[PATCH 00/10] q800: migration fixes

[PATCH 00/10] q800: migration fixes

[Mark Cave-Ayland]

This patchset contains fixes for the macfb and esp devices which enable
migration of the q800 machine to succeed here in local testing.

Patches 1-5 contain fixes and improvements for migrating the macfb device
whilst patches 6-9 change the ESPState pdma_cb field from being a
function pointer to an integer index that can be included in the migration
stream.

Finally patch 10 ensures that any in-flight SCSI requests active during
migration are resumed correctly post-migration. This is required because
PDMA requires the guest to read/write DMA data and hence an active
SCSI request cannot run to completion before migration starts.

NOTE: this patchset is based upon my previous mos6522 patchset posted at
https://lists.gnu.org/archive/html/qemu-devel/2022-02/msg05248.html.

Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Based-on: <20220224115956.29997-1-mark.cave-ayland@ilande.co.uk>


Mark Cave-Ayland (10):
  macfb: add VMStateDescription for MacfbNubusState and MacfbSysBusState
  macfb: don't use special irq_state and irq_mask variables in
    MacfbState
  macfb: increase number of registers saved in MacfbState
  macfb: add VMStateDescription fields for display type and VBL timer
  macfb: set initial value of mode control registers in
    macfb_common_realize()
  esp: introduce esp_set_pdma_cb() function
  esp: introduce esp_pdma_cb() function
  esp: convert ESPState pdma_cb from a function pointer to an integer
  esp: include the current PDMA callback in the migration stream
  esp: recreate ESPState current_req after migration

 hw/display/macfb.c         | 49 ++++++++++++++++++-----
 hw/scsi/esp.c              | 80 +++++++++++++++++++++++++++++++++-----
 include/hw/display/macfb.h |  4 +-
 include/hw/scsi/esp.h      | 11 +++++-
 4 files changed, 121 insertions(+), 23 deletions(-)

-- 
2.20.1

[PATCH 01/10] macfb: add VMStateDescription for MacfbNubusState and MacfbSysBusState

[Mark Cave-Ayland]

Currently when QEMU tries to migrate the macfb framebuffer it crashes randomly
because the opaque provided by the DeviceClass vmsd property for both devices
is set to MacfbState rather than MacfbNubusState or MacfbSysBusState as
appropriate.

Resolve the issue by adding new VMStateDescriptions for MacfbNubusState and
MacfbSysBusState which embed the existing vmstate_macfb VMStateDescription
within them using VMSTATE_STRUCT.

Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
---
 hw/display/macfb.c | 24 ++++++++++++++++++++++--
 1 file changed, 22 insertions(+), 2 deletions(-)

diff --git a/hw/display/macfb.c b/hw/display/macfb.c
index c9b468c10e..66ceacf1ae 100644
--- a/hw/display/macfb.c
+++ b/hw/display/macfb.c
@@ -746,6 +746,16 @@ static Property macfb_sysbus_properties[] = {
     DEFINE_PROP_END_OF_LIST(),
 };
 
+static const VMStateDescription vmstate_macfb_sysbus = {
+    .name = "macfb-sysbus",
+    .version_id = 1,
+    .minimum_version_id = 1,
+    .fields = (VMStateField[]) {
+        VMSTATE_STRUCT(macfb, MacfbSysBusState, 1, vmstate_macfb, MacfbState),
+        VMSTATE_END_OF_LIST()
+    }
+};
+
 static Property macfb_nubus_properties[] = {
     DEFINE_PROP_UINT32("width", MacfbNubusState, macfb.width, 640),
     DEFINE_PROP_UINT32("height", MacfbNubusState, macfb.height, 480),
@@ -755,6 +765,16 @@ static Property macfb_nubus_properties[] = {
     DEFINE_PROP_END_OF_LIST(),
 };
 
+static const VMStateDescription vmstate_macfb_nubus = {
+    .name = "macfb-nubus",
+    .version_id = 1,
+    .minimum_version_id = 1,
+    .fields = (VMStateField[]) {
+        VMSTATE_STRUCT(macfb, MacfbNubusState, 1, vmstate_macfb, MacfbState),
+        VMSTATE_END_OF_LIST()
+    }
+};
+
 static void macfb_sysbus_class_init(ObjectClass *klass, void *data)
 {
     DeviceClass *dc = DEVICE_CLASS(klass);
@@ -762,7 +782,7 @@ static void macfb_sysbus_class_init(ObjectClass *klass, void *data)
     dc->realize = macfb_sysbus_realize;
     dc->desc = "SysBus Macintosh framebuffer";
     dc->reset = macfb_sysbus_reset;
-    dc->vmsd = &vmstate_macfb;
+    dc->vmsd = &vmstate_macfb_sysbus;
     device_class_set_props(dc, macfb_sysbus_properties);
 }
 
@@ -777,7 +797,7 @@ static void macfb_nubus_class_init(ObjectClass *klass, void *data)
                                       &ndc->parent_unrealize);
     dc->desc = "Nubus Macintosh framebuffer";
     dc->reset = macfb_nubus_reset;
-    dc->vmsd = &vmstate_macfb;
+    dc->vmsd = &vmstate_macfb_nubus;
     set_bit(DEVICE_CATEGORY_DISPLAY, dc->categories);
     device_class_set_props(dc, macfb_nubus_properties);
 }
-- 
2.20.1

[PATCH 02/10] macfb: don't use special irq_state and irq_mask variables in MacfbState

[Mark Cave-Ayland]

The current IRQ state and IRQ mask are handled exactly the same as standard
register accesses, so store these values directly in the regs array rather
than having separate variables for them.

Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
---
 hw/display/macfb.c         | 15 +++++++--------
 include/hw/display/macfb.h |  2 --
 2 files changed, 7 insertions(+), 10 deletions(-)

diff --git a/hw/display/macfb.c b/hw/display/macfb.c
index 66ceacf1ae..fb54b460c1 100644
--- a/hw/display/macfb.c
+++ b/hw/display/macfb.c
@@ -476,7 +476,8 @@ static void macfb_update_display(void *opaque)
 
 static void macfb_update_irq(MacfbState *s)
 {
-    uint32_t irq_state = s->irq_state & s->irq_mask;
+    uint32_t irq_state = s->regs[DAFB_INTR_STAT >> 2] &
+                         s->regs[DAFB_INTR_MASK >> 2];
 
     if (irq_state) {
         qemu_irq_raise(s->irq);
@@ -496,7 +497,7 @@ static void macfb_vbl_timer(void *opaque)
     MacfbState *s = opaque;
     int64_t next_vbl;
 
-    s->irq_state |= DAFB_INTR_VBL;
+    s->regs[DAFB_INTR_STAT >> 2] |= DAFB_INTR_VBL;
     macfb_update_irq(s);
 
     /* 60 Hz irq */
@@ -530,10 +531,8 @@ static uint64_t macfb_ctrl_read(void *opaque,
     case DAFB_MODE_VADDR2:
     case DAFB_MODE_CTRL1:
     case DAFB_MODE_CTRL2:
-        val = s->regs[addr >> 2];
-        break;
     case DAFB_INTR_STAT:
-        val = s->irq_state;
+        val = s->regs[addr >> 2];
         break;
     case DAFB_MODE_SENSE:
         val = macfb_sense_read(s);
@@ -568,7 +567,7 @@ static void macfb_ctrl_write(void *opaque,
         macfb_sense_write(s, val);
         break;
     case DAFB_INTR_MASK:
-        s->irq_mask = val;
+        s->regs[addr >> 2] = val;
         if (val & DAFB_INTR_VBL) {
             next_vbl = macfb_next_vbl();
             timer_mod(s->vbl_timer, next_vbl);
@@ -577,12 +576,12 @@ static void macfb_ctrl_write(void *opaque,
         }
         break;
     case DAFB_INTR_CLEAR:
-        s->irq_state &= ~DAFB_INTR_VBL;
+        s->regs[DAFB_INTR_STAT >> 2] &= ~DAFB_INTR_VBL;
         macfb_update_irq(s);
         break;
     case DAFB_RESET:
         s->palette_current = 0;
-        s->irq_state &= ~DAFB_INTR_VBL;
+        s->regs[DAFB_INTR_STAT >> 2] &= ~DAFB_INTR_VBL;
         macfb_update_irq(s);
         break;
     case DAFB_LUT:
diff --git a/include/hw/display/macfb.h b/include/hw/display/macfb.h
index e52775aa21..6d9f0f7869 100644
--- a/include/hw/display/macfb.h
+++ b/include/hw/display/macfb.h
@@ -66,8 +66,6 @@ typedef struct MacfbState {
     uint32_t regs[MACFB_NUM_REGS];
     MacFbMode *mode;
 
-    uint32_t irq_state;
-    uint32_t irq_mask;
     QEMUTimer *vbl_timer;
     qemu_irq irq;
 } MacfbState;
-- 
2.20.1

Re: [PATCH 02/10] macfb: don't use special irq_state and irq_mask variables in MacfbState

[Philippe Mathieu-Daudé]

On 28/2/22 23:25, Mark Cave-Ayland wrote:
> The current IRQ state and IRQ mask are handled exactly the same as standard
> register accesses, so store these values directly in the regs array rather
> than having separate variables for them.
> 
> Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
> ---
>   hw/display/macfb.c         | 15 +++++++--------
>   include/hw/display/macfb.h |  2 --
>   2 files changed, 7 insertions(+), 10 deletions(-)

Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>

[PATCH 03/10] macfb: increase number of registers saved in MacfbState

[Mark Cave-Ayland]

The MacOS toolbox ROM accesses a number of addresses between 0x0 and 0x200 during
initialisation and resolution changes. Whilst the function of many of these
registers is unknown, it is worth the minimal cost of saving these extra values as
part of migration to help future-proof the migration stream for the q800 machine
as it starts to stabilise.

Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
---
 include/hw/display/macfb.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/include/hw/display/macfb.h b/include/hw/display/macfb.h
index 6d9f0f7869..c0e2df8dc6 100644
--- a/include/hw/display/macfb.h
+++ b/include/hw/display/macfb.h
@@ -48,7 +48,7 @@ typedef struct MacFbMode {
     uint32_t offset;
 } MacFbMode;
 
-#define MACFB_NUM_REGS      8
+#define MACFB_NUM_REGS      (0x200 / sizeof(uint32_t))
 
 typedef struct MacfbState {
     MemoryRegion mem_vram;
-- 
2.20.1

Re: [PATCH 03/10] macfb: increase number of registers saved in MacfbState

[Mark Cave-Ayland]

On 28/02/2022 22:25, Mark Cave-Ayland wrote:

> The MacOS toolbox ROM accesses a number of addresses between 0x0 and 0x200 during
> initialisation and resolution changes. Whilst the function of many of these
> registers is unknown, it is worth the minimal cost of saving these extra values as
> part of migration to help future-proof the migration stream for the q800 machine
> as it starts to stabilise.
> 
> Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
> ---
>   include/hw/display/macfb.h | 2 +-
>   1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/include/hw/display/macfb.h b/include/hw/display/macfb.h
> index 6d9f0f7869..c0e2df8dc6 100644
> --- a/include/hw/display/macfb.h
> +++ b/include/hw/display/macfb.h
> @@ -48,7 +48,7 @@ typedef struct MacFbMode {
>       uint32_t offset;
>   } MacFbMode;
>   
> -#define MACFB_NUM_REGS      8
> +#define MACFB_NUM_REGS      (0x200 / sizeof(uint32_t))
>   
>   typedef struct MacfbState {
>       MemoryRegion mem_vram;

Sigh. This patch isn't right: it's missing a later commit in my local branch that 
needs to be squashed into it, and even then I can now see a bug in that implementation.

I'll send an updated version of this patch in v2.


ATB,

Mark.

[PATCH 04/10] macfb: add VMStateDescription fields for display type and VBL timer

[Mark Cave-Ayland]

These fields are required in the migration stream to restore macfb state
correctly.

Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
---
 hw/display/macfb.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/hw/display/macfb.c b/hw/display/macfb.c
index fb54b460c1..f9ceea532e 100644
--- a/hw/display/macfb.c
+++ b/hw/display/macfb.c
@@ -617,9 +617,11 @@ static const VMStateDescription vmstate_macfb = {
     .minimum_version_id = 1,
     .post_load = macfb_post_load,
     .fields = (VMStateField[]) {
+        VMSTATE_UINT8(type, MacfbState),
         VMSTATE_UINT8_ARRAY(color_palette, MacfbState, 256 * 3),
         VMSTATE_UINT32(palette_current, MacfbState),
         VMSTATE_UINT32_ARRAY(regs, MacfbState, MACFB_NUM_REGS),
+        VMSTATE_TIMER_PTR(vbl_timer, MacfbState),
         VMSTATE_END_OF_LIST()
     }
 };
-- 
2.20.1

[PATCH 05/10] macfb: set initial value of mode control registers in macfb_common_realize()

[Mark Cave-Ayland]

If booting Linux directly in the q800 machine using -kernel rather than using a
MacOS toolbox ROM, the mode control registers are never initialised,
causing macfb_mode_write() to fail to determine the current resolution after
migration. Resolve this by always setting the initial values of the mode control
registers based upon the initial macfb properties during realize.

Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
---
 hw/display/macfb.c | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/hw/display/macfb.c b/hw/display/macfb.c
index f9ceea532e..bcfb2d0dab 100644
--- a/hw/display/macfb.c
+++ b/hw/display/macfb.c
@@ -647,6 +647,14 @@ static bool macfb_common_realize(DeviceState *dev, MacfbState *s, Error **errp)
         return false;
     }
 
+    /*
+     * Set mode control registers to match the mode found above so that
+     * macfb_mode_write() does the right thing if no MacOS toolbox ROM
+     * is present to initialise them
+     */
+    s->regs[DAFB_MODE_CTRL1 >> 2] = s->mode->mode_ctrl1;
+    s->regs[DAFB_MODE_CTRL2 >> 2] = s->mode->mode_ctrl2;
+
     s->con = graphic_console_init(dev, 0, &macfb_ops, s);
     surface = qemu_console_surface(s->con);
 
-- 
2.20.1

[PATCH 06/10] esp: introduce esp_set_pdma_cb() function

[Mark Cave-Ayland]

This function is to be used to set the current PDMA callback rather than
accessing the ESPState pdma_cb function pointer directly.

Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
---
 hw/scsi/esp.c | 19 ++++++++++++-------
 1 file changed, 12 insertions(+), 7 deletions(-)

diff --git a/hw/scsi/esp.c b/hw/scsi/esp.c
index 58d0edbd56..d1640218c4 100644
--- a/hw/scsi/esp.c
+++ b/hw/scsi/esp.c
@@ -195,6 +195,11 @@ static void esp_pdma_write(ESPState *s, uint8_t val)
     esp_set_tc(s, dmalen);
 }
 
+static void esp_set_pdma_cb(ESPState *s, void (*cb)(ESPState *))
+{
+    s->pdma_cb = cb;
+}
+
 static int esp_select(ESPState *s)
 {
     int target;
@@ -356,7 +361,7 @@ static void handle_satn(ESPState *s)
         s->dma_cb = handle_satn;
         return;
     }
-    s->pdma_cb = satn_pdma_cb;
+    esp_set_pdma_cb(s, satn_pdma_cb);
     cmdlen = get_cmd(s, ESP_CMDFIFO_SZ);
     if (cmdlen > 0) {
         s->cmdfifo_cdb_offset = 1;
@@ -387,7 +392,7 @@ static void handle_s_without_atn(ESPState *s)
         s->dma_cb = handle_s_without_atn;
         return;
     }
-    s->pdma_cb = s_without_satn_pdma_cb;
+    esp_set_pdma_cb(s, s_without_satn_pdma_cb);
     cmdlen = get_cmd(s, ESP_CMDFIFO_SZ);
     if (cmdlen > 0) {
         s->cmdfifo_cdb_offset = 0;
@@ -422,7 +427,7 @@ static void handle_satn_stop(ESPState *s)
         s->dma_cb = handle_satn_stop;
         return;
     }
-    s->pdma_cb = satn_stop_pdma_cb;
+    esp_set_pdma_cb(s, satn_stop_pdma_cb);
     cmdlen = get_cmd(s, 1);
     if (cmdlen > 0) {
         trace_esp_handle_satn_stop(fifo8_num_used(&s->cmdfifo));
@@ -464,7 +469,7 @@ static void write_response(ESPState *s)
             s->rregs[ESP_RINTR] |= INTR_BS | INTR_FC;
             s->rregs[ESP_RSEQ] = SEQ_CD;
         } else {
-            s->pdma_cb = write_response_pdma_cb;
+            esp_set_pdma_cb(s, write_response_pdma_cb);
             esp_raise_drq(s);
             return;
         }
@@ -604,7 +609,7 @@ static void esp_do_dma(ESPState *s)
             s->dma_memory_read(s->dma_opaque, buf, len);
             fifo8_push_all(&s->cmdfifo, buf, len);
         } else {
-            s->pdma_cb = do_dma_pdma_cb;
+            esp_set_pdma_cb(s, do_dma_pdma_cb);
             esp_raise_drq(s);
             return;
         }
@@ -646,7 +651,7 @@ static void esp_do_dma(ESPState *s)
         if (s->dma_memory_read) {
             s->dma_memory_read(s->dma_opaque, s->async_buf, len);
         } else {
-            s->pdma_cb = do_dma_pdma_cb;
+            esp_set_pdma_cb(s, do_dma_pdma_cb);
             esp_raise_drq(s);
             return;
         }
@@ -678,7 +683,7 @@ static void esp_do_dma(ESPState *s)
             }
 
             esp_set_tc(s, esp_get_tc(s) - len);
-            s->pdma_cb = do_dma_pdma_cb;
+            esp_set_pdma_cb(s, do_dma_pdma_cb);
             esp_raise_drq(s);
 
             /* Indicate transfer to FIFO is complete */
-- 
2.20.1

Re: [PATCH 06/10] esp: introduce esp_set_pdma_cb() function

[Philippe Mathieu-Daudé]

On 28/2/22 23:25, Mark Cave-Ayland wrote:
> This function is to be used to set the current PDMA callback rather than
> accessing the ESPState pdma_cb function pointer directly.
> 
> Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
> ---
>   hw/scsi/esp.c | 19 ++++++++++++-------
>   1 file changed, 12 insertions(+), 7 deletions(-)

Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>

[PATCH 07/10] esp: introduce esp_pdma_cb() function

[Mark Cave-Ayland]

This function is to be used to execute the current PDMA callback rather than
dereferencing the ESPState pdma_cb function pointer directly.

Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
---
 hw/scsi/esp.c | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)

diff --git a/hw/scsi/esp.c b/hw/scsi/esp.c
index d1640218c4..035fb0d1f6 100644
--- a/hw/scsi/esp.c
+++ b/hw/scsi/esp.c
@@ -200,6 +200,11 @@ static void esp_set_pdma_cb(ESPState *s, void (*cb)(ESPState *))
     s->pdma_cb = cb;
 }
 
