Hi Russell,

On 8/18/2016 7:24 AM, Russell King - ARM Linux wrote:

On Wed, Aug 17, 2016 at 03:05:17PM +0300, Roger Quadros wrote:

quoted

Since commit 6ce0d2001692 ("ARM: dma: Use dma_pfn_offset for dma address translation"),
dma_to_pfn() already returns the PFN with the physical memory start offset
so we don't need to add it again.

This fixes USB mass storage lock-up problem on systems that can't do DMA
over the entire physical memory range (e.g.) Keystone 2 systems with 4GB RAM
can only do DMA over the first 2GB. [K2E-EVM].

What happens there is that without this patch SCSI layer sets a wrong
bounce buffer limit in scsi_calculate_bounce_limit() for the USB mass
storage device. dma_max_pfn() evaluates to 0x8fffff and bounce_limit
is set to 0x8fffff000 whereas maximum DMA'ble physical memory on Keystone 2
is 0x87fffffff. This results in non DMA'ble pages being given to the
USB controller and hence the lock-up.

NOTE: in the above case, USB-SCSI-device's dma_pfn_offset was showing as 0.
This should have really been 0x780000 as on K2e, LOWMEM_START is 0x80000000
and HIGHMEM_START is 0x800000000. DMA zone is 2GB so dma_max_pfn should be
0x87ffff. The incorrect dma_pfn_offset for the USB storage device is because
USB devices are not correctly inheriting the dma_pfn_offset from the
USB host controller. This will be fixed by a separate patch.

I'd like to hear from Santosh, as the author of the original change.
The original commit doesn't mention which platform it was intended for
or what the problem was, which would've been helpful.

 From what I recollect, we did these changes to make the max pfn behave
same on ARM arch as other archs. This patch was evolved as part of
fixing the max*pfn assumption.

commit 26ba47b18318abe7dadbe9294a611c0e932651d8
Author: Santosh Shilimkar [off-list ref]
Date:   Thu Aug 1 03:12:01 2013 +0100

     ARM: 7805/1: mm: change max*pfn to include the physical offset of 
memory

     Most of the kernel code assumes that max*pfn is maximum pfns because
     the physical start of memory is expected to be PFN0. Since this
     assumption is not true on ARM architectures, the meaning of max*pfn
     is number of memory pages. This is done to keep drivers happy which
     are making use of of these variable to calculate the dma bounce limit
     using dma_mask.

     Now since we have a architecture override possibility for DMAable
     maximum pfns, lets make meaning of max*pfns as maximum pnfs on ARM
     as well.

quoted

Fixes: 6ce0d2001692 ("ARM: dma: Use dma_pfn_offset for dma address translation")
Cc: stable at vger.kernel.org
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Russell King <redacted>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Olof Johansson <redacted>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Linus Walleij <redacted>
Reported-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: Roger Quadros <redacted>
---
 arch/arm/include/asm/dma-mapping.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/arch/arm/include/asm/dma-mapping.h b/arch/arm/include/asm/dma-mapping.h
index d009f79..bf02dbd 100644
--- a/arch/arm/include/asm/dma-mapping.h
+++ b/arch/arm/include/asm/dma-mapping.h

@@ -111,7 +111,7 @@ static inline dma_addr_t virt_to_dma(struct device *dev, void *addr)
 /* The ARM override for dma_max_pfn() */
 static inline unsigned long dma_max_pfn(struct device *dev)
 {
-	return PHYS_PFN_OFFSET + dma_to_pfn(dev, *dev->dma_mask);
+	return dma_to_pfn(dev, *dev->dma_mask);
 }
 #define dma_max_pfn(dev) dma_max_pfn(dev)

By doing this change I hope we don't break other drivers on Keystone so
am not sure about the change.