Skip to contentSkip to conversation

hw/usb/ehci: Add 64-bit descriptor addressing support

PATCH Superseded Jamin Lin 5 participants 32 messages May 4 – Jul 13 Arm/Aspeed machine support

v6 of 7 Superseded Latest is v7 →

lore
Share & tools
lore
author
#1 lore raw

PATCH v6 1/11 tests/functional/arm/test_aspeed_ast2600_sdk: Add USB EHCI test for AST2600 SDK

In reply to Jamin Lin

Add a functional test to verify USB EHCI support on the AST2600 SDK machine by attaching a USB keyboard device and checking its enumeration via lsusb.

This introduces a helper routine that runs lsusb in the guest and validates that the emulated "QEMU USB Keyboard" is detected.

---
 tests/functional/arm/test_aspeed_ast2600_sdk.py | 7 +++++++
 1 file changed, 7 insertions(+)
tests/functional/arm/test_aspeed_ast2600_sdk.py +7 -0
index d787e90d10..01548dd135 100755
@@ -40,6 +40,11 @@ def do_ast2600_i3c_test(self):
             'i3ctransfer -d /dev/bus/i3c/5-1234567890ab -r 8 | grep 0x | xargs',
             '0x12 0x34 0x56 0x78 0x90 0xab 0xcd 0xef')
 
+    def do_ast2600_usb_ehci_test(self):
+        exec_command_and_wait_for_pattern(self,
+            'lsusb',
+            'QEMU QEMU USB Keyboard')
+
     def test_arm_ast2600_evb_sdk(self):
         self.set_machine('ast2600-evb')
         self.require_netdev('user')
@@ -54,6 +59,7 @@ def test_arm_ast2600_evb_sdk(self):
         self.vm.add_args('-netdev', 'user,id=net1')
         self.vm.add_args('-device',
             'mock-i3c-target,bus=dw.i3c.5,pid=0xab9078563412')
+        self.vm.add_args('-device', 'usb-kbd,bus=usb-bus.1')
         self.do_test_arm_aspeed_sdk_start(
             self.scratch_file("ast2600-default-image", "image-bmc"))
 
@@ -81,6 +87,7 @@ def test_arm_ast2600_evb_sdk(self):
              '/sbin/hwclock -f /dev/rtc1', year)
         self.do_ast2600_pcie_test()
         self.do_ast2600_i3c_test()
+        self.do_ast2600_usb_ehci_test()
 
 
 if __name__ == '__main__':
Signature · 1 line
2.43.0
author
#7 lore raw

PATCH v6 3/11 hw/usb/hcd-ehci: Add property to advertise 64-bit addressing capability

In reply to Jamin Lin

Introduce a new boolean property, "caps-64bit-addr", to control HCCPARAMS[0] (64-bit Addressing Capability).

When enabled, the EHCI controller advertises support for 64-bit address memory pointers as defined in the EHCI specification (Table 2-7, HCCPARAMS). This allows software to use the 64-bit data structure formats described in Appendix B.

When disabled (default), the controller reports 32-bit addressing capability and uses the standard 32-bit data structures.

The EHCI CTRLDSSEGMENT register provides the upper 32 bits [63:32] used to form 64-bit addresses for EHCI control data structures. Per EHCI 1.0 spec section 2.3.5, when the HCCPARAMS 64-bit Addressing Capability bit is zero, CTRLDSSEGMENT is not used: software cannot write it and reads must return zero.

Add a capability check in the operational register write handler and reject guest writes to CTRLDSSEGMENT when 64-bit addressing is not enabled.

---
 hw/usb/hcd-ehci.h |  5 ++++-
 hw/usb/hcd-ehci.c | 14 +++++++++++++-
 2 files changed, 17 insertions(+), 2 deletions(-)
hw/usb/hcd-ehci.h +4 -1
index 3acbddfc5c..f1f2fde578 100644
@@ -257,6 +257,7 @@ struct EHCIState {
     /* properties */
     uint32_t maxframes;
     bool migrate_fetch_addr_64bit;
+    bool caps_64bit_addr;
 
     /*
      *  EHCI spec version 1.0 Section 2.3
@@ -313,7 +314,9 @@ struct EHCIState {
 #define DEFINE_EHCI_COMMON_PROPERTIES(_state) \
     DEFINE_PROP_UINT32("maxframes", _state, ehci.maxframes, 128), \
     DEFINE_PROP_BOOL("x-migrate-fetch-addr-64bit", _state, \
-                     ehci.migrate_fetch_addr_64bit, true)
+                     ehci.migrate_fetch_addr_64bit, true), \
+    DEFINE_PROP_BOOL("caps-64bit-addr", _state, \
+                     ehci.caps_64bit_addr, false)
 
 extern const VMStateDescription vmstate_ehci;
 
hw/usb/hcd-ehci.c +13 -1
index 9d74259289..7166a93fd2 100644
@@ -1109,6 +1109,15 @@ static void ehci_opreg_write(void *ptr, hwaddr addr,
         }
         break;
 
+    case CTRLDSSEGMENT:
+        if (!s->caps_64bit_addr) {
+            qemu_log_mask(LOG_GUEST_ERROR,
+                          "ehci: write to CTRLDSSEGMENT while "
+                          "64-bit addressing capability is disabled\n");
+            return;
+        }
+        break;
+
     case ASYNCLISTADDR:
         if (ehci_async_enabled(s)) {
             qemu_log_mask(LOG_GUEST_ERROR,
@@ -2554,6 +2563,9 @@ void usb_ehci_realize(EHCIState *s, DeviceState *dev, Error **errp)
                    s->maxframes);
         return;
     }
+    if (s->caps_64bit_addr) {
+        s->caps[0x08] |= BIT(0);
+    }
 
     memory_region_add_subregion(&s->mem, s->capsbase, &s->mem_caps);
     memory_region_add_subregion(&s->mem, s->opregbase, &s->mem_opreg);
@@ -2613,7 +2625,7 @@ void usb_ehci_init(EHCIState *s, DeviceState *dev)
     s->caps[0x05] = 0x00;        /* No companion ports at present */
     s->caps[0x06] = 0x00;
     s->caps[0x07] = 0x00;
-    s->caps[0x08] = 0x80;        /* We can cache whole frame, no 64-bit */
+    s->caps[0x08] = 0x80;        /* We can cache whole frame */
     s->caps[0x0a] = 0x00;
     s->caps[0x0b] = 0x00;
 
Signature · 1 line
2.43.0
author
#2 lore raw

PATCH v6 2/11 hw/usb/hcd-ehci: Change descriptor addresses to 64-bit with migration compatibility

In reply to Jamin Lin

Change internal EHCI descriptor addresses from uint32_t to uint64_t.

The following fields are updated:
- EHCIPacket::qtdaddr
- EHCIQueue::{qhaddr, qtdaddr}
- EHCIState::{a_fetch_addr, p_fetch_addr}

Update get_dwords() and put_dwords() to take 64-bit addresses and propagate the type change through the descriptor traversal paths.

Adjust NLPTR_GET() to operate on 64-bit values:

    #define NLPTR_GET(x) ((x) & ~0x1fULL)

so that link pointer masking works correctly when descriptor addresses exceed 32-bit space. The previous mask (0xffffffe0) implicitly truncated addresses to 32 bits.

This patch does not change the on-wire descriptor layout yet. It only removes the internal 32-bit address limit and prepares for later patches that will add full 64-bit QH/qTD/iTD/siTD support.

Update the EHCI trace-events prototypes for QH, qTD, iTD, and siTD to use uint64_t for the address argument and print it with PRIx64. This ensures full 64-bit addresses are shown in trace output and improves debugging of queue heads and transfer descriptors.

Migration compatibility:

To preserve backward migration compatibility, keep the legacy 32-bit fetch address fields (a_fetch_addr_32, p_fetch_addr_32) alongside the new 64-bit fields.

Migration format is selected using a machine compat property "x-migrate-fetch-addr-64bit":

- Old machine types migrate 32-bit fetch addresses
- New machine types migrate full 64-bit fetch addresses

This is implemented using VMSTATE_UINT32_TEST() and VMSTATE_UINT64_TEST() so that only the appropriate format is migrated.

In pre_save, the 32-bit shadow fields are populated when migrating to old machine types. In post_load, the 32-bit values are restored into the 64-bit fields when loading old migration streams.

No functional change.

---
 hw/usb/hcd-ehci.h   | 17 +++++++----
 hw/core/machine.c   |  5 +++-
 hw/usb/hcd-ehci.c   | 72 ++++++++++++++++++++++++++++++++-------------
 hw/usb/trace-events | 24 +++++++--------
 4 files changed, 78 insertions(+), 40 deletions(-)
hw/usb/hcd-ehci.h +11 -6
index d038ee1e31..3acbddfc5c 100644
@@ -208,7 +208,7 @@ struct EHCIPacket {
     QTAILQ_ENTRY(EHCIPacket) next;
 
     EHCIqtd qtd;           /* copy of current QTD (being worked on) */
-    uint32_t qtdaddr;      /* address QTD read from                 */
+    uint64_t qtdaddr;      /* address QTD read from                 */
 
     USBPacket packet;
     QEMUSGList sgl;
@@ -229,8 +229,8 @@ struct EHCIQueue {
      * when guest removes an entry (doorbell, handshake sequence)
      */
     EHCIqh qh;             /* copy of current QH (being worked on) */
-    uint32_t qhaddr;       /* address QH read from                 */
-    uint32_t qtdaddr;      /* address QTD read from                */
+    uint64_t qhaddr;       /* address QH read from                 */
+    uint64_t qtdaddr;      /* address QTD read from                */
     int last_pid;          /* pid of last packet executed          */
     USBDevice *dev;
     QTAILQ_HEAD(, EHCIPacket) packets;
@@ -256,6 +256,7 @@ struct EHCIState {
 
     /* properties */
     uint32_t maxframes;
+    bool migrate_fetch_addr_64bit;
 
     /*
      *  EHCI spec version 1.0 Section 2.3
@@ -294,8 +295,10 @@ struct EHCIState {
     EHCIQueueHead pqueues;
 
     /* which address to look at next */
-    uint32_t a_fetch_addr;
-    uint32_t p_fetch_addr;
+    uint32_t a_fetch_addr_32;
+    uint32_t p_fetch_addr_32;
+    uint64_t a_fetch_addr;
+    uint64_t p_fetch_addr;
 
     USBPacket ipacket;
     QEMUSGList isgl;
@@ -308,7 +311,9 @@ struct EHCIState {
 };
 
 #define DEFINE_EHCI_COMMON_PROPERTIES(_state) \
-    DEFINE_PROP_UINT32("maxframes", _state, ehci.maxframes, 128)
+    DEFINE_PROP_UINT32("maxframes", _state, ehci.maxframes, 128), \
+    DEFINE_PROP_BOOL("x-migrate-fetch-addr-64bit", _state, \
+                     ehci.migrate_fetch_addr_64bit, true)
 
 extern const VMStateDescription vmstate_ehci;
 
hw/core/machine.c +4 -1
index 1b661fd36a..54336da294 100644
@@ -39,7 +39,10 @@
 #include "hw/acpi/generic_event_device.h"
 #include "qemu/audio.h"
 
-GlobalProperty hw_compat_11_0[] = {};
+GlobalProperty hw_compat_11_0[] = {
+    { "sysbus-ehci-usb", "x-migrate-fetch-addr-64bit", "off" },
+    { "pci-ehci-usb", "x-migrate-fetch-addr-64bit", "off" },
+};
 const size_t hw_compat_11_0_len = G_N_ELEMENTS(hw_compat_11_0);
 
 GlobalProperty hw_compat_10_2[] = {
hw/usb/hcd-ehci.c +51 -21

208 lines not on this page · raw message

hw/usb/trace-events +12 -12
index 0d4318dcf1..8c90688bb3 100644
@@ -86,15 +86,15 @@ usb_ehci_portsc_write(uint32_t addr, uint32_t port, uint32_t val) "wr mmio 0x%04
 usb_ehci_portsc_change(uint32_t addr, uint32_t port, uint32_t new, uint32_t old) "ch mmio 0x%04x [port %d] = 0x%x (old: 0x%x)"
 usb_ehci_usbsts(const char *sts, int state) "usbsts %s %d"
 usb_ehci_state(const char *schedule, const char *state) "%s schedule %s"
-usb_ehci_qh_ptrs(void *q, uint32_t addr, uint32_t nxt, uint32_t c_qtd, uint32_t n_qtd, uint32_t a_qtd) "q %p - QH @ 0x%08x: next 0x%08x qtds 0x%08x,0x%08x,0x%08x"
-usb_ehci_qh_fields(uint32_t addr, int rl, int mplen, int eps, int ep, int devaddr) "QH @ 0x%08x - rl %d, mplen %d, eps %d, ep %d, dev %d"
-usb_ehci_qh_bits(uint32_t addr, int c, int h, int dtc, int i) "QH @ 0x%08x - c %d, h %d, dtc %d, i %d"
+usb_ehci_qh_ptrs(void *q, uint64_t addr, uint32_t nxt, uint32_t c_qtd, uint32_t n_qtd, uint32_t a_qtd) "q %p - QH @ 0x%" PRIx64 ": next 0x%08x qtds 0x%08x,0x%08x,0x%08x"
+usb_ehci_qh_fields(uint64_t addr, int rl, int mplen, int eps, int ep, int devaddr) "QH @ 0x%" PRIx64 " - rl %d, mplen %d, eps %d, ep %d, dev %d"
+usb_ehci_qh_bits(uint64_t addr, int c, int h, int dtc, int i) "QH @ 0x%" PRIx64 " - c %d, h %d, dtc %d, i %d"
 usb_ehci_qh_tbytes(uint32_t tbytes) "updating tbytes to %d"
-usb_ehci_qtd_ptrs(void *q, uint32_t addr, uint32_t nxt, uint32_t altnext) "q %p - QTD @ 0x%08x: next 0x%08x altnext 0x%08x"
-usb_ehci_qtd_fields(uint32_t addr, int tbytes, int cpage, int cerr, int pid) "QTD @ 0x%08x - tbytes %d, cpage %d, cerr %d, pid %d"
-usb_ehci_qtd_bits(uint32_t addr, int ioc, int active, int halt, int babble, int xacterr) "QTD @ 0x%08x - ioc %d, active %d, halt %d, babble %d, xacterr %d"
-usb_ehci_itd(uint32_t addr, uint32_t nxt, uint32_t mplen, uint32_t mult, uint32_t ep, uint32_t devaddr) "ITD @ 0x%08x: next 0x%08x - mplen %d, mult %d, ep %d, dev %d"
-usb_ehci_sitd(uint32_t addr, uint32_t nxt, uint32_t active) "ITD @ 0x%08x: next 0x%08x - active %d"
+usb_ehci_qtd_ptrs(void *q, uint64_t addr, uint32_t nxt, uint32_t altnext) "q %p - QTD @ 0x%" PRIx64 ": next 0x%08x altnext 0x%08x"
+usb_ehci_qtd_fields(uint64_t addr, int tbytes, int cpage, int cerr, int pid) "QTD @ 0x%" PRIx64 " - tbytes %d, cpage %d, cerr %d, pid %d"
+usb_ehci_qtd_bits(uint64_t addr, int ioc, int active, int halt, int babble, int xacterr) "QTD @ 0x%" PRIx64 " - ioc %d, active %d, halt %d, babble %d, xacterr %d"
+usb_ehci_itd(uint64_t addr, uint32_t nxt, uint32_t mplen, uint32_t mult, uint32_t ep, uint32_t devaddr) "ITD @ 0x%" PRIx64 ": next 0x%08x - mplen %d, mult %d, ep %d, dev %d"
+usb_ehci_sitd(uint64_t addr, uint32_t nxt, uint32_t active) "SITD @ 0x%" PRIx64 ": next 0x%08x - active %d"
 usb_ehci_port_attach(uint32_t port, const char *owner, const char *device) "attach port #%d, owner %s, device %s"
 usb_ehci_port_detach(uint32_t port, const char *owner) "detach port #%d, owner %s"
 usb_ehci_port_reset(uint32_t port, int enable) "reset port #%d - %d"
@@ -104,15 +104,15 @@ usb_ehci_port_resume(uint32_t port) "port #%d"
 usb_ehci_port_disable(uint32_t port) "port #%d"
 usb_ehci_queue_action(void *q, const char *action) "q %p: %s"
 usb_ehci_packet_action(void *q, void *p, const char *action) "q %p p %p: %s"
-usb_ehci_packet_submit(uint32_t qhaddr, uint32_t next, uint32_t qtdaddr, int pid, size_t len, int endp, int status, int actual_length) "qh=0x%x, next=0x%x, qtd=0x%x, pid=0x%x, len=%zd, endp=0x%x, status=%d, actual_length=%d"
+usb_ehci_packet_submit(uint64_t qhaddr, uint32_t next, uint64_t qtdaddr, int pid, size_t len, int endp, int status, int actual_length) "qh=0x%" PRIx64 ", next=0x%x, qtd=0x%" PRIx64 ", pid=0x%x, len=%zd, endp=0x%x, status=%d, actual_length=%d"
 usb_ehci_irq(uint32_t level, uint32_t frindex, uint32_t sts, uint32_t mask) "level %d, frindex 0x%04x, sts 0x%x, mask 0x%x"
 usb_ehci_guest_bug(const char *reason) "%s"
 usb_ehci_doorbell_ring(void) ""
 usb_ehci_doorbell_ack(void) ""
 usb_ehci_dma_error(void) ""
-usb_ehci_execute_complete(uint32_t qhaddr, uint32_t next, uint32_t qtdaddr, int status, int actual_length) "qhaddr=0x%x, next=0x%x, qtdaddr=0x%x, status=%d, actual_length=%d"
-usb_ehci_fetchqh_reclaim_done(uint32_t qhaddr) "QH 0x%08x H-bit set, reclamation status reset - done processing"
-usb_ehci_fetchqh_dbg(uint32_t qhaddr, uint32_t h, uint32_t halt, uint32_t active, uint32_t next) "QH 0x%08x (h 0x%x halt 0x%x active 0x%x) next 0x%08x"
+usb_ehci_execute_complete(uint64_t qhaddr, uint32_t next, uint64_t qtdaddr, int status, int actual_length) "qhaddr=0x%" PRIx64 ", next=0x%x, qtdaddr=0x%" PRIx64 ", status=%d, actual_length=%d"
+usb_ehci_fetchqh_reclaim_done(uint64_t qhaddr) "QH 0x%" PRIx64 " H-bit set, reclamation status reset - done processing"
+usb_ehci_fetchqh_dbg(uint64_t qhaddr, uint32_t h, uint32_t halt, uint32_t active, uint32_t next) "QH 0x%" PRIx64 " (h 0x%x halt 0x%x active 0x%x) next 0x%08x"
 usb_ehci_periodic_state_advance(uint32_t frame, uint32_t list, uint32_t entry) "frame=%d, list=0x%x, entry=0x%x"
 usb_ehci_skipped_uframes(uint64_t skipped_uframes) "skipped %" PRIu64 " uframes"
 usb_ehci_log(const char *msg) "%s"
Signature · 1 line
2.43.0
author
#22 lore raw

PATCH v6 11/11 tests/functional/aarch64/test_aspeed_ast2700: Add USB EHCI test for AST2700 A1/A2

In reply to Jamin Lin

Add a functional test to verify USB EHCI support on the AST2700 A2/A1 by attaching a USB keyboard device and checking its enumeration via lsusb.

This introduces a helper routine that runs lsusb in the guest and validates that the emulated "QEMU USB Keyboard" is detected.

---
 tests/functional/aarch64/test_aspeed_ast2700a1.py | 7 +++++++
 tests/functional/aarch64/test_aspeed_ast2700a2.py | 7 +++++++
 2 files changed, 14 insertions(+)
tests/functional/aarch64/test_aspeed_ast2700a1.py +7 -0
index b0c08854da..42b8b200cf 100755
@@ -118,6 +118,11 @@ def do_ast2700_pcie_test(self):
             'ip addr show dev eth2',
             'inet 10.0.2.15/24')
 
+    def do_ast2700_usb_ehci_test(self):
+        exec_command_and_wait_for_pattern(self,
+            'lsusb',
+            'QEMU QEMU USB Keyboard')
+
     def start_ast2700_test(self, name, bus_id):
         num_cpu = 4
         load_images_list = [
@@ -161,10 +166,12 @@ def test_aarch64_ast2700a1_evb_sdk_v11_01(self):
         self.archive_extract(self.ASSET_SDK_V1101_AST2700A1)
         self.vm.add_args('-device', 'e1000e,netdev=net1,bus=pcie.2')
         self.vm.add_args('-netdev', 'user,id=net1')
+        self.vm.add_args('-device', 'usb-kbd,bus=usb-bus.3')
         self.start_ast2700_test('ast2700-a1-image', 1)
         self.verify_openbmc_boot_and_login('ast2700-a1')
         self.do_ast2700_i2c_test(1)
         self.do_ast2700_pcie_test()
+        self.do_ast2700_usb_ehci_test()
 
     def test_aarch64_ast2700a1_evb_sdk_vbootrom_v11_01(self):
         self.set_machine('ast2700a1-evb')
tests/functional/aarch64/test_aspeed_ast2700a2.py +7 -0
index ed414999f4..40e0ad81ed 100755
@@ -121,6 +121,11 @@ def do_ast2700_pcie_test(self):
             'ip addr show dev eth2',
             'inet 10.0.2.15/24')
 
+    def do_ast2700_usb_ehci_test(self):
+        exec_command_and_wait_for_pattern(self,
+            'lsusb',
+            'QEMU QEMU USB Keyboard')
+
     def start_ast2700_test(self, name, bus_id):
         num_cpu = 4
         load_images_list = [
@@ -164,10 +169,12 @@ def test_aarch64_ast2700a2_evb_sdk_v11_01(self):
         self.archive_extract(self.ASSET_SDK_V1101_AST2700A2)
         self.vm.add_args('-device', 'e1000e,netdev=net1,bus=pcie.2')
         self.vm.add_args('-netdev', 'user,id=net1')
+        self.vm.add_args('-device', 'usb-kbd,bus=usb-bus.3')
         self.start_ast2700_test('ast2700-default-image', 1)
         self.verify_openbmc_boot_and_login('ast2700-default')
         self.do_ast2700_i2c_test(1)
         self.do_ast2700_pcie_test()
+        self.do_ast2700_usb_ehci_test()
 
     def test_aarch64_ast2700a2_evb_sdk_vbootrom_v11_01(self):
         self.set_machine('ast2700a2-evb')
Signature · 1 line
2.43.0
author
#21 lore raw

PATCH v6 10/11 hw/arm/aspeed_ast27x0: Enable 64-bit EHCI DMA addressing

In reply to Jamin Lin

AST2700 supports a 64-bit DRAM address space. Therefore, DMA transactions must be capable of accessing 64-bit addresses.

Enable the "caps-64bit-addr" property for the EHCI controllers on AST2700 so that USB DMA operations can correctly handle 64-bit memory addresses.

---
 hw/arm/aspeed_ast27x0.c | 2 ++
 1 file changed, 2 insertions(+)
hw/arm/aspeed_ast27x0.c +2 -0
index ea527479e6..a8f0bc389c 100644
@@ -859,6 +859,8 @@ static void aspeed_soc_ast2700_realize(DeviceState *dev, Error **errp)
         object_property_set_int(OBJECT(&s->ehci[i]), "ctrldssegment-default",
                                 sc->memmap[ASPEED_DEV_SDRAM] >> 32,
                                 &error_abort);
+        object_property_set_bool(OBJECT(&s->ehci[i]), "caps-64bit-addr", true,
+                                 &error_abort);
         if (!sysbus_realize(SYS_BUS_DEVICE(&s->ehci[i]), errp)) {
             return;
         }
Signature · 1 line
2.43.0
author
#9 lore raw

PATCH v6 5/11 hw/usb/hcd-ehci: Implement 64-bit qTD descriptor addressing

In reply to Jamin Lin

EHCI supports 64-bit addressing when the 64-bit Addressing Capability bit in HCCPARAMS is set. In that mode, the CTRLDSSEGMENT register provides the upper 32 bits that are concatenated with 32-bit link pointer values to form 64-bit control data structure addresses (EHCI 1.0, section 2.3.5 and Appendix B).

qTD link pointers (current_qtd/next_qtd/altnext_qtd and qTD.next) are stored as 32-bit values in the data structures and must be expanded to full 64-bit descriptor addresses when 64-bit mode is enabled. Update the qTD traversal paths to use ehci_get_desc_addr() when following link pointers.

Appendix B also defines high dword fields for qTD buffer pointers. Add bufptr_hi[5] to EHCIqtd and extend qTD fetch and QH overlay handling to load and propagate the high buffer pointer fields.

When 64-bit capability is disabled, descriptor and buffer addresses remain 32-bit and existing behaviour is unchanged.

