Skip to contentSkip to conversation

Fixes and updates for hppa target

PATCH Superseded Helge Deller 3 participants 9 messages Jan 12, 2022 – Jan 15, 2022

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

cover Fixes and updates for hppa target

This patchset fixes some important bugs in the hppa artist graphics driver:
- Fix framebuffer access for Linux
- Mouse cursor fixes for HP-UX

New qemu features for hppa:
- Allow up to 16 emulated CPUs (instead of 8)
- Add support for an emulated TOC/NMI button

A new Seabios-hppa firmware:
- Update SeaBIOS-hppa to VERSION 3
- New opt/hostid fw_cfg option to change hostid
- Add opt/console fw_cfg option to select default console
- Added 16x32 font to STI firmware

Helge Deller (5):
seabios-hppa: Update SeaBIOS-hppa to VERSION 3 hw/hppa: Allow up to 16 emulated CPUs
hppa: Add support for an emulated TOC/NMI button. hw/display/artist: Mouse cursor fixes for HP-UX hw/display/artist: Fix framebuffer access for Linux

 hw/display/artist.c       |  58 +++++++++++++++++++++++++++-----------
 hw/hppa/hppa_hardware.h   |   5 ++--
 hw/hppa/machine.c         |  35 ++++++++++++++++++++++-
 pc-bios/hppa-firmware.img | Bin 757144 -> 701964 bytes
 roms/seabios-hppa         |   2 +-
 target/hppa/cpu.c         |   2 +-
 target/hppa/cpu.h         |   5 ++++
 target/hppa/helper.h      |   1 +
 target/hppa/insns.decode  |   1 +
 target/hppa/int_helper.c  |  19 ++++++++++++-
 target/hppa/op_helper.c   |   7 ++++-
 target/hppa/translate.c   |  10 +++++++
 12 files changed, 122 insertions(+), 23 deletions(-)
Signature · 1 line
2.31.1
#1 lore raw

PATCH 1/5 seabios-hppa: Update SeaBIOS-hppa to VERSION 3

New firmware features and fixes:

* Allow up to 16 CPUs

* Add TOC button support:
To trigger a TOC, execute "nmi" in the qemu monitor (Ctrl-A C)

* New opt/hostid fw_cfg option to change hostid:

    -fw_cfg opt/hostid,string=334455

* Add opt/console fw_cfg option to select default console:
-fw_cfg opt/console,string=serial
-fw_cfg opt/console,string=graphics

* Add Linux TER16x32 font to STI firmware:
-fw_cfg opt/font,string=2

* Leave IRQs disabled after rendevouz

---
 pc-bios/hppa-firmware.img | Bin 757144 -> 701964 bytes
 roms/seabios-hppa         |   2 +-
 2 files changed, 1 insertion(+), 1 deletion(-)
pc-bios/hppa-firmware.img binary +0 -0
index a34a8c84f5d96419bbe7baf0b2fed1842a450cdd..b5b682db5d4c1c155379c0a511f5c57da0af7f14 100644
GIT binary patch
roms/seabios-hppa +1 -1
index b12acac4be..ca4add2568 160000
@@ -1 +1 @@
-Subproject commit b12acac4be27b6d5d9fbe48c4be1286dcc245fbb
+Subproject commit ca4add256818419bc85a14d5d2c57ffed40a81fe

--
2.31.1

#2 lore raw

PATCH 2/5 hw/hppa: Allow up to 16 emulated CPUs

In reply to Helge Deller

This brings the hppa_hardware.h file in sync with the copy in the SeaBIOS-hppa sources.

In order to support up to 16 CPUs, it's required to move the HPA for MEMORY_HPA out of the address space of the 16th CPU.

