s390/pci: fix max size calculation in zpci_memcpy_toio()

The zpci_get_max_write_size() helper is used to determine the maximum
size a PCI store or load can use at a given __iomem address.

For the PCI block store the following restrictions apply:

1. The dst + len must not cross a 4K boundary in the (pseudo-)MMIO space
2. len must not exceed ZPCI_MAX_WRITE_SIZE
3. len must be a multiple of 8 bytes
4. The src address must be double word (8 byte) aligned
5. The dst address must be double word (8 byte) aligned

Otherwise only a normal PCI store which takes its src value from
a register can be used. For these PCI store restriction 1 still applies.
Similarly 1 also applies to PCI loads.

It turns out zpci_max_write_size() instead implements stricter
conditions which prevents PCI block stores from being used where they
can and should be used. In particular instead of conditions 4 and 5 it
wrongly enforces both dst and src to be size aligned. This indirectly
covers condition 1 but also prevents many legal PCI block stores.

On top of the functional shortcomings the zpci_get_max_write_size() is
misnamed as it is used for both read and write size calculations. Rename
it to zpci_get_max_io_size() and implement the listed conditions
explicitly.

Reviewed-by: Matthew Rosato <[email protected]>
Fixes: cd24834 ("s390/pci: base support")
Signed-off-by: Niklas Schnelle <[email protected]>
[[email protected] replaced spaces with tabs]
Signed-off-by: Alexander Gordeev <[email protected]>

arch/s390/include/asm/pci_io.h

@@ -11,6 +11,8 @@
 /* I/O size constraints */
 #define ZPCI_MAX_READ_SIZE 8
 #define ZPCI_MAX_WRITE_SIZE 128
+#define ZPCI_BOUNDARY_SIZE (1 << 12)
+#define ZPCI_BOUNDARY_MASK (ZPCI_BOUNDARY_SIZE - 1)
 
 /* I/O Map */
 #define ZPCI_IOMAP_SHIFT 48
@@ -125,16 +127,18 @@ static inline int zpci_read_single(void *dst, const volatile void __iomem *src,
 int zpci_write_block(volatile void __iomem *dst, const void *src,
  unsigned long len);
 
-static inline u8 zpci_get_max_write_size(u64 src, u64 dst, int len, int max)
+static inline int zpci_get_max_io_size(u64 src, u64 dst, int len, int max)
 {
- int count = len > max ? max : len, size = 1;
+ int offset = dst & ZPCI_BOUNDARY_MASK;
+ int size;
 
- while (!(src & 0x1) && !(dst & 0x1) && ((size << 1) <= count)) {
- dst = dst >> 1;
- src = src >> 1;
- size = size << 1;
- }
- return size;
+ size = min3(len, ZPCI_BOUNDARY_SIZE - offset, max);
+ if (IS_ALIGNED(src, 8) && IS_ALIGNED(dst, 8) && IS_ALIGNED(size, 8))
+ return size;
+
+ if (size >= 8)
+ return 8;
+ return rounddown_pow_of_two(size);
 }
 
 static inline int zpci_memcpy_fromio(void *dst,
@@ -144,9 +148,9 @@ static inline int zpci_memcpy_fromio(void *dst,
  int size, rc = 0;
 
  while (n > 0) {
- size = zpci_get_max_write_size((u64 __force) src,
-  (u64) dst, n,
-  ZPCI_MAX_READ_SIZE);
+ size = zpci_get_max_io_size((u64 __force) src,
+ (u64) dst, n,
+ ZPCI_MAX_READ_SIZE);
  rc = zpci_read_single(dst, src, size);
  if (rc)
  break;
@@ -166,9 +170,9 @@ static inline int zpci_memcpy_toio(volatile void __iomem *dst,
  return -EINVAL;
 
  while (n > 0) {
- size = zpci_get_max_write_size((u64 __force) dst,
-  (u64) src, n,
-  ZPCI_MAX_WRITE_SIZE);
+ size = zpci_get_max_io_size((u64 __force) dst,
+ (u64) src, n,
+ ZPCI_MAX_WRITE_SIZE);
  if (size > 8) /* main path */
  rc = zpci_write_block(dst, src, size);
  else

arch/s390/pci/pci_mmio.c

@@ -97,9 +97,9 @@ static inline int __memcpy_toio_inuser(void __iomem *dst,
  return -EINVAL;
 
  while (n > 0) {
- size = zpci_get_max_write_size((u64 __force) dst,
-  (u64 __force) src, n,
-  ZPCI_MAX_WRITE_SIZE);
+ size = zpci_get_max_io_size((u64 __force) dst,
+ (u64 __force) src, n,
+ ZPCI_MAX_WRITE_SIZE);
  if (size > 8) /* main path */
  rc = __pcistb_mio_inuser(dst, src, size, &status);
  else
@@ -242,9 +242,9 @@ static inline int __memcpy_fromio_inuser(void __user *dst,
  u8 status;
 
  while (n > 0) {
- size = zpci_get_max_write_size((u64 __force) src,
-  (u64 __force) dst, n,
-  ZPCI_MAX_READ_SIZE);
+ size = zpci_get_max_io_size((u64 __force) src,
+ (u64 __force) dst, n,
+ ZPCI_MAX_READ_SIZE);
  rc = __pcilg_mio_inuser(dst, src, size, &status);
  if (rc)
  break;