Re: [Qemu-devel] [PATCH RFC 3/4] exec: don't exit unconditionally if failed to allocate memory

[Michael S. Tsirkin]

On Sat, Jun 14, 2014 at 07:07:39PM +0200, Paolo Bonzini wrote:
> Il 14/06/2014 06:48, Hu Tao ha scritto:
> >return -1 instead.
> >
> >Now user can add objects memory-backend-ram on-the-fly, fail it if
> >cannot allocate memory rather than quit qemu.
> >
> >Signed-off-by: Hu Tao <address@hidden>
> 
> This needs an audit of all callers or, alternatively, we need to add
> memory_region_init_ram_nofail.  Better leave it for after the merge.
> 
> Paolo

Specifically memory_region_init_ram_from_file does not seem to
handle failures.

qemu_ram_free chunk also looks weird. Can we not avoid calling
free on invalid addresses?

> >---
> > backends/hostmem-ram.c | 3 +++
> > exec.c                 | 6 +++++-
> > 2 files changed, 8 insertions(+), 1 deletion(-)
> >
> >diff --git a/backends/hostmem-ram.c b/backends/hostmem-ram.c
> >index d9a8290..afb305d 100644
> >--- a/backends/hostmem-ram.c
> >+++ b/backends/hostmem-ram.c
> >@@ -28,6 +28,9 @@ ram_backend_memory_alloc(HostMemoryBackend *backend, Error 
> >**errp)
> >     path = object_get_canonical_path_component(OBJECT(backend));
> >     memory_region_init_ram(&backend->mr, OBJECT(backend), path,
> >                            backend->size);
> >+    if (backend->mr.ram_addr == -1) {
> >+        error_setg(errp, "can't allocate memory");
> >+    }
> >     g_free(path);
> > }
> >
> >diff --git a/exec.c b/exec.c
> >index 8705cc5..74560e5 100644
> >--- a/exec.c
> >+++ b/exec.c
> >@@ -1228,7 +1228,7 @@ static ram_addr_t ram_block_add(RAMBlock *new_block)
> >             if (!new_block->host) {
> >                 fprintf(stderr, "Cannot set up guest memory '%s': %s\n",
> >                         new_block->mr->name, strerror(errno));
> >-                exit(1);
> >+                return -1;
> >             }
> >             memory_try_enable_merging(new_block->host, new_block->length);
> >         }
> >@@ -1356,6 +1356,10 @@ void qemu_ram_free(ram_addr_t addr)
> > {
> >     RAMBlock *block;
> >
> >+    if (addr == -1) {
> >+        return;
> >+    }
> >+
> >     /* This assumes the iothread lock is taken here too.  */
> >     qemu_mutex_lock_ramlist();
> >     QTAILQ_FOREACH(block, &ram_list.blocks, next) {
> >