| Message ID | 1569217425-23777-3-git-send-email-anshuman.khandual@arm.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | arm64/mm: Enable memory hot remove | expand |
On Mon, Sep 23, 2019 at 11:13:45AM +0530, Anshuman Khandual wrote: > +#ifdef CONFIG_MEMORY_HOTPLUG > +static void free_hotplug_page_range(struct page *page, size_t size) > +{ > + WARN_ON(!page || PageReserved(page)); WARN_ON(!page) isn't terribly useful. You're going to crash on the very next line when you call page_address() anyway. If this line were if (WARN_ON(!page || PageReserved(page))) return; it would make sense, or if it were just WARN_ON(PageReserved(page)) it would also make sense. > + free_pages((unsigned long)page_address(page), get_order(size)); > +}
On 09/23/2019 04:47 PM, Matthew Wilcox wrote: > On Mon, Sep 23, 2019 at 11:13:45AM +0530, Anshuman Khandual wrote: >> +#ifdef CONFIG_MEMORY_HOTPLUG >> +static void free_hotplug_page_range(struct page *page, size_t size) >> +{ >> + WARN_ON(!page || PageReserved(page)); > > WARN_ON(!page) isn't terribly useful. You're going to crash on the very > next line when you call page_address() anyway. If this line were > > if (WARN_ON(!page || PageReserved(page))) > return; > > it would make sense, or if it were just > > WARN_ON(PageReserved(page)) > > it would also make sense. I guess WARN_ON(PageReserved(page)) should be good enough to make sure that page being freed here was originally allocated at runtime for a previous memory hot add operation and did not some how come from the memblock reserved area. That was the original objective for this check. Will change it. > >> + free_pages((unsigned long)page_address(page), get_order(size)); >> +} >
On Mon, Sep 23, 2019 at 11:13:45AM +0530, Anshuman Khandual wrote: > The arch code for hot-remove must tear down portions of the linear map and > vmemmap corresponding to memory being removed. In both cases the page > tables mapping these regions must be freed, and when sparse vmemmap is in > use the memory backing the vmemmap must also be freed. > > This patch adds unmap_hotplug_range() and free_empty_tables() helpers which > can be used to tear down either region and calls it from vmemmap_free() and > ___remove_pgd_mapping(). The sparse_vmap argument determines whether the > backing memory will be freed. Can you change the 'sparse_vmap' name to something more meaningful which would suggest freeing of the backing memory? > It makes two distinct passes over the kernel page table. In the first pass > with unmap_hotplug_range() it unmaps, invalidates applicable TLB cache and > frees backing memory if required (vmemmap) for each mapped leaf entry. In > the second pass with free_empty_tables() it looks for empty page table > sections whose page table page can be unmapped, TLB invalidated and freed. > > While freeing intermediate level page table pages bail out if any of its > entries are still valid. This can happen for partially filled kernel page > table either from a previously attempted failed memory hot add or while > removing an address range which does not span the entire page table page > range. > > The vmemmap region may share levels of table with the vmalloc region. > There can be conflicts between hot remove freeing page table pages with > a concurrent vmalloc() walking the kernel page table. This conflict can > not just be solved by taking the init_mm ptl because of existing locking > scheme in vmalloc(). So free_empty_tables() implements a floor and ceiling > method which is borrowed from user page table tear with free_pgd_range() > which skips freeing page table pages if intermediate address range is not > aligned or maximum floor-ceiling might not own the entire page table page. > > While here update arch_add_memory() to handle __add_pages() failures by > just unmapping recently added kernel linear mapping. Now enable memory hot > remove on arm64 platforms by default with ARCH_ENABLE_MEMORY_HOTREMOVE. > > This implementation is overall inspired from kernel page table tear down > procedure on X86 architecture and user page table tear down method. > > Acked-by: Steve Capper <steve.capper@arm.com> > Acked-by: David Hildenbrand <david@redhat.com> > Signed-off-by: