| Message ID | 20240325204515.250203-6-jason.andryuk@amd.com (mailing list archive) |
|---|---|
| State | Superseded |
| Headers | show |
| Series | x86/pvh: Support relocating dom0 kernel | expand |
On 25.03.2024 21:45, Jason Andryuk wrote: > +/* Find an e820 RAM region that fits the kernel at a suitable alignment. */ > +static paddr_t __init find_kernel_memory( > + const struct domain *d, struct elf_binary *elf, > + const struct elf_dom_parms *parms) > +{ > + paddr_t kernel_size = elf->dest_size; > + unsigned int i; > + > + for ( i = 0; i < d->arch.nr_e820; i++ ) > + { > + paddr_t start = d->arch.e820[i].addr; > + paddr_t end = start + d->arch.e820[i].size; > + paddr_t kstart, kend; > + > + if ( d->arch.e820[i].type != E820_RAM || > + d->arch.e820[i].size < kernel_size ) > + continue; > + > + kstart = ROUNDUP(start, parms->phys_align); > + kstart = max_t(paddr_t, kstart, parms->phys_min); > + kend = kstart + kernel_size; > + > + if ( kend - 1 > parms->phys_max ) > + return 0; > + > + if ( kend <= end ) > + return kstart; IOW within a suitable region the lowest suitable part is selected. Often low memory is deemed more precious than higher one, so if this choice is indeed intentional, I'd like to ask for a brief comment towards the reasons. > --- a/xen/common/libelf/libelf-dominfo.c > +++ b/xen/common/libelf/libelf-dominfo.c > @@ -17,6 +17,16 @@ > > #include "libelf-private.h" > > +#if defined(__i386__) || defined(__x86_64__) > +#define ARCH_PHYS_MIN_DEFAULT 0; > +#define ARCH_PHYS_MAX_DEFAULT (GB(4) - 1); > +#define ARCH_PHYS_ALIGN_DEFAULT MB(2); > +#else > +#define ARCH_PHYS_MIN_DEFAULT 0; > +#define ARCH_PHYS_MAX_DEFAULT 0; > +#define ARCH_PHYS_ALIGN_DEFAULT 0; > +#endif None of the semicolons should really be here. > @@ -227,6 +239,27 @@ elf_errorstatus elf_xen_parse_note(struct elf_binary *elf, > case XEN_ELFNOTE_PHYS32_ENTRY: > parms->phys_entry = val; > break; > + > + case XEN_ELFNOTE_PHYS32_RELOC: > + parms->phys_reloc = true; > + > + if ( descsz >= 4 ) > + { > + parms->phys_max = elf_note_numeric_array(elf, note, 4, 0); > + elf_msg(elf, " = max: %#"PRIx32, parms->phys_max); As indicated before, I consider the = here a little odd. > + } > + if ( descsz >= 8 ) > + { > + parms->phys_min = elf_note_numeric_array(elf, note, 4, 1); > + elf_msg(elf, " min: %#"PRIx32, parms->phys_min); > + } > + if ( descsz >= 12 ) > + { > + parms->phys_align = elf_note_numeric_array(elf, note, 4, 2); > + elf_msg(elf, " align: %#"PRIx32, parms->phys_align); > + } I'd like us to reconsider this ordering: I'm inclined to say that MAX isn't the most likely one a guest may find a need to use. Instead I'd expect both MIN and ALIGN wanting to be given higher priority; what I'm less certain about is the ordering between the two. To keep MIN and MAX adjacent, how about ALIGN, MIN, MAX? > --- a/xen/include/public/elfnote.h > +++ b/xen/include/public/elfnote.h > @@ -194,10 +194,27 @@ > */ > #define XEN_ELFNOTE_PHYS32_ENTRY 18 > > +/* > + * Physical loading constraints for PVH kernels > + * > + * The presence of this note indicates the kernel supports relocating itself. > + * > + * The note may include up to three 32bit values to place constraints on the > + * guest physical loading addresses and alignment for a PVH kernel. Values > + * are read in the following order: > + * - a maximum address for the entire image to be loaded below (default > + * 0xffffffff) "below" isn't exactly true anymore with this now being an inclusive value. Perhaps "up to", or perhaps more of a re-wording. I also think the wrapped line's indentation is too deep (by 2 blanks). > + * - a minimum address for the start of the image (default 0) > + * - a required start alignment (default 0x200000) > + * > + * This note is only valid for x86 binaries. Maybe s/valid/recognized/ (or honored or some such)? Jan
