| Message ID | 1415937567-14510-1-git-send-email-yinghai@kernel.org (mailing list archive) |
|---|---|
| State | New, archived |
| Delegated to: | Bjorn Helgaas |
| Headers | show |
On Thu, Nov 13, 2014 at 07:59:27PM -0800, Yinghai Lu wrote: > Aaron reported 32bit/PAE mode, has problem with 64bit resource. > > [ 6.610012] pci 0000:03:00.0: reg 0x10: [mem 0x383fffc00000-0x383fffdfffff 64bit pref] > [ 6.622195] pci 0000:03:00.0: reg 0x20: [mem 0x383fffe04000-0x383fffe07fff 64bit pref] > [ 6.656112] pci 0000:03:00.1: reg 0x10: [mem 0x383fffa00000-0x383fffbfffff 64bit pref] > [ 6.668293] pci 0000:03:00.1: reg 0x20: [mem 0x383fffe00000-0x383fffe03fff 64bit pref] > ... > [ 12.374143] calling ixgbe_init_module+0x0/0x51 @ 1 > [ 12.378130] ixgbe: Intel(R) 10 Gigabit PCI Express Network Driver - version 3.19.1-k > [ 12.385318] ixgbe: Copyright (c) 1999-2014 Intel Corporation. > [ 12.390578] ixgbe 0000:03:00.0: Adapter removed > [ 12.394247] ixgbe: probe of 0000:03:00.0 failed with error -5 > [ 12.399369] ixgbe 0000:03:00.1: Adapter removed > [ 12.403036] ixgbe: probe of 0000:03:00.1 failed with error -5 > [ 12.408017] initcall ixgbe_init_module+0x0/0x51 returned 0 after 29200 usecs > > root cause is ioremap can not handle mmio range that is more than 44bits on > 32bit PAE mode. > > We are using pfn with unsigned long like pfn_pte(), so those 0x383fffc00000 will > overflow in pfn format with unsigned long (that is 32bits in 32bit x86 kernel, > and pfn only can support 44bits). > > | static inline pte_t pfn_pte(unsigned long page_nr, pgprot_t pgprot) > | { > | return __pte(((phys_addr_t)page_nr << PAGE_SHIFT) | > | massage_pgprot(pgprot)); > | } > > We could limit iomem to 44 bits so we can reject them early from root bus. > but xhci is not happy with resource allocation (hang ?) > > Change phys_addr_t for pfn_pte, and add overflow check to skip ram checking, > as the mmio is too big to be ram. > At last, can not use PHYSICAL_PAGE_MASK to get aligned phys_addr. If I understand correctly, the problem is that we have a 46-bit physical address (0x383fffc00000) and with 4KB pages, we're trying to put a 36-bit PFN into a 32-bit unsigned long, which obviously doesn't fit. This patch fixes a few spots, but aren't there many more places with the same problem? I guess this is a VM issue anyway, not a PCI problem, so I'll defer to the VM folks and let them handle this. Bjorn > Link: https://bugzilla.kernel.org/show_bug.cgi?id=88131 > Reported-by: Aaron Ma <mapengyu@gmail.com> > Tested-by: Aaron Ma <mapengyu@gmail.com> > Signed-off-by: Yinghai Lu <yinghai@kernel.org> > > --- > arch/x86/include/asm/page.h | 8 ++++++++ > arch/x86/include/asm/pgtable.h | 4 ++-- > arch/x86/mm/ioremap.c | 6 ++++-- > arch/x86/mm/pat.c | 3 +++ > 4 files changed, 17 insertions(+), 4 deletions(-) > > Index: linux-2.6/arch/x86/include/asm/page.h > =================================================================== > --- linux-2.6.orig/arch/x86/include/asm/page.h > +++ linux-2.6/arch/x86/include/asm/page.h > @@ -15,6 +15,14 @@ > > #ifndef __ASSEMBLY__ > > +static inline int pfn_overflow(dma_addr_t phy_addr) > +{ > + dma_addr_t real_pfn = phy_addr >> PAGE_SHIFT; > + unsigned long pfn = (unsigned long)real_pfn; > + > + return pfn != real_pfn; > +} > + > struct page; > > #include <linux/range.h> > Index: linux-2.6/arch/x86/include/asm/pgtable.h > =================================================================== > --- linux-2.6.orig/arch/x86/include/asm/pgtable.h > +++ linux-2.6/arch/x86/include/asm/pgtable.h > @@ -354,9 +354,9 @@ static inline pgprotval_t massage_pgprot > return protval; > } > > -static inline pte_t pfn_pte(unsigned long page_nr, pgprot_t pgprot) > +static inline pte_t pfn_pte(phys_addr_t page_nr, pgprot_t pgprot) > { > - return __pte(((phys_addr_t)page_nr << PAGE_SHIFT) | > + return __pte((page_nr << PAGE_SHIFT) | > massage_pgprot(pgprot)); > } > > Index: linux-2.6/arch/x86/mm/ioremap.c > =================================================================== > --- linux-2.6.orig/arch/x86/mm/ioremap.c > +++ linux-2.6/arch/x86/mm/ioremap.c > @@ -122,7 +122,9 @@ static void __iomem *__ioremap_caller(re > if (ram_region < 0) { > pfn = phys_addr >> PAGE_SHIFT; > last_pfn = last_addr >> PAGE_SHIFT; > - if (walk_system_ram_range(pfn, last_pfn - pfn + 1, NULL, > + /* pfn overflow, don't need to check */ > + if (!pfn_overflow(last_addr) && > + walk_system_ram_range(pfn, last_pfn - pfn + 1, NULL, > __ioremap_check_ram) == 1) > return NULL; > } > @@ -130,7 +132,7 @@ static void __iomem *__ioremap_caller(re > * Mappings have to be page-aligned > */ > offset = phys_addr & ~PAGE_MASK; > - phys_addr &= PHYSICAL_PAGE_MASK; > + phys_addr -= offset; > size = PAGE_ALIGN(last_addr+1) - phys_addr; > > retval = reserve_memtype(phys_addr, (u64)phys_addr + size, > Index: linux-2.6/arch/x86/mm/pat.c > =================================================================== > --- linux-2.6.orig/arch/x86/mm/pat.c > +++ linux-2.6/arch/x86/mm/pat.c > @@ -183,6 +183,9 @@ static int pat_pagerange_is_ram(resource > unsigned long end_pfn = (end + PAGE_SIZE - 1) >> PAGE_SHIFT; > struct pagerange_state state = {start_pfn, 0, 0}; > > + /* pfn overflow, don't need to check */ > + if (pfn_overflow(end + PAGE_SIZE - 1)) > + return 0; > /* > * For legacy reasons, physical address range in the legacy ISA > * region is tracked as non-RAM. This will allow users of -- To unsubscribe from this list: send the line "unsubscribe linux-pci" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Index: linux-2.6/arch/x86/include/asm/page.h =================================================================== --- linux-2.6.orig/arch/x86/include/asm/page.h +++ linux-2.6/arch/x86/include/asm/page.h @@ -15,6 +15,14 @@ #ifndef __ASSEMBLY__ +static inline int pfn_overflow(dma_addr_t phy_addr) +{ + dma_addr_t real_pfn = phy_addr >> PAGE_SHIFT; + unsigned long pfn = (unsigned long)real_pfn; + + return pfn != real_pfn; +} + struct page; #include <linux/range.h> Index: linux-2.6/arch/x86/include/asm/pgtable.h =================================================================== --- linux-2.6.orig/arch/x86/include/asm/pgtable.h +++ linux-2.6/arch/x86/include/asm/pgtable.h @@ -354,9 +354,9 @@ static inline pgprotval_t massage_pgprot return protval; } -static inline pte_t pfn_pte(unsigned long page_nr, pgprot_t pgprot) +static inline pte_t pfn_pte(phys_addr_t page_nr, pgprot_t pgprot) { - return __pte(((phys_addr_t)page_nr << PAGE_SHIFT) | + return __pte((page_nr << PAGE_SHIFT) | massage_pgprot(pgprot)); } Index: linux-2.6/arch/x86/mm/ioremap.c =================================================================== --- linux-2.6.orig/arch/x86/mm/ioremap.c +++ linux-2.6/arch/x86/mm/ioremap.c @@ -122,7 +122,9 @@ static void __iomem *__ioremap_caller(re if (ram_region < 0) { pfn = phys_addr >> PAGE_SHIFT; last_pfn = last_addr >> PAGE_SHIFT; - if (walk_system_ram_range(pfn, last_pfn - pfn + 1, NULL, + /* pfn overflow, don't need to check */ + if (!pfn_overflow(last_addr) && + walk_system_ram_range(pfn, last_pfn - pfn + 1, NULL, __ioremap_check_ram) == 1) return NULL; } @@ -130,7 +132,7 @@ static void __iomem *__ioremap_caller(re * Mappings have to be page-aligned */ offset = phys_addr & ~PAGE_MASK; - phys_addr &= PHYSICAL_PAGE_MASK; + phys_addr -= offset; size = PAGE_ALIGN(last_addr+1) - phys_addr; retval = reserve_memtype(phys_addr, (u64)phys_addr + size, Index: linux-2.6/arch/x86/mm/pat.c =================================================================== --- linux-2.6.orig/arch/x86/mm/pat.c +++ linux-2.6/arch/x86/mm/pat.c @@ -183,6 +183,9 @@ static int pat_pagerange_is_ram(resource unsigned long end_pfn = (end + PAGE_SIZE - 1) >> PAGE_SHIFT; struct pagerange_state state = {start_pfn, 0, 0}; + /* pfn overflow, don't need to check */ + if (pfn_overflow(end + PAGE_SIZE - 1)) + return 0; /* * For legacy reasons, physical address range in the legacy ISA * region is tracked as non-RAM. This will allow users of
Aaron reported 32bit/PAE mode, has problem with 64bit resource. [ 6.610012] pci 0000:03:00.0: reg 0x10: [mem 0x383fffc00000-0x383fffdfffff 64bit pref] [ 6.622195] pci 0000:03:00.0: reg 0x20: [mem 0x383fffe04000-0x383fffe07fff 64bit pref] [ 6.656112] pci 0000:03:00.1: reg 0x10: [mem 0x383fffa00000-0x383fffbfffff 64bit pref] [ 6.668293] pci 0000:03:00.1: reg 0x20: [mem 0x383fffe00000-0x383fffe03fff 64bit pref] ... [ 12.374143] calling ixgbe_init_module+0x0/0x51 @ 1 [ 12.378130] ixgbe: Intel(R) 10 Gigabit PCI Express Network Driver - version 3.19.1-k [ 12.385318] ixgbe: Copyright (c) 1999-2014 Intel Corporation. [ 12.390578] ixgbe 0000:03:00.0: Adapter removed [ 12.394247] ixgbe: probe of 0000:03:00.0 failed with error -5 [ 12.399369] ixgbe 0000:03:00.1: Adapter removed [ 12.403036] ixgbe: probe of 0000:03:00.1 failed with error -5 [ 12.408017] initcall ixgbe_init_module+0x0/0x51 returned 0 after 29200 usecs root cause is ioremap can not handle mmio range that is more than 44bits on 32bit PAE mode. We are using pfn with unsigned long like pfn_pte(), so those 0x383fffc00000 will overflow in pfn format with unsigned long (that is 32bits in 32bit x86 kernel, and pfn only can support 44bits). | static inline pte_t pfn_pte(unsigned long page_nr, pgprot_t pgprot) | { | return __pte(((phys_addr_t)page_nr << PAGE_SHIFT) | | massage_pgprot(pgprot)); | } We could limit iomem to 44 bits so we can reject them early from root bus. but xhci is not happy with resource allocation (hang ?) Change phys_addr_t for pfn_pte, and add overflow check to skip ram checking, as the mmio is too big to be ram. At last, can not use PHYSICAL_PAGE_MASK to get aligned phys_addr. Link: https://bugzilla.kernel.org/show_bug.cgi?id=88131 Reported-by: Aaron Ma <mapengyu@gmail.com> Tested-by: Aaron Ma <mapengyu@gmail.com> Signed-off-by: Yinghai Lu <yinghai@kernel.org> --- arch/x86/include/asm/page.h | 8 ++++++++ arch/x86/include/asm/pgtable.h | 4 ++-- arch/x86/mm/ioremap.c | 6 ++++-- arch/x86/mm/pat.c | 3 +++ 4 files changed, 17 insertions(+), 4 deletions(-) -- To unsubscribe from this list: send the line "unsubscribe linux-pci" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html