From mboxrd@z Thu Jan 1 00:00:00 1970 From: FUJITA Tomonori Subject: Re: [regression]2.6.25-rc4: boot panic on alpha Date: Sat, 8 Mar 2008 01:45:42 +0900 Message-ID: <20080308014543E.tomof@acm.org> References: <20080307141439.GB23178@elte.hu> <20080307235616T.tomof@acm.org> <20080307162116.GA18563@jurassic.park.msu.ru> Mime-Version: 1.0 Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <20080307162116.GA18563@jurassic.park.msu.ru> Sender: linux-alpha-owner@vger.kernel.org List-ID: Content-Type: Text/Plain; charset="us-ascii" To: ink@jurassic.park.msu.ru Cc: tomof@acm.org, rct@frus.com, mingo@elte.hu, linux-kernel@vger.kernel.org, linux-alpha@vger.kernel.org, fujita.tomonori@lab.ntt.co.jp, akpm@linux-foundation.org On Fri, 7 Mar 2008 19:21:16 +0300 Ivan Kokshaysky wrote: > On Fri, Mar 07, 2008 at 11:56:40PM +0900, FUJITA Tomonori wrote: > > Very sorry, I probably broke the IOMMU. I don't have the hardware so > > I've not tested the patches. > > Unfortunately, I was not able to test these patches either, as my > alpha with iommmu and scsi is disassembled at the moment, thanks to > dead CPU fan... > > > - BUG_ON(arena->dma_base & ~PAGE_MASK); > > + if (arena->dma_base & ~PAGE_MASK) > > + printk("%s %d: %lx\n", __FUNCTION__, __LINE__, arena->dma_base); > > arena->dma_base is always properly aligned - it's pretty fundamental thing. Thanks, I see. > > base = arena->dma_base >> PAGE_SHIFT; > > if (dev) > > boundary_size = ALIGN(dma_get_max_seg_size(dev) + 1, PAGE_SIZE) > ^^^^^^^^^^^^^^^^^^^^^^^^^ > Must be dma_get_seg_boundary(dev). Hopefully it's the only typo ;-) Ah, a stupid typo... > > @@ -152,7 +155,8 @@ iommu_arena_find_pages(struct device *dev, struct pci_iommu_arena *arena, > > else > > boundary_size = ALIGN(1UL << 32, PAGE_SIZE) >> PAGE_SHIFT; > > A bit of abuse of ALIGN() macro, to my taste ;-) > > > - BUG_ON(!is_power_of_2(boundary_size)); > > + if (!is_power_of_2(boundary_size)) > > + printk("%s %d: %lx\n", __FUNCTION__, __LINE__, boundary_size); > > Perhaps it was that, if the scsi driver set max_seg_size to something > other than default 65536 (and not a power of two)... Yeah, probably (and hopefully) it was. > Bob, can you give it a try? > > Ivan. > > diff --git a/arch/alpha/kernel/pci_iommu.c b/arch/alpha/kernel/pci_iommu.c > index be6fa10..e07a23f 100644 > --- a/arch/alpha/kernel/pci_iommu.c > +++ b/arch/alpha/kernel/pci_iommu.c > @@ -144,15 +144,14 @@ iommu_arena_find_pages(struct device *dev, struct pci_iommu_arena *arena, > unsigned long base; > unsigned long boundary_size; > > - BUG_ON(arena->dma_base & ~PAGE_MASK); > base = arena->dma_base >> PAGE_SHIFT; > - if (dev) > - boundary_size = ALIGN(dma_get_max_seg_size(dev) + 1, PAGE_SIZE) > - >> PAGE_SHIFT; > - else > - boundary_size = ALIGN(1UL << 32, PAGE_SIZE) >> PAGE_SHIFT; > - > - BUG_ON(!is_power_of_2(boundary_size)); > + if (dev) { > + boundary_size = dma_get_seg_boundary(dev) + 1; > + BUG_ON(!is_power_of_2(boundary_size)); > + boundary_size >>= PAGE_SHIFT; > + } else { > + boundary_size = 1UL << (32 - PAGE_SHIFT); > + } > > /* Search forward for the first mask-aligned sequence of N free ptes */ > ptes = arena->ptes; Thanks!