On 2024-03-26 03:50, Jan Beulich wrote: > On 25.03.2024 21:45, Jason Andryuk wrote: >> +/* Find an e820 RAM region that fits the kernel at a suitable alignment. */ >> +static paddr_t __init find_kernel_memory( >> + const struct domain *d, struct elf_binary *elf, >> + const struct elf_dom_parms *parms) >> +{ >> + paddr_t kernel_size = elf->dest_size; >> + unsigned int i; >> + >> + for ( i = 0; i < d->arch.nr_e820; i++ ) >> + { >> + paddr_t start = d->arch.e820[i].addr; >> + paddr_t end = start + d->arch.e820[i].size; >> + paddr_t kstart, kend; >> + >> + if ( d->arch.e820[i].type != E820_RAM || >> + d->arch.e820[i].size < kernel_size ) >> + continue; >> + >> + kstart = ROUNDUP(start, parms->phys_align); >> + kstart = max_t(paddr_t, kstart, parms->phys_min); >> + kend = kstart + kernel_size; >> + >> + if ( kend - 1 > parms->phys_max ) >> + return 0; >> + >> + if ( kend <= end ) >> + return kstart; > > IOW within a suitable region the lowest suitable part is selected. Often > low memory is deemed more precious than higher one, so if this choice is > indeed intentional, I'd like to ask for a brief comment towards the > reasons. It is not particularly intentional. I'll look into locating at a higher address. >> --- a/xen/common/libelf/libelf-dominfo.c >> +++ b/xen/common/libelf/libelf-dominfo.c >> @@ -17,6 +17,16 @@ >> >> #include "libelf-private.h" >> >> +#if defined(__i386__) || defined(__x86_64__) >> +#define ARCH_PHYS_MIN_DEFAULT 0; >> +#define ARCH_PHYS_MAX_DEFAULT (GB(4) - 1); >> +#define ARCH_PHYS_ALIGN_DEFAULT MB(2); >> +#else >> +#define ARCH_PHYS_MIN_DEFAULT 0; >> +#define ARCH_PHYS_MAX_DEFAULT 0; >> +#define ARCH_PHYS_ALIGN_DEFAULT 0; >> +#endif > > None of the semicolons should really be here. Yes, sorry. I inadvertently retained them when reworking this. >> @@ -227,6 +239,27 @@ elf_errorstatus elf_xen_parse_note(struct elf_binary *elf, >> case XEN_ELFNOTE_PHYS32_ENTRY: >> parms->phys_entry = val; >> break; >> + >> + case XEN_ELFNOTE_PHYS32_RELOC: >> + parms->phys_reloc = true; >> + >> + if ( descsz >= 4 ) >> + { >> + parms->phys_max = elf_note_numeric_array(elf, note, 4, 0); >> + elf_msg(elf, " = max: %#"PRIx32, parms->phys_max); > > As indicated before, I consider the = here a little odd. I retained = for consistency with other notes: ELF: note: PHYS32_RELOC = max: 0x40000000 min: 0x1000000 align: 0x200000 ELF: note: PHYS32_ENTRY = 0x1000000 ELF: note: GUEST_OS = "linux" I guess whitespace and labels makes it clear, so I'll drop the '='. >> + } >> + if ( descsz >= 8 ) >> + { >> + parms->phys_min = elf_note_numeric_array(elf, note, 4, 1); >> + elf_msg(elf, " min: %#"PRIx32, parms->phys_min); >> + } >> + if ( descsz >= 12 ) >> + { >> + parms->phys_align = elf_note_numeric_array(elf, note, 4, 2); >> + elf_msg(elf, " align: %#"PRIx32, parms->phys_align); >> + } > > I'd like us to reconsider this ordering: I'm inclined to say that MAX isn't > the most likely one a guest may find a need to use. Instead I'd expect both > MIN and ALIGN wanting to be given higher priority; what I'm less certain > about is the ordering between the two. To keep MIN and MAX adjacent, how > about ALIGN, MIN, MAX? ALIGN, MIN, MAX works for me. On the Linux side, I'm expecting them all to be