| Message ID | 20220503194843.1379101-69-richard.henderson@linaro.org |
|---|---|
| State | New |
| Headers | show |
| Series | semihosting cleanup | expand |
On Tue, 3 May 2022 at 21:52, Richard Henderson <richard.henderson@linaro.org> wrote: > > Attempt to load the kernel with load_elf. If this fails with > ELF_LOAD_NOT_ELF, continue to treat the kernel as a raw image. > > This will be handy for running semihosting programs. > > Signed-off-by: Richard Henderson <richard.henderson@linaro.org> I know nothing of the specifics of the rx target, but I'm always a bit dubious about adding more behaviour to the -kernel option, which is (a) already a morass of undocumented target specific behaviours (b) nominally supposed to be "load a Linux kernel", not "load any random thing". Can you do what you need with the generic-loader device instead? That has the benefit of both being documented and also doing the same thing on all target architectures. thanks -- PMM
On 5/13/22 08:44, Peter Maydell wrote: > On Tue, 3 May 2022 at 21:52, Richard Henderson > <richard.henderson@linaro.org> wrote: >> >> Attempt to load the kernel with load_elf. If this fails with >> ELF_LOAD_NOT_ELF, continue to treat the kernel as a raw image. >> >> This will be handy for running semihosting programs. >> >> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> > > I know nothing of the specifics of the rx target, but I'm > always a bit dubious about adding more behaviour to the > -kernel option, which is (a) already a morass of undocumented > target specific behaviours (b) nominally supposed to be > "load a Linux kernel", not "load any random thing". > Can you do what you need with the generic-loader device instead? > That has the benefit of both being documented and also doing > the same thing on all target architectures. I'll give that a try. r~
On 5/13/22 08:48, Richard Henderson wrote: > On 5/13/22 08:44, Peter Maydell wrote: >> On Tue, 3 May 2022 at 21:52, Richard Henderson >> <richard.henderson@linaro.org> wrote: >>> >>> Attempt to load the kernel with load_elf. If this fails with >>> ELF_LOAD_NOT_ELF, continue to treat the kernel as a raw image. >>> >>> This will be handy for running semihosting programs. >>> >>> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> >> >> I know nothing of the specifics of the rx target, but I'm >> always a bit dubious about adding more behaviour to the >> -kernel option, which is (a) already a morass of undocumented >> target specific behaviours (b) nominally supposed to be >> "load a Linux kernel", not "load any random thing". >> Can you do what you need with the generic-loader device instead? >> That has the benefit of both being documented and also doing >> the same thing on all target architectures. > > I'll give that a try. It works, editing the board not to reject lack of -bios and lack of -kernel. And running by hand because the syntax of the test harness does not allow the test file to be joined with $(QEMU_OPTS), as in -device loader,cpu-num=0,file=testcase I'll work with Alex to figure out how best to restructure the test harness. In the meantime I certainly don't mind putting the rx part on the back burner if it'll help get the previous 67 patches reviewed... r~
diff --git a/hw/rx/rx-gdbsim.c b/hw/rx/rx-gdbsim.c index be147b4bd9..64b533181d 100644 --- a/hw/rx/rx-gdbsim.c +++ b/hw/rx/rx-gdbsim.c @@ -26,6 +26,7 @@ #include "sysemu/device_tree.h" #include "hw/boards.h" #include "qom/object.h" +#include "elf.h" /* Same address of GDB integrated simulator */ #define SDRAM_BASE EXT_CS_BASE @@ -57,15 +58,32 @@ static void rx_load_image(RXCPU *cpu, const char *filename, uint32_t start, uint32_t size) { static uint32_t extable[32]; - long kernel_size; + ssize_t kernel_size; + uint64_t kernel_entry; int i; + /* Try an ELF image first. */ + + kernel_size = load_elf(filename, NULL, NULL, NULL, &kernel_entry, + NULL, NULL, NULL, false, EM_RX, false, false); + if (kernel_size >= 0) { + cpu_set_pc(CPU(cpu), kernel_entry); + return; + } + if (kernel_size != ELF_LOAD_NOT_ELF) { + error_report("could not load kernel '%s': %s", + filename, load_elf_strerror(kernel_size)); + exit(1); + } + + /* Not ELF: load a raw image, e.g. zImage. */ + kernel_size = load_image_targphys(filename, start, size); if (kernel_size < 0) { - fprintf(stderr, "qemu: could not load kernel '%s'\n", filename); + error_report("could not load kernel '%s'", filename); exit(1); } - cpu->env.pc = start; + cpu_set_pc(CPU(cpu), start); /* setup exception trap trampoline */ /* linux kernel only works little-endian mode */
Attempt to load the kernel with load_elf. If this fails with ELF_LOAD_NOT_ELF, continue to treat the kernel as a raw image. This will be handy for running semihosting programs. Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- hw/rx/rx-gdbsim.c | 24 +++++++++++++++++++++--- 1 file changed, 21 insertions(+), 3 deletions(-)