<p>Hi Christian,</p><p><br></p><p>Can you help explain more detail under what circumstances userspace mapping</p><p>will be problematic, you mean cached mappings also need adjustment in</p><p>function ttm_prot_from_caching?</p><p><br></p><p>Regards,</p><p>Cong.</p><div id="cs2c_mail_sigature"><p></p></div><p></p><div id="re" style="margin-left:0.5em;padding-left:0.5em;border-left:1px solid green;"><br><br><br>
        <div style="background-color:#f5f7fa"><b>主 题:</b><span id="subject">Re: 回复: Re: 回复: Re: [PATCH v1 1/2] drm/qxl: replace ioremap by ioremap_cache on arm64</span>
            <br><b>日 期:</b><span id="date">2022-03-23 18:17</span>
            <br><b>发件人:</b><span id="from">Christian König</span>
            <br><b>收件人:</b><span id="to">liucong2@kylinos.cnairlied@redhat.comkraxel@redhat.comairlied@linux.iedaniel@ffwll.chray.huang@amd.comvirtualization@lists.linux-foundation.orgspice-devel@lists.freedesktop.orgdri-devel@lists.freedesktop.org</span>
            
        </div><br>
        <div id="content"><div class="viewer_part"><div><meta http-equiv="Content-Type" content="text/html; charset=utf-8">Hi Cong,<br>
    <br>
    yes I've seen that, but that is still not sufficient.<br>
    <br>
    You need to update the check in ttm_module.c as well or otherwise
    your userspace mapping might not work correctly either.<br>
    <br>
    Regards,<br>
    Christian. <br>
    <br>
    <div class="moz-cite-prefix">Am 23.03.22 um 11:00 schrieb      <a class="moz-txt-link-abbreviated" href="mailto:liucong2@kylinos.cn">liucong2@kylinos.cn</a>:<br>
    </div><blockquote type="cite" cite="mid:odhwkkfdf7-odj6iduf90@nsmail6.0">   <br>
      <p>Hi <span style="background-color: rgb(245, 247, 250);">Christian,</span></p><p><br>
      </p><p>another commit fix the case in ttm. I send two patches at the
        same time, but seems I miss</p><p> '--cover-letter' when run format-patch or some other bad
        operation.</p><p>----</p><div id="cs2c_mail_sigature"><p><br>
        </p><p>Regards,</p><p>Cong.</p></div><div id="re" style="margin-left:0.5em;padding-left:0.5em;border-left:1px
        solid green;"><br>
        <br>
        <br>
               <div style="background-color:#f5f7fa"><b>主 题:</b><span id="subject">Re: 回复: Re: [PATCH v1 1/2] drm/qxl: replace
            ioremap by ioremap_cache on arm64</span>            <br>
          <b>日 期:</b><span id="date">2022-03-23 17:34</span>            <br>
          <b>发件人:</b><span id="from">Christian König</span>            <br>
          <b>收件人:</b><span id="to"><a class="moz-txt-link-abbreviated" href="mailto:liucong2@kylinos.cnairlied@redhat.comkraxel@redhat.comairlied@linux.iedaniel@ffwll.chray.huang@amd.comvirtualization@lists.linux-foundation.orgspice-devel@lists.freedesktop.orgdri-devel@lists.freedesktop.org">liucong2@kylinos.cnairlied@redhat.comkraxel@redhat.comairlied@linux.iedaniel@ffwll.chray.huang@amd.comvirtualization@lists.linux-foundation.orgspice-devel@lists.freedesktop.orgdri-devel@lists.freedesktop.org</a></span>
                             </div><br>
               <div id="content"><div class="viewer_part"><div>Hi Cong,<br>
                 <br>
                 well than Dave must decide what to do here.<br>
                 <br>
                 When QXL emulates a device it should also not use
              memory accesses    which won't work on a physical device.<br>
                 <br>
                 BTW: Your patch is really buggy, it misses the cases in
              ttm_module.c<br>
                 <br>
                 Regards,<br>
                 Christian.<br>
                 <br>
                 <div class="moz-cite-prefix">Am 23.03.22 um 09:48 schrieb
                     <a class="moz-txt-link-abbreviated
                  moz-txt-link-freetext" href="mailto:liucong2@kylinos.cn" moz-do-not-send="true">liucong2@kylinos.cn</a>:<br>
                   </div><blockquote type="cite" cite="mid:18a8ijpwp6j-18a9shjbr0c@nsmail6.0"><p>Hi Christian,</p><p><br>
                       </p><p>according to 'Arm Architecture Reference Manual
                  Armv8,for        Armv8-A</p><p>architecture profile' pdf E2.6.5</p><p><br>
                       </p><p>E2.6.5 Generation of Alignment faults by load/store
                  multiple        accesses to</p><p> Device memory</p><p><br>
                       </p><p><span style="white-space:pre">    </span>When
                  FEAT_LSMAOC is        implemented and the value of the
                  applicable nTLSMD</p><p><span style="white-space:pre">       </span>field is 0, any
                  memory        access by an AArch32 Load Multiple or
                  Store </p><p><span style="white-space:pre">     </span>Multiple
                  instruction to        an address that the stage 1
                  translation </p><p><span style="white-space:pre">       </span>assigns as
                  Device-nGRE,        Device-nGnRE, or Device-nGnRnE
                  generates </p><p><span style="white-space: pre;">       </span>an Alignment