set: ALIGN = CONFIG_PHYSICAL_ALIGN MIN = LOAD_PHYSICAL_ADDR MAX = KERNEL_IMAGE_SIZE You need enough identity page tables to cover up to MAX. LOAD_PHYSICAL_ADDR is used as a minimum, so requesting placement above MIN makes sense to me. >> --- a/xen/include/public/elfnote.h >> +++ b/xen/include/public/elfnote.h >> @@ -194,10 +194,27 @@ >> */ >> #define XEN_ELFNOTE_PHYS32_ENTRY 18 >> >> +/* >> + * Physical loading constraints for PVH kernels >> + * >> + * The presence of this note indicates the kernel supports relocating itself. >> + * >> + * The note may include up to three 32bit values to place constraints on the >> + * guest physical loading addresses and alignment for a PVH kernel. Values >> + * are read in the following order: >> + * - a maximum address for the entire image to be loaded below (default >> + * 0xffffffff) > > "below" isn't exactly true anymore with this now being an inclusive value. > Perhaps "up to", or perhaps more of a re-wording. Yes, good point. > I also think the wrapped line's indentation is too deep (by 2 blanks). Yes, thanks. >> + * - a minimum address for the start of the image (default 0) >> + * - a required start alignment (default 0x200000) >> + * >> + * This note is only valid for x86 binaries. > > Maybe s/valid/recognized/ (or honored or some such)? Would a comment at the top of the file saying Notes are only used with x86 be better instead of this one-off comment? Roger already said that, and elf_xen_note_check() has a successful early exit with "ELF: Not bothering with notes on ARM\n" Thanks, Jason
On 26.03.2024 14:24, Jason Andryuk wrote: > On 2024-03-26 03:50, Jan Beulich wrote: >> On 25.03.2024 21:45, Jason Andryuk wrote: >>> + * - a minimum address for the start of the image (default 0) >>> + * - a required start alignment (default 0x200000) >>> + * >>> + * This note is only valid for x86 binaries. >> >> Maybe s/valid/recognized/ (or honored or some such)? > > Would a comment at the top of the file saying Notes are only used with > x86 be better instead of this one-off comment? Roger already said that, > and elf_xen_note_check() has a successful early exit with "ELF: Not > bothering with notes on ARM\n" If truly none of the notes are of interest for Arm, then yes, such a more general comment would likely make sense. Jan
On 26.03.2024 14:24, Jason Andryuk wrote: > On 2024-03-26 03:50, Jan Beulich wrote: >> On 25.03.2024 21:45, Jason Andryuk wrote: >>> @@ -227,6 +239,27 @@ elf_errorstatus elf_xen_parse_note(struct elf_binary *elf, >>> case XEN_ELFNOTE_PHYS32_ENTRY: >>> parms->phys_entry = val; >>> break; >>> + >>> + case XEN_ELFNOTE_PHYS32_RELOC: >>> + parms->phys_reloc = true; >>> + >>> + if ( descsz >= 4 ) >>> + { >>> + parms->phys_max = elf_note_numeric_array(elf, note, 4, 0); >>> + elf_msg(elf, " = max: %#"PRIx32, parms->phys_max); >> >> As indicated before, I consider the = here a little odd. > > I retained = for consistency with other notes: > ELF: note: PHYS32_RELOC = max: 0x40000000 min: 0x1000000 align: 0x200000 > ELF: note: PHYS32_ENTRY = 0x1000000 > ELF: note: GUEST_OS = "linux" > > I guess whitespace and labels makes it clear, so I'll drop the '='. > >>> + } >>> + if ( descsz >= 8 ) >>> + { >>> + parms->phys_min = elf_note_numeric_array(elf, note, 4, 1); >>> + elf_msg(elf, " min: %#"PRIx32, parms->phys_min); >>> + } >>> + if ( descsz >= 12 ) >>> + { >>> + parms->phys_align = elf_note_numeric_array(elf, note, 4, 2); >>> + elf_msg(elf, " align: %#"PRIx32, parms->phys_align); >>> + } >> >> I'd like us to reconsider this