Handle armv8m unpriv_access with SAU #486

orenc17 · 2017-08-14T11:59:32Z

No description provided.

Patater · 2017-08-14T12:00:41Z

core/vmpu/src/mpu_armv8m/vmpu_armv8m_unpriv_access.c

-        if ((vmpu_unpriv_test_range(addr, UVISOR_UNPRIV_ACCESS_SIZE(size)) & (TT_RESP_NSRW_Msk | TT_RESP_SRVALID_Msk)) == (TT_RESP_NSRW_Msk | TT_RESP_SRVALID_Msk)) {
+        //if ((vmpu_unpriv_test_range(addr, UVISOR_UNPRIV_ACCESS_SIZE(size)) & (TT_RESP_NSRW_Msk | TT_RESP_SRVALID_Msk)) == (TT_RESP_NSRW_Msk | TT_RESP_SRVALID_Msk)) {
+        //this should be slower substantualy
+        if (vmpu_buffer_access_is_ok(g_active_box, addr, UVISOR_UNPRIV_ACCESS_SIZE(size))){


We shouldn't need to use the loop anymore, nor try to recover.

orenc17 · 2017-08-14T15:19:18Z

works well on v8

alzix

squash the commits and write appropriate commit message describing the problem and the fix

alzix · 2017-08-14T15:38:46Z

core/vmpu/src/mpu_armv8m/vmpu_armv8m_unpriv_access.c

-        }
-        if (++tries > 1 || !vmpu_fault_recovery_mpu(0, 0, addr, 0)) {
-            break;
+    //if ((vmpu_unpriv_test_range(addr, UVISOR_UNPRIV_ACCESS_SIZE(size)) & (TT_RESP_NSRW_Msk | TT_RESP_SRVALID_Msk)) == (TT_RESP_NSRW_Msk | TT_RESP_SRVALID_Msk)) {


please remove legacy code

Patater · 2017-08-14T16:06:54Z

core/vmpu/src/mpu_armv8m/vmpu_armv8m_unpriv_access.c

-        if (++tries > 1 || !vmpu_fault_recovery_mpu(0, 0, addr, 0)) {
-            break;
+    //if ((vmpu_unpriv_test_range(addr, UVISOR_UNPRIV_ACCESS_SIZE(size)) & (TT_RESP_NSRW_Msk | TT_RESP_SRVALID_Msk)) == (TT_RESP_NSRW_Msk | TT_RESP_SRVALID_Msk)) {
+    //this should be slower substantualy


Please explain how and why this is slower in this comment.

Patater · 2017-08-14T16:07:40Z

core/vmpu/src/mpu_armv8m/vmpu_armv8m_unpriv_access.c

        }
    }
+


Make whitespace changes in a separate commit.

Don't add trailing whitespace.

Patater · 2017-08-14T16:08:29Z

core/vmpu/src/mpu_armv8m/vmpu_armv8m_unpriv_access.c

+    //if ((vmpu_unpriv_test_range(addr, UVISOR_UNPRIV_ACCESS_SIZE(size)) & (TT_RESP_NSRW_Msk | TT_RESP_SRVALID_Msk)) == (TT_RESP_NSRW_Msk | TT_RESP_SRVALID_Msk)) {
+    //this should be slower substantualy
+    if (vmpu_buffer_access_is_ok(g_active_box, (const void *) addr, UVISOR_UNPRIV_ACCESS_SIZE(size))){
+        switch(size) {


Add a space after keywords like switch, before the (. It was wrong before. Please make style changes in a separate commit.

Patater · 2017-08-14T17:04:18Z

What testing has been done with v8-M? All v8-M tests pass?

orenc17 · 2017-08-15T08:40:23Z

@niklas-arm could you review the changes and compare it to your work?

Patater · 2017-08-15T09:47:37Z

Looks like this is suffering from the same problem on the page_heap_one_page_test_invalid_access_0 test as Niklas's implementation. Please debug.

niklarm · 2017-08-15T12:25:47Z

This is pretty similar to what I implemented. The issue is related to the SAU not being updated with the region containing the accessed memory anymore. For some reason this now leads to a hardfault.

Patater · 2017-08-22T13:04:55Z

core/vmpu/src/mpu_armv8m/vmpu_armv8m_unpriv_access.c

        }
+
        if (++tries > 1 || !vmpu_fault_recovery_mpu(0, 0, addr, 0)) {


We should have to recover. vmpu_buffer_access_is_ok should tell you what access you would have if you did recover. The access can happen from secure mode without the recovery happening.