---
 hw/usb/hcd-ehci.h |  1 +
 hw/usb/hcd-ehci.c | 19 ++++++++++++-------
 2 files changed, 13 insertions(+), 7 deletions(-)
hw/usb/hcd-ehci.h +1 -0
index 3428839ec6..fc66aacd9f 100644
@@ -139,6 +139,7 @@ typedef struct EHCIqtd {
     uint32_t bufptr[5];               /* Standard buffer pointer */
 #define QTD_BUFPTR_MASK               0xfffff000
 #define QTD_BUFPTR_SH                 12
+    uint32_t bufptr_hi[5];
 } EHCIqtd;
 
 /* QH overlay: altnext_qtd, token, bufptr[5], bufptr_hi[5] */
hw/usb/hcd-ehci.c +12 -7
index e770f0a7fa..238f08b4a5 100644
@@ -472,7 +472,8 @@ static bool ehci_verify_qtd(EHCIPacket *p, EHCIqtd *qtd)
             (p->qtd.next != qtd->next)) ||
         (!NLPTR_TBIT(p->qtd.altnext) && (p->qtd.altnext != qtd->altnext)) ||
         p->qtd.token != qtd->token ||
-        p->qtd.bufptr[0] != qtd->bufptr[0]) {
+        p->qtd.bufptr[0] != qtd->bufptr[0] ||
+        p->qtd.bufptr_hi[0] != qtd->bufptr_hi[0]) {
         return false;
     } else {
         return true;
@@ -1200,6 +1201,7 @@ static int ehci_qh_do_overlay(EHCIQueue *q)
 
     for (i = 0; i < 5; i++) {
         q->qh.bufptr[i] = p->qtd.bufptr[i];
+        q->qh.bufptr_hi[i] = p->qtd.bufptr_hi[i];
     }
 
     if (!(q->qh.epchar & QH_EPCHAR_DTC)) {
@@ -1233,7 +1235,8 @@ static int ehci_init_transfer(EHCIPacket *p)
             return -1;
         }
 
-        page  = p->qtd.bufptr[cpage] & QTD_BUFPTR_MASK;
+        page = ehci_get_buf_addr(p->queue->ehci, p->qtd.bufptr_hi[cpage],
+                                 p->qtd.bufptr[cpage], QTD_BUFPTR_MASK);
         page += offset;
         plen  = bytes;
         if (plen > 4096 - offset) {
@@ -1727,7 +1730,7 @@ static EHCIQueue *ehci_state_fetchqh(EHCIState *ehci, int async)
     } else if ((q->qh.token & QTD_TOKEN_ACTIVE) &&
                (NLPTR_TBIT(q->qh.current_qtd) == 0) &&
                (q->qh.current_qtd != 0)) {
-        q->qtdaddr = q->qh.current_qtd;
+        q->qtdaddr = ehci_get_desc_addr(ehci, q->qh.current_qtd);
         ehci_set_state(ehci, async, EST_FETCHQTD);
 
     } else {
@@ -1805,14 +1808,14 @@ static int ehci_state_advqueue(EHCIQueue *q)
      */
     if (((q->qh.token & QTD_TOKEN_TBYTES_MASK) != 0) &&
         (NLPTR_TBIT(q->qh.altnext_qtd) == 0)) {
-        q->qtdaddr = q->qh.altnext_qtd;
+        q->qtdaddr = ehci_get_desc_addr(q->ehci, q->qh.altnext_qtd);
         ehci_set_state(q->ehci, q->async, EST_FETCHQTD);
 
     /*
      *  next qTD is valid
      */
     } else if (NLPTR_TBIT(q->qh.next_qtd) == 0) {
-        q->qtdaddr = q->qh.next_qtd;
+        q->qtdaddr = ehci_get_desc_addr(q->ehci, q->qh.next_qtd);
         ehci_set_state(q->ehci, q->async, EST_FETCHQTD);
 
     /*
@@ -1841,7 +1844,9 @@ static int ehci_state_fetchqtd(EHCIQueue *q)
     if (get_dwords(q->ehci, addr +  0, &qtd.next,    1) < 0 ||
         get_dwords(q->ehci, addr +  4, &qtd.altnext, 1) < 0 ||
         get_dwords(q->ehci, addr + 12, qtd.bufptr,
-                   ARRAY_SIZE(qtd.bufptr)) < 0) {
+                   ARRAY_SIZE(qtd.bufptr)) < 0 ||
+        get_dwords(q->ehci, addr + 32, qtd.bufptr_hi,
+                   ARRAY_SIZE(qtd.bufptr_hi)) < 0) {
         return 0;
     }
     ehci_trace_qtd(q, NLPTR_GET(q->qtdaddr), &qtd);
@@ -1922,7 +1927,7 @@ static int ehci_fill_queue(EHCIPacket *p)
         if (NLPTR_TBIT(qtd.next) != 0) {
             break;
         }
-        qtdaddr = qtd.next;
+        qtdaddr = ehci_get_desc_addr(q->ehci, qtd.next);
         /*
          * Detect circular td lists, Windows creates these, counting on the
          * active bit going low after execution to make the queue stop.
Signature · 1 line
2.43.0
author
#8 lore raw

PATCH v6 4/11 hw/usb/hcd-ehci: Implement 64-bit QH descriptor addressing

In reply to Jamin Lin

EHCI supports 64-bit control data structure addressing when the 64-bit Addressing Capability bit in HCCPARAMS is set. In that mode, the CTRLDSSEGMENT register supplies the upper 32 bits which are concatenated with 32-bit link pointer fields to form full 64-bit descriptor addresses (EHCI 1.0, section 2.3.5 and Appendix B).

The current implementation assumes 32-bit QH descriptor addresses and directly uses link pointer values without applying the CTRLDSSEGMENT upper dword.

Introduce a helper, ehci_get_desc_addr(), to construct full 64-bit descriptor addresses when 64-bit capability is enabled. Update QH traversal paths (async list walk, horizontal QH link, and periodic schedule entry handling) to use the translated 64-bit addresses.

EHCI 64-bit buffer pointer fields are defined in Appendix B as split 32-bit low/high parts located at separate offsets, rather than a single contiguous 64-bit field. Therefore, the buffer pointers cannot be represented as uint64_t bufptr[5] without violating the descriptor layout defined by the specification.

Introduce ehci_get_buf_addr() to construct full 64-bit buffer addresses from bufptr[] and bufptr_hi[] fields. Use this helper when calculating transfer buffer addresses so that data buffers above 4GB are correctly handled.

Also add bufptr_hi[5] to EHCIqh to support 64-bit buffer pointer fields as defined in Appendix B.

When 64-bit capability is disabled, descriptor addresses remain 32-bit and existing behaviour is unchanged.

Note: Similar split 64-bit buffer pointer handling is required for qTD, iTD and siTD descriptors, which will be addressed in follow-up changes.

---
 hw/usb/hcd-ehci.h   |  4 ++++
 hw/usb/hcd-ehci.c   | 47 ++++++++++++++++++++++++++++++++++-----------
 hw/usb/trace-events |  2 +-
 3 files changed, 41 insertions(+), 12 deletions(-)
hw/usb/hcd-ehci.h +4 -0
index f1f2fde578..3428839ec6 100644
@@ -141,6 +141,9 @@ typedef struct EHCIqtd {
 #define QTD_BUFPTR_SH                 12
 } EHCIqtd;
 
+/* QH overlay: altnext_qtd, token, bufptr[5], bufptr_hi[5] */
+#define EHCI_QH_OVERLAY_COUNT 12
+
 /*
  * EHCI spec version 1.0 Section 3.6
  */
@@ -194,6 +197,7 @@ typedef struct EHCIqh {
 #define BUFPTR_FRAMETAG_MASK          0x0000001f
 #define BUFPTR_SBYTES_MASK            0x00000fe0
 #define BUFPTR_SBYTES_SH              5
+    uint32_t bufptr_hi[5];
 } EHCIqh;
 
 enum async_state {
hw/usb/hcd-ehci.c +36 -11

99 lines not on this page · raw message

hw/usb/trace-events +1 -1
index 8c90688bb3..67249d69c2 100644
@@ -113,7 +113,7 @@ usb_ehci_dma_error(void) ""
 usb_ehci_execute_complete(uint64_t qhaddr, uint32_t next, uint64_t qtdaddr, int status, int actual_length) "qhaddr=0x%" PRIx64 ", next=0x%x, qtdaddr=0x%" PRIx64 ", status=%d, actual_length=%d"
 usb_ehci_fetchqh_reclaim_done(uint64_t qhaddr) "QH 0x%" PRIx64 " H-bit set, reclamation status reset - done processing"
 usb_ehci_fetchqh_dbg(uint64_t qhaddr, uint32_t h, uint32_t halt, uint32_t active, uint32_t next) "QH 0x%" PRIx64 " (h 0x%x halt 0x%x active 0x%x) next 0x%08x"
-usb_ehci_periodic_state_advance(uint32_t frame, uint32_t list, uint32_t entry) "frame=%d, list=0x%x, entry=0x%x"
+usb_ehci_periodic_state_advance(uint32_t frame, uint64_t list, uint64_t entry) "frame=%d, list=0x%" PRIx64 ", entry=0x%" PRIx64
 usb_ehci_skipped_uframes(uint64_t skipped_uframes) "skipped %" PRIu64 " uframes"
 usb_ehci_log(const char *msg) "%s"
 
Signature · 1 line
2.43.0
author
#11 lore raw

PATCH v6 7/11 hw/usb/hcd-ehci: Implement 64-bit siTD descriptor addressing

In reply to Jamin Lin

EHCI supports 64-bit control data structure addressing when the 64-bit Addressing Capability bit in HCCPARAMS is set. In that mode, the CTRLDSSEGMENT register provides the upper 32 bits that are concatenated with 32-bit link pointer values to form full 64-bit descriptor addresses (EHCI 1.0, section 2.3.5 and Appendix B).

siTD link pointers are stored as 32-bit values and must be expanded to full 64-bit descriptor addresses when 64-bit mode is enabled. Update the siTD traversal path to use ehci_get_desc_addr() when following link pointers.

When 64-bit capability is disabled, descriptor addresses remain 32-bit and existing behaviour is unchanged.

---
 hw/usb/hcd-ehci.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
hw/usb/hcd-ehci.c +1 -1
index 611882ba1d..b1c7bfd90c 100644
@@ -1792,7 +1792,7 @@ static int ehci_state_fetchsitd(EHCIState *ehci, int async)
         warn_report("Skipping active siTD");
     }
 
-    ehci_set_fetch_addr(ehci, async, sitd.next);
+    ehci_set_fetch_addr(ehci, async, ehci_get_desc_addr(ehci, sitd.next));
     ehci_set_state(ehci, async, EST_FETCHENTRY);
     return 1;
 }
Signature · 1 line
2.43.0

Jamin Lin

author
#0 lore raw

v6 cover hw/usb/ehci: Add 64-bit descriptor addressing support

EHCI supports 64-bit addressing through the CTRLDSSEGMENT register, which provides the upper 32 bits of descriptor addresses when the controller advertises 64-bit capability.

Currently QEMU EHCI model only partially supports this functionality and descriptor addresses are effectively treated as 32-bit. This becomes problematic on systems where system memory is located above the 4GB boundary.

The Linux EHCI driver enables 64-bit addressing if the controller advertises the capability. During initialization it programs the segment register to zero:

https://github.com/torvalds/linux/blob/master/drivers/usb/host/ehci-hcd.c#L600

The driver also notes that descriptor structures allocated from the DMA pool use segment zero semantics. Descriptor memory is allocated using the DMA API and platforms may configure a 64-bit DMA mask, allowing descriptor memory to be placed above 4GB.

On AST2700 platforms, system DRAM is mapped at 0x400000000. As a result, descriptor addresses constructed directly from the EHCI registers do not match the actual system addresses used by the controller when accessing queue heads (QH) and queue element transfer descriptors (qTD).

This patch series implements full 64-bit descriptor addressing support in the EHCI emulation. Descriptor address handling is updated to use 64-bit values and the descriptor structures (QH, qTD, iTD and siTD) are extended to support the upper address bits provided by the segment register.

Add a ctrldssegment-default property so platforms can provide a descriptor address offset when constructing descriptor addresses. This allows systems where DRAM resides above 4GB to access EHCI descriptors correctly.

The AST2700 machine uses this property to account for its DRAM mapping at 0x400000000 and enables 64-bit EHCI DMA addressing.

Test Result:
1. EHCI 32bits with ast2600-evb machine
Command line:
./build/qemu-system-arm \
-machine ast2600-evb \
-m 1G \
-drive file=image-bmc,if=mtd,format=raw \
-nographic \
-device usb-kbd,bus=usb-bus.1,id=mykbd \
-drive id=usbdisk,if=none,file=image0.ext4,format=raw \
-device usb-storage,bus=usb-bus.1,id=mystorage,drive=usbdisk
-snapshot \
-nographic
Result:
unable to initialize usb specBus 001 Device 001: ID 1d6b:0002 Linux 6.18.3-v00.08.01-g172b7e27a30d ehci_hcd EHCI Host Controller Bus 001 Device 002: ID 0627:0001 QEMU QEMU USB Keyboard
Bus 001 Device 003: ID 46f4:0001 QEMU QEMU USB HARDDRIVE
Bus 002 Device 001: ID 1d6b:0001 Linux 6.18.3-v00.08.01-g172b7e27a30d uhci_hcd Generic UHCI Host Controller

2. EHCI 64bits with ast2700a2-evb machine
Command line:
./build/qemu-system-aarch64 -M ast2700a2-evb -nographic\
-bios ast27x0_bootrom.bin \
-drive file=image-bmc,format=raw,if=mtd \
-snapshot \
-device usb-kbd,bus=usb-bus.3,id=mykbd \
-drive id=usbdisk,if=none,file=image0.ext4,format=raw \
-device usb-storage,bus=usb-bus.3,id=mystorage,drive=usbdisk
Result:

root@ast2700-default:~# lsusb
unable to initialize usb specBus 001 Device 001: ID 1d6b:0001 Linux 6.18.3-v00.08.01-g172b7e27a30d uhci_hcd Generic UHCI Host Controller
Bus 002 Device 001: ID 1d6b:0002 Linux 6.18.3-v00.08.01-g172b7e27a30d ehci_hcd EHCI Host Controller
Bus 002 Device 002: ID 0627:0001 QEMU QEMU USB Keyboard
Bus 002 Device 003: ID 46f4:0001 QEMU QEMU USB HARDDRIVE

v1
1. Fix checkpatch coding style issues
2. Implement 64-bit addressing for QH/qTD/iTD/siTD descriptors
3. Add descriptor address offset property
4. Enable 64-bit EHCI DMA addressing on AST2700
5. Configure descriptor address offset for AST2700

v2
1. Remove unused EHCIfstn structure and dead code
2. Replace fprintf(stderr, ...) with qemu_log_mask(LOG_GUEST_ERROR)
3. Replace DPRINTF debug logs with trace events
4. Add functional tests for USB EHCI on AST2600 and AST2700 A1/A2
5. Fix review issue

v3:
1. Add Migration version test function
2. Add EHCI 64-bit buffer pointer fields description in commit log

v4:
1. Reorder patches in the series
2. Fix EHCI migration issues
3. Introduce a common properties macro for both sysbus and PCI
4. Drop the descriptor address offset property
5. Add ctrldssegment-default property
6. Address review comments

v5:
1. Add 11.0 machine compatibility properties

v6:
1. Update reviewer suggested improvements.

Jamin Lin (11):
tests/functional/arm/test_aspeed_ast2600_sdk: Add USB EHCI test for

    AST2600 SDK
  hw/usb/hcd-ehci: Change descriptor addresses to 64-bit with migration
    compatibility
  hw/usb/hcd-ehci: Add property to advertise 64-bit addressing
    capability
  hw/usb/hcd-ehci: Implement 64-bit QH descriptor addressing
  hw/usb/hcd-ehci: Implement 64-bit qTD descriptor addressing
  hw/usb/hcd-ehci: Implement 64-bit iTD descriptor addressing
  hw/usb/hcd-ehci: Implement 64-bit siTD descriptor addressing
  hw/usb/hcd-ehci: Add ctrldssegment-default property
  hw/arm/aspeed_ast27x0: Set EHCI ctrldssegment-default
  hw/arm/aspeed_ast27x0: Enable 64-bit EHCI DMA addressing
  tests/functional/aarch64/test_aspeed_ast2700: Add USB EHCI test for
    AST2700 A1/A2
 hw/usb/hcd-ehci.h                             |  29 +++-
 hw/arm/aspeed_ast27x0.c                       |   5 +
 hw/core/machine.c                             |   5 +-
 hw/usb/hcd-ehci.c                             | 162 +++++++++++++-----
 hw/usb/trace-events                           |  26 +--
 .../aarch64/test_aspeed_ast2700a1.py          |   7 +
 .../aarch64/test_aspeed_ast2700a2.py          |   7 +
 .../functional/arm/test_aspeed_ast2600_sdk.py |   7 +
 8 files changed, 185 insertions(+), 63 deletions(-)
Signature · 1 line
2.43.0
author
#12 lore raw

PATCH v6 8/11 hw/usb/hcd-ehci: Add ctrldssegment-default property

When 64-bit addressing is supported, the Linux EHCI driver programs the segment register to zero. See ehci_run function:
https://github.com/torvalds/linux/blob/master/drivers/usb/host/ehci-hcd.c

The driver comment also notes that descriptor structures allocated from the DMA pool use segment zero semantics.

Descriptor memory is allocated using the DMA API. The platform driver configures a 64-bit DMA mask so memory can be allocated above 4GB. See ehci_platform_probe function:
https://github.com/torvalds/linux/blob/master/drivers/usb/host/ehci-platform.c

On AST2700 platforms, system DRAM is mapped above 4GB at 0x400000000. As a result, descriptor addresses constructed directly from the guest EHCI registers do not match the actual system address used by the controller when fetching queue heads (QH) and queue element transfer descriptors (qTD).

Add a ctrldssegment-default property so platforms can provide a descriptor address offset when constructing descriptor addresses. This allows systems where DRAM resides above 4GB to access EHCI descriptors correctly.

The default value is zero, so existing machines are not affected.

---
 hw/usb/hcd-ehci.h | 5 ++++-
 hw/usb/hcd-ehci.c | 1 +
 2 files changed, 5 insertions(+), 1 deletion(-)
hw/usb/hcd-ehci.h +4 -1
index 8e6a8cdfb0..3d57e1de5f 100644
@@ -264,6 +264,7 @@ struct EHCIState {
     uint32_t maxframes;
     bool migrate_fetch_addr_64bit;
     bool caps_64bit_addr;
+    uint32_t ctrldssegment_default;
 
     /*
      *  EHCI spec version 1.0 Section 2.3
@@ -322,7 +323,9 @@ struct EHCIState {
     DEFINE_PROP_BOOL("x-migrate-fetch-addr-64bit", _state, \
                      ehci.migrate_fetch_addr_64bit, true), \
     DEFINE_PROP_BOOL("caps-64bit-addr", _state, \
-                     ehci.caps_64bit_addr, false)
+                     ehci.caps_64bit_addr, false), \
+    DEFINE_PROP_UINT32("ctrldssegment-default", _state, \
+                       ehci.ctrldssegment_default, 0)
 
 extern const VMStateDescription vmstate_ehci;
 
hw/usb/hcd-ehci.c +1 -0
index b1c7bfd90c..62c3770ac5 100644
@@ -1134,6 +1134,7 @@ static void ehci_opreg_write(void *ptr, hwaddr addr,
                           "64-bit addressing capability is disabled\n");
             return;
         }
+        val |= s->ctrldssegment_default;
         break;
 
     case ASYNCLISTADDR:
Signature · 1 line
2.43.0
author
#19 lore raw

PATCH v6 9/11 hw/arm/aspeed_ast27x0: Set EHCI ctrldssegment-default

In reply to Jamin Lin

On AST2700 platforms, system DRAM is mapped above 4GB with a base address at 0x400000000.

The Linux EHCI driver programs the segment register to zero when 64-bit addressing is supported. As a result, descriptor addresses derived from the EHCI registers do not include the DRAM base address.

Descriptor memory is allocated through the DMA API with a 64-bit DMA mask, allowing descriptors to reside in DRAM above 4GB. On AST2700, EHCI queue heads (QH) and queue element transfer descriptors (qTD) are therefore placed at addresses starting from 0x400000000.

Set the ctrldssegment-default property to "sc->memmap[ASPEED_DEV_SDRAM] >> 32" so the upper 32 bits of descriptor addresses are adjusted accordingly. This allows the emulated EHCI controller to construct correct system addresses when accessing descriptors in DRAM above 4GB.

---
 hw/arm/aspeed_ast27x0.c | 3 +++
 1 file changed, 3 insertions(+)
hw/arm/aspeed_ast27x0.c +3 -0
index 87dcb82e1b..ea527479e6 100644
@@ -856,6 +856,9 @@ static void aspeed_soc_ast2700_realize(DeviceState *dev, Error **errp)
 
     /* EHCI */
     for (i = 0; i < sc->ehcis_num; i++) {
+        object_property_set_int(OBJECT(&s->ehci[i]), "ctrldssegment-default",
+                                sc->memmap[ASPEED_DEV_SDRAM] >> 32,
+                                &error_abort);
         if (!sysbus_realize(SYS_BUS_DEVICE(&s->ehci[i]), errp)) {
             return;
         }
Signature · 1 line
2.43.0
author
#10 lore raw

PATCH v6 6/11 hw/usb/hcd-ehci: Implement 64-bit iTD descriptor addressing

In reply to Jamin Lin

EHCI supports 64-bit control data structure addressing when the 64-bit Addressing Capability bit in HCCPARAMS is set. In that mode, the CTRLDSSEGMENT register provides the upper 32 bits that are concatenated with 32-bit link pointer values to form full 64-bit descriptor addresses (EHCI 1.0, section 2.3.5 and Appendix B).

iTD link pointers are stored as 32-bit values and must be expanded to full 64-bit descriptor addresses when 64-bit mode is enabled. Update the iTD traversal path to use ehci_get_desc_addr() when following link pointers.

Appendix B also defines high dword fields for iTD buffer pointers. Add bufptr_hi[7] to EHCIitd and use ehci_get_buf_addr() to construct full 64-bit buffer addresses from bufptr[] and bufptr_hi[] fields when processing isochronous transfers. This allows buffers above 4GB to be handled correctly.

When 64-bit capability is disabled, descriptor and buffer addresses remain 32-bit and existing behaviour is unchanged.

---
 hw/usb/hcd-ehci.h | 1 +
 hw/usb/hcd-ehci.c | 9 ++++++---
 2 files changed, 7 insertions(+), 3 deletions(-)
hw/usb/hcd-ehci.h +1 -0
index fc66aacd9f..8e6a8cdfb0 100644
@@ -63,6 +63,7 @@ typedef struct EHCIitd {
 #define ITD_BUFPTR_MAXPKT_SH     0
 #define ITD_BUFPTR_MULT_MASK     0x00000003
 #define ITD_BUFPTR_MULT_SH       0
+    uint32_t bufptr_hi[7];
 } EHCIitd;
 
 /*
hw/usb/hcd-ehci.c +6 -3
index 238f08b4a5..611882ba1d 100644
@@ -1464,7 +1464,8 @@ static int ehci_process_itd(EHCIState *ehci,
                 return -1;
             }
 
-            ptr1 = (itd->bufptr[pg] & ITD_BUFPTR_MASK);
+            ptr1 = ehci_get_buf_addr(ehci, itd->bufptr_hi[pg],
+                                     itd->bufptr[pg], ITD_BUFPTR_MASK);
             qemu_sglist_init(&ehci->isgl, ehci->device, 2, ehci->as);
             if (off + len > 4096) {
                 /* transfer crosses page border */
@@ -1472,7 +1473,9 @@ static int ehci_process_itd(EHCIState *ehci,
                     qemu_sglist_destroy(&ehci->isgl);
                     return -1;  /* avoid page pg + 1 */
                 }
-                ptr2 = (itd->bufptr[pg + 1] & ITD_BUFPTR_MASK);
+                ptr2 = ehci_get_buf_addr(ehci, itd->bufptr_hi[pg + 1],
+                                         itd->bufptr[pg + 1],
+                                         ITD_BUFPTR_MASK);
                 uint32_t len2 = off + len - 4096;
                 uint32_t len1 = len - len2;
                 qemu_sglist_add(&ehci->isgl, ptr1 + off, len1);
@@ -1762,7 +1765,7 @@ static int ehci_state_fetchitd(EHCIState *ehci, int async)
 
     put_dwords(ehci, NLPTR_GET(entry), (uint32_t *) &itd,
                sizeof(EHCIitd) >> 2);
-    ehci_set_fetch_addr(ehci, async, itd.next);
+    ehci_set_fetch_addr(ehci, async, ehci_get_desc_addr(ehci, itd.next));
     ehci_set_state(ehci, async, EST_FETCHENTRY);
 
     return 1;
Signature · 1 line
2.43.0
Reviewed-by on v6 2/11
#3 lore raw

Re: [PATCH v6 02/11] hw/usb/hcd-ehci: Change descriptor addresses to 64-bit with migration compatibility

In reply to Jamin Lin
Quoted 239 lines · Jamin Lin

Change internal EHCI descriptor addresses from uint32_t to uint64_t.

The following fields are updated:
- EHCIPacket::qtdaddr
- EHCIQueue::{qhaddr, qtdaddr}
- EHCIState::{a_fetch_addr, p_fetch_addr}

Update get_dwords() and put_dwords() to take 64-bit addresses and propagate the type change through the descriptor traversal paths.

Adjust NLPTR_GET() to operate on 64-bit values:

     #define NLPTR_GET(x) ((x) & ~0x1fULL)

so that link pointer masking works correctly when descriptor addresses exceed 32-bit space. The previous mask (0xffffffe0) implicitly truncated addresses to 32 bits.

This patch does not change the on-wire descriptor layout yet. It only removes the internal 32-bit address limit and prepares for later patches that will add full 64-bit QH/qTD/iTD/siTD support.

Update the EHCI trace-events prototypes for QH, qTD, iTD, and siTD to use uint64_t for the address argument and print it with PRIx64. This ensures full 64-bit addresses are shown in trace output and improves debugging of queue heads and transfer descriptors.

Migration compatibility:

To preserve backward migration compatibility, keep the legacy 32-bit fetch address fields (a_fetch_addr_32, p_fetch_addr_32) alongside the new 64-bit fields.

Migration format is selected using a machine compat property "x-migrate-fetch-addr-64bit":

- Old machine types migrate 32-bit fetch addresses
- New machine types migrate full 64-bit fetch addresses

This is implemented using VMSTATE_UINT32_TEST() and VMSTATE_UINT64_TEST() so that only the appropriate format is migrated.