---
 hw/hppa/hppa_hardware.h | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
hw/hppa/hppa_hardware.h +3 -2
index bc258895c9..5edf577563 100644
@@ -25,7 +25,7 @@
 #define LASI_GFX_HPA    0xf8000000
 #define ARTIST_FB_ADDR  0xf9000000
 #define CPU_HPA         0xfffb0000
-#define MEMORY_HPA      0xfffbf000
+#define MEMORY_HPA      0xfffff000
 
 #define PCI_HPA         DINO_HPA        /* PCI bus */
 #define IDE_HPA         0xf9000000      /* Boot disc controller */
@@ -43,9 +43,10 @@
 #define PORT_SERIAL1    (DINO_UART_HPA + 0x800)
 #define PORT_SERIAL2    (LASI_UART_HPA + 0x800)
 
-#define HPPA_MAX_CPUS   8       /* max. number of SMP CPUs */
+#define HPPA_MAX_CPUS   16      /* max. number of SMP CPUs */
 #define CPU_CLOCK_MHZ   250     /* emulate a 250 MHz CPU */
 
 #define CPU_HPA_CR_REG  7       /* store CPU HPA in cr7 (SeaBIOS internal) */
+#define PIM_STORAGE_SIZE 600	/* storage size of pdc_pim_toc_struct (64bit) */
 
 #endif

--
2.31.1

#3 lore raw

PATCH 3/5 hppa: Add support for an emulated TOC/NMI button.

In reply to Helge Deller

Almost all PA-RISC machines have either a button that is labeled with 'TOC' or a BMC/GSP function to trigger a TOC. TOC is a non-maskable interrupt that is sent to the processor. This can be used for diagnostic purposes like obtaining a stack trace/register dump or to enter KDB/KGDB in Linux.

This patch adds support for such an emulated TOC button.

It wires up the qemu monitor "nmi" command to trigger a TOC. For that it provides the hppa_nmi function which is assigned to the nmi_monitor_handler function pointer. When called it raises the EXCP_TOC hardware interrupt in the hppa_cpu_do_interrupt() function. The interrupt function then calls the architecturally defined TOC function in SeaBIOS-hppa firmware (at fixed address 0xf0000000).

According to the PA-RISC PDC specification, the SeaBIOS firmware then writes the CPU registers into PIM (processor internal memmory) for later analysis. In order to write all registers it needs to know the contents of the CPU "shadow registers" and the IASQ- and IAOQ-back values. The IAOQ/IASQ values are provided by qemu in shadow registers when entering the SeaBIOS TOC function. This patch adds a new aritificial opcode "getshadowregs" (0xfffdead2) which restores the original values of the shadow registers. With this opcode SeaBIOS can store those registers as well into PIM before calling an OS-provided TOC handler.

To trigger a TOC, switch to the qemu monitor with Ctrl-A C, and type in the command "nmi". After the TOC started the OS-debugger, exit the qemu monitor with Ctrl-A C.

---
 hw/hppa/machine.c        | 35 ++++++++++++++++++++++++++++++++++-
 target/hppa/cpu.c        |  2 +-
 target/hppa/cpu.h        |  5 +++++
 target/hppa/helper.h     |  1 +
 target/hppa/insns.decode |  1 +
 target/hppa/int_helper.c | 19 ++++++++++++++++++-
 target/hppa/op_helper.c  |  7 ++++++-
 target/hppa/translate.c  | 10 ++++++++++
 8 files changed, 76 insertions(+), 4 deletions(-)
hw/hppa/machine.c +34 -1
index 2a46af5bc9..98b30e0395 100644
@@ -17,6 +17,7 @@
 #include "hw/timer/i8254.h"
 #include "hw/char/serial.h"
 #include "hw/net/lasi_82596.h"
+#include "hw/nmi.h"
 #include "hppa_sys.h"
 #include "qemu/units.h"
 #include "qapi/error.h"
@@ -355,6 +356,14 @@ static void hppa_machine_reset(MachineState *ms)
     cpu[0]->env.gr[19] = FW_CFG_IO_BASE;
 }
 
+static void hppa_nmi(NMIState *n, int cpu_index, Error **errp)
+{
+    CPUState *cs;
+
+    CPU_FOREACH(cs) {
+        cpu_interrupt(cs, CPU_INTERRUPT_NMI);
+    }
+}
 
 static void machine_hppa_machine_init(MachineClass *mc)
 {
@@ -371,4 +380,28 @@ static void machine_hppa_machine_init(MachineClass *mc)
     mc->default_ram_id = "ram";
 }
 