Anshuman Khandual <anshuman.khandual@arm.com> Given the amount of changes since version 7, do the acks still stand? [...] > +static void free_pte_table(pmd_t *pmdp, unsigned long addr, unsigned long end, > + unsigned long floor, unsigned long ceiling) > +{ > + struct page *page; > + pte_t *ptep; > + int i; > + > + if (!pgtable_range_aligned(addr, end, floor, ceiling, PMD_MASK)) > + return; > + > + ptep = pte_offset_kernel(pmdp, 0UL); > + for (i = 0; i < PTRS_PER_PTE; i++) { > + if (!pte_none(READ_ONCE(ptep[i]))) > + return; > + } > + > + page = pmd_page(READ_ONCE(*pmdp)); Arguably, that's not the pmd page we are freeing here. Even if you get the same result, pmd_page() is normally used for huge pages pointed at by the pmd entry. Since you have the ptep already, why not use virt_to_page(ptep)? > + pmd_clear(pmdp); > + __flush_tlb_kernel_pgtable(addr); > + free_hotplug_pgtable_page(page); > +} > + > +static void free_pmd_table(pud_t *pudp, unsigned long addr, unsigned long end, > + unsigned long floor, unsigned long ceiling) > +{ > + struct page *page; > + pmd_t *pmdp; > + int i; > + > + if (CONFIG_PGTABLE_LEVELS <= 2) > + return; > + > + if (!pgtable_range_aligned(addr, end, floor, ceiling, PUD_MASK)) > + return; > + > + pmdp = pmd_offset(pudp, 0UL); > + for (i = 0; i < PTRS_PER_PMD; i++) { > + if (!pmd_none(READ_ONCE(pmdp[i]))) > + return; > + } > + > + page = pud_page(READ_ONCE(*pudp)); Same here, virt_to_page(pmdp). > + pud_clear(pudp); > + __flush_tlb_kernel_pgtable(addr); > + free_hotplug_pgtable_page(page); > +} > + > +static void free_pud_table(pgd_t *pgdp, unsigned long addr, unsigned long end, > + unsigned long floor, unsigned long ceiling) > +{ > + struct page *page; > + pud_t *pudp; > + int i; > + > + if (CONFIG_PGTABLE_LEVELS <= 3) > + return; > + > + if (!pgtable_range_aligned(addr, end, floor, ceiling, PGDIR_MASK)) > + return; > + > + pudp = pud_offset(pgdp, 0UL); > + for (i = 0; i < PTRS_PER_PUD; i++) { > + if (!pud_none(READ_ONCE(pudp[i]))) > + return; > + } > + > + page = pgd_page(READ_ONCE(*pgdp)); As above. > + pgd_clear(pgdp); > + __flush_tlb_kernel_pgtable(addr); > + free_hotplug_pgtable_page(page); > +} > + > +static void unmap_hotplug_pte_range(pmd_t *pmdp, unsigned long addr, > + unsigned long end, bool sparse_vmap) > +{ > + struct page *page; > + pte_t *ptep, pte; > + > + do { > + ptep = pte_offset_kernel(pmdp, addr); > + pte = READ_ONCE(*ptep); > + if (pte_none(pte)) > + continue; > + > + WARN_ON(!pte_present(pte)); > + page = sparse_vmap ? pte_page(pte) : NULL; > + pte_clear(&init_mm, addr, ptep); > + flush_tlb_kernel_range(addr, addr + PAGE_SIZE); > + if (sparse_vmap) > + free_hotplug_page_range(page, PAGE_SIZE); You could only set 'page' if sparse_vmap (or even drop 'page' entirely). The compiler is probably smart enough to optimise it but using a pointless ternary operator just makes the code harder to follow. > + } while (addr += PAGE_SIZE, addr < end); > +} [...] > +static void free_empty_pte_table(pmd_t *pmdp, unsigned long addr, > + unsigned long end) > +{ > + pte_t *ptep, pte; > + > + do { > + ptep = pte_offset_kernel(pmdp, addr); > + pte = READ_ONCE(*ptep); > + WARN_ON(!pte_none(pte)); > + } while (addr += PAGE_SIZE, addr < end); > +} > + > +static void free_empty_pmd_table(pud_t *pudp, unsigned long addr, > + unsigned long end, unsigned long floor, > + unsigned long ceiling) > +{ > + unsigned long next; > + pmd_t *pmdp, pmd; > + > + do { > + next = pmd_addr_end(addr, end); > + pmdp = pmd_offset(pudp, addr); > + pmd = READ_ONCE(*pmdp); > + if (pmd_none(pmd)) > + continue; > + > + WARN_ON(!pmd_present(pmd) || !pmd_table(pmd) || pmd_sect(pmd)); > + free_empty_pte_table(pmdp, addr, next); > + free_pte_table(pmdp, addr, next, floor, ceiling); Do we need two closely named functions here? Can you not collapse free_empty_pud_table() and free_pte_table() into a single one? The same comment for the pmd/pud variants. I just find this confusing. > + } while (addr = next, addr < end); You could make these function in two steps: first, as above, invoke the next level recursively; second, after the do..while loop, check whether it's empty and free the pmd page as in free_pmd_table(). > +} [...]
