qemu-e2k/tests/ivshmem-test.c

517 lines
12 KiB
C
Raw Normal View History

/*
* QTest testcase for ivshmem
*
* Copyright (c) 2014 SUSE LINUX Products GmbH
* Copyright (c) 2015 Red Hat, Inc.
*
* This work is licensed under the terms of the GNU GPL, version 2 or later.
* See the COPYING file in the top-level directory.
*/
#include "qemu/osdep.h"
#include <glib/gstdio.h>
#include "contrib/ivshmem-server/ivshmem-server.h"
#include "libqos/pci-pc.h"
#include "libqtest.h"
#include "qemu-common.h"
#define TMPSHMSIZE (1 << 20)
static char *tmpshm;
static void *tmpshmem;
static char *tmpdir;
static char *tmpserver;
static void save_fn(QPCIDevice *dev, int devfn, void *data)
{
QPCIDevice **pdev = (QPCIDevice **) data;
*pdev = dev;
}
static QPCIDevice *get_device(QPCIBus *pcibus)
{
QPCIDevice *dev;
dev = NULL;
qpci_device_foreach(pcibus, 0x1af4, 0x1110, save_fn, &dev);
g_assert(dev != NULL);
return dev;
}
typedef struct _IVState {
QTestState *qtest;
void *reg_base, *mem_base;
QPCIBus *pcibus;
QPCIDevice *dev;
} IVState;
enum Reg {
INTRMASK = 0,
INTRSTATUS = 4,
IVPOSITION = 8,
DOORBELL = 12,
};
static const char* reg2str(enum Reg reg) {
switch (reg) {
case INTRMASK:
return "IntrMask";
case INTRSTATUS:
return "IntrStatus";
case IVPOSITION:
return "IVPosition";
case DOORBELL:
return "DoorBell";
default:
return NULL;
}
}
static inline unsigned in_reg(IVState *s, enum Reg reg)
{
const char *name = reg2str(reg);
QTestState *qtest = global_qtest;
unsigned res;
global_qtest = s->qtest;
res = qpci_io_readl(s->dev, s->reg_base + reg);
g_test_message("*%s -> %x\n", name, res);
global_qtest = qtest;
return res;
}
static inline void out_reg(IVState *s, enum Reg reg, unsigned v)
{
const char *name = reg2str(reg);
QTestState *qtest = global_qtest;
global_qtest = s->qtest;
g_test_message("%x -> *%s\n", v, name);
qpci_io_writel(s->dev, s->reg_base + reg, v);
global_qtest = qtest;
}
static void cleanup_vm(IVState *s)
{
g_free(s->dev);
qpci_free_pc(s->pcibus);
qtest_quit(s->qtest);
}
static void setup_vm_cmd(IVState *s, const char *cmd, bool msix)
{
uint64_t barsize;
s->qtest = qtest_start(cmd);
s->pcibus = qpci_init_pc();
s->dev = get_device(s->pcibus);
s->reg_base = qpci_iomap(s->dev, 0, &barsize);
g_assert_nonnull(s->reg_base);
g_assert_cmpuint(barsize, ==, 256);
if (msix) {
qpci_msix_enable(s->dev);
}
s->mem_base = qpci_iomap(s->dev, 2, &barsize);
g_assert_nonnull(s->mem_base);
g_assert_cmpuint(barsize, ==, TMPSHMSIZE);
qpci_device_enable(s->dev);
}
static void setup_vm(IVState *s)
{
ivshmem: Split ivshmem-plain, ivshmem-doorbell off ivshmem ivshmem can be configured with and without interrupt capability (a.k.a. "doorbell"). The two configurations have largely disjoint options, which makes for a confusing (and badly checked) user interface. Moreover, the device can't tell the guest whether its doorbell is enabled. Create two new device models ivshmem-plain and ivshmem-doorbell, and deprecate the old one. Changes from ivshmem: * PCI revision is 1 instead of 0. The new revision is fully backwards compatible for guests. Guests may elect to require at least revision 1 to make sure they're not exposed to the funny "no shared memory, yet" state. * Property "role" replaced by "master". role=master becomes master=on, role=peer becomes master=off. Default is off instead of auto. * Property "use64" is gone. The new devices always have 64 bit BARs. Changes from ivshmem to ivshmem-plain: * The Interrupt Pin register in PCI config space is zero (does not use an interrupt pin) instead of one (uses INTA). * Property "x-memdev" is renamed to "memdev". * Properties "shm" and "size" are gone. Use property "memdev" instead. * Property "msi" is gone. The new device can't have MSI-X capability. It can't interrupt