Re: [PATCH v14 06/12] swiotlb: Use is_swiotlb_force_bounce for swiotlb data bouncing

[Will Deacon <will@xxxxxxxxxx>]

On Thu, Jun 24, 2021 at 12:34:09PM +0100, Robin Murphy wrote:
> On 2021-06-24 12:18, Will Deacon wrote:
> > On Thu, Jun 24, 2021 at 12:14:39PM +0100, Robin Murphy wrote:
> > > On 2021-06-24 07:05, Claire Chang wrote:
> > > > On Thu, Jun 24, 2021 at 1:43 PM Christoph Hellwig <hch@xxxxxx> wrote:
> > > > > 
> > > > > On Wed, Jun 23, 2021 at 02:44:34PM -0400, Qian Cai wrote:
> > > > > > is_swiotlb_force_bounce at 
> > > > > > /usr/src/linux-next/./include/linux/swiotlb.h:119
> > > > > > 
> > > > > > is_swiotlb_force_bounce() was the new function introduced in this 
> > > > > > patch here.
> > > > > > 
> > > > > > +static inline bool is_swiotlb_force_bounce(struct device *dev)
> > > > > > +{
> > > > > > +     return dev->dma_io_tlb_mem->force_bounce;
> > > > > > +}
> > > > > 
> > > > > To me the crash looks like dev->dma_io_tlb_mem is NULL.  Can you
> > > > > turn this into :
> > > > > 
> > > > >           return dev->dma_io_tlb_mem && 
> > > > > dev->dma_io_tlb_mem->force_bounce;
> > > > > 
> > > > > for a quick debug check?
> > > > 
> > > > I just realized that dma_io_tlb_mem might be NULL like Christoph
> > > > pointed out since swiotlb might not get initialized.
> > > > However,  `Unable to handle kernel paging request at virtual address
> > > > dfff80000000000e` looks more like the address is garbage rather than
> > > > NULL?
> > > > I wonder if that's because dev->dma_io_tlb_mem is not assigned
> > > > properly (which means device_initialize is not called?).
> > > 
> > > What also looks odd is that the base "address" 0xdfff800000000000 is held 
> > > in
> > > a couple of registers, but the offset 0xe looks too small to match up to 
> > > any
> > > relevant structure member in that dereference chain :/
> > 
> > FWIW, I've managed to trigger a NULL dereference locally when swiotlb hasn't
> > been initialised but we dereference 'dev->dma_io_tlb_mem', so I think
> > Christoph's suggestion is needed regardless.
> 
> Ack to that - for SWIOTLB_NO_FORCE, io_tlb_default_mem will remain NULL. The
> massive jump in KernelCI baseline failures as of yesterday looks like every
> arm64 machine with less than 4GB of RAM blowing up...

Ok, diff below which attempts to tackle the offset issue I mentioned as
well. Qian Cai -- please can you try with these changes?

Will

--->8

diff --git a/include/linux/swiotlb.h b/include/linux/swiotlb.h
index 175b6c113ed8..39284ff2a6cd 100644
--- a/include/linux/swiotlb.h
+++ b/include/linux/swiotlb.h
@@ -116,7 +116,9 @@ static inline bool is_swiotlb_buffer(struct device *dev, 
phys_addr_t paddr)
 
 static inline bool is_swiotlb_force_bounce(struct device *dev)
 {
-       return dev->dma_io_tlb_mem->force_bounce;
+       struct io_tlb_mem *mem = dev->dma_io_tlb_mem;
+
+       return mem && mem->force_bounce;
 }
 
 void __init swiotlb_exit(void);
diff --git a/kernel/dma/swiotlb.c b/kernel/dma/swiotlb.c
index 44be8258e27b..0ffbaae9fba2 100644
--- a/kernel/dma/swiotlb.c
+++ b/kernel/dma/swiotlb.c
@@ -449,6 +449,7 @@ static int swiotlb_find_slots(struct device *dev, 
phys_addr_t orig_addr,
                dma_get_min_align_mask(dev) & ~(IO_TLB_SIZE - 1);
        unsigned int nslots = nr_slots(alloc_size), stride;
        unsigned int index, wrap, count = 0, i;
+       unsigned int offset = swiotlb_align_offset(dev, orig_addr);
        unsigned long flags;
 
        BUG_ON(!nslots);
@@ -497,7 +498,7 @@ static int swiotlb_find_slots(struct device *dev, 
phys_addr_t orig_addr,
        for (i = index; i < index + nslots; i++) {
                mem->slots[i].list = 0;
                mem->slots[i].alloc_size =
-                       alloc_size - ((i - index) << IO_TLB_SHIFT);
+                       alloc_size - (offset + ((i - index) << IO_TLB_SHIFT));
        }
        for (i = index - 1;
             io_tlb_offset(i) != IO_TLB_SEGSIZE - 1 &&