-DEFINE_MACHINE("hppa", machine_hppa_machine_init)
+static void machine_hppa_machine_init_class_init(ObjectClass *oc, void *data)
+{
+    MachineClass *mc = MACHINE_CLASS(oc);
+    machine_hppa_machine_init(mc);
+
+    NMIClass *nc = NMI_CLASS(oc);
+    nc->nmi_monitor_handler = hppa_nmi;
+}
+
+static const TypeInfo machine_hppa_machine_init_typeinfo = {
+    .name = ("hppa" "-machine"),
+    .parent = "machine",
+    .class_init = machine_hppa_machine_init_class_init,
+    .interfaces = (InterfaceInfo[]) {
+        { TYPE_NMI },
+        { }
+    },
+};
+
+static void machine_hppa_machine_init_register_types(void)
+{
+    type_register_static(&machine_hppa_machine_init_typeinfo);
+}
+
+type_init(machine_hppa_machine_init_register_types)
target/hppa/cpu.c +1 -1
index 23eb254228..37b763fca0 100644
@@ -62,7 +62,7 @@ static void hppa_cpu_synchronize_from_tb(CPUState *cs,
 
 static bool hppa_cpu_has_work(CPUState *cs)
 {
-    return cs->interrupt_request & CPU_INTERRUPT_HARD;
+    return cs->interrupt_request & (CPU_INTERRUPT_HARD | CPU_INTERRUPT_NMI);
 }
 
 static void hppa_cpu_disas_set_info(CPUState *cs, disassemble_info *info)
target/hppa/cpu.h +5 -0
index 45fd338b02..93c119532a 100644
@@ -69,6 +69,11 @@
 #define EXCP_SYSCALL             30
 #define EXCP_SYSCALL_LWS         31
 
+/* Emulated hardware TOC button */
+#define EXCP_TOC                 32 /* TOC = Transfer of control (NMI) */
+
+#define CPU_INTERRUPT_NMI       CPU_INTERRUPT_TGT_EXT_3         /* TOC */
+
 /* Taken from Linux kernel: arch/parisc/include/asm/psw.h */
 #define PSW_I            0x00000001
 #define PSW_D            0x00000002
target/hppa/helper.h +1 -0
index 0a629ffa7c..fe8a9ce493 100644
@@ -80,6 +80,7 @@ DEF_HELPER_FLAGS_0(read_interval_timer, TCG_CALL_NO_RWG, tr)
 #ifndef CONFIG_USER_ONLY
 DEF_HELPER_1(halt, noreturn, env)
 DEF_HELPER_1(reset, noreturn, env)
+DEF_HELPER_1(getshadowregs, void, env)
 DEF_HELPER_1(rfi, void, env)
 DEF_HELPER_1(rfi_r, void, env)
 DEF_HELPER_FLAGS_2(write_interval_timer, TCG_CALL_NO_RWG, void, env, tr)
target/hppa/insns.decode +1 -0
index d4eefc0d48..c7a7e997f9 100644
@@ -111,6 +111,7 @@ rfi_r           000000 ----- ----- --- 01100101 00000
 # They are allocated from the unassigned instruction space.
 halt            1111 1111 1111 1101 1110 1010 1101 0000
 reset           1111 1111 1111 1101 1110 1010 1101 0001
+getshadowregs   1111 1111 1111 1101 1110 1010 1101 0010
 
 ####
 # Memory Management
target/hppa/int_helper.c +18 -1
index 13073ae2bd..f599dccfff 100644
@@ -23,6 +23,7 @@
 #include "cpu.h"
 #include "exec/helper-proto.h"
 #include "hw/core/cpu.h"
+#include "hw/hppa/hppa_hardware.h"
 
 #ifndef CONFIG_USER_ONLY
 static void eval_interrupt(HPPACPU *cpu)
@@ -181,7 +182,14 @@ void hppa_cpu_do_interrupt(CPUState *cs)
     }
 
