Skip to contentSkip to conversation

hw/ppc: SPI model - coverity fixes

PATCH In review Chalapathi V 3 participants 8 messages Aug 6, 2024 – Aug 8, 2024

lore
Share & tools
lore
#0 lore raw

cover hw/ppc: SPI model - coverity fixes

Hello,

In this patchset below defects detected during coverity scan for latest spi model changes were fixed.

- CID 1558831: Resource leaks (RESOURCE_LEAK)
Variable "rsp_payload" going out of scope leaks the storage it points to.
- CID 1558827: (OVERRUN)
Overrunning array "s->seq_op" of 8 bytes at byte offset 16 using index "get_seq_index(s) + 1" (which evaluates to 16).

Tested:
passed make check and make check-avocado

Chalapathi V (2):
Fixes: Coverity CID 1558827
Fixes: Coverity CID 1558831

 hw/ssi/pnv_spi.c | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)
Signature · 1 line
2.34.1
#1 lore raw

PATCH 1/2 Fixes: Coverity CID 1558827

In this commit the following coverity scan defect has been fixed.
CID 1558827: (OVERRUN)
Overrunning array "s->seq_op" of 8 bytes at byte offset 16
using index "get_seq_index(s) + 1" (which evaluates to 16).

---
 hw/ssi/pnv_spi.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
hw/ssi/pnv_spi.c +3 -3
index c1297ab733..a33f682897 100644
@@ -729,7 +729,7 @@ static void operation_sequencer(PnvSpi *s)
      * some operations may cause more than one frame to be sequenced.
      */
     while (get_seq_index(s) < NUM_SEQ_OPS) {
-        opcode = s->seq_op[get_seq_index(s)];
+        opcode = s->seq_op[(get_seq_index(s) & 0x7)];
         /* Set sequencer state to decode */
         s->status = SETFIELD(SPI_STS_SEQ_FSM, s->status, SEQ_STATE_DECODE);
         /*
@@ -834,8 +834,8 @@ static void operation_sequencer(PnvSpi *s)
                  * transmission to the responder without requiring a refill of
                  * the TDR between the two operations.
                  */
-                if (PNV_SPI_MASKED_OPCODE(s->seq_op[get_seq_index(s) + 1])
-                                == SEQ_OP_SHIFT_N2) {
+                if (PNV_SPI_MASKED_OPCODE(s->seq_op[((get_seq_index(s) + 1) &
+                                                0x7)]) == SEQ_OP_SHIFT_N2) {
                     send_n1_alone = false;
                 }
                 s->status = SETFIELD(SPI_STS_SHIFTER_FSM, s->status,
Signature · 1 line
2.34.1
#2 lore raw

Re: [PATCH v1 1/2] Fixes: Coverity CID 1558827

Hi Chalapathi,

Please prefix subject with "hw/ssi/pnv".

Quoted 15 lines · Chalapathi V

In this commit the following coverity scan defect has been fixed.
CID 1558827: (OVERRUN)
Overrunning array "s->seq_op" of 8 bytes at byte offset 16
using index "get_seq_index(s) + 1" (which evaluates to 16).

---
  hw/ssi/pnv_spi.c | 6 +++---
  1 file changed, 3 insertions(+), 3 deletions(-)
hw/ssi/pnv_spi.c +0 -0
index c1297ab733..a33f682897 100644

hw/ssi/pnv_spi.c static void operation_sequencer(PnvSpi *s) · line 732

Quoted 5 lines
hw/ssi/pnv_spi.c +1 -1
@@ -729,7 +729,7 @@ static void operation_sequencer(PnvSpi *s)
       * some operations may cause more than one frame to be sequenced.
       */
      while (get_seq_index(s) < NUM_SEQ_OPS) {
-        opcode = s->seq_op[get_seq_index(s)];
+        opcode = s->seq_op[(get_seq_index(s) & 0x7)];

seq_op[] has PNV_SPI_REG_SIZE elements, PNV_SPI_REG_SIZE being 8.

We also have NUM_SEQ_OPS defined as 8.

get_seq_index() returns SPI_STS_SEQ_INDEX. Being defined as PPC_BITMASK(28, 31), it is 4-bit width. (I was wondering why not have get_seq_index return a masked value).

I don't know this area, but this code is not very clear...

