Why does Map() fail on default texture despite UnifiedMemoryArchitecture and MapOnDefaultTextures

Docs suggest, that default usage textures can be mapped on UMA architectures like Intel integrated graphics with Direct3D 11.3.

I tried to achieve this, but Map() always fails with E_INVALIDARG.

I am quite new to C++ and DirectX, but below is what I believe to be a minimal test case. Please don't hesitate to point out any stupidity I am committing.

I am running this on a notebook with Windows 10 1809, Intel Skylake i5-6300U with HD Graphics 520.

#include "pch.h"
#include <iostream>
#include <dxgi1_6.h>
#include <d3d.h>
#include <d3d11_4.h>
#include <assert.h>

int main()
{
    HRESULT res = S_OK;
    ID3D11Device *Dev = nullptr;
    ID3D11DeviceContext *Ctx = nullptr;
    D3D_FEATURE_LEVEL Fl;
    D3D_FEATURE_LEVEL fls[1] = { D3D_FEATURE_LEVEL_11_1 };

    res = D3D11CreateDevice(nullptr, D3D_DRIVER_TYPE_HARDWARE, nullptr, D3D11_CREATE_DEVICE_DEBUG | D3D11_CREATE_DEVICE_BGRA_SUPPORT, fls, 1, D3D11_SDK_VERSION, &Dev, &Fl, &Ctx);
    assert(res == S_OK);
    assert(Fl == D3D_FEATURE_LEVEL_11_1);

    ID3D11Device5 *Dev5 = nullptr;
    res = Dev->QueryInterface<ID3D11Device5>(&Dev5);
    assert(res == S_OK);
    Dev->Release();
    Dev = nullptr;

    ID3D11DeviceContext4 *Ctx4;
    res = Ctx->QueryInterface<ID3D11DeviceContext4>(&Ctx4);
    assert(res == S_OK);
    Ctx->Release();
    Ctx = nullptr;

    D3D11_FEATURE_DATA_D3D11_OPTIONS2 opts2;
    res = Dev5->CheckFeatureSupport(D3D11_FEATURE_D3D11_OPTIONS2, &opts2, sizeof(opts2));
    assert(res == S_OK);
    assert(opts2.MapOnDefaultTextures);
    assert(opts2.UnifiedMemoryArchitecture);

    D3D11_TEXTURE2D_DESC1 texDesc = { 0 };
    texDesc.ArraySize = 1;
    texDesc.BindFlags = D3D11_BIND_SHADER_RESOURCE | D3D11_BIND_UNORDERED_ACCESS;
    texDesc.CPUAccessFlags = D3D11_CPU_ACCESS_READ | D3D11_CPU_ACCESS_WRITE;
    texDesc.Format = DXGI_FORMAT_R8G8B8A8_UNORM;
    texDesc.Height = 256;
    texDesc.Width = 256;
    texDesc.MipLevels = 1;
    texDesc.MiscFlags = 0;
    texDesc.SampleDesc.Count = 1;
    texDesc.SampleDesc.Quality = 0;
    texDesc.TextureLayout = D3D11_TEXTURE_LAYOUT_UNDEFINED;
    texDesc.Usage = D3D11_USAGE_DEFAULT;

    byte mem[256 * 256 * 4];
    ZeroMemory(mem, 256 * 256 * 4);

    D3D11_SUBRESOURCE_DATA data = { 0 };
    data.pSysMem = mem;
    data.SysMemPitch = 256 * 4;

    ID3D11Texture2D1 *tex2d;
    res = Dev5->CreateTexture2D1(&texDesc, &data, &tex2d);
    assert(res == S_OK);

    D3D11_MAPPED_SUBRESOURCE map = { 0 };
    // I believe at least one of these should succeed, but all fail
    res = Ctx4->Map(tex2d, 0, D3D11_MAP_READ, 0, &map);
    //res = Ctx4->Map(tex2d, 0, D3D11_MAP_WRITE, 0, &map);
    //res = Ctx4->Map(tex2d, 0, D3D11_MAP_READ_WRITE, 0, &map);
    assert(res == S_OK); // E_INVALIDARG
}

I believe the Map() call should succeed, but it fails with E_INVALIDARG.

EDIT: I tried D3D11_TEXTURE_LAYOUT_ROW_MAJOR and D3D11_TEXTURE_LAYOUT_64K_STANDARD_SWIZZLE too, but then CreateTexture2D1() fails with E_INVALIDARG. Maybe my hardware doesn't support those modes?

I think the issue is described in the documentation:

It is illegal to set CPU access flags on default textures without also setting TextureLayout to a value other than D3D11_TEXTURE_LAYOUT_UNDEFINED.