ordering: I'm inclined to say that MAX isn't >> the most likely one a guest may find a need to use. Instead I'd expect both >> MIN and ALIGN wanting to be given higher priority; what I'm less certain >> about is the ordering between the two. To keep MIN and MAX adjacent, how >> about ALIGN, MIN, MAX? > > ALIGN, MIN, MAX works for me. > > On the Linux side, I'm expecting them all to be set: > ALIGN = CONFIG_PHYSICAL_ALIGN > MIN = LOAD_PHYSICAL_ADDR > MAX = KERNEL_IMAGE_SIZE > > You need enough identity page tables to cover up to MAX. > LOAD_PHYSICAL_ADDR is used as a minimum, so requesting placement above > MIN makes sense to me. Hmm, setting MIN like this means moving down is precluded. Why would it not be possible to move a kernel to lower than the default of 16M, when CONFIG_PHYSICAL_START can be as low as 0? (In fact, I doubt 0 would work if chosen, but 2M surely does work, as I build some of my Dom0 kernels that way.) MAX, otoh, I guess really wants setting as you say, for KERNEL_IMAGE_SIZE actually being commented upon as mis-named. Just that it now really wants to be KERNEL_IMAGE_SIZE-1. Jan
On 2024-03-26 09:40, Jan Beulich wrote: > On 26.03.2024 14:24, Jason Andryuk wrote: >> On 2024-03-26 03:50, Jan Beulich wrote: >>> On 25.03.2024 21:45, Jason Andryuk wrote: >>>> @@ -227,6 +239,27 @@ elf_errorstatus elf_xen_parse_note(struct elf_binary *elf, >>>> + } >>>> + if ( descsz >= 8 ) >>>> + { >>>> + parms->phys_min = elf_note_numeric_array(elf, note, 4, 1); >>>> + elf_msg(elf, " min: %#"PRIx32, parms->phys_min); >>>> + } >>>> + if ( descsz >= 12 ) >>>> + { >>>> + parms->phys_align = elf_note_numeric_array(elf, note, 4, 2); >>>> + elf_msg(elf, " align: %#"PRIx32, parms->phys_align); >>>> + } >>> >>> I'd like us to reconsider this ordering: I'm inclined to say that MAX isn't >>> the most likely one a guest may find a need to use. Instead I'd expect both >>> MIN and ALIGN wanting to be given higher priority; what I'm less certain >>> about is the ordering between the two. To keep MIN and MAX adjacent, how >>> about ALIGN, MIN, MAX? >> >> ALIGN, MIN, MAX works for me. >> >> On the Linux side, I'm expecting them all to be set: >> ALIGN = CONFIG_PHYSICAL_ALIGN >> MIN = LOAD_PHYSICAL_ADDR >> MAX = KERNEL_IMAGE_SIZE >> >> You need enough identity page tables to cover up to MAX. >> LOAD_PHYSICAL_ADDR is used as a minimum, so requesting placement above >> MIN makes sense to me. > > Hmm, setting MIN like this means moving down is precluded. Why would it > not be possible to move a kernel to lower than the default of 16M, when > CONFIG_PHYSICAL_START can be as low as 0? (In fact, I doubt 0 would work > if chosen, but 2M surely does work, as I build some of my Dom0 kernels > that way.) I successfully booted at a lower address when testing, so it's possible. The bzImage early boot code uses LOAD_PHYSICAL_ADDR as a minimum for extracting vmlinux, so I matched that. It's not clear to me exactly why that is used, though it avoids using the 16MB ISA DMA region. Kconfig RELOCATABLE has this: Note: If CONFIG_RELOCATABLE=y, then the kernel runs from the address it has been loaded at and the compile time physical address (CONFIG_PHYSICAL_START) is used as the minimum location. Which is again why I thought to use it as MIN. > MAX, otoh, I guess really wants setting as you say, for KERNEL_IMAGE_SIZE > actually being commented upon as mis-named. Just that it now really wants > to be KERNEL_IMAGE_SIZE-1. Yes. If placement changes to favor higher addresses, then ALIGN, MAX, MIN becomes a little more important since that should be consulted first. Thanks, Jason