On 10/07/2019 07:47 PM, Catalin Marinas wrote: > On Mon, Sep 23, 2019 at 11:13:45AM +0530, Anshuman Khandual wrote: >> The arch code for hot-remove must tear down portions of the linear map and >> vmemmap corresponding to memory being removed. In both cases the page >> tables mapping these regions must be freed, and when sparse vmemmap is in >> use the memory backing the vmemmap must also be freed. >> >> This patch adds unmap_hotplug_range() and free_empty_tables() helpers which >> can be used to tear down either region and calls it from vmemmap_free() and >> ___remove_pgd_mapping(). The sparse_vmap argument determines whether the >> backing memory will be freed. > > Can you change the 'sparse_vmap' name to something more meaningful which > would suggest freeing of the backing memory? free_mapped_mem or free_backed_mem ? Even shorter forms like free_mapped or free_backed might do as well. Do you have a particular preference here ? But yes, sparse_vmap has been very much specific to vmemmap for these functions which are now very generic in nature. > >> It makes two distinct passes over the kernel page table. In the first pass >> with unmap_hotplug_range() it unmaps, invalidates applicable TLB cache and >> frees backing memory if required (vmemmap) for each mapped leaf entry. In >> the second pass with free_empty_tables() it looks for empty page table >> sections whose page table page can be unmapped, TLB invalidated and freed. >> >> While freeing intermediate level page table pages bail out if any of its >> entries are still valid. This can happen for partially filled kernel page >> table either from a previously attempted failed memory hot add or while >> removing an address range which does not span the entire page table page >> range. >> >> The vmemmap region may share levels of table with the vmalloc region. >> There can be conflicts between hot remove freeing page table pages with >> a concurrent vmalloc() walking the kernel page table. This conflict can >> not just be solved by taking the init_mm ptl because of existing locking >> scheme in vmalloc(). So free_empty_tables() implements a floor and ceiling >> method which is borrowed from user page table tear with free_pgd_range() >> which skips freeing page table pages if intermediate address range is not >> aligned or maximum floor-ceiling might not own the entire page table page. >> >> While here update arch_add_memory() to handle __add_pages() failures by >> just unmapping recently added kernel linear mapping. Now enable memory hot >> remove on arm64 platforms by default with ARCH_ENABLE_MEMORY_HOTREMOVE. >> >> This implementation is overall inspired from kernel page table tear down >> procedure on X86 architecture and user page table tear down method. >> >> Acked-by: Steve Capper <steve.capper@arm.com> >> Acked-by: David Hildenbrand <david@redhat.com> >> Signed-off-by: Anshuman Khandual <anshuman.khandual@arm.com> > > Given the amount of changes since version 7, do the acks still stand? I had taken the liberty to carry them till V7 where the implementation has been sort of structurally similar but as you mention, there have been some basic changes in the approach since V7. Will drop these tags in next version and request their fresh ACKs once again. > > [...] >> +static void free_pte_table(pmd_t *pmdp, unsigned long addr, unsigned long end, >> + unsigned long floor, unsigned long ceiling) >> +{ >> + struct page *page; >> + pte_t *ptep; >> + int i; >> + >> + if (!pgtable_range_aligned(addr, end, floor, ceiling, PMD_MASK)) >> + return; >> + >> + ptep = pte_offset_kernel(pmdp, 0UL); >> + for (i = 0; i < PTRS_PER_PTE; i++) { >> + if (!pte_none(READ_ONCE(ptep[i]))) >> + return; >> + } >> + >> + page = pmd_page(READ_ONCE(*pmdp)); > > Arguably, that's not the pmd page we are freeing here. Even if you get > the same result, pmd_page() is normally used for huge pages pointed at > by the pmd entry. Since you have the ptep already, why not use > virt_to_page(ptep)? Makes sense, will do. > >> + pmd_clear(pmdp); >> + __flush_tlb_kernel_pgtable(addr); >> + free_hotplug_pgtable_page(page); >> +} >> + >> +static void free_pmd_table(pud_t *pudp, unsigned long addr, unsigned long end, >> + unsigned long floor, unsigned long ceiling) >> +{ >> + struct page *page; >> + pmd_t *pmdp; >> + int i; >> + >> + if (CONFIG_PGTABLE_LEVELS <= 2) >> + return; >> + >> + if (!pgtable_range_aligned(addr, end, floor, ceiling, PUD_MASK)) >> + return; >> + >> + pmdp = pmd_offset(pudp, 0UL); >> + for (i = 0; i < PTRS_PER_PMD; i++) { >> + if (!pmd_none(READ_ONCE(pmdp[i]))) >> + return; >> + } >> + >> + page = pud_page(READ_ONCE(*pudp)); > > Same here, virt_to_page(pmdp). Will do. > >> + pud_clear(pudp); >> + __flush_tlb_kernel_pgtable(addr); >> + free_hotplug_pgtable_page(page); >> +} >> + >> +static void free_pud_table(pgd_t *pgdp, unsigned long addr, unsigned long end, >> + unsigned long floor, unsigned long ceiling) >> +{ >> + struct page *page; >> + pud_t *pudp; >> + int i; >> + >> + if (CONFIG_PGTABLE_LEVELS <= 3) >> + return; >> + >> + if (!pgtable_range_aligned(addr, end, floor, ceiling, PGDIR_MASK)) >> + return; >> + >> + pudp = pud_offset(pgdp, 0UL); >> + for (i = 0; i < PTRS_PER_PUD; i++) { >> + if (!pud_none(READ_ONCE(pudp[i]))) >> + return; >> + } >> + >> + page = pgd_page(READ_ONCE(*pgdp)); > > As above. Will do. > >> + pgd_clear(pgdp); >> + __flush_tlb_kernel_pgtable(addr); >> + free_hotplug_pgtable_page(page); >> +} >> + >> +static void unmap_hotplug_pte_range(pmd_t *pmdp, unsigned long addr, >> + unsigned long end, bool sparse_vmap) >> +{ >> + struct page *page; >> + pte_t *ptep, pte; >> + >> + do { >> + ptep = pte_offset_kernel(pmdp, addr); >> + pte = READ_ONCE(*ptep); >> + if (pte_none(pte)) >> + continue; >> + >> + WARN_ON(!pte_present(pte)); >> + page = sparse_vmap ? pte_page(pte) : NULL; >> + pte_clear(&init_mm, addr, ptep); >> + flush_tlb_kernel_range(addr, addr + PAGE_SIZE); >> + if (sparse_vmap) >> + free_hotplug_page_range(page, PAGE_SIZE); > > You could only set 'page' if sparse_vmap (or even drop 'page' entirely). I am afraid 'page' is being used to hold pte_page(pte) extraction which needs to be freed (sparse_vmap) as we are going to clear the ptep entry in the next statement and lose access to it for good. We will need some where to hold onto pte_page(pte) across pte_clear() as we cannot free it before clearing it's entry and flushing the TLB. Hence wondering how the 'page' can be completely dropped. > The compiler is probably smart enough to optimise it but using a > pointless ternary operator just makes the code harder to follow. Not sure I got this but are you suggesting for an 'if' statement here if (sparse_vmap) page = pte_page(pte); instead of the current assignment ? page = sparse_vmap ? pte_page(pte) : NULL; > >> + } while (addr += PAGE_SIZE, addr < end); >> +} > [...] >> +static void free_empty_pte_table(pmd_t *pmdp, unsigned long addr, >> + unsigned long end) >> +{ >> + pte_t *ptep, pte; >> + >> + do { >> + ptep = pte_offset_kernel(pmdp, addr); >> + pte = READ_ONCE(*ptep); >> + WARN_ON(!pte_none(pte)); >> + } while (addr += PAGE_SIZE, addr < end); >> +} >> + >> +static void free_empty_pmd_table(pud_t *pudp, unsigned long addr, >> + unsigned long end, unsigned long floor, >> + unsigned long ceiling) >> +{ >> + unsigned long next; >> + pmd_t *pmdp, pmd; >> + >> + do { >> + next = pmd_addr_end(addr, end); >> + pmdp = pmd_offset(pudp, addr); >> + pmd = READ_ONCE(*pmdp); >> + if (pmd_none(pmd)) >> + continue; >> + >> + WARN_ON(!pmd_present(pmd) || !pmd_table(pmd) || pmd_sect(pmd)); >> + free_empty_pte_table(pmdp, addr, next); >> + free_pte_table(pmdp, addr, next, floor, ceiling); > > Do we need two closely named functions here? Can you not collapse > free_empty_pud_table() and free_pte_table() into a single one? The same > comment for the pmd/pud variants. I just find this confusing. The two functions could be collapsed into a single one. But just wanted to keep free_pxx_table() part which checks floor/ceiling alignment, non-zero entries clear off the actual page table walking. > >> + } while (addr = next, addr < end); > > You could make these function in two steps: first, as above, invoke the > next level recursively; second, after the do..while loop, check whether > it's empty and free the pmd page as in free_pmd_table(). free_pte_table() freeing attempt actually belongs to free_empty_pte_table(). Yes, free_pte_table() part can be moved inside free_empty_pte_table() after it's do..while(). Also s/free_pte_table/free_pte_page to make it sound more distinct with respect to free_empty_pte_table(). Just that the pgtable page freeing part is still wrapped in a helper function to hide it's details. But if you prefer not to have these helpers free_pxx_page() and directly encode the second step in free_empty_pxx_table(), then will that instead. > >> +} > [...] >