Patater · 2017-08-22T13:05:06Z

core/vmpu/src/mpu_armv8m/vmpu_armv8m_unpriv_access.c

-
-    return response_lower & (TT_RESP_NSRW_Msk | TT_RESP_NSR_Msk | TT_RESP_RW_Msk | TT_RESP_R_Msk |
-                             TT_RESP_S_Msk | TT_RESP_SRVALID_Msk | TT_RESP_SREGION_Msk);
-}

 extern int vmpu_fault_recovery_mpu(uint32_t pc, uint32_t sp, uint32_t fault_addr, uint32_t fault_status);

 uint32_t vmpu_unpriv_access(uint32_t addr, uint32_t size, uint32_t data)
 {
    unsigned int tries = 0;
    while(1) {


We shouldn't have to loop here either.

orenc17 · 2017-08-23T11:44:50Z

core/debug/src/core_armv8m/debug_box_armv8m.c

+    /* Upon execution return debug_handler will be executed. Upon return from
+    * debug_handler, return_handler will be executed. */
+
+    //uint32_t caller = UVISOR_GET_NS_ALIAS(UVISOR_GET_NS_ADDRESS((uint32_t) debug_handler));


@Patater @niklas-arm
my problem is how can i proceed from here

Previously when we got to unpriv_access we verfied the operation with the TT assembly command, we now switch to using the SAU with uVisor vmpu functions. This could slow down the operation since we switch an assembly command with a C function which scans the virtual regions.

Patater · 2017-09-06T09:10:26Z

This is waiting for #497 to merge first, so that the debug box has a good context in which to run.

orenc17 · 2017-09-07T09:44:27Z

@Patater this PR is independent from #497
the PR that should wait for #497 is PR #496

Patater · 2017-09-07T10:04:47Z

This needs to wait, as not all the v8-M tests are passing if I merge this.

Patater reviewed Aug 14, 2017

View reviewed changes

orenc17 force-pushed the v8_upriv_access branch from a35d01b to 818ec82 Compare August 14, 2017 14:50

orenc17 force-pushed the v8_upriv_access branch from f7a284b to 07febfa Compare August 14, 2017 15:22

alzix suggested changes Aug 14, 2017

View reviewed changes

Patater suggested changes Aug 14, 2017

View reviewed changes

orenc17 force-pushed the v8_upriv_access branch 2 times, most recently from c9f3515 to c0c6db1 Compare August 15, 2017 08:26

orenc17 force-pushed the v8_upriv_access branch 2 times, most recently from 0f7c8ac to df08917 Compare August 15, 2017 08:56

Patater changed the title ~~change behavior of vmpu_unpriv_access in v8~~ Handle armv8m unpriv_access with SAU Aug 15, 2017

orenc17 force-pushed the v8_upriv_access branch from 02a49a0 to 5499cc7 Compare August 20, 2017 14:26

Patater reviewed Aug 22, 2017

View reviewed changes

orenc17 force-pushed the v8_upriv_access branch from 98230cf to 980c649 Compare August 23, 2017 11:43

orenc17 commented Aug 23, 2017

View reviewed changes

orenc17 force-pushed the v8_upriv_access branch 2 times, most recently from af7667e to 249bea0 Compare August 24, 2017 13:24

orenc17 force-pushed the v8_upriv_access branch from 249bea0 to eb32d63 Compare August 27, 2017 07:34

Patater added needs: ci needs: preceding PR labels Sep 28, 2017

theamirocohen mentioned this pull request Oct 15, 2017

Add msplim_ns for box 0 #520

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Handle armv8m unpriv_access with SAU #486

Handle armv8m unpriv_access with SAU #486

orenc17 commented Aug 14, 2017

Patater Aug 14, 2017

orenc17 commented Aug 14, 2017

alzix left a comment

alzix Aug 14, 2017

Patater Aug 14, 2017

Patater Aug 14, 2017

Patater Aug 15, 2017

Patater Aug 14, 2017

Patater commented Aug 14, 2017

orenc17 commented Aug 15, 2017

Patater commented Aug 15, 2017

niklarm commented Aug 15, 2017

Patater Aug 22, 2017

Patater Aug 22, 2017

orenc17 Aug 23, 2017

Patater commented Sep 6, 2017

orenc17 commented Sep 7, 2017

Patater commented Sep 7, 2017

		}

		if (++tries > 1 \|\| !vmpu_fault_recovery_mpu(0, 0, addr, 0)) {

Handle armv8m unpriv_access with SAU #486

Are you sure you want to change the base?

Handle armv8m unpriv_access with SAU #486

Conversation

orenc17 commented Aug 14, 2017

Choose a reason for hiding this comment

orenc17 commented Aug 14, 2017

alzix left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Patater commented Aug 14, 2017

orenc17 commented Aug 15, 2017

Patater commented Aug 15, 2017

niklarm commented Aug 15, 2017

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Patater commented Sep 6, 2017

orenc17 commented Sep 7, 2017

Patater commented Sep 7, 2017