In pre_save, the 32-bit shadow fields are populated when migrating to old machine types. In post_load, the 32-bit values are restored into the 64-bit fields when loading old migration streams.

No functional change.

---
  hw/usb/hcd-ehci.h   | 17 +++++++----
  hw/core/machine.c   |  5 +++-
  hw/usb/hcd-ehci.c   | 72 ++++++++++++++++++++++++++++++++-------------
  hw/usb/trace-events | 24 +++++++--------
  4 files changed, 78 insertions(+), 40 deletions(-)
hw/usb/hcd-ehci.h +11 -6
index d038ee1e31..3acbddfc5c 100644
@@ -208,7 +208,7 @@ struct EHCIPacket {
      QTAILQ_ENTRY(EHCIPacket) next;
 
      EHCIqtd qtd;           /* copy of current QTD (being worked on) */
-    uint32_t qtdaddr;      /* address QTD read from                 */
+    uint64_t qtdaddr;      /* address QTD read from                 */
 
      USBPacket packet;
      QEMUSGList sgl;
@@ -229,8 +229,8 @@ struct EHCIQueue {
       * when guest removes an entry (doorbell, handshake sequence)
       */
      EHCIqh qh;             /* copy of current QH (being worked on) */
-    uint32_t qhaddr;       /* address QH read from                 */
-    uint32_t qtdaddr;      /* address QTD read from                */
+    uint64_t qhaddr;       /* address QH read from                 */
+    uint64_t qtdaddr;      /* address QTD read from                */
      int last_pid;          /* pid of last packet executed          */
      USBDevice *dev;
      QTAILQ_HEAD(, EHCIPacket) packets;
@@ -256,6 +256,7 @@ struct EHCIState {
 
      /* properties */
      uint32_t maxframes;
+    bool migrate_fetch_addr_64bit;
 
      /*
       *  EHCI spec version 1.0 Section 2.3
@@ -294,8 +295,10 @@ struct EHCIState {
      EHCIQueueHead pqueues;
 
      /* which address to look at next */
-    uint32_t a_fetch_addr;
-    uint32_t p_fetch_addr;
+    uint32_t a_fetch_addr_32;
+    uint32_t p_fetch_addr_32;
+    uint64_t a_fetch_addr;
+    uint64_t p_fetch_addr;
 
      USBPacket ipacket;
      QEMUSGList isgl;
@@ -308,7 +311,9 @@ struct EHCIState {
  };
 
  #define DEFINE_EHCI_COMMON_PROPERTIES(_state) \
-    DEFINE_PROP_UINT32("maxframes", _state, ehci.maxframes, 128)
+    DEFINE_PROP_UINT32("maxframes", _state, ehci.maxframes, 128), \
+    DEFINE_PROP_BOOL("x-migrate-fetch-addr-64bit", _state, \
+                     ehci.migrate_fetch_addr_64bit, true)
 
  extern const VMStateDescription vmstate_ehci;
 
hw/core/machine.c +4 -1
index 1b661fd36a..54336da294 100644
@@ -39,7 +39,10 @@
  #include "hw/acpi/generic_event_device.h"
  #include "qemu/audio.h"
 
-GlobalProperty hw_compat_11_0[] = {};
+GlobalProperty hw_compat_11_0[] = {
+    { "sysbus-ehci-usb", "x-migrate-fetch-addr-64bit", "off" },
+    { "pci-ehci-usb", "x-migrate-fetch-addr-64bit", "off" },
+};
  const size_t hw_compat_11_0_len = G_N_ELEMENTS(hw_compat_11_0);
 
  GlobalProperty hw_compat_10_2[] = {
hw/usb/hcd-ehci.c +13 -12
index 28a60e4c1a..9d74259289 100644
@@ -72,7 +72,7 @@ typedef enum {
  } EHCI_STATES;
 
  /* macros for accessing fields within next link pointer entry */
-#define NLPTR_GET(x)             ((x) & 0xffffffe0)
+#define NLPTR_GET(x)             ((x) & ~0x1fULL)
  #define NLPTR_TYPE_GET(x)        (((x) >> 1) & 3)
  #define NLPTR_TBIT(x)            ((x) & 1)  /* 1=invalid, 0=valid */
 
@@ -287,7 +287,7 @@ static int ehci_get_state(EHCIState *s, int async)
      return async ? s->astate : s->pstate;
  }
 
-static void ehci_set_fetch_addr(EHCIState *s, int async, uint32_t addr)
+static void ehci_set_fetch_addr(EHCIState *s, int async, uint64_t addr)
  {
      if (async) {
          s->a_fetch_addr = addr;
@@ -296,7 +296,7 @@ static void ehci_set_fetch_addr(EHCIState *s, int async, uint32_t addr)
      }
  }
 
-static int ehci_get_fetch_addr(EHCIState *s, int async)
+static uint64_t ehci_get_fetch_addr(EHCIState *s, int async)
  {
      return async ? s->a_fetch_addr : s->p_fetch_addr;
  }
@@ -373,7 +373,7 @@ static inline bool ehci_periodic_enabled(EHCIState *s)
  }
 
  /* Get an array of dwords from main memory */
-static inline int get_dwords(EHCIState *ehci, uint32_t addr,
+static inline int get_dwords(EHCIState *ehci, uint64_t addr,
                               uint32_t *buf, int num)
  {
      int i;
@@ -395,7 +395,7 @@ static inline int get_dwords(EHCIState *ehci, uint32_t addr,
  }
 
  /* Put an array of dwords in to main memory */
-static inline int put_dwords(EHCIState *ehci, uint32_t addr,
+static inline int put_dwords(EHCIState *ehci, uint64_t addr,
                               uint32_t *buf, int num)
  {
      int i;
@@ -549,7 +549,7 @@ static void ehci_free_packet(EHCIPacket *p)
 
  /* queue management */
 
-static EHCIQueue *ehci_alloc_queue(EHCIState *ehci, uint32_t addr, int async)
+static EHCIQueue *ehci_alloc_queue(EHCIState *ehci, uint64_t addr, int async)
  {
      EHCIQueueHead *head = async ? &ehci->aqueues : &ehci->pqueues;
      EHCIQueue *q;
@@ -622,7 +622,7 @@ static void ehci_free_queue(EHCIQueue *q, const char *warn)
      g_free(q);
  }
 
-static EHCIQueue *ehci_find_queue_by_qh(EHCIState *ehci, uint32_t addr,
+static EHCIQueue *ehci_find_queue_by_qh(EHCIState *ehci, uint64_t addr,
                                          int async)
  {
      EHCIQueueHead *head = async ? &ehci->aqueues : &ehci->pqueues;
@@ -1135,7 +1135,7 @@ static void ehci_flush_qh(EHCIQueue *q)
  {
      uint32_t *qh = (uint32_t *) &q->qh;
      uint32_t dwords = sizeof(EHCIqh) >> 2;
-    uint32_t addr = NLPTR_GET(q->qhaddr);
+    uint64_t addr = NLPTR_GET(q->qhaddr);
 
      put_dwords(q->ehci, addr + 3 * sizeof(uint32_t), qh + 3, dwords - 3);
  }
@@ -1406,12 +1406,13 @@ static int ehci_execute(EHCIPacket *p, const char *action)
  /* 4.7.2 */
  static int ehci_process_itd(EHCIState *ehci,
                              EHCIitd *itd,
-                            uint32_t addr)
+                            uint64_t addr)
  {
      USBDevice *dev;
      USBEndpoint *ep;
      uint32_t i, len, pid, dir, devaddr, endp;
-    uint32_t pg, off, ptr1, ptr2, max, mult;
+    uint32_t pg, off, max, mult;
+    uint64_t ptr1, ptr2;
 
      ehci->periodic_sched_active = PERIODIC_ACTIVE;
 
@@ -1528,7 +1529,7 @@ static int ehci_state_waitlisthead(EHCIState *ehci,  int async)
      EHCIqh qh;
      int i = 0;
      int again = 0;
-    uint32_t entry = ehci->asynclistaddr;
+    uint64_t entry = ehci->asynclistaddr;
 
      /* set reclamation flag at start event (4.8.6) */
      if (async) {
@@ -1578,7 +1579,7 @@ out:
  static int ehci_state_fetchentry(EHCIState *ehci, int async)
  {
      int again = 0;
-    uint32_t entry = ehci_get_fetch_addr(ehci, async);
+    uint64_t entry = ehci_get_fetch_addr(ehci, async);
 
      if (NLPTR_TBIT(entry)) {
          ehci_set_state(ehci, async, EST_ACTIVE);

hw/usb/hcd-ehci.c static int ehci_state_fetchentry(EHCIState *ehci, int async) · line 1615

Quoted 5 lines
hw/usb/hcd-ehci.c +1 -1
@@ -1611,7 +1612,7 @@ static int ehci_state_fetchentry(EHCIState *ehci, int async)
      default:
          /* TODO: handle FSTN type */
          qemu_log_mask(LOG_GUEST_ERROR,
-                      "FETCHENTRY: entry at 0x%x is of type %u "
+                      "FETCHENTRY: entry at %" PRIx64 "is of type %" PRIu64 " "
                                                          ^
                                                     missing space.

I will amend the patch when applying.

Quoted 156 lines
hw/usb/hcd-ehci.c +37 -8
                        "which is not supported yet\n",
                        entry, NLPTR_TYPE_GET(entry));
          return -1;
@@ -1623,7 +1624,7 @@ out:
 
  static EHCIQueue *ehci_state_fetchqh(EHCIState *ehci, int async)
  {
-    uint32_t entry;
+    uint64_t entry;
      EHCIQueue *q;
      EHCIqh qh;
 
@@ -1712,7 +1713,7 @@ out:
 
  static int ehci_state_fetchitd(EHCIState *ehci, int async)
  {
-    uint32_t entry;
+    uint64_t entry;
      EHCIitd itd;
 
      assert(!async);
@@ -1738,7 +1739,7 @@ static int ehci_state_fetchitd(EHCIState *ehci, int async)
 
  static int ehci_state_fetchsitd(EHCIState *ehci, int async)
  {
-    uint32_t entry;
+    uint64_t entry;
      EHCIsitd sitd;
 
      assert(!async);
@@ -1802,7 +1803,7 @@ static int ehci_state_fetchqtd(EHCIQueue *q)
      EHCIqtd qtd;
      EHCIPacket *p;
      int again = 1;
-    uint32_t addr;
+    uint64_t addr;
 
      addr = NLPTR_GET(q->qtdaddr);
      if (get_dwords(q->ehci, addr +  8, &qtd.token,   1) < 0) {
@@ -1885,7 +1886,7 @@ static int ehci_fill_queue(EHCIPacket *p)
      USBEndpoint *ep = p->packet.ep;
      EHCIQueue *q = p->queue;
      EHCIqtd qtd = p->qtd;
-    uint32_t qtdaddr;
+    uint64_t qtdaddr;
 
      for (;;) {
          if (NLPTR_TBIT(qtd.next) != 0) {
@@ -2008,7 +2009,8 @@ static int ehci_state_executing(EHCIQueue *q)
  static int ehci_state_writeback(EHCIQueue *q)
  {
      EHCIPacket *p = QTAILQ_FIRST(&q->packets);
-    uint32_t *qtd, addr;
+    uint32_t *qtd;
+    uint64_t addr;
      int again = 0;
 
      /*  Write back the QTD from the QH area */
@@ -2414,6 +2416,18 @@ static USBBusOps ehci_bus_ops_standalone = {
      .wakeup_endpoint = ehci_wakeup_endpoint,
  };
 
+static bool ehci_fetch_addr_64_needed(void *opaque, int version_id)
+{
+    EHCIState *s = opaque;
+
+    return s->migrate_fetch_addr_64bit;
+}
+
+static bool ehci_fetch_addr_32_needed(void *opaque, int version_id)
+{
+    return !ehci_fetch_addr_64_needed(opaque, version_id);
+}
+
  static int usb_ehci_pre_save(void *opaque)
  {
      EHCIState *ehci = opaque;
@@ -2424,6 +2438,11 @@ static int usb_ehci_pre_save(void *opaque)
      ehci->last_run_ns -= (ehci->frindex - new_frindex) * UFRAME_TIMER_NS;
      ehci->frindex = new_frindex;
 
+    if (!ehci->migrate_fetch_addr_64bit) {
+        ehci->a_fetch_addr_32 = ehci->a_fetch_addr;
+        ehci->p_fetch_addr_32 = ehci->p_fetch_addr;
+    }
+
      return 0;
  }
 
@@ -2444,6 +2463,11 @@ static int usb_ehci_post_load(void *opaque, int version_id)
          }
      }
 
+    if (!s->migrate_fetch_addr_64bit) {
+        s->a_fetch_addr = s->a_fetch_addr_32;
+        s->p_fetch_addr = s->p_fetch_addr_32;
+    }
+
      return 0;
  }
 
@@ -2504,8 +2528,14 @@ const VMStateDescription vmstate_ehci = {
          /* schedule state */
          VMSTATE_UINT32(astate, EHCIState),
          VMSTATE_UINT32(pstate, EHCIState),
-        VMSTATE_UINT32(a_fetch_addr, EHCIState),
-        VMSTATE_UINT32(p_fetch_addr, EHCIState),
+        VMSTATE_UINT32_TEST(a_fetch_addr_32, EHCIState,
+                            ehci_fetch_addr_32_needed),
+        VMSTATE_UINT32_TEST(p_fetch_addr_32, EHCIState,
+                            ehci_fetch_addr_32_needed),
+        VMSTATE_UINT64_TEST(a_fetch_addr, EHCIState,
+                            ehci_fetch_addr_64_needed),
+        VMSTATE_UINT64_TEST(p_fetch_addr, EHCIState,
+                            ehci_fetch_addr_64_needed),
          VMSTATE_END_OF_LIST()
      }
  };
hw/usb/trace-events +9 -9
index 0d4318dcf1..8c90688bb3 100644
@@ -86,15 +86,15 @@ usb_ehci_portsc_write(uint32_t addr, uint32_t port, uint32_t val) "wr mmio 0x%04
  usb_ehci_portsc_change(uint32_t addr, uint32_t port, uint32_t new, uint32_t old) "ch mmio 0x%04x [port %d] = 0x%x (old: 0x%x)"
  usb_ehci_usbsts(const char *sts, int state) "usbsts %s %d"
  usb_ehci_state(const char *schedule, const char *state) "%s schedule %s"
-usb_ehci_qh_ptrs(void *q, uint32_t addr, uint32_t nxt, uint32_t c_qtd, uint32_t n_qtd, uint32_t a_qtd) "q %p - QH @ 0x%08x: next 0x%08x qtds 0x%08x,0x%08x,0x%08x"
-usb_ehci_qh_fields(uint32_t addr, int rl, int mplen, int eps, int ep, int devaddr) "QH @ 0x%08x - rl %d, mplen %d, eps %d, ep %d, dev %d"
-usb_ehci_qh_bits(uint32_t addr, int c, int h, int dtc, int i) "QH @ 0x%08x - c %d, h %d, dtc %d, i %d"
+usb_ehci_qh_ptrs(void *q, uint64_t addr, uint32_t nxt, uint32_t c_qtd, uint32_t n_qtd, uint32_t a_qtd) "q %p - QH @ 0x%" PRIx64 ": next 0x%08x qtds 0x%08x,0x%08x,0x%08x"
+usb_ehci_qh_fields(uint64_t addr, int rl, int mplen, int eps, int ep, int devaddr) "QH @ 0x%" PRIx64 " - rl %d, mplen %d, eps %d, ep %d, dev %d"
+usb_ehci_qh_bits(uint64_t addr, int c, int h, int dtc, int i) "QH @ 0x%" PRIx64 " - c %d, h %d, dtc %d, i %d"
  usb_ehci_qh_tbytes(uint32_t tbytes) "updating tbytes to %d"
-usb_ehci_qtd_ptrs(void *q, uint32_t addr, uint32_t nxt, uint32_t altnext) "q %p - QTD @ 0x%08x: next 0x%08x altnext 0x%08x"
-usb_ehci_qtd_fields(uint32_t addr, int tbytes, int cpage, int cerr, int pid) "QTD @ 0x%08x - tbytes %d, cpage %d, cerr %d, pid %d"
-usb_ehci_qtd_bits(uint32_t addr, int ioc, int active, int halt, int babble, int xacterr) "QTD @ 0x%08x - ioc %d, active %d, halt %d, babble %d, xacterr %d"
-usb_ehci_itd(uint32_t addr, uint32_t nxt, uint32_t mplen, uint32_t mult, uint32_t ep, uint32_t devaddr) "ITD @ 0x%08x: next 0x%08x - mplen %d, mult %d, ep %d, dev %d"
-usb_ehci_sitd(uint32_t addr, uint32_t nxt, uint32_t active) "ITD @ 0x%08x: next 0x%08x - active %d"
+usb_ehci_qtd_ptrs(void *q, uint64_t addr, uint32_t nxt, uint32_t altnext) "q %p - QTD @ 0x%" PRIx64 ": next 0x%08x altnext 0x%08x"
+usb_ehci_qtd_fields(uint64_t addr, int tbytes, int cpage, int cerr, int pid) "QTD @ 0x%" PRIx64 " - tbytes %d, cpage %d, cerr %d, pid %d"
+usb_ehci_qtd_bits(uint64_t addr, int ioc, int active, int halt, int babble, int xacterr) "QTD @ 0x%" PRIx64 " - ioc %d, active %d, halt %d, babble %d, xacterr %d"
+usb_ehci_itd(uint64_t addr, uint32_t nxt, uint32_t mplen, uint32_t mult, uint32_t ep, uint32_t devaddr) "ITD @ 0x%" PRIx64 ": next 0x%08x - mplen %d, mult %d, ep %d, dev %d"
+usb_ehci_sitd(uint64_t addr, uint32_t nxt, uint32_t active) "SITD @ 0x%" PRIx64 ": next 0x%08x - active %d"
  usb_ehci_port_attach(uint32_t port, const char *owner, const char *device) "attach port #%d, owner %s, device %s"
  usb_ehci_port_detach(uint32_t port, const char *owner) "detach port #%d, owner %s"
  usb_ehci_port_reset(uint32_t port, int enable) "reset port #%d - %d"
@@ -104,15 +104,15 @@ usb_ehci_port_resume(uint32_t port) "port #%d"
  usb_ehci_port_disable(uint32_t port) "port #%d"
  usb_ehci_queue_action(void *q, const char *action) "q %p: %s"
  usb_ehci_packet_action(void *q, void *p, const char *action) "q %p p %p: %s"
-usb_ehci_packet_submit(uint32_t qhaddr, uint32_t next, uint32_t qtdaddr, int pid, size_t len, int endp, int status, int actual_length) "qh=0x%x, next=0x%x, qtd=0x%x, pid=0x%x, len=%zd, endp=0x%x, status=%d, actual_length=%d"
+usb_ehci_packet_submit(uint64_t qhaddr, uint32_t next, uint64_t qtdaddr, int pid, size_t len, int endp, int status, int actual_length) "qh=0x%" PRIx64 ", next=0x%x, qtd=0x%" PRIx64 ", pid=0x%x, len=%zd, endp=0x%x, status=%d, actual_length=%d"
  usb_ehci_irq(uint32_t level, uint32_t frindex, uint32_t sts, uint32_t mask) "level %d, frindex 0x%04x, sts 0x%x, mask 0x%x"
  usb_ehci_guest_bug(const char *reason) "%s"
  usb_ehci_doorbell_ring(void) ""
  usb_ehci_doorbell_ack(void) ""

hw/usb/trace-events usb_ehci_port_resume(uint32_t port) "port #%d" · line 118

Quoted 10 lines
hw/usb/trace-events +3 -3
  usb_ehci_dma_error(void) ""
-usb_ehci_execute_complete(uint32_t qhaddr, uint32_t next, uint32_t qtdaddr, int status, int actual_length) "qhaddr=0x%x, next=0x%x, qtdaddr=0x%x, status=%d, actual_length=%d"
-usb_ehci_fetchqh_reclaim_done(uint32_t qhaddr) "QH 0x%08x H-bit set, reclamation status reset - done processing"
-usb_ehci_fetchqh_dbg(uint32_t qhaddr, uint32_t h, uint32_t halt, uint32_t active, uint32_t next) "QH 0x%08x (h 0x%x halt 0x%x active 0x%x) next 0x%08x"
+usb_ehci_execute_complete(uint64_t qhaddr, uint32_t next, uint64_t qtdaddr, int status, int actual_length) "qhaddr=0x%" PRIx64 ", next=0x%x, qtdaddr=0x%" PRIx64 ", status=%d, actual_length=%d"
+usb_ehci_fetchqh_reclaim_done(uint64_t qhaddr) "QH 0x%" PRIx64 " H-bit set, reclamation status reset - done processing"
+usb_ehci_fetchqh_dbg(uint64_t qhaddr, uint32_t h, uint32_t halt, uint32_t active, uint32_t next) "QH 0x%" PRIx64 " (h 0x%x halt 0x%x active 0x%x) next 0x%08x"
  usb_ehci_periodic_state_advance(uint32_t frame, uint32_t list, uint32_t entry) "frame=%d, list=0x%x, entry=0x%x"
  usb_ehci_skipped_uframes(uint64_t skipped_uframes) "skipped %" PRIu64 " uframes"
  usb_ehci_log(const char *msg) "%s"

Thanks,

C.

#23 lore raw
In reply to Jamin Lin
Quoted 124 lines · Jamin Lin

EHCI supports 64-bit addressing through the CTRLDSSEGMENT register, which provides the upper 32 bits of descriptor addresses when the controller advertises 64-bit capability.

Currently QEMU EHCI model only partially supports this functionality and descriptor addresses are effectively treated as 32-bit. This becomes problematic on systems where system memory is located above the 4GB boundary.

The Linux EHCI driver enables 64-bit addressing if the controller advertises the capability. During initialization it programs the segment register to zero:

https://github.com/torvalds/linux/blob/master/drivers/usb/host/ehci-hcd.c#L600

The driver also notes that descriptor structures allocated from the DMA pool use segment zero semantics. Descriptor memory is allocated using the DMA API and platforms may configure a 64-bit DMA mask, allowing descriptor memory to be placed above 4GB.

On AST2700 platforms, system DRAM is mapped at 0x400000000. As a result, descriptor addresses constructed directly from the EHCI registers do not match the actual system addresses used by the controller when accessing queue heads (QH) and queue element transfer descriptors (qTD).

This patch series implements full 64-bit descriptor addressing support in the EHCI emulation. Descriptor address handling is updated to use 64-bit values and the descriptor structures (QH, qTD, iTD and siTD) are extended to support the upper address bits provided by the segment register.

Add a ctrldssegment-default property so platforms can provide a descriptor address offset when constructing descriptor addresses. This allows systems where DRAM resides above 4GB to access EHCI descriptors correctly.

The AST2700 machine uses this property to account for its DRAM mapping at 0x400000000 and enables 64-bit EHCI DMA addressing.

Test Result:
1. EHCI 32bits with ast2600-evb machine
Command line:

./build/qemu-system-arm \
   -machine ast2600-evb \
   -m 1G \
   -drive file=image-bmc,if=mtd,format=raw \
   -nographic \
   -device usb-kbd,bus=usb-bus.1,id=mykbd \
   -drive id=usbdisk,if=none,file=image0.ext4,format=raw \
   -device usb-storage,bus=usb-bus.1,id=mystorage,drive=usbdisk
   -snapshot \
   -nographic

Result:
unable to initialize usb specBus 001 Device 001: ID 1d6b:0002 Linux 6.18.3-v00.08.01-g172b7e27a30d ehci_hcd EHCI Host Controller Bus 001 Device 002: ID 0627:0001 QEMU QEMU USB Keyboard
Bus 001 Device 003: ID 46f4:0001 QEMU QEMU USB HARDDRIVE
Bus 002 Device 001: ID 1d6b:0001 Linux 6.18.3-v00.08.01-g172b7e27a30d uhci_hcd Generic UHCI Host Controller

2. EHCI 64bits with ast2700a2-evb machine
Command line:
./build/qemu-system-aarch64 -M ast2700a2-evb -nographic\
-bios ast27x0_bootrom.bin \
-drive file=image-bmc,format=raw,if=mtd \
-snapshot \
-device usb-kbd,bus=usb-bus.3,id=mykbd \
-drive id=usbdisk,if=none,file=image0.ext4,format=raw \
-device usb-storage,bus=usb-bus.3,id=mystorage,drive=usbdisk
Result:

root@ast2700-default:~# lsusb
unable to initialize usb specBus 001 Device 001: ID 1d6b:0001 Linux 6.18.3-v00.08.01-g172b7e27a30d uhci_hcd Generic UHCI Host Controller
Bus 002 Device 001: ID 1d6b:0002 Linux 6.18.3-v00.08.01-g172b7e27a30d ehci_hcd EHCI Host Controller
Bus 002 Device 002: ID 0627:0001 QEMU QEMU USB Keyboard
Bus 002 Device 003: ID 46f4:0001 QEMU QEMU USB HARDDRIVE

v1
1. Fix checkpatch coding style issues
2. Implement 64-bit addressing for QH/qTD/iTD/siTD descriptors
3. Add descriptor address offset property
4. Enable 64-bit EHCI DMA addressing on AST2700
5. Configure descriptor address offset for AST2700

v2
1. Remove unused EHCIfstn structure and dead code
2. Replace fprintf(stderr, ...) with qemu_log_mask(LOG_GUEST_ERROR)
3. Replace DPRINTF debug logs with trace events
4. Add functional tests for USB EHCI on AST2600 and AST2700 A1/A2
5. Fix review issue

v3:
1. Add Migration version test function
2. Add EHCI 64-bit buffer pointer fields description in commit log

v4:
1. Reorder patches in the series
2. Fix EHCI migration issues
3. Introduce a common properties macro for both sysbus and PCI
4. Drop the descriptor address offset property
5. Add ctrldssegment-default property
6. Address review comments

v5:
1. Add 11.0 machine compatibility properties

v6:
1. Update reviewer suggested improvements.

