From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:54567) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1gWmsV-0000Sq-MI for qemu-devel@nongnu.org; Tue, 11 Dec 2018 13:35:33 -0500 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1gWmsR-0007Aa-F3 for qemu-devel@nongnu.org; Tue, 11 Dec 2018 13:35:31 -0500 Received: from userp2130.oracle.com ([156.151.31.86]:60954) by eggs.gnu.org with esmtps (TLS1.0:RSA_AES_256_CBC_SHA1:32) (Exim 4.71) (envelope-from ) id 1gWmsO-0006yA-1L for qemu-devel@nongnu.org; Tue, 11 Dec 2018 13:35:24 -0500 References: <1544049446-6359-1-git-send-email-liam.merwick@oracle.com> <1544049446-6359-4-git-send-email-liam.merwick@oracle.com> From: Maran Wilson Message-ID: <33fb9ea4-d6c5-ed23-2fb5-6f818e250043@oracle.com> Date: Tue, 11 Dec 2018 10:35:01 -0800 MIME-Version: 1.0 In-Reply-To: Content-Type: text/plain; charset=utf-8; format=flowed Content-Transfer-Encoding: 7bit Content-Language: en-US Subject: Re: [Qemu-devel] [RFC 3/3] pvh: Boot uncompressed kernel using direct boot ABI List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Stefano Garzarella , liam.merwick@oracle.com Cc: qemu-devel@nongnu.org, Paolo Bonzini , Eduardo Habkost , rth@twiddle.net, xen-devel@lists.xenproject.org, mst@redhat.com, Stefan Hajnoczi On 12/11/2018 9:11 AM, Stefano Garzarella wrote: > Hi Liam, > in order to support PVH also with SeaBIOS, I'm going to work on a new > option rom (like linuxboot/multiboot) that can be used in this case. That is awesome. Yes, please keep us posted when you have something working. Just FYI, before switching over to using Qemu+qboot, we had been using a Qemu only solution (but not using an option rom) internally that worked very well using no FW at all. We had Qemu simply parse the ELF file and jump to the PVH entry point if one is found. The only gotcha was that we had to include a pair of patches that were originally written by folks at Intel as part of the clear containers work. Specifically, in order to be able to skip firmware entirely, we had to do 2 additional things: (1) ACPI tables generated by Qemu are usually patched up by FW. Since we were running no FW, we needed to do that patching up of the ACPI tables in Qemu when it was detected that we were going to enter the OS via the PVH entry point. (2) We also needed to add a patch to Qemu to enable a few PM registers -- something typically done by FW. But if SeaBIOS is involved in the solution you are working on, I guess you won't really need those extra patches. Just figured I'd mention it so you have the full picture. Thanks, -Maran > I'll keep you updated on it! > > Cheers, > Stefano > On Wed, Dec 5, 2018 at 11:38 PM Liam Merwick wrote: >> These changes (along with corresponding qboot and Linux kernel changes) >> enable a guest to be booted using the x86/HVM direct boot ABI. >> >> This commit adds a load_elfboot() routine to pass the size and >> location of the kernel entry point to qboot (which will fill in >> the start_info struct information needed to to boot the guest). >> Having loaded the ELF binary, load_linux() will run qboot >> which continues the boot. >> >> The address for the kernel entry point has already been read >> from an ELF Note in the uncompressed kernel binary earlier >> in pc_memory_init(). >> >> Signed-off-by: George Kennedy >> Signed-off-by: Liam Merwick >> --- >> hw/i386/pc.c | 72 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ >> 1 file changed, 72 insertions(+) >> >> diff --git a/hw/i386/pc.c b/hw/i386/pc.c >> index 056aa46d99b9..d3012cbd8597 100644 >> --- a/hw/i386/pc.c >> +++ b/hw/i386/pc.c >> @@ -54,6 +54,7 @@ >> #include "sysemu/qtest.h" >> #include "kvm_i386.h" >> #include "hw/xen/xen.h" >> +#include "hw/xen/start_info.h" >> #include "ui/qemu-spice.h" >> #include "exec/memory.h" >> #include "exec/address-spaces.h" >> @@ -1098,6 +1099,50 @@ done: >> return pvh_start_addr != 0; >> } >> >> +static bool load_elfboot(const char *kernel_filename, >> + int kernel_file_size, >> + uint8_t *header, >> + size_t pvh_xen_start_addr, >> + FWCfgState *fw_cfg) >> +{ >> + uint32_t flags = 0; >> + uint32_t mh_load_addr = 0; >> + uint32_t elf_kernel_size = 0; >> + uint64_t elf_entry; >> + uint64_t elf_low, elf_high; >> + int kernel_size; >> + >> + if (ldl_p(header) != 0x464c457f) { >> + return false; /* no elfboot */ >> + } >> + >> + bool elf_is64 = header[EI_CLASS] == ELFCLASS64; >> + flags = elf_is64 ? >> + ((Elf64_Ehdr *)header)->e_flags : ((Elf32_Ehdr *)header)->e_flags; >> + >> + if (flags & 0x00010004) { /* LOAD_ELF_HEADER_HAS_ADDR */ >> + error_report("elfboot unsupported flags = %x", flags); >> + exit(1); >> + } >> + >> + kernel_size = load_elf(kernel_filename, NULL, NULL, &elf_entry, >> + &elf_low, &elf_high, 0, I386_ELF_MACHINE, >> + 0, 0); >> + >> + if (kernel_size < 0) { >> + error_report("Error while loading elf kernel"); >> + exit(1); >> + } >> + mh_load_addr = elf_low; >> + elf_kernel_size = elf_high - elf_low; >> + >> + fw_cfg_add_i32(fw_cfg, FW_CFG_KERNEL_ENTRY, pvh_xen_start_addr); >> + fw_cfg_add_i32(fw_cfg, FW_CFG_KERNEL_ADDR, mh_load_addr); >> + fw_cfg_add_i32(fw_cfg, FW_CFG_KERNEL_SIZE, elf_kernel_size); >> + >> + return true; >> +} >> + >> static void load_linux(PCMachineState *pcms, >> FWCfgState *fw_cfg) >> { >> @@ -1138,6 +1183,33 @@ static void load_linux(PCMachineState *pcms, >> if (ldl_p(header+0x202) == 0x53726448) { >> protocol = lduw_p(header+0x206); >> } else { >> + /* If the kernel address for using the x86/HVM direct boot ABI has >> + * been saved then proceed with booting the uncompressed kernel */ >> + if (pvh_start_addr) { >> + if (load_elfboot(kernel_filename, kernel_size, >> + header, pvh_start_addr, fw_cfg)) { >> + struct hvm_modlist_entry ramdisk_mod = { 0 }; >> + >> + fclose(f); >> + >> + fw_cfg_add_i32(fw_cfg, FW_CFG_CMDLINE_SIZE, >> + strlen(kernel_cmdline) + 1); >> + fw_cfg_add_string(fw_cfg, FW_CFG_CMDLINE_DATA, kernel_cmdline); >> + >> + assert(machine->device_memory != NULL); >> + ramdisk_mod.paddr = machine->device_memory->base; >> + ramdisk_mod.size = >> + memory_region_size(&machine->device_memory->mr); >> + >> + fw_cfg_add_bytes(fw_cfg, FW_CFG_KERNEL_DATA, &ramdisk_mod, >> + sizeof(ramdisk_mod)); >> + fw_cfg_add_i32(fw_cfg, FW_CFG_SETUP_SIZE, sizeof(header)); >> + fw_cfg_add_bytes(fw_cfg, FW_CFG_SETUP_DATA, >> + header, sizeof(header)); >> + >> + return; >> + } >> + } >> /* This looks like a multiboot kernel. If it is, let's stop >> treating it like a Linux kernel. */ >> if (load_multiboot(fw_cfg, f, kernel_filename, initrd_filename, >> -- >> 1.8.3.1 >> >