anyway. * Properties "ioeventfd" and "vectors" are gone. They're meaningless without interrupts anyway. Changes from ivshmem to ivshmem-doorbell: * Property "msi" is gone. The new device always has MSI-X capability. * Property "ioeventfd" defaults to on instead of off. * Property "size" is gone. The new device can only map all the shared memory received from the server. Guests can easily find out whether the device is configured for interrupts by checking for MSI-X capability. Note: some code added in sub-optimal places to make the diff easier to review. The next commit will move it to more sensible places. Signed-off-by: Markus Armbruster <armbru@redhat.com> Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Message-Id: <1458066895-20632-37-git-send-email-armbru@redhat.com>
2016-03-15 19:34:51 +01:00
char *cmd = g_strdup_printf("-object memory-backend-file"
",id=mb1,size=1M,share,mem-path=/dev/shm%s"
" -device ivshmem-plain,memdev=mb1", tmpshm);
setup_vm_cmd(s, cmd, false);
g_free(cmd);
}
static void test_ivshmem_single(void)
{
IVState state, *s;
uint32_t data[1024];
int i;
setup_vm(&state);
s = &state;
/* initial state of readable registers */
g_assert_cmpuint(in_reg(s, INTRMASK), ==, 0);
g_assert_cmpuint(in_reg(s, INTRSTATUS), ==, 0);
g_assert_cmpuint(in_reg(s, IVPOSITION), ==, 0);
/* trigger interrupt via registers */
out_reg(s, INTRMASK, 0xffffffff);
g_assert_cmpuint(in_reg(s, INTRMASK), ==, 0xffffffff);
out_reg(s, INTRSTATUS, 1);
/* check interrupt status */
g_assert_cmpuint(in_reg(s, INTRSTATUS), ==, 1);
/* reading clears */
g_assert_cmpuint(in_reg(s, INTRSTATUS), ==, 0);
/* TODO intercept actual interrupt (needs qtest work) */
/* invalid register access */
out_reg(s, IVPOSITION, 1);
in_reg(s, DOORBELL);
/* ring the (non-functional) doorbell */
out_reg(s, DOORBELL, 8 << 16);
/* write shared memory */
for (i = 0; i < G_N_ELEMENTS(data); i++) {
data[i] = i;
}
qtest_memwrite(s->qtest, (uintptr_t)s->mem_base, data, sizeof(data));
/* verify write */
for (i = 0; i < G_N_ELEMENTS(data); i++) {
g_assert_cmpuint(((uint32_t *)tmpshmem)[i], ==, i);
}
/* read it back and verify read */
memset(data, 0, sizeof(data));
qtest_memread(s->qtest, (uintptr_t)s->mem_base, data, sizeof(data));
for (i = 0; i < G_N_ELEMENTS(data); i++) {
g_assert_cmpuint(data[i], ==, i);
}
cleanup_vm(s);
}
static void test_ivshmem_pair(void)
{
IVState state1, state2, *s1, *s2;
char *data;
int i;
setup_vm(&state1);
s1 = &state1;
setup_vm(&state2);
s2 = &state2;
data = g_malloc0(TMPSHMSIZE);
/* host write, guest 1 & 2 read */
memset(tmpshmem, 0x42, TMPSHMSIZE);
qtest_memread(s1->qtest, (uintptr_t)s1->mem_base, data, TMPSHMSIZE);
for (i = 0; i < TMPSHMSIZE; i++) {
g_assert_cmpuint(data[i], ==, 0x42);
}
qtest_memread(s2->qtest, (uintptr_t)s2->mem_base, data, TMPSHMSIZE);
for (i = 0; i < TMPSHMSIZE; i++) {
g_assert_cmpuint(data[i], ==, 0x42);
}
/* guest 1 write, guest 2 read */
memset(data, 0x43, TMPSHMSIZE);
qtest_memwrite(s1->qtest, (uintptr_t)s1->mem_base, data, TMPSHMSIZE);
memset(data, 0, TMPSHMSIZE);
qtest_memread(s2->qtest, (uintptr_t)s2->mem_base, data, TMPSHMSIZE);
for (i = 0; i < TMPSHMSIZE; i++) {
g_assert_cmpuint(data[i], ==, 0x43);
}
/* guest 2 write, guest 1 read */
memset(data, 0x44, TMPSHMSIZE);
qtest_memwrite(s2->qtest, (uintptr_t)s2->mem_base, data, TMPSHMSIZE);
memset(data, 0, TMPSHMSIZE);
qtest_memread(s1->qtest, (uintptr_t)s2->mem_base, data, TMPSHMSIZE);
for (i = 0; i < TMPSHMSIZE; i++) {
g_assert_cmpuint(data[i], ==, 0x44);
}
cleanup_vm(s1);
cleanup_vm(s2);
g_free(data);
}
typedef struct ServerThread {
GThread *thread;
IvshmemServer *server;
int pipe[2]; /* to handle quit */
} ServerThread;
static void *server_thread(void *data)
{
ServerThread *t = data;
IvshmemServer *server = t->server;