Jamin Lin (11):
tests/functional/arm/test_aspeed_ast2600_sdk: Add USB EHCI test for

     AST2600 SDK
   hw/usb/hcd-ehci: Change descriptor addresses to 64-bit with migration
     compatibility
   hw/usb/hcd-ehci: Add property to advertise 64-bit addressing
     capability
   hw/usb/hcd-ehci: Implement 64-bit QH descriptor addressing
   hw/usb/hcd-ehci: Implement 64-bit qTD descriptor addressing
   hw/usb/hcd-ehci: Implement 64-bit iTD descriptor addressing
   hw/usb/hcd-ehci: Implement 64-bit siTD descriptor addressing
   hw/usb/hcd-ehci: Add ctrldssegment-default property
   hw/arm/aspeed_ast27x0: Set EHCI ctrldssegment-default
   hw/arm/aspeed_ast27x0: Enable 64-bit EHCI DMA addressing
   tests/functional/aarch64/test_aspeed_ast2700: Add USB EHCI test for
     AST2700 A1/A2
Quoted 10 lines
  hw/usb/hcd-ehci.h                             |  29 +++-
  hw/arm/aspeed_ast27x0.c                       |   5 +
  hw/core/machine.c                             |   5 +-
  hw/usb/hcd-ehci.c                             | 162 +++++++++++++-----
  hw/usb/trace-events                           |  26 +--
  .../aarch64/test_aspeed_ast2700a1.py          |   7 +
  .../aarch64/test_aspeed_ast2700a2.py          |   7 +
  .../functional/arm/test_aspeed_ast2600_sdk.py |   7 +
  8 files changed, 185 insertions(+), 63 deletions(-)

Applied to

https://github.com/legoater/qemu aspeed-next

Please base future contributions on this branch.

Thanks,

C.

Reviewed-by on v6 8/11
#13 lore raw

Re: [PATCH v6 08/11] hw/usb/hcd-ehci: Add ctrldssegment-default property

In reply to Jamin Lin
Quoted 45 lines · Jamin Lin

When 64-bit addressing is supported, the Linux EHCI driver programs the segment register to zero. See ehci_run function:
https://github.com/torvalds/linux/blob/master/drivers/usb/host/ehci-hcd.c

The driver comment also notes that descriptor structures allocated from the DMA pool use segment zero semantics.

Descriptor memory is allocated using the DMA API. The platform driver configures a 64-bit DMA mask so memory can be allocated above 4GB. See ehci_platform_probe function:
https://github.com/torvalds/linux/blob/master/drivers/usb/host/ehci-platform.c

On AST2700 platforms, system DRAM is mapped above 4GB at 0x400000000. As a result, descriptor addresses constructed directly from the guest EHCI registers do not match the actual system address used by the controller when fetching queue heads (QH) and queue element transfer descriptors (qTD).

Add a ctrldssegment-default property so platforms can provide a descriptor address offset when constructing descriptor addresses. This allows systems where DRAM resides above 4GB to access EHCI descriptors correctly.

The default value is zero, so existing machines are not affected.

---
  hw/usb/hcd-ehci.h | 5 ++++-
  hw/usb/hcd-ehci.c | 1 +
  2 files changed, 5 insertions(+), 1 deletion(-)
hw/usb/hcd-ehci.h +1 -0
index 8e6a8cdfb0..3d57e1de5f 100644
@@ -264,6 +264,7 @@ struct EHCIState {
      uint32_t maxframes;
      bool migrate_fetch_addr_64bit;
      bool caps_64bit_addr;
+    uint32_t ctrldssegment_default;
 
      /*
       *  EHCI spec version 1.0 Section 2.3

hw/usb/hcd-ehci.h struct EHCIState { · line 328

Quoted 7 lines
hw/usb/hcd-ehci.h +3 -1
@@ -322,7 +323,9 @@ struct EHCIState {
      DEFINE_PROP_BOOL("x-migrate-fetch-addr-64bit", _state, \
                       ehci.migrate_fetch_addr_64bit, true), \
      DEFINE_PROP_BOOL("caps-64bit-addr", _state, \
-                     ehci.caps_64bit_addr, false)
+                     ehci.caps_64bit_addr, false), \
+    DEFINE_PROP_UINT32("ctrldssegment-default", _state, \
+                       ehci.ctrldssegment_default, 0)

ctrl-ds-segment-default and ctrl_ds_segment_default?

Otherwise,

Reviewed-by on v6 9/11
#20 lore raw

Re: [PATCH v6 09/11] hw/arm/aspeed_ast27x0: Set EHCI ctrldssegment-default

In reply to Jamin Lin
Quoted 19 lines · Jamin Lin

On AST2700 platforms, system DRAM is mapped above 4GB with a base address at 0x400000000.

The Linux EHCI driver programs the segment register to zero when 64-bit addressing is supported. As a result, descriptor addresses derived from the EHCI registers do not include the DRAM base address.

Descriptor memory is allocated through the DMA API with a 64-bit DMA mask, allowing descriptors to reside in DRAM above 4GB. On AST2700, EHCI queue heads (QH) and queue element transfer descriptors (qTD) are therefore placed at addresses starting from 0x400000000.

Set the ctrldssegment-default property to "sc->memmap[ASPEED_DEV_SDRAM] >> 32" so the upper 32 bits of descriptor addresses are adjusted accordingly. This allows the emulated EHCI controller to construct correct system addresses when accessing descriptors in DRAM above 4GB.

Quoted 5 lines
---
  hw/arm/aspeed_ast27x0.c | 3 +++
  1 file changed, 3 insertions(+)
#4 lore raw

Re: [PATCH v6 02/11] hw/usb/hcd-ehci: Change descriptor addresses to 64-bit with migration compatibility

In reply to Jamin Lin
Quoted 66 lines · Jamin Lin

Change internal EHCI descriptor addresses from uint32_t to uint64_t.

The following fields are updated:
- EHCIPacket::qtdaddr
- EHCIQueue::{qhaddr, qtdaddr}
- EHCIState::{a_fetch_addr, p_fetch_addr}

Update get_dwords() and put_dwords() to take 64-bit addresses and propagate the type change through the descriptor traversal paths.

Adjust NLPTR_GET() to operate on 64-bit values:

     #define NLPTR_GET(x) ((x) & ~0x1fULL)

so that link pointer masking works correctly when descriptor addresses exceed 32-bit space. The previous mask (0xffffffe0) implicitly truncated addresses to 32 bits.

This patch does not change the on-wire descriptor layout yet. It only removes the internal 32-bit address limit and prepares for later patches that will add full 64-bit QH/qTD/iTD/siTD support.

Update the EHCI trace-events prototypes for QH, qTD, iTD, and siTD to use uint64_t for the address argument and print it with PRIx64. This ensures full 64-bit addresses are shown in trace output and improves debugging of queue heads and transfer descriptors.

Migration compatibility:

To preserve backward migration compatibility, keep the legacy 32-bit fetch address fields (a_fetch_addr_32, p_fetch_addr_32) alongside the new 64-bit fields.

Migration format is selected using a machine compat property "x-migrate-fetch-addr-64bit":

- Old machine types migrate 32-bit fetch addresses
- New machine types migrate full 64-bit fetch addresses

This is implemented using VMSTATE_UINT32_TEST() and VMSTATE_UINT64_TEST() so that only the appropriate format is migrated.

In pre_save, the 32-bit shadow fields are populated when migrating to old machine types. In post_load, the 32-bit values are restored into the 64-bit fields when loading old migration streams.

No functional change.

---
  hw/usb/hcd-ehci.h   | 17 +++++++----
  hw/core/machine.c   |  5 +++-
  hw/usb/hcd-ehci.c   | 72 ++++++++++++++++++++++++++++++++-------------
  hw/usb/trace-events | 24 +++++++--------
  4 files changed, 78 insertions(+), 40 deletions(-)
hw/usb/hcd-ehci.h +1 -0
@@ -256,6 +256,7 @@ struct EHCIState {
 
      /* properties */
      uint32_t maxframes;
+    bool migrate_fetch_addr_64bit;
 
      /*
       *  EHCI spec version 1.0 Section 2.3

hw/usb/hcd-ehci.h struct EHCIState { · line 299

Quoted 7 lines
hw/usb/hcd-ehci.h +2 -2
@@ -294,8 +295,10 @@ struct EHCIState {
      EHCIQueueHead pqueues;
 
      /* which address to look at next */
-    uint32_t a_fetch_addr;
-    uint32_t p_fetch_addr;
+    uint32_t a_fetch_addr_32;
+    uint32_t p_fetch_addr_32;

Technically '?_fetch_addr_32' aren't "which address to look at next" anymore, but simple migration helpers. Could be worth adding a comment? And rename including 'migrat{e,ion}'? Can be cleared on top.

Quoted 21 lines
hw/usb/hcd-ehci.h +5 -1
+    uint64_t a_fetch_addr;
+    uint64_t p_fetch_addr;
 
      USBPacket ipacket;
      QEMUSGList isgl;
@@ -308,7 +311,9 @@ struct EHCIState {
  };
 
  #define DEFINE_EHCI_COMMON_PROPERTIES(_state) \
-    DEFINE_PROP_UINT32("maxframes", _state, ehci.maxframes, 128)
+    DEFINE_PROP_UINT32("maxframes", _state, ehci.maxframes, 128), \
+    DEFINE_PROP_BOOL("x-migrate-fetch-addr-64bit", _state, \
+                     ehci.migrate_fetch_addr_64bit, true)
 
  extern const VMStateDescription vmstate_ehci;
 
hw/core/machine.c +0 -0
index 1b661fd36a..54336da294 100644

hw/core/machine.c line 46

Quoted 9 lines
hw/core/machine.c +4 -1
@@ -39,7 +39,10 @@
  #include "hw/acpi/generic_event_device.h"
  #include "qemu/audio.h"
 
-GlobalProperty hw_compat_11_0[] = {};
+GlobalProperty hw_compat_11_0[] = {
+    { "sysbus-ehci-usb", "x-migrate-fetch-addr-64bit", "off" },
+    { "pci-ehci-usb", "x-migrate-fetch-addr-64bit", "off" },
+};
  const size_t hw_compat_11_0_len = G_N_ELEMENTS(hw_compat_11_0);

Good.

Quoted 43 lines
hw/core/machine.c +22 -0
+static bool ehci_fetch_addr_64_needed(void *opaque, int version_id)
+{
+    EHCIState *s = opaque;
+
+    return s->migrate_fetch_addr_64bit;
+}
+
+static bool ehci_fetch_addr_32_needed(void *opaque, int version_id)
+{
+    return !ehci_fetch_addr_64_needed(opaque, version_id);
+}
+
  static int usb_ehci_pre_save(void *opaque)
  {
      EHCIState *ehci = opaque;
@@ -2424,6 +2438,11 @@ static int usb_ehci_pre_save(void *opaque)
      ehci->last_run_ns -= (ehci->frindex - new_frindex) * UFRAME_TIMER_NS;
      ehci->frindex = new_frindex;
 
+    if (!ehci->migrate_fetch_addr_64bit) {
+        ehci->a_fetch_addr_32 = ehci->a_fetch_addr;
+        ehci->p_fetch_addr_32 = ehci->p_fetch_addr;
+    }
+
      return 0;
  }
 
@@ -2444,6 +2463,11 @@ static int usb_ehci_post_load(void *opaque, int version_id)
          }
      }
 
+    if (!s->migrate_fetch_addr_64bit) {
+        s->a_fetch_addr = s->a_fetch_addr_32;
+        s->p_fetch_addr = s->p_fetch_addr_32;
+    }
+
      return 0;
  }
 
@@ -2504,8 +2528,14 @@ const VMStateDescription vmstate_ehci = {
          /* schedule state */
          VMSTATE_UINT32(astate, EHCIState),
          VMSTATE_UINT32(pstate, EHCIState),

hw/core/machine.c const VMStateDescription vmstate_ehci = { · line 2538

Quoted 10 lines
hw/core/machine.c +8 -2
-        VMSTATE_UINT32(a_fetch_addr, EHCIState),
-        VMSTATE_UINT32(p_fetch_addr, EHCIState),
+        VMSTATE_UINT32_TEST(a_fetch_addr_32, EHCIState,
+                            ehci_fetch_addr_32_needed),
+        VMSTATE_UINT32_TEST(p_fetch_addr_32, EHCIState,
+                            ehci_fetch_addr_32_needed),
+        VMSTATE_UINT64_TEST(a_fetch_addr, EHCIState,
+                            ehci_fetch_addr_64_needed),
+        VMSTATE_UINT64_TEST(p_fetch_addr, EHCIState,
+                            ehci_fetch_addr_64_needed),

Hmmmm. I didn't review all previous versions, for my own education, why don't we use VMSTATE_UINT64_V() here? Did you test (pre-11 <-> this series) migration? My understanding was we need to maintain the stream size to avoid mismatch.

Quoted 3 lines
hw/core/machine.c +0 -0
          VMSTATE_END_OF_LIST()
      }
  };
#14 lore raw

Re: [PATCH v6 08/11] hw/usb/hcd-ehci: Add ctrldssegment-default property

In reply to Philippe Mathieu-Daudé
Quoted 55 lines · Philippe Mathieu-Daudé
Quoted 44 lines · Jamin Lin

When 64-bit addressing is supported, the Linux EHCI driver programs the segment register to zero. See ehci_run function:
https://github.com/torvalds/linux/blob/master/drivers/usb/host/ehci-hcd.c

The driver comment also notes that descriptor structures allocated from the DMA pool use segment zero semantics.

Descriptor memory is allocated using the DMA API. The platform driver configures a 64-bit DMA mask so memory can be allocated above 4GB. See ehci_platform_probe function:
https://github.com/torvalds/linux/blob/master/drivers/usb/host/ehci-platform.c

On AST2700 platforms, system DRAM is mapped above 4GB at 0x400000000. As a result, descriptor addresses constructed directly from the guest EHCI registers do not match the actual system address used by the controller when fetching queue heads (QH) and queue element transfer descriptors (qTD).

Add a ctrldssegment-default property so platforms can provide a descriptor address offset when constructing descriptor addresses. This allows systems where DRAM resides above 4GB to access EHCI descriptors correctly.

The default value is zero, so existing machines are not affected.

---
  hw/usb/hcd-ehci.h | 5 ++++-
  hw/usb/hcd-ehci.c | 1 +

  2 files changed, 5 insertions(+), 1 deletion(-)
hw/usb/hcd-ehci.h +1 -0
index 8e6a8cdfb0..3d57e1de5f 100644
@@ -264,6 +264,7 @@ struct EHCIState {
       uint32_t maxframes;
       bool migrate_fetch_addr_64bit;
       bool caps_64bit_addr;
+    uint32_t ctrldssegment_default;
       /*
        *  EHCI spec version 1.0 Section 2.3

hw/usb/hcd-ehci.h struct EHCIState { · line 328

Quoted 7 lines
hw/usb/hcd-ehci.h +3 -1
@@ -322,7 +323,9 @@ struct EHCIState {
       DEFINE_PROP_BOOL("x-migrate-fetch-addr-64bit", _state, \
                        ehci.migrate_fetch_addr_64bit, true), \
       DEFINE_PROP_BOOL("caps-64bit-addr", _state, \
-                     ehci.caps_64bit_addr, false)
+                     ehci.caps_64bit_addr, false), \
+    DEFINE_PROP_UINT32("ctrldssegment-default", _state, \
+                       ehci.ctrldssegment_default, 0)

ctrl-ds-segment-default and ctrl_ds_segment_default?

Quoted 3 lines

Otherwise,

yeah. I agree. That's the general pattern for property names.

Thanks,

C.

#15 lore raw

Re: [PATCH v6 08/11] hw/usb/hcd-ehci: Add ctrldssegment-default property

Quoted 56 lines · Cédric Le Goater
Quoted 55 lines · Philippe Mathieu-Daudé
Quoted 44 lines · Jamin Lin

When 64-bit addressing is supported, the Linux EHCI driver programs the segment register to zero. See ehci_run function:
https://github.com/torvalds/linux/blob/master/drivers/usb/host/ehci- hcd.c

The driver comment also notes that descriptor structures allocated from the DMA pool use segment zero semantics.

Descriptor memory is allocated using the DMA API. The platform driver configures a 64-bit DMA mask so memory can be allocated above 4GB. See ehci_platform_probe function:
https://github.com/torvalds/linux/blob/master/drivers/usb/host/ehci- platform.c

On AST2700 platforms, system DRAM is mapped above 4GB at 0x400000000. As a result, descriptor addresses constructed directly from the guest EHCI registers do not match the actual system address used by the controller when fetching queue heads (QH) and queue element transfer descriptors (qTD).

Add a ctrldssegment-default property so platforms can provide a descriptor address offset when constructing descriptor addresses. This allows systems where DRAM resides above 4GB to access EHCI descriptors correctly.

The default value is zero, so existing machines are not affected.

---
  hw/usb/hcd-ehci.h | 5 ++++-
  hw/usb/hcd-ehci.c | 1 +

  2 files changed, 5 insertions(+), 1 deletion(-)
hw/usb/hcd-ehci.h +1 -0
index 8e6a8cdfb0..3d57e1de5f 100644
@@ -264,6 +264,7 @@ struct EHCIState {
       uint32_t maxframes;
       bool migrate_fetch_addr_64bit;
       bool caps_64bit_addr;
+    uint32_t ctrldssegment_default;
       /*
        *  EHCI spec version 1.0 Section 2.3

hw/usb/hcd-ehci.h struct EHCIState { · line 328

Quoted 7 lines
hw/usb/hcd-ehci.h +3 -1
@@ -322,7 +323,9 @@ struct EHCIState {
       DEFINE_PROP_BOOL("x-migrate-fetch-addr-64bit", _state, \
                        ehci.migrate_fetch_addr_64bit, true), \
       DEFINE_PROP_BOOL("caps-64bit-addr", _state, \
-                     ehci.caps_64bit_addr, false)
+                     ehci.caps_64bit_addr, false), \
+    DEFINE_PROP_UINT32("ctrldssegment-default", _state, \
+                       ehci.ctrldssegment_default, 0)

ctrl-ds-segment-default and ctrl_ds_segment_default?

Quoted 5 lines
Quoted 3 lines

Otherwise,

yeah. I agree. That's the general pattern for property names.

I don't get the '-default' suffix. Alternatives:

dram-segment-base-addr or even x-dram-segment-base-addr since I still believe this is a AST2700 issue, likely not mapping RAM appropriately.

Quoted 5 lines

Thanks,

C.

#16 lore raw

Re: [PATCH v6 08/11] hw/usb/hcd-ehci: Add ctrldssegment-default property

Quoted 61 lines · Philippe Mathieu-Daudé
Quoted 60 lines · Cédric Le Goater
Quoted 58 lines · Philippe Mathieu-Daudé
Quoted 44 lines · Jamin Lin

When 64-bit addressing is supported, the Linux EHCI driver programs the segment register to zero. See ehci_run function:
https://github.com/torvalds/linux/blob/master/drivers/usb/host/ehci- hcd.c

The driver comment also notes that descriptor structures allocated from the DMA pool use segment zero semantics.

Descriptor memory is allocated using the DMA API. The platform driver configures a 64-bit DMA mask so memory can be allocated above 4GB. See ehci_platform_probe function:
https://github.com/torvalds/linux/blob/master/drivers/usb/host/ehci- platform.c

On AST2700 platforms, system DRAM is mapped above 4GB at 0x400000000. As a result, descriptor addresses constructed directly from the guest EHCI registers do not match the actual system address used by the controller when fetching queue heads (QH) and queue element transfer descriptors (qTD).

Add a ctrldssegment-default property so platforms can provide a descriptor address offset when constructing descriptor addresses. This allows systems where DRAM resides above 4GB to access EHCI descriptors correctly.

The default value is zero, so existing machines are not affected.

---
  hw/usb/hcd-ehci.h | 5 ++++-
  hw/usb/hcd-ehci.c | 1 +

  2 files changed, 5 insertions(+), 1 deletion(-)
hw/usb/hcd-ehci.h +1 -0
index 8e6a8cdfb0..3d57e1de5f 100644
@@ -264,6 +264,7 @@ struct EHCIState {
       uint32_t maxframes;
       bool migrate_fetch_addr_64bit;
       bool caps_64bit_addr;
+    uint32_t ctrldssegment_default;
       /*
        *  EHCI spec version 1.0 Section 2.3

hw/usb/hcd-ehci.h struct EHCIState { · line 328

Quoted 7 lines
hw/usb/hcd-ehci.h +3 -1
@@ -322,7 +323,9 @@ struct EHCIState {
       DEFINE_PROP_BOOL("x-migrate-fetch-addr-64bit", _state, \
                        ehci.migrate_fetch_addr_64bit, true), \
       DEFINE_PROP_BOOL("caps-64bit-addr", _state, \
-                     ehci.caps_64bit_addr, false)
+                     ehci.caps_64bit_addr, false), \
+    DEFINE_PROP_UINT32("ctrldssegment-default", _state, \
+                       ehci.ctrldssegment_default, 0)

ctrl-ds-segment-default and ctrl_ds_segment_default?

Otherwise,

Quoted 5 lines
Quoted 1 line

yeah. I agree. That's the general pattern for property names.

I don't get the '-default' suffix. Alternatives:

dram-segment-base-addr or even x-dram-segment-base-addr

What ever makes sense. EHCI specs have :

CTRLDSSEGMENT 4G Segment Selector

The purpose of the "ctrldssegment-default" property is to force a default value of the CTRLDSSEGMENT register for the ast2700.

Jamin, how are we sure the DMAs are allocated in this 4G segment ? because RAM covers 8G :

      0000000400000000-00000005ffffffff (prio 0, ram): ram
Quoted 1 line

since I still believe this is a AST2700 issue, likely not mapping RAM appropriately.

Other controllers would be impacted no ?
Linux (upstream) forces the CTRLDSSEGMENT value :

/*
* hcc_params controls whether ehci->regs->segment must (!!!)
* be used; it constrains QH/ITD/SITD and QTD locations.
* dma_pool consistent memory always uses segment zero.
* streaming mappings for I/O buffers, like dma_map_single(),
* can return segments above 4GB, if the device allows. *
* NOTE: the dma mask is visible through dev->dma_mask, so
* drivers can pass this info along ... like NETIF_F_HIGHDMA,
* Scsi_Host.highmem_io, and so forth. It's readonly to all
* host side drivers though.
*/
hcc_params = ehci_readl(ehci, &ehci->caps->hcc_params); if (HCC_64BIT_ADDR(hcc_params)) {
ehci_writel(ehci, 0, &ehci->regs->segment); <------- here
#if 0
// this is deeply broken on almost all architectures

		if (!dma_set_mask(hcd->self.controller, DMA_BIT_MASK(64)))
			ehci_info(ehci, "enabled 64bit DMA\n");
#endif
	}

C.

author
#17 lore raw

RE: [PATCH v6 08/11] hw/usb/hcd-ehci: Add ctrldssegment-default property

Hi Cédric, Philippe

Quoted 80 lines

Subject: Re: [PATCH v6 08/11] hw/usb/hcd-ehci: Add ctrldssegment-default property

Quoted 8 lines · Philippe Mathieu-Daudé
Quoted 7 lines · Cédric Le Goater
Quoted 6 lines · Philippe Mathieu-Daudé
Quoted 5 lines · Jamin Lin

When 64-bit addressing is supported, the Linux EHCI driver programs the segment register to zero. See ehci_run function:
https://github.com/torvalds/linux/blob/master/drivers/usb/host/ehci
- hcd.c

Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines

The driver comment also notes that descriptor structures allocated from the DMA pool use segment zero semantics.

Descriptor memory is allocated using the DMA API. The platform driver configures a 64-bit DMA mask so memory can be allocated above

4GB.

Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines

See ehci_platform_probe function:
https://github.com/torvalds/linux/blob/master/drivers/usb/host/ehci
- platform.c

On AST2700 platforms, system DRAM is mapped above 4GB at

0x400000000.

Quoted 19 lines
Quoted 19 lines
Quoted 19 lines
Quoted 19 lines

As a result, descriptor addresses constructed directly from the guest EHCI registers do not match the actual system address used by the controller when fetching queue heads (QH) and queue element transfer descriptors (qTD).

Add a ctrldssegment-default property so platforms can provide a descriptor address offset when constructing descriptor addresses. This allows systems where DRAM resides above 4GB to access EHCI descriptors correctly.

The default value is zero, so existing machines are not affected.

---
  hw/usb/hcd-ehci.h | 5 ++++-
  hw/usb/hcd-ehci.c | 1 +

  2 files changed, 5 insertions(+), 1 deletion(-)
Quoted 14 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
hw/usb/hcd-ehci.h index renamed from hw/usb/hcd-ehci.h +0 -0

8e6a8cdfb0..3d57e1de5f 100644

hw/usb/hcd-ehci.h +0 -0
@@ -264,6 +264,7 @@ struct EHCIState {
       uint32_t maxframes;
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
hw/usb/hcd-ehci.h +1 -0
       bool migrate_fetch_addr_64bit;
       bool caps_64bit_addr;
+    uint32_t ctrldssegment_default;
       /*
        *  EHCI spec version 1.0 Section 2.3 @@ -322,7 +323,9 @@
 struct EHCIState {

      DEFINE_PROP_BOOL("x-migrate-fetch-addr-64bit", _state, \
                       ehci.migrate_fetch_addr_64bit, true),

\

Quoted 12 lines
Quoted 12 lines
Quoted 11 lines
Quoted 5 lines

      DEFINE_PROP_BOOL("caps-64bit-addr", _state, \
-                     ehci.caps_64bit_addr, false) +                     ehci.caps_64bit_addr, false), \
+    DEFINE_PROP_UINT32("ctrldssegment-default", _state, \
+                       ehci.ctrldssegment_default, 0)

ctrl-ds-segment-default and ctrl_ds_segment_default?

