|
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] Re: [PATCH v4 2/3] x86/EFI: replace ebmalloc()
On 18.11.2026 14:26, Andrew Cooper wrote:
> On 17/11/2026 7:37 am, Jan Beulich wrote:
>> On 16.11.2026 19:13, Andrew Cooper wrote:
>>> On 03/11/2026 12:50 pm, Jan Beulich wrote:
>>>> --- a/xen/arch/x86/include/asm/setup.h
>>>> +++ b/xen/arch/x86/include/asm/setup.h
>>>> @@ -9,6 +9,8 @@ extern const char __2M_rodata_start[], _
>>>> extern char __2M_init_start[], __2M_init_end[];
>>>> extern char __2M_rwdata_start[], __2M_rwdata_end[];
>>>>
>>>> +extern unsigned long brk_end;
>>>> +
>>>> extern unsigned long xenheap_initial_phys_start;
>>>> extern uint64_t boot_tsc_stamp;
>>>>
>>>> --- a/xen/arch/x86/setup.c
>>>> +++ b/xen/arch/x86/setup.c
>>>> @@ -1,6 +1,7 @@
>>>> #include <xen/acpi.h>
>>>> #include <xen/alternative-call.h>
>>>> #include <xen/bitops.h>
>>>> +#include <xen/brk.h>
>>>> #include <xen/console.h>
>>>> #include <xen/cpu.h>
>>>> #include <xen/cpuidle.h>
>>>> @@ -164,6 +165,8 @@ cpumask_t __read_mostly cpu_present_map;
>>>>
>>>> unsigned long __read_mostly xen_phys_start;
>>>>
>>>> +unsigned long __ro_after_init brk_end;
>>>> +
>>> I know it's only x86 which uses brk_end right now, but given uses such
>>> as ...
>>>
>>>
>>>> /* Only used in asm code and within this source file */
>>>> char asmlinkage __section(".init.bss.stack_aligned") __aligned(STACK_SIZE)
>>>> cpu0_stack[STACK_SIZE];
>>>> @@ -1157,7 +1160,6 @@ void asmlinkage __init noreturn __start_
>>>> struct boot_info *bi;
>>>> unsigned long raw_max_page;
>>>> int i, j, bytes = 0;
>>>> - unsigned long eb_start, eb_end;
>>>> bool acpi_boot_table_init_done = false, relocated = false;
>>>> bool vm_init_done = false;
>>>> int ret;
>>>> @@ -1523,7 +1525,7 @@ void asmlinkage __init noreturn __start_
>>>> /*
>>>> * This needs to remain in sync with remove_xen_ranges() and the
>>>> * respective reserve_e820_ram() invocation below. No need to
>>>> - * query efi_boot_mem_unused() here, though.
>>>> + * query brk_cease() here, though.
>>>> */
>>>> xen->start = virt_to_maddr(_stext);
>>>> xen->size = __2M_rwdata_end - _stext;
>>>> @@ -1667,20 +1669,12 @@ void asmlinkage __init noreturn __start_
>>>> if ( !xen_phys_start )
>>>> panic("Not enough memory to relocate Xen\n");
>>>>
>>>> - /* FIXME: Putting a hole in .bss would shatter the large page
>>>> mapping. */
>>>> - if ( using_2M_mapping() )
>>>> - efi_boot_mem_unused(NULL, NULL);
>>>> -
>>>> /* This needs to remain in sync with remove_xen_ranges(). */
>>>> - if ( efi_boot_mem_unused(&eb_start, &eb_end) )
>>>> - {
>>>> - if ( !reserve_e820_ram(&boot_e820, __pa(_stext), __pa(eb_start))
>>>> ||
>>>> - !reserve_e820_ram(&boot_e820, __pa(eb_end),
>>>> __pa(__2M_rwdata_end)) )
>>>> - panic("Failed to reserve Xen in E820\n");
>>>> - }
>>>> - else
>>>> - if ( !reserve_e820_ram(&boot_e820, __pa(_stext),
>>>> __pa(__2M_rwdata_end)) )
>>>> - panic("Failed to reserve Xen in E820\n");
>>>> + brk_end = brk_cease();
>>>> + if ( using_2M_mapping() )
>>>> + brk_end = PAGE_ALIGN_2M(brk_end);
>>>> + if ( !reserve_e820_ram(&boot_e820, __pa(_stext), __pa(brk_end)) )
>>>> + panic("Failed to reserve Xen in E820\n");
>>>>
>>>> /* Late kexec reservation (dynamic start address). */
>>>> kexec_reserve_area();
>>>> @@ -2274,7 +2268,6 @@ __initcall(init_xen_cap_info);
>>>>
>>>> int __hwdom_init remove_xen_ranges(struct rangeset *r)
>>>> {
>>>> - paddr_t start, end;
>>>> int rc;
>>>>
>>>> /* S3 resume code (and other real mode trampoline code) */
>>>> @@ -2296,26 +2289,10 @@ int __hwdom_init remove_xen_ranges(struc
>>>> return rc;
>>>>
>>>> /* hypervisor .data + .bss */
>>>> - if ( efi_boot_mem_unused(&start, &end) )
>>>> - {
>>>> - ASSERT(__pa(start) >= __pa(&__2M_rwdata_start));
>>>> - rc = rangeset_remove_range(r, PFN_DOWN(__pa(&__2M_rwdata_start)),
>>>> - PFN_DOWN(__pa(start) - 1));
>>>> - if ( rc )
>>>> - return rc;
>>>> - ASSERT(__pa(end) <= __pa(&__2M_rwdata_end));
>>>> - rc = rangeset_remove_range(r, PFN_DOWN(__pa(end)),
>>>> - PFN_DOWN(__pa(&__2M_rwdata_end) - 1));
>>>> - if ( rc )
>>>> - return rc;
>>>> - }
>>>> - else
>>>> - {
>>>> - rc = rangeset_remove_range(r, PFN_DOWN(__pa(&__2M_rwdata_start)),
>>>> - PFN_DOWN(__pa(&__2M_rwdata_end) - 1));
>>>> - if ( rc )
>>>> - return rc;
>>>> - }
>>>> + rc = rangeset_remove_range(r, PFN_DOWN(__pa(&__2M_rwdata_start)),
>>>> + PFN_DOWN(__pa(brk_end) - 1));
>>>> + if ( rc )
>>>> + return rc;
>>> ... this, it clearly needs to be common in order for other architectures
>>> to start using the allocator.
>> I don't see why this would necessarily be this way. This is x86-specific
>> memory management, and other ports may have no need. I'd therefore prefer
>> to leave brk_end x86-only until proven otherwise. It was, after all, a
>> deliberate decision to make this x86-only.
>
> I'm very unconvinced.
>
> An architecture that doesn't care to reclaim unused space just uses
> plain data/bss objects, which is what happens today.
>
> An architecture will only care to switch to brk in order to reclaim
> space, and to do so it *must* replace all of it's current end markers
> with brk_end.
Or we want to start using brk_alloc() in common code, without forcing
all ports to deal with the freeing of excess space. They then don't
need to record the return value of brk_cease(), they don't need to
call brk_free_unused(), and they would have no need for any form of
brk_end.
> brk_end will probably only be consumed by arch-specific code, but it's
> produced by common code and therefore it needs to be part of the common
> brk infrastructure. This also has the important property of preventing
> it from getting named differently in different architectures.
Well, as almost always (no matter that as the patch author I think I
should have some room to choose between options) it's going to be me
to give in. Just to finally get this done.
Jan
|
![]() |
Lists.xenproject.org is hosted with RackSpace, monitoring our |