Alternative to make Coverity happy:

  seq_index = get_seq_index(s);
  assert(seq_index < NUM_SEQ_OPS);
  opcode = s->seq_op[seq_index];
Quoted 14 lines
hw/ssi/pnv_spi.c +2 -2
          /* Set sequencer state to decode */
          s->status = SETFIELD(SPI_STS_SEQ_FSM, s->status, SEQ_STATE_DECODE);
          /*
@@ -834,8 +834,8 @@ static void operation_sequencer(PnvSpi *s)
                   * transmission to the responder without requiring a refill of
                   * the TDR between the two operations.
                   */
-                if (PNV_SPI_MASKED_OPCODE(s->seq_op[get_seq_index(s) + 1])
-                                == SEQ_OP_SHIFT_N2) {
+                if (PNV_SPI_MASKED_OPCODE(s->seq_op[((get_seq_index(s) + 1) &
+                                                0x7)]) == SEQ_OP_SHIFT_N2) {
                      send_n1_alone = false;
                  }
                  s->status = SETFIELD(SPI_STS_SHIFTER_FSM, s->status,
#3 lore raw

Re: [PATCH v1 1/2] Fixes: Coverity CID 1558827

In reply to Chalapathi V

On Tue, 6 Aug 2024 at 14:50, Chalapathi V <[email protected]> wrote:

The Subject email for a patch should say what it does in terms of code changes, not just list the bug number or coverity issue number, and it should start with the prefix showing what part of the codebase it is changing. You can look through other commit messages with "git log" to see the general style.

Quoted 15 lines

In this commit the following coverity scan defect has been fixed.
CID 1558827: (OVERRUN)
Overrunning array "s->seq_op" of 8 bytes at byte offset 16
using index "get_seq_index(s) + 1" (which evaluates to 16).

---
 hw/ssi/pnv_spi.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
hw/ssi/pnv_spi.c +0 -0
index c1297ab733..a33f682897 100644

hw/ssi/pnv_spi.c static void operation_sequencer(PnvSpi *s) · line 732

Quoted 5 lines
hw/ssi/pnv_spi.c +1 -1
@@ -729,7 +729,7 @@ static void operation_sequencer(PnvSpi *s)
      * some operations may cause more than one frame to be sequenced.
      */
     while (get_seq_index(s) < NUM_SEQ_OPS) {
-        opcode = s->seq_op[get_seq_index(s)];
+        opcode = s->seq_op[(get_seq_index(s) & 0x7)];

This doesn't seem like the right fix, as Philippe points out. It's also not any of the possible approaches I suggested in my email in the other thread.

* if we're confident that this value really can't be more than 7, then we should assert() that
* if the value might be more than 7 if the guest has done something silly, we should arrange to detect and handle that error
* if the hardware really ignores the high bit of the field, that should be implemented in get_seq_index(), not in its callers
* we should consider whether using a local variable instead of repeatedly calling get_seq_index() might make the code easier to read (as well as helping Coverity)

hw/ssi/pnv_spi.c static void operation_sequencer(PnvSpi *s) · line 838

Quoted 11 lines
hw/ssi/pnv_spi.c +2 -2
         /* Set sequencer state to decode */
         s->status = SETFIELD(SPI_STS_SEQ_FSM, s->status, SEQ_STATE_DECODE);
         /*
@@ -834,8 +834,8 @@ static void operation_sequencer(PnvSpi *s)
                  * transmission to the responder without requiring a refill of
                  * the TDR between the two operations.
                  */
-                if (PNV_SPI_MASKED_OPCODE(s->seq_op[get_seq_index(s) + 1])
-                                == SEQ_OP_SHIFT_N2) {
+                if (PNV_SPI_MASKED_OPCODE(s->seq_op[((get_seq_index(s) + 1) &
+                                                0x7)]) == SEQ_OP_SHIFT_N2) {

This doesn't look right. If operation 7 is SHIFT_N1 then we
do not want to look at operation 0 (which is what "(get_seq_index(s) + 1) & 0x7" will cause us to look at), because operation 0 is unrelated. What we want to do is have this condition be "if (sequence index != 7 && s->seq_op[sequence index + 1] is SHIFT_N2)".

hw/ssi/pnv_spi.c static void operation_sequencer(PnvSpi *s) · line 841

Quoted 3 lines
hw/ssi/pnv_spi.c +0 -0
                     send_n1_alone = false;
                 }
                 s->status = SETFIELD(SPI_STS_SHIFTER_FSM, s->status,

thanks
-- PMM

#4 lore raw

PATCH 2/2 Fixes: Coverity CID 1558831

In reply to Chalapathi V

In this commit the following coverity scan defect has been fixed CID 1558831: Resource leaks (RESOURCE_LEAK)
Variable "rsp_payload" going out of scope leaks the storage it points to.