Otherwise,

Quoted 5 lines
Quoted 1 line

yeah. I agree. That's the general pattern for property names.

I don't get the '-default' suffix. Alternatives:

dram-segment-base-addr or even x-dram-segment-base-addr

What ever makes sense. EHCI specs have :

CTRLDSSEGMENT 4G Segment Selector

Quoted 8 lines

The purpose of the "ctrldssegment-default" property is to force a default value of the CTRLDSSEGMENT register for the ast2700.

Jamin, how are we sure the DMAs are allocated in this 4G segment ? because RAM covers 8G :

       0000000400000000-00000005ffffffff (prio 0, ram): ram

It is a firmware issue. After discussing it with the USB driver owner, we found that this issue occurs when users use 8GB DRAM because the driver does not configure CTRLDSSEGMENT. Currently, our SDK only supports up to 2GB DRAM, so users should not encounter this issue under normal conditions.
However, this could become a potential issue if users configure the DRAM size to 8GB.

Quoted 7 lines
Quoted 3 lines

since I still believe this is a AST2700 issue, likely not mapping RAM appropriately.

Other controllers would be impacted no ?
Linux (upstream) forces the CTRLDSSEGMENT value :

Linux (upstream) forces CTRLDSSEGMENT to 0, which means only DRAM addresses below 4GB are allowed.

Therefore, AST2700 hardware sets the default value of CTRLDSSEGMENT to 4, which maps to the AST2700 DRAM required base address 0x400000000.

In other words, the AST2700 firmware only needs to configure bits 0 and 1 of CTRLDSSEGMENT.

However, setting only bits 0 and 1 in the firmware is not sufficient for QEMU, because QEMU is not aware that the AST2700 hardware initializes the default value of CTRLDSSEGMENT to 4.

In conclusion, should I rename this property to ctrl-ds-segment-default and ctrl_ds_segment_default? Also, should I create a new patch for this change, or resend the entire patch series?

Thanks,
Jamin

Quoted 23 lines

/*
* hcc_params controls whether ehci->regs->segment must (!!!)
* be used; it constrains QH/ITD/SITD and QTD locations.
* dma_pool consistent memory always uses segment zero.
* streaming mappings for I/O buffers, like dma_map_single(),
* can return segments above 4GB, if the device allows. *
* NOTE: the dma mask is visible through dev->dma_mask, so
* drivers can pass this info along ... like NETIF_F_HIGHDMA,
* Scsi_Host.highmem_io, and so forth. It's readonly to all
* host side drivers though.
*/
hcc_params = ehci_readl(ehci, &ehci->caps->hcc_params); if (HCC_64BIT_ADDR(hcc_params)) {
ehci_writel(ehci, 0, &ehci->regs->segment); <------- here
#if 0
// this is deeply broken on almost all architectures

		if (!dma_set_mask(hcd->self.controller, DMA_BIT_MASK(64)))
			ehci_info(ehci, "enabled 64bit DMA\n"); #endif
	}

C.

author
#5 lore raw

RE: [PATCH v6 02/11] hw/usb/hcd-ehci: Change descriptor addresses to 64-bit with migration compatibility

In reply to Philippe Mathieu-Daudé

Hi Philippe

Quoted 73 lines

Subject: Re: [PATCH v6 02/11] hw/usb/hcd-ehci: Change descriptor addresses to 64-bit with migration compatibility

Quoted 46 lines · Jamin Lin

Change internal EHCI descriptor addresses from uint32_t to uint64_t.

The following fields are updated:
- EHCIPacket::qtdaddr
- EHCIQueue::{qhaddr, qtdaddr}
- EHCIState::{a_fetch_addr, p_fetch_addr}

Update get_dwords() and put_dwords() to take 64-bit addresses and propagate the type change through the descriptor traversal paths.

Adjust NLPTR_GET() to operate on 64-bit values:

     #define NLPTR_GET(x) ((x) & ~0x1fULL)

so that link pointer masking works correctly when descriptor addresses exceed 32-bit space. The previous mask (0xffffffe0) implicitly truncated addresses to 32 bits.

This patch does not change the on-wire descriptor layout yet. It only removes the internal 32-bit address limit and prepares for later patches that will add full 64-bit QH/qTD/iTD/siTD support.

Update the EHCI trace-events prototypes for QH, qTD, iTD, and siTD to use uint64_t for the address argument and print it with PRIx64. This ensures full 64-bit addresses are shown in trace output and improves debugging of queue heads and transfer descriptors.

Migration compatibility:

To preserve backward migration compatibility, keep the legacy 32-bit fetch address fields (a_fetch_addr_32, p_fetch_addr_32) alongside the new 64-bit fields.

Migration format is selected using a machine compat property "x-migrate-fetch-addr-64bit":

- Old machine types migrate 32-bit fetch addresses
- New machine types migrate full 64-bit fetch addresses

This is implemented using VMSTATE_UINT32_TEST() and VMSTATE_UINT64_TEST() so that only the appropriate format is migrated.

In pre_save, the 32-bit shadow fields are populated when migrating to old machine types. In post_load, the 32-bit values are restored into the 64-bit fields when loading old migration streams.

Quoted 7 lines

No functional change.

---
  hw/usb/hcd-ehci.h   | 17 +++++++----
  hw/core/machine.c   |  5 +++-
  hw/usb/hcd-ehci.c   | 72

++++++++++++++++++++++++++++++++-------------

Quoted 15 lines
  hw/usb/trace-events | 24 +++++++--------
  4 files changed, 78 insertions(+), 40 deletions(-)
(quoted diff) +1 -0
@@ -256,6 +256,7 @@ struct EHCIState {
 
      /* properties */
      uint32_t maxframes;
+    bool migrate_fetch_addr_64bit;
 
      /*
       *  EHCI spec version 1.0 Section 2.3 @@ -294,8 +295,10 @@
struct EHCIState {
      EHCIQueueHead pqueues;
Quoted 10 lines

Review

Quoted 5 lines
      /* which address to look at next */
(quoted diff) +2 -2
-    uint32_t a_fetch_addr;
-    uint32_t p_fetch_addr;
+    uint32_t a_fetch_addr_32;
+    uint32_t p_fetch_addr_32;

Technically '?_fetch_addr_32' aren't "which address to look at next" anymore, but simple migration helpers. Could be worth adding a comment? And rename including 'migrat{e,ion}'? Can be cleared on top.

Will update to

/*
* Migration compatibility fields for old machine types that only
* support 32-bit fetch addresses in the migration stream. *
* New machine types migrate the full 64-bit runtime fetch address. */
uint32_t migrate_a_fetch_addr;
uint32_t migrate_p_fetch_addr;

/*
* Controls migration stream compatibility for old machine types.
* Old machine types only transfer 32-bit fetch addresses. */
bool migrate_fetch_addr_64bit;

Quoted 80 lines
Quoted 21 lines
(quoted diff) +5 -1
+    uint64_t a_fetch_addr;
+    uint64_t p_fetch_addr;
 
      USBPacket ipacket;
      QEMUSGList isgl;
@@ -308,7 +311,9 @@ struct EHCIState {
  };
 
  #define DEFINE_EHCI_COMMON_PROPERTIES(_state) \
-    DEFINE_PROP_UINT32("maxframes", _state, ehci.maxframes, 128)
+    DEFINE_PROP_UINT32("maxframes", _state, ehci.maxframes, 128), \
+    DEFINE_PROP_BOOL("x-migrate-fetch-addr-64bit", _state, \
+                     ehci.migrate_fetch_addr_64bit, true)
 
  extern const VMStateDescription vmstate_ehci;
 
hw/core/machine.c index renamed from hw/core/machine.c +0 -0

1b661fd36a..54336da294 100644

hw/core/machine.c line 45

Quoted 8 lines
hw/core/machine.c +3 -1
@@ -39,7 +39,10 @@
  #include "hw/acpi/generic_event_device.h"
  #include "qemu/audio.h"
 
-GlobalProperty hw_compat_11_0[] = {};
+GlobalProperty hw_compat_11_0[] = {
+    { "sysbus-ehci-usb", "x-migrate-fetch-addr-64bit", "off" },
+    { "pci-ehci-usb", "x-migrate-fetch-addr-64bit", "off" }, };
  const size_t hw_compat_11_0_len = G_N_ELEMENTS(hw_compat_11_0);

Good.

hw/core/machine.c static int usb_ehci_pre_save(void *opaque) · line 2438

Quoted 13 lines
hw/core/machine.c +8 -0
+static bool ehci_fetch_addr_64_needed(void *opaque, int version_id) {
+    EHCIState *s = opaque;
+
+    return s->migrate_fetch_addr_64bit; }
+
+static bool ehci_fetch_addr_32_needed(void *opaque, int version_id) {
+    return !ehci_fetch_addr_64_needed(opaque, version_id); }
+
  static int usb_ehci_pre_save(void *opaque)
  {
      EHCIState *ehci = opaque;
@@ -2424,6 +2438,11 @@ static int usb_ehci_pre_save(void *opaque)
      ehci->last_run_ns -= (ehci->frindex - new_frindex) *

UFRAME_TIMER_NS;

hw/core/machine.c static int usb_ehci_post_load(void *opaque, int

Quoted 11 lines
hw/core/machine.c +5 -0
      ehci->frindex = new_frindex;
 
+    if (!ehci->migrate_fetch_addr_64bit) {
+        ehci->a_fetch_addr_32 = ehci->a_fetch_addr;
+        ehci->p_fetch_addr_32 = ehci->p_fetch_addr;
+    }
+
      return 0;
  }
 
@@ -2444,6 +2463,11 @@ static int usb_ehci_post_load(void *opaque, int

version_id)

Quoted 22 lines
hw/core/machine.c +10 -2
          }
      }
 
+    if (!s->migrate_fetch_addr_64bit) {
+        s->a_fetch_addr = s->a_fetch_addr_32;
+        s->p_fetch_addr = s->p_fetch_addr_32;
+    }
+
      return 0;
  }
 
@@ -2504,8 +2528,14 @@ const VMStateDescription vmstate_ehci = {
          /* schedule state */
          VMSTATE_UINT32(astate, EHCIState),
          VMSTATE_UINT32(pstate, EHCIState),
-        VMSTATE_UINT32(a_fetch_addr, EHCIState),
-        VMSTATE_UINT32(p_fetch_addr, EHCIState),
+        VMSTATE_UINT32_TEST(a_fetch_addr_32, EHCIState,
+                            ehci_fetch_addr_32_needed),
+        VMSTATE_UINT32_TEST(p_fetch_addr_32, EHCIState,
+                            ehci_fetch_addr_32_needed),
+        VMSTATE_UINT64_TEST(a_fetch_addr, EHCIState,
Quoted 9 lines

hw/core/machine.c const VMStateDescription vmstate_ehci = { · line 2538

Quoted 3 lines
hw/core/machine.c +3 -0
+                            ehci_fetch_addr_64_needed),
+        VMSTATE_UINT64_TEST(p_fetch_addr, EHCIState,
+                            ehci_fetch_addr_64_needed),

Hmmmm. I didn't review all previous versions, for my own education, why don't we use VMSTATE_UINT64_V() here? Did you test (pre-11 <-> this series) migration? My understanding was we need to maintain the stream size to avoid mismatch.

Please see the migration test steps here:
https://patchwork.kernel.org/project/qemu-devel/patch/[email protected]/#26911317
I tested migration compatibility with both v11 and v10.2.

I originally tried using VMSTATE_UINT64_V(). That implementation was similar to the v3 version of this series, which you previously reviewed:
https://patchwork.kernel.org/project/qemu-devel/patch/[email protected]/

Unfortunately, that approach caused migration failures.

The issue is that older machine types do not contain the 64-bit fetch address fields in their migration ABI. Simply adding VMSTATE_UINT64_V() changes the migration stream layout, which breaks compatibility with older machine types / older QEMU versions.

The migration failure looks like this:
qemu-system-x86_64: Missing section footer for 0000:00:03.0/ehci
qemu-system-x86_64: load of migration failed: Invalid argument: Section footer error, section_id: 42

Because of this, I used migrate_fetch_addr_64bit to preserve the original 32-bit migration stream for older compatible machine types while enabling the 64-bit migration format only for newer machine types.

Thanks,
Jamin

hw/usb/hcd-ehci.h +2 -2
index 3d57e1de5f..7b3e92895a 100644
@@ -303,8 +303,8 @@ struct EHCIState {
     EHCIQueueHead pqueues;
 
     /* which address to look at next */
-    uint32_t a_fetch_addr_32;
-    uint32_t p_fetch_addr_32;
+    uint32_t migrate_a_fetch_addr;
+    uint32_t migrate_p_fetch_addr;
     uint64_t a_fetch_addr;
     uint64_t p_fetch_addr;
 
hw/usb/hcd-ehci.c +13 -22
index 62c3770ac5..d3b9890726 100644
@@ -2459,16 +2459,9 @@ static USBBusOps ehci_bus_ops_standalone = {
     .wakeup_endpoint = ehci_wakeup_endpoint,
 };
 
-static bool ehci_fetch_addr_64_needed(void *opaque, int version_id)
+static bool ehci_core_is_before_version_3(void *opaque, int version_id)
 {
-    EHCIState *s = opaque;
-
-    return s->migrate_fetch_addr_64bit;
-}
-
-static bool ehci_fetch_addr_32_needed(void *opaque, int version_id)
-{
-    return !ehci_fetch_addr_64_needed(opaque, version_id);
+    return version_id < 3;
 }
 
 static int usb_ehci_pre_save(void *opaque)
@@ -2482,8 +2475,8 @@ static int usb_ehci_pre_save(void *opaque)
     ehci->frindex = new_frindex;
 
     if (!ehci->migrate_fetch_addr_64bit) {
-        ehci->a_fetch_addr_32 = ehci->a_fetch_addr;
-        ehci->p_fetch_addr_32 = ehci->p_fetch_addr;
+        ehci->migrate_a_fetch_addr = ehci->a_fetch_addr;
+        ehci->migrate_p_fetch_addr = ehci->p_fetch_addr;
     }
 
     return 0;
@@ -2507,8 +2500,8 @@ static int usb_ehci_post_load(void *opaque, int version_id)
     }
 
     if (!s->migrate_fetch_addr_64bit) {
-        s->a_fetch_addr = s->a_fetch_addr_32;
-        s->p_fetch_addr = s->p_fetch_addr_32;
+        s->a_fetch_addr = s->migrate_a_fetch_addr;
+        s->p_fetch_addr = s->migrate_p_fetch_addr;
     }
 
     return 0;
@@ -2542,7 +2535,7 @@ static void usb_ehci_vm_state_change(void *opaque, bool running, RunState state)
 
 const VMStateDescription vmstate_ehci = {
     .name        = "ehci-core",
-    .version_id  = 2,
+    .version_id  = 3,
     .minimum_version_id  = 1,
     .pre_save    = usb_ehci_pre_save,
     .post_load   = usb_ehci_post_load,
@@ -2571,14 +2564,12 @@ const VMStateDescription vmstate_ehci = {
         /* schedule state */
         VMSTATE_UINT32(astate, EHCIState),
         VMSTATE_UINT32(pstate, EHCIState),
-        VMSTATE_UINT32_TEST(a_fetch_addr_32, EHCIState,
-                            ehci_fetch_addr_32_needed),
-        VMSTATE_UINT32_TEST(p_fetch_addr_32, EHCIState,
-                            ehci_fetch_addr_32_needed),
-        VMSTATE_UINT64_TEST(a_fetch_addr, EHCIState,
-                            ehci_fetch_addr_64_needed),
-        VMSTATE_UINT64_TEST(p_fetch_addr, EHCIState,
-                            ehci_fetch_addr_64_needed),
+        VMSTATE_UINT32_TEST(migrate_a_fetch_addr, EHCIState,
+                            ehci_core_is_before_version_3),
+        VMSTATE_UINT32_TEST(migrate_p_fetch_addr, EHCIState,
+                            ehci_core_is_before_version_3),
+        VMSTATE_UINT64_V(a_fetch_addr, EHCIState, 3),
+        VMSTATE_UINT64_V(p_fetch_addr, EHCIState, 3),
         VMSTATE_END_OF_LIST()
     }
 };
author
#18 lore raw

RE: [PATCH v6 08/11] hw/usb/hcd-ehci: Add ctrldssegment-default property

In reply to Jamin Lin

Hi Cédric, Philippe

Please help to review.

hw/usb/ehci: Improve migration helper and property naming
https://patchwork.kernel.org/project/qemu-devel/list/?series=1090233

Thanks,
Jamin

Quoted 154 lines

Hi Cédric, Philippe

Quoted 46 lines

Subject: Re: [PATCH v6 08/11] hw/usb/hcd-ehci: Add ctrldssegment-default property

Quoted 9 lines · Philippe Mathieu-Daudé
Quoted 8 lines · Cédric Le Goater
Quoted 7 lines · Philippe Mathieu-Daudé
Quoted 6 lines · Jamin Lin

When 64-bit addressing is supported, the Linux EHCI driver programs the segment register to zero. See ehci_run function:
https://github.com/torvalds/linux/blob/master/drivers/usb/host/eh
ci
- hcd.c

Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines

The driver comment also notes that descriptor structures allocated from the DMA pool use segment zero semantics.

Descriptor memory is allocated using the DMA API. The platform driver configures a 64-bit DMA mask so memory can be allocated above

4GB.

Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines

See ehci_platform_probe function:
https://github.com/torvalds/linux/blob/master/drivers/usb/host/eh
ci
- platform.c

On AST2700 platforms, system DRAM is mapped above 4GB at

0x400000000.

Quoted 19 lines
Quoted 19 lines
Quoted 19 lines
Quoted 19 lines

As a result, descriptor addresses constructed directly from the guest EHCI registers do not match the actual system address used by the controller when fetching queue heads (QH) and queue element transfer descriptors (qTD).

Add a ctrldssegment-default property so platforms can provide a descriptor address offset when constructing descriptor addresses. This allows systems where DRAM resides above 4GB to access EHCI descriptors correctly.

The default value is zero, so existing machines are not affected.

---
  hw/usb/hcd-ehci.h | 5 ++++-
  hw/usb/hcd-ehci.c | 1 +

  2 files changed, 5 insertions(+), 1 deletion(-)
Quoted 11 lines
Quoted 11 lines
Quoted 11 lines
Quoted 11 lines

hw/usb/hcd-ehci.h struct EHCIState { · line 269

Quoted 11 lines
hw/usb/hcd-ehci.h index renamed from hw/usb/hcd-ehci.h +0 -0

8e6a8cdfb0..3d57e1de5f 100644

hw/usb/hcd-ehci.h +1 -0
@@ -264,6 +264,7 @@ struct EHCIState {
       uint32_t maxframes;
       bool migrate_fetch_addr_64bit;
       bool caps_64bit_addr;
+    uint32_t ctrldssegment_default;
       /*
        *  EHCI spec version 1.0 Section 2.3 @@ -322,7 +323,9

@@

Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
hw/usb/hcd-ehci.h +0 -0
 struct EHCIState {

      DEFINE_PROP_BOOL("x-migrate-fetch-addr-64bit", _state, \
                       ehci.migrate_fetch_addr_64bit,

true),

Quoted 23 lines

\

Quoted 12 lines
Quoted 12 lines
Quoted 11 lines
Quoted 5 lines

      DEFINE_PROP_BOOL("caps-64bit-addr", _state, \
-                     ehci.caps_64bit_addr, false) +                     ehci.caps_64bit_addr, false), \
+    DEFINE_PROP_UINT32("ctrldssegment-default", _state, \
+                       ehci.ctrldssegment_default, 0)

ctrl-ds-segment-default and ctrl_ds_segment_default?

Otherwise,

Quoted 5 lines
Quoted 1 line

yeah. I agree. That's the general pattern for property names.

I don't get the '-default' suffix. Alternatives:

dram-segment-base-addr or even x-dram-segment-base-addr

What ever makes sense. EHCI specs have :

CTRLDSSEGMENT 4G Segment Selector

Quoted 8 lines

The purpose of the "ctrldssegment-default" property is to force a default value of the CTRLDSSEGMENT register for the ast2700.

Jamin, how are we sure the DMAs are allocated in this 4G segment ? because RAM covers 8G :

       0000000400000000-00000005ffffffff (prio 0, ram): ram

It is a firmware issue. After discussing it with the USB driver owner, we found that this issue occurs when users use 8GB DRAM because the driver does not configure CTRLDSSEGMENT.
Currently, our SDK only supports up to 2GB DRAM, so users should not encounter this issue under normal conditions.
However, this could become a potential issue if users configure the DRAM size to 8GB.

Quoted 7 lines
Quoted 3 lines

since I still believe this is a AST2700 issue, likely not mapping RAM appropriately.

Other controllers would be impacted no ?
Linux (upstream) forces the CTRLDSSEGMENT value :

Linux (upstream) forces CTRLDSSEGMENT to 0, which means only DRAM addresses below 4GB are allowed.

Therefore, AST2700 hardware sets the default value of CTRLDSSEGMENT to 4, which maps to the AST2700 DRAM required base address 0x400000000.

In other words, the AST2700 firmware only needs to configure bits 0 and 1 of CTRLDSSEGMENT.

However, setting only bits 0 and 1 in the firmware is not sufficient for QEMU, because QEMU is not aware that the AST2700 hardware initializes the default value of CTRLDSSEGMENT to 4.

In conclusion, should I rename this property to ctrl-ds-segment-default and ctrl_ds_segment_default?
Also, should I create a new patch for this change, or resend the entire patch series?

Thanks,
Jamin

Quoted 23 lines

/*
* hcc_params controls whether ehci->regs->segment must (!!!)
* be used; it constrains QH/ITD/SITD and QTD locations.
* dma_pool consistent memory always uses segment zero.
* streaming mappings for I/O buffers, like dma_map_single(),
* can return segments above 4GB, if the device allows. *
* NOTE: the dma mask is visible through dev->dma_mask, so
* drivers can pass this info along ... like NETIF_F_HIGHDMA,
* Scsi_Host.highmem_io, and so forth. It's readonly to all
* host side drivers though.
*/
hcc_params = ehci_readl(ehci, &ehci->caps->hcc_params); if (HCC_64BIT_ADDR(hcc_params)) {
ehci_writel(ehci, 0, &ehci->regs->segment); <------- here
#if 0
// this is deeply broken on almost all architectures

		if (!dma_set_mask(hcd->self.controller, DMA_BIT_MASK(64)))
			ehci_info(ehci, "enabled 64bit DMA\n"); #endif
	}

C.

author
#6 lore raw

RE: [PATCH v6 02/11] hw/usb/hcd-ehci: Change descriptor addresses to 64-bit with migration compatibility

In reply to Jamin Lin

Hi Cédric, Philippe

Please help to review.

hw/usb/ehci: Improve migration helper and property naming
https://patchwork.kernel.org/project/qemu-devel/list/?series=1090233

Thanks,
Jamin

Quoted 316 lines

Subject: RE: [PATCH v6 02/11] hw/usb/hcd-ehci: Change descriptor addresses to 64-bit with migration compatibility

Hi Philippe

Quoted 73 lines

Subject: Re: [PATCH v6 02/11] hw/usb/hcd-ehci: Change descriptor addresses to 64-bit with migration compatibility

Quoted 46 lines · Jamin Lin

Change internal EHCI descriptor addresses from uint32_t to uint64_t.

The following fields are updated:
- EHCIPacket::qtdaddr
- EHCIQueue::{qhaddr, qtdaddr}
- EHCIState::{a_fetch_addr, p_fetch_addr}

Update get_dwords() and put_dwords() to take 64-bit addresses and propagate the type change through the descriptor traversal paths.

Adjust NLPTR_GET() to operate on 64-bit values:

     #define NLPTR_GET(x) ((x) & ~0x1fULL)

so that link pointer masking works correctly when descriptor addresses exceed 32-bit space. The previous mask (0xffffffe0) implicitly truncated addresses to 32 bits.

This patch does not change the on-wire descriptor layout yet. It only removes the internal 32-bit address limit and prepares for later patches that will add full 64-bit QH/qTD/iTD/siTD support.

Update the EHCI trace-events prototypes for QH, qTD, iTD, and siTD to use uint64_t for the address argument and print it with PRIx64. This ensures full 64-bit addresses are shown in trace output and improves debugging of queue heads and transfer descriptors.

Migration compatibility:

To preserve backward migration compatibility, keep the legacy 32-bit fetch address fields (a_fetch_addr_32, p_fetch_addr_32) alongside the new 64-bit fields.

Migration format is selected using a machine compat property "x-migrate-fetch-addr-64bit":

- Old machine types migrate 32-bit fetch addresses
- New machine types migrate full 64-bit fetch addresses

This is implemented using VMSTATE_UINT32_TEST() and VMSTATE_UINT64_TEST() so that only the appropriate format is migrated.

In pre_save, the 32-bit shadow fields are populated when migrating to old machine types. In post_load, the 32-bit values are restored into the 64-bit fields when loading old migration streams.

Quoted 7 lines

No functional change.

---
  hw/usb/hcd-ehci.h   | 17 +++++++----
  hw/core/machine.c   |  5 +++-
  hw/usb/hcd-ehci.c   | 72

++++++++++++++++++++++++++++++++-------------

Quoted 15 lines
  hw/usb/trace-events | 24 +++++++--------
  4 files changed, 78 insertions(+), 40 deletions(-)
(quoted diff) +1 -0
@@ -256,6 +256,7 @@ struct EHCIState {
 
      /* properties */
      uint32_t maxframes;
+    bool migrate_fetch_addr_64bit;
 
      /*
       *  EHCI spec version 1.0 Section 2.3 @@ -294,8 +295,10 @@
struct EHCIState {
      EHCIQueueHead pqueues;
Quoted 10 lines

Review

Quoted 5 lines
      /* which address to look at next */
(quoted diff) +2 -2
-    uint32_t a_fetch_addr;
-    uint32_t p_fetch_addr;
+    uint32_t a_fetch_addr_32;
+    uint32_t p_fetch_addr_32;

Technically '?_fetch_addr_32' aren't "which address to look at next" anymore, but simple migration helpers. Could be worth adding a comment? And rename including 'migrat{e,ion}'? Can be cleared on top.

