| Message ID | 20211005202529.213812-1-mike.kravetz@oracle.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | arm64/hugetlb: fix CMA gigantic page order for non-4K PAGE_SIZE | expand |
On Tue, 5 Oct 2021 13:25:29 -0700 Mike Kravetz <mike.kravetz@oracle.com> wrote: > For non-4K PAGE_SIZE configs, the largest gigantic huge page size is > CONT_PMD_SHIFT order. What are the user visible effects of this bug?
On 10/5/21 1:54 PM, Andrew Morton wrote: > On Tue, 5 Oct 2021 13:25:29 -0700 Mike Kravetz <mike.kravetz@oracle.com> wrote: > >> For non-4K PAGE_SIZE configs, the largest gigantic huge page size is >> CONT_PMD_SHIFT order. > > What are the user visible effects of this bug? > > Sorry, I only recently got easy access to arm64 platforms. This is what I saw as a user: The largest gigantic huge page size on arm64 with 64K PAGE_SIZE 64K is 16G. Therefore, one should be able to specify 'hugetlb_cma=16G' on the kernel command line so that 1 gigantic page can be allocated from CMA. However, when adding such an option the following message is produced: hugetlb_cma: cma area should be at least 8796093022208 MiB This is because the calculation for non-4K gigantic page order is incorrect in the arm64 specific routine arm64_hugetlb_cma_reserve(). Would you like me to send a new version with this in the commit message? Or, is it easier for you to just add it?
On Tue, 5 Oct 2021 14:28:03 -0700 Mike Kravetz <mike.kravetz@oracle.com> wrote: > On 10/5/21 1:54 PM, Andrew Morton wrote: > > On Tue, 5 Oct 2021 13:25:29 -0700 Mike Kravetz <mike.kravetz@oracle.com> wrote: > > > >> For non-4K PAGE_SIZE configs, the largest gigantic huge page size is > >> CONT_PMD_SHIFT order. > > > > What are the user visible effects of this bug? > > > > > > Sorry, > I only recently got easy access to arm64 platforms. This is what I saw > as a user: > > The largest gigantic huge page size on arm64 with 64K PAGE_SIZE 64K is > 16G. Therefore, one should be able to specify 'hugetlb_cma=16G' on the > kernel command line so that 1 gigantic page can be allocated from CMA. > However, when adding such an option the following message is produced: > > hugetlb_cma: cma area should be at least 8796093022208 MiB > > This is because the calculation for non-4K gigantic page order is > incorrect in the arm64 specific routine arm64_hugetlb_cma_reserve(). Cool, thanks. > Would you like me to send a new version with this in the commit message? > Or, is it easier for you to just add it? I assumed that it would be merged via the same path as the offending abb7962adc80. Catalin's arm tree, it appears.
On 10/6/21 1:55 AM, Mike Kravetz wrote: > For non-4K PAGE_SIZE configs, the largest gigantic huge page size is > CONT_PMD_SHIFT order. > > Fixes: abb7962adc80 ("arm64/hugetlb: Reserve CMA areas for gigantic > pages on 16K and 64K configs") > Signed-off-by: Mike Kravetz <mike.kravetz@oracle.com> > Cc: <stable@vger.kernel.org> Reviewed-by: Anshuman Khandual <anshuman.khandual@arm.com> > --- > arch/arm64/mm/hugetlbpage.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/arch/arm64/mm/hugetlbpage.c b/arch/arm64/mm/hugetlbpage.c > index 23505fc35324..a8158c948966 100644 > --- a/arch/arm64/mm/hugetlbpage.c > +++ b/arch/arm64/mm/hugetlbpage.c > @@ -43,7 +43,7 @@ void __init arm64_hugetlb_cma_reserve(void) > #ifdef CONFIG_ARM64_4K_PAGES > order = PUD_SHIFT - PAGE_SHIFT; > #else > - order = CONT_PMD_SHIFT + PMD_SHIFT - PAGE_SHIFT; > + order = CONT_PMD_SHIFT - PAGE_SHIFT; > #endif > /* > * HugeTLB CMA reservation is required for gigantic > The commit a1634a542f74 ("arm64/mm: Redefine CONT_{PTE, PMD}_SHIFT") which got merged during the exact same week, broke the above commit as both were in flight. The commit here updated hugetlbpage_init() but did not update the new incoming arm64_hugetlb_cma_reserve().
On Tue, 5 Oct 2021 13:25:29 -0700, Mike Kravetz wrote: > For non-4K PAGE_SIZE configs, the largest gigantic huge page size is > CONT_PMD_SHIFT order. Applied to arm64 (for-next/fixes), thanks! [1/1] arm64/hugetlb: fix CMA gigantic page order for non-4K PAGE_SIZE https://git.kernel.org/arm64/c/0350419b14b9
diff --git a/arch/arm64/mm/hugetlbpage.c b/arch/arm64/mm/hugetlbpage.c index 23505fc35324..a8158c948966 100644 --- a/arch/arm64/mm/hugetlbpage.c +++ b/arch/arm64/mm/hugetlbpage.c @@ -43,7 +43,7 @@ void __init arm64_hugetlb_cma_reserve(void) #ifdef CONFIG_ARM64_4K_PAGES order = PUD_SHIFT - PAGE_SHIFT; #else - order = CONT_PMD_SHIFT + PMD_SHIFT - PAGE_SHIFT; + order = CONT_PMD_SHIFT - PAGE_SHIFT; #endif /* * HugeTLB CMA reservation is required for gigantic
For non-4K PAGE_SIZE configs, the largest gigantic huge page size is CONT_PMD_SHIFT order. Fixes: abb7962adc80 ("arm64/hugetlb: Reserve CMA areas for gigantic pages on 16K and 64K configs") Signed-off-by: Mike Kravetz <mike.kravetz@oracle.com> Cc: <stable@vger.kernel.org> --- arch/arm64/mm/hugetlbpage.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)