     /* step 7 */
-    env->iaoq_f = env->cr[CR_IVA] + 32 * i;
+    if (i == EXCP_TOC) {
+        env->iaoq_f = FIRMWARE_START;
+        /* help SeaBIOS and provide iaoq_b and iasq_back in shadow regs */
+        env->gr[24] = env->cr_back[0];
+        env->gr[25] = env->cr_back[1];
+    } else {
+        env->iaoq_f = env->cr[CR_IVA] + 32 * i;
+    }
     env->iaoq_b = env->iaoq_f + 4;
     env->iasq_f = 0;
     env->iasq_b = 0;
@@ -219,6 +227,7 @@ void hppa_cpu_do_interrupt(CPUState *cs)
             [EXCP_PER_INTERRUPT] = "performance monitor interrupt",
             [EXCP_SYSCALL]       = "syscall",
             [EXCP_SYSCALL_LWS]   = "syscall-lws",
+            [EXCP_TOC]           = "TOC (transfer of control)",
         };
         static int count;
         const char *name = NULL;
@@ -248,6 +257,14 @@ bool hppa_cpu_exec_interrupt(CPUState *cs, int interrupt_request)
     HPPACPU *cpu = HPPA_CPU(cs);
     CPUHPPAState *env = &cpu->env;
 
+    if (interrupt_request & CPU_INTERRUPT_NMI) {
+        /* Raise TOC (NMI) interrupt */
+        cpu_reset_interrupt(cs, CPU_INTERRUPT_NMI);
+        cs->exception_index = EXCP_TOC;
+        hppa_cpu_do_interrupt(cs);
+        return true;
+    }
+
     /* If interrupts are requested and enabled, raise them.  */
     if ((env->psw & PSW_I) && (interrupt_request & CPU_INTERRUPT_HARD)) {
         cs->exception_index = EXCP_EXT_INTERRUPT;
target/hppa/op_helper.c +6 -1
index 1b86557d5d..b0dec4ebf4 100644
@@ -694,7 +694,7 @@ void HELPER(rfi)(CPUHPPAState *env)
     cpu_hppa_put_psw(env, env->cr[CR_IPSW]);
 }
 
-void HELPER(rfi_r)(CPUHPPAState *env)
+void HELPER(getshadowregs)(CPUHPPAState *env)
 {
     env->gr[1] = env->shadow[0];
     env->gr[8] = env->shadow[1];
@@ -703,6 +703,11 @@ void HELPER(rfi_r)(CPUHPPAState *env)
     env->gr[17] = env->shadow[4];
     env->gr[24] = env->shadow[5];
     env->gr[25] = env->shadow[6];
+}
+
+void HELPER(rfi_r)(CPUHPPAState *env)
+{
+    helper_getshadowregs(env);
     helper_rfi(env);
 }
 #endif
target/hppa/translate.c +10 -0
index c6195590f8..5c0b1eb274 100644
@@ -2393,6 +2393,16 @@ static bool trans_reset(DisasContext *ctx, arg_reset *a)
 #endif
 }
 
+static bool trans_getshadowregs(DisasContext *ctx, arg_getshadowregs *a)
+{
+    CHECK_MOST_PRIVILEGED(EXCP_PRIV_OPR);
+#ifndef CONFIG_USER_ONLY
+    nullify_over(ctx);
+    gen_helper_getshadowregs(cpu_env);
+    return nullify_end(ctx);
+#endif
+}
+
 static bool trans_nop_addrx(DisasContext *ctx, arg_ldst *a)
 {
     if (a->m) {

--
2.31.1

#4 lore raw

PATCH 4/5 hw/display/artist: Mouse cursor fixes for HP-UX

In reply to Helge Deller

This patch fix the behaviour and positioning of the X11 mouse cursor in HP-UX.

The current code missed to subtract the offset of the CURSOR_CTRL register from the current mouse cursor position. The HP-UX graphics driver stores in this register the offset of the mouse graphics compared to the current cursor position. Without this adjustment the mouse behaves strange at the screen borders.