+static void esp_pdma_cb(ESPState *s)
+{
+    s->pdma_cb(s);
+}
+
 static int esp_select(ESPState *s)
 {
     int target;
@@ -1268,7 +1273,7 @@ static void sysbus_esp_pdma_write(void *opaque, hwaddr addr,
         esp_pdma_write(s, val);
         break;
     }
-    s->pdma_cb(s);
+    esp_pdma_cb(s);
 }
 
 static uint64_t sysbus_esp_pdma_read(void *opaque, hwaddr addr,
@@ -1290,7 +1295,7 @@ static uint64_t sysbus_esp_pdma_read(void *opaque, hwaddr addr,
         break;
     }
     if (fifo8_num_used(&s->fifo) < 2) {
-        s->pdma_cb(s);
+        esp_pdma_cb(s);
     }
     return val;
 }
-- 
2.20.1

Re: [PATCH 07/10] esp: introduce esp_pdma_cb() function

[Philippe Mathieu-Daudé]

On 28/2/22 23:25, Mark Cave-Ayland wrote:
> This function is to be used to execute the current PDMA callback rather than
> dereferencing the ESPState pdma_cb function pointer directly.
> 
> Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
> ---
>   hw/scsi/esp.c | 9 +++++++--
>   1 file changed, 7 insertions(+), 2 deletions(-)

Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>

[PATCH 08/10] esp: convert ESPState pdma_cb from a function pointer to an integer

[Mark Cave-Ayland]

This prepares for the inclusion of the current PDMA callback in the migration
stream since the callback is referenced by an integer instead of a function
pointer.

Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
---
 hw/scsi/esp.c         | 47 ++++++++++++++++++++++++++++++-------------
 include/hw/scsi/esp.h | 11 +++++++++-
 2 files changed, 43 insertions(+), 15 deletions(-)

diff --git a/hw/scsi/esp.c b/hw/scsi/esp.c
index 035fb0d1f6..e8b6f25dbb 100644
--- a/hw/scsi/esp.c
+++ b/hw/scsi/esp.c
@@ -195,14 +195,10 @@ static void esp_pdma_write(ESPState *s, uint8_t val)
     esp_set_tc(s, dmalen);
 }
 
-static void esp_set_pdma_cb(ESPState *s, void (*cb)(ESPState *))
+static void esp_set_pdma_cb(ESPState *s, int pdma_cb)
 {
-    s->pdma_cb = cb;
-}
-
-static void esp_pdma_cb(ESPState *s)
-{
-    s->pdma_cb(s);
+    assert(pdma_cb >= 0 && pdma_cb < PDMA_NUM_CBS);
+    s->pdma_cb = pdma_cb;
 }
 
 static int esp_select(ESPState *s)
@@ -366,7 +362,7 @@ static void handle_satn(ESPState *s)
         s->dma_cb = handle_satn;
         return;
     }
-    esp_set_pdma_cb(s, satn_pdma_cb);
+    esp_set_pdma_cb(s, PDMA_SATN_PDMA_CB);
     cmdlen = get_cmd(s, ESP_CMDFIFO_SZ);
     if (cmdlen > 0) {
         s->cmdfifo_cdb_offset = 1;
@@ -397,7 +393,7 @@ static void handle_s_without_atn(ESPState *s)
         s->dma_cb = handle_s_without_atn;
         return;
     }
-    esp_set_pdma_cb(s, s_without_satn_pdma_cb);
+    esp_set_pdma_cb(s, PDMA_S_WITHOUT_SATN_PDMA_CB);
     cmdlen = get_cmd(s, ESP_CMDFIFO_SZ);
     if (cmdlen > 0) {
         s->cmdfifo_cdb_offset = 0;
@@ -432,7 +428,7 @@ static void handle_satn_stop(ESPState *s)
         s->dma_cb = handle_satn_stop;
         return;
     }
-    esp_set_pdma_cb(s, satn_stop_pdma_cb);
+    esp_set_pdma_cb(s, PDMA_SATN_STOP_PDMA_CB);
     cmdlen = get_cmd(s, 1);
     if (cmdlen > 0) {
         trace_esp_handle_satn_stop(fifo8_num_used(&s->cmdfifo));
@@ -474,7 +470,7 @@ static void write_response(ESPState *s)
             s->rregs[ESP_RINTR] |= INTR_BS | INTR_FC;
             s->rregs[ESP_RSEQ] = SEQ_CD;
         } else {
-            esp_set_pdma_cb(s, write_response_pdma_cb);
+            esp_set_pdma_cb(s, PDMA_WRITE_RESPONSE_PDMA_CB);
             esp_raise_drq(s);
             return;
         }
@@ -614,7 +610,7 @@ static void esp_do_dma(ESPState *s)
             s->dma_memory_read(s->dma_opaque, buf, len);
             fifo8_push_all(&s->cmdfifo, buf, len);
         } else {
-            esp_set_pdma_cb(s, do_dma_pdma_cb);
+            esp_set_pdma_cb(s, PDMA_DO_DMA_PDMA_CB);
             esp_raise_drq(s);
             return;
         }
@@ -656,7 +652,7 @@ static void esp_do_dma(ESPState *s)
         if (s->dma_memory_read) {
             s->dma_memory_read(s->dma_opaque, s->async_buf, len);
         } else {
-            esp_set_pdma_cb(s, do_dma_pdma_cb);
+            esp_set_pdma_cb(s, PDMA_DO_DMA_PDMA_CB);
             esp_raise_drq(s);
             return;
         }
@@ -688,7 +684,7 @@ static void esp_do_dma(ESPState *s)
             }
 
             esp_set_tc(s, esp_get_tc(s) - len);
-            esp_set_pdma_cb(s, do_dma_pdma_cb);
+            esp_set_pdma_cb(s, PDMA_DO_DMA_PDMA_CB);
             esp_raise_drq(s);
 
