Message ID | alpine.LRH.2.02.1503181057340.14516@file01.intranet.prod.int.rdu2.redhat.com (mailing list archive) |
---|---|
State | Superseded |
Headers | show |
On Wed, Mar 18, 2015 at 11:16:42AM -0400, Mikulas Patocka wrote: > Hi > > Here I'm sending a patch that fixes numerous "BUG: non-zero nr_pmds on > freeing mm: -1" errors on 64-bit PA-RISC kernel. > > I think the patch posted here > http://www.spinics.net/lists/linux-parisc/msg05981.html is incorrect, it > wouldn't work if the affected address range is freed and allocated > multiple times. > - 1. alloc pgd with built-in pmd, the count of pmds is 1 > - 2. free the range covered by the built-in pmd, the count of pmds > is 0, but the built-in pmd is still present Hm. Okay. I didn't realize you have special case in pmd_clear() for these pmds. What about adding mm_inc_nr_pmds() in pmd_clear() for PxD_FLAG_ATTACHED to compensate mm_dec_nr_pmds() in free_pmd_range()? I don't like pmd_preallocated() in generic code. It's too specific to parisc. > - 3. alloc some memory in the range affected by the built-in pmd, > the count of pmds is still 0 > - 4. free the range covered by the built-in pmd, the counter > underflows to -1 > > Mikulas > > > From: Mikulas Patocka <mpatocka@redhat.com> > > The patch dc6c9a35b66b520cf67e05d8ca60ebecad3b0479 that counts pmds > allocated for a process introduced a bug on 64-bit PA-RISC kernels. There > are many "BUG: non-zero nr_pmds on freeing mm: -1" messages. > > The PA-RISC architecture preallocates one pmd with each pgd. This > preallocated pmd can never be freed - pmd_free does nothing when it is > called with this pmd. When the kernel attempts to free this preallocated > pmd, it decreases the count of allocated pmds. The result is that the > counter underflows and this error is reported. > > This patch fixes the bug by introducing a macro pmd_preallocated and > making sure that the counter is not decremented when this preallocated pmd > is freed. > > Signed-off-by: Mikulas Patocka <mpatocka@redhat.com> > > --- > arch/parisc/include/asm/pgalloc.h | 2 ++ > mm/memory.c | 5 ++++- > 2 files changed, 6 insertions(+), 1 deletion(-) > > Index: linux-4.0-rc4/arch/parisc/include/asm/pgalloc.h > =================================================================== > --- linux-4.0-rc4.orig/arch/parisc/include/asm/pgalloc.h 2015-03-18 15:31:10.000000000 +0100 > +++ linux-4.0-rc4/arch/parisc/include/asm/pgalloc.h 2015-03-18 15:33:20.000000000 +0100 > @@ -81,6 +81,8 @@ static inline void pmd_free(struct mm_st > free_pages((unsigned long)pmd, PMD_ORDER); > } > > +#define pmd_preallocated(pmd) (pmd_flag(*(pmd)) & PxD_FLAG_ATTACHED) > + > #else > > /* Two Level Page Table Support for pmd's */ > Index: linux-4.0-rc4/mm/memory.c > =================================================================== > --- linux-4.0-rc4.orig/mm/memory.c 2015-03-18 15:30:42.000000000 +0100 > +++ linux-4.0-rc4/mm/memory.c 2015-03-18 15:32:33.000000000 +0100 > @@ -427,8 +427,11 @@ static inline void free_pmd_range(struct > > pmd = pmd_offset(pud, start); > pud_clear(pud); > +#ifdef pmd_preallocated > + if (!pmd_preallocated(pmd)) > +#endif > + mm_dec_nr_pmds(tlb->mm); > pmd_free_tlb(tlb, pmd, start); > - mm_dec_nr_pmds(tlb->mm); > } > > static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd, > > -- > To unsubscribe, send a message with 'unsubscribe linux-mm' in > the body to majordomo@kvack.org. For more info on Linux MM, > see: http://www.linux-mm.org/ . > Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>