Additionally, depending on the HP-UX version, the mouse cursor position in the cursor_pos register reports different values. To accommodate this track the current min and max reported values and auto-adjust at runtime.

With this fix the mouse now behaves as expected on HP-UX 10 and 11.

---
 hw/display/artist.c | 42 ++++++++++++++++++++++++++++++++++--------
 1 file changed, 34 insertions(+), 8 deletions(-)
hw/display/artist.c +34 -8
index 21b7fd1b44..6384076c60 100644
@@ -80,6 +80,7 @@ struct ARTISTState {
     uint32_t line_pattern_skip;
 
     uint32_t cursor_pos;
+    uint32_t cursor_cntrl;
 
     uint32_t cursor_height;
     uint32_t cursor_width;
@@ -328,19 +329,42 @@ static void artist_get_cursor_pos(ARTISTState *s, int *x, int *y)
 {
     /*
      * Don't know whether these magic offset values are configurable via
-     * some register. They are the same for all resolutions, so don't
-     * bother about it.
+     * some register. They seem to be the same for all resolutions.
+     * The cursor values provided in the registers are:
+     * X-value: -295 (for HP-UX 11) and 338 (for HP-UX 10.20) up to 2265
+     * Y-value: 1146 down to 0
+     * The emulated Artist graphic is like a CRX graphic, and as such
+     * it's usually fixed at 1280x1024 pixels.
+     * Because of the maximum Y-value of 1146 you can not choose a higher
+     * vertical resolution on HP-UX (unless you disable the mouse).
      */
 
-    *y = 0x47a - artist_get_y(s->cursor_pos);
-    *x = ((artist_get_x(s->cursor_pos) - 338) / 2);
+    static int offset = 338;
+    int lx;
+
+    /* ignore if uninitialized */
+    if (s->cursor_pos == 0) {
+        *x = *y = 0;
+        return;
+    }
+
+    lx = artist_get_x(s->cursor_pos);
+    if (lx < offset)
+        offset = lx;
+    *x = (lx - offset) / 2;
+
+    *y = 1146 - artist_get_y(s->cursor_pos);
+
+    /* subtract cursor offset from cursor control register */
+    *x -= (s->cursor_cntrl & 0xf0) >> 4;
+    *y -= (s->cursor_cntrl & 0x0f);
 
     if (*x > s->width) {
-        *x = 0;
+        *x = s->width;
     }
 
     if (*y > s->height) {
-        *y = 0;
+        *y = s->height;
     }
 }
 
@@ -1034,6 +1058,7 @@ static void artist_reg_write(void *opaque, hwaddr addr, uint64_t val,
         break;
 
     case CURSOR_CTRL:
+        combine_write_reg(addr, val, size, &s->cursor_cntrl);
         break;
 
     case IMAGE_BITMAP_OP:
@@ -1422,8 +1447,8 @@ static int vmstate_artist_post_load(void *opaque, int version_id)
 
 static const VMStateDescription vmstate_artist = {
     .name = "artist",
-    .version_id = 1,
-    .minimum_version_id = 1,
+    .version_id = 2,
+    .minimum_version_id = 2,
     .post_load = vmstate_artist_post_load,
     .fields = (VMStateField[]) {
         VMSTATE_UINT16(height, ARTISTState),
@@ -1443,6 +1468,7 @@ static const VMStateDescription vmstate_artist = {
         VMSTATE_UINT32(line_end, ARTISTState),
         VMSTATE_UINT32(line_xy, ARTISTState),
         VMSTATE_UINT32(cursor_pos, ARTISTState),
+        VMSTATE_UINT32(cursor_cntrl, ARTISTState),
         VMSTATE_UINT32(cursor_height, ARTISTState),
         VMSTATE_UINT32(cursor_width, ARTISTState),
         VMSTATE_UINT32(plane_mask, ARTISTState),

--
2.31.1

#5 lore raw

PATCH 5/5 hw/display/artist: Fix framebuffer access for Linux

In reply to Helge Deller

This patch fixes two problems which prevented Linux to access the artist graphics framebuffer:

The check if the framebuffer or the color map should be accessed was incomplete. By using the vram_read/write_bufidx() functions we now check correctly if ARTIST_BUFFER_CMAP should be accessed.