On Tue, Oct 08, 2019 at 10:06:26AM +0530, Anshuman Khandual wrote: > On 10/07/2019 07:47 PM, Catalin Marinas wrote: > > On Mon, Sep 23, 2019 at 11:13:45AM +0530, Anshuman Khandual wrote: > >> The arch code for hot-remove must tear down portions of the linear map and > >> vmemmap corresponding to memory being removed. In both cases the page > >> tables mapping these regions must be freed, and when sparse vmemmap is in > >> use the memory backing the vmemmap must also be freed. > >> > >> This patch adds unmap_hotplug_range() and free_empty_tables() helpers which > >> can be used to tear down either region and calls it from vmemmap_free() and > >> ___remove_pgd_mapping(). The sparse_vmap argument determines whether the > >> backing memory will be freed. > > > > Can you change the 'sparse_vmap' name to something more meaningful which > > would suggest freeing of the backing memory? > > free_mapped_mem or free_backed_mem ? Even shorter forms like free_mapped or > free_backed might do as well. Do you have a particular preference here ? But > yes, sparse_vmap has been very much specific to vmemmap for these functions > which are now very generic in nature. free_mapped would do. > >> +static void unmap_hotplug_pte_range(pmd_t *pmdp, unsigned long addr, > >> + unsigned long end, bool sparse_vmap) > >> +{ > >> + struct page *page; > >> + pte_t *ptep, pte; > >> + > >> + do { > >> + ptep = pte_offset_kernel(pmdp, addr); > >> + pte = READ_ONCE(*ptep); > >> + if (pte_none(pte)) > >> + continue; > >> + > >> + WARN_ON(!pte_present(pte)); > >> + page = sparse_vmap ? pte_page(pte) : NULL; > >> + pte_clear(&init_mm, addr, ptep); > >> + flush_tlb_kernel_range(addr, addr + PAGE_SIZE); > >> + if (sparse_vmap) > >> + free_hotplug_page_range(page, PAGE_SIZE); > > > > You could only set 'page' if sparse_vmap (or even drop 'page' entirely). > > I am afraid 'page' is being used to hold pte_page(pte) extraction which > needs to be freed (sparse_vmap) as we are going to clear the ptep entry > in the next statement and lose access to it for good. You clear *ptep, not pte. > We will need some > where to hold onto pte_page(pte) across pte_clear() as we cannot free it > before clearing it's entry and flushing the TLB. Hence wondering how the > 'page' can be completely dropped. > > > The compiler is probably smart enough to optimise it but using a > > pointless ternary operator just makes the code harder to follow. > > Not sure I got this but are you suggesting for an 'if' statement here > > if (sparse_vmap) > page = pte_page(pte); > > instead of the current assignment ? > > page = sparse_vmap ? pte_page(pte) : NULL; I suggest: if (sparse_vmap) free_hotplug_pgtable_page(pte_page(pte), PAGE_SIZE); > >> + } while (addr += PAGE_SIZE, addr < end); > >> +} > > [...] > >> +static void free_empty_pte_table(pmd_t *pmdp, unsigned long addr, > >> + unsigned long end) > >> +{ > >> + pte_t *ptep, pte; > >> + > >> + do { > >> + ptep = pte_offset_kernel(pmdp, addr); > >> + pte = READ_ONCE(*ptep); > >> + WARN_ON(!pte_none(pte)); > >> + } while (addr += PAGE_SIZE, addr < end); > >> +} > >> + > >> +static void free_empty_pmd_table(pud_t *pudp, unsigned long addr, > >> + unsigned long end, unsigned long floor, > >> + unsigned long ceiling) > >> +{ > >> + unsigned long next; > >> + pmd_t *pmdp, pmd; > >> + > >> + do { > >> + next = pmd_addr_end(addr, end); > >> + pmdp = pmd_offset(pudp, addr); > >> + pmd = READ_ONCE(*pmdp); > >> + if (pmd_none(pmd)) > >> + continue; > >> + > >> + WARN_ON(!pmd_present(pmd) || !pmd_table(pmd) || pmd_sect(pmd)); > >> + free_empty_pte_table(pmdp, addr, next); > >> + free_pte_table(pmdp, addr, next, floor, ceiling); > > > > Do we need two closely named functions here? Can you not collapse > > free_empty_pud_table() and free_pte_table() into a single one? The same > > comment for the pmd/pud variants. I just find this confusing. > > The two functions could be collapsed into a single one. But just wanted to > keep free_pxx_table() part which checks floor/ceiling alignment, non-zero > entries clear off the actual page table walking. With the pmd variant, they both take the floor/ceiling argument while the free_empty_pte_table() doesn't even free anything. So not entirely consistent. Can you not just copy the free_pgd_range() functions but instead of p*d_free_tlb() just do the TLB invalidation followed by page freeing? That seems to be an easier pattern to follow.
