powerpc/pseries/svm: Allow IOMMU to work in SVM

H_PUT_TCE_INDIRECT uses a shared page to send up to 512 TCE to
a hypervisor in a single hypercall. This does not work for secure VMs
as the page needs to be shared or the VM should use H_PUT_TCE instead.

This disables H_PUT_TCE_INDIRECT by clearing the FW_FEATURE_PUT_TCE_IND
feature bit so SVMs will map TCEs using H_PUT_TCE.

This is not a part of init_svm() as it is called too late after FW
patching is done and may result in a warning like this:

[    3.727716] Firmware features changed after feature patching!
[    3.727965] WARNING: CPU: 0 PID: 1 at (...)arch/powerpc/lib/feature-fixups.c:466 check_features+0xa4/0xc0

Signed-off-by: Alexey Kardashevskiy <[email protected]>
Reviewed-by: Thiago Jung Bauermann <[email protected]>
Tested-by: Thiago Jung Bauermann <[email protected]>
Signed-off-by: Michael Ellerman <[email protected]>
Link: https://lore.kernel.org/r/[email protected]

arch/powerpc/platforms/pseries/firmware.c

@@ -22,6 +22,7 @@
 #include <asm/firmware.h>
 #include <asm/prom.h>
 #include <asm/udbg.h>
+#include <asm/svm.h>
 
 #include "pseries.h"
 
@@ -101,6 +102,12 @@ static void __init fw_hypertas_feature_init(const char *hypertas,
 		}
 	}
 
+	if (is_secure_guest() &&
+	    (powerpc_firmware_features & FW_FEATURE_PUT_TCE_IND)) {
+		powerpc_firmware_features &= ~FW_FEATURE_PUT_TCE_IND;
+		pr_debug("SVM: disabling PUT_TCE_IND firmware feature\n");
+	}
+
 	pr_debug(" <- fw_hypertas_feature_init()\n");
 }