~mirabilos/klibc/master : changes

~mirabilos/klibc/master » Changes from revision 2226

From Revision 2226 to 2207

Rev	Summary	Authors	Tags	Date
2226	[klibc] 2.0.13 released, next version is 2.0.14 [klibc] 2.0.13 released, next version is 2.0.14	Ben Hutchings		9 months ago
2225	[klibc] Install command links as appropriate [klibc] Install command links as appropriate We build gzip to support being called as "gunzip" or "zcat", and similarly we build halt to support being called as "poweroff" or "reboot". In the build directory, we create the aliases as links to the same binary. When installing, however, we pass all the names for the same binary to the "install" command, creating multiple copies of it in the install directory. To fix this: - Define an install_links command that creates links as specified in an install-links-y variable. - Invoke install_links after installing binaries. - In usr/gzip/Kbuild and usr/utils/Kbuild, specify the links in install-links-y instead of install-y. Reported-by: Benjamin Drung <benjamin.drung@canonical.com> References: https://bugs.launchpad.net/ubuntu/+source/klibc/+bug/2028571 Signed-off-by: Ben Hutchings <ben@decadent.org.uk>	Ben Hutchings	klibc-2.0.13	9 months ago
2224	[klibc] kinit: Support specifying root with PARTLABEL [klibc] kinit: Support specifying root with PARTLABEL Allow specifying the root device to be mounted as root=PARTLABEL=label. The label is the GPT partition label of the intended root partition, as presented by the kernel with the uevent entry PARTNAME. Signed-off-by: Rob Vandermeulen <rvandermeulen@google.com> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>	Rob Vandermeulen		9 months ago
2223	[klibc] riscv64: Move shared library below executables [klibc] riscv64: Move shared library below executables The default base address for riscv64 executables is 64 kiB, and I originally selected a base address for the shared library of 2 MiB. This limits the size of executables using it to be < 2 MiB, and it also interacts with a bug in QEMU's user-mode emulation, causing executables to immediately crash. It doesn't help code generation, because single-instruction jumps have a range of only ±1 MiB and two-instruction jumps have a range of ±2 GiB. Move the shared library to 64 kiB and executables to 576 kiB, to allow for generation of single-instruction jumps. (klibc.so is currently much smaller than the 512 kiB this allows for.) References: https://bugs.debian.org/1040981 Signed-off-by: Ben Hutchings <ben@decadent.org.uk>	Ben Hutchings		10 months ago
2222	[klibc] arm: Move shared library below executables [klibc] arm: Move shared library below executables The default base address for arm executables is 64 kiB, and the base address for the shared library is 3.5 MiB when generating Thumb instructions. This limits the size of executables using it to be < 3.5 MiB, and it also interacts with a bug in QEMU's user-mode emulation, causing executables to immediately crash. Moving the shared library to the end of the 16 MiB range didn't help. Instead, move the shared library to 2 MiB and executables to 4 MiB. This works for both Thumb and non-Thumb configurations, so make it unconditional. References: https://bugs.debian.org/1040981 Signed-off-by: Ben Hutchings <ben@decadent.org.uk>	Ben Hutchings		10 months ago
2221	[klibc] mips64: Compatibility with R6 compact branches [klibc] mips64: Compatibility with R6 compact branches MIPS R6 introduced compact branch instructions, including BALC as a replacement for JAL. Both BALC and JAL have a 26-bit offset field that's shifted 2 bits left. However, in BALC this is treated as a PC-relative signed offset (±128 MiB) rather than a segment-relative unsigned offset (0-256 MiB). The base address of the klibc shared library therefore needs to change depending on whether the executable calls it with JAL or BALC. gcc (but not Clang, currently) generates BALC instructions for external function calls on R6 targets, which breaks shared library builds of klibc for these targets. Add a config symbol to control whether compact branches are used and to change the shared library base address accordingly. Signed-off-by: Ben Hutchings <ben@decadent.org.uk>	Ben Hutchings		10 months ago
2220	[klibc] ipconfig: Retry failed send after 1 second [klibc] ipconfig: Retry failed send after 1 second Currently we check whether a device has link down before attempting to send a packet, but the send can still fail with ENETDOWN - perhaps due to racing with the kernel's linkwatch work. After such an error we'll wait about 10 seconds before retrying. Since the link check isn't reliable, remove it. Instead, wait only 1 second if we fail to send a packet. Signed-off-by: Ben Hutchings <ben@decadent.org.uk>	Ben Hutchings		11 months ago
2219	[klibc] ipconfig: Handle error events from poll [klibc] ipconfig: Handle error events from poll When poll returns events for a socket other than POLLRDNORM, such as POLLERR, we do not read from the socket but keep polling on it. This results in busy-polling while the error state persists. In case some other event is returned, put the device into DEVST_ERROR state for 1 second. Signed-off-by: Ben Hutchings <ben@decadent.org.uk>	Ben Hutchings		11 months ago
