Re: [PATCH] [GDB,AArch64] Fix off-by-one when calculating tag granules.

[Alan Hayward <Alan.Hayward@arm.com>]

On 13 May 2021, at 11:56, Luis Machado <luis.machado@linaro.org<mailto:luis.machado@linaro.org>> wrote:

On 5/13/21 7:50 AM, Alan Hayward wrote:
On 11 May 2021, at 14:20, Luis Machado via Gdb-patches <gdb-patches@sourceware.org<mailto:gdb-patches@sourceware.org>> wrote:

When we want to fetch tags from a memory range, the last address in that
range is not included.

There is a off-by-one error in aarch64_mte_get_tag_granules, which this
patch fixes.

gdb/ChangeLog:

YYYY-MM-DD  Luis Machado  <luis.machado@linaro.org<mailto:luis.machado@linaro.org>>

* arch/aarch64-mte-linux.c (aarch64_mte_get_tag_granules): Don't
include the last address in the range.
---
gdb/arch/aarch64-mte-linux.c | 5 +++--
1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/gdb/arch/aarch64-mte-linux.c b/gdb/arch/aarch64-mte-linux.c
index 959c0247ed5..7c2ae9a7058 100644
--- a/gdb/arch/aarch64-mte-linux.c
+++ b/gdb/arch/aarch64-mte-linux.c
@@ -31,9 +31,10 @@ aarch64_mte_get_tag_granules (CORE_ADDR addr, size_t len, size_t granule_size)
  /* Start address */
  CORE_ADDR s_addr = align_down (addr, granule_size);
  /* End address */
-  CORE_ADDR e_addr = align_down (addr + len, granule_size);
+  CORE_ADDR e_addr = align_down (addr + len - 1, granule_size);

-  /* We always have at least 1 granule.  */
+  /* We always have at least 1 granule because len is non-zero at this
+     point.  */
  return 1 + (e_addr - s_addr) / granule_size;
Ok,  I’m happy with round down because we never want to read/write a partial tag
But if len is greater than zero, but less than the size of a one tag, then it returns 1. Is that correct?

The range is specified by [address, address + len). If len is non-zero, we will always have a non-empty range and therefore will always have tag granules.

Should it instead be:
  if (len < granule_size)
    return 0;

We want to know how many tag granules exist in a particular range. We only get 0 granules if the range is empty.

Assuming addr is aligned, then calculating length -> return value
3 * granulesize -> 3
2.5 * granulesize -> 2
2 * granulesize -> 2
1.5 * granulesize -> 1
1 * granulesize -> 1
0.5 * granulesize -> 1.   - this doesn’t match the others.
0 -> 0

?

Alan.