                  fault.</p><br>
                     <p>so it seems not just some ARM boards doesn't allow
                  unaligned        access to MMIO </p><p>space, all pci memory mapped as Device-nGnRE in arm64
                  cannot        support</p><p>unaligned access. and qxl is a device simulated by
                  qemu, it        should be able to access </p><p>to MMIO space in a more flexible way(PROT_NORMAL)
                  than the real        physical </p><p>graphics card.</p><p><br>
                       </p><p>----</p><p><br>
                       </p><p><br>
                       </p><p>Cong.<br>
                       </p><p><br>
                       </p><p><br>
                         <br>
                         <br>
                                </p><p><strong>主 题:</strong><span id="subject">Re: [PATCH v1
                    1/2]          drm/qxl: replace ioremap by
                    ioremap_cache on arm64</span>                   <br>
                         <strong>日 期:</strong><span id="date">2022-03-23
                    15:15</span>                   <br>
                         <strong>发件人:</strong><span id="from">Christian
                    König</span>                   <br>
                         <strong>收件人:</strong><span id="to">Cong<a class="moz-txt-link-abbreviated
                      moz-txt-link-freetext" href="mailto:Liuairlied@redhat.comkraxel@redhat.comairlied@linux.iedaniel@ffwll.chray.huang@amd.comvirtualization@lists.linux-foundation.orgspice-devel@lists.freedesktop.orgdri-devel@lists.freedesktop.org" moz-do-not-send="true">Liuairlied@redhat.comkraxel@redhat.comairlied@linux.iedaniel@ffwll.chray.huang@amd.comvirtualization@lists.linux-foundation.orgspice-devel@lists.freedesktop.orgdri-devel@lists.freedesktop.org</a></span>
                                            </p><p><br>
                                </p><p>Am 22.03.22 um 10:34 schrieb Cong Liu:<br>
                         > qxl use ioremap to map ram_header and rom,
                  in the arm64        implementation,<br>
                         > the device is mapped as DEVICE_nGnRE, it
                  can not support        unaligned<br>
                         > access.<br>
                         <br>
                         Well that some ARM boards doesn't allow
                  unaligned access to MMIO        space <br>
                         is a well known bug of those ARM boards.<br>
                         <br>
                         So as far as I know this is a hardware bug you
                  are trying to        workaround <br>
                         here and I'm not 100% sure that this is
                  correct.<br>
                         <br>
                         Christian.<br>
                         <br>
                         ><br>
                         >    6.620515] pc : setup_hw_slot+0x24/0x60
                  [qxl]<br>
                         > [    6.620961] lr : setup_slot+0x34/0xf0
                  [qxl]<br>
                         > [    6.621376] sp : ffff800012b73760<br>
                         > [    6.621701] x29: ffff800012b73760 x28:
                  0000000000000001        x27: 0000000010000000<br>
                         > [    6.622400] x26: 0000000000000001 x25:
                  0000000004000000        x24: ffffcf376848c000<br>
                         > [    6.623099] x23: ffff0000c4087400 x22:
                  ffffcf3718e17140        x21: 0000000000000000<br>
                         > [    6.623823] x20: ffff0000c4086000 x19:
                  ffff0000c40870b0        x18: 0000000000000014<br>
                         > [    6.624519] x17: 000000004d3605ab x16:
                  00000000bb3b6129        x15: 000000006e771809<br>
                         > [    6.625214] x14: 0000000000000001 x13:
                  007473696c5f7974        x12: 696e696666615f65<br>
                         > [    6.625909] x11: 00000000d543656a x10:
                  0000000000000000        x9 : ffffcf3718e085a4<br>
                         > [    6.626616] x8 : 00000000006c7871 x7 :
                  000000000000000a        x6 : 0000000000000017<br>
                         > [    6.627343] x5 : 0000000000001400 x4 :
                  ffff800011f63400        x3 : 0000000014000000<br>
                         > [    6.628047] x2 : 0000000000000000 x1 :
                  ffff0000c40870b0        x0 : ffff0000c4086000<br>
                         > [    6.628751] Call trace:<br>
                         > [    6.628994]  setup_hw_slot+0x24/0x60
                  [qxl]<br>
                         > [    6.629404]  setup_slot+0x34/0xf0 [qxl]<br>
                         > [    6.629790]
                   qxl_device_init+0x6f0/0x7f0 [qxl]<br>
                         > [    6.630235]  qxl_pci_probe+0xdc/0x1d0
                  [qxl]<br>
                         > [    6.630654]  local_pci_probe+0x48/0xb8<br>