Will update to

/*
* Migration compatibility fields for old machine types that only
* support 32-bit fetch addresses in the migration stream. *
* New machine types migrate the full 64-bit runtime fetch address. */
uint32_t migrate_a_fetch_addr;
uint32_t migrate_p_fetch_addr;

/*
* Controls migration stream compatibility for old machine types.
* Old machine types only transfer 32-bit fetch addresses. */
bool migrate_fetch_addr_64bit;

Quoted 24 lines
Quoted 24 lines
(quoted diff) +5 -1
+    uint64_t a_fetch_addr;
+    uint64_t p_fetch_addr;
 
      USBPacket ipacket;
      QEMUSGList isgl;
@@ -308,7 +311,9 @@ struct EHCIState {
  };
 
  #define DEFINE_EHCI_COMMON_PROPERTIES(_state) \
-    DEFINE_PROP_UINT32("maxframes", _state, ehci.maxframes, 128)
+    DEFINE_PROP_UINT32("maxframes", _state, ehci.maxframes, 128), \
+    DEFINE_PROP_BOOL("x-migrate-fetch-addr-64bit", _state, \
+                     ehci.migrate_fetch_addr_64bit, true)
 
  extern const VMStateDescription vmstate_ehci;
 
hw/core/machine.c index renamed from hw/core/machine.c +0 -0

1b661fd36a..54336da294 100644

hw/core/machine.c +0 -0
@@ -39,7 +39,10 @@
  #include "hw/acpi/generic_event_device.h"
  #include "qemu/audio.h"
 
Quoted 5 lines

hw/core/machine.c line 45

Quoted 5 lines
hw/core/machine.c +3 -1
-GlobalProperty hw_compat_11_0[] = {};
+GlobalProperty hw_compat_11_0[] = {
+    { "sysbus-ehci-usb", "x-migrate-fetch-addr-64bit", "off" },
+    { "pci-ehci-usb", "x-migrate-fetch-addr-64bit", "off" }, };
  const size_t hw_compat_11_0_len =

G_N_ELEMENTS(hw_compat_11_0);

Quoted 53 lines

Good.

hw/core/machine.c static int usb_ehci_pre_save(void *opaque) · line 2438

Quoted 13 lines
hw/core/machine.c +8 -0
+static bool ehci_fetch_addr_64_needed(void *opaque, int version_id) {
+    EHCIState *s = opaque;
+
+    return s->migrate_fetch_addr_64bit; }
+
+static bool ehci_fetch_addr_32_needed(void *opaque, int version_id) {
+    return !ehci_fetch_addr_64_needed(opaque, version_id); }
+
  static int usb_ehci_pre_save(void *opaque)
  {
      EHCIState *ehci = opaque;
@@ -2424,6 +2438,11 @@ static int usb_ehci_pre_save(void *opaque)
      ehci->last_run_ns -= (ehci->frindex - new_frindex) *

UFRAME_TIMER_NS;

Quoted 11 lines
hw/core/machine.c +5 -0
      ehci->frindex = new_frindex;
 
+    if (!ehci->migrate_fetch_addr_64bit) {
+        ehci->a_fetch_addr_32 = ehci->a_fetch_addr;
+        ehci->p_fetch_addr_32 = ehci->p_fetch_addr;
+    }
+
      return 0;
  }
 
@@ -2444,6 +2463,11 @@ static int usb_ehci_post_load(void *opaque,
Quoted 1 line

int

version_id)

Quoted 23 lines
          }
      }
hw/core/machine.c +11 -2
+    if (!s->migrate_fetch_addr_64bit) {
+        s->a_fetch_addr = s->a_fetch_addr_32;
+        s->p_fetch_addr = s->p_fetch_addr_32;
+    }
+
      return 0;
  }
@@ -2504,8 +2528,14 @@ const VMStateDescription vmstate_ehci = {
          /* schedule state */
          VMSTATE_UINT32(astate, EHCIState),
          VMSTATE_UINT32(pstate, EHCIState),
-        VMSTATE_UINT32(a_fetch_addr, EHCIState),
-        VMSTATE_UINT32(p_fetch_addr, EHCIState),
+        VMSTATE_UINT32_TEST(a_fetch_addr_32, EHCIState,
+                            ehci_fetch_addr_32_needed),
+        VMSTATE_UINT32_TEST(p_fetch_addr_32, EHCIState,
+                            ehci_fetch_addr_32_needed),
+        VMSTATE_UINT64_TEST(a_fetch_addr, EHCIState,
+                            ehci_fetch_addr_64_needed),
Quoted 9 lines

hw/core/machine.c const VMStateDescription vmstate_ehci = { · line 2538

Quoted 2 lines
hw/core/machine.c +2 -0
+        VMSTATE_UINT64_TEST(p_fetch_addr, EHCIState,
+                            ehci_fetch_addr_64_needed),

Hmmmm. I didn't review all previous versions, for my own education, why don't we use VMSTATE_UINT64_V() here? Did you test (pre-11 <-> this
series) migration? My understanding was we need to maintain the stream size to avoid mismatch.

Please see the migration test steps here:
https://patchwork.kernel.org/project/qemu-devel/patch/20260423051804.362
[email protected]/#26911317
I tested migration compatibility with both v11 and v10.2.

I originally tried using VMSTATE_UINT64_V(). That implementation was similar to the v3 version of this series, which you previously reviewed:
https://patchwork.kernel.org/project/qemu-devel/patch/20260416014928.127
[email protected]/

Unfortunately, that approach caused migration failures.

The issue is that older machine types do not contain the 64-bit fetch address fields in their migration ABI. Simply adding VMSTATE_UINT64_V() changes the migration stream layout, which breaks compatibility with older machine types / older QEMU versions.

The migration failure looks like this:
qemu-system-x86_64: Missing section footer for 0000:00:03.0/ehci
qemu-system-x86_64: load of migration failed: Invalid argument: Section footer error, section_id: 42

Because of this, I used migrate_fetch_addr_64bit to preserve the original 32-bit migration stream for older compatible machine types while enabling the 64-bit migration format only for newer machine types.

Thanks,
Jamin

hw/usb/hcd-ehci.h index renamed from hw/usb/hcd-ehci.h +0 -0

3d57e1de5f..7b3e92895a 100644

hw/usb/hcd-ehci.h +2 -2
@@ -303,8 +303,8 @@ struct EHCIState {
     EHCIQueueHead pqueues;
 
     /* which address to look at next */
-    uint32_t a_fetch_addr_32;
-    uint32_t p_fetch_addr_32;
+    uint32_t migrate_a_fetch_addr;
+    uint32_t migrate_p_fetch_addr;
     uint64_t a_fetch_addr;
     uint64_t p_fetch_addr;
 
hw/usb/hcd-ehci.c index renamed from hw/usb/hcd-ehci.c +0 -0

62c3770ac5..d3b9890726 100644

hw/usb/hcd-ehci.c +6 -12
@@ -2459,16 +2459,9 @@ static USBBusOps ehci_bus_ops_standalone = {
     .wakeup_endpoint = ehci_wakeup_endpoint,  };
 
-static bool ehci_fetch_addr_64_needed(void *opaque, int version_id)
+static bool ehci_core_is_before_version_3(void *opaque, int version_id)
 {
-    EHCIState *s = opaque;
-
-    return s->migrate_fetch_addr_64bit;
-}
-
-static bool ehci_fetch_addr_32_needed(void *opaque, int version_id) -{
-    return !ehci_fetch_addr_64_needed(opaque, version_id);
+    return version_id < 3;
 }
 
 static int usb_ehci_pre_save(void *opaque) @@ -2482,8 +2475,8 @@ static
 int usb_ehci_pre_save(void *opaque)
     ehci->frindex = new_frindex;
 
     if (!ehci->migrate_fetch_addr_64bit) {
-        ehci->a_fetch_addr_32 = ehci->a_fetch_addr;
-        ehci->p_fetch_addr_32 = ehci->p_fetch_addr;
+        ehci->migrate_a_fetch_addr = ehci->a_fetch_addr;
+        ehci->migrate_p_fetch_addr = ehci->p_fetch_addr;
     }
 
     return 0;
@@ -2507,8 +2500,8 @@ static int usb_ehci_post_load(void *opaque, int
 version_id)
     }
 
     if (!s->migrate_fetch_addr_64bit) {
-        s->a_fetch_addr = s->a_fetch_addr_32;
-        s->p_fetch_addr = s->p_fetch_addr_32;
+        s->a_fetch_addr = s->migrate_a_fetch_addr;
+        s->p_fetch_addr = s->migrate_p_fetch_addr;
     }
 
     return 0;
hw/usb/hcd-ehci.c +1 -1
@@ -2542,7 +2535,7 @@ static void usb_ehci_vm_state_change(void
 *opaque, bool running, RunState state)
 
 const VMStateDescription vmstate_ehci = {
     .name        = "ehci-core",
-    .version_id  = 2,
+    .version_id  = 3,
     .minimum_version_id  = 1,
     .pre_save    = usb_ehci_pre_save,

.post_load = usb_ehci_post_load,

hw/usb/hcd-ehci.c +6 -8
@@ -2571,14 +2564,12 @@ const VMStateDescription vmstate_ehci = {
         /* schedule state */
         VMSTATE_UINT32(astate, EHCIState),
         VMSTATE_UINT32(pstate, EHCIState),
-        VMSTATE_UINT32_TEST(a_fetch_addr_32, EHCIState,
-                            ehci_fetch_addr_32_needed),
-        VMSTATE_UINT32_TEST(p_fetch_addr_32, EHCIState,
-                            ehci_fetch_addr_32_needed),
-        VMSTATE_UINT64_TEST(a_fetch_addr, EHCIState,
-                            ehci_fetch_addr_64_needed),
-        VMSTATE_UINT64_TEST(p_fetch_addr, EHCIState,
-                            ehci_fetch_addr_64_needed),
+        VMSTATE_UINT32_TEST(migrate_a_fetch_addr, EHCIState,
+                            ehci_core_is_before_version_3),
+        VMSTATE_UINT32_TEST(migrate_p_fetch_addr, EHCIState,
+                            ehci_core_is_before_version_3),
+        VMSTATE_UINT64_V(a_fetch_addr, EHCIState, 3),
+        VMSTATE_UINT64_V(p_fetch_addr, EHCIState, 3),
         VMSTATE_END_OF_LIST()
     }
 };
#24 lore raw
In reply to Cédric Le Goater
Quoted 134 lines · Cédric Le Goater
Quoted 130 lines · Jamin Lin

EHCI supports 64-bit addressing through the CTRLDSSEGMENT register, which provides the upper 32 bits of descriptor addresses when the controller advertises 64-bit capability.

Currently QEMU EHCI model only partially supports this functionality and descriptor addresses are effectively treated as 32-bit. This becomes problematic on systems where system memory is located above the 4GB boundary.

The Linux EHCI driver enables 64-bit addressing if the controller advertises the capability. During initialization it programs the segment register to zero:

   https://github.com/torvalds/linux/blob/master/drivers/usb/host/ehci-hcd.c#L600

The driver also notes that descriptor structures allocated from the DMA pool use segment zero semantics. Descriptor memory is allocated using the DMA API and platforms may configure a 64-bit DMA mask, allowing descriptor memory to be placed above 4GB.

On AST2700 platforms, system DRAM is mapped at 0x400000000. As a result, descriptor addresses constructed directly from the EHCI registers do not match the actual system addresses used by the controller when accessing queue heads (QH) and queue element transfer descriptors (qTD).

This patch series implements full 64-bit descriptor addressing support in the EHCI emulation. Descriptor address handling is updated to use 64-bit values and the descriptor structures (QH, qTD, iTD and siTD) are extended to support the upper address bits provided by the segment register.

Add a ctrldssegment-default property so platforms can provide a descriptor address offset when constructing descriptor addresses. This allows systems where DRAM resides above 4GB to access EHCI descriptors correctly.

The AST2700 machine uses this property to account for its DRAM mapping at 0x400000000 and enables 64-bit EHCI DMA addressing.

Test Result:
1. EHCI 32bits with ast2600-evb machine
Command line:
./build/qemu-system-arm \
   -machine ast2600-evb \
   -m 1G \
   -drive file=image-bmc,if=mtd,format=raw \
   -nographic \
   -device usb-kbd,bus=usb-bus.1,id=mykbd \
   -drive id=usbdisk,if=none,file=image0.ext4,format=raw \
   -device usb-storage,bus=usb-bus.1,id=mystorage,drive=usbdisk
   -snapshot \
   -nographic
Result:
unable to initialize usb specBus 001 Device 001: ID 1d6b:0002 Linux 6.18.3-v00.08.01-g172b7e27a30d ehci_hcd EHCI Host Controller Bus 001 Device 002: ID 0627:0001 QEMU QEMU USB Keyboard
Bus 001 Device 003: ID 46f4:0001 QEMU QEMU USB HARDDRIVE
Bus 002 Device 001: ID 1d6b:0001 Linux 6.18.3-v00.08.01-g172b7e27a30d uhci_hcd Generic UHCI Host Controller

2. EHCI 64bits with ast2700a2-evb machine
Command line:
./build/qemu-system-aarch64 -M ast2700a2-evb -nographic\
  -bios ast27x0_bootrom.bin \
  -drive file=image-bmc,format=raw,if=mtd \
  -snapshot \
  -device usb-kbd,bus=usb-bus.3,id=mykbd \
  -drive id=usbdisk,if=none,file=image0.ext4,format=raw \
  -device usb-storage,bus=usb-bus.3,id=mystorage,drive=usbdisk Result:

root@ast2700-default:~# lsusb
unable to initialize usb specBus 001 Device 001: ID 1d6b:0001 Linux 6.18.3-v00.08.01-g172b7e27a30d uhci_hcd Generic UHCI Host Controller
Bus 002 Device 001: ID 1d6b:0002 Linux 6.18.3-v00.08.01-g172b7e27a30d ehci_hcd EHCI Host Controller
Bus 002 Device 002: ID 0627:0001 QEMU QEMU USB Keyboard
Bus 002 Device 003: ID 46f4:0001 QEMU QEMU USB HARDDRIVE
v1
  1. Fix checkpatch coding style issues
  2. Implement 64-bit addressing for QH/qTD/iTD/siTD descriptors
  3. Add descriptor address offset property
  4. Enable 64-bit EHCI DMA addressing on AST2700
  5. Configure descriptor address offset for AST2700

v2
  1. Remove unused EHCIfstn structure and dead code
  2. Replace fprintf(stderr, ...) with qemu_log_mask(LOG_GUEST_ERROR)
  3. Replace DPRINTF debug logs with trace events
  4. Add functional tests for USB EHCI on AST2600 and AST2700 A1/A2
  5. Fix review issue

v3:
  1. Add Migration version test function
  2. Add EHCI 64-bit buffer pointer fields description in commit log

v4:
  1. Reorder patches in the series
  2. Fix EHCI migration issues
  3. Introduce a common properties macro for both sysbus and PCI
  4. Drop the descriptor address offset property
  5. Add ctrldssegment-default property
  6. Address review comments

v5:
  1. Add 11.0 machine compatibility properties

v6:
  1. Update reviewer suggested improvements.
Jamin Lin (11):
   tests/functional/arm/test_aspeed_ast2600_sdk: Add USB EHCI test for AST2600 SDK
   hw/usb/hcd-ehci: Change descriptor addresses to 64-bit with migration compatibility
   hw/usb/hcd-ehci: Add property to advertise 64-bit addressing capability
   hw/usb/hcd-ehci: Implement 64-bit QH descriptor addressing hw/usb/hcd-ehci: Implement 64-bit qTD descriptor addressing hw/usb/hcd-ehci: Implement 64-bit iTD descriptor addressing hw/usb/hcd-ehci: Implement 64-bit siTD descriptor addressing hw/usb/hcd-ehci: Add ctrldssegment-default property hw/arm/aspeed_ast27x0: Set EHCI ctrldssegment-default hw/arm/aspeed_ast27x0: Enable 64-bit EHCI DMA addressing tests/functional/aarch64/test_aspeed_ast2700: Add USB EHCI test for AST2700 A1/A2

  hw/usb/hcd-ehci.h                             |  29 +++- hw/arm/aspeed_ast27x0.c                       |   5 + hw/core/machine.c                             |   5 +- hw/usb/hcd-ehci.c                             | 162 +++++++++++++----- hw/usb/trace-events                           |  26 +-- .../aarch64/test_aspeed_ast2700a1.py          |   7 + .../aarch64/test_aspeed_ast2700a2.py          |   7 + .../functional/arm/test_aspeed_ast2600_sdk.py |   7 +

Quoted 2 lines
  8 files changed, 185 insertions(+), 63 deletions(-)
Quoted 3 lines

Applied to

    https://github.com/legoater/qemu aspeed-next

Jamin,

I dropped the series from aspeed-next and kept :

tests/functional/arm/test_aspeed_ast2600_sdk: Add USB EHCI test for AST2600 SDK

The USB subsystem is orphan but it is still used in the virt world. This series is modifying the EHCI internals in such way that migration is impacted and the CTRLDSSEGMENT (Control Data Structure Segment Register) implementation is incomplete AFAICT. From EHCI specs :

This register allows the host software to locate all control data structures within the *same 4 Gigabyte memory segment*.

Linux driver seems buggy too.

Overall, the risk of regression is too high given the time I can dedicate to this topic.

Thanks for all the good work you done there, specially for migration. If someone can Ack the series and step forward to become USB maintainer, then I will reconsider.

C.

author
#25 lore raw

Hi flwu, Nabih

Quoted 17 lines

Subject: Re: [PATCH v6 00/11] hw/usb/ehci: Add 64-bit descriptor addressing support
Jamin,

I dropped the series from aspeed-next and kept :

tests/functional/arm/test_aspeed_ast2600_sdk: Add USB EHCI test for
AST2600 SDK

The USB subsystem is orphan but it is still used in the virt world. This series is modifying the EHCI internals in such way that migration is impacted and the CTRLDSSEGMENT (Control Data Structure Segment Register) implementation is incomplete AFAICT. From EHCI specs :

This register allows the host software to locate all control data structures within the *same 4 Gigabyte memory segment*.

Quoted 10 lines

Linux driver seems buggy too.

Overall, the risk of regression is too high given the time I can dedicate to this topic.

Thanks for all the good work you done there, specially for migration. If someone can Ack the series and step forward to become USB maintainer, then I will reconsider.

Just to let you know the upstream status.
If you want to use USB EHCI on AST2700, please apply this patch series to your codebase.

Thanks,
Jamin

Quoted 1 line

C.

#26 lore raw
In reply to Cédric Le Goater

+ Gerd, Marc-André,

May be they can provide some more feedback.

Quoted 153 lines · Cédric Le Goater
Quoted 134 lines · Cédric Le Goater
Quoted 130 lines · Jamin Lin

EHCI supports 64-bit addressing through the CTRLDSSEGMENT register, which provides the upper 32 bits of descriptor addresses when the controller advertises 64-bit capability.

Currently QEMU EHCI model only partially supports this functionality and descriptor addresses are effectively treated as 32-bit. This becomes problematic on systems where system memory is located above the 4GB boundary.

The Linux EHCI driver enables 64-bit addressing if the controller advertises the capability. During initialization it programs the segment register to zero:

   https://github.com/torvalds/linux/blob/master/drivers/usb/host/ehci-hcd.c#L600

The driver also notes that descriptor structures allocated from the DMA pool use segment zero semantics. Descriptor memory is allocated using the DMA API and platforms may configure a 64-bit DMA mask, allowing descriptor memory to be placed above 4GB.

On AST2700 platforms, system DRAM is mapped at 0x400000000. As a result, descriptor addresses constructed directly from the EHCI registers do not match the actual system addresses used by the controller when accessing queue heads (QH) and queue element transfer descriptors (qTD).

This patch series implements full 64-bit descriptor addressing support in the EHCI emulation. Descriptor address handling is updated to use 64-bit values and the descriptor structures (QH, qTD, iTD and siTD) are extended to support the upper address bits provided by the segment register.

Add a ctrldssegment-default property so platforms can provide a descriptor address offset when constructing descriptor addresses. This allows systems where DRAM resides above 4GB to access EHCI descriptors correctly.

The AST2700 machine uses this property to account for its DRAM mapping at 0x400000000 and enables 64-bit EHCI DMA addressing.

Test Result:
1. EHCI 32bits with ast2600-evb machine
Command line:
./build/qemu-system-arm \
   -machine ast2600-evb \
   -m 1G \
   -drive file=image-bmc,if=mtd,format=raw \
   -nographic \
   -device usb-kbd,bus=usb-bus.1,id=mykbd \
   -drive id=usbdisk,if=none,file=image0.ext4,format=raw \
   -device usb-storage,bus=usb-bus.1,id=mystorage,drive=usbdisk
   -snapshot \
   -nographic
Result:
unable to initialize usb specBus 001 Device 001: ID 1d6b:0002 Linux 6.18.3-v00.08.01-g172b7e27a30d ehci_hcd EHCI Host Controller Bus 001 Device 002: ID 0627:0001 QEMU QEMU USB Keyboard
Bus 001 Device 003: ID 46f4:0001 QEMU QEMU USB HARDDRIVE
Bus 002 Device 001: ID 1d6b:0001 Linux 6.18.3-v00.08.01-g172b7e27a30d uhci_hcd Generic UHCI Host Controller

2. EHCI 64bits with ast2700a2-evb machine
Command line:
./build/qemu-system-aarch64 -M ast2700a2-evb -nographic\
  -bios ast27x0_bootrom.bin \
  -drive file=image-bmc,format=raw,if=mtd \
  -snapshot \
  -device usb-kbd,bus=usb-bus.3,id=mykbd \
  -drive id=usbdisk,if=none,file=image0.ext4,format=raw \
  -device usb-storage,bus=usb-bus.3,id=mystorage,drive=usbdisk Result:

root@ast2700-default:~# lsusb
unable to initialize usb specBus 001 Device 001: ID 1d6b:0001 Linux 6.18.3-v00.08.01-g172b7e27a30d uhci_hcd Generic UHCI Host Controller
Bus 002 Device 001: ID 1d6b:0002 Linux 6.18.3-v00.08.01-g172b7e27a30d ehci_hcd EHCI Host Controller
Bus 002 Device 002: ID 0627:0001 QEMU QEMU USB Keyboard
Bus 002 Device 003: ID 46f4:0001 QEMU QEMU USB HARDDRIVE
v1
  1. Fix checkpatch coding style issues
  2. Implement 64-bit addressing for QH/qTD/iTD/siTD descriptors
  3. Add descriptor address offset property
  4. Enable 64-bit EHCI DMA addressing on AST2700
  5. Configure descriptor address offset for AST2700

v2
  1. Remove unused EHCIfstn structure and dead code
  2. Replace fprintf(stderr, ...) with qemu_log_mask(LOG_GUEST_ERROR)
  3. Replace DPRINTF debug logs with trace events
  4. Add functional tests for USB EHCI on AST2600 and AST2700 A1/A2
  5. Fix review issue

v3:
  1. Add Migration version test function
  2. Add EHCI 64-bit buffer pointer fields description in commit log

v4:
  1. Reorder patches in the series
  2. Fix EHCI migration issues
  3. Introduce a common properties macro for both sysbus and PCI
  4. Drop the descriptor address offset property
  5. Add ctrldssegment-default property
  6. Address review comments

v5:
  1. Add 11.0 machine compatibility properties

v6:
  1. Update reviewer suggested improvements.
Jamin Lin (11):
   tests/functional/arm/test_aspeed_ast2600_sdk: Add USB EHCI test for AST2600 SDK
   hw/usb/hcd-ehci: Change descriptor addresses to 64-bit with migration compatibility
   hw/usb/hcd-ehci: Add property to advertise 64-bit addressing capability
   hw/usb/hcd-ehci: Implement 64-bit QH descriptor addressing hw/usb/hcd-ehci: Implement 64-bit qTD descriptor addressing hw/usb/hcd-ehci: Implement 64-bit iTD descriptor addressing hw/usb/hcd-ehci: Implement 64-bit siTD descriptor addressing hw/usb/hcd-ehci: Add ctrldssegment-default property hw/arm/aspeed_ast27x0: Set EHCI ctrldssegment-default hw/arm/aspeed_ast27x0: Enable 64-bit EHCI DMA addressing tests/functional/aarch64/test_aspeed_ast2700: Add USB EHCI test for AST2700 A1/A2

  hw/usb/hcd-ehci.h                             |  29 +++- hw/arm/aspeed_ast27x0.c                       |   5 + hw/core/machine.c                             |   5 +- hw/usb/hcd-ehci.c                             | 162 +++++++++++++----- hw/usb/trace-events                           |  26 +-- .../aarch64/test_aspeed_ast2700a1.py          |   7 + .../aarch64/test_aspeed_ast2700a2.py          |   7 + .../functional/arm/test_aspeed_ast2600_sdk.py |   7 +

Quoted 2 lines
  8 files changed, 185 insertions(+), 63 deletions(-)
Quoted 3 lines

Applied to

     https://github.com/legoater/qemu aspeed-next

Jamin,

I dropped the series from aspeed-next and kept :

  tests/functional/arm/test_aspeed_ast2600_sdk: Add USB EHCI test for AST2600 SDK

The USB subsystem is orphan but it is still used in the virt world. This series is modifying the EHCI internals in such way that migration is impacted and the CTRLDSSEGMENT (Control Data Structure Segment Register) implementation is incomplete AFAICT. From EHCI specs :

   This register allows the host software to locate all control data structures within the *same 4 Gigabyte memory segment*.