diff --git a/xen/arch/x86/hvm/dom0_build.c b/xen/arch/x86/hvm/dom0_build.c index 0ceda4140b..d925fc7417 100644 --- a/xen/arch/x86/hvm/dom0_build.c +++ b/xen/arch/x86/hvm/dom0_build.c @@ -537,6 +537,97 @@ static paddr_t __init find_memory( return INVALID_PADDR; } +static bool __init check_load_address( + const struct domain *d, const struct elf_binary *elf) +{ + paddr_t kernel_start = (uintptr_t)elf->dest_base; + paddr_t kernel_end = kernel_start + elf->dest_size; + unsigned int i; + + /* Relies on a sorted memory map with adjacent entries merged. */ + for ( i = 0; i < d->arch.nr_e820; i++ ) + { + paddr_t start = d->arch.e820[i].addr; + paddr_t end = start + d->arch.e820[i].size; + + if ( start >= kernel_end ) + return false; + + if ( d->arch.e820[i].type == E820_RAM && + start <= kernel_start && + end >= kernel_end ) + return true; + } + + return false; +} + +/* Find an e820 RAM region that fits the kernel at a suitable alignment. */ +static paddr_t __init find_kernel_memory( + const struct domain *d, struct elf_binary *elf, + const struct elf_dom_parms *parms) +{ + paddr_t kernel_size = elf->dest_size; + unsigned int i; + + for ( i = 0; i < d->arch.nr_e820; i++ ) + { + paddr_t start = d->arch.e820[i].addr; + paddr_t end = start + d->arch.e820[i].size; + paddr_t kstart, kend; + + if ( d->arch.e820[i].type != E820_RAM || + d->arch.e820[i].size < kernel_size ) + continue; + + kstart = ROUNDUP(start, parms->phys_align); + kstart = max_t(paddr_t, kstart, parms->phys_min); + kend = kstart + kernel_size; + + if ( kend - 1 > parms->phys_max ) + return 0; + + if ( kend <= end ) + return kstart; + } + + return 0; +} + +/* Check the kernel load address, and adjust if necessary and possible. */ +static bool __init check_and_adjust_load_address( + const struct domain *d, struct elf_binary *elf, struct elf_dom_parms *parms) +{ + paddr_t reloc_base; + + if ( check_load_address(d, elf) ) + return true; + + if ( !parms->phys_reloc ) + { + printk("%pd kernel: Address conflict and not relocatable\n", d); + return false; + } + + reloc_base = find_kernel_memory(d, elf, parms); + if ( !reloc_base ) + { + printk("%pd kernel: Failed find a load address\n", d); + return false; + } + + if ( opt_dom0_verbose ) + printk("%pd kernel: Moving [%p, %p] -> [%"PRIpaddr", %"PRIpaddr"]\n", d, + elf->dest_base, elf->dest_base + elf->dest_size - 1, + reloc_base, reloc_base + elf->dest_size - 1); + + parms->phys_entry = reloc_base + + (parms->phys_entry - (uintptr_t)elf->dest_base); + elf->dest_base = (char *)reloc_base; + + return true; +} + static int __init pvh_load_kernel(struct domain *d, const module_t *image, unsigned long image_headroom, module_t *initrd, void *image_base, @@ -585,6 +676,9 @@ static int __init pvh_load_kernel(struct domain *d, const module_t *image, elf.dest_base = (void *)(parms.virt_kstart - parms.virt_base); elf.dest_size = parms.virt_kend - parms.virt_kstart; + if ( !check_and_adjust_load_address(d, &elf, &parms) ) + return -ENOSPC; + elf_set_vcpu(&elf, v); rc = elf_load_binary(&elf); if ( rc < 0 ) diff --git a/xen/common/libelf/libelf-dominfo.c b/xen/common/libelf/libelf-dominfo.c index e7b44d238b..b47d540023 100644 --- a/xen/common/libelf/libelf-dominfo.c +++ b/xen/common/libelf/libelf-dominfo.c @@ -17,6 +17,16 @@ #include "libelf-private.h" +#if defined(__i386__) || defined(__x86_64__) +#define ARCH_PHYS_MIN_DEFAULT 0; +#define ARCH_PHYS_MAX_DEFAULT (GB(4) - 1); +#define ARCH_PHYS_ALIGN_DEFAULT