The second fix is to correctly calculate the X- and Y-coordinates and check against the graphics resolution.

With this fix in place, the Linux stifb driver now works correctly, shows the penguins at bootup and uses the stifb as graphics console. I haven't seen any negative side effects when running HP-UX.

---
 hw/display/artist.c | 16 ++++++++--------
 1 file changed, 8 insertions(+), 8 deletions(-)
hw/display/artist.c +8 -8
index 6384076c60..fbf5525334 100644
@@ -1186,7 +1186,7 @@ static void artist_vram_write(void *opaque, hwaddr addr, uint64_t val,
     unsigned int offset;
     trace_artist_vram_write(size, addr, val);
 
-    if (s->cmap_bm_access) {
+    if (vram_write_bufidx(s) == ARTIST_BUFFER_CMAP) {
         buf = &s->vram_buffer[ARTIST_BUFFER_CMAP];
         if (addr + 3 < buf->size) {
             *(uint32_t *)(buf->data + addr) = val;
@@ -1195,14 +1195,14 @@ static void artist_vram_write(void *opaque, hwaddr addr, uint64_t val,
     }
 
     buf = vram_write_buffer(s);
-    posy = ADDR_TO_Y(addr >> 2);
-    posx = ADDR_TO_X(addr >> 2);
+    posy = ADDR_TO_Y(addr);
+    posx = ADDR_TO_X(addr);
 
     if (!buf->size) {
         return;
     }
 
-    if (posy > buf->height || posx > buf->width) {
+    if (posy >= buf->height || posx >= buf->width) {
         return;
     }
 
@@ -1242,7 +1242,7 @@ static uint64_t artist_vram_read(void *opaque, hwaddr addr, unsigned size)
     uint64_t val;
     unsigned int posy, posx;
 
-    if (s->cmap_bm_access) {
+    if (vram_read_bufidx(s) == ARTIST_BUFFER_CMAP) {
         buf = &s->vram_buffer[ARTIST_BUFFER_CMAP];
         val = 0;
         if (addr < buf->size && addr + 3 < buf->size) {
@@ -1257,10 +1257,10 @@ static uint64_t artist_vram_read(void *opaque, hwaddr addr, unsigned size)
         return 0;
     }
 
-    posy = ADDR_TO_Y(addr >> 2);
-    posx = ADDR_TO_X(addr >> 2);
+    posy = ADDR_TO_Y(addr);
+    posx = ADDR_TO_X(addr);
 
-    if (posy > buf->height || posx > buf->width) {
+    if (posy >= buf->height || posx >= buf->width) {
         return 0;
     }
 

--
2.31.1

#6 lore raw

Re: [PATCH 5/5] hw/display/artist: Fix framebuffer access for Linux

+Sven

Quoted 7 lines · Helge Deller

This patch fixes two problems which prevented Linux to access the artist graphics framebuffer:

The check if the framebuffer or the color map should be accessed was incomplete. By using the vram_read/write_bufidx() functions we now check correctly if ARTIST_BUFFER_CMAP should be accessed.

Quoted 5 lines

The second fix is to correctly calculate the X- and Y-coordinates and check against the graphics resolution.

With this fix in place, the Linux stifb driver now works correctly, shows the penguins at bootup and uses the stifb as graphics console.

Cool, could you add a test similar to these?

$ git grep Tux tests/avocado/
tests/avocado/machine_arm_integratorcp.py:69:        Boot Linux and verify the Tux logo is displayed on the framebuffer.
tests/avocado/machine_mips_malta.py:44:        Boot Linux kernel and check Tux logo is displayed on the framebuffer.
Quoted 66 lines

I haven't seen any negative side effects when running HP-UX.