while (true) {
fd_set fds;
int maxfd, ret;
FD_ZERO(&fds);
FD_SET(t->pipe[0], &fds);
maxfd = t->pipe[0] + 1;
ivshmem_server_get_fds(server, &fds, &maxfd);
ret = select(maxfd, &fds, NULL, NULL, NULL);
if (ret < 0) {
if (errno == EINTR) {
continue;
}
g_critical("select error: %s\n", strerror(errno));
break;
}
if (ret == 0) {
continue;
}
if (FD_ISSET(t->pipe[0], &fds)) {
break;
}
if (ivshmem_server_handle_fds(server, &fds, maxfd) < 0) {
g_critical("ivshmem_server_handle_fds() failed\n");
break;
}
}
return NULL;
}
static void setup_vm_with_server(IVState *s, int nvectors, bool msi)
{
char *cmd = g_strdup_printf("-chardev socket,id=chr0,path=%s,nowait "
ivshmem: Split ivshmem-plain, ivshmem-doorbell off ivshmem ivshmem can be configured with and without interrupt capability (a.k.a. "doorbell"). The two configurations have largely disjoint options, which makes for a confusing (and badly checked) user interface. Moreover, the device can't tell the guest whether its doorbell is enabled. Create two new device models ivshmem-plain and ivshmem-doorbell, and deprecate the old one. Changes from ivshmem: * PCI revision is 1 instead of 0. The new revision is fully backwards compatible for guests. Guests may elect to require at least revision 1 to make sure they're not exposed to the funny "no shared memory, yet" state. * Property "role" replaced by "master". role=master becomes master=on, role=peer becomes master=off. Default is off instead of auto. * Property "use64" is gone. The new devices always have 64 bit BARs. Changes from ivshmem to ivshmem-plain: * The Interrupt Pin register in PCI config space is zero (does not use an interrupt pin) instead of one (uses INTA). * Property "x-memdev" is renamed to "memdev". * Properties "shm" and "size" are gone. Use property "memdev" instead. * Property "msi" is gone. The new device can't have MSI-X capability. It can't interrupt anyway. * Properties "ioeventfd" and "vectors" are gone. They're meaningless without interrupts anyway. Changes from ivshmem to ivshmem-doorbell: * Property "msi" is gone. The new device always has MSI-X capability. * Property "ioeventfd" defaults to on instead of off. * Property "size" is gone. The new device can only map all the shared memory received from the server. Guests can easily find out whether the device is configured for interrupts by checking for MSI-X capability. Note: some code added in sub-optimal places to make the diff easier to review. The next commit will move it to more sensible places. Signed-off-by: Markus Armbruster <armbru@redhat.com> Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Message-Id: <1458066895-20632-37-git-send-email-armbru@redhat.com>
2016-03-15 19:34:51 +01:00
"-device ivshmem%s,chardev=chr0,vectors=%d",
tmpserver,
msi ? "-doorbell" : ",size=1M,msi=off",
nvectors);
setup_vm_cmd(s, cmd, msi);
g_free(cmd);
}
static void test_ivshmem_server(bool msi)
{
IVState state1, state2, *s1, *s2;
ServerThread thread;
IvshmemServer server;
int ret, vm1, vm2;
int nvectors = 2;
guint64 end_time = g_get_monotonic_time() + 5 * G_TIME_SPAN_SECOND;
ret = ivshmem_server_init(&server, tmpserver, tmpshm, true,
TMPSHMSIZE, nvectors,
g_test_verbose());
g_assert_cmpint(ret, ==, 0);
ret = ivshmem_server_start(&server);
g_assert_cmpint(ret, ==, 0);
thread.server = &server;
ret = pipe(thread.pipe);
g_assert_cmpint(ret, ==, 0);
thread.thread = g_thread_new("ivshmem-server", server_thread, &thread);
g_assert(thread.thread != NULL);