On 10/08/2019 04:25 PM, Catalin Marinas wrote: > On Tue, Oct 08, 2019 at 10:06:26AM +0530, Anshuman Khandual wrote: >> On 10/07/2019 07:47 PM, Catalin Marinas wrote: >>> On Mon, Sep 23, 2019 at 11:13:45AM +0530, Anshuman Khandual wrote: >>>> The arch code for hot-remove must tear down portions of the linear map and >>>> vmemmap corresponding to memory being removed. In both cases the page >>>> tables mapping these regions must be freed, and when sparse vmemmap is in >>>> use the memory backing the vmemmap must also be freed. >>>> >>>> This patch adds unmap_hotplug_range() and free_empty_tables() helpers which >>>> can be used to tear down either region and calls it from vmemmap_free() and >>>> ___remove_pgd_mapping(). The sparse_vmap argument determines whether the >>>> backing memory will be freed. >>> >>> Can you change the 'sparse_vmap' name to something more meaningful which >>> would suggest freeing of the backing memory? >> >> free_mapped_mem or free_backed_mem ? Even shorter forms like free_mapped or >> free_backed might do as well. Do you have a particular preference here ? But >> yes, sparse_vmap has been very much specific to vmemmap for these functions >> which are now very generic in nature. > > free_mapped would do. Sure. > >>>> +static void unmap_hotplug_pte_range(pmd_t *pmdp, unsigned long addr, >>>> + unsigned long end, bool sparse_vmap) >>>> +{ >>>> + struct page *page; >>>> + pte_t *ptep, pte; >>>> + >>>> + do { >>>> + ptep = pte_offset_kernel(pmdp, addr); >>>> + pte = READ_ONCE(*ptep); >>>> + if (pte_none(pte)) >>>> + continue; >>>> + >>>> + WARN_ON(!pte_present(pte)); >>>> + page = sparse_vmap ? pte_page(pte) : NULL; >>>> + pte_clear(&init_mm, addr, ptep); >>>> + flush_tlb_kernel_range(addr, addr + PAGE_SIZE); >>>> + if (sparse_vmap) >>>> + free_hotplug_page_range(page, PAGE_SIZE); >>> >>> You could only set 'page' if sparse_vmap (or even drop 'page' entirely). >> >> I am afraid 'page' is being used to hold pte_page(pte) extraction which >> needs to be freed (sparse_vmap) as we are going to clear the ptep entry >> in the next statement and lose access to it for good. > > You clear *ptep, not pte. Ahh, missed that. We have already captured the contents with READ_ONCE(). > >> We will need some >> where to hold onto pte_page(pte) across pte_clear() as we cannot free it >> before clearing it's entry and flushing the TLB. Hence wondering how the >> 'page' can be completely dropped. >> >>> The compiler is probably smart enough to optimise it but using a >>> pointless ternary operator just makes the code harder to follow. >> >> Not sure I got this but are you suggesting for an 'if' statement here >> >> if (sparse_vmap) >> page = pte_page(pte); >> >> instead of the current assignment ? >> >> page = sparse_vmap ? pte_page(pte) : NULL; > > I suggest: > > if (sparse_vmap) > free_hotplug_pgtable_page(pte_page(pte), PAGE_SIZE); Sure, will do. > >>>> + } while (addr += PAGE_SIZE, addr < end); >>>> +} >>> [...] >>>> +static void free_empty_pte_table(pmd_t *pmdp, unsigned long addr, >>>> + unsigned long end) >>>> +{ >>>> + pte_t *ptep, pte; >>>> + >>>> + do { >>>> + ptep = pte_offset_kernel(pmdp, addr); >>>> + pte = READ_ONCE(*ptep); >>>> + WARN_ON(!pte_none(pte)); >>>> + } while (addr += PAGE_SIZE, addr < end); >>>> +} >>>> + >>>> +static void free_empty_pmd_table(pud_t *pudp, unsigned long addr, >>>> + unsigned long end, unsigned long floor, >>>> + unsigned long ceiling) >>>> +{ >>>> + unsigned long next; >>>> + pmd_t *pmdp, pmd; >>>> + >>>> + do { >>>> + next = pmd_addr_end(addr, end); >>>> + pmdp = pmd_offset(pudp, addr); >>>> + pmd = READ_ONCE(*pmdp); >>>> + if (pmd_none(pmd)) >>>> + continue; >>>> + >>>> + WARN_ON(!pmd_present(pmd) || !pmd_table(pmd) || pmd_sect(pmd)); >>>> + free_empty_pte_table(pmdp, addr, next); >>>> + free_pte_table(pmdp, addr, next, floor, ceiling); >>> >>> Do we need two closely named functions here? Can you not collapse >>> free_empty_pud_table() and free_pte_table() into a single one? The same >>> comment for the pmd/pud variants. I just find this confusing. >> >> The two functions could be collapsed into a single one. But just wanted to >> keep free_pxx_table() part which checks floor/ceiling alignment, non-zero >> entries clear off the actual page table walking. > > With the pmd variant, they both take the floor/ceiling argument while > the free_empty_pte_table() doesn't even free anything. So not entirely > consistent.> > Can you not just copy the free_pgd_range() functions but instead of > p*d_free_tlb() just do the TLB invalidation followed by page freeing? > That seems to be an easier pattern to follow. > Sure, will follow that pattern.