             /* Indicate transfer to FIFO is complete */
@@ -787,6 +783,29 @@ static void esp_do_nodma(ESPState *s)
     esp_raise_irq(s);
 }
 
+static void esp_pdma_cb(ESPState *s)
+{
+    switch (s->pdma_cb) {
+    case PDMA_SATN_PDMA_CB:
+        satn_pdma_cb(s);
+        break;
+    case PDMA_S_WITHOUT_SATN_PDMA_CB:
+        s_without_satn_pdma_cb(s);
+        break;
+    case PDMA_SATN_STOP_PDMA_CB:
+        satn_stop_pdma_cb(s);
+        break;
+    case PDMA_WRITE_RESPONSE_PDMA_CB:
+        write_response_pdma_cb(s);
+        break;
+    case PDMA_DO_DMA_PDMA_CB:
+        do_dma_pdma_cb(s);
+        break;
+    default:
+        g_assert_not_reached();
+    }
+}
+
 void esp_command_complete(SCSIRequest *req, size_t resid)
 {
     ESPState *s = req->hba_private;
diff --git a/include/hw/scsi/esp.h b/include/hw/scsi/esp.h
index b1ec27612f..885ccf4660 100644
--- a/include/hw/scsi/esp.h
+++ b/include/hw/scsi/esp.h
@@ -51,7 +51,7 @@ struct ESPState {
     ESPDMAMemoryReadWriteFunc dma_memory_write;
     void *dma_opaque;
     void (*dma_cb)(ESPState *s);
-    void (*pdma_cb)(ESPState *s);
+    uint8_t pdma_cb;
 
     uint8_t mig_version_id;
 
@@ -150,6 +150,15 @@ struct SysBusESPState {
 #define TCHI_FAS100A 0x4
 #define TCHI_AM53C974 0x12
 
+/* PDMA callbacks */
+#define PDMA_SATN_PDMA_CB              0
+#define PDMA_S_WITHOUT_SATN_PDMA_CB    1
+#define PDMA_SATN_STOP_PDMA_CB         2
+#define PDMA_WRITE_RESPONSE_PDMA_CB    3
+#define PDMA_DO_DMA_PDMA_CB            4
+
+#define PDMA_NUM_CBS                   5
+
 void esp_dma_enable(ESPState *s, int irq, int level);
 void esp_request_cancelled(SCSIRequest *req);
 void esp_command_complete(SCSIRequest *req, size_t resid);
-- 
2.20.1

Re: [PATCH 08/10] esp: convert ESPState pdma_cb from a function pointer to an integer

[Philippe Mathieu-Daudé]

On 28/2/22 23:25, Mark Cave-Ayland wrote:
> This prepares for the inclusion of the current PDMA callback in the migration
> stream since the callback is referenced by an integer instead of a function
> pointer.
> 
> Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
> ---
>   hw/scsi/esp.c         | 47 ++++++++++++++++++++++++++++++-------------
>   include/hw/scsi/esp.h | 11 +++++++++-
>   2 files changed, 43 insertions(+), 15 deletions(-)
> 
> diff --git a/hw/scsi/esp.c b/hw/scsi/esp.c
> index 035fb0d1f6..e8b6f25dbb 100644
> --- a/hw/scsi/esp.c
> +++ b/hw/scsi/esp.c
> @@ -195,14 +195,10 @@ static void esp_pdma_write(ESPState *s, uint8_t val)
>       esp_set_tc(s, dmalen);
>   }
>   
> -static void esp_set_pdma_cb(ESPState *s, void (*cb)(ESPState *))
> +static void esp_set_pdma_cb(ESPState *s, int pdma_cb)

Why signed?

>   {
> -    s->pdma_cb = cb;
> -}
> -
> -static void esp_pdma_cb(ESPState *s)
> -{
> -    s->pdma_cb(s);
> +    assert(pdma_cb >= 0 && pdma_cb < PDMA_NUM_CBS);

No need to check >=0 if using unsigned.

