Re: [PATCH v4 1/5] powerpc/rtas: Handle ibm,open-errinjct return format

Hi Sourabh,

Thanks for the detailed review.

On 01/09/26 2:52 PM, Sourabh Jain wrote:

On 31/08/26 12:24, Narayana Murty N wrote:

quoted

ibm,open-errinjct uses a non-standard RTAS return layout:

   rets[0] = session token  (output parameter)
   rets[1] = status code

Unlike all other RTAS functions which use:

   rets[0] = status code
   rets[1..] = output parameters

Add rtas_token_is_open_errinjct() to identify this call, and
rtas_status_from_args() to extract status from the correct position.

Add an early guard in rtas_call() that rejects ibm,open-errinjct
invocations where nret < 2, since reading rets[1] would be out of
bounds:

   if (rtas_token_is_open_errinjct(token) && nret < 2) {
           WARN_ON_ONCE(1);
           return RTAS_INVALID_PARAMETER;
   }

Adjust the output-copy loop so that for ibm,open-errinjct:

   return value = rets[1]   (status)
   outputs[0]   = rets[0]   (session token)

For all other functions the existing convention is preserved:

   return value = rets[0]   (status)
   outputs[0..] = rets[1..] (non-status outputs)

Move the "/* A -1 return code... */" comment immediately before the
ret == -1 check so it documents the check it guards.

Also fix sys_rtas() last-error status detection: ibm,open-errinjct
places status at rets[1], so the -1 sentinel check must use rets[1]
for that function rather than always using rets[0].

Reference: OpenPOWER PAPR documentation
            https://files.openpower.foundation/s/XFgfMaqLMD5Bcm8
Signed-off-by: Narayana Murty N <redacted>
---
  arch/powerpc/kernel/rtas.c | 78 +++++++++++++++++++++++++++++++++-----
  1 file changed, 68 insertions(+), 10 deletions(-)

diff --git a/arch/powerpc/kernel/rtas.c b/arch/powerpc/kernel/rtas.c
index 8d81c1e7a8db..7131870655c6 100644
--- a/arch/powerpc/kernel/rtas.c
+++ b/arch/powerpc/kernel/rtas.c

@@ -1117,6 +1117,28 @@ static bool token_is_restricted_errinjct(s32 

token)
             token == rtas_function_token(RTAS_FN_IBM_ERRINJCT);
  }
+/*
+ * ibm,open-errinjct uses a non-standard return layout:
+ *   rets[0] = session token  (output parameter)
+ *   rets[1] = status code
+ *
+ * All other RTAS functions use the standard layout:
+ *   rets[0] = status code
+ *   rets[1..] = output parameters
+ */
+static inline bool rtas_token_is_open_errinjct(int token)
+{
+    return token == rtas_function_token(RTAS_FN_IBM_OPEN_ERRINJCT);
+}
+
+static int rtas_status_from_args(int token, struct rtas_args *args, 
int nret)
+{
+    if (rtas_token_is_open_errinjct(token))
+        return be32_to_cpu(args->rets[1]);
+
+    return nret > 0 ? be32_to_cpu(args->rets[0]) : 0;
+}
+
  /**
   * rtas_call() - Invoke an RTAS firmware function.
   * @token: Identifies the function being invoked.

@@ -1198,6 +1220,16 @@ int rtas_call(int token, int nargs, int nret, 

int *outputs, ...)
              return -1;
      }
+    /*
+     * ibm,open-errinjct returns rets[0]=session_token, rets[1]=status.
+     * We need nret >= 2 to read status from rets[1].  Reject early if
+     * the caller forgot to account for the extra return cell.
+     */
+    if (rtas_token_is_open_errinjct(token) && nret < 2) {
+        WARN_ON_ONCE(1);
+        return RTAS_INVALID_PARAMETER;

Nit: I would prefer -EINVAL instead. RTAS_INVALID_PARAMETER is RTAS 
error code but here kernel is validating the parameter so I think - 
EINVAL would be better.

I agree this is a kernel-side validation before entering RTAS, but for
rtas_call() I think keeping RTAS_INVALID_PARAMETER is safer.