ivshmem: Receive shared memory synchronously in realize() When configured for interrupts (property "chardev" given), we receive the shared memory from an ivshmem server. We do so asynchronously after realize() completes, by setting up callbacks with qemu_chr_add_handlers(). Keeping server I/O out of realize() that way avoids delays due to a slow server. This is probably relevant only for hot plug. However, this funny "no shared memory, yet" state of the device also causes a raft of issues that are hard or impossible to work around: * The guest is exposed to this state: when we enter and leave it its shared memory contents is apruptly replaced, and device register IVPosition changes. This is a known issue. We document that guests should not access the shared memory after device initialization until the IVPosition register becomes non-negative. For cold plug, the funny state is unlikely to be visible in practice, because we normally receive the shared memory long before the guest gets around to mess with the device. For hot plug, the timing is tighter, but the relative slowness of PCI device configuration has a good chance to hide the funny state. In either case, guests complying with the documented procedure are safe. * Migration becomes racy. If migration completes before the shared memory setup completes on the source, shared memory contents is silently lost. Fortunately, migration is rather unlikely to win this race. If the shared memory's ramblock arrives at the destination before shared memory setup completes, migration fails. There is no known way for a management application to wait for shared memory setup to complete. All you can do is retry failed migration. You can improve your chances by leaving more time between running the destination QEMU and the migrate command. To mitigate silent memory loss, you need to ensure the server initializes shared memory exactly the same on source and destination. These issues are entirely undocumented so far. I'd expect the server to be almost always fast enough to hide these issues. But then rare catastrophic races are in a way the worst kind. This is way more trouble than I'm willing to take from any device. Kill the funny state by receiving shared memory synchronously in realize(). If your hot plug hangs, go kill your ivshmem server. For easier review, this commit only makes the receive synchronous, it doesn't add the necessary error propagation. Without that, the funny state persists. The next commit will do that, and kill it off for real. Signed-off-by: Markus Armbruster <armbru@redhat.com> Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Message-Id: <1458066895-20632-26-git-send-email-armbru@redhat.com>
2016-03-15 19:34:40 +01:00
setup_vm_with_server(&state1, nvectors, msi);
s1 = &state1;
setup_vm_with_server(&state2, nvectors, msi);
s2 = &state2;
/* check got different VM ids */
vm1 = in_reg(s1, IVPOSITION);
vm2 = in_reg(s2, IVPOSITION);
ivshmem: Receive shared memory synchronously in realize() When configured for interrupts (property "chardev" given), we receive the shared memory from an ivshmem server. We do so asynchronously after realize() completes, by setting up callbacks with qemu_chr_add_handlers(). Keeping server I/O out of realize() that way avoids delays due to a slow server. This is probably relevant only for hot plug. However, this funny "no shared memory, yet" state of the device also causes a raft of issues that are hard or impossible to work around: * The guest is exposed to this state: when we enter and leave it its shared memory contents is apruptly replaced, and device register IVPosition changes. This is a known issue. We document that guests should not access the shared memory after device initialization until the IVPosition register becomes non-negative. For cold plug, the funny state is unlikely to be visible in practice, because we normally receive the shared memory long before the guest gets around to mess with the device. For hot plug, the