> +    s->pdma_cb = pdma_cb;
>   }

> +static void esp_pdma_cb(ESPState *s)
> +{
> +    switch (s->pdma_cb) {
> +    case PDMA_SATN_PDMA_CB:
> +        satn_pdma_cb(s);
> +        break;
> +    case PDMA_S_WITHOUT_SATN_PDMA_CB:
> +        s_without_satn_pdma_cb(s);
> +        break;
> +    case PDMA_SATN_STOP_PDMA_CB:
> +        satn_stop_pdma_cb(s);
> +        break;
> +    case PDMA_WRITE_RESPONSE_PDMA_CB:
> +        write_response_pdma_cb(s);
> +        break;
> +    case PDMA_DO_DMA_PDMA_CB:
> +        do_dma_pdma_cb(s);
> +        break;
> +    default:
> +        g_assert_not_reached();
> +    }
> +}
> +

> index b1ec27612f..885ccf4660 100644
> --- a/include/hw/scsi/esp.h
> +++ b/include/hw/scsi/esp.h
> @@ -51,7 +51,7 @@ struct ESPState {
>       ESPDMAMemoryReadWriteFunc dma_memory_write;
>       void *dma_opaque;
>       void (*dma_cb)(ESPState *s);
> -    void (*pdma_cb)(ESPState *s);
> +    uint8_t pdma_cb;

And here you use unsigned :)

>       uint8_t mig_version_id;
>   
> @@ -150,6 +150,15 @@ struct SysBusESPState {
>   #define TCHI_FAS100A 0x4
>   #define TCHI_AM53C974 0x12
>   
> +/* PDMA callbacks */
> +#define PDMA_SATN_PDMA_CB              0
> +#define PDMA_S_WITHOUT_SATN_PDMA_CB    1
> +#define PDMA_SATN_STOP_PDMA_CB         2
> +#define PDMA_WRITE_RESPONSE_PDMA_CB    3
> +#define PDMA_DO_DMA_PDMA_CB            4

I'd rather use an enum (and in esp_pdma_cb's switch).

> +#define PDMA_NUM_CBS                   5
> +
>   void esp_dma_enable(ESPState *s, int irq, int level);
>   void esp_request_cancelled(SCSIRequest *req);
>   void esp_command_complete(SCSIRequest *req, size_t resid);

Preferrably using unsigned:
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>

Re: [PATCH 08/10] esp: convert ESPState pdma_cb from a function pointer to an integer

[Mark Cave-Ayland]

On 28/02/2022 23:10, Philippe Mathieu-Daudé wrote:

> On 28/2/22 23:25, Mark Cave-Ayland wrote:
>> This prepares for the inclusion of the current PDMA callback in the migration
>> stream since the callback is referenced by an integer instead of a function
>> pointer.
>>
>> Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
>> ---
>>   hw/scsi/esp.c         | 47 ++++++++++++++++++++++++++++++-------------
>>   include/hw/scsi/esp.h | 11 +++++++++-
>>   2 files changed, 43 insertions(+), 15 deletions(-)
>>
>> diff --git a/hw/scsi/esp.c b/hw/scsi/esp.c
>> index 035fb0d1f6..e8b6f25dbb 100644
>> --- a/hw/scsi/esp.c
>> +++ b/hw/scsi/esp.c
>> @@ -195,14 +195,10 @@ static void esp_pdma_write(ESPState *s, uint8_t val)
>>       esp_set_tc(s, dmalen);
>>   }
>> -static void esp_set_pdma_cb(ESPState *s, void (*cb)(ESPState *))
>> +static void esp_set_pdma_cb(ESPState *s, int pdma_cb)
> 
> Why signed?

Ooops.

>>   {
>> -    s->pdma_cb = cb;
>> -}
>> -
>> -static void esp_pdma_cb(ESPState *s)
>> -{
>> -    s->pdma_cb(s);
>> +    assert(pdma_cb >= 0 && pdma_cb < PDMA_NUM_CBS);
> 
> No need to check >=0 if using unsigned.

Agreed.

>> +    s->pdma_cb = pdma_cb;
>>   }
> 
>> +static void esp_pdma_cb(ESPState *s)
>> +{
>> +    switch (s->pdma_cb) {
>> +    case PDMA_SATN_PDMA_CB:
>> +        satn_pdma_cb(s);
>> +        break;
>> +    case PDMA_S_WITHOUT_SATN_PDMA_CB:
>> +        s_without_satn_pdma_cb(s);
>> +        break;
>> +    case PDMA_SATN_STOP_PDMA_CB:
>> +        satn_stop_pdma_cb(s);
>> +        break;
>> +    case PDMA_WRITE_RESPONSE_PDMA_CB:
>> +        write_response_pdma_cb(s);
>> +        break;
>> +    case PDMA_DO_DMA_PDMA_CB:
>> +        do_dma_pdma_cb(s);
>> +        break;
>> +    default:
>> +        g_assert_not_reached();
>> +    }
>> +}
>> +
> 
>> index b1ec27612f..885ccf4660 100644
>> --- a/include/hw/scsi/esp.h
>> +++ b/include/hw/scsi/esp.h
>> @@ -51,7 +51,7 @@ struct ESPState {
>>       ESPDMAMemoryReadWriteFunc dma_memory_write;
>>       void *dma_opaque;
>>       void (*dma_cb)(ESPState *s);
>> -    void (*pdma_cb)(ESPState *s);
>> +    uint8_t pdma_cb;
> 
> And here you use unsigned :)
> 
>>       uint8_t mig_version_id;
>> @@ -150,6 +150,15 @@ struct SysBusESPState {
>>   #define TCHI_FAS100A 0x4
>>   #define TCHI_AM53C974 0x12
>> +/* PDMA callbacks */
>> +#define PDMA_SATN_PDMA_CB              0
>> +#define PDMA_S_WITHOUT_SATN_PDMA_CB    1
>> +#define PDMA_SATN_STOP_PDMA_CB         2
>> +#define PDMA_WRITE_RESPONSE_PDMA_CB    3
>> +#define PDMA_DO_DMA_PDMA_CB            4
> 
> I'd rather use an enum (and in esp_pdma_cb's switch).
> 
>> +#define PDMA_NUM_CBS                   5
>> +
>>   void esp_dma_enable(ESPState *s, int irq, int level);
>>   void esp_request_cancelled(SCSIRequest *req);
>>   void esp_command_complete(SCSIRequest *req, size_t resid);
> 
> Preferrably using unsigned:
> Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>