On Wed, 18 Mar 2015, Kirill A. Shutemov wrote: > On Wed, Mar 18, 2015 at 11:16:42AM -0400, Mikulas Patocka wrote: > > Hi > > > > Here I'm sending a patch that fixes numerous "BUG: non-zero nr_pmds on > > freeing mm: -1" errors on 64-bit PA-RISC kernel. > > > > I think the patch posted here > > http://www.spinics.net/lists/linux-parisc/msg05981.html is incorrect, it > > wouldn't work if the affected address range is freed and allocated > > multiple times. > > - 1. alloc pgd with built-in pmd, the count of pmds is 1 > > - 2. free the range covered by the built-in pmd, the count of pmds > > is 0, but the built-in pmd is still present > > Hm. Okay. I didn't realize you have special case in pmd_clear() for these > pmds. > > What about adding mm_inc_nr_pmds() in pmd_clear() for PxD_FLAG_ATTACHED > to compensate mm_dec_nr_pmds() in free_pmd_range()? pmd_clear clears one entry in the pmd, it wouldn't work. You need to add it to pgd_clear. That clears the pointer to the pmd (and does nothing if it is asked to clear the pointer to the preallocated pmd). But pgd_clear doesn't receive the pointer to mm. > I don't like pmd_preallocated() in generic code. It's too specific to > parisc. The question is if it is better to use pmd_preallocated, or pass the pointer to the mm to pgd_clear (that would affect all architectures). Mikulas -- To unsubscribe from this list: send the line "unsubscribe linux-parisc" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Wed, Mar 18, 2015 at 12:25:11PM -0400, Mikulas Patocka wrote: > > > On Wed, 18 Mar 2015, Kirill A. Shutemov wrote: > > > On Wed, Mar 18, 2015 at 11:16:42AM -0400, Mikulas Patocka wrote: > > > Hi > > > > > > Here I'm sending a patch that fixes numerous "BUG: non-zero nr_pmds on > > > freeing mm: -1" errors on 64-bit PA-RISC kernel. > > > > > > I think the patch posted here > > > http://www.spinics.net/lists/linux-parisc/msg05981.html is incorrect, it > > > wouldn't work if the affected address range is freed and allocated > > > multiple times. > > > - 1. alloc pgd with built-in pmd, the count of pmds is 1 > > > - 2. free the range covered by the built-in pmd, the count of pmds > > > is 0, but the built-in pmd is still present > > > > Hm. Okay. I didn't realize you have special case in pmd_clear() for these > > pmds. > > > > What about adding mm_inc_nr_pmds() in pmd_clear() for PxD_FLAG_ATTACHED > > to compensate mm_dec_nr_pmds() in free_pmd_range()? > > pmd_clear clears one entry in the pmd, it wouldn't work. You need to add > it to pgd_clear. That clears the pointer to the pmd (and does nothing if > it is asked to clear the pointer to the preallocated pmd). But pgd_clear > doesn't receive the pointer to mm. I meant pmd_free(), not pmd_clear(). This should work fine. > > > I don't like pmd_preallocated() in generic code. It's too specific to > > parisc. > > The question is if it is better to use pmd_preallocated, or pass the > pointer to the mm to pgd_clear (that would affect all architectures). > > Mikulas
Index: linux-4.0-rc4/arch/parisc/include/asm/pgalloc.h =================================================================== --- linux-4.0-rc4.orig/arch/parisc/include/asm/pgalloc.h 2015-03-18 15:31:10.000000000 +0100 +++ linux-4.0-rc4/arch/parisc/include/asm/pgalloc.h 2015-03-18 15:33:20.000000000 +0100 @@ -81,6 +81,8 @@ static inline void pmd_free(struct mm_st free_pages((unsigned long)pmd, PMD_ORDER); } +#define pmd_preallocated(pmd) (pmd_flag(*(pmd)) & PxD_FLAG_ATTACHED) + #else /* Two Level Page Table Support for pmd's */ Index: linux-4.0-rc4/mm/memory.c =================================================================== --- linux-4.0-rc4.orig/mm/memory.c 2015-03-18 15:30:42.000000000 +0100 +++ linux-4.0-rc4/mm/memory.c 2015-03-18 15:32:33.000000000 +0100 @@ -427,8 +427,11 @@ static inline void free_pmd_range(struct pmd = pmd_offset(pud, start); pud_clear(pud); +#ifdef pmd_preallocated + if (!pmd_preallocated(pmd)) +#endif + mm_dec_nr_pmds(tlb->mm); pmd_free_tlb(tlb, pmd, start); - mm_dec_nr_pmds(tlb->mm); } static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd,