<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 Liuairlied@redhat.comkraxel@redhat.comairlied@linux.iedaniel@ffwll.chray.huang@amd.comvirtualization@lists.linux-foundation.orgspice-devel@lists.freedesktop.orgdri-devel@lists.freedesktop.org</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>