I'll aim to keep the unsigned type in ESPState, and experiment with changing the 
#defines to an enum. If it looks good, I'll include the enum version in v2.


ATB,

Mark.

[PATCH 09/10] esp: include the current PDMA callback in the migration stream

[Mark Cave-Ayland]

This involves (re)adding a PDMA-specific subsection to hold the reference to the
current PDMA callback.

Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
---
 hw/scsi/esp.c | 23 +++++++++++++++++++++++
 1 file changed, 23 insertions(+)

diff --git a/hw/scsi/esp.c b/hw/scsi/esp.c
index e8b6f25dbb..7998dcfb97 100644
--- a/hw/scsi/esp.c
+++ b/hw/scsi/esp.c
@@ -1210,6 +1210,25 @@ static int esp_post_load(void *opaque, int version_id)
     return 0;
 }
 
+static bool esp_pdma_needed(void *opaque)
+{
+    ESPState *s = ESP(opaque);
+
+    return s->dma_memory_read == NULL && s->dma_memory_write == NULL &&
+           s->dma_enabled;
+}
+
+static const VMStateDescription vmstate_esp_pdma = {
+    .name = "esp/pdma",
+    .version_id = 0,
+    .minimum_version_id = 0,
+    .needed = esp_pdma_needed,
+    .fields = (VMStateField[]) {
+        VMSTATE_UINT8(pdma_cb, ESPState),
+        VMSTATE_END_OF_LIST()
+    }
+};
+
 const VMStateDescription vmstate_esp = {
     .name = "esp",
     .version_id = 6,
@@ -1244,6 +1263,10 @@ const VMStateDescription vmstate_esp = {
         VMSTATE_UINT8_TEST(lun, ESPState, esp_is_version_6),
         VMSTATE_END_OF_LIST()
     },
+    .subsections = (const VMStateDescription * []) {
+        &vmstate_esp_pdma,
+        NULL
+    }
 };
 
 static void sysbus_esp_mem_write(void *opaque, hwaddr addr,
-- 
2.20.1

[PATCH 10/10] esp: recreate ESPState current_req after migration

[Mark Cave-Ayland]

Since PDMA reads/writes are driven by the guest, it is possible that migration
can occur whilst a SCSIRequest is still active. Fortunately active SCSIRequests
are already included in the migration stream and restarted post migration but
this still leaves the reference in ESPState uninitialised.

Implement the SCSIBusInfo .load_request callback to obtain a reference to the
currently active SCSIRequest and use it to recreate ESPState current_req
after migration.

Suggested-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
---
 hw/scsi/esp.c | 10 ++++++++++
 1 file changed, 10 insertions(+)

diff --git a/hw/scsi/esp.c b/hw/scsi/esp.c
index 7998dcfb97..b232af4224 100644
--- a/hw/scsi/esp.c
+++ b/hw/scsi/esp.c
@@ -1342,6 +1342,15 @@ static uint64_t sysbus_esp_pdma_read(void *opaque, hwaddr addr,
     return val;
 }
 
+static void *esp_load_request(QEMUFile *f, SCSIRequest *req)
+{
+    ESPState *s = container_of(req->bus, ESPState, bus);
+
+    scsi_req_ref(req);
+    s->current_req = req;
+    return s;
+}
+
 static const MemoryRegionOps sysbus_esp_pdma_ops = {
     .read = sysbus_esp_pdma_read,
     .write = sysbus_esp_pdma_write,
@@ -1357,6 +1366,7 @@ static const struct SCSIBusInfo esp_scsi_info = {
     .max_target = ESP_MAX_DEVS,
     .max_lun = 7,
 
+    .load_request = esp_load_request,
     .transfer_data = esp_transfer_data,
     .complete = esp_command_complete,
     .cancel = esp_request_cancelled
-- 
2.20.1