diff --git a/arch/arm64/Kconfig b/arch/arm64/Kconfig index 41a9b42..511249f 100644 --- a/arch/arm64/Kconfig +++ b/arch/arm64/Kconfig @@ -274,6 +274,9 @@ config ZONE_DMA32 config ARCH_ENABLE_MEMORY_HOTPLUG def_bool y +config ARCH_ENABLE_MEMORY_HOTREMOVE + def_bool y + config SMP def_bool y diff --git a/arch/arm64/include/asm/memory.h b/arch/arm64/include/asm/memory.h index b61b50b..615dcd0 100644 --- a/arch/arm64/include/asm/memory.h +++ b/arch/arm64/include/asm/memory.h @@ -54,6 +54,7 @@ #define MODULES_VADDR (BPF_JIT_REGION_END) #define MODULES_VSIZE (SZ_128M) #define VMEMMAP_START (-VMEMMAP_SIZE - SZ_2M) +#define VMEMMAP_END (VMEMMAP_START + VMEMMAP_SIZE) #define PCI_IO_END (VMEMMAP_START - SZ_2M) #define PCI_IO_START (PCI_IO_END - PCI_IO_SIZE) #define FIXADDR_TOP (PCI_IO_START - SZ_2M) diff --git a/arch/arm64/mm/mmu.c b/arch/arm64/mm/mmu.c index 60c929f..6178837 100644 --- a/arch/arm64/mm/mmu.c +++ b/arch/arm64/mm/mmu.c @@ -725,6 +725,277 @@ int kern_addr_valid(unsigned long addr) return pfn_valid(pte_pfn(pte)); } + +#ifdef CONFIG_MEMORY_HOTPLUG +static void free_hotplug_page_range(struct page *page, size_t size) +{ + WARN_ON(!page || PageReserved(page)); + free_pages((unsigned long)page_address(page), get_order(size)); +} + +static void free_hotplug_pgtable_page(struct page *page) +{ + free_hotplug_page_range(page, PAGE_SIZE); +} + +static bool pgtable_range_aligned(unsigned long start, unsigned long end, + unsigned long floor, unsigned long ceiling, + unsigned long mask) +{ + start &= mask; + if (start < floor) + return false; + + if (ceiling) { + ceiling &= mask; + if (!ceiling) + return false; + } + + if (end - 1 > ceiling - 1) + return false; + return true; +} + +static void free_pte_table(pmd_t *pmdp, unsigned long addr, unsigned long end, + unsigned long floor, unsigned long ceiling) +{ + struct page *page; + pte_t *ptep; + int i; + + if (!pgtable_range_aligned(addr, end, floor, ceiling, PMD_MASK)) + return; + + ptep = pte_offset_kernel(pmdp, 0UL); + for (i = 0; i < PTRS_PER_PTE; i++) { + if (!pte_none(READ_ONCE(ptep[i]))) + return; + } + + page = pmd_page(READ_ONCE(*pmdp)); + pmd_clear(pmdp); + __flush_tlb_kernel_pgtable(addr); + free_hotplug_pgtable_page(page); +} + +static void free_pmd_table(pud_t *pudp, unsigned long addr, unsigned long end, + unsigned long floor, unsigned long ceiling) +{ + struct page *page; + pmd_t *pmdp; + int i; + + if (CONFIG_PGTABLE_LEVELS <= 2) + return; + + if (!pgtable_range_aligned(addr, end, floor, ceiling, PUD_MASK)) + return; + + pmdp = pmd_offset(pudp, 0UL); + for (i = 0; i < PTRS_PER_PMD; i++) { + if (!pmd_none(READ_ONCE(pmdp[i]))) + return; + } + + page = pud_page(READ_ONCE(*pudp)); + pud_clear(pudp); + __flush_tlb_kernel_pgtable(addr); + free_hotplug_pgtable_page(page); +} + +static void free_pud_table(pgd_t *pgdp, unsigned long addr, unsigned long end, + unsigned long floor, unsigned long ceiling) +{ + struct page *page; + pud_t *pudp; + int i; + + if (CONFIG_PGTABLE_LEVELS <= 3) + return; + + if (!pgtable_range_aligned(addr, end, floor, ceiling, PGDIR_MASK)) + return; + + pudp = pud_offset(pgdp, 0UL); + for (i = 0; i < PTRS_PER_PUD; i++) { + if (!pud_none(READ_ONCE(pudp[i]))) + return; + } + + page = pgd_page(READ_ONCE(*pgdp)); + pgd_clear(pgdp); + __flush_tlb_kernel_pgtable(addr); + free_hotplug_pgtable_page(page); +} + +static void unmap_hotplug_pte_range(pmd_t *pmdp, unsigned long addr, + unsigned long end, bool sparse_vmap) +{ + struct page *page; + pte_t *ptep, pte; + + do { + ptep = pte_offset_kernel(pmdp, addr); + pte = READ_ONCE(*ptep); + if (pte_none(pte)) + continue; + + WARN_ON(!pte_present(pte)); + page = sparse_vmap ? pte_page(pte) : NULL; + pte_clear(&init_mm, addr, ptep); + flush_tlb_kernel_range(addr, addr + PAGE_SIZE); + if (sparse_vmap) + free_hotplug_page_range(page, PAGE_SIZE); + } while (addr += PAGE_SIZE, addr < end); +} + +static void unmap_hotplug_pmd_range(pud_t *pudp, unsigned long addr, + unsigned long end, bool sparse_vmap) +{ + unsigned long next; + struct page *page; + pmd_t *pmdp, pmd; + + do { + next = pmd_addr_end(addr, end); + pmdp = pmd_offset(pudp, addr); + pmd = READ_ONCE(*pmdp); + if (pmd_none(pmd)) + continue; + + WARN_ON(!pmd_present(pmd)); + if (pmd_sect(pmd)) { + page = sparse_vmap ? pmd_page(pmd) : NULL; + pmd_clear(pmdp); + flush_tlb_kernel_range(addr, next); + if (sparse_vmap) + free_hotplug_page_range(page, PMD_SIZE); + continue; + } + WARN_ON(!pmd_table(pmd)); + unmap_hotplug_pte_range(pmdp, addr, next, sparse_vmap); + } while (addr = next, addr < end); +} + +static void unmap_hotplug_pud_range(pgd_t *pgdp, unsigned long addr, + unsigned long end, bool sparse_vmap) +{ + unsigned long next; + struct page *page; + pud_t *pudp, pud; + + do { + next = pud_addr_end(addr, end); + pudp = pud_offset(pgdp, addr); + pud = READ_ONCE(*pudp); + if (pud_none(pud)) + continue; + + WARN_ON(!pud_present(pud)); + if (pud_sect(pud)) { + page = sparse_vmap ? pud_page(pud) : NULL; + pud_clear(pudp); + flush_tlb_kernel_range(addr, next); + if (sparse_vmap) + free_hotplug_page_range(page, PUD_SIZE); + continue; + } + WARN_ON(!pud_table(pud)); + unmap_hotplug_pmd_range(pudp, addr, next, sparse_vmap); + } while (addr = next, addr < end); +} + +static void unmap_hotplug_range(unsigned long addr, unsigned long end, + bool sparse_vmap) +{ + unsigned long next; + pgd_t *pgdp, pgd; + + do { + next = pgd_addr_end(addr, end); + pgdp = pgd_offset_k(addr); + pgd = READ_ONCE(*pgdp); + if (pgd_none(pgd)) + continue; + + WARN_ON(!pgd_present(pgd)); + unmap_hotplug_pud_range(pgdp, addr, next, sparse_vmap); + } while (addr = next, addr < end); +} + +static void free_empty_pte_table(pmd_t *pmdp, unsigned long addr, + unsigned long end) +{ + pte_t *ptep, pte; + + do { + ptep = pte_offset_kernel(pmdp, addr); + pte = READ_ONCE(*ptep); + WARN_ON(!pte_none(pte)); + } while (addr += PAGE_SIZE, addr < end); +} + +static void free_empty_pmd_table(pud_t *pudp, unsigned long addr, + unsigned long end, unsigned long floor, + unsigned long ceiling) +{ + unsigned long next; + pmd_t *pmdp, pmd; + + do { + next = pmd_addr_end(addr, end); + pmdp = pmd_offset(pudp, addr); + pmd = READ_ONCE(*pmdp); + if (pmd_none(pmd)) + continue; + + WARN_ON(!pmd_present(pmd) || !pmd_table(pmd) || pmd_sect(pmd)); + free_empty_pte_table(pmdp, addr, next); + free_pte_table(pmdp, addr, next, floor, ceiling); + } while (addr = next, addr < end); +} + +static void free_empty_pud_table(pgd_t *pgdp, unsigned long addr, + unsigned long end, unsigned long floor, + unsigned long ceiling) +{ + unsigned long next; + pud_t *pudp, pud; + + do { + next = pud_addr_end(addr, end); + pudp = pud_offset(pgdp, addr); + pud = READ_ONCE(*pudp); + if (pud_none(pud)) + continue; + + WARN_ON(!pud_present(pud) || !pud_table(pud) || pud_sect(pud)); + free_empty_pmd_table(pudp, addr, next, floor, ceiling); + free_pmd_table(pudp, addr, next, floor, ceiling); + } while (addr = next, addr < end); +} + +static void free_empty_tables(unsigned long addr, unsigned long end, + unsigned long floor, unsigned long ceiling) +{ + unsigned long next; + pgd_t *pgdp, pgd; + + do { + next = pgd_addr_end(addr, end); + pgdp = pgd_offset_k(addr); + pgd = READ_ONCE(*pgdp); + if (pgd_none(pgd)) + continue; + + WARN_ON(!pgd_present(pgd)); + free_empty_pud_table(pgdp, addr, next, floor, ceiling); + free_pud_table(pgdp, addr, next, floor, ceiling); + } while (addr = next, addr < end); +} +#endif + #ifdef CONFIG_SPARSEMEM_VMEMMAP #if !ARM64_SWAPPER_USES_SECTION_MAPS int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node, @@ -772,6 +1043,12 @@ int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node, void vmemmap_free(unsigned long start, unsigned long end, struct vmem_altmap *altmap) { +#ifdef CONFIG_MEMORY_HOTPLUG + WARN_ON((start < VMEMMAP_START) || (end > VMEMMAP_END)); + + unmap_hotplug_range(start, end, true); + free_empty_tables(start, end, VMEMMAP_START, VMEMMAP_END); +#endif } #endif /* CONFIG_SPARSEMEM_VMEMMAP */ @@ -1050,10 +1327,21 @@ int p4d_free_pud_page(p4d_t *p4d, unsigned long addr) } #ifdef CONFIG_MEMORY_HOTPLUG +static void __remove_pgd_mapping(pgd_t *pgdir, unsigned long start, u64 size) +{ + unsigned long end = start + size; + + WARN_ON(pgdir != init_mm.pgd); + WARN_ON((start < PAGE_OFFSET) || (end > PAGE_END)); + + unmap_hotplug_range(start, end, false); + free_empty_tables(start, end, PAGE_OFFSET, PAGE_END); +} + int arch_add_memory(int nid, u64 start, u64 size, struct mhp_restrictions *restrictions) { - int flags = 0; + int ret, flags = 0; if (rodata_full || debug_pagealloc_enabled()) flags = NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS; @@ -1061,9 +1349,14 @@ int arch_add_memory(int nid, u64 start, u64 size, __create_pgd_mapping(swapper_pg_dir, start, __phys_to_virt(start), size, PAGE_KERNEL, __pgd_pgtable_alloc, flags); - return __add_pages(nid, start >> PAGE_SHIFT, size >> PAGE_SHIFT, + ret = __add_pages(nid, start >> PAGE_SHIFT, size >> PAGE_SHIFT, restrictions); + if (ret) + __remove_pgd_mapping(swapper_pg_dir, + __phys_to_virt(start), size); + return ret; } + void arch_remove_memory(int nid, u64 start, u64 size, struct vmem_altmap *altmap) { @@ -1071,14 +1364,8 @@ void arch_remove_memory(int nid, u64 start, u64 size, unsigned long nr_pages = size >> PAGE_SHIFT; struct zone *zone; - /* - * FIXME: Cleanup page tables (also in arch_add_memory() in case - * adding fails). Until then, this function should only be used - * during memory hotplug (adding memory), not for memory - * unplug. ARCH_ENABLE_MEMORY_HOTREMOVE must not be - * unlocked yet. - */ zone = page_zone(pfn_to_page(start_pfn)); __remove_pages(zone, start_pfn, nr_pages, altmap); + __remove_pgd_mapping(swapper_pg_dir, __phys_to_virt(start), size); } #endif