arm64/ptrace: Don't flush ZA/ZT storage when writing ZA via ptrace

[ Upstream commit b7c510d049049409e8945b932f4b0b357fa17415 ]

When writing ZA we currently unconditionally flush the buffer used to store
it as part of ensuring that it is allocated. Since this buffer is shared
with ZT0 this means that a write to ZA when PSTATE.ZA is already set will
corrupt the value of ZT0 on a SME2 system. Fix this by only flushing the
backing storage if PSTATE.ZA was not previously set.

This will mean that short or failed writes may leave stale data in the
buffer, this seems as correct as our current behaviour and unlikely to be
something that userspace will rely on.

Fixes: f90b529bcbe5 ("arm64/sme: Implement ZT0 ptrace support")
Signed-off-by: Mark Brown <broonie@kernel.org>
Link: https://lore.kernel.org/r/20240115-arm64-fix-ptrace-za-zt-v1-1-48617517028a@kernel.org
Signed-off-by: Will Deacon <will@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>

diff --git a/arch/arm64/kernel/ptrace.c b/arch/arm64/kernel/ptrace.c
index 20d7ef8..b3f6414 100644 (file)
--- a/arch/arm64/kernel/ptrace.c
+++ b/arch/arm64/kernel/ptrace.c
@@ -1107,12 +1107,13 @@ static int za_set(struct task_struct *target,
                }
        }
 
-       /* Allocate/reinit ZA storage */
-       sme_alloc(target, true);
-       if (!target->thread.sme_state) {
-               ret = -ENOMEM;
-               goto out;
-       }
+       /*
+        * Only flush the storage if PSTATE.ZA was not already set,
+        * otherwise preserve any existing data.
+        */
+       sme_alloc(target, !thread_za_enabled(&target->thread));
+       if (!target->thread.sme_state)
+               return -ENOMEM;
 
        /* If there is no data then disable ZA */
        if (!count) {