---
 hw/ssi/pnv_spi.c | 1 +
 1 file changed, 1 insertion(+)
hw/ssi/pnv_spi.c +1 -0
index a33f682897..dbe06df224 100644
@@ -237,6 +237,7 @@ static void transfer(PnvSpi *s, PnvXferBuffer *payload)
     }
     if (rsp_payload != NULL) {
         spi_response(s, s->N1_bits, rsp_payload);
+        pnv_spi_xfer_buffer_free(rsp_payload);
     }
 }
 
Signature · 1 line
2.34.1
#5 lore raw

Re: [PATCH v1 2/2] Fixes: Coverity CID 1558831

Quoted 9 lines · Chalapathi V

In this commit the following coverity scan defect has been fixed CID 1558831: Resource leaks (RESOURCE_LEAK)
Variable "rsp_payload" going out of scope leaks the storage it points to.

---
  hw/ssi/pnv_spi.c | 1 +
  1 file changed, 1 insertion(+)
Fixes b4cb930e40 ("hw/ssi: Extend SPI model")
#6 lore raw

Re: [PATCH v1 2/2] Fixes: Coverity CID 1558831

In reply to Chalapathi V

Back at this patch since Cédric asked me to look at it.

Quoted 15 lines · Chalapathi V

In this commit the following coverity scan defect has been fixed CID 1558831: Resource leaks (RESOURCE_LEAK)
Variable "rsp_payload" going out of scope leaks the storage it points to.

---
  hw/ssi/pnv_spi.c | 1 +
  1 file changed, 1 insertion(+)
hw/ssi/pnv_spi.c +0 -0
index a33f682897..dbe06df224 100644

hw/ssi/pnv_spi.c static void transfer(PnvSpi *s, PnvXferBuffer *payload) · line 243

Quoted 7 lines
hw/ssi/pnv_spi.c +1 -0
@@ -237,6 +237,7 @@ static void transfer(PnvSpi *s, PnvXferBuffer *payload)
      }
      if (rsp_payload != NULL) {
          spi_response(s, s->N1_bits, rsp_payload);
+        pnv_spi_xfer_buffer_free(rsp_payload);
      }
  }
 

Or bigger patch but simpler logic:

-- >8 --

hw/ssi/pnv_spi.c +3 -0
index c1297ab733..aaedba84af 100644
@@ -217,6 +217,9 @@ static void transfer(PnvSpi *s, PnvXferBuffer *payload)
     PnvXferBuffer *rsp_payload = NULL;
 
     rsp_payload = pnv_spi_xfer_buffer_new();
+    if (!rsp_payload) {
+        return;
+    }
     for (int offset = 0; offset < payload->len; offset += 
s->transfer_len) {
         tx = 0;
         for (int i = 0; i < s->transfer_len; i++) {
@@ -235,9 +238,8 @@ static void transfer(PnvSpi *s, PnvXferBuffer *payload)
                     (rx >> (8 * (s->transfer_len - 1) - i * 8)) & 0xFF;
         }
     }
-    if (rsp_payload != NULL) {
-        spi_response(s, s->N1_bits, rsp_payload);
-    }
+    spi_response(s, s->N1_bits, rsp_payload);
+    pnv_spi_xfer_buffer_free(rsp_payload);

}

---

I note few things is odd here:

1/ pnv_spi_xfer_buffer_new() uses the GLib g_malloc0() call
while pnv_spi_xfer_buffer_free() uses plan free().

2/ pnv_spi_xfer_buffer_free() frees payload->data so doesn't
match pnv_spi_xfer_buffer_new().

This is a bit disappointing.

#7 lore raw

Re: [PATCH v1 2/2] Fixes: Coverity CID 1558831

Quoted 35 lines · Philippe Mathieu-Daudé

Back at this patch since Cédric asked me to look at it.

Quoted 15 lines · Chalapathi V

In this commit the following coverity scan defect has been fixed CID 1558831: Resource leaks (RESOURCE_LEAK)
Variable "rsp_payload" going out of scope leaks the storage it points to.