timing is tighter, but the relative slowness of PCI device configuration has a good chance to hide the funny state. In either case, guests complying with the documented procedure are safe. * Migration becomes racy. If migration completes before the shared memory setup completes on the source, shared memory contents is silently lost. Fortunately, migration is rather unlikely to win this race. If the shared memory's ramblock arrives at the destination before shared memory setup completes, migration fails. There is no known way for a management application to wait for shared memory setup to complete. All you can do is retry failed migration. You can improve your chances by leaving more time between running the destination QEMU and the migrate command. To mitigate silent memory loss, you need to ensure the server initializes shared memory exactly the same on source and destination. These issues are entirely undocumented so far. I'd expect the server to be almost always fast enough to hide these issues. But then rare catastrophic races are in a way the worst kind. This is way more trouble than I'm willing to take from any device. Kill the funny state by receiving shared memory synchronously in realize(). If your hot plug hangs, go kill your ivshmem server. For easier review, this commit only makes the receive synchronous, it doesn't add the necessary error propagation. Without that, the funny state persists. The next commit will do that, and kill it off for real. Signed-off-by: Markus Armbruster <armbru@redhat.com> Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Message-Id: <1458066895-20632-26-git-send-email-armbru@redhat.com>
2016-03-15 19:34:40 +01:00
g_assert_cmpint(vm1, >=, 0);
g_assert_cmpint(vm2, >=, 0);
g_assert_cmpint(vm1, !=, vm2);
/* check number of MSI-X vectors */
global_qtest = s1->qtest;
if (msi) {
ret = qpci_msix_table_size(s1->dev);
g_assert_cmpuint(ret, ==, nvectors);
}
/* TODO test behavior before MSI-X is enabled */
/* ping vm2 -> vm1 on vector 0 */
if (msi) {
ret = qpci_msix_pending(s1->dev, 0);
g_assert_cmpuint(ret, ==, 0);
} else {
g_assert_cmpuint(in_reg(s1, INTRSTATUS), ==, 0);
}
out_reg(s2, DOORBELL, vm1 << 16);
do {
g_usleep(10000);
ret = msi ? qpci_msix_pending(s1->dev, 0) : in_reg(s1, INTRSTATUS);
} while (ret == 0 && g_get_monotonic_time() < end_time);
g_assert_cmpuint(ret, !=, 0);
/* ping vm1 -> vm2 on vector 1 */
global_qtest = s2->qtest;
if (msi) {
ret = qpci_msix_pending(s2->dev, 1);
g_assert_cmpuint(ret, ==, 0);
} else {
g_assert_cmpuint(in_reg(s2, INTRSTATUS), ==, 0);
}
out_reg(s1, DOORBELL, vm2 << 16 | 1);
do {
g_usleep(10000);
ret = msi ? qpci_msix_pending(s2->dev, 1) : in_reg(s2, INTRSTATUS);
} while (ret == 0 && g_get_monotonic_time() < end_time);
g_assert_cmpuint(ret, !=, 0);
cleanup_vm(s2);
cleanup_vm(s1);
if (qemu_write_full(thread.pipe[1], "q", 1) != 1) {
g_error("qemu_write_full: %s", g_strerror(errno));
}
g_thread_join(thread.thread);
ivshmem_server_close(&server);
close(thread.pipe[1]);
close(thread.pipe[0]);
}
static void test_ivshmem_server_msi(void)
{
test_ivshmem_server(true);
}
static void test_ivshmem_server_irq(void)
{
test_ivshmem_server(false);
}
#define PCI_SLOT_HP 0x06
static void test_ivshmem_hotplug(void)
{
gchar *opts;
qtest_start("");
opts = g_strdup_printf("'shm': '%s', 'size': '1M'", tmpshm);
qpci_plug_device_test("ivshmem", "iv1", PCI_SLOT_HP, opts);
qpci_unplug_acpi_device_test("iv1", PCI_SLOT_HP);
qtest_end();
g_free(opts);
}
static void test_ivshmem_memdev(void)
{
IVState state;
/* just for the sake of checking memory-backend property */
setup_vm_cmd(&state, "-object memory-backend-ram,size=1M,id=mb1"