MB(2); +#else +#define ARCH_PHYS_MIN_DEFAULT 0; +#define ARCH_PHYS_MAX_DEFAULT 0; +#define ARCH_PHYS_ALIGN_DEFAULT 0; +#endif + /* ------------------------------------------------------------------------ */ /* xen features */ @@ -125,6 +135,7 @@ elf_errorstatus elf_xen_parse_note(struct elf_binary *elf, [XEN_ELFNOTE_SUSPEND_CANCEL] = { "SUSPEND_CANCEL", ELFNOTE_INT }, [XEN_ELFNOTE_MOD_START_PFN] = { "MOD_START_PFN", ELFNOTE_INT }, [XEN_ELFNOTE_PHYS32_ENTRY] = { "PHYS32_ENTRY", ELFNOTE_INT }, + [XEN_ELFNOTE_PHYS32_RELOC] = { "PHYS32_RELOC", ELFNOTE_NAME }, }; /* *INDENT-ON* */ @@ -132,6 +143,7 @@ elf_errorstatus elf_xen_parse_note(struct elf_binary *elf, uint64_t val = 0; unsigned int i; unsigned type = elf_uval(elf, note, type); + unsigned descsz = elf_uval(elf, note, descsz); if ( (type >= sizeof(note_desc) / sizeof(note_desc[0])) || (note_desc[type].name == NULL) ) @@ -227,6 +239,27 @@ elf_errorstatus elf_xen_parse_note(struct elf_binary *elf, case XEN_ELFNOTE_PHYS32_ENTRY: parms->phys_entry = val; break; + + case XEN_ELFNOTE_PHYS32_RELOC: + parms->phys_reloc = true; + + if ( descsz >= 4 ) + { + parms->phys_max = elf_note_numeric_array(elf, note, 4, 0); + elf_msg(elf, " = max: %#"PRIx32, parms->phys_max); + } + if ( descsz >= 8 ) + { + parms->phys_min = elf_note_numeric_array(elf, note, 4, 1); + elf_msg(elf, " min: %#"PRIx32, parms->phys_min); + } + if ( descsz >= 12 ) + { + parms->phys_align = elf_note_numeric_array(elf, note, 4, 2); + elf_msg(elf, " align: %#"PRIx32, parms->phys_align); + } + + break; } if ( note_desc[type].type == ELFNOTE_NAME) @@ -542,6 +575,10 @@ elf_errorstatus elf_xen_parse(struct elf_binary *elf, parms->p2m_base = UNSET_ADDR; parms->elf_paddr_offset = UNSET_ADDR; parms->phys_entry = UNSET_ADDR32; + parms->phys_min = ARCH_PHYS_MIN_DEFAULT; + parms->phys_max = ARCH_PHYS_MAX_DEFAULT; + parms->phys_align = ARCH_PHYS_ALIGN_DEFAULT; + parms->phys_reloc = false; /* Find and parse elf notes. */ count = elf_phdr_count(elf); diff --git a/xen/common/libelf/libelf-private.h b/xen/common/libelf/libelf-private.h index 47db679966..98cac65bc5 100644 --- a/xen/common/libelf/libelf-private.h +++ b/xen/common/libelf/libelf-private.h @@ -71,6 +71,7 @@ #endif #include <xen/elfnote.h> #include <xen/libelf/libelf.h> +#include <xen-tools/common-macros.h> #ifndef FUZZ_NO_LIBXC #include "xenctrl.h" diff --git a/xen/include/public/elfnote.h b/xen/include/public/elfnote.h index 8bf54d035b..ed87d5575d 100644 --- a/xen/include/public/elfnote.h +++ b/xen/include/public/elfnote.h @@ -194,10 +194,27 @@ */ #define XEN_ELFNOTE_PHYS32_ENTRY 18 +/* + * Physical loading constraints for PVH kernels + * + * The presence of this note indicates the kernel supports relocating itself. + * + * The note may include up to three 32bit values to place constraints on the + * guest physical loading addresses and alignment for a PVH kernel. Values + * are read in the following order: + * - a maximum address for the entire image to be loaded below (default + * 0xffffffff) + * - a minimum address for the start of the image (default 0) + * - a required start alignment (default 0x200000) + * + * This note is only valid for x86 binaries. + */ +#define XEN_ELFNOTE_PHYS32_RELOC 19 + /* * The number of the highest elfnote defined. */ -#define XEN_ELFNOTE_MAX XEN_ELFNOTE_PHYS32_ENTRY +#define XEN_ELFNOTE_MAX XEN_ELFNOTE_PHYS32_RELOC /* * System information exported through crash notes. diff --git a/xen/include/xen/libelf.h b/xen/include/xen/libelf.h index 1c77e3df31..777c5008ca 100644 --- a/xen/include/xen/libelf.h +++ b/xen/include/xen/libelf.h @@ -434,10 +434,14 @@ struct elf_dom_parms { uint32_t f_supported[XENFEAT_NR_SUBMAPS]; uint32_t f_required[XENFEAT_NR_SUBMAPS]; uint32_t phys_entry; + uint32_t phys_min; + uint32_t phys_max; + uint32_t phys_align; /* calculated */ uint64_t virt_kstart; uint64_t virt_kend; + bool phys_reloc; }; static inline void elf_xen_feature_set(int nr, uint32_t * addr)