                         > [    6.631027]
                   pci_device_probe+0x194/0x208<br>
                         > [    6.631464]  really_probe+0xd0/0x458<br>
                         > [    6.631818]
                   __driver_probe_device+0x124/0x1c0<br>
                         > [    6.632256]
                   driver_probe_device+0x48/0x130<br>
                         > [    6.632669]  __driver_attach+0xc4/0x1a8<br>
                         > [    6.633049]  bus_for_each_dev+0x78/0xd0<br>
                         > [    6.633437]  driver_attach+0x2c/0x38<br>
                         > [    6.633789]  bus_add_driver+0x154/0x248<br>
                         > [    6.634168]  driver_register+0x6c/0x128<br>
                         > [    6.635205]
                   __pci_register_driver+0x4c/0x58<br>
                         > [    6.635628]  qxl_init+0x48/0x1000 [qxl]<br>
                         > [    6.636013]  do_one_initcall+0x50/0x240<br>
                         > [    6.636390]  do_init_module+0x60/0x238<br>
                         > [    6.636768]  load_module+0x2458/0x2900<br>
                         > [    6.637136]
                   __do_sys_finit_module+0xbc/0x128<br>
                         > [    6.637561]
                   __arm64_sys_finit_module+0x28/0x38<br>
                         > [    6.638004]  invoke_syscall+0x74/0xf0<br>
                         > [    6.638366]
                   el0_svc_common.constprop.0+0x58/0x1a8<br>
                         > [    6.638836]  do_el0_svc+0x2c/0x90<br>
                         > [    6.639216]  el0_svc+0x40/0x190<br>
                         > [    6.639526]
                   el0t_64_sync_handler+0xb0/0xb8<br>
                         > [    6.639934]  el0t_64_sync+0x1a4/0x1a8<br>
                         > [    6.640294] Code: 910003fd f9484804
                  f9400c23 8b050084        (f809c083)<br>
                         > [    6.640889] ---[ end trace
                  95615d89b7c87f95 ]---<br>
                         ><br>
                         > Signed-off-by: Cong Liu<liucong2@kylinos.cn><br>
                             > ---<br>
                             >   drivers/gpu/drm/qxl/qxl_kms.c | 10
                    ++++++++++<br>
                             >   1 file changed, 10 insertions(+)<br>
                             ><br>
                             > diff --git
                    a/drivers/gpu/drm/qxl/qxl_kms.c        
                     b/drivers/gpu/drm/qxl/qxl_kms.c<br>
                             > index 4dc5ad13f12c..0e61ac04d8ad
                    100644<br>
                             > --- a/drivers/gpu/drm/qxl/qxl_kms.c<br>
                             > +++ b/drivers/gpu/drm/qxl/qxl_kms.c<br>
                             > @@ -165,7 +165,11 @@ int
                    qxl_device_init(struct          qxl_device *qdev,<br>
                             >   (int)qdev->surfaceram_size /
                    1024,<br>
                             >   (sb == 4) ? "64bit" : "32bit");<br>
                             >   <br>
                             > +#ifdef CONFIG_ARM64<br>
                             > + qdev->rom =
                    ioremap_cache(qdev->rom_base,        
                     qdev->rom_size);<br>
                             > +#else<br>
                             >   qdev->rom =
                    ioremap(qdev->rom_base,        
                     qdev->rom_size);<br>
                             > +#endif<br>
                             >   if (!qdev->rom) {<br>
                             >   pr_err("Unable to ioremap ROM\n");<br>
                             >   r = -ENOMEM;<br>
                             > @@ -183,9 +187,15 @@ int
                    qxl_device_init(struct          qxl_device *qdev,<br>
                             >   goto rom_unmap;<br>
                             >   }<br>
                             >   <br>
                             > +#ifdef CONFIG_ARM64<br>
                             > + qdev->ram_header =
                    ioremap_cache(qdev->vram_base          +<br>
                             > +  
                    qdev->rom->ram_header_offset,<br>
                             > +   sizeof(*qdev->ram_header));<br>
                             > +#else<br>
                             >   qdev->ram_header =
                    ioremap(qdev->vram_base +<br>
                             >    
                    qdev->rom->ram_header_offset,<br>
                             >     sizeof(*qdev->ram_header));<br>
                             > +#endif<br>
                             >   if (!qdev->ram_header) {<br>
                             >   DRM_ERROR("Unable to ioremap RAM
                    header\n");<br>
                             >   r = -ENOMEM;<br>
                             <br>
                           </liucong2@kylinos.cn></p></blockquote><br>
               </div></div></div></div></blockquote><br>
  </div></div></div></div>