POLLHUP will be reported when no more data will + * be available (such as when a connection is lost). In either + * case, the read() can then be performed without blocking. + */ + if (readfds) { - if (pollset[ndx].revents & POLLIN) + if (pollset[ndx].revents & (POLLIN|POLLHUP)) { FD_SET(pollset[ndx].fd, readfds); ret++; } } + /* Check for write conditions */ + if (writefds) { if (pollset[ndx].revents & POLLOUT) @@ -197,6 +218,8 @@ int select(int nfds, FAR fd_set *readfds, FAR fd_set *writefds, } } + /* Check for exceptions */ + if (exceptfds) { if (pollset[ndx].revents & POLLERR) |