---
  hw/display/artist.c | 16 ++++++++--------
  1 file changed, 8 insertions(+), 8 deletions(-)
hw/display/artist.c +8 -9
index 6384076c60..fbf5525334 100644
@@ -1186,7 +1186,7 @@ static void artist_vram_write(void *opaque, hwaddr addr, uint64_t val,
      unsigned int offset;
      trace_artist_vram_write(size, addr, val);
 
-    if (s->cmap_bm_access) {
+    if (vram_write_bufidx(s) == ARTIST_BUFFER_CMAP) {
          buf = &s->vram_buffer[ARTIST_BUFFER_CMAP];
          if (addr + 3 < buf->size) {
              *(uint32_t *)(buf->data + addr) = val;
@@ -1195,14 +1195,14 @@ static void artist_vram_write(void *opaque, hwaddr addr, uint64_t val,
      }
 
      buf = vram_write_buffer(s);
-    posy = ADDR_TO_Y(addr >> 2);
-    posx = ADDR_TO_X(addr >> 2);
+    posy = ADDR_TO_Y(addr);
+    posx = ADDR_TO_X(addr);
 
      if (!buf->size) {
          return;
      }
 
-    if (posy > buf->height || posx > buf->width) {
+    if (posy >= buf->height || posx >= buf->width) {
          return;
      }
 
@@ -1242,7 +1242,7 @@ static uint64_t artist_vram_read(void *opaque, hwaddr addr, unsigned size)
      uint64_t val;
      unsigned int posy, posx;
 
-    if (s->cmap_bm_access) {
+    if (vram_read_bufidx(s) == ARTIST_BUFFER_CMAP) {
          buf = &s->vram_buffer[ARTIST_BUFFER_CMAP];
          val = 0;
          if (addr < buf->size && addr + 3 < buf->size) {
@@ -1257,10 +1257,10 @@ static uint64_t artist_vram_read(void *opaque, hwaddr addr, unsigned size)
          return 0;
      }
 
-    posy = ADDR_TO_Y(addr >> 2);
-    posx = ADDR_TO_X(addr >> 2);
+    posy = ADDR_TO_Y(addr);
+    posx = ADDR_TO_X(addr);
 
-    if (posy > buf->height || posx > buf->width) {
+    if (posy >= buf->height || posx >= buf->width) {
          return 0;
      }
 
--

2.31.1

#7 lore raw

Re: [PATCH 5/5] hw/display/artist: Fix framebuffer access for Linux

Quoted 14 lines · Philippe Mathieu-Daudé

+Sven

Quoted 10 lines · Helge Deller

This patch fixes two problems which prevented Linux to access the artist graphics framebuffer:
The check if the framebuffer or the color map should be accessed was incomplete. By using the vram_read/write_bufidx() functions we now check correctly if ARTIST_BUFFER_CMAP should be accessed.
The second fix is to correctly calculate the X- and Y-coordinates and
check against the graphics resolution.
With this fix in place, the Linux stifb driver now works correctly, shows the penguins at bootup and uses the stifb as graphics console.

Quoted 9 lines

Cool, could you add a test similar to these?

$ git grep Tux tests/avocado/
tests/avocado/machine_arm_integratorcp.py:69:        Boot Linux and
verify the Tux logo is displayed on the framebuffer.
tests/avocado/machine_mips_malta.py:44:        Boot Linux kernel and
check Tux logo is displayed on the framebuffer.
Quoted 1 line

I haven't seen any negative side effects when running HP-UX.

Hmm, the patch below breaks hp-ux 10.20 for me, please see the attached screenshot.