2218	[klibc] ipconfig: Do not poll sockets we don't intend to rea... [klibc] ipconfig: Do not poll sockets we don't intend to read from When a device is in state DEVST_COMPLETE or DEVST_ERROR, we never read packets from its socket, but we still poll it. If the socket is readable or in an error state, this results in busy-polling until a timeout is reached. Only add a device's socket to the fds array when the device is in some other state. Signed-off-by: Ben Hutchings <ben@decadent.org.uk>	Ben Hutchings		11 months ago
2217	[klibc] Add LoongArch64 port [klibc] Add LoongArch64 port Add LoongArch64 support. I've cribbed most of this from the RISC-V and MIPS ports. Signed-off-by: Feiyang Chen <chenfeiyang@loongson.cn> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>	Feiyang Chen		1 year ago
2216	Merge branch 'time64' Merge branch 'time64' This switches all architectures to using 64-bit time_t, struct timespec, and the related new system calls. - Since there is no pselect variant using 64-bit time and the old signal API, this switches the remaining architectures to the RT signal API and makes it mandatory. - Since the only stat variant upporting 64-bit time on 32-bit architectures and mips64 is statx(), all stat functions are changed to be wrappers for that. - Some time system calls that don't have direct replacements available on 32-bit architectures. They are changed to be wrappers for the newer system calls, e.g. nanosleep() becomes a wrapper for clock_nanosleep_time64(). To keep things simple, the system call wrapper changes are applied on all architectures.	Ben Hutchings		1 year ago
2215	[klibc] 2.0.12 released, next version is 2.0.13 [klibc] 2.0.12 released, next version is 2.0.13	Ben Hutchings		1 year ago
2214	[klibc] mips64: Fix struct stat layout [klibc] mips64: Fix struct stat layout On mips64, struct stat has 32-bit time fields instead of following struct timespec. Signed-off-by: Ben Hutchings <ben@decadent.org.uk>	Ben Hutchings		1 year ago
2213	[klibc] ls: Handle relative directory names correctly [klibc] ls: Handle relative directory names correctly ls currently starts to list a directory with chdir(path) and opendir(path). This obviously fails for relative names. That could be fixed by using opendir(".") instead. However, if multiple relative directory names are given, this would still fail after the first such directory. Instead, leave the current directory unchanged and use the fstatat() and readlinkat() functions to get information about files in each directory. Signed-off-by: Ben Hutchings <ben@decadent.org.uk>	Ben Hutchings		1 year ago
2212	[klibc] strstr, memmem: Handle zero-length needle correctly [klibc] strstr, memmem: Handle zero-length needle correctly strstr(haystack, "") and memmem(haystack, n, needle, 0) should return haystack, not NULL. - Handle the !m and (m > n \|\| !n) cases separately at the top. - Delete the !n condition. After handling !m we know m > 0, so checking for !n is redundant with m > n. Signed-off-by: Ben Hutchings <ben@decadent.org.uk>	Ben Hutchings		1 year ago
2211	[klibc] strrchr: Handle c == 0 correctly [klibc] strrchr: Handle c == 0 correctly strrchr(s, 0) should return a pointer to the null terminator, not NULL. Move the null terminator check to make this work. Reported-by: Bugs Reporter <bugs@compiler.ai> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>	Ben Hutchings		1 year ago
2210	[klibc] tests: Add test program for string search functions [klibc] tests: Add test program for string search functions Add some test cases for str{,r}chr(), str{,c}spn(), strpbrk(), and strstr(). One test case for strrchr() and one for strstr() currently fail. Signed-off-by: Ben Hutchings <ben@decadent.org.uk>	Ben Hutchings		1 year ago
2209	[klibc] mips/mips64: Get rid of delay slot hacks [klibc] mips/mips64: Get rid of delay slot hacks There are some MIPS assembly code trying to take advantage of delay slots. However in MIPS Release 6, assembler (LLVM) will try to relax traditional branch to compat branch, and having noreroder code will block assembler from doing so. Moreover, nowadays assemblers are smart enough to fill delay slots properly. Most of those hacks can actually be archived by assemblers. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Signed-off-by: Ben Hutchings <ben@decadent.org.uk>	Jiaxun Yang		1 year ago
2208	[klibc] 2.0.11 released, next version is 2.0.12 [klibc] 2.0.11 released, next version is 2.0.12	Ben Hutchings		1 year ago
2207	[klibc] zalloc: Fix infinite recursion with Clang 14 [klibc] zalloc: Fix infinite recursion with Clang 14 zalloc() calls malloc() followed by memset(), which Clang 14 optimises to a call to calloc(). But our calloc() calls zalloc(), so this results in recursion until the stack is exhausted. Use the option -fno-builtin-malloc to inhibit this optimisation for zalloc.c. Signed-off-by: Ben Hutchings <ben@decadent.org.uk>	Ben Hutchings		1 year ago

Older »