The return value from rtas_call() is normally interpreted by callers as 
an RTAS status code. If we return -EINVAL from rtas_call(), the caller 
may still treat it as an RTAS return value. Since this is an invalid 
RTAS call layout for ibm,open-errinjct, I would prefer to keep the 
return value as RTAS_INVALID_PARAMETER in rtas_call().

quoted

+    }
+
      if ((mfmsr() & (MSR_IR|MSR_DR)) != (MSR_IR|MSR_DR)) {
          WARN_ON_ONCE(1);
          return -1;

@@ -1213,15 +1245,33 @@ int rtas_call(int token, int nargs, int nret, 

int *outputs, ...)
      va_rtas_call_unlocked(args, token, nargs, nret, list);
      va_end(list);
+    ret = rtas_status_from_args(token, args, nret);
+
      /* A -1 return code indicates that the last command couldn't
-       be completed due to a hardware error. */
-    if (be32_to_cpu(args->rets[0]) == -1)
+     * be completed due to a hardware error.
+     */
+    if (ret == -1)
          buff_copy = __fetch_rtas_last_error(NULL);
-    if (nret > 1 && outputs != NULL)
-        for (i = 0; i < nret-1; ++i)
-            outputs[i] = be32_to_cpu(args->rets[i + 1]);
-    ret = (nret > 0) ? be32_to_cpu(args->rets[0]) : 0;
+    /*
+     * Copy non-status outputs to the caller's buffer.
+     *
+     * For ibm,open-errinjct the layout is:
+     *   rets[0] = session token  -> outputs[0]
+     *   rets[1] = status         (returned, not copied)
+     *
+     * For all other RTAS functions:
+     *   rets[0] = status         (returned, not copied)
+     *   rets[1..nret-1] -> outputs[0..nret-2]
+     */
+    if (outputs != NULL) {
+        if (rtas_token_is_open_errinjct(token)) {
+            outputs[0] = be32_to_cpu(args->rets[0]);
+        } else if (nret > 1) {
+            for (i = 0; i < nret - 1; ++i)
+                outputs[i] = be32_to_cpu(args->rets[i + 1]);
+        }
+    }
      lockdep_unpin_lock(&rtas_lock, cookie);
      raw_spin_unlock_irqrestore(&rtas_lock, flags);

@@ -1942,10 +1992,18 @@ SYSCALL_DEFINE1(rtas, struct rtas_args __user 

*, uargs)
      do_enter_rtas(&rtas_args);
      args = rtas_args;
-    /* A -1 return code indicates that the last command couldn't
-       be completed due to a hardware error. */
-    if (be32_to_cpu(args.rets[0]) == -1)
-        errbuf = __fetch_rtas_last_error(buff_copy);
+    /*
+     * A -1 return code indicates that the last command couldn't
+     * be completed due to a hardware error.  ibm,open-errinjct
+     * places status at rets[1] rather than rets[0]; check the
+     * correct position for the -1 sentinel.
+     */
+    {
+        __be32 status_cell = (token == 
rtas_function_token(RTAS_FN_IBM_OPEN_ERRINJCT) &&
+                      nret >= 2) ? args.rets[1] : args.rets[0];
+        if (be32_to_cpu(status_cell) == -1)
+            errbuf = __fetch_rtas_last_error(buff_copy);

Do we know what happens when the ibm,open-errinjct RTAS call is made
with nret < 2?

The reason I’m asking is that, even with the above changes,
args.rets[0] is used as the return code if the ibm,open-errinjct call
is made with nret < 2.

I like the approach you took in rtas_call() of pre-validating nret for
the ibm,open-errinjct RTAS call and returning early if it is less than
2. I think we can use a similar approach here as well. If we do that,
the above code changes will be much cleaner. In that case, we don't
have to figure out how RTAS processes ibm,open-errinjct with nret < 2.

The only concern I have is that this change would alter the system
call behavior. Right now, the kernel accepts nret < 2 for
ibm,open-errinjct and makes the RTAS call, but with the above suggested
change, the kernel would return early if nret < 2.

The prominent user of this system call is librtas, which passes
nret = 2 for ibm,open-errinjct:

https://github.com/ibm-power-utilities/librtas/blob/ 
d321a1f5ae3d528ba027fc748d1cc1123dd4ae29/librtas_src/syscall_calls.c#L488

Also, as per PAPR, users are supposed to pass nret = 2 for this RTAS
call. So I think it should be fine to validate nret in sys_rtas for
ibm,open-errinjct and return early if it is found to be less than 2.

Since this is a change in system call behavior, I want to be a
little cautious. So, I’d like to hear your thoughts and would also
like to know what others think about making the above change.

Agreed.

For ibm,open-errinjct, rets[0] is the session token and rets[1] is the
status. So when nret < 2, we should not fall back to rets[0], because 
that can interpret a session token as a status code.

I will add an early validation in sys_rtas() before entering RTAS for 
this case and return -EINVAL there, since sys_rtas() is the userspace 
syscall boundary.

So the split will be:
rtas_call(): return RTAS_INVALID_PARAMETER for invalid RTAS return layout

sys_rtas(): return -EINVAL for invalid userspace syscall arguments

This should not affect the valid librtas path, since librtas already 
passes nret = 2 for ibm,open-errinjct, as required by PAPR.

Thanks,
Narayana

Thanks, Sourabh Jain

quoted

+    }
      lockdep_unpin_lock(&rtas_lock, cookie);
      raw_spin_unlock_irqrestore(&rtas_lock, flags);