Re: [Qemu-devel] [ARM/FDPIC v2 3/4] linux-user: ARM-FDPIC: Add support of FDPIC for ARM.

[Christophe Lyon]

On 23/04/2018 14:49, Peter Maydell wrote:

On 23 April 2018 at 08:51, Christophe Lyon <christophe.l...@st.com> wrote:

Add FDPIC info into image_info structure since interpreter info is on
stack and needs to be saved to be accessed later on.

Co-Authored-By:  Mickaël Guêné <mickael.gu...@st.com>
Signed-off-by: Christophe Lyon <christophe.l...@st.com>

diff --git a/linux-user/elfload.c b/linux-user/elfload.c
index 76d7718..1ee1e38 100644
--- a/linux-user/elfload.c
+++ b/linux-user/elfload.c
@@ -78,6 +78,11 @@ enum {
   */
  #define personality(pers)       (pers & PER_MASK)

+int info_is_fdpic(struct image_info *info)
+{
+    return (info->personality == PER_LINUX_FDPIC);
+}
+
  /* this flag is uneffective under linux too, should be deleted */
  #ifndef MAP_DENYWRITE
  #define MAP_DENYWRITE 0
@@ -287,6 +292,24 @@ static inline void init_thread(struct target_pt_regs *regs,
      /* For uClinux PIC binaries.  */
      /* XXX: Linux does this only on ARM with no MMU (do we care ?) */
      regs->uregs[10] = infop->start_data;
+
+    /* Support ARM FDPIC.  */
+    if (info_is_fdpic(infop)) {
+            /* As described in the ABI document, r7 points to the loadmap info
+             * prepared by the kernel. If an interpreter is needed, r8 points
+             * to the interpreter loadmap and r9 points to the interpreter
+             * PT_DYNAMIC info. If no interpreter is needed, r8 is zer0, and

"zero"

+             * r9 points to the main program PT_DYNAMIC info.  */

Something odd seems to have happened to the indentation here.

Nit: */ should be on a line of its own.

+            regs->uregs[7] = infop->loadmap_addr;
+            if (infop->interpreter_loadmap_addr) {
+                /* Executable is dynamically loaded.  */
+                regs->uregs[8] = infop->interpreter_loadmap_addr;
+                regs->uregs[9] = infop->interpreter_pt_dynamic_addr;
+            } else {
+                regs->uregs[8] = 0;
+                regs->uregs[9] = infop->pt_dynamic_addr;
+            }
+        }
  }

  #define ELF_NREG    18
@@ -1745,6 +1768,11 @@ static abi_ulong create_elf_tables(abi_ulong p, int 
argc, int envc,
          if (interp_info) {
              interp_info->other_info = info;
              sp = loader_build_fdpic_loadmap(interp_info, sp);
+            info->interpreter_loadmap_addr = interp_info->loadmap_addr;
+            info->interpreter_pt_dynamic_addr = interp_info->pt_dynamic_addr;
+        } else {
+            info->interpreter_loadmap_addr = 0;
+            info->interpreter_pt_dynamic_addr = 0;
          }
      }

diff --git a/linux-user/main.c b/linux-user/main.c
index 2acac36..3579f0e 100644
--- a/linux-user/main.c
+++ b/linux-user/main.c
@@ -4893,6 +4893,9 @@ int main(int argc, char **argv, char **envp)
              env->cp15.sctlr_el[1] |= SCTLR_B;
          }
  #endif
+
+        /* Are we running an FDPIC binary?  */
+        ((TaskState *)thread_cpu->opaque)->is_fdpic = info_is_fdpic(info);
      }
  #elif defined(TARGET_SPARC)
      {
diff --git a/linux-user/qemu.h b/linux-user/qemu.h
index da3b517..a2ed148 100644
--- a/linux-user/qemu.h
+++ b/linux-user/qemu.h
@@ -57,6 +57,8 @@ struct image_info {
          uint16_t        nsegs;
          void           *loadsegs;
          abi_ulong       pt_dynamic_addr;
+        abi_ulong       interpreter_loadmap_addr;
+        abi_ulong       interpreter_pt_dynamic_addr;
          struct image_info *other_info;
  };

@@ -145,6 +147,9 @@ typedef struct TaskState {
       */
      int signal_pending;

+    /* We need to know if we have an FDPIC binary to adapt signal
+     * syscalls.  */
+    int is_fdpic;

Looking at the TaskState struct it already has a pointer
to the image_info struct. So we could just have something like
     bool task_is_fdpic(TaskState *ts)
     {
         return ts->info->personality == PER_LINUX_FDPIC;
     }

and save the need to set up and test a separate flag, I think.

This patch defines and uses info_is_fdpic, the next one in the
series uses information for TaskState, so I suppose it's OK
not to add this new flag as you suggest, and in the next patch
use info_is_fdpic(ts->info) instead of adding task_is_fdpic() ?

Thanks,

Christophe

Otherwise looks good.

thanks
-- PMM
.