ivshmem: Split ivshmem-plain, ivshmem-doorbell off ivshmem ivshmem can be configured with and without interrupt capability (a.k.a. "doorbell"). The two configurations have largely disjoint options, which makes for a confusing (and badly checked) user interface. Moreover, the device can't tell the guest whether its doorbell is enabled. Create two new device models ivshmem-plain and ivshmem-doorbell, and deprecate the old one. Changes from ivshmem: * PCI revision is 1 instead of 0. The new revision is fully backwards compatible for guests. Guests may elect to require at least revision 1 to make sure they're not exposed to the funny "no shared memory, yet" state. * Property "role" replaced by "master". role=master becomes master=on, role=peer becomes master=off. Default is off instead of auto. * Property "use64" is gone. The new devices always have 64 bit BARs. Changes from ivshmem to ivshmem-plain: * The Interrupt Pin register in PCI config space is zero (does not use an interrupt pin) instead of one (uses INTA). * Property "x-memdev" is renamed to "memdev". * Properties "shm" and "size" are gone. Use property "memdev" instead. * Property "msi" is gone. The new device can't have MSI-X capability. It can't interrupt anyway. * Properties "ioeventfd" and "vectors" are gone. They're meaningless without interrupts anyway. Changes from ivshmem to ivshmem-doorbell: * Property "msi" is gone. The new device always has MSI-X capability. * Property "ioeventfd" defaults to on instead of off. * Property "size" is gone. The new device can only map all the shared memory received from the server. Guests can easily find out whether the device is configured for interrupts by checking for MSI-X capability. Note: some code added in sub-optimal places to make the diff easier to review. The next commit will move it to more sensible places. Signed-off-by: Markus Armbruster <armbru@redhat.com> Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Message-Id: <1458066895-20632-37-git-send-email-armbru@redhat.com>
2016-03-15 19:34:51 +01:00
" -device ivshmem-plain,memdev=mb1", false);
cleanup_vm(&state);
}
static void cleanup(void)
{
if (tmpshmem) {
munmap(tmpshmem, TMPSHMSIZE);
tmpshmem = NULL;
}
if (tmpshm) {
shm_unlink(tmpshm);
g_free(tmpshm);
tmpshm = NULL;
}
if (tmpserver) {
g_unlink(tmpserver);
g_free(tmpserver);
tmpserver = NULL;
}
if (tmpdir) {
g_rmdir(tmpdir);
tmpdir = NULL;
}
}
static void abrt_handler(void *data)
{
cleanup();
}
static gchar *mktempshm(int size, int *fd)
{
while (true) {
gchar *name;
name = g_strdup_printf("/qtest-%u-%u", getpid(), g_random_int());
*fd = shm_open(name, O_CREAT|O_RDWR|O_EXCL,
S_IRWXU|S_IRWXG|S_IRWXO);
if (*fd > 0) {
g_assert(ftruncate(*fd, size) == 0);
return name;
}
g_free(name);
if (errno != EEXIST) {
perror("shm_open");
return NULL;
}
}
}
int main(int argc, char **argv)
{
int ret, fd;
gchar dir[] = "/tmp/ivshmem-test.XXXXXX";
#if !GLIB_CHECK_VERSION(2, 31, 0)
if (!g_thread_supported()) {
g_thread_init(NULL);
}
#endif
g_test_init(&argc, &argv, NULL);
qtest_add_abrt_handler(abrt_handler, NULL);
/* shm */
tmpshm = mktempshm(TMPSHMSIZE, &fd);
if (!tmpshm) {
return 0;
}
tmpshmem = mmap(0, TMPSHMSIZE, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0);
g_assert(tmpshmem != MAP_FAILED);
/* server */
if (mkdtemp(dir) == NULL) {
g_error("mkdtemp: %s", g_strerror(errno));
}
tmpdir = dir;
tmpserver = g_strconcat(tmpdir, "/server", NULL);
qtest_add_func("/ivshmem/single", test_ivshmem_single);
qtest_add_func("/ivshmem/hotplug", test_ivshmem_hotplug);
qtest_add_func("/ivshmem/memdev", test_ivshmem_memdev);
if (g_test_slow()) {
qtest_add_func("/ivshmem/pair", test_ivshmem_pair);
qtest_add_func("/ivshmem/server-msi", test_ivshmem_server_msi);
qtest_add_func("/ivshmem/server-irq", test_ivshmem_server_irq);
}
ret = g_test_run();
cleanup();
return ret;
}