Xen tries to load a PVH dom0 kernel at the fixed guest physical address from the elf headers. For Linux, this defaults to 0x1000000 (16MB), but it can be configured. Unfortunately there exist firmwares that have reserved regions at this address, so Xen fails to load the dom0 kernel since it's not RAM. The PVH entry code is not relocatable - it loads from absolute addresses, which fail when the kernel is loaded at a different address. With a suitably modified kernel, a reloctable entry point is possible. Add XEN_ELFNOTE_PHYS32_RELOC which specifies optional minimum, maximum and alignment needed for the kernel. The presence of the NOTE indicates the kernel supports a relocatable entry path. Change the loading to check for an acceptable load address. If the kernel is relocatable, support finding an alternate load address. The primary motivation for an explicit align field is that Linux has a configurable CONFIG_PHYSICAL_ALIGN field. This value is present in the bzImage setup header, but not the ELF program headers p_align, which report 2MB even when CONFIG_PHYSICAL_ALIGN is greater. Since a kernel is only considered relocatable if the PHYS32_RELOC elf note is present, the alignment contraints can just be specified within the note instead of searching for an alignment value via a heuristic. libelf-private.h includes common-macros.h to satisfy the fuzzer build. Link: https://gitlab.com/xen-project/xen/-/issues/180 Signed-off-by: Jason Andryuk <jason.andryuk@amd.com> --- ELF Note printing looks like: (XEN) ELF: note: PHYS32_RELOC = max: 0xffffffff min: 0x1000000 align: 0x200000 v2: Use elfnote for min, max & align - use 64bit values. Print original and relocated memory addresses Use check_and_adjust_load_address() name Return relocated base instead of offset Use PAGE_ALIGN Don't load above max_phys (expected to be 4GB in kernel elf note) Use single line comments Exit check_load_address loop earlier Add __init to find_kernel_memory() v3: Remove kernel_start/end page rounding Change loop comment to rely on a sorted memory map. Reorder E820_RAM check first Use %p for dest_base Use PRIpaddr Use 32bit phys_min/max/align Consolidate to if ( x || y ) continue Use max_t Add parms->phys_reloc Use common "%pd kernel: " prefix for messages Re-order phys_entry assignment Print range ends inclusively Remove extra "Unable to load kernel" message s/PVH_RELOCATION/PHYS32_RELOC/ Make PHYS32_RELOC contents optional Use 2MB default alignment Update ELF Note comment Update XEN_ELFNOTE_MAX v4: Cast dest_base to uintptr_t Use local start variable Mention e820 requiring adjacent entries merged Remove extra whitespace Re-word elfnote comment & mention x86 Use ELFNOTE_NAME Return -ENOSPC Use ! instead of == 0 Check kend - 1 to avoid off by one libelf: Use MB/GB() to define the size. Use ARCH_PHYS_* defines --- xen/arch/x86/hvm/dom0_build.c | 94 ++++++++++++++++++++++++++++++ xen/common/libelf/libelf-dominfo.c | 37 ++++++++++++ xen/common/libelf/libelf-private.h | 1 + xen/include/public/elfnote.h | 19 +++++- xen/include/xen/libelf.h | 4 ++ 5 files changed, 154 insertions(+), 1 deletion(-)