Quoted 8 lines

Linux driver seems buggy too.

Overall, the risk of regression is too high given the time I can dedicate to this topic.

Thanks for all the good work you done there, specially for migration. If someone can Ack the series and step forward to become USB maintainer, then I will reconsider.

Thanks,

C.

#27 lore raw
Quoted 167 lines · Cédric Le Goater

+ Gerd, Marc-André,

May be they can provide some more feedback.

Quoted 162 lines · Cédric Le Goater
Quoted 135 lines · Cédric Le Goater
Quoted 126 lines · Jamin Lin

EHCI supports 64-bit addressing through the CTRLDSSEGMENT register, which provides the upper 32 bits of descriptor addresses when the controller advertises 64-bit capability.

Currently QEMU EHCI model only partially supports this functionality and descriptor addresses are effectively treated as 32-bit. This becomes problematic on systems where system memory is located above the 4GB boundary.

The Linux EHCI driver enables 64-bit addressing if the controller advertises the capability. During initialization it programs the segment register to zero:

   https://github.com/torvalds/linux/blob/master/drivers/usb/host/ ehci-hcd.c#L600

The driver also notes that descriptor structures allocated from the DMA pool use segment zero semantics. Descriptor memory is allocated using the DMA API and platforms may configure a 64-bit DMA mask, allowing descriptor memory to be placed above 4GB.

On AST2700 platforms, system DRAM is mapped at 0x400000000. As a result, descriptor addresses constructed directly from the EHCI registers do not match the actual system addresses used by the controller when accessing queue heads (QH) and queue element transfer descriptors (qTD).

This patch series implements full 64-bit descriptor addressing support in the EHCI emulation. Descriptor address handling is updated to use 64-bit values and the descriptor structures (QH, qTD, iTD and siTD) are extended to support the upper address bits provided by the segment register.

Add a ctrldssegment-default property so platforms can provide a descriptor address offset when constructing descriptor addresses. This allows systems where DRAM resides above 4GB to access EHCI descriptors correctly.

The AST2700 machine uses this property to account for its DRAM mapping at 0x400000000 and enables 64-bit EHCI DMA addressing.

Test Result:
1. EHCI 32bits with ast2600-evb machine
Command line:
./build/qemu-system-arm \
   -machine ast2600-evb \
   -m 1G \
   -drive file=image-bmc,if=mtd,format=raw \
   -nographic \
   -device usb-kbd,bus=usb-bus.1,id=mykbd \
   -drive id=usbdisk,if=none,file=image0.ext4,format=raw \
   -device usb-storage,bus=usb-bus.1,id=mystorage,drive=usbdisk
   -snapshot \
   -nographic
Result:
unable to initialize usb specBus 001 Device 001: ID 1d6b:0002 Linux 6.18.3-v00.08.01-g172b7e27a30d ehci_hcd EHCI Host Controller Bus 001 Device 002: ID 0627:0001 QEMU QEMU USB Keyboard
Bus 001 Device 003: ID 46f4:0001 QEMU QEMU USB HARDDRIVE
Bus 002 Device 001: ID 1d6b:0001 Linux 6.18.3-v00.08.01- g172b7e27a30d uhci_hcd Generic UHCI Host Controller

2. EHCI 64bits with ast2700a2-evb machine
Command line:
./build/qemu-system-aarch64 -M ast2700a2-evb -nographic\
  -bios ast27x0_bootrom.bin \
  -drive file=image-bmc,format=raw,if=mtd \
  -snapshot \
  -device usb-kbd,bus=usb-bus.3,id=mykbd \
  -drive id=usbdisk,if=none,file=image0.ext4,format=raw \
  -device usb-storage,bus=usb-bus.3,id=mystorage,drive=usbdisk Result:

root@ast2700-default:~# lsusb
unable to initialize usb specBus 001 Device 001: ID 1d6b:0001 Linux 6.18.3-v00.08.01-g172b7e27a30d uhci_hcd Generic UHCI Host Controller
Bus 002 Device 001: ID 1d6b:0002 Linux 6.18.3-v00.08.01- g172b7e27a30d ehci_hcd EHCI Host Controller
Bus 002 Device 002: ID 0627:0001 QEMU QEMU USB Keyboard
Bus 002 Device 003: ID 46f4:0001 QEMU QEMU USB HARDDRIVE
v1
  1. Fix checkpatch coding style issues
  2. Implement 64-bit addressing for QH/qTD/iTD/siTD descriptors
  3. Add descriptor address offset property
  4. Enable 64-bit EHCI DMA addressing on AST2700
  5. Configure descriptor address offset for AST2700

v2
  1. Remove unused EHCIfstn structure and dead code
  2. Replace fprintf(stderr, ...) with qemu_log_mask(LOG_GUEST_ERROR)
  3. Replace DPRINTF debug logs with trace events
  4. Add functional tests for USB EHCI on AST2600 and AST2700 A1/A2
  5. Fix review issue

v3:
  1. Add Migration version test function
  2. Add EHCI 64-bit buffer pointer fields description in commit log

v4:
  1. Reorder patches in the series
  2. Fix EHCI migration issues
  3. Introduce a common properties macro for both sysbus and PCI
  4. Drop the descriptor address offset property
  5. Add ctrldssegment-default property
  6. Address review comments

v5:
  1. Add 11.0 machine compatibility properties

v6:
  1. Update reviewer suggested improvements.
Jamin Lin (11):
   tests/functional/arm/test_aspeed_ast2600_sdk: Add USB EHCI test for AST2600 SDK
   hw/usb/hcd-ehci: Change descriptor addresses to 64-bit with migration compatibility
   hw/usb/hcd-ehci: Add property to advertise 64-bit addressing capability
   hw/usb/hcd-ehci: Implement 64-bit QH descriptor addressing hw/usb/hcd-ehci: Implement 64-bit qTD descriptor addressing hw/usb/hcd-ehci: Implement 64-bit iTD descriptor addressing hw/usb/hcd-ehci: Implement 64-bit siTD descriptor addressing hw/usb/hcd-ehci: Add ctrldssegment-default property hw/arm/aspeed_ast27x0: Set EHCI ctrldssegment-default hw/arm/aspeed_ast27x0: Enable 64-bit EHCI DMA addressing tests/functional/aarch64/test_aspeed_ast2700: Add USB EHCI test for AST2700 A1/A2

  hw/usb/hcd-ehci.h                             |  29 +++- hw/arm/aspeed_ast27x0.c                       |   5 + hw/core/machine.c                             |   5 +- hw/usb/hcd-ehci.c                             | 162 ++++++++++++

Quoted 7 lines
+-----
  hw/usb/trace-events                           |  26 +--
  .../aarch64/test_aspeed_ast2700a1.py          |   7 +
  .../aarch64/test_aspeed_ast2700a2.py          |   7 +
  .../functional/arm/test_aspeed_ast2600_sdk.py |   7 +
  8 files changed, 185 insertions(+), 63 deletions(-)
Quoted 3 lines

Applied to

     https://github.com/legoater/qemu aspeed-next

Jamin,

I dropped the series from aspeed-next and kept :

   tests/functional/arm/test_aspeed_ast2600_sdk: Add USB EHCI test for AST2600 SDK

The USB subsystem is orphan but it is still used in the virt world. This series is modifying the EHCI internals in such way that migration is impacted and the CTRLDSSEGMENT (Control Data Structure Segment Register) implementation is incomplete AFAICT. From EHCI specs :

    This register allows the host software to locate all control data structures within the *same 4 Gigabyte memory segment*.

Linux driver seems buggy too.

Overall, the risk of regression is too high given the time I can dedicate to this topic.

Thanks for all the good work you done there, specially for migration. If someone can Ack the series and step forward to become USB maintainer, then I will reconsider.

I'm not an USB expert / maintainer but FWIW this series looked good enough to me when I looked at it. Maybe consider to merge after the 11.1 release?

#28 lore raw
Quoted 164 lines · Philippe Mathieu-Daudé
Quoted 163 lines · Cédric Le Goater

+ Gerd, Marc-André,

May be they can provide some more feedback.

Quoted 158 lines · Cédric Le Goater
Quoted 134 lines · Cédric Le Goater
Quoted 130 lines · Jamin Lin

EHCI supports 64-bit addressing through the CTRLDSSEGMENT register, which provides the upper 32 bits of descriptor addresses when the controller advertises 64-bit capability.

Currently QEMU EHCI model only partially supports this functionality and descriptor addresses are effectively treated as 32-bit. This becomes problematic on systems where system memory is located above the 4GB boundary.

The Linux EHCI driver enables 64-bit addressing if the controller advertises the capability. During initialization it programs the segment register to zero:

   https://github.com/torvalds/linux/blob/master/drivers/usb/host/ ehci-hcd.c#L600

The driver also notes that descriptor structures allocated from the DMA pool use segment zero semantics. Descriptor memory is allocated using the DMA API and platforms may configure a 64-bit DMA mask, allowing descriptor memory to be placed above 4GB.

On AST2700 platforms, system DRAM is mapped at 0x400000000. As a result, descriptor addresses constructed directly from the EHCI registers do not match the actual system addresses used by the controller when accessing queue heads (QH) and queue element transfer descriptors (qTD).

This patch series implements full 64-bit descriptor addressing support in the EHCI emulation. Descriptor address handling is updated to use 64-bit values and the descriptor structures (QH, qTD, iTD and siTD) are extended to support the upper address bits provided by the segment register.

Add a ctrldssegment-default property so platforms can provide a descriptor address offset when constructing descriptor addresses. This allows systems where DRAM resides above 4GB to access EHCI descriptors correctly.

The AST2700 machine uses this property to account for its DRAM mapping at 0x400000000 and enables 64-bit EHCI DMA addressing.

Test Result:
1. EHCI 32bits with ast2600-evb machine
Command line:
./build/qemu-system-arm \
   -machine ast2600-evb \
   -m 1G \
   -drive file=image-bmc,if=mtd,format=raw \
   -nographic \
   -device usb-kbd,bus=usb-bus.1,id=mykbd \
   -drive id=usbdisk,if=none,file=image0.ext4,format=raw \
   -device usb-storage,bus=usb-bus.1,id=mystorage,drive=usbdisk
   -snapshot \
   -nographic
Result:
unable to initialize usb specBus 001 Device 001: ID 1d6b:0002 Linux 6.18.3-v00.08.01-g172b7e27a30d ehci_hcd EHCI Host Controller Bus 001 Device 002: ID 0627:0001 QEMU QEMU USB Keyboard
Bus 001 Device 003: ID 46f4:0001 QEMU QEMU USB HARDDRIVE
Bus 002 Device 001: ID 1d6b:0001 Linux 6.18.3-v00.08.01- g172b7e27a30d uhci_hcd Generic UHCI Host Controller

2. EHCI 64bits with ast2700a2-evb machine
Command line:
./build/qemu-system-aarch64 -M ast2700a2-evb -nographic\
  -bios ast27x0_bootrom.bin \
  -drive file=image-bmc,format=raw,if=mtd \
  -snapshot \
  -device usb-kbd,bus=usb-bus.3,id=mykbd \
  -drive id=usbdisk,if=none,file=image0.ext4,format=raw \
  -device usb-storage,bus=usb-bus.3,id=mystorage,drive=usbdisk Result:

root@ast2700-default:~# lsusb
unable to initialize usb specBus 001 Device 001: ID 1d6b:0001 Linux 6.18.3-v00.08.01-g172b7e27a30d uhci_hcd Generic UHCI Host Controller
Bus 002 Device 001: ID 1d6b:0002 Linux 6.18.3-v00.08.01- g172b7e27a30d ehci_hcd EHCI Host Controller
Bus 002 Device 002: ID 0627:0001 QEMU QEMU USB Keyboard
Bus 002 Device 003: ID 46f4:0001 QEMU QEMU USB HARDDRIVE
v1
  1. Fix checkpatch coding style issues
  2. Implement 64-bit addressing for QH/qTD/iTD/siTD descriptors
  3. Add descriptor address offset property
  4. Enable 64-bit EHCI DMA addressing on AST2700
  5. Configure descriptor address offset for AST2700

v2
  1. Remove unused EHCIfstn structure and dead code
  2. Replace fprintf(stderr, ...) with qemu_log_mask(LOG_GUEST_ERROR)
  3. Replace DPRINTF debug logs with trace events
  4. Add functional tests for USB EHCI on AST2600 and AST2700 A1/A2
  5. Fix review issue

v3:
  1. Add Migration version test function
  2. Add EHCI 64-bit buffer pointer fields description in commit log

v4:
  1. Reorder patches in the series
  2. Fix EHCI migration issues
  3. Introduce a common properties macro for both sysbus and PCI
  4. Drop the descriptor address offset property
  5. Add ctrldssegment-default property
  6. Address review comments

v5:
  1. Add 11.0 machine compatibility properties

v6:
  1. Update reviewer suggested improvements.
Jamin Lin (11):
   tests/functional/arm/test_aspeed_ast2600_sdk: Add USB EHCI test for AST2600 SDK
   hw/usb/hcd-ehci: Change descriptor addresses to 64-bit with migration compatibility
   hw/usb/hcd-ehci: Add property to advertise 64-bit addressing capability
   hw/usb/hcd-ehci: Implement 64-bit QH descriptor addressing hw/usb/hcd-ehci: Implement 64-bit qTD descriptor addressing hw/usb/hcd-ehci: Implement 64-bit iTD descriptor addressing hw/usb/hcd-ehci: Implement 64-bit siTD descriptor addressing hw/usb/hcd-ehci: Add ctrldssegment-default property hw/arm/aspeed_ast27x0: Set EHCI ctrldssegment-default hw/arm/aspeed_ast27x0: Enable 64-bit EHCI DMA addressing tests/functional/aarch64/test_aspeed_ast2700: Add USB EHCI test for AST2700 A1/A2

  hw/usb/hcd-ehci.h                             |  29 +++- hw/arm/aspeed_ast27x0.c                       |   5 + hw/core/machine.c                             |   5 +- hw/usb/hcd-ehci.c                             | 162 ++++++++++++ +----- hw/usb/trace-events                           |  26 +-- .../aarch64/test_aspeed_ast2700a1.py          |   7 + .../aarch64/test_aspeed_ast2700a2.py          |   7 + .../functional/arm/test_aspeed_ast2600_sdk.py |   7 +

Quoted 2 lines
  8 files changed, 185 insertions(+), 63 deletions(-)
Quoted 3 lines

Applied to

     https://github.com/legoater/qemu aspeed-next

Jamin,

I dropped the series from aspeed-next and kept :

   tests/functional/arm/test_aspeed_ast2600_sdk: Add USB EHCI test for AST2600 SDK

The USB subsystem is orphan but it is still used in the virt world. This series is modifying the EHCI internals in such way that migration is impacted and the CTRLDSSEGMENT (Control Data Structure Segment Register) implementation is incomplete AFAICT. From EHCI specs :

    This register allows the host software to locate all control data structures within the *same 4 Gigabyte memory segment*.

Linux driver seems buggy too.

Overall, the risk of regression is too high given the time I can dedicate to this topic.

Quoted 7 lines
Quoted 3 lines
Quoted 3 lines

Thanks for all the good work you done there, specially for migration. If someone can Ack the series and step forward to become USB maintainer, then I will reconsider.

I'm not an USB expert / maintainer but FWIW this series looked good enough to me when I looked at it. Maybe consider to merge after the 11.1 release?

Yes. I would like to too but ...

The change in size of the EHCIqh and EHCIitd descriptors is a bit worrying since these are used for guest DMA, see ehci_flush_qh(), or compare, see ehci_verify_qh().

This should be fine for Linux, which has support since day one. Other older guest OSes may not allocate the extra space. This is a 20+ year old spec so the risk is probably low but I don't have the expertise to assess the impact tbh.

Anyhow if we want the model to be correct, we should not write beyond spec-defined descriptor boundaries when in 32-bit mode.

More work ...

Thanks,

C.

author
#29 lore raw

Hi Cédric, Philippe

Quoted 206 lines

Subject: Re: [PATCH v6 00/11] hw/usb/ehci: Add 64-bit descriptor addressing support

Quoted 58 lines · Philippe Mathieu-Daudé
Quoted 57 lines · Cédric Le Goater

+ Gerd, Marc-André,

May be they can provide some more feedback.

Quoted 52 lines · Cédric Le Goater
Quoted 51 lines · Cédric Le Goater
Quoted 50 lines · Jamin Lin

EHCI supports 64-bit addressing through the CTRLDSSEGMENT register, which provides the upper 32 bits of descriptor addresses when the controller advertises 64-bit capability.

Currently QEMU EHCI model only partially supports this functionality and descriptor addresses are effectively treated as 32-bit. This becomes problematic on systems where system memory is located above the 4GB boundary.

The Linux EHCI driver enables 64-bit addressing if the controller advertises the capability. During initialization it programs the segment register to zero:

   https://github.com/torvalds/linux/blob/master/drivers/usb/host/
ehci-hcd.c#L600

The driver also notes that descriptor structures allocated from the DMA pool use segment zero semantics. Descriptor memory is allocated using the DMA API and platforms may configure a 64-bit DMA mask, allowing descriptor memory to be placed above 4GB.

On AST2700 platforms, system DRAM is mapped at 0x400000000. As a result, descriptor addresses constructed directly from the EHCI registers do not match the actual system addresses used by the controller when accessing queue heads (QH) and queue element transfer descriptors (qTD).

This patch series implements full 64-bit descriptor addressing support in the EHCI emulation. Descriptor address handling is updated to use 64-bit values and the descriptor structures (QH, qTD, iTD and siTD) are extended to support the upper address bits provided by the segment register.

Add a ctrldssegment-default property so platforms can provide a descriptor address offset when constructing descriptor addresses. This allows systems where DRAM resides above 4GB to access EHCI descriptors correctly.

The AST2700 machine uses this property to account for its DRAM mapping at 0x400000000 and enables 64-bit EHCI DMA addressing.

Test Result:
1. EHCI 32bits with ast2600-evb machine Command line:
./build/qemu-system-arm \
   -machine ast2600-evb \
   -m 1G \
   -drive file=image-bmc,if=mtd,format=raw \
   -nographic \
   -device usb-kbd,bus=usb-bus.1,id=mykbd \
   -drive id=usbdisk,if=none,file=image0.ext4,format=raw \

Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines

   -device usb-storage,bus=usb-bus.1,id=mystorage,drive=usbdisk -snapshot \
   -nographic
Result:
unable to initialize usb specBus 001 Device 001: ID 1d6b:0002 Linux 6.18.3-v00.08.01-g172b7e27a30d ehci_hcd EHCI Host Controller Bus 001 Device 002: ID 0627:0001 QEMU QEMU USB Keyboard Bus 001 Device 003: ID 46f4:0001 QEMU QEMU USB HARDDRIVE Bus 002

Device

Quoted 18 lines
Quoted 18 lines
Quoted 18 lines
Quoted 18 lines
Quoted 12 lines

001: ID 1d6b:0001 Linux 6.18.3-v00.08.01- g172b7e27a30d uhci_hcd Generic UHCI Host Controller

2. EHCI 64bits with ast2700a2-evb machine Command line:
./build/qemu-system-aarch64 -M ast2700a2-evb -nographic\
  -bios ast27x0_bootrom.bin \
  -drive file=image-bmc,format=raw,if=mtd \
  -snapshot \
  -device usb-kbd,bus=usb-bus.3,id=mykbd \
  -drive id=usbdisk,if=none,file=image0.ext4,format=raw \
  -device usb-storage,bus=usb-bus.3,id=mystorage,drive=usbdisk Result:

Quoted 6 lines
root@ast2700-default:~# lsusb
unable to initialize usb specBus 001 Device 001: ID 1d6b:0001
Linux 6.18.3-v00.08.01-g172b7e27a30d uhci_hcd Generic UHCI Host
Controller Bus 002 Device 001: ID 1d6b:0002 Linux
6.18.3-v00.08.01- g172b7e27a30d ehci_hcd EHCI Host Controller Bus
002 Device 002: ID 0627:0001 QEMU QEMU USB Keyboard Bus 002

Device

Quoted 55 lines
Quoted 55 lines
Quoted 55 lines
Quoted 55 lines
Quoted 55 lines

003: ID 46f4:0001 QEMU QEMU USB HARDDRIVE
v1
  1. Fix checkpatch coding style issues
  2. Implement 64-bit addressing for QH/qTD/iTD/siTD descriptors
  3. Add descriptor address offset property
  4. Enable 64-bit EHCI DMA addressing on AST2700
  5. Configure descriptor address offset for AST2700

v2
  1. Remove unused EHCIfstn structure and dead code
  2. Replace fprintf(stderr, ...) with qemu_log_mask(LOG_GUEST_ERROR)
  3. Replace DPRINTF debug logs with trace events
  4. Add functional tests for USB EHCI on AST2600 and AST2700 A1/A2
  5. Fix review issue

v3:
  1. Add Migration version test function
  2. Add EHCI 64-bit buffer pointer fields description in commit log

v4:
  1. Reorder patches in the series
  2. Fix EHCI migration issues
  3. Introduce a common properties macro for both sysbus and PCI
  4. Drop the descriptor address offset property
  5. Add ctrldssegment-default property
  6. Address review comments

v5:
  1. Add 11.0 machine compatibility properties

v6:
  1. Update reviewer suggested improvements.
Jamin Lin (11):
   tests/functional/arm/test_aspeed_ast2600_sdk: Add USB EHCI test for
     AST2600 SDK
   hw/usb/hcd-ehci: Change descriptor addresses to 64-bit with migration
     compatibility
   hw/usb/hcd-ehci: Add property to advertise 64-bit addressing capability
   hw/usb/hcd-ehci: Implement 64-bit QH descriptor addressing hw/usb/hcd-ehci: Implement 64-bit qTD descriptor addressing hw/usb/hcd-ehci: Implement 64-bit iTD descriptor addressing hw/usb/hcd-ehci: Implement 64-bit siTD descriptor addressing hw/usb/hcd-ehci: Add ctrldssegment-default property hw/arm/aspeed_ast27x0: Set EHCI ctrldssegment-default hw/arm/aspeed_ast27x0: Enable 64-bit EHCI DMA addressing tests/functional/aarch64/test_aspeed_ast2700: Add USB EHCI test for
     AST2700 A1/A2

Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

  hw/usb/hcd-ehci.h                             |  2

9 +++-

Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

  hw/arm/aspeed_ast27x0.c                       |   5

+

Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

  hw/core/machine.c                             |

5 +-

Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

  hw/usb/hcd-ehci.c                             | 162

++++++++++++

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
+-----
  hw/usb/trace-events                           |  26
+--
>>>>>   .../aarch64/test_aspeed_ast2700a1.py          |   7 +
>>>>>   .../aarch64/test_aspeed_ast2700a2.py          |   7 +
>>>>>   .../functional/arm/test_aspeed_ast2600_sdk.py |   7 +
Quoted 29 lines
Quoted 29 lines
Quoted 29 lines
Quoted 6 lines
Quoted 2 lines
  8 files changed, 185 insertions(+), 63 deletions(-)

Applied to

     https://github.com/legoater/qemu aspeed-next

Jamin,

I dropped the series from aspeed-next and kept :

   tests/functional/arm/test_aspeed_ast2600_sdk: Add USB EHCI test for AST2600 SDK

The USB subsystem is orphan but it is still used in the virt world. This series is modifying the EHCI internals in such way that migration is impacted and the CTRLDSSEGMENT (Control Data Structure Segment
Register) implementation is incomplete AFAICT. From EHCI specs :

    This register allows the host software to locate all control data
    structures within the *same 4 Gigabyte memory segment*.

Linux driver seems buggy too.

Overall, the risk of regression is too high given the time I can dedicate to this topic.

Quoted 7 lines
Quoted 3 lines
Quoted 3 lines

Thanks for all the good work you done there, specially for migration. If someone can Ack the series and step forward to become USB maintainer, then I will reconsider.

I'm not an USB expert / maintainer but FWIW this series looked good enough to me when I looked at it. Maybe consider to merge after the 11.1 release?

Yes. I would like to too but ...

The change in size of the EHCIqh and EHCIitd descriptors is a bit worrying since these are used for guest DMA, see ehci_flush_qh(), or compare, see ehci_verify_qh().

This should be fine for Linux, which has support since day one. Other older guest OSes may not allocate the extra space. This is a 20+ year old spec so the risk is probably low but I don't have the expertise to assess the impact tbh.

Quoted 8 lines

Anyhow if we want the model to be correct, we should not write beyond spec-defined descriptor boundaries when in 32-bit mode.

More work ...

Thanks,

C.

Thanks for pointing this out.

I agree that even though the QEMU internal EHCIqh, EHCIqtd, and EHCIitd structures include the extended high buffer pointer fields, we must not access those fields in guest memory when 64-bit addressing capability is not advertised.

I have updated the patch accordingly.

The descriptor DMA size is now selected based on caps_64bit_addr:

When 64-bit addressing is supported and advertised, the complete descriptor, including the high buffer pointer fields, is transferred. In 32-bit mode, DMA reads and writes stop at the beginning of bufptr_hi, so QEMU does not access memory beyond the descriptor layout defined for 32-bit addressing.

I also clear the local descriptor structures before partial DMA reads. Therefore, the high buffer pointer fields remain zero in 32-bit mode rather than containing stale or uninitialized values.