---
  hw/ssi/pnv_spi.c | 1 +
  1 file changed, 1 insertion(+)
hw/ssi/pnv_spi.c +0 -0
index a33f682897..dbe06df224 100644

hw/ssi/pnv_spi.c static void transfer(PnvSpi *s, PnvXferBuffer *payload) · line 243

Quoted 7 lines
hw/ssi/pnv_spi.c +1 -0
@@ -237,6 +237,7 @@ static void transfer(PnvSpi *s, PnvXferBuffer *payload)
      }
      if (rsp_payload != NULL) {
          spi_response(s, s->N1_bits, rsp_payload);
+        pnv_spi_xfer_buffer_free(rsp_payload);
      }
  }
 

Or bigger patch but simpler logic:

-- >8 --

hw/ssi/pnv_spi.c +0 -0
index c1297ab733..aaedba84af 100644

hw/ssi/pnv_spi.c static void transfer(PnvSpi *s, PnvXferBuffer *payload) · line 222

Quoted 6 lines
hw/ssi/pnv_spi.c +3 -0
@@ -217,6 +217,9 @@ static void transfer(PnvSpi *s, PnvXferBuffer *payload)
      PnvXferBuffer *rsp_payload = NULL;
 
      rsp_payload = pnv_spi_xfer_buffer_new();
+    if (!rsp_payload) {
+        return;
+    }

pnv_spi_xfer_buffer_new() cannot fail (it calls g_malloc0, which aborts on not having enough memory) so we don't need to check it for NULL.

hw/ssi/pnv_spi.c static void transfer(PnvSpi *s, PnvXferBuffer *payload) · line 240

Quoted 11 lines
hw/ssi/pnv_spi.c +0 -0
      for (int offset = 0; offset < payload->len; offset +=
 s->transfer_len) {
          tx = 0;
          for (int i = 0; i < s->transfer_len; i++) {
hw/ssi/pnv_spi.c +0 -3
@@ -235,9 +238,8 @@ static void transfer(PnvSpi *s, PnvXferBuffer *payload)
                      (rx >> (8 * (s->transfer_len - 1) - i * 8)) & 0xFF;
          }
      }
-    if (rsp_payload != NULL) {
-        spi_response(s, s->N1_bits, rsp_payload);
-    }

(So this NULL check was unnecessary in the old code.)

Quoted 10 lines
hw/ssi/pnv_spi.c +2 -1
+    spi_response(s, s->N1_bits, rsp_payload);
+    pnv_spi_xfer_buffer_free(rsp_payload);
  }
 
---

I note few things is odd here:

1/ pnv_spi_xfer_buffer_new() uses the GLib g_malloc0() call
while pnv_spi_xfer_buffer_free() uses plan free().

This does work (glib guarantees now that g_malloc and friends can be paired with free), but it is not great style.

Quoted 2 lines

2/ pnv_spi_xfer_buffer_free() frees payload->data so doesn't
match pnv_spi_xfer_buffer_new().

This part I think is OK -- the payload->data buffer is allocated with g_realloc(), so pnv_spi_xfer_buffer_new() creates a valid initial state (payload->data = NULL and payload->len = 0), pnv_spi_xfer_buffer_write_ptr() may allocate or extend the buffer (using g_realloc and updating the payload->len to match), and pnv_spi_xfer_buffer_free() will free the buffer (including handling the "buffer never allocated so payload->data = NULL" case, because free(NULL) is OK).

What it doesn't get right I think is that when pnv_spi_xfer_buffer_write_ptr() extends the payload->data buffer it doesn't zero the newly added memory, so I think we might end up giving the guest back the contents of uninitialized memory.

Another style issue is:

    PnvXferBuffer *payload = g_malloc0(sizeof(*payload));

g_new0(PnvXferBuffer, 1) is the better way to say "allocate me a zeroed out PnvXferBuffer", rather than manual use of sizeof.

The various places which do

    if (*payload == NULL) {
        *payload = pnv_spi_xfer_buffer_new();
    }

also look odd to me -- I'm pretty sure that
in operation_shiftn1() and operation_shiftn2() it is impossible for them to be called with a NULL payload.

The PnvXferBuffer struct is only 12 bytes large (a length and a pointer), so it seems rather overkill to be allocating and freeing it -- is it possible to have it be a local variable instead?

My other question here is whether we need to be doing dynamic memory allocation, extension and freeing of the payload->data at all. This is quite an unusual thing to need to do in a hardware device model, because usually the real hardware we're modelling doesn't have infinite resources, it has some fixed amount of memory that it has to work with. If the guest can effectively ask us to allocate arbitrary amounts of memory, do we have appropriate guards in place to forbid that?

Is there a datasheet for this device?

thanks
-- PMM