Quoted 55 lines
Quoted 55 lines
---
  hw/display/artist.c | 16 ++++++++--------
  1 file changed, 8 insertions(+), 8 deletions(-)
hw/display/artist.c +8 -9
index 6384076c60..fbf5525334 100644
@@ -1186,7 +1186,7 @@ static void artist_vram_write(void *opaque, hwaddr addr, uint64_t val,
      unsigned int offset;
      trace_artist_vram_write(size, addr, val);
-    if (s->cmap_bm_access) {
+    if (vram_write_bufidx(s) == ARTIST_BUFFER_CMAP) {
          buf = &s->vram_buffer[ARTIST_BUFFER_CMAP];
          if (addr + 3 < buf->size) {
              *(uint32_t *)(buf->data + addr) = val;
@@ -1195,14 +1195,14 @@ static void artist_vram_write(void *opaque, hwaddr addr, uint64_t val,
      }
      buf = vram_write_buffer(s);
-    posy = ADDR_TO_Y(addr >> 2);
-    posx = ADDR_TO_X(addr >> 2);
+    posy = ADDR_TO_Y(addr);
+    posx = ADDR_TO_X(addr);
      if (!buf->size) {
          return;
      }
-    if (posy > buf->height || posx > buf->width) {
+    if (posy >= buf->height || posx >= buf->width) {
          return;
      }
@@ -1242,7 +1242,7 @@ static uint64_t artist_vram_read(void *opaque,
 hwaddr addr, unsigned size)
      uint64_t val;
      unsigned int posy, posx;
-    if (s->cmap_bm_access) {
+    if (vram_read_bufidx(s) == ARTIST_BUFFER_CMAP) {
          buf = &s->vram_buffer[ARTIST_BUFFER_CMAP];
          val = 0;
          if (addr < buf->size && addr + 3 < buf->size) {
@@ -1257,10 +1257,10 @@ static uint64_t artist_vram_read(void *opaque, hwaddr addr, unsigned size)
          return 0;
      }
-    posy = ADDR_TO_Y(addr >> 2);
-    posx = ADDR_TO_X(addr >> 2);
+    posy = ADDR_TO_Y(addr);
+    posx = ADDR_TO_X(addr);
-    if (posy > buf->height || posx > buf->width) {
+    if (posy >= buf->height || posx >= buf->width) {
          return 0;
      }
--

2.31.1

#8 lore raw

Re: [PATCH 5/5] hw/display/artist: Fix framebuffer access for Linux

Quoted 24 lines · Sven Schnelle
Quoted 22 lines · Philippe Mathieu-Daudé

+Sven

Quoted 10 lines · Helge Deller

This patch fixes two problems which prevented Linux to access the artist graphics framebuffer:
The check if the framebuffer or the color map should be accessed was incomplete. By using the vram_read/write_bufidx() functions we now check correctly if ARTIST_BUFFER_CMAP should be accessed.
The second fix is to correctly calculate the X- and Y-coordinates and
check against the graphics resolution.
With this fix in place, the Linux stifb driver now works correctly, shows the penguins at bootup and uses the stifb as graphics console.

Cool, could you add a test similar to these?

$ git grep Tux tests/avocado/
tests/avocado/machine_arm_integratorcp.py:69:        Boot Linux and
verify the Tux logo is displayed on the framebuffer.
tests/avocado/machine_mips_malta.py:44:        Boot Linux kernel and
check Tux logo is displayed on the framebuffer.
Quoted 5 lines
Quoted 1 line
Quoted 1 line

I haven't seen any negative side effects when running HP-UX.

Hmm, the patch below breaks hp-ux 10.20 for me, please see the attached screenshot.

I think my initial thought that the register 118000 is the buffer access mode for the color map is just wrong. I think it's setting the SRC & DST buffer access mode at the same time because on the Visualize FX cards we have similar registers:

#define B2_FBC_BABoth                     0x00920804 /* DBA & SBA (reads return DBA) (RW) */
#define B2_FBC_DBA                        0x00920808 /* Destination Bitmap Access Register (RW) */
#define B2_FBC_SBA                        0x0092080C /* Source Bitmap Access Register (RW) */

Looking at Artist, we have:

CMAP_BM_ACCESS = 0x118000
DST_BM_ACCESS = 0x118004
SRC_BM_ACCESS = 0x118008

Given that artist and visualize fx are very similar when it comes to 2D acceleration, i think CMAP_BM_ACCESS is just changing both registers, completely unrelated to the color map. I tried changing the code, but of course that breaks a lot of things. Let me see whether i can make that work.

/Sven