This has been applied to:

QH fetching, verification, and writeback
qTD fetching and verification
iTD fetching and writeback

For example:
/*
* EHCIqh / EHCIqtd / EHCIitd are sized to always include the extended
* high buffer pointer fields from EHCI 1.0 Appendix B. When 64-bit
* addressing capability is not advertised to the guest, the descriptors
* in guest memory only have the classic 32-bit layout, so DMA transfers
* must not read or write past that boundary.
*/

#define EHCI_QH_DWORDS_32   (offsetof(EHCIqh,  bufptr_hi) / sizeof(uint32_t))
#define EHCI_QTD_DWORDS_32  (offsetof(EHCIqtd, bufptr_hi) / sizeof(uint32_t))
#define EHCI_ITD_DWORDS_32  (offsetof(EHCIitd, bufptr_hi) / sizeof(uint32_t))

#define EHCI_QH_DWORDS_32 \
(offsetof(EHCIqh, bufptr_hi) / sizeof(uint32_t))

static uint32_t ehci_qh_dwords(const EHCIState *s)

{
    return s->caps_64bit_addr ?
           (sizeof(EHCIqh) >> 2) : EHCI_QH_DWORDS_32;
}

ehci_flush_qh() now uses this selected size, so in 32-bit mode it will not write the extended high buffer pointer fields back to guest memory.

Similarly, ehci_state_fetchqtd() only reads qtd.bufptr_hi when caps_64bit_addr is enabled:

memset(qtd.bufptr_hi, 0, sizeof(qtd.bufptr_hi));

if (... ||

    (q->ehci->caps_64bit_addr &&
         get_dwords(q->ehci, addr + offsetof(EHCIqtd, bufptr_hi),
                    qtd.bufptr_hi, ARRAY_SIZE(qtd.bufptr_hi)) < 0)) {    return 0;
}

This should preserve compatibility with guests that allocate only the original 32-bit EHCI descriptor size, while still allowing the model to support the extended 64-bit descriptor layout. If you think this approach looks OK, I'll include it in the next revision (v7).

Thanks,
Jamin

author
#30 lore raw

+ Philippe Mathieu-Daudé <[email protected]>

Quoted 298 lines

Subject: RE: [PATCH v6 00/11] hw/usb/ehci: Add 64-bit descriptor addressing support

Hi Cédric, Philippe

Quoted 35 lines

Subject: Re: [PATCH v6 00/11] hw/usb/ehci: Add 64-bit descriptor addressing support

Quoted 31 lines · Philippe Mathieu-Daudé
Quoted 30 lines · Cédric Le Goater

+ Gerd, Marc-André,

May be they can provide some more feedback.

Quoted 25 lines · Cédric Le Goater
Quoted 24 lines · Cédric Le Goater
Quoted 23 lines · Jamin Lin

EHCI supports 64-bit addressing through the CTRLDSSEGMENT register, which provides the upper 32 bits of descriptor addresses when the controller advertises 64-bit capability.

Currently QEMU EHCI model only partially supports this functionality and descriptor addresses are effectively treated as 32-bit. This becomes problematic on systems where system memory is located above the 4GB boundary.

The Linux EHCI driver enables 64-bit addressing if the controller advertises the capability. During initialization it programs the segment register to zero:

https://github.com/torvalds/linux/blob/master/drivers/usb/host/
ehci-hcd.c#L600

The driver also notes that descriptor structures allocated from the DMA pool use segment zero semantics. Descriptor memory is allocated using the DMA API and platforms may configure a 64-bit DMA mask, allowing descriptor memory to be placed above 4GB.

On AST2700 platforms, system DRAM is mapped at 0x400000000. As

a

Quoted 114 lines
Quoted 29 lines
Quoted 29 lines
Quoted 29 lines
Quoted 29 lines
Quoted 29 lines

result, descriptor addresses constructed directly from the EHCI registers do not match the actual system addresses used by the controller when accessing queue heads (QH) and queue element transfer descriptors (qTD).

This patch series implements full 64-bit descriptor addressing support in the EHCI emulation. Descriptor address handling is updated to use 64-bit values and the descriptor structures (QH, qTD, iTD and siTD) are extended to support the upper address bits provided by the segment register.

Add a ctrldssegment-default property so platforms can provide a descriptor address offset when constructing descriptor addresses. This allows systems where DRAM resides above 4GB to access EHCI descriptors correctly.

The AST2700 machine uses this property to account for its DRAM mapping at 0x400000000 and enables 64-bit EHCI DMA addressing.

Test Result:
1. EHCI 32bits with ast2600-evb machine Command line:
./build/qemu-system-arm \
   -machine ast2600-evb \
   -m 1G \
   -drive file=image-bmc,if=mtd,format=raw \
   -nographic \
   -device usb-kbd,bus=usb-bus.1,id=mykbd \
   -drive id=usbdisk,if=none,file=image0.ext4,format=raw \
   -device usb-storage,bus=usb-bus.1,id=mystorage,drive=usbdisk

Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines

   -snapshot \
   -nographic
Result:
unable to initialize usb specBus 001 Device 001: ID 1d6b:0002 Linux 6.18.3-v00.08.01-g172b7e27a30d ehci_hcd EHCI Host Controller Bus 001 Device 002: ID 0627:0001 QEMU QEMU USB Keyboard Bus 001 Device 003: ID 46f4:0001 QEMU QEMU USB HARDDRIVE Bus 002

Device

Quoted 19 lines
Quoted 19 lines
Quoted 19 lines
Quoted 19 lines
Quoted 12 lines

001: ID 1d6b:0001 Linux 6.18.3-v00.08.01- g172b7e27a30d uhci_hcd Generic UHCI Host Controller

2. EHCI 64bits with ast2700a2-evb machine Command line:
./build/qemu-system-aarch64 -M ast2700a2-evb -nographic\
  -bios ast27x0_bootrom.bin \
  -drive file=image-bmc,format=raw,if=mtd \
  -snapshot \
  -device usb-kbd,bus=usb-bus.3,id=mykbd \
  -drive id=usbdisk,if=none,file=image0.ext4,format=raw \
  -device usb-storage,bus=usb-bus.3,id=mystorage,drive=usbdisk Result:

Quoted 7 lines
root@ast2700-default:~# lsusb
unable to initialize usb specBus 001 Device 001: ID 1d6b:0001
Linux 6.18.3-v00.08.01-g172b7e27a30d uhci_hcd Generic UHCI Host
Controller Bus 002 Device 001: ID 1d6b:0002 Linux
6.18.3-v00.08.01- g172b7e27a30d ehci_hcd EHCI Host Controller
Bus
002 Device 002: ID 0627:0001 QEMU QEMU USB Keyboard Bus 002

Device

Quoted 55 lines
Quoted 55 lines
Quoted 55 lines
Quoted 55 lines
Quoted 55 lines

003: ID 46f4:0001 QEMU QEMU USB HARDDRIVE
v1
  1. Fix checkpatch coding style issues
  2. Implement 64-bit addressing for QH/qTD/iTD/siTD descriptors
  3. Add descriptor address offset property
  4. Enable 64-bit EHCI DMA addressing on AST2700
  5. Configure descriptor address offset for AST2700

v2
  1. Remove unused EHCIfstn structure and dead code
  2. Replace fprintf(stderr, ...) with qemu_log_mask(LOG_GUEST_ERROR)
  3. Replace DPRINTF debug logs with trace events
  4. Add functional tests for USB EHCI on AST2600 and AST2700 A1/A2
  5. Fix review issue

v3:
  1. Add Migration version test function
  2. Add EHCI 64-bit buffer pointer fields description in commit log

v4:
  1. Reorder patches in the series
  2. Fix EHCI migration issues
  3. Introduce a common properties macro for both sysbus and PCI
  4. Drop the descriptor address offset property
  5. Add ctrldssegment-default property
  6. Address review comments

v5:
  1. Add 11.0 machine compatibility properties

v6:
  1. Update reviewer suggested improvements.
Jamin Lin (11):
   tests/functional/arm/test_aspeed_ast2600_sdk: Add USB EHCI test for
     AST2600 SDK
   hw/usb/hcd-ehci: Change descriptor addresses to 64-bit with migration
     compatibility
   hw/usb/hcd-ehci: Add property to advertise 64-bit addressing capability
   hw/usb/hcd-ehci: Implement 64-bit QH descriptor addressing hw/usb/hcd-ehci: Implement 64-bit qTD descriptor addressing hw/usb/hcd-ehci: Implement 64-bit iTD descriptor addressing hw/usb/hcd-ehci: Implement 64-bit siTD descriptor addressing hw/usb/hcd-ehci: Add ctrldssegment-default property hw/arm/aspeed_ast27x0: Set EHCI ctrldssegment-default hw/arm/aspeed_ast27x0: Enable 64-bit EHCI DMA addressing tests/functional/aarch64/test_aspeed_ast2700: Add USB EHCI test for
     AST2700 A1/A2

Quoted 8 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

  hw/usb/hcd-ehci.h                             |  2

9 +++-

Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

  hw/arm/aspeed_ast27x0.c                       |   5

+

Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

  hw/core/machine.c                             |

5 +-

Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

  hw/usb/hcd-ehci.c                             |

162

Quoted 47 lines

++++++++++++

Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 1 line

+-----

  hw/usb/trace-events                           |  26

+--
>>>>>   .../aarch64/test_aspeed_ast2700a1.py          |   7 +
>>>>>   .../aarch64/test_aspeed_ast2700a2.py          |   7 +
>>>>>   .../functional/arm/test_aspeed_ast2600_sdk.py |   7 +
Quoted 29 lines
Quoted 29 lines
Quoted 29 lines
Quoted 6 lines
Quoted 2 lines
  8 files changed, 185 insertions(+), 63 deletions(-)

Applied to

     https://github.com/legoater/qemu aspeed-next

Jamin,

I dropped the series from aspeed-next and kept :

   tests/functional/arm/test_aspeed_ast2600_sdk: Add USB EHCI test for AST2600 SDK

The USB subsystem is orphan but it is still used in the virt world. This series is modifying the EHCI internals in such way that migration is impacted and the CTRLDSSEGMENT (Control Data Structure Segment
Register) implementation is incomplete AFAICT. From EHCI specs :

    This register allows the host software to locate all control data
    structures within the *same 4 Gigabyte memory segment*.

Linux driver seems buggy too.

Overall, the risk of regression is too high given the time I can dedicate to this topic.

Quoted 7 lines
Quoted 3 lines
Quoted 3 lines

Thanks for all the good work you done there, specially for migration. If someone can Ack the series and step forward to become USB maintainer, then I will reconsider.

I'm not an USB expert / maintainer but FWIW this series looked good enough to me when I looked at it. Maybe consider to merge after the 11.1 release?

Yes. I would like to too but ...

Quoted 7 lines

The change in size of the EHCIqh and EHCIitd descriptors is a bit worrying since these are used for guest DMA, see ehci_flush_qh(), or compare, see ehci_verify_qh().

This should be fine for Linux, which has support since day one. Other older guest OSes may not allocate the extra space. This is a 20+ year old spec so the risk is probably low but I don't have the expertise to assess

the impact tbh.

Quoted 9 lines

Anyhow if we want the model to be correct, we should not write beyond spec-defined descriptor boundaries when in 32-bit mode.

More work ...

Thanks,

C.

Thanks for pointing this out.

I agree that even though the QEMU internal EHCIqh, EHCIqtd, and EHCIitd structures include the extended high buffer pointer fields, we must not access those fields in guest memory when 64-bit addressing capability is not advertised.

I have updated the patch accordingly.

The descriptor DMA size is now selected based on caps_64bit_addr:

When 64-bit addressing is supported and advertised, the complete descriptor, including the high buffer pointer fields, is transferred.
In 32-bit mode, DMA reads and writes stop at the beginning of bufptr_hi, so QEMU does not access memory beyond the descriptor layout defined for 32-bit addressing.

I also clear the local descriptor structures before partial DMA reads. Therefore, the high buffer pointer fields remain zero in 32-bit mode rather than containing stale or uninitialized values.

This has been applied to:

QH fetching, verification, and writeback qTD fetching and verification iTD fetching and writeback

For example:
/*
* EHCIqh / EHCIqtd / EHCIitd are sized to always include the extended
* high buffer pointer fields from EHCI 1.0 Appendix B. When 64-bit
* addressing capability is not advertised to the guest, the descriptors
* in guest memory only have the classic 32-bit layout, so DMA transfers
* must not read or write past that boundary.
*/
#define EHCI_QH_DWORDS_32 (offsetof(EHCIqh, bufptr_hi) /
sizeof(uint32_t))
#define EHCI_QTD_DWORDS_32 (offsetof(EHCIqtd, bufptr_hi) / sizeof(uint32_t)) #define EHCI_ITD_DWORDS_32 (offsetof(EHCIitd, bufptr_hi) / sizeof(uint32_t))

#define EHCI_QH_DWORDS_32 \
(offsetof(EHCIqh, bufptr_hi) / sizeof(uint32_t))

static uint32_t ehci_qh_dwords(const EHCIState *s) {
    return s->caps_64bit_addr ?
           (sizeof(EHCIqh) >> 2) : EHCI_QH_DWORDS_32; }

ehci_flush_qh() now uses this selected size, so in 32-bit mode it will not write the extended high buffer pointer fields back to guest memory.

Similarly, ehci_state_fetchqtd() only reads qtd.bufptr_hi when caps_64bit_addr is enabled:

memset(qtd.bufptr_hi, 0, sizeof(qtd.bufptr_hi));

if (... ||

    (q->ehci->caps_64bit_addr &&
         get_dwords(q->ehci, addr + offsetof(EHCIqtd, bufptr_hi),
                    qtd.bufptr_hi, ARRAY_SIZE(qtd.bufptr_hi)) < 0))

{ return 0;
}

This should preserve compatibility with guests that allocate only the original 32-bit EHCI descriptor size, while still allowing the model to support the extended 64-bit descriptor layout.
If you think this approach looks OK, I'll include it in the next revision (v7).

Thanks,
Jamin

author
#31 lore raw

Hi Cédric, Philippe

Quoted 40 lines

Subject: RE: [PATCH v6 00/11] hw/usb/ehci: Add 64-bit descriptor addressing support

+ Philippe Mathieu-Daudé <[email protected]>

Quoted 35 lines

Subject: RE: [PATCH v6 00/11] hw/usb/ehci: Add 64-bit descriptor addressing support

Hi Cédric, Philippe

Quoted 30 lines

Subject: Re: [PATCH v6 00/11] hw/usb/ehci: Add 64-bit descriptor addressing support

Quoted 26 lines · Philippe Mathieu-Daudé
Quoted 25 lines · Cédric Le Goater

+ Gerd, Marc-André,

May be they can provide some more feedback.

Quoted 20 lines · Cédric Le Goater
Quoted 19 lines · Cédric Le Goater
Quoted 18 lines · Jamin Lin

EHCI supports 64-bit addressing through the CTRLDSSEGMENT register, which provides the upper 32 bits of descriptor addresses when the controller advertises 64-bit capability.

Currently QEMU EHCI model only partially supports this functionality and descriptor addresses are effectively treated as 32-bit. This becomes problematic on systems where system memory is located above the 4GB boundary.

The Linux EHCI driver enables 64-bit addressing if the controller advertises the capability. During initialization it programs the segment register to zero:

https://github.com/torvalds/linux/blob/master/drivers/usb/host
/
ehci-hcd.c#L600

Quoted 278 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines

The driver also notes that descriptor structures allocated from the DMA pool use segment zero semantics. Descriptor memory is allocated using the DMA API and platforms may configure a 64-bit DMA mask, allowing descriptor memory to be

placed above 4GB.

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

On AST2700 platforms, system DRAM is mapped at 0x400000000.

As

Quoted 270 lines

a

Quoted 120 lines
Quoted 30 lines
Quoted 30 lines
Quoted 30 lines
Quoted 30 lines
Quoted 30 lines

result, descriptor addresses constructed directly from the EHCI registers do not match the actual system addresses used by the controller when accessing queue heads (QH) and queue element transfer descriptors (qTD).

This patch series implements full 64-bit descriptor addressing support in the EHCI emulation. Descriptor address handling is updated to use 64-bit values and the descriptor structures (QH, qTD, iTD and siTD) are extended to support the upper address bits provided by the segment register.

Add a ctrldssegment-default property so platforms can provide a descriptor address offset when constructing descriptor addresses. This allows systems where DRAM resides above 4GB to access EHCI descriptors correctly.

The AST2700 machine uses this property to account for its DRAM mapping at 0x400000000 and enables 64-bit EHCI DMA addressing.

Test Result:
1. EHCI 32bits with ast2600-evb machine Command line:
./build/qemu-system-arm \
   -machine ast2600-evb \
   -m 1G \
   -drive file=image-bmc,if=mtd,format=raw \
   -nographic \
   -device usb-kbd,bus=usb-bus.1,id=mykbd \
   -drive id=usbdisk,if=none,file=image0.ext4,format=raw \
   -device usb-storage,bus=usb-bus.1,id=mystorage,drive=usbdisk

Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines

   -snapshot \
   -nographic
Result:
unable to initialize usb specBus 001 Device 001: ID 1d6b:0002 Linux 6.18.3-v00.08.01-g172b7e27a30d ehci_hcd EHCI Host Controller Bus 001 Device 002: ID 0627:0001 QEMU QEMU USB Keyboard Bus 001 Device 003: ID 46f4:0001 QEMU QEMU USB HARDDRIVE Bus 002

Device

Quoted 19 lines
Quoted 19 lines
Quoted 19 lines
Quoted 19 lines
Quoted 12 lines

001: ID 1d6b:0001 Linux 6.18.3-v00.08.01- g172b7e27a30d uhci_hcd Generic UHCI Host Controller

2. EHCI 64bits with ast2700a2-evb machine Command line:
./build/qemu-system-aarch64 -M ast2700a2-evb -nographic\
  -bios ast27x0_bootrom.bin \
  -drive file=image-bmc,format=raw,if=mtd \
  -snapshot \
  -device usb-kbd,bus=usb-bus.3,id=mykbd \
  -drive id=usbdisk,if=none,file=image0.ext4,format=raw \
  -device usb-storage,bus=usb-bus.3,id=mystorage,drive=usbdisk Result:

Quoted 7 lines
root@ast2700-default:~# lsusb
unable to initialize usb specBus 001 Device 001: ID 1d6b:0001
Linux 6.18.3-v00.08.01-g172b7e27a30d uhci_hcd Generic UHCI
Host Controller Bus 002 Device 001: ID 1d6b:0002 Linux
6.18.3-v00.08.01- g172b7e27a30d ehci_hcd EHCI Host Controller
Bus
002 Device 002: ID 0627:0001 QEMU QEMU USB Keyboard Bus 002

Device

Quoted 59 lines
Quoted 59 lines
Quoted 59 lines
Quoted 59 lines
Quoted 59 lines

003: ID 46f4:0001 QEMU QEMU USB HARDDRIVE
v1
  1. Fix checkpatch coding style issues
  2. Implement 64-bit addressing for QH/qTD/iTD/siTD descriptors
  3. Add descriptor address offset property
  4. Enable 64-bit EHCI DMA addressing on AST2700
  5. Configure descriptor address offset for AST2700

v2
  1. Remove unused EHCIfstn structure and dead code
  2. Replace fprintf(stderr, ...) with qemu_log_mask(LOG_GUEST_ERROR)
  3. Replace DPRINTF debug logs with trace events
  4. Add functional tests for USB EHCI on AST2600 and AST2700 A1/A2
  5. Fix review issue

v3:
  1. Add Migration version test function
  2. Add EHCI 64-bit buffer pointer fields description in commit log

v4:
  1. Reorder patches in the series
  2. Fix EHCI migration issues
  3. Introduce a common properties macro for both sysbus and PCI
  4. Drop the descriptor address offset property
  5. Add ctrldssegment-default property
  6. Address review comments

v5:
  1. Add 11.0 machine compatibility properties

v6:
  1. Update reviewer suggested improvements.
Jamin Lin (11):
   tests/functional/arm/test_aspeed_ast2600_sdk: Add USB EHCI test for
     AST2600 SDK
   hw/usb/hcd-ehci: Change descriptor addresses to 64-bit with migration
     compatibility
   hw/usb/hcd-ehci: Add property to advertise 64-bit addressing
     capability
   hw/usb/hcd-ehci: Implement 64-bit QH descriptor addressing hw/usb/hcd-ehci: Implement 64-bit qTD descriptor addressing hw/usb/hcd-ehci: Implement 64-bit iTD descriptor addressing hw/usb/hcd-ehci: Implement 64-bit siTD descriptor addressing
   hw/usb/hcd-ehci: Add ctrldssegment-default property hw/arm/aspeed_ast27x0: Set EHCI ctrldssegment-default hw/arm/aspeed_ast27x0: Enable 64-bit EHCI DMA addressing tests/functional/aarch64/test_aspeed_ast2700: Add USB EHCI test for
     AST2700 A1/A2

Quoted 9 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line

  hw/usb/hcd-ehci.h                             |  2

9 +++-

Quoted 1 line
Quoted 1 line
Quoted 1 line

  hw/arm/aspeed_ast27x0.c                       |   5

+

Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

  hw/core/machine.c                             |

5 +-

Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

  hw/usb/hcd-ehci.c                             |

162

Quoted 49 lines

++++++++++++

Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 1 line
Quoted 1 line

+-----

  hw/usb/trace-events                           |  26

+--
>>>>>   .../aarch64/test_aspeed_ast2700a1.py          |   7 +
>>>>>   .../aarch64/test_aspeed_ast2700a2.py          |   7 +
>>>>>   .../functional/arm/test_aspeed_ast2600_sdk.py |   7 +
Quoted 29 lines
Quoted 29 lines
Quoted 29 lines
Quoted 6 lines
Quoted 2 lines
  8 files changed, 185 insertions(+), 63 deletions(-)

Applied to

     https://github.com/legoater/qemu aspeed-next

Jamin,

I dropped the series from aspeed-next and kept :

   tests/functional/arm/test_aspeed_ast2600_sdk: Add USB EHCI test for AST2600 SDK

The USB subsystem is orphan but it is still used in the virt world. This series is modifying the EHCI internals in such way that migration is impacted and the CTRLDSSEGMENT (Control Data Structure Segment
Register) implementation is incomplete AFAICT. From EHCI specs :

    This register allows the host software to locate all control data
    structures within the *same 4 Gigabyte memory segment*.

Linux driver seems buggy too.

Overall, the risk of regression is too high given the time I can dedicate to this topic.

Quoted 8 lines
Quoted 3 lines
Quoted 3 lines

Thanks for all the good work you done there, specially for migration. If someone can Ack the series and step forward to become USB maintainer, then I will reconsider.

I'm not an USB expert / maintainer but FWIW this series looked good enough to me when I looked at it. Maybe consider to merge after the
11.1 release?

Yes. I would like to too but ...

Quoted 8 lines

The change in size of the EHCIqh and EHCIitd descriptors is a bit worrying since these are used for guest DMA, see ehci_flush_qh(), or compare, see ehci_verify_qh().

This should be fine for Linux, which has support since day one. Other older guest OSes may not allocate the extra space. This is a 20+ year old spec so the risk is probably low but I don't have the expertise to assess

the impact tbh.

Quoted 9 lines

Anyhow if we want the model to be correct, we should not write beyond spec-defined descriptor boundaries when in 32-bit mode.

More work ...

Thanks,

C.

Thanks for pointing this out.

I agree that even though the QEMU internal EHCIqh, EHCIqtd, and EHCIitd structures include the extended high buffer pointer fields, we must not access those fields in guest memory when 64-bit addressing capability is not advertised.

I have updated the patch accordingly.

The descriptor DMA size is now selected based on caps_64bit_addr:

When 64-bit addressing is supported and advertised, the complete descriptor, including the high buffer pointer fields, is transferred. In 32-bit mode, DMA reads and writes stop at the beginning of bufptr_hi, so QEMU does not access memory beyond the descriptor layout defined for 32-bit addressing.

I also clear the local descriptor structures before partial DMA reads. Therefore, the high buffer pointer fields remain zero in 32-bit mode rather than containing stale or uninitialized values.

This has been applied to:

QH fetching, verification, and writeback qTD fetching and verification iTD fetching and writeback

For example:
/*
* EHCIqh / EHCIqtd / EHCIitd are sized to always include the extended
* high buffer pointer fields from EHCI 1.0 Appendix B. When 64-bit
* addressing capability is not advertised to the guest, the
descriptors
* in guest memory only have the classic 32-bit layout, so DMA
transfers
* must not read or write past that boundary.
*/
#define EHCI_QH_DWORDS_32 (offsetof(EHCIqh, bufptr_hi) /
sizeof(uint32_t))
#define EHCI_QTD_DWORDS_32 (offsetof(EHCIqtd, bufptr_hi) / sizeof(uint32_t)) #define EHCI_ITD_DWORDS_32 (offsetof(EHCIitd, bufptr_hi) / sizeof(uint32_t))

#define EHCI_QH_DWORDS_32 \
(offsetof(EHCIqh, bufptr_hi) / sizeof(uint32_t))

static uint32_t ehci_qh_dwords(const EHCIState *s) {
    return s->caps_64bit_addr ?
           (sizeof(EHCIqh) >> 2) : EHCI_QH_DWORDS_32; }

ehci_flush_qh() now uses this selected size, so in 32-bit mode it will not write the extended high buffer pointer fields back to guest memory.

Similarly, ehci_state_fetchqtd() only reads qtd.bufptr_hi when caps_64bit_addr is enabled:

memset(qtd.bufptr_hi, 0, sizeof(qtd.bufptr_hi));

if (... ||

    (q->ehci->caps_64bit_addr &&
         get_dwords(q->ehci, addr + offsetof(EHCIqtd, bufptr_hi),
                    qtd.bufptr_hi, ARRAY_SIZE(qtd.bufptr_hi)) < 0))

{ return 0;
}

This should preserve compatibility with guests that allocate only the original 32-bit EHCI descriptor size, while still allowing the model to support the extended 64-bit descriptor layout.
If you think this approach looks OK, I'll include it in the next revision (v7).

Thanks,
Jamin

I sent v7 here, https://patchwork.kernel.org/project/qemu-devel/